You are on page 1of 288

Computational Chemistry Study

of Solvents for Carbon Dioxide


Absorption

Eirik Falck da Silva

Doctoral Thesis

Norwegian University of
Science and Technology

Fakultet for Naturvitenskap og Teknologi


Institutt for Kjemisk Prosessteknologi
Trondheim, August 2005
Contents
List of Papers................................................................................................................5
Abstract.........................................................................................................................7
Acknowledgements ......................................................................................................9
1 Introduction........................................................................................................11
1.1 Purpose.................................................................................................... 11
1.2 Global Warming...................................................................................... 12
1.3 Mitigation................................................................................................ 13
1.3.1 Background ..................................................................................... 13
1.3.2 CO2 Capture and Storage ................................................................ 13
2 CO2 Absorption ..................................................................................................17
2.1 Introduction............................................................................................. 17
2.2 Solvents................................................................................................... 19
2.3 Challenges............................................................................................... 20
2.4 Current Understanding of the CO2 Absorption Process.......................... 21
3 Computational Chemistry.................................................................................25
3.1 Introduction............................................................................................. 25
3.2 Quantum Mechanics ............................................................................... 26
3.2.1 Introduction..................................................................................... 26
3.2.2 The Born-Oppenheimer Approximation......................................... 28
3.2.3 Hartree-Fock Self-Consistent Field Method ................................... 28
3.2.4 Post-HF Methods ............................................................................ 28
3.2.5 Density Functional Theory.............................................................. 29
3.2.6 Basis Sets ........................................................................................ 30
3.2.7 Basis Set Superposition Error ......................................................... 31
3.2.8 Temperature .................................................................................... 32
3.2.9 Performance .................................................................................... 32
3.3 Molecular Mechanics.............................................................................. 33
3.3.1 Introduction..................................................................................... 33
3.3.2 Force Field Parameterization .......................................................... 35
3.3.3 Atomic Charges............................................................................... 36
3.3.4 Polarizable Force Fields.................................................................. 38
3.4 Simulations ............................................................................................. 39
3.4.1 Introduction..................................................................................... 39
3.4.2 Free Energy Perturbations............................................................... 41
3.5 QM/MM .................................................................................................. 43
3.6 Computational Chemistry and Experiment............................................. 43
3.7 Review of Computational Chemistry Work on CO2 Absorption............ 45
4 Modeling of Solvation Energy...........................................................................47
4.1 Introduction............................................................................................. 47
4.2 The liquid state........................................................................................ 48

1
4.3 Statistical Mechanics .............................................................................. 49
4.4 Models to Calculate the Free Energy of Solution ................................... 52
4.4.1 Introduction..................................................................................... 52
4.4.2 Equation of State and Lattice Models ............................................. 53
4.4.3 Continuum Models.......................................................................... 56
4.4.4 Molecular Simulation...................................................................... 62
4.4.5 RISM and RISM-SCF..................................................................... 63
4.4.6 Supermolecule Approach................................................................ 64
4.4.7 Hybrids of Computational Chemistry approaches .......................... 65
4.4.8 Descriptor Models........................................................................... 66
4.4.9 Other Models................................................................................... 67
4.4.10 Hybridization of Gibbs Energy Models and Computational
Chemistry Based Models ................................................................................ 68
4.5 Comparison of Methods to Calculate the Free Energy of Solution ........ 69
4.5.1 Methods........................................................................................... 69
4.5.2 The Basicity .................................................................................... 71
4.5.3 Amines ............................................................................................ 72
4.5.4 Results............................................................................................. 73
4.5.5 Conclusion ...................................................................................... 79
5 Reaction Mechanisms and Equilibrium...........................................................81
5.1 Introduction............................................................................................. 81
5.2 Reaction Mechanisms ............................................................................. 81
5.2.1 Introduction..................................................................................... 81
5.2.2 Bicarbonate Formation.................................................................... 82
5.2.3 Carbamate formation....................................................................... 84
5.2.4 Bases ............................................................................................... 87
5.2.5 Alcohol-Group Bonding to CO2 ..................................................... 87
5.2.6 Carbamate as Reaction Intermediate............................................... 88
5.2.7 Molecules with Multiple Amine Functionalities ............................ 89
5.2.8 Shuttle Mechanism.......................................................................... 91
5.2.9 Summary and Conclusion ............................................................... 92
5.3 Determining Equilibrium ........................................................................ 94
5.3.1 Equilibrium and Kinetics ................................................................ 94
5.3.2 Temperature Dependency of Equilibrium Constants...................... 95
5.3.3 Activity Coefficients ....................................................................... 96
5.3.4 Process Energy Consumption ......................................................... 97
5.3.5 Summary ......................................................................................... 98
6 Other Solvent Properties ...................................................................................99
6.1 Introduction............................................................................................. 99
6.2 Solubility in Water .................................................................................. 99
6.3 Solvent Degradation.............................................................................. 100
6.4 Corrosion............................................................................................... 101

2
6.5 Foaming ................................................................................................ 102
6.6 Toxicology ............................................................................................ 103
6.7 Cost ....................................................................................................... 103
6.8 Precipitations......................................................................................... 104
7 Present and Potential Solvents........................................................................105
7.1 Introduction........................................................................................... 105
7.2 Solvents in Use ..................................................................................... 105
7.2.1 Ethanolamine ................................................................................ 105
7.2.2 Tertiary Amines ............................................................................ 106
7.2.3 Sterically Hindered Amines .......................................................... 106
7.2.4 Multiple Amine Functionalities .................................................... 107
7.2.5 Ionic solvents ................................................................................ 107
7.2.6 Patented Solvents .......................................................................... 107
7.3 Ideal Solvent Properties ........................................................................ 108
7.3.1 Equilibrium Constants .................................................................. 108
7.3.2 Other Properties ............................................................................ 110
7.4 Comparison with Ethanolamine............................................................ 111
7.5 Conclusion ............................................................................................ 111
8 Future Work.....................................................................................................113
8.1 Continuation of the Present Research ................................................... 113
8.2 Other Applications of Present Work..................................................... 113
8.3 Beyond Amines..................................................................................... 114
References .................................................................................................................115

3
List of Papers
I. da Silva, E. F. and Svendsen, H. F. (2003) Prediction of the pKa Values of
Amines Using ab Initio Methods and Free Energy Perturbations Ind. Eng.
Chem. Res. 42, 4414-4421.

II. da Silva, E. F. and Svendsen, H. F. (2004) Ab Initio study of the reaction of


carbamate formation from CO2 and alkanolamines Ind. Eng. Chem. Res. 43,
3413-3418.

III. da Silva, E. F. (2004) Use of Free Energy Simulations to predict Infinite


Dilution Activity Coefficients Fluid Phase Eq. 221, 15-24.

IV. da Silva, E. F. (2005) Comparison of Quantum Mechanical and Experimental


Gas Phase Basicities of Amines J. Phys. Chem. A 109, 1603-1607.

V. da Silva, E. F. and Svendsen, H. F. (2005) Study of the Carbamate Stability


of Amines Using ab Initio Methods and Free-Energy Perturbations, Accepted
in Ind. Eng. Chem. Res.

VI. da Silva, E. F., Kuznetsova, T. and Kvamme, B. (2005) Molecular Dynamics


Study of Ethanolamine as a Pure Liquid and in Aqueous Solution.

VII. da Silva, E. F., Yamazaki, T. and Hirata, F. (2005) Comparison of Solvation


Models in the Calculation of Amine Basicity.

5
Abstract
Absorption with aqueous amine solvents is at present the most viable technology
for CO2 capture. While this is a proven technology, efforts are ongoing to improve
it in order to make it a more attractive technology for large scale use to reduce CO2
emissions. Finding solvents with better properties is one approach to improving the
technology.
In this thesis methods in computational chemistry are used to improve the
understanding of the chemistry of CO2 absorption in amine-water systems. The
work is also intended to provide models that can be used to predict the performance
of new solvents. Such predictive models are intended to facilitate the screening for
new solvents.
The main focus of the computational chemistry work has been to model solvent
effects. Most of the work has been based on use of quantum mechanical
calculations to determine gas phase properties and different models to determine
the solvation energy. Most of the solvation energy calculations have been based on
molecular simulations and continuum models. In addition solvation energies
calculated with the RISM-SCF model have been studied.
The reaction mechanisms of the process have been studied in detail. Calculations
have been used to attempt to resolve uncertainties regarding mechanisms. The
work is in most cases in agreement with the consensus in the literature, but it is
concluded that carbamate formation is most likely to be a single-step mechanism.
From the study of reaction mechanisms it is concluded that the reactivity of an
amine solvent with CO2 is governed by two equilibrium constants: the base
stability and carbamate stability.
These two equilibrium constants have been modeled with gas phase quantum
mechanical calculations and different solvation models. Comparison with
experimental data suggests that both equilibrium constants can be modeled with a

7
semi-quantitative accuracy. The models are not entirely accurate but do mostly
capture trends observed in experimental data.
In addition to the equilibrium constants there are other properties that may affect
the overall performance and viability of a solvent in large scale industrial use.
These properties are also discussed and the possibility of modeling them is
assessed.
The ideal values for the main equilibrium constants are unknown and the present
work does therefore not reach any specific conclusions on what the ideal solvent is.
This thesis does however offer a fairly detailed plan of how to find optimal
solvents and tools to carry out the screening.

8
Acknowledgements
I would like to express my gratitude to all those people that in different ways have
helped me with my work on this doctor thesis.
First of all I would like to thank Professor Hallvard Svendsen for giving me the
opportunity to do this work. I am also grateful to him for continuous support,
interesting discussions and giving me freedom in choosing how to approach
different issues. I would also like to thank people in the Department of Chemical
Engineering, in particular in the reactor technology group, for providing a good
working environment. Special thank goes to Karl Anders Hoff and Jana
Poplsteinova Jakobsen for helping to introduce me to the field of gas processing.
My own work has been based on theory and models not studied in the chemical
engineering department. I have therefore often relied on help from researchers at
other groups and institutions.
At NTNU I would especially thank Professor Per-Olof Åstrand at the Department
of Chemistry for insight and discussions on computational chemistry.
Professor Bjørn Kvamme and Tatyana Kuznetsova at the University of Bergen
have been of great help in understanding, and working with, molecular simulation.
Professor Fumio Hirata and Takeshi Yamazaki at the Institute for Molecular
Science in Okazaki, Japan, have helped me with use and understanding of RISM
models. I am also grateful to Professor Hirata for being a very good host during my
visit to the Institute for Molecular Science.
I should also like to thank Professor Bjørn Hafskjold and Yasuo Oishi for first
introducing me to computational chemistry.
I am also grateful to numerous reviewers that have often provided constructive
criticism of my work.
Finally I would like to express special thanks to Mami, for love and support.

9
1 Introduction

1 Introduction
Considering these and many other major and still growing impacts of human
activities on earth and atmosphere, and at all, including global, scales, it seems to
us more than appropriate to emphasize the central role of mankind in geology and
ecology by proposing to use the term 'anthropocene' for the current geological
epoch.
Paul J. Crutzen and Eugene F. Stormer

1.1 Purpose
The main aim of my thesis has been to contribute to the selection of optimal
solvents for CO2 capture from exhaust gases. The main tools of the work have been
various forms of computational chemistry. Selection of solvents for CO2 capture is
not a simple task. There are a number of properties that contribute to determine if a
solvent can be applied economically at an industrial scale. At the same time it is
not a priori given what the ideal properties are. Extensive and time-consuming
experimental work is also required to draw confident conclusions on the viability
of a given solvent. The process of solvent selection or design can perhaps in some
ways be compared to the process of drug design.
Computational chemistry has come to play a significant role in drug design, and
my ambition with this work has been to apply computational chemistry in a similar
fashion in CO2 capture. This work can not resolve all issues regarding solvent
selection, rather it is a part of a larger project where different modeling and
experimental tools are applied together. I have attempted to focus my work on the
central issues where computational chemistry can contribute the most to the
process or solvent selection. Part of the work has gone into making predictions of
properties that can be used to find promising solvent molecules. Another part of the

11
1 Introduction

work has been to contribute to the general understanding of the chemistry of the
systems.
This thesis has three main parts. The first chapters deal with the how-and-why of
this work. First the issue of global warming and mitigation is briefly presented in
the present chapter. An introduction is then given to amine-based CO2 absorption,
the technology which is the topic of the present work. Then computational
chemistry is presented, with special attention given to issue of modeling energies in
solution.
The second part is a set of chapters drawing conclusions from the present work
and looking at remaining issues. At the end is included the papers with the results
of research carried out as a part of this thesis. Most of the actual research findings
are in the papers.

1.2 Global Warming


The United Nations Panel on Climate Change has concluded that if no steps are
taken, human emissions of greenhouse-gases are likely to result in a warming of
1.4 to 5.8 ° C over the next 100 years (IPCC 2001a). Human emissions are already
believed to be affecting the climate, and to have been the main cause of observed
warming during the last century. While it is almost certain that our emissions of
greenhouse-gases will result in the planet becoming warmer, the specific
consequences for life on our planet are more difficult to predict. It has been
observed by some that what we are doing amounts to carrying out a laboratory
experiment with our entire planet. Some of the more likely consequences are
stronger heat-waves, changing precipitation patterns, disappearing glaciers, loss of
biodiversity and rising sea levels (IPCC 2001b). More recent work such as the
Arctic Climate Impact Assessment Report (ACIA 2004) strengthens these
conclusions.

12
1 Introduction

1.3 Mitigation

1.3.1 Background
The issue of anthropogenic global warming leads us to the question of what, if
anything, we can do to combat it. The answer is to reduce our emissions of
greenhouse gases. While the answer is simple there is a significant challenge
involved in carrying out such reductions. Consumption of fossil fuels is at present a
necessity in the industrialized world. Use of fossil fuels inevitably leads to carbon
dioxide (CO2) being formed and CO2 is the greenhouse gas responsible for most of
the anthropogenic global warming (IPCC 2001a). As new countries develop, their
consumption of energy and CO2 emissions are expected to increase.
There are three main alternatives to reducing our CO2 emissions without
hampering economic growth (Haug 2004). One is to use energy more efficiently,
thereby reducing the energy consumption. The second option is to change to
consumption of renewable energy sources and the final option is to burn fossil fuels
while capturing and storing the CO2 instead of releasing it unto the atmosphere.
None of these options are by themselves likely to be enough to stabilize our
emissions of CO2 and the viable way forward is therefore likely to be a
combination of these approaches (IPCC 2001c, Herzog et al. 2000 and Kuuskra et.
al 2004). The option of burning fossil fuels while storing the CO2 instead of
releasing it unto the atmosphere is referred to as “CO2 capture and storage”,
“Carbon capture and storage” or “CO2 sequestration”.

1.3.2 CO2 Capture and Storage


Combustion of fossil fuels takes place as a set of reactions between oxygen and
hydrocarbons with CO2 as one of the products. Usually the reaction is carried out,
not with pure oxygen, but with air as a reactant. The exhaust gas that is produced
resembles air but with a higher concentration of CO2. The total amount of exhaust

13
1 Introduction

gas produced is very large and to store all of it is not an option (Thambimuthu and
Davidson 2004). The CO2 must therefore be separated from the other exhaust gas
components. Once the CO2 has been isolated it can be transported and stored. The
main option for storage being explored at present is geological storage, i.e. to pump
the CO2 in to geological formations below the earth’s surface (Hepple and Benson
2005). The most demanding part of this approach is the capture, how to separate
the CO2 from other exhaust gas components. Other aspects such as compression of
the gas and transport do however also contribute to the overall cost.
There are a number of technologies available for capturing CO2. The technologies
vary in complexity, degree of maturity and cost. The technologies can be separated
into different categories (Thambimuthu and Davidson 2004 and Bolland 2004), a
overview is given in Figure 1.1.

Figure 1.1 Technologies for CO2 capture (Bolland 2004).

14
1 Introduction

Technologies based on capturing the CO2 from the exhaust gas are referred to as
post-combustion technologies. Technologies based on using pure oxygen as a fuel
are called oxyfuel processes. Finally there are technologies based on converting
hydrocarbons to hydrogen and CO2. This approach is called pre-combustion.
Another classification of separation technologies is presented in Figure 1.2.

Figure 1.2 Technologies for CO2 capture (Rao and Rubin 2002).

The present thesis deals with chemical absorption of CO2, this technology is today
the most important post-combustion CO2 capture technology. Of all the available
CO2 capture technologies this also represents at present the most efficient
technology for capturing CO2. This in part reflects technological maturity, the
technology having been patented for natural gas sweetening as early as 1930 (Kohl
and Nielsen 1997). It has also been used in small-scale removal of CO2 from
exhaust gas (Reddy et al. 2003 and Yagi et al. 2004). Chemical absorption is also a
technology that can be fairly easily installed. Existing power-plants can be
retrofitted with equipment for chemical absorption (Thambimuthu and Davidson
2004); whereas many other technologies involve new forms of power plant
technology. Research is being carried out to improve the different technologies,

15
1 Introduction

and improvements are likely to change the relative performance of different


technologies. Recent investigations (Kvamsdal et al. 2004 and de Koeijer 2004)
have however suggested that chemical absorption of CO2 is likely to remain a
highly competitive technology for CO2 capture in the future.

16
2 CO2 Absorption

2 CO2 Absorption
We've learned from experience that the truth will come out. Other experimenters
will repeat your experiment and find out whether you were wrong or right.
Nature's phenomena will agree or they'll disagree with your theory. And, although
you may gain some temporary fame and excitement, you will not gain a good
reputation as a scientist if you haven't tried to be very careful in this kind of work.

Richard P. Feynman

2.1 Introduction
In the present chapter a general presentation will be made of the CO2 capture
technology and the nature of available experimental data for the process will be
summerized. Where nothing else is indicated the material is drawn from the
textbook “Gas Purification” (Kohl and Nielsen 1997). Figure 2.1 illustrates the
apparatus commonly used for CO2 capure.

Figure 2.1 CO2 Absorber columns.

17
2 CO2 Absorption

A cooled exhaust gas is led into the bottom of the absorber column. The gas rises
through the column meeting a counter-current liquid stream. The CO2 absorbs and
reacts with components in the liquid, and the gas stream gradually loses its CO2
while moving up the column.
At the top the gas with low CO2 content is released into the atmosphere. The CO2
content of the liquid increases as the liquid moves down the column. The liquid
stream is typically at 90-95% of equilibrium with incoming exhaust gas at the
column bottom. At the bottom the liquid is taken out and is pumped to the top of a
second column, the stripper (also called desorber). In the stripper the temperature
and/or pressure are set so that the chemical equilibrium in the liquid are reversed
and the CO2 is released into the gas phase. Pressure release is very common in
natural gas applications whereas changing the temperature is the most common
approach for exhaust gas treatment. Change in temperature is usually achieved by
adding heat as steam in the reboiler below the stripper column. A gas phase
consisting only of CO2 and steam is taken out at the top of the column. The steam
is separated from CO2 in the overhead condenser and the CO2 can be compressed
and sent to storage. The liquid at the bottom of the stripper column will have a low
concentration of CO2; and is again ready to be used for CO2 absorption. It is sent
back to the top of the absorber column. The liquid keeps circulating between
absorber and stripping column, transporting the CO2 between the columns. In an
industrial process the absorber will often be operated at temperatures around 40-
55 ° C while the stripper will be operating at around 120 ° C.
Exhaust gas pressure is usually much lower than encountered in natural gas
processing. The Sleipner natural gas sweetening process is for example operated at
a 100 bar (de Koeijer and Solbraa 2004) while exhaust gas is usually at
atmospheric pressure. This means that while the technology for CO2 removal from
natural gas removal can be applied for exhaust gas treatment, optimal operating
conditions and solvents are quite different.

18
2 CO2 Absorption

The concentration of CO2 in the exhaust gases will vary with the nature of the
fossil fuel used. Typically a coal fired power station will have an exhaust
containing 10-12 volume percent CO2 (Rao and Rubin 2001), whereas a natural gas
fired power station can produce an exhaust gas with a CO2 concentration as low as
3 volume percent (Reddy et al. 2003). This means that even for different exhaust
gases optimal process settings may vary.

2.2 Solvents
There are a number of different solvents that are, and have been, applied in CO2
absorption. Most solvents are mixtures of water and base molecules. The bases can
either be organic or inorganic compounds. Almost all organic bases are amine
molecules, and such amine solvents are the topic of the present thesis.
The standard amine solvent for exhaust gases is ethanolamine; this molecule is
shown in the Figure 2.2. Most of the other common solvents are also
alkanolamines, i.e. molecules with both amino and hydroxyl functional groups.
Different amines vary significantly in how they react with CO2. Ethanolamine is a
particularly important solvent because it is the most widely used at present, and it
represents the benchmark that new solvents will be compared with. Amines can be
classified into different groups depending on the number of carbon atoms directly
bonding to the nitrogen atom; there are primary, secondary and tertiary amines.
Tertiary amines differ from the others in how they can react with CO2. I will return
to the various amines and reaction mechanisms in later chapters.

19
2 CO2 Absorption

Figure 2.2 Ethanolamine.

In addition to solvents that react chemically with CO2 there are other solvents that
have a capacity to absorb CO2 without reacting, these are called physical solvents.
There is to my knowledge no physical solvent that is being considered for the
treatment of exhaust gases and chemical absorption is believed to be the viable
option. Solvents with some degree of both chemical and physical absorption might
however be an interesting option. An example of a physical solvent that can be
applied in such a way is Sulfolane (Jenab et al. 2005).

2.3 Challenges
The CO2 absorption process is an established and proven technology, the overall
challenge is to bring the costs down to the point where it becomes an attractive
option in mitigation of global warming. The cost of the process stems from several
components. There is the cost of building the plant and purchasing the solvents.
There are also the operational costs, in particular the energy required to run the
process. Thermal energy is required to heat the solvent entering the stripper, to
generate the heat required to release the CO2 and to generate steam for dilution in
the stripper (Erga, Juliussen and Lidal 1995). In addition electric energy is required
for blowing the exhaust gas through the absorber and for solvent pumping.
One approach to making the process more efficient is to find solvents with more
favorable characteristics than the ones presently in use. This is the topic of the

20
2 CO2 Absorption

present thesis. The performance of the process can also be improved in other ways;
a better design can be proposed or the conditions under which the process is run
may be optimized.
There are other solvent properties besides the reactivity towards CO2 that are of
importance for the overall economy of the process. The solvent may be corrosive,
and the effect on the equipment must be considered. The solvent can also degrade
over time as a consequence of undesired reactions taking place. If the solvent has a
high vapor-pressure it will also evaporate in the absorber and in the stripper. Water
washes are for this reason mounted on the absorber and stripper to recover the
solvent. A recent study has suggested that the water wash operation significantly
affects the overall system performance (Tobiesen, Svendsen and Hoff 2005). The
cost of producing the solvent is also a factor to be considered, particularly if the
solvent has a high degradation rate and must be replaced often. It is also of great
importance that a solvent to be utilized in large quantities in an industrial process is
not toxic. I will return to the various solvent characteristics in a later chapter.

2.4 Current Understanding of the CO2 Absorption Process


The foundation for my work lies in the current level of understanding of the CO2
absorption process. The nature, quality and quantity of experimental data available
is therefore of importance. I have already noted that the CO2 absorption technology
is quite mature, it does however not necessarily follow that the process is well
understood. A lot of experimental work has been published on various aspects of
the absorption process. Here I will briefly go through the nature of published data.
A number of studies deal with the gas-liquid equilibrium of CO2-amine-water
systems. In such equilibrium experiments the system pressure and temperature is
set. The CO2 partial pressure and CO2 concentration in the liquid is then

21
2 CO2 Absorption

determined. Results from such experiments can be used to generate plots as the one
shown in Figure 2.3.
10000

1000
p CO2 / kPa

100

10

1
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

CO2 loading / (m ol of CO2/m ol of MDEA)

Figure 2.3 Plot from Ma’mun, Nilsen and Svendsen (2005).

Another important form of experiment is study of the kinetics. The experimental


setup for kinetics studies varies significantly but the general approach is to measure
the rate of CO2 uptake in the liquid at a given set of conditions. The conditions set
are usually temperature, pressure and liquid composition. Versteeg et al. (1996)
provide a fairly comprehensive review of kinetic data. The CO2 uptake does not
necessarily reflect a single rate of reaction and some analysis work is usually
required to extract reaction kinetics data from the experimental results.
Calorimetric experiments can be used to obtain information on the enthalpy of
CO2 absorption. An example is the work by Oscarson et al. (1989). They report
enthalpy data based on such calorimetric measurements for the CO2–aqueous
diethanolamine system. The same kind of measurements can also be used to
determine heat capacities.

22
2 CO2 Absorption

Physical properties such as density, boiling point and viscosity have also been
measured for many pure amines and amine-water systems. The extent of such data
is however more limited than for common organic molecules.
The types of experiment described so far do not give much direct insight into the
composition of the liquid phase. In recent years some effort has however gone into
experimental work that can provide such information about the nature and
composition of the liquid. Most important is the use of NMR-measurements
(Bishnoi 2000 and Poplsteinova 2004). NMR-data can provide information on the
presence and concentration of various species. From such measurements more
robust and quantitative conclusions can be drawn on the liquid composition. Such
experiments do however also have their limitations. There can be difficulties in
precisely measuring concentrations, protonation equilibriums are difficult to study
and measurement at higher temperatures is difficult.
A lot of the experimental work is done as part of the development of modelling
tools. Models are constructed to predict the overall performance of the CO2
absorption process. The work of Hoff (2004) is an example of such model
development. Such models are based on fitting parameters to experimental data for
a given system and their applicability for conditions beyond those at which they
were fitted can be uncertain.
The main aspects of the absorption process are at present fairly well understood.
The species formed are known and the reactions taking place are fairly well
understood. Equilibrium and kinetic constants are however not that well
established. The liquid consists of a significant number of components in
equilibrium. Determining the relative concentrations of species in the liquid only
from knowledge of amine concentration and CO2 uptake is clearly not an easy task.
The equilibriums and thermodynamics of these systems are for this reason to some
extent uncertain.

23
2 CO2 Absorption

Most of the experimental work reported in the open literature has been carried out
in the context of direct application to process-development and process-modelling.
There is much less published work dedicated to studying the systems in terms of
chemistry and physical chemistry. Data on liquid structure and molecular structure
is for example very sparse. This means that there is little experimental data that can
be used directly in the validation or parameterization of molecular level models
such as the ones used in the present work.

24
3 Computational Chemistry

3 Computational Chemistry
The underlying physical laws necessary for the mathematical theory of a large part
of physics and the whole of chemistry are thus completely known, and the difficulty
is only that exact application of these laws leads to equations much too
complicated to be soluble.
Paul A. M. Dirac (1929)

3.1 Introduction
The present chapter will be devoted to introducing computational chemistry and
it’s various branches. Computational chemistry can perhaps be loosely defined as
chemistry modeling based on a molecular or atomic level description. The term
covers a fairly broad range of theories and methods.
The various methods of computational chemistry can be thought of as offering a
toolbox. For different problems studied choices must be made as to what methods
are best suited. In the present work several different issues related to CO2
absorption are studied. In studying these problems an eclectic approach has been
chosen; namely to attempt to find the method best suited to each problem.
As a result a fairly large number of methods have been employed. And these
methods again draw on different fields of theory. No attempt will be made to cover
the underlying theory in any detail here. The present chapter will rather focus on
introducing the various branches of computational chemistry in general terms. The
introduction is intended to give general insight into the various methods, in
particular their strengths, weaknesses and limitations. In addition the terminology
to be used in the following chapters and papers will be introduced.
The chapter will be divided into two parts, first a brief presentation of the main
elements of computational chemistry will be made. This presentation will follow
the outline used by Grant and Richards (1995) in their textbook “Computational

25
3 Computational Chemistry

Chemistry”. In the second part a discussion will be made on the general issue of the
application of computational chemistry. The issue of modeling of chemistry in
solution will be discussed in the next chapter. Where nothing else is indicated the
material in this chapter is drawn from Grant and Richards (1995) and the textbook
“Essentials of Computational Chemistry” by Cramer (2002).

3.2 Quantum Mechanics

3.2.1 Introduction
The claim made by Dirac regarding the laws of chemistry and physics (quoted in
the beginning of this chapter) referred mainly to the postulation of the Schrödinger
equation. In its barest, and most innocent, form it can be written as:
H ψ = Eψ (3.1)
H is the shorthand form of the Hamilton operator which takes into account the
contributions to the energy of the system. E is the energy of the system and ψ is
the wave function. The energy has only certain allowed values, with a
corresponding wave function for each allowed energy level.
The Hamiltonian consists of the potential and kinetic energy contributions. In the
absence of external magnetic and electrical fields and ignoring relativistic effects it
takes the following form:
= 2 = 2 e2 Z k e2 e2 Z k Zl
H = −∑ ∇i − ∑ ∇k − ∑∑ +∑ +∑ (3.2)
i 2me k 2mk i k rij i< j rij k <l rkl

where i and j run over electrons, k and l run over nuclei, = is the Planck’s constant
divided by 2 π , me is the mass of the electron, mk is the mass of the nucleus k.

∇2 is the Laplacian operator, e is the charge on the electron, Z is an atomic number


and rab is the distance between particles a and b . From the Hamiltonian it can be
seen that the Schrödinger equation is a set of differential equations.

26
3 Computational Chemistry

For the wave function itself it is difficult to give a simple definition or direct
physical interpretation. The product of the wave function with its complex
conjugate ψ * ψ does however have a physical interpretation; it gives the

probability density for the system. For an electron ψ * ψ multiplied with a volume

element would give the probability of the electron being in that volume element.
2
The normalized integral of ψ over all space must be unity. The wave function

can therefore be thought of as a kind of road-map to how the electrons are


localized.
There is no way to directly derive the wave function itself, but there are some
conditions it must meet. It must be “well-behaved”, displaying only smooth
changes and going to zero at infinity. The variational principle states that the lower
the ground state energy calculated by a wave function is, the higher is the quality of
the wave function.
Assuming that each electron can be treated separately one can operate with one-
electron wave functions also called orbitals. In a system with more than one atom,
i. e. a molecule, we deal with molecular orbitals.
All electrons are characterized by a spin quantum number, with two possible
eigenvalues. The Pauli principle states that two electrons can not have the same
quantum numbers. One molecular orbital is therefore limited to two electrons with
opposite spin.
The Schrödinger equation is a postulate, believed to be entirely accurate. The
complexity of it is however such that the largest system for which it is analytically
solvable is the hydrogen atom. It was this state of affairs that led Dirac to make his
observation.
Since Dirac made his observation a lot of work has gone into making
approximations that make it possible to make calculations on systems of practical
interest. Some of the main approximations will be briefly outlined here.

27
3 Computational Chemistry

3.2.2 The Born-Oppenheimer Approximation


The atoms in a system are much heavier and move much more slowly than the
electrons. It is therefore assumed that the movements can be decoupled. The energy
of the electrons is calculated with the atoms in fixed positions. This approximation
is in most cases entirely reasonable and universally applied.

3.2.3 Hartree-Fock Self-Consistent Field Method


Much of the difficulty of solving the Schrödinger equation stems from the need to
simultaneously determine the energy of each electron in the presence all other
electrons. In the Hartree-Fock (HF) method this is avoided by calculating the
energy of each electron in the averaged static field of the others. Initially a guess is
made of the electron energies. The energy of each electron is then calculated in the
field of the initial electron configuration. This procedure is repeated in an iterative
loop until convergence (Self-Consistent referring to this iterative calculation).
The Hartree-Fock method can therefore be thought of as a kind of mean-spherical
approximation at the electron level. The difference between the Hartree-Fock
energy and the energy for the full Schrödinger equation is called the correlation
energy. Hartree-Fock calculations are sufficiently accurate to provide insight into
many problems and they are widely used. As Hartree-Fock calculations have been
applied to different problems it has however become increasingly clear that the
correlation energy is of great significance in determining the properties of a system.
Efforts have therefore been made to improve on the Hartee-Fock energy.

3.2.4 Post-HF Methods


There a number of different methods that go beyond Hartree-Fock calculations, one
of the widely used approaches is perturbation theory. In perturbation theory the
Hartree-Fock solution is treated as the first term in a Taylor series. The

28
3 Computational Chemistry

perturbation terms added involve the electron repulsion. One of the more common
forms was developed by Møller and Plesset. The second order perturbation form is
referred to as MP2. This form will be utilized in the present work.
It should be noted that the electron-electron repulsion energy is not necessarily a
small perturbation. In cases in which this term is large the application of
perturbation theory can become more difficult.
There are a number of other techniques to include electron correlation that can
potentially provide very accurate results, such calculations can however become
very time consuming and at present they tend to be used for small molecules with
maybe 3-4 heavy (non-hydrogen) atoms. The molecules studied in the present work
are somewhat larger and the decision has been made not to use such time-
consuming methods.

3.2.5 Density Functional Theory


Density Functional Theory (DFT) is based on determining the electron density
rather than the wave function. The electron density unlike the wave function is a
physically observable quantity. It has been proven that given the electron density
the Hamilitonian operator is also determined. A variational principle has also been
established for DFT. Unlike HF theory DFT in itself contains no approximations.
There is however no way to derive an energy contribution in DFT known as the
exchange-correlation energy. The quality of the models is usually determined by
some form of comparison with experimental data. DFT models are therefore in a
sense semi-empirical models and once assumptions about the exchange-correlation
energy are introduced (as they must be) there is no variational principle. This
means that for DFT, unlike HF and post-HF methods, there is no a priori way to
establish how good a given method is and no systematic way to improve upon it.
This state of affairs led many researchers to look at DFT with skepticism. It has

29
3 Computational Chemistry

however become clear that DFT methods often produce results of comparable
quality to much more expensive post-HF methods. It has also become fairly well
established for what type of molecules and properties DFT methods are reliable.
One of the most common DFT methods is the so-called B3LYP method, which is a
form of hybrid between DFT and HF methods. It is considered to be fairly robust,
perhaps because it balances some of the weaknesses of DFT and HF methods.
B3LYP is the DFT method that will be used in the present work.

3.2.6 Basis Sets


HF and Perturbation Theory have taken us from the Schrödinger equation to a
solvable set of equations (DFT offering an alternative route). In order to carry out
calculations a representation of the wave function is also needed. Each molecular
orbital is constructed from linear combinations of basis functions.
2
For computational reasons gaussian type orbitals ( e−r ) are commonly used.
Gaussian type orbitals do however not have the correct shape required to reproduce
the form of a electron distribution. Orbitals are therefore usually constructed as
combinations of a set of gaussians in order to reproduce the correct shape.
Basis-sets must be sufficiently flexible to allow the description of electron
distribution in various forms of molecules and the quality of the results obtained do
in general improve with increasing size and flexibility of the functions employed.
On the other hand calculations will also become more time consuming with
increasing basis set size. One of the common approaches is to add more basis sets
for the valence electrons compared to inner orbitals.
In the present work the common 3-21G, 6-31G and 6-311G basis sets will be
utilized. 3-21G indicates a single basis set consisting of 3 gaussian functions for
inner electrons and two separate basisfunctions, one consisting of 2 gaussians
functions and the other 1 gaussian function for valence electrons. In 6-31G the

30
3 Computational Chemistry

number of gaussian functions to represent the basis sets is increased and in 6-311G
the number of separate basis functions for valence electrons is also increased.
It is common to add further sets of basis functions. One approach is to add higher
level orbitals to electrons at a given level, one may for example add d-orbitals to
electons in a p-orbital and p-orbitals to electrons occupying s-orbitals. Such orbitals
are called polarizable orbitals and the inclusion of such d-orbitals will in the
present work be indicated with a (d) and p-orbitals with (p). It is common to
indicate the use of polarizable orbitals with a (x,y) notation, where x is the number
of polarizable orbitals on heavy (non-hydrogen) atoms and y indicates the
polarizable functions on the hydrogen atoms. Another notation that is sometimes
used is to indicate (d) polarization with a “*” and (d,p) polarization with “**”. This
notation is utilized in one of the papers in the present work.
Finally there is in some cases a special need to allow electrons to localize far
from the atom center. Standard basis-sets are in such cases augmented with so-
called diffuse basis sets. In the present work such diffuse basis-sets on heavy (non-
hydrogen) atoms are indicated with a “+”, if they are also included on hydrogen
atoms it is indicated with “++”. One of the circumstances in which such diffuse
basis sets are required is in the accurate modeling of hydrogen bonds.
Some basis-sets are regarded as being better than others in providing quality
results for a given amount of computation time. Some basis sets have therefore
become standard for calculations. In the present work all calculations will be done
with such widely used basis-sets.

3.2.7 Basis Set Superposition Error


When atoms interact the basis sets allocated to each of them will overlap. This
overlapping gives electrons greater freedom to localize and can result in a
reduction of the energy. This reduction in energy would however not have occurred

31
3 Computational Chemistry

if the basis sets had been infinitely large. This energy reduction is therefore an
artifact of working with limited basis sets. This is called the basis set superposition
error (BSSE).
For atoms on different molecules there are schemes to correct for the BSSE. Most
common is the so-called Counterpoise correction. For interactions within the same
molecule application of such corrections is however more difficult (Reiling et al.
1996 and Lii et al. 1999). This is of importance in the present work because many
alkanolamine molecules display intramolecular hydrogen bonding. The BSSE is
expected to become smaller with increasing basis set and in calculating
intramolecular hydrogen bonds it would therefore seem that larger basis sets are
more reliable. In the present work the general approach will therefore be to use
large basis-sets in order to obtain more accurate results.

3.2.8 Temperature
Standard quantum mechanical calculations are usually carried out on a single or
small number of molecules at 0 K, and without accounting for the zero-point
energy. The intramolecular effects of temperature are usually calculated by using
the harmonic oscillator approximation. This relies on calculating the second
derivative of the energy with respect to the displacement ( r ).
It should be noted that because quantum mechanical calculations are usually
carried out on very small number of molecules in vacuum, pressure effects are not
accounted for.

3.2.9 Performance
The quantum mechanics based methods are often referred to as ab inito methods,
as none of the methods rely on parameterization to experimental data. This is an
important point because it distinguishes quantum mechanical calculations from
many other forms of modeling carried out in science. The development of such

32
3 Computational Chemistry

calculations has however not taken place without experimental input (this being
particularly true for DFT methods). Comparison with experimental data is used to
validate the calculations. Sometimes comparisons with experiment have shown
methods to be less reliable than expected, while others have proven more reliable.
This partly happens because there can occur various forms of fortuitous
cancellation of errors.
It has already been noted that quantum mechanical calculations can be time-
consuming. Some of the calculations in the present work took 2-4 days of CPU
time.
Quantum mechanical calculations can today be carried out for systems of up to
maybe a 100 atoms. The calculation time can increase quite steeply when
increasing the size of the basis set or using more advanced methods.
In the present work quantum mechanical calculations will mainly be used to
calculate geometries and energies of molecules.
For geometry optimization most quantum mechanical methods are fairly reliable.
High level calculations are of quality comparable to experimental data, HF
calculations with smaller basis sets also tend to be reasonably accurate.
Calculation of energy is in general more difficult. Results can vary quite
significantly with the level of theory. Prediction of absolute energy values are
difficult, but relative trends in energies can usually be calculated with reasonable
accuracy.

3.3 Molecular Mechanics

3.3.1 Introduction
Quantum mechanical calculations are, as have already been mentioned, time-
consuming. Molecular Mechanics (MM) offer a simplified form of molecular
representation that makes it possible to perform significantly faster calculations.

33
3 Computational Chemistry

A molecular mechanics representation can best be summarized as soft spheres


attached by springs to represent bonds. The potential energy between non-bonded
atoms is usually expressed as the sum of Lennard-Jones and Coulomb potential
functions:
⎡⎛ ⎞12 ⎛ ⎞6 ⎤
⎢ σ ⎟ σ ⎟⎥ qq
U = ∑ 4εij ⎢⎜⎜⎜ ij ⎟⎟ − ⎜⎜⎜ ij ⎟⎟ ⎥ + ∑ i j (3.3)
i< j ⎢⎜⎝ rij ⎟⎠ ⎜⎝ rij ⎠⎟ ⎥ i< j rij
⎣ ⎦
where the sums are over all pairs of interaction sites and ε and σ are the Lennard-
Jones potential parameters, qi is the partial electric charge of interaction site i and
rij is the separation between interaction sites. Interaction sites are usually, but not
always, atomic centers. This form of representation only accounts for two-body
interactions. In a real system many-body effects, such as three-body and four-body
interactions, can also play a part. There is therefore an approximation involved in
the form of such potential functions. To correct for this, parameters can be set to
implicitly account for the many-body effects.
For bond-lengths simple harmonic stretching functions are often used where the
energy increases as the bond-length deviates from some equilibrium bond-length.
For bond angles harmonic functions of the following form are often utilized:
2
U (θ) = kθ (θ −θ 0 ) (3.4)

where θ is the bond angle and the subscript 0 denotes the equilibrium value. kθ is
the spring constant. Dihedral angle energies around bonds are given by some form
a fourier series. One of the common forms is the following:
5
U (φ) = ∑ Ci cos (φ)
i
(3.5)
i =1

where φ is the dihedral angle and the Ci are constants.

34
3 Computational Chemistry

3.3.2 Force Field Parameterization


The quality of the results produced by MM calculations obviously depends on the
parameters chosen for the various interactions and some form of parameterization
must be undertaken. A set of parameters for a single molecule or groups of
molecules are called force fields.
A fairly large number of schemes have been proposed to develop force fields,
partly reflecting the fact that different researchers are looking at different
applications. One of the main applications is biological systems, in which case the
focus is often on reproducing the structural characteristics of molecules of
biological importance. A second application is the modeling of liquids.
In the modeling of liquids, parameters are often chosen to reproduce the
properties of liquids as determined from experimental work. There is however a
number of different properties one can choose to reproduce. Among them are
density, diffusion rates, dielectric constants and radial distribution functions. An
example of such work is the “Optimized Parameters for Liquid Simulations”
(OPLS) force field developed by Jorgensen and coworkers (Jorgensen et al. 1996
and Rizzo and Jorgensen 1999). In addition there is the choice of attempting to
reproduce the properties for a given temperature, or to attempt the more ambitious
task of reproducing properties over a range of temperatures as done by Walser et al.
(2000). For some solvents such as water, a fairly large body of experimental data is
available. In other cases, such as for the amines of interest in the present study,
experimental data is however more sparse.
For a solvent such as water there exist a remarkably large number of force fields
(Guillot 2002). This confronts the practitioner with some difficult choices when
selecting force fields for a specific problem. It would seem that some work remains
on determining which force fields are more reliably for specific tasks, and if any
can be regarded as “better” in a general sense.

35
3 Computational Chemistry

Quantum mechanical calculations are often used for setting force field
parameters. They can for example be used for determining molecular geometries
and atomic charges.

3.3.3 Atomic Charges


One of the most important and difficult issues in the design of a force field is the
selection of charges. In most common force fields fixed charges are used and they
are often, but not always, located at the atom centers. These atomic charges are
intended to reproduce the net effect of electrons and nuclei for a given atom. As
electrons are not located at a single point operating with charges situated at the
atomic centers does represent an approximation. Operating with fixed charges is
also an approximation, as the location of electrons can be effected by the
environment the molecule finds itself in.
Two main approaches to determining atomic charges can be identified in the
literature. One approach is to fit the charges in simulations intended to reproduce
various experimental properties. Such fitting is usually done for small organic
molecules. For larger molecules experimental data is often more sparse and the
number of charges to fit is much larger. In such cases one will often rely on atomic
charges being transferable parameters. Having determined charges for alcohol-
groups, amine-groups and alkane-groups in small molecules one assumes these to
be the same in larger molecules. The OPLS force field is based on this approach
(Jorgensen et al. 1996). The second approach is to determine the atomic charges
from quantum mechanical calculations, an example of this is the work by Kollman
and coworkers (Cornell et al. 1995). This second approach is very appealing
because it reduces the need for experimental data and gives the modeling a stronger
predictive character (provided it works).

36
3 Computational Chemistry

Even if a quantum mechanical calculation contains information about position of


nucleus and electrons, the task of determining atomic charges is still a difficult one.
Atomic charges are not uniquely defined and the task of assigning parts of the
electron distribution to atoms in a molecule is ambiguous. The first such scheme
was the Mullikan population, which is based on determining how much each
atomic basis set contribute to the wave function. While Mullikan populations have
been widely used, they have come to be regarded as unreliable (Franckl and
Chirlian 2000). One of the newer schemes is to reproduce the electrostatic potential
around the solute, even for this approach there are however a number of different
implementations (Franckl and Chirlian 2000). Singh and Kollmann (1984)
developed a procedure based on reproducing the electrostatic potential on
gridpoints distributed spherically around each solute atom center, outside the van
der Waals volume of the solute. This type of charges will be utilized in the present
work, these will be referred to by their common acronym “MK”. It has become
clear that the calculated charges are sensitive to the specific procedure chosen to fit
the electrostatic potential (Franckl and Chirlian 2000). Other difficulties are that
such procedures tend to work poorly for atoms buried inside a molecule and that
charges can display a high degree of conformer dependency (Bayly et al. 1993).
There are also schemes that attempt to use the quantum mechanical
representation while at the same time reproducing some experimentally measured
property. One such hybrid scheme is CM2 charges (Li, Zhu, Cramer and Truhlar
1998) that reproduces experimental dipole moments. This type of charges will also
be used in the present work.
While selection of atomic charges is a difficult issue, it should also be noted that
in many contexts different force fields produce quite similar results. In such cases
one does not have to worry too much about the selection of charges. In general
different schemes to calculate atomic charges do also produce charges that are in
reasonable qualitative agreement (da Silva, Yamazaki and Hirata 2005).

37
3 Computational Chemistry

3.3.4 Polarizable Force Fields


One of the most important approximations in standard simulations with molecular
mechanics representation is the use of fixed charges. Introduction of polarizability
is one way to improve the representation while avoiding the expense of quantum
mechanical calculations. Such models were recently reviewed by Rick and Stuart
(2002). There are some main approaches to adding polarization in simulations.
Among them are shell models based on polarizable point dipoles, were fixed
charges are attached to each other with harmonic springs. Another form of model is
based on charges being allowed to fluctuate between sites in a molecule.
The charges in a molecule do depend on the surrounding environment. It is
therefore to be expected that a model with fixed charges will have problems
representing a molecule in different states such as solids, liquids and gases.
Polarizable models should have the potential to represent a molecule in different
states. Another difficulty with fixed charges is that they can not reflect changes in
charge distribution that may take place as a molecule changes conformer. This is
again something that a polarizable model has the potential to handle. On the other
hand simulations with polarizable models do take longer time than simulation with
fixed charges.
Rick and Stuart (2002) conclude that polarizable models in several respects do
perform better than models with fixed charges. Compared to a model with fixed
charges there is however a greater number of parameters to be set in a polarizable
model and this does offer some added challenges.
While a polarizable model is in form more realistic than a model with fixed
charges, it is not given that it will produce more realistic results. In using a ball-
and-stick representation of molecules there is a number of assumptions involved,
and in it is not given that overall performance will improve by improving on one of
the approximations.

38
3 Computational Chemistry

In the present work simulations with polarizable molecular representations are


not used. Mainly because I feel that such advanced and time-consuming modeling
should only be utilized when simpler fixed charge models are shown to be
inadequate. Such forms of models should however be considered if fixed charge
models are found wanting in a given context.

3.4 Simulations

3.4.1 Introduction
There are two forms of simulations that are used in computational chemistry:
Molecular Dynamics (MD) and Monte Carlo (MC). These are used for calculations
of ensembles of molecules. Simulation techniques are described in detail in the
textbooks “Computer Simulations of Liquids” (Allen and Tildesley 1987) and
“Understanding molecular simulation” (Frenkel and Smit 2002).
Molecular Dynamics calculations are based on calculating the forces between
molecules and atoms in a system and allowing them to move according to Newtons
laws of motion. From the calculated forces the acceleration and velocity of the
particles in the system are calculated. The particles are moved over a small time-
step, forces and velocities are recalculated and the system is moved forward a new
time-step. For each time-step the properties of the system such as energy and
temperature are monitored. A simulation is carried out for whatever number of
time-steps is deemed necessary to obtain reliable averages.
Monte Carlo simulations are on the other hand based on random alterations of the
coordinates of the system. The energy change for each alteration is calculated, the
probability of the alteration being accepted depending on the associated change in
energy. For changes leading to lower energy the probability is higher. The standard
approach is Metropolis sampling in which the sampling has a Boltzmann-weighted

39
3 Computational Chemistry

probability. As in MD the simulation is continued for whatever number of steps


deemed necessary for sampling.
Often in simulations the purpose is to simulate the bulk behavior of liquids.
Simply placing a number of molecules in a vacuum would produce a cluster that
might have properties different from bulk liquid. It is therefore customary both in
MD and MC to do calculations with periodic boundary conditions. The cell
containing the ensemble is then surrounded by replicas of itself.
Simulations are usually carried out with a molecular mechanics level
representation. Simulations with such a molecular representation can be carried out
on ensembles of thousands of molecules. This is significantly more than in QM
calculations, but is still an extremely small number compared to the number of
molecules present in even the smallest droplet of water. In the present work most
simulations are done on ensembles of 256 molecules. Such an ensemble is usually
regarded as large enough for reliable calculations, at the same time as such
calculations can be carried out in reasonable amounts of time.
The Lennard-Jones and Coulomb interactions are usually truncated at some
value. This is mainly done to save time in the calculations. The Lennard-Jones
potential decays steeply as a function of distance and its truncation is
unproblematic. Coulomb interactions have a slower decay, but for neutral species
the truncation can still be a reasonable simplification. For ionic species truncation
is however usually not an acceptable option. Although there are schemes to handle
long-range forces, simulations of ionic systems are challenging. The standard way
of handling long-range electrostatics is the use of Ewald sums.
Simulations are often carried out in microcanonical ( NVE ), canonical ( NVT )
and constant number of particles-constant pressure-constant temperature ( NPT )
ensembles. The grand-canonical ensemble with constant free energy, volume and
temperature ( µVT ) is also used in some types of simulation.

40
3 Computational Chemistry

When carrying out simulations the statistical sampling is always a concern. The
question of whether the system has sampled sufficiently the full set of
configurations it can take on (the phase space) must be addressed. Verifying this is
difficult, and is not only an issue of doing a long enough simulation. If one can
imagine the potential energy of phase space as a kind of landscape, one can think
of the simulation as risking becoming trapped in a valley. As simulations are based
on statistical averaging of properties, there is always some degree of statistical
uncertainty in the results obtained.
While MD and MC can be used to calculate many of the same properties, they
differ in some important respects. In MC simulations time is not a parameter and
there is no easy way to obtain time-dependent properties such as diffusion-rates.
MD has been developed further in handling of long-range electrostatic interactions.
MC calculations can sometimes provide more efficient sampling of phase space.

3.4.2 Free Energy Perturbations


The free energy of solvation is the energy associated with a molecule going from
the gas phase to solution and is a central concept in understanding chemistry in
solution. Determination of free energies is also one of the main issues in this thesis.
Kollman (1993) provided a review of the main simulation techniques for
determining free energies. The main techniques are free energy perturbations
(FEP), thermodynamic integration and slow growth. FEP sees the widest use and is
also the technique I have adopted in my work. The fundamental equation that free
energy perturbations are based on is the following (Kollman 1993):
GB − GA = ∆G = −RT ln e−∆H / RT (3.6)
A

where ∆H = H B − H A and A
refers to an ensemble average over a system

represented with the Hamiltonian H A . If the systems differ in a significant way the

41
3 Computational Chemistry

equations will however not lead to a meaningful result. This problem can be
overcome by performing multiple simulations over intermediate steps between A
and B. A coupling parameter ( λ ) is introduced that allows the smooth conversion
of system A to B. The mutation of any geometry or potential parameter of the
system can then be represented in the following form (Jorgensen and Ravimohan
1985):
ξ (λ ) = ξ0 + λ (ξ1 − ξ0 ) (3.7)

The total free-energy change is obtained by adding together the contributions from
each single perturbation. In the present work calculations are performed with the
double-wide sampling scheme (Jorgensen and Ravimohan 1985). In this scheme
the free energy difference for λi → λi+1 and λi → λi−1 are evaluated in a single
ensemble.
The mutation will usually be between two different solute molecules. The essence
of the FEP is to mutate one molecule into another and compute the energy
associated with the transformation. FEPs are in general more accurate for
calculating differences between molecules with similar properties. More accurate
results are for example obtained when perturbating between molecules of the same
charge, as opposed to between species with different charges. Calculating the
absolute free energy of solvation is more difficult than simply calculating relative
free energies, the two systems one is calculating the energy between in this case
being very different (differing by the number of molecules). In this work most
calculations are done as relative free energy perturbations.
While the general concept of FEP calculations is easily understood, the technical
issues involved are far from trivial. Kofke and Cummings (1997, 1998) have
concluded that perturbations that involve growth are superior to those involving
shrinking or deletion of molecules. They conclude that shrinking offers loss of
accuracy due to biases in the sampling. More recently Kofke (2005) has however
concluded that the error involved in insertion and deletion approaches may vary

42
3 Computational Chemistry

from system to system. The development of optimal FEP methods does therefore
appear to be a work in progress.
The selection of methods in this work has been dictated mainly by available
simulation codes, rather than any assessments of the merits of various methods to
determine the free energy.

3.5 QM/MM
QM/MM is a general term for calculations combining quantum mechanical (QM)
and molecular mechanics (MM) representations. Such calculations can be
advantageous when one wishes to model parts of a system with greater accuracy.
For studies of a reaction in solution one might for example represent the reacting
molecules at a QM level, while solvent molecules are represented at MM level. In a
more sophisticated calculation one might also represent the closest solvent
molecules at QM level, while solvent molecules at a greater distance are
represented at a MM level. While QM/MM calculations are less time consuming
than pure QM calculations, they can still be prohibitively expensive.
Gao (1996) has reviewed the various types and levels of QM/MM coupling. In
the present work QM geometries and atomic charges will be used in simulations.
This can be regarded as a very weak form of QM/MM coupling.

3.6 Computational Chemistry and Experiment


Computational chemistry will never be a full replacement for doing experiments,
but can often supplement experimental work. Very often neither experiment nor
computational chemistry can by itself give us the full insight we could desire. In
many cases one must therefore piece together whatever information can be drawn
from either source, to draw whatever conclusions can be drawn. Sometimes
computational chemistry can be used to calculate properties that are not at all
available from experimental work, while some issues can be difficult both in the

43
3 Computational Chemistry

laboratory and on the computer. When working with results from modeling and the
laboratory one should have a feeling for the quality of the data from different
sources. Computational chemistry has grown as a field rather quickly, and many
researchers are perhaps not fully aware of the potential of its methods. On the other
hand I have the impression that researchers sometimes are more aware of the
limitations of the tools they work with, than that of other methods, and may come
to overestimate the quality of the work in a different field.
In the field of CO2 absorption there is as noted in the last chapter considerable
amount of experimental data. On the other hand it must also be noted that the
systems in the process display considerable complexity. The systems consist of
many components in an aqueous solution. Several of the components are ionic and
many components can have large numbers of potential conformers. In addition the
process runs at temperatures ranging from 50ºC to 130ºC. There are also
degradation products, impurities and surface effects that may change the chemistry,
or affect the process operation in some other way. Given the complexity of the
system, the amount of experimental data available must actually be considered to
be rather sparse. The multi-component nature of the liquid in the system also
present challenges for computational chemistry work. The absence of experimental
data at the molecular level also means that it can be difficult to validate models
being used and conclusions drawn regarding the nature of the system.
The approach in this work has therefore been somewhat pragmatic. The most
effort has been put into areas where computational chemistry was thought to give
the most valuable new insight, where as some aspects have been approached
tentatively.

44
3 Computational Chemistry

3.7 Review of Computational Chemistry Work on CO2 Absorption


There has been published some computational chemistry work on chemistry
directly related to the CO2 absorption process. Ohno and co-workers (Ohno et al.
1998 and Ohno et al. 1999) have done some very detailed work on 2-(N,N-
Dimethylamino)ethanol and 2-(N-Methylamino)ethanol. These papers combine
quantum mechanical calculations with infrared and Raman spectroscopy. These
papers offers valuable results in terms of the adopted conformers of amines and
CO2 bound amine molecules.
Papers by Chakraborty et al. (1988) and Jamroz et al. (1997) deal with
interactions between CO2 and amine molecules. While interesting these papers
provide no clear conclusions of direct relevance to CO2 absorption. Suda et al.
(1998) attempt to correlate the amount of CO2 absorbed in a liquid with frontier
orbital properties. While this is an interesting approach the correlations obtained
were not very good. In summary may be concluded that very little computational
chemistry work has been done for the specific purpose of understanding the CO2-
absorption process.
There has been done some computational chemistry work on the liquid structure
of ethanolamine (Button et al. 1996, Alejandro et al. 2000 and Gubskaya and
Kusilik 2004a) and ethanolamine in aqueous solution (Gubskaya and Kusilik
2004b). These papers do give some insight into the liquid structure of
ethanolamine, but they do not deal with the more complex multi-component
systems encountered in CO2 absorption. For other amine solvents of interest in CO2
capture I am not aware of any computational chemistry work having been done.
While there has been done very little computational chemistry work on CO2
absorption, there has been done a lot of work on molecules similar to those utilized
in this process. Molecules with amino- and hydroxyl-groups are important in all
forms of living organism, and the modeling of them is subject of substantial work

45
3 Computational Chemistry

in biochemistry. Most of the models applied in this thesis originated in studies of


systems of importance in biochemistry.

46
4 Modeling of Solvation Energy

4 Modeling of Solvation Energy


Realistic relative gas-phase energies of ionization may soon be estimated by
molecular quantum mechanics. If an accurate set of solution energies can be
determined, we can expect that the next generation of calculations may include
solvation energies, and thus complete quantum mechanical calculations for
protonation reactions in solution might become possible.
Jones and Arnett (1974)

4.1 Introduction
This chapter is intended as a review of available approaches to modeling energies
in solution. It will be mainly focused on issues directly relevant to CO2 absorption,
but will also touch upon some general issues. In addition to briefly describing
different solvation models, results will also be presented comparing different
models ability to predict relative base strength.
Liquids do in general probably represent the most difficult phase to model. In the
gas phase intermolecular interactions are limited and it is often sufficient to look at
the characteristics of a single molecule. In such a case quantum mechanical
calculations can be applied successfully, examples of this will be shown later in
this chapter. Solids are more complex, but the modeling of them is in some cases
made easier by their periodic nature.
In a liquid a single molecule interacts with a large number of neighbor molecules
that do not display any orderly structure. The interactions are also shifting
continuously as molecules move around, and the observed properties of liquids
represent averages of these interactions. Rigorous quantum mechanical calculations
of all these interactions are simply not feasible at present or in the foreseeable
future. This leaves scientists with the challenge of finding the best approximations
in simplifying a complex problem.

47
4 Modeling of Solvation Energy

4.2 The liquid state


While liquids do by nature not display any long term periodicity or regular
structure, they do have some structural characteristics. One common way to
describe such structural features is by use of radial distribution functions ( g (r ) ).

The radial distribution describes the probability of encountering two atoms at given
distances ( r12 ). Usually it is given in a normalized form so that the probability
density of 1 is the average density of the system. A radial distribution function will
often look something like the one shown in Figure 4.1.

Figure 4.1 Radial distribution function.

Atoms do not overlap and at very short range the probability of finding a second
atom will be zero. At slightly larger distances there is the ideal distance for direct
interaction between the atoms and this usually results a peak in the function, atoms
at this distance form what is called the first solvation shell. Because other atoms do
not overlap with the ones in the first solvation shell the radial distribution function

48
4 Modeling of Solvation Energy

will usually show a marked drop after the first peak. At longer distances the radial
distribution will average out to the average density of the system, which in other
words means that positions of atoms at long range are not correlated. With radial
distributions for all types of atomic sites in a liquid one can form a picture of the
interactions. In addition to the radial distribution functions there are other
properties that can be used to summarize the characteristics of liquids. An example
of such a property is the angular pair correlation function, this carries information
about how molecules in a liquid orientate relative to each other (Gray and Gubbins
1984). Another example is spatial distribution functions, these provide three
dimensional pictures of interactions in a liquid, the work by Gubskaya and Kusilik
(2004a) is an example of the use of such distribution functions.

4.3 Statistical Mechanics


Statistical mechanics offers tools for analytically describing interactions in liquids.
Here a brief outline will be given based on a compendium by Kjellander (1992).
One central parameter is the pair correlation function:

h(r1 , r2 ) = g (r1 , r2 ) −1 (4.1)

where g is the radial distribution function for atoms (or other form of site) located
at r1 and r2 . At the heart of much of the statistical mechanics works on liquids is
the Ornstein-Zernike equation. The Ornstein-Zernike equation states that the pair
correlation function can be written in the following form:

h (r1 , r2 ) = c (r1 , r2 ) + ∫ c (r1 , r3 )n (r3 ) h (r3 , r2 ) dr3 (4.2)

Here c (r1 , r3 ) is the direct correlation function and n (r3 ) is the density at a given

point. What this equation states is that the correlation between two particles in a
liquid ( h(r1 , r2 ) ) can be divided into two parts, one direct and one indirect. The
direct part is called the direct correlation function. The indirect part comes from

49
4 Modeling of Solvation Energy

particle 2 again correlating with the other particles in the system, these particles
again correlating directly with particle 1.
The h(r3 , r2 ) in the integral can be written in the same form as the Ornstein-
Zernike equation:

h (r3 , r2 ) = c (r3 , r2 ) + ∫ c (r3 , r4 )n (r4 ) h (r4 , r2 ) dr4 (4.3)

Inserting this into the Ornstein-Zernike equation the following equation is


obtained:

h (r1 , r2 ) = c (r1 , r2 ) + ∫ c (r1 , r3 )n (r3 ) c (r3 , r2 ) dr3 + ∫ c (r3 , r4 )n (r4 ) h (r4 , r2 ) dr4 dr3 (4.4)

This procedure can be repeated and the result is clearly an infinite sum of integral
equations. The Ornstein-Zernike equation is by itself the definition of the direct
correlation function and it should be noted that no approximations are required in
formulating it. This function is expected to have a simpler form than the pair
correlation function. If one combines the Ornstein-Zernike equation with a closure
equation one can obtain a solvable set of equations. One of the simpler closure
equations is the mean-spherical approximation:
h(r ) = −1 for r ≤ σ (4.5)
U
c(r ) = − for r > σ
kT
where U is the interatomic potential energy and σ is the radius of the atom.
A widely used closure is the Percus-Yevick equation:
g (r ) ≈ e
−u(r ) / kT
(1 + h (r )− c (r )) (4.6)

A second widely used closure equation is the so-called Hypernetted Chain(HNC)


approximation:

g (r ) ≈ e−u(r ) / kT +h(r )−c(r ) (4.7)

50
4 Modeling of Solvation Energy

These closure equations represent approximations and the solutions obtained with
these equations are therefore approximate. For simple idealized liquids such as
hard spheres or soft spheres these equations can be solved and results can be
compared with molecular simulations utilizing the same molecular representation.
Such comparisons have revealed that most of the closure equations produce
satisfactory results in some cases but fail in others. The Percus-Yevick closure is
known to work well in describing interactions between hard spheres, but performs
less well for systems with electrostatic interactions. The HNC closure performs
well for a number of different types of systems, but can fail for some mixtures of
different solvent molecules and can sometimes lead to diverging calculations
(Hirata 2003). Efforts have been made to develop closure equations that produce
results in better general agreement with those obtained from simulations. Hansen
and McDonald (1990) provide a brief review of such efforts. A more recent
example is the KH closure suggested by Hirata and co-workers (Hirata 2003).
The Ornstein-Zernike equation is for independent single-sites (atoms) and the
extension to molecules consisting of several-sites is not trivial. One Ornstein-
Zernike based model that is easily extended to molecules is the reference
interaction site model (RISM). This model was first derived by Chandler and
Anderson (1972). It can be written as:

h = ω * c* ω + ρω * c* h (4.8)
Here “ * ” represents convolutions and c and h are in this case matrixes of the
various correlation functions for the different sites in molecules. ω is a matrix of
intramolecular correlations. Hansen and McDonald (1990) refer to this theory as
the “RISM approximation” and unlike the Ornstein-Zernike equation itself
approximations have been made in deriving it. Hansen and McDonald (1990)
summarize some of the failures of RISM and look at efforts to come up with more
accurate models. They do observe that while there have appeared models that in a
formal sense would appear more correct than RISM, they have not yielded better

51
4 Modeling of Solvation Energy

results. Work by Lue and Blankschtein (1995) provides an example of more recent
efforts to develop integral equation theories. They present results for water for the
site-site Ornstein-Zernike equation and the Chandler-Silbey-Ladanyi equations. It
should also be noted that efforts have been made to improve or correct the RISM
formulation (Hirata 2003 and Kvamme 2002). Development of such integral
equation theories is clearly a difficult issue, the mathematics of formulating the
equations is in itself demanding and so is the issue of solving these equations for a
given system.
It should be emphasized that these integral equation models do not say anything
about the nature of molecular interactions, rather they can be seen as an alternative
to simulations that when coupled with a molecular representation offers a model of
the liquid state. RISM as a model for solvation will be discussed in section 4.4 5.

4.4 Models to Calculate the Free Energy of Solution

4.4.1 Introduction

The free energy of the solvation ( ∆Gs0 ) is the free energy change associated with a
molecule leaving the gas phase and entering a condensed phase. It is the liquid
property that will be of greatest interest in the present work. It may for a given
species A be determined from equilibrium data by the following equation (Cramer
2002):

⎪⎧⎪ [ A] ⎪⎫⎪
∆G ( A) = lim ⎪⎨−RT ln sol
0 ⎪⎬ (4.9)
s
[ A]sol →0 ⎪
⎪⎪ [ A]gas ⎪⎪
⎩ eq ⎭ ⎪
This formulation draws on work by Ben-Naim (Ben-Naim and Marcus 1984 and
Ben-Naim 1992). While the equation in the form it is stated here is for infinite
dilution, it can also be formulated for other conditions. While the free energy of
solvation is perhaps not a quantity that most people in chemical engineering are

52
4 Modeling of Solvation Energy

familiar with, it does contain information relevant to several aspects of liquids


equilibrium that in chemical engineering is represented in other forms. If the
energy for the reaction in the gas phase and the solvation energy for each
component is known the reaction energy in solution can be calculated. The free
energy of solution also conveys information about the vapor-pressure of a given
species (Winget et al. 2000) and solvation energies can also be converted into
activity coefficient data (da Silva 2004).
The present review is intended to cover all the main approaches to calculating
solvation energies. The form of the present outline draws mainly on reviews by
Bacskay and Reimers (1998) and Orozco and Luque (2000). While I have tried to
make this review fairly broad the main focus will be on models that may be
suitable for modeling CO2 absorption. As noted in Chapter 2 the CO2 capture
processes is usually run with base molecules in water and most attention will also
be given to the modeling of aqueous solution. Water is also, for obvious reasons,
the solvent that has received the greatest attention in the literature.

4.4.2 Equation of State and Lattice Models


The interest of chemists in understanding, modeling and predicting the behavior of
liquids predates computational chemistry and there are models not based on any
form of computational chemistry. The present outline of such models is drawn
from the textbook “Molecular Thermodynamics of Fluid-Phase Equilibria
(Prausnitz, Lichtentaler and Gomes de Azevedo 1999). The authors distinguish
between two main approaches. One is based on approaching a liquid with a gas
model and the second on approaching it with a solid-like lattice model.
The first approach leads us to the equation of state models. From an equation of
state the free energy (Helmholtz or Gibbs) can be determined and from an equation
of state for mixtures the free energy of each component can be derived.

53
4 Modeling of Solvation Energy

The lattice models are based on thinking of molecules in liquids as occupying


spaces in a grid. Mixing of two pure components can be envisioned as molecules
exchanging places in the grid. If interaction parameters are introduced the energy
and entropy change associated with mixing can be calculated.
Considering a liquid as a dense gas or a rigid lattice is a fairly drastic
approximation and the first such models were not all that successful. More
advanced models have however been developed that successfully capture the
behavior of many types of solvents. Such models are however in general not able to
predict properties for systems where no experimental data is available. The
parameters in the equations are fitted to reproduce experimental data, and the
parameters together with the equations can perhaps be thought of as a way to
summarize experimental data for a given system. What such models can do is
predict the behavior for a system over changing concentrations, temperature and
pressure. Examples of such applications on CO2 capture related issues can be found
in the work of Solbraa (2002) and Hoff (2003).
There is however a family of models derived from equation of state and lattice
models to do predictive calculations. These are group contribution models. The
most famous of these is perhaps the UNIFAC model (Fredenslund, Gmehling and
Rasmussen 1977). These are based on breaking up a molecule into units such as
hydroxyl-, alkane- and amino-groups. The properties of the solvent are attributed to
its various groups. Some form of database is constructed of groups and their
parameters, when a new molecule is encountered the groups it is composed of is
identified and from these the properties are predicted. While such models see fairly
wide use they do have clear limitations.
In Figure 4.2 is shown two possible conformers of ethanolamine. In one of the
conformers there is an intramolecular hydrogen bond, if such a bond is formed in
solution then this will reduce the extent of bonding formed with other species. The
interactions with other species will therefore depend on which conformer form is

54
4 Modeling of Solvation Energy

preferred. Yet present group contribution methods do not take account of such
conformer effects and their reliability in cases where such issues arise would seem
questionable, a conclusion also drawn by Wu and Sandler (1991).

Figure 4.2 Conformers of ethanolamine.

Another, and perhaps more fundamental, issue is if the parameters in a model such
as UNIFAC are sufficient to capture effects of all interactions in a liquid. UNIFAC
is in form similar to traditional equations of state models such as UNIQUAC and
NRTL. These are all sets of quite simple equations, energy interactions between
different species are usually represented by a single parameter. It would seem
questionable if such a small number of parameters can accurately capture the
complex interactions in a liquid. In Figure 4.3 three pairs of water molecules are
shown. The interaction energy is obviously different for the different pairs, the
energy depending on distance and orientation of the molecules. A single parameter
can perhaps express the average of all interactions in a given system, but if the
composition of a system changes the average interaction for a given component
might also change.

55
4 Modeling of Solvation Energy

Figure 4.3 Interactions of water molecules.

More advanced models have been proposed in recent years, one notable model
being the Statistical Associated-Fluid Theory (SAFT) (Chapman et al. 1989 and
Chapman et al. 1990). This model has expressions to take account of contributions
from short-range repulsion, long-range dispersion, chemical bonding and
association or hydrogen bonding between different molecules. It is also noteworthy
that the model was tested for consistency with molecular simulation during
development.
While advanced models such as SAFT seem better formulated to capture the
various effects that play a part in solution the number of parameters in the model is
also inevitably greater than for the simpler models such as UNIQUAC. In the
absence of more detailed experimental data the determination of parameters for a
more advanced model can be a difficult task. While such advanced models do
represent progress and may in general be considered to be better than simpler
models, the underlying issues of parameter fitting must always be kept in mind.

4.4.3 Continuum Models


Continuum models are the group of solvation models in computational chemistry
that have seen the greatest use. These models have a long history, some of the early
work was carried out by the Norwegian Lars Onsager (Onsager 1936). Rather then
explicitly representing solvent molecules, the solvent is represented by a

56
4 Modeling of Solvation Energy

continuous electric field that represents a statistical average over all solvent degrees
of freedom. Such models are also known as implicit solvation models. The outline
given here will be based on the textbook “Essentials of Computational Chemistry”
(Cramer 2002). Continuum models have been the subject of numerous reviews in
recent years (Tomasi and Persico 1994, Cramer and Truhlar 1999, Orozco and
Luque 2000) and I will to some extent draw on these too in the present outline.
At the heart of the continuum models is the Poisson equation. This is valid when
a surrounding dielectric responds linearly to embedding charges:
4πρ (r )
∇2φ (r ) = − (4.10)
ε
where ρ is the charge density, φ is the electrostatic potential and ε is the dielectric
constant. The Poisson equation is valid under conditions of zero ionic strength, for
charged species the Possion-Boltzmann equation applies:

k BT ⎡ qφ (r )⎤
∇ε (r )⋅∇φ (r ) − ε (r )λ (r )κ 2 sinh ⎢ ⎥ = −4πρ (r ) (4.11)
q ⎢ kT ⎥
⎣ B ⎦
where λ is a function that switches from zero in areas not accessible to the
electrolyte to one in areas that are accessible and q is the magnitude of the charge.

κ 2 is the Debye-Huckel parameter:

8πq 2 I
κ2 = (4.12)
εk B T
where I is the ionic strength. The work needed to create a charge distribution can
be determined from the following equation.

1
ρ (r )φ (r ) dr
2∫
G =− (4.13)

In continuum models the solute is placed in a cavity representing the space


occupied by the solute in the solvent. Such a procedure is illustrated in Figure 4.4.

57
4 Modeling of Solvation Energy

Figure 4.4 Creation of Cavity and Insertion of Solute in Solvent.

In the first continuum models cavities were simple spheres, while in most present
models they are usually built as a set of spheres around each atomic centre.
Continuum model calculations can be carried out with either a molecular
mechanical or quantum mechanical representation of the solute, the latter does
however represent a more rigors model. In quantum mechanical continuum
calculations an iterative cycle is often used. First the solute with its gas phase
electron distribution is inserted into the cavity. The solvent electrostatic field that
arises from the solute charges is calculated, this is usually called the reaction field.
The reaction field is then introduced as an external potential in the quantum
mechanical calculation. This is done by solving the following form of the
Schrödinger equation:
⎛ ⎞
⎜⎜ H − 1 V ⎟⎟ ψ = E ψ (4.14)
⎜⎝ 2 ⎠⎟
where V is the reaction field inside the cavity. The reaction field is then
recalculated based on the new solute charge distribution. This is repeated in an
iterative procedure until the energy of the system converges. The energy change
obtained from such a calculation is usually called the electrostatic energy. There
are however other contributions to the solvation energy.
The free energy of solvation can be thought of as consisting of three main
contributions (Orozco and Luque 2000):

58
4 Modeling of Solvation Energy

∆Gs = ∆Gelectrostatic + ∆Gcavitation + ∆Gvan der Waals (4.15)

The electrostatic contribution is the term just described. The cavitation term
represents the energy that is needed to create room for the solute in the solvent.
Finally the van der Waals term is the energy that comes from dispersion and
repulsion interactions between solute and solvent molecules. In addition it should
be noted that if the solute causes rearrangement in the solvent structure that can
also give a contribution to the solvation energy (Cramer and Truhlar 1999). The
continuum models are in essence electrostatic models and estimates for the other
contributions must be sought elsewhere. The cavitation and van der Waals terms
are usually included as semi-empirical terms which size is made proportional to the
size of the solute cavity.
It should be clear from this outline that determining the size and shape of cavities
is a central issue in developing and applying continuum models. Cossi et al. (1996)
describe the selection of cavities as “one of the most delicate steps in defining a
continuum solvation model”. While cavities can be easily understood as the space
occupied by a solute molecule in a solvent this is not an unambiguously defined
property, nor is it experimentally observable. Many definitions have been
proposed, one of the more common is the van der Waals surface, defined as the
cavity formed by van der Waals spheres centered on the atoms (Cossi et al. 1996).
There is also no requirement that the same definition of the cavity be used in the
calculation of electrostatic and other terms, in the PCM model (to which I will
return) different definitions are in fact utilized (Cossi et al. 1996).
As noted some of the energy contributions in continuum models are usually
determined in some form of fitting to experimental data and there is also some
arbitrariness in the selection of cavities. Continuum models are often directly fitted
to free energies of solvation. This fitting process does mean that these models tend

59
4 Modeling of Solvation Energy

to be semi-empirical in nature. One consequence is that the models tend to be


limited to the solvent and conditions for which they were parameterized. It also
means that the validation of different models can be somewhat difficult. The
empirical contributions may mask errors in the electrostatic model, this means that
the underlying quality of the model is not that easily determined from comparisons
with experimental data. This can perhaps create challenges in the further
development of continuum models.
Continuum models are in general reasonably good at reproducing the free
energies of neutral solutes, but less accurate for ions. The nature of the models does
raise some questions about how reliable they can be. One of the issues often
debated is to what extent they can capture effects of interactions in the first
solvation shell such as hydrogen bonding (Cramer and Truhlar 1999 and Kawata et
al. 1996).
In a continuum model a solvent is represented by a fairly small number of
parameters, among them the dielectric constant. With such a form of solvent
representation it is unclear if and how such models can be extended to mixtures of
solvents. Many continuum models do also tend to be dedicated to calculations at
infinite dilution.
A great number of variations of continuum models have been proposed, here I
will briefly review some of the more common models.
PCM The Polar Continuum Models (Miertus, Scrocco and Tomasi 1981, Cossi,
Barone, Cammi and Tomasi 1996 and Cances, Mennucci and Tomasi 1997) are the
continuum models that see the greatest use in the context of quantum mechanical
calculations. One of the main reasons for this is that the models are implemented in
the widely used Gaussian programs (Frisch et al. 1998). These models are also
regarded as being fairly robust and flexible in the sense that they can be easily
applied to different molecules and systems. In the PCM the cavity surface is
divided into surface areas called tessarae.

60
4 Modeling of Solvation Energy

The reaction field of the solvent is represented by charges placed on the cavity
surface. PCM calculations can be carried out on in combination with most forms of
quantum mechanical calculations.
SM The “Solvent Model” are a series of models developed by Cramer, Truhlar
and co-workers (Chambers et al. 1996 Li et al. 1999). The philosophy behind these
models is somewhat different from that behind the PCM models. In the
development of SM models a number of different forms have been tested. The
developers, rather than focusing on developing a single form of rigorous model,
have implemented various models. They have studied how different models, when
properly parameterized, are able to reproduce the experimental free energies. A
particular feature is the use of atomic surface tensions ( σk ) (Li et al. 1999). The
free energy is in this model the sum of electrostatic and surface tension
contributions. These atomic surface tensions are purely empirical terms intended to
capture cavitation, van der Waals contributions and any other contributions arising
from the first solvation shell.
The SM models are in general based on simpler electrostatic calculations than the
PCM models and offers results of comparable quality from calculations that are
much less time consuming. They do however rely on more extensive use of
empirical data.
COSMO-RS The COSMO model was originally only one type of
implementation of a continuum model (Klamt 1995 and Andzelm, Kölmer and
Klamt 1995). In the COSMO calculations the solvent was considered to be an
electric conductor with an infinite dielectric constant, this having the virtue of
facilitating the calculations. COSMO-RS (“Realistic Solvent”) represents the
development by Klamt and co-workers (Eckart and Klamt 2002 ) of a model
designed to study properties both for pure components and liquid mixtures. The
model can therefore be used for a number of problems of interest in chemical
engineering. While results published with the COSMO-RS model seem

61
4 Modeling of Solvation Energy

encouraging (Eckart and Klamt 2002) it must be noted that this is a proprietary
model and that there might therefore perhaps be less transparency regarding its
strength and weaknesses than is the case for other models. All the publications on
this model have to my knowledge been by the developers themselves.

4.4.4 Molecular Simulation


In the last chapter MC and MD simulations were briefly introduced. In such
calculations the solvent molecules are explicitly represented and such models are
also called “explicit models” (while the continuum models are implicit). As noted
in the last chapter simulations are time consuming, the simulation time increasing
with the number of molecules being studied and the level of molecular
representation. Today most simulations are done with molecular mechanics
representation of solvent and solute. In both MD and MC there are techniques to
obtain the free energy of the solvation, one of the most common methods being
free energy perturbations.
As noted in the last chapter a key approximation made when using a molecular
mechanics representation is that charges are fixed. Polarization is handled
implicitly by using values for the charges intended to reflect the average
polarization in the liquid. Compared to a continuum model the polarization effects
are handled in a simpler manner in standard simulations.
Simulations are not packaged as models in the same way continuum models are.
Most simulation codes give the user great freedom in selecting the force field
parameters.
The parameters for many force fields are derived from fitting to empirical data.
The parameterization is usually based on reproducing liquid properties, and is thus
very different in nature from that used for the continuum models.

62
4 Modeling of Solvation Energy

As indicated in the last chapter there is also many ways in which quantum
mechanical calculations can be used to determine force field parameters. The most
important use is in determination of molecular geometry and atomic charges.
As noted in the last chapter there are a number of ways in which partial atomic
charges can be derived from quantum mechanical calculations. What kind of
schemes will lead to more accurate solvation energies is still an open question.
In the last chapter it was also observed that polarizable force fields are available.
While such simulations are still too time-consuming to be done on a routine basis it
is clearly an option to consider if a more realistic solvent representation is desired.
With simulations the study of liquids composed of two or more components is in
principle no problem. As a system becomes less homogeneous it would however
seem very likely that the sampling required for satisfactory average values will
increase. Simulations can also quite readily be used to study systems with different
temperature. Pressure conditions can also be controlled, although this is still
perhaps somewhat more difficult than temperature control (Allen and Tildesley
1987).

4.4.5 RISM and RISM-SCF


In the statistical mechanics section of this chapter the RISM equations were
introduced. RISM can be regarded as an alternative to simulations, that when
combined with a molecular representation offers a solvent model from which the
free energy of solution can be calculated. Calculations where the solute and solvent
have a molecular mechanics representation is called a classical RISM calculation.
RISM-SCF (Hirata 2003 and references therein) combines a classical solvent
representation with a quantum mechanical solute representation. The calculation
procedure is analogue to the one described for the continuum models. From a
quantum mechanical calculation in the gas phase partial atomic charges are

63
4 Modeling of Solvation Energy

calculated and a RISM calculation is carried out to determine the solvent structure
around the solute. From the solvent structure the reaction field is calculated and a
new quantum mechanical calculation is carried out to determine the atomic charges
in the presence of the field. This is repeated until convergence.
An RISM-SCF calculation has the virtue of combining the explicit solvent
representation of a classical simulation with the solute polarisation term found in
the continuum models. As in a standard simulation polarization of solvent
molecules is only included implicitly in present forms of RISM-SCF. As noted
previously the RISM equations involve some approximations, and compared to
simulations this adds a layer of uncertainty. The issues regarding force fields and
atomic charges are the same in RISM calculations as in simulations. In present
implementations of RISM-SCF (Kawata et al. 1996) the solute charges are
determined by fitting the charges to the electrostatic potential.
The complexity of the underlying equations and problems that can appear in
obtaining converged solutions have probably contributed to this type of model
seeing less use than continuum models and simulations.

4.4.6 Supermolecule Approach


The supermolecule approach is a term used for calculating the entire system
quantum mechanically. Traditionally this has been done by performing calculations
with some few solvent molecules in vacuum. While such an approach can be useful
in determining the direct effect of solvent molecules on the solute it is more
difficult to use such an approach for systematic calculations of solvation energy.
Solute molecules can vary in shape and size and the number of solvent molecules
that is needed to represent the main interactions will vary. As noted in the first part
of this chapter the observed properties of a liquid represent averages over large
numbers of configurations, it is clear that a calculation based on a single geometry
will often not reflect such averages.

64
4 Modeling of Solvation Energy

A more advanced form of supermolecule calculations are QM/MD (Quantum


Mechanical Molecular Dynamics) and QM/MC (Quantum Mechanical Monte
Carlo) calculations. In such calculations simulations are carried out on molecules
represented at a quantum mechanical level. Such calculations are seeing increasing
use. The most famous, but far from the only, such approach is the Car-Parrinello
Molecular Dynamics scheme (Car and Parrinello 1985). Such calculations are
obviously very time-consuming and I am not aware of any such model being used
to calculate free energies of solution. Free energies from such simulations can
perhaps be determined by a scheme similar to what has been suggested for
QM/MM calculations (Cummins and Gready 1997).

4.4.7 Hybrids of Computational Chemistry approaches


Looking at a radial distribution function it is clear that the short-range and long-
range interactions are different in nature. At short range the interactions take on the
nature of weak bonding or repulsion, while at long range the solute molecule only
“sees” the averaged effect of a large number of solvent molecules. This would
suggest that one is perhaps best served by using a model that has a better solvent
representation at short range than at long range.
One such approach is to combine the supermolecule approach with continuum
models. A small number of explicit solvent molecules can be included inside the
cavity. While such calculations see some use there is an issue of how many solvent
molecules to include, and how to do it in a consistent way for molecules of
different size and shape (Cramer and Truhlar 1999).
Another form of hybridization is QM/MM models introduced in the last chapter.
In such models part of the system is described with a molecular mechanics
representation and other parts have a quantum mechanical representation. The
simplest form would involve representing the solute quantum mechanically and
using a MM representation of the solvent. In more advanced calculations some of

65
4 Modeling of Solvation Energy

the solvent molecules closest to the solute can also be represented at a quantum
mechanical level. Calculations are carried out using some kind of simulation. To
obtain sufficient statistical sampling a simulation must usually be carried out for
many thousand steps, if one needed a QM calculation for each step and each QM
calculation took something like an hour, it is easy to see that such simulations can
become very time-consuming. One approach taken to reduce the time consumption
is to use very low level semi-empirical QM models (Kaminski and Jorgensen 1998
and Cummins and Gready 1997). Such an approach does on the other hand also
reduce the quality of the results obtained. Another interesting proposal is to use
MM simulations to calculate geometries and do set of QM calculations on energies
only (Wood et al. 1999). This proposal does suggest that there is room for
refinement of QM/MM methods by only using the QM calculations for the most
vital part of the calculations.

4.4.8 Descriptor Models


One of the approaches often taken in science when confronted with predicting a
certain property is to correlate it to another known property. While this approach
has perhaps not been used that much to determine the free energy of solvation
itself, it has often been used in estimating different equilibrium constants in
solution. An example of this can be found in the work of Eimer (1994) that finds a
correlation between an equilibrium constant and molecular weight, and uses this to
estimate the value for molecules where the equilibrium constant is not known. Such
applications involve a form of implicit estimation of solvation energies.
Approaches based on correlations with experimentally observable properties have
been in use for a long time and do not require any input from computational
chemistry.
Computational chemistry methods can however be used to determine a large
number of molecular characteristics that can otherwise not be obtained. This offers

66
4 Modeling of Solvation Energy

new sets of descriptors that can be applied in finding correlations with the solvation
energy or equilibrium constants. A motivation for using such an approach is that
computational chemistry methods may calculate some property with greater
accuracy than the free energy itself can be calculated. One descriptor that can be
determined from simulations and that is used in finding correlations with solvation
energies is the solvent-accessible surface area (SASA) of a molecule, this can be
thought of as a form of “efficient surface area”. Duffy and Jorgensen (2000)
obtained encouraging results when looking at correlations between SASA, other
descriptors related to extent of hydrogen bonding and solvation energies. Another
model based on SASA has been developed by Kollman and coworkers (Wang et al.
2001). It can also be noted that some forms of the SM continuum models are
similar in nature to descriptor models.

4.4.9 Other Models


The outline and classification of models so far covers the most widely applied
approaches to determining solvation energies. There are however some methods
that do not fall within any of these categories. The most notable is perhaps the
Langevin-Dipole method (Warshel and Levitt 1976). The solvent is in this model
represented as a grid of dipoles. This method lies somewhere between a continuum
model and an explicit solvent model.
There are also a number of ways in which one can imagine combining different
approaches to calculating solvation energies. New variations do appear from time
to time in the literature, and combining aspects of different models is one way
forward to obtaining models that produce more accurate results within a given
timeframe.

67
4 Modeling of Solvation Energy

4.4.10 Hybridization of Gibbs Energy Models and Computational Chemistry


Based Models
It is likely to take some time before computational chemistry methods can provide
robust a priori quantitative predictions of solvation energy for multi-component
systems. An option that might be worth exploring is to pursue a general semi-
empirical approach, drawing on whatever information is available experimentally
and using computational chemistry to fill in the gaps.
One way to do this would be to use methods in computational chemistry to
determine parameters for a lattice or equation of state model while at the same
maintaining the models reproduction of experimental observations.
Some early efforts along these lines are work by Jonsdottir et al. (1996) and Sum
and Sandler (1999). Both these works are based on determining the parameters for
the UNIQUAC model. The general problem with this work is that the UNIQUAC
parameters do not have any clear definition. Fischer (1983) warns against assigning
a physical meaning to these empirically determined parameters. Attempting to
predict ill-defined parameters is obviously a most unpromising line of research. I
comment in greater detail on the work by Sum and Sandler in a paper appearing as
a part of this thesis (da Silva 2004).
While UNIQUAC and similar equations are not suitable for combinations with
data from computational chemistry, more advanced models may be. The SAFT
equations would seem to be a promising candidate.
It would seem that there is a potential to combine information from
computational chemistry with some form of free energy model. This is however
likely to be a time-consuming task, requiring careful work and patience.

68
4 Modeling of Solvation Energy

4.5 Comparison of Methods to Calculate the Free Energy of Solution

4.5.1 Methods
In the last section a fairly broad review was made of models available to calculate
the free energy of solvation. In the present section a direct comparison will be
made between a series of models in the calculation of amine basicity. This
comparison should give an idea of how far computational chemistry has come in
meeting the expectations of Jones and Arnett (quoted in the beginning of this
chapter).
In choosing models for this comparison an effort has been made to include as
many as possible of what seems to be the most promising methods. Not all solvent
models are however distributed as ready to use software. Some only exist as codes
in the laboratories developing the model, and can only be obtained by cooperation
with the researchers behind the model. Of such models the RISM-SCF model has
been included in the present comparison. Two of the models chosen are among the
widely used continuum models. The final models are MC simulations with free
energy perturbation. As noted in the previous chapter, there are numerous ways to
calculate the solute atomic charges from QM calculations. In the present work two
somewhat different types of charges are tested.
The comparison is intended to assess the models on an equal basis. All models
have been run with settings that were expected to be close to optimal, in general
that has meant using types of calculations similar to the ones in the papers first
describing the models. No effort has however been made to tune any of the models
to be in better agreement with the experimental data in consideration. It should be
emphasized that the present comparison does not attempt to address the overall
potential or quality of different models.

69
4 Modeling of Solvation Energy

The RISM-SCF and simulation results appear in a paper in this thesis (da Silva,
Yamazaki and Hirata 2005). A detailed discussion of methods and results for these
calculations can be found in that paper.
PCM. In the present work the IEFPCM (Cances, Mennucci and Tomasi 1997)
model was used with its default settings in Gaussian 98 (Frisch et al. 1998) with 60
tesserae per atomic sphere. Calculations were carried out on HF/6-31G(d) gas
phase geometries (IEFPCM/HF 6-31G(d)//HF 6-31G(d)).
SM. The SM 5.42R model implemented in Gamesol (Xidos et al. 1993) has been
utilized. This model uses gas phase geometries and is parameterized for a series of
basis-sets. Calculations are carried out with HF/6-31G(d) gas phase geometries and
energies (SM 5.42R/HF 6-31G(d)//HF 6-31G(d)). Data in the Gamesol distribution
manual (Xidos et al. 1993) suggests that this level of calculation should be close to
optimal for the SM solvation model.
RISM-SCF. RISM-SCF calculations were carried out at the HF/6-31G(d,p) level.
The solute Lennard-Jones parameters are from the all-atom OPLS force field
(Jorgensen et al. 1996 and Rizzo and Jorgensen 1999) while the solvent was
represented with the TIP3P (Jorgensen et al. 1983) water model. Solute atomic
charges were determined by fitting to the electrostatic potential.
FEP-MK. A set of free energy perturbations were carried out with MK type
charges (described in the last chapter). The solute Lennard-Jones parameters are
from the all-atom OPLS force field (Jorgensen et al. 1996 and Rizzo and Jorgensen
1999) while the solvent in this case was represented with the TIP4P (Jorgensen et
al. 1983) water model. Both the solute and solvent representation in these free
energy perturbations were similar to what is utilized in the RISM-SCF calculation.
FEP-CM2. A second set of free energy perturbations were carried out with CM2
type charges (also described in the last chapter). The solute and solvent
representation was otherwise the same as for the FEP-MK simulations.

70
4 Modeling of Solvation Energy

4.5.2 The Basicity


The property that has been chosen for the present comparison is the relative
basicity of amines. Amines are the family of molecules that are of main interest in
the present work. The basicity is in itself one of the amine properties of greater
interest. The calculation of basicity requires the calculation of solvation energy of a
neutral molecule and it’s protonated (and ionic) form and the gas phase basicity.
This can therefore provide a fairly rigorous test of how good a solvation model is,
provided accurate gas phase basicities are available. In addition there is accurate
experimental data available with which to compare the model results. For base
reaction the following equilibrium can be set up:
aB a H +
Ka = (4.16 )
aBH +

The free energy of protonation in aqueous solution ( ∆G ps ) is related to Ka by the

following equation:

∆G ps = −2.303RT log K a (4.17)

The calculations are based on the following thermodynamic cycle.

Figure 4.5. Thermodynamic cycle.

71
4 Modeling of Solvation Energy

Quantum mechanical calculations are used to determine gas phase basicities, while
solvation models are used to determine solvation energies ( ∆Gs ). These results can
be added together to give the free energy in aqueous solution. In the present
comparison only the relative basicities will be studied. The quality of a model is
given by how close calculated relative basicities are to the relative experimental
values.

4.5.3 Amines
The amines included in the present study are shown in Figure 4.6. To reduce the
complexity of the comparison amines with uncertain conformer forms are not
included. This study does however include amine molecules that display significant
variations in geometry.

Figure 4.6 Amines in present study.

72
4 Modeling of Solvation Energy

4.5.4 Results
In Table 4.1 the calculated gas phase basicity data are shown. These can be seen to
be in good agreement with experimental data.

Table 4.1 Relative Gas Phase Protonation Energies. Data in [kcal/mol].


Molecule Theoreticala Experimentalb

Ammonia 0.0 0.0


Methylamine 10.2 10.9
Ethylamine 14.3 14.1
Dimethylamine 17.9 18.5
Trimethylamine 22.2 23.7
Piperidine 24.2 24.4
Piperazine 23.3 22.9
Morpholine 17.0 17.3
Pyrrolidine 23.6 23.0
2,2,6,6-Tetramethyl-4-piperidinol (TMP) 29.7
a
B3LYP/6-311++G(d,p) energy with thermal correction and ZPE calculated at HF/6-
31G(d) level.b Data from Hunter and Lias(2003).

In Table 4.2 and 4.3 the solvation energies determined with the different models
are shown. The simulations have only been used to calculate relative solvation
energies between the amines. To facilitate the comparison with the other data in the
tables the absolute energy from the SM continuum model has been added. The
RISM-SCF calculations have a systematic error that leads to an overestimation of
the size effect on the solvation energy (da Silva, Yamazaki and Hirata 2005). In
determining the basicity it is the difference of solvation energies between two

73
4 Modeling of Solvation Energy

species of approximately the same size that is important and this error is therefore
expected to cancel out.

Table 4.2 Free Energy of Solvation for Neutral Amines. Data in [kcal/mol].
Molecule RISM-SCF FEP-MK FEP-CM2 PCM SM Exp.a
Ammonia -0.3 -4.87b -4.87b -4.30 -4.87 -2.41
Methylamine 5.1 -4.26 -6.72 -4.75 -5.14 -2.68
Ethylamine 10.1 -5.13 -6.36 -4.48 -4.90 -2.61
Dimethylamine 13.6 -1.40 -4.24 -4.30 -3.78 -2.41
Trimethylamine 23.3 1.44 -3.67 -2.76 -2.98 -1.34
Piperidine 24.1 -2.26 -4.65 -4.91 -4.33 -0.83
Piperazine 16.7 -7.41 -5.8 -10.65 -9.08
Morpholine 15.8 -5.34 -7.14 -9.01 -7.24
Pyrrolidine 21.5 -2.08 -9.50 -5.56 -6.03
TMP 47.9 -25.36 -6.98 -8.34 -5.62
a
Data from Jones and Arnett (1974).b SM model values.

74
4 Modeling of Solvation Energy

Table 4.3 Free Energy of Solvation for Protonated Amines. Data in [kcal/mol].
Molecule RISM-SCF FEP-MK FEP-CM2 PCM SM
Ammonia -57.6 -87.19a -87.19a -79.9 -87.19
Methylamine -43.5 -76.23 -80.90 -70.25 -76.39
Ethylamine -37.8 -75.48 -79.63 -67.79 -72.70
Dimethylamine -29.7 -69.14 -71.78 -65.45 -67.07
Trimethylamine -16.6 -62.60 -63.57 -59.12 -59.38
Piperidine -9.5 -61.60 -69.09 -62.35 -61.70
Piperazine -21.3 -61.60 -70.92 -69.22 -66.20
Morpholine -22.1 -72.43 -75.08 -71.78 -67.73
Pyrrolidine -17.2 -66.06 -70.89 -63.79 -63.91
TMP 23.9 -86.64 -65.29 -57.61 -56.23
a
SM model values.

In Table 4.4 the relative free energy of protonation determined from experimental
data and the various solvation models are shown. All data are given relative to
ammonia.

75
4 Modeling of Solvation Energy

Table 4.4 Basicity in solution. Data in [kcal/mol].


Molecule exptl pKa Gpsa Gpsa
exptl RISM-SCF FEP-MK FEP-CM2 PCM SM
Ammonia 9.24b 0.00 0.0 0.0 0.0 0.0 0.0
Methylamine 10.65b 1.92 1.5 -0.2 2.1 0.1 -0.9
Ethylamine 10.78b 2.10 5.0 2.3 5.3 2.0 -0.2
Dimethylamine 10.8b 2.13 3.9 3.3 3.1 3.5 -1.1
Trimethylamine 9.80b 0.90 4.8 3.9 -0.2 3.0 -3.7
Piperidine 11.12c 2.56 0.5 1.2 6.3 6.0 -0.8
Piperazine 9.83c 0.80 4.0 6.0 10.2 6.3 -1.9
Morpholine 8.49c -1.02 -2.3 1.7 2.6 4.2 -4.8
Pyrrolidine 11.30c 2.81 5.1 5.2 2.7 6.2 -0.9
TMP 10.05c 1.10 -3.5 8.6 5.7 3.4 -2.0
a
Energies relative to ammonia. b Data from Jones and Arnett (1974). c Data from
Perrin (1965).

In Figure 4.7 the gas phase basicity is plotted against the experimental pKa. This
figure illustrates the importance of solvation energies in determining relative
basicities in solution. Figure 4.8 shows the results for the RISM-SCF, FEP-MK and
FEP-CM2 models. Finally in Figure 4.9 the PCM and SM results are shown.

76
4 Modeling of Solvation Energy

Figure 4.7 Calculated gas phase basicity versus experimental pKa. The stippled
line indicates the theoretical trend (in solution) relative to ammonia.

77
4 Modeling of Solvation Energy

Figure 4.8 Calculated solution phase basicity versus experimental pKa Crosshairs are
RISM-SCF, open circles are FEP-MK and squares are FEP-CM2. The stippled line
indicates the theoretical trend relative to ammonia.

78
4 Modeling of Solvation Energy

Figure 4.9 Calculated solution phase basicity versus experimental pKa. Triangles are PCM
results while diamonds are SM model. The stippled line indicates the theoretical trend
relative to ammonia.

4.5.5 Conclusion
Looking at Figure 4.8 and 4.9 it is clear that none of the models produce results in
full agreement with experimental data. All the models do however successfully
close the large differences between relative basicities in gas phase and solution
(Figure 4.7). It must however be concluded that the expectation expressed by Jones
and Arnett (quotation in beginning of this chapter) over 30 years ago have only
partly been met.
Cramer and Truhlar (1999) did in their review note that accurately predicting the
relative basicity in solution between ammonia, methylamine, dimethylamine and
trimethylamine has proven to be difficult. Amines distinguish themselves from
most other functional groups in having varying number of hydrogen atoms, thereby

79
4 Modeling of Solvation Energy

also having a varying potential to form intermolecular hydrogen bonds. None of the
present models are accurate enough to capture differences caused by such changes
at a quantitative level.
The agreement with experimental data is comparable for the different models.
The SM model does produce the best trend, the main error being in the solvation
energy of ammonia. A rather odd feature of Figure 5.8 is that the two continuum
models appear to have errors of same dimension and opposite sign for the different
molecules. Agreement with experimental data for the SM model is consistent with
its reported uncertainty for ionic species (Li et al. 1999). The level of agreement
with experimental data for the other models is also generally in line with what is
reported in the literature. It should be emphasized that all models are reasonably
robust in a qualitative sense, capturing larger trends in solvation energies.

80
5 Reaction Mechanisms and Equilibrium

5 Reaction Mechanisms and Equilibrium


Kinetics is nature’s way of preventing everything from happening all at once.
S. E. LeBlanc

5.1 Introduction
This chapter is intended to give an overview of the reaction mechanisms of CO2
absorption in aqueous amine systems. On several key points it draws on my own
work. Most of it published (da Silva and Svendsen 2004a and da Silva and
Svendsen 2004b), but some not previously presented. This chapter also serves as a
review of observations in the literature. In addition to providing an overview of the
important aspects of CO2 capture, this chapter is also intended to show to what
extent central equilibrium constants can be modeled. This part is mainly based on
my own modeling work.

5.2 Reaction Mechanisms


5.2.1 Introduction
CO2 reacts in aqueous amine systems to form either bicarbonate or carbamate.
These species are shown in the figure below. The R groups in NR2 can be a proton
or any form of substituent group.

Figure 5.1 Bicarbonate and carbamate.

81
5 Reaction Mechanisms and Equilibrium

Bicarbonate is produced by the reaction of a CO2 molecule with a water molecule,


while carbamate is formed by the reaction of a CO2 molecule with an amine
molecule. CO2 in the liquid is bound almost entirely in one of these two forms, and
only a small fraction is found as free CO2.

5.2.2 Bicarbonate Formation


The formation of bicarbonate from CO2 and water is a well known reaction in
chemistry. There are three (related) mechanisms for this reaction.
CO2 + H 2O U H 2CO3 (5.1)

CO2 + OH − U HCO3− (5.2)

H 2CO3 + OH − U HCO3− + H 2O (5.3)

Bicarbonate can again be deprotonated by a base molecule (B).

HCO3− + B U CO3−2 + BH + (5.4)

The base molecule is usually an amine molecule or a hydroxyl ion ( OH − ). By


itself bicarbonate formation is however a rather slow reaction. It has been observed
that this reaction proceeds more quickly in the presence of amine molecules, an
effect beside the direct effect of the amines acting as bases (Donaldsen and Nguyen
1980). It is also known that Brønsted bases can act to catalyze the formation of
bicarbonate (Sharma and Danckwerts 1963).

Calculations have been performed to identify a mechanism that might account for
this increased reaction-rate. These calculations were performed with PC Spartan
(1999) at the HF/6-31G(d,p) level. Calculations were performed with one water
molecule and one CO2 molecule in the presence of an ethanolamine molecule. In

82
5 Reaction Mechanisms and Equilibrium

this case the reaction coordinate chosen was O(H2O)-C(CO2). The determined
transition-state geometry is shown in Figure 5.2.

Figure 5.2 Transition state of bicarbonate formation.

The transition state identified in these calculations is consistent with the


mechanism proposed by Donaldsen and Nguyen (1980):

Figure 5.3

At the HF/6-31G(d,p) level this mechanism was found to have a barrier of 29.5
kcal/mol. For the bicarbonate formation in water a reaction-barrier of 42.5 kcal/mol
has been reported with the same type of calculation at the same level of theory
(Nguyen et al. 1997). The presence of base-molecules can therefore be seen to give
a significantly lower barrier to bicarbonate formation.

83
5 Reaction Mechanisms and Equilibrium

While this mechanism is usually mentioned in the context of tertiary amines I do


not believe it is unique to them. As observed above, it is known that base catalysis
can take place for a number of Brønstad bases. The calculations suggest that this
reaction will take place with any base molecule of appropriate strength. For
primary and secondary amines the kinetics will however often be dominated by
carbamate formation and base catalysis will play a lesser role and perhaps be more
difficult to detect experimentally. Base catalysis may however be significant for
primary and secondary amines in systems where bicarbonate formation dominates.

5.2.3 Carbamate formation


The carbamate formation is one of the main reactions of CO2 absorption. Two
mechanisms have been proposed for this in the literature. One is the zwitterion
mechanism originally proposed by Caplow (1968):

Figure 5.4

In this mechanism CO2 forms a bond to the amine functionality in a first step. In a
second step an amine-proton is transferred to a second molecule. In Caplows article
the second molecule was water, but this can be any base-molecule.
The intermediate species in the reaction is a zwitterion. Caplow assumed (as shown
in Figure 5.4) that a hydrogen bond is formed between the amine and a water
(base) molecule before the amine reacts with the CO2 molecule. This feature has
however been omitted in the recent literature, as can be seen in the work by
Danckwerts (1979), Versteeg et al. (1996) and Kumar et al. (2003).

84
5 Reaction Mechanisms and Equilibrium

Crooks and Donnellan (1989) proposed the following single-step mechanism:

Figure 5.5

Here B is a base molecule. In this mechanism the bonding between amine and CO2
and the proton transfer take place simultaneously.
It has been argued that the rate-expression of the zwitterion mechanism can be
better used to account for experimental observations such as broken order kinetics
in some solvents (Versteeg et al. 1996). In a paper appearing in this thesis (da Silva
and Svendsen 2004a) the experimental evidence was reviewed and quantum
mechanical calculations were carried out to determine the most likely mechanism.
A central finding from these calculations was that a CO2 molecule does not react
with an ethanolamine molecule in gas phase. The calculations strongly suggest that
base molecules must be present for CO2 to bond with amine molecules. This would
again suggest that if any reaction intermediate exists it can not be very stable and is
likely to be short-lived. Calculations also suggest that if a strong base (such as
another amine molecule) is interacting with the amine functionality there is no
barrier to the proton-transfer. This is consistent with a single-step mechanism.
What could not be resolved by the calculations is what would happen if the CO2-
amine complex was solvated entirely by water molecules. In this case it is possible
that the water molecules could transfer a proton to a base molecule located further
away. Alternatively the CO2-amine complex can remain stable, awaiting the
approach of a base molecule. The calculations, while not entirely conclusive,
suggest a single-step mechanism, or a short-lived reaction intermediate.

85
5 Reaction Mechanisms and Equilibrium

The two reaction mechanisms give rise to similar expressions for reaction
kinetics. It was found in the review (da Silva and Svendsen 2004a) that most
experimental data can be accounted for with both mechanisms. The zwitterion
mechanism based rate-expression is somewhat more flexible than expressions
based on the single-step mechanism, i. e. it can be fitted to a greater variety of data.
There is however no experimental data that I am aware of suggesting this flexibility
to be required. An argument against the zwitterion mechanism is that some
parameters at times take on values that do not seem plausible (Crooks and
Donnellan 1989 and Aboudheir et al. 2003). If the overall order of reaction is three
it follows from the zwitterion mechanism that a proton transfer is rate-determining,
a conclusion that would seem somewhat implausible.
The difficulty in drawing any definite conclusion as to what mechanism is correct
stems in part from the fact that they are very similar. The zwitterion mechanism
becomes equivalent to the single-step mechanism when the lifetime of the
zwitterion-intermediate approaches zero. This has been somewhat obscured in the
literature where the zwitterion mechanism has been written without hydrogen
bonds to base molecules. The calculations would suggest that while the zwitterion
mechanism is not necessarily wrong the single-step mechanism is more suited to
conveying the nature of the reaction taking place.
From the quantum mechanical calculations potential reaction-barriers were also
identified. The mechanism was found to have no intrinsic barrier. The kinetics is
therefore likely to be dominated by the need for molecular encounters for the
reaction to take place. One barrier may arise from the CO2 molecule having to
displace the solvation-shell around the amine functionality. A second barrier may
be caused by the need for the CO2, amine and a base-molecule to be aligned for
reaction to take place. A study on the liquid structure of ethanolamine-water
mixtures (da Silva, Kuznetosova and Kvamme 2005, also part of the present thesis)
suggest that the latter of these barriers is the most significant.

86
5 Reaction Mechanisms and Equilibrium

As can be seen from the mechanism in Figure 5.5, the carbamate-formation


involves the transfer of a proton from the amine functionality. It is well-known that
amines without such protons, such as tertiary amines, do not form carbamate
(Versteeg et al. 1996 and da Silva and Svendsen 2005).

5.2.4 Bases
A base molecule can by definition undergo the following reaction:

B + H 2O U BH + + OH − (5.5)

Amine molecules are all bases and they are usually the strongest bases present in
the system. Water itself is a weak base, the hydroxyl-ion ( OH − ) is a strong base,
but usually present in small quantities. Bicarbonate is a very weak base. The
carbamate species might act as a base, bonding a proton to one of the CO2 oxygen
atoms. There is however nothing in the experimental data to suggest that such
protonation takes place.

5.2.5 Alcohol-Group Bonding to CO2


It has been suggested that at very high pH values, CO2 can bond to alcohol-groups
(Jørgensen and Faurholt 1954). To investigate the possibility of such a reaction
taking place calculations were carried out at HF/3-21G(d) level with PC Spartan
(1999). The calculations were carried out for two ethanolamine molecules and a
CO2 molecule in vacuum. Calculations were carried out with C(CO2)-
O(ethanolamine) as a restrained reaction coordinate. The results suggest a
mechanism analog to carbamate formation:

87
5 Reaction Mechanisms and Equilibrium

Figure 5.6

This reaction is however in general not expected to play a significant role in


industrial CO2 absorption processes as the pH of the system is usually not high
enough (Versteeg et al. 1996).

5.2.6 Carbamate as Reaction Intermediate


It has been suggested that carbamate undergoes a direct hydrolysis reaction (Smith
and Quinn 1979), meaning a direct reaction with water to form bicarbonate and
amine. The figure below shows a HF/3-21G(d) geometry of a water molecule
placed next to MEA carbamate.

Figure 5.7 Ethanolamine carbamate interacting with a water molecule.

From such calculations no reaction path or transition-state was found. It is


apparently not possible for a water molecule to bond to the CO2-group as long as it
remains bonded to the amine functionality. The water oxygen molecule must

88
5 Reaction Mechanisms and Equilibrium

interact with the CO2 group in the same site as it interacts with the amine-group. I
am also not aware of any experimental data that suggest that direct carbamate
hydrolysis takes place. Shifts in concentration between carbamate and bicarbonate
are not particularly fast and can be readily explained as the shift of equilibrium
through a set of reversible reactions. If equilibrium conditions change some
carbamate will revert back to amine and CO2. This CO2 can then go on to form
bicarbonate.

In general it would seem that the bonding of carbamate species is such that it is
unlikely to act as any form of reaction intermediate.

5.2.7 Molecules with Multiple Amine Functionalities


Molecules can have more than one amine functionality. Among the solvents being
considered for CO2 capture piperazine, and piperazine-derivatives have multiple
amine functionalities. So does N-(2-hydroxyethyl)ethylenediamine (AEEA) which
has received some attention recently (Ma’mun, Svendsen, Hoff and Juliussen 2004
and Bonenfant et al. 2005). The nature of the functional groups is the same in such
molecules as in simpler amines. The form of interactions with CO2 is therefore also
likely to be the same. In the case of multiple amine functionalities there is however
a greater number of species that can be formed. In Figure 5.8 the species that can
be formed by piperazine are shown.

89
5 Reaction Mechanisms and Equilibrium

Figure 5.8 Piperazine species.

In addition to the carbamate and protonated amine there is a diprotonated species, a


dicarbamate, and a form with combined protonation and carbamate formation.
Experimental work and analyses by Bishnoi (2000) suggest that these species can
exist in significant concentration. In piperazine the two amine functionalities are
equivalent, for AEEA and other molecules with non-equivalent amine sites the
number of potential species would be even greater. Systems with such amines are
more difficult to study experimentally, as the number of experimentally observable
properties often remains the same while the number of equilibrium constants to be
determined increases significantly.

90
5 Reaction Mechanisms and Equilibrium

5.2.8 Shuttle Mechanism


In the absorption process CO2 must be transferred from the gas-liquid interphase to
the bulk of the liquid. The CO2 can diffuse as a free molecule or in one of the
bound forms. The diffusion process takes a significant amount of time, and may
under some conditions determine the overall absorption rate.
It has been observed that mixtures of amines can absorb CO2 more quickly than
would be expected from considering the kinetics of the different species involved
(Versteeg et al. 1996). This has been explained as resulting from carbamate
forming species transporting CO2 from the interface to the bulk of the liquid, where
it goes on to undergo bicarbonate formation. This process has been demonstrated in
absorption models and it is referred to as the shuttle mechanism (Versteeg et al.
1990). This is not a chemical reaction, but rather a phenomenon arising from the
differing diffusion-rates of different species. The carbamate forming amine in this
process is usually referred to as a “promoter” or “activator”.

Figure 5.9 The shuttle mechanism.

91
5 Reaction Mechanisms and Equilibrium

5.2.9 Summary and Conclusion


As is shown in this chapter the number of reactions involved in the CO2 absorption
process is quite small. It can not be ruled out that other reactions take place, but the
present set of reactions can to my knowledge account for all the observed behavior
of CO2 absorption in amine systems. Thus, all the reactions involving CO2 can be
generalized in the following simple form:

Figure 5.10

B is again a base molecule and AH is any molecule with a free-electron pair and a
hydrogen atom on the same site. If AH is an amine molecule and B is a water
molecule or a second amine molecule this is the carbamate formation mechanism.
If AH is a water molecule and B an amine molecule the reaction is base catalysed
bicarbonate formation. Finally with both AH and B as water the reaction becomes
standard bicarbonate formation.
Below is given a summary of the main reaction mechanisms. Alcohol-group
bonding to CO2 is in most conditions expected not to be significant, and is omitted
from this list.

92
5 Reaction Mechanisms and Equilibrium

CO2 + Am + B U AmCO2− + BH + I

CO2 + H 2O + B U HCO3− + BH + II

B + H 2O U BH + + OH − III

CO2 + H 2O U H 2CO3 IV

CO2 + OH − U HCO3− V

HCO3− + B U CO3−2 + BH + VI

B is again used to indicate a base molecule, usually an amine molecule or hydroxyl


ion. This set of reactions is in itself quite small, and there are only four reactions (I,
II, III and VI) that are amine specific. Three of the four amine reactions depend on
the base strength of the amine, the final reactions depend on the stability of the
carbamate molecule formed by a given amine. The base strength is usually given as
the pKa, and the equilibrium constant for carbamate formation will be written as Kc.
The CO2 absorption process is usually based on temperature ( T ) variation, and the
temperature dependency of the equilibrium constants is clearly required to predict
the performance of a given solvent. The equilibrium constants in the system do also
change as a function of the composition ( c ) of the liquid, and the concentration
dependency of the equilibrium constants is clearly also of importance. This means
that if pK a (T , c) and K c (T , c ) can be determined the absorption process for

different solvents can be predicted. In other (and more correct) words it means
determining the equilibrium constants, their temperature dependencies and the
activity coefficients of the various components present in the liquid. This is the
main goal of the present thesis. Tertiary amines do not form carbamate and for such
molecules the base strength is the only equilibrium constant governing the
reactivity towards CO2.

93
5 Reaction Mechanisms and Equilibrium

5.3 Determining Equilibrium

5.3.1 Equilibrium and Kinetics


Comparison of calculated and experimental base strengths of a series of amines
(results in Chapter 4 and da Silva and Svendsen 2003) showed the models to be
reasonably accurate. Determination and prediction of base strength can however be
an easier task than implied by the results of these papers. Determination of base
strength is fairly routine experimental work and fairly accurate data is available for
a large number of molecules (Perrin 1965). Very often the pKa of a molecule being
considered as a solvent has been reported in the literature. In most of the cases
where the pKa for the molecule itself has not been reported, it has been reported for
some closely related substance. The modeling task can then be reduced to
determining the difference in base strength for two closely related compounds; this
can usually be done with a fairly high degree of confidence.
The modeling results for carbamate stability (da Silva and Svendsen 2005) appear
to be of the same quality as the results for base strength. The experimental
determination of carbamate stability is much more difficult than the determination
of base strength, and it is in this case much more difficult to draw on experimental
data to refine estimation of equilibrium constants for new molecules.
The set of modeling results for base strength and carbamate stability together
provide what can perhaps be summarized as a semi-quantitative model for new
solvents.
In modeling work (da Silva and Svendsen 2005) a linear correlation was found
between carbamate stability and rate of reaction. For tertiary amines a similar
correlation has been found between the base strength and rate of reaction (Versteeg
et al. 1996). This is not entirely surprising since all the molecules undergo analog
reactions. This simple relationship between equilibrium constants and reaction

94
5 Reaction Mechanisms and Equilibrium

rates, suggest that predictions of equilibrium constants can also serve to predict
reaction rates.
CO2 in aqueous amine solutions reacts, as already noted, to form either carbamate
or bicarbonate. Carbamate formation is in general a far more exothermic reaction
than bicarbonate formation. It is in general faster than bicarbonate formation, but
more energy is also required to reverse the reaction in the stripper. Two amine
molecules are required to form a carbamate molecule. The loading (mol CO2/mol
amine solvent) will therefore not be much higher than 0.5 in a system dominated by
carbamate formation.
An issue I have not addressed so far is what the ideal values are for base strength
and carbamate stability, or to which extent carbamate formation or bicarbonate
formation is preferable. At present this remains unknown. This is not a question
that can be addressed with the tools of computational chemistry. A complete model
of the CO2 absorption process is required. While such models exist they have so far
not been applied to this question. In Chapter 7 I will look at what the ideal
characteristics might be.

5.3.2 Temperature Dependency of Equilibrium Constants


In the work on basicity (da Silva and Svendsen 2003) it was found that entropies
calculated from quantum mechanical calculations could be used to predict the
change in basicity over temperature. The quality of the correlation obtained was
very good. The results suggest that the change in basicity over temperature depends
on the nature of the intramolecular hydrogen bonding of a given molecule. Only
eight molecules were however included in the study of temperature effects. A
difficulty in expanding the comparison with experimental data is that experimental
data for basicity is not always consistent at higher temperatures (Kamps and
Maurer 1996). The prediction of changes in basicity over temperature can be a

95
5 Reaction Mechanisms and Equilibrium

valuable tool, particularly for higher temperatures where very little, if any,
experimental data is available.
Predicting the change in the carbamate stability over temperature is more
difficult. Almost no experimental data of quantitative quality is available for
carbamate stability at any temperature, and this is required to obtain the change
over temperature (da Silva and Svendsen 2005). Computational chemistry could be
used to determine both the absolute equilibrium constant and absolute entropy
values, but without any experimental data with which to compare the results, the
quality of predictions made would be uncertain.

5.3.3 Activity Coefficients


The equilibrium constant models results are for infinite dilution, far from the
composition that is utilized in industrial CO2 absorption. There is therefore a need
to predict how the equilibrium constants change as function of system composition.
Such concentration dependency is usually given as activity coefficients for the
various components
I have made some effort to model the activity coefficients for the ethanolamine-
water mixture (da Silva 2004). The quality of the results was however not
particularly good, no effort had been made to find an optimal force field for
ethanolamine and the accuracy of the simulations could have been better. In later
work I have looked at force field parameterization for ethanolamine (da Silva,
Kuznetsova and Kvamme 2005), this force field has however yet not been applied
in determination of activity coefficients. While I am not entirely confident that the
force field presented in that paper will produce activity coefficients in full
agreement with values derived from experiment, I am confident that a force field
can be developed that reproduces both the properties of ethanolamine-water
mixtures and pure ethanolamine.

96
5 Reaction Mechanisms and Equilibrium

The CO2 absorption system consists of many components, not only water and
ethanolamine. I am not aware of any work having been done to determine activity
coefficients in such multi-component systems. Long simulations would probably
be required to produce results with a reasonable degree of confidence; this is
something that can be carried out with sufficient patience and computer-resources.
Another issue is how accurate activity coefficients from simulations will be.
Vorholz et al. (2004) were able to demonstrate the salting-out effect in molecular
simulation. This is an interesting result, suggesting that simulations are able to
capture different effects of changing concentrations at a qualitative, if not a
quantitative level.
Activity coefficients are not likely to be one of the most important factors in
determining the overall performance of a solvent. A first approximation for a new
solvent could be to assume that the system has the same activity coefficients as
some structurally similar solvent for which experimental data is available.

5.3.4 Process Energy Consumption


I have not dealt specifically with modeling of the reaction energies and energy
consumption of the system, but it can be readily determined from the modeling
results for equilibrium constants. Once the constants are determined, reaction
energies can be derived with thermodynamic calculations.
Reaction energies determined from computational chemistry work in the present
thesis are not likely to be close to experimental values. As with the equilibrium
constants I believe the best results will be obtained by using computational
chemistry results to obtain relative values, these can then be anchored to results for
a solvent where experimental data are available.

97
5 Reaction Mechanisms and Equilibrium

5.3.5 Summary
The models presented in this thesis can be used to predict the main equilibrium
constants for CO2 absorption in aqueous amine systems with a fair degree of
confidence. From such predictions the solvents effectiveness as an absorber for
CO2 capture can be estimated. Results also suggest that quantum mechanical
calculations can be used to predict changes in equilibrium constants over
temperature. Less work has been done on determining the activity coefficients of
the systems in question. Confident prediction of activity coefficients is likely to
require considerable simulation work. While knowledge of activity coefficients is
clearly desirable, they are not likely to be that important in predictions of the
relative differences between solvents.

98
6 Other Solvent Properties

6 Other Solvent Properties


Why, a four-year-old child could understand this. Someone get me a four-year-old.
Grocho Marx

6.1 Introduction
In the last chapter the reaction mechanisms and equilibrium of amine-CO2-water
systems were discussed. While the equilibrium is central to the applicability of an
amine solvent to CO2 absorption there are other issues that are also of importance.
The present chapter will deal with these other properties.
Properties is here used in a loose sense, issues ranging from solvability to cost
will be covered. The outline will follow a prioritized order, with the properties I
believe to be of greatest importance presented first. This chapter is essentially a
review; the current state of knowledge for each property will be briefly discussed
and the issue of modeling and prediction of behavior of new solvents will be
considered.

6.2 Solubility in Water


Loss of solvent by evaporation in the stripper and absorber can be a problem. If a
solvent has low solubility in water that will also limit the amine concentration
under which the process can be operated. The calculation of solvation energy has
been one of the main topics of my work, for the solvent itself (which is a neutral
molecule) any of the solvation models utilized in the present work can produce
results with a fairly high degree of confidence. The continuum models are
particularly well suited for this task.

99
6 Other Solvent Properties

6.3 Solvent Degradation


There is in a aqueous amine system the possibility that the amine will undergo
undesirable side reactions. Such reactions can result in the amine molecules
degrading irreversibly. The degradation rate of the solvent can greatly influence the
overall cost of operating a CO2 absorption plant.
The solvent can be subject to three types of degradation: thermal, carbamate
polymerization and oxidative (Goff and Rochelle 2004). Exhaust gases usually
have a high content of oxygen, oxidative degradation can in such systems be
expected to be the main cause of degradation. Thermal degradation can however be
an issue for some solvent molecules (Nagao et al. 1998). Degradation is a problem
for several reasons (Goff and Rochelle 2004); because the solvent is degraded new
solvent must be added at regular intervals. Degradation products might also have
unfavorable characteristics that the solvent itself does not have; degradation
products may be toxic and corrosive. There is therefore an interest in understanding
degradation mechanisms.
Degradation reactions are inherently difficult to study. Experimental work can be
time-consuming as the reactions may be relatively slow, degradation may also be
catalyzed by impurities in the system. Experimental monitoring of degradation is
also made difficult by the lack of a priori knowledge of what degradation products
are formed in a given system.
It is today believed that the oxidation process proceeds by radical reactions (Chi
and Rochelle 2002 and Goff and Rochelle 2004). Radical chemistry is often
complex and a large number of different products can be formed. Goff and
Rochelle (2004) present a series of possible steps in degradation mechanisms for
ethanolamine, but the mechanism remains uncertain.
Blanc et al. (1982) did experimental work on the degradation of diethanolamine
and N-methyldiethanolamine. They separated the degradation products into basic
and acid products. The basic products were amine molecules of complex structure,

100
6 Other Solvent Properties

while the acid products were small carboxyl acid molecules such as formic acid.
Ammonia is also known to be formed in amine degradation (Goff and Rochelle
2004).
There is little reliable data on the relative degradation of different amines in the
open literature. The work by Blanc et al. (1982) suggests that ethanolamine (a
primary amine) degrades more quickly than secondary and tertiary amines. Kohl
and Nielsen (1997) accepted this conclusion in their textbook. Substituent groups
on the amine functionality itself, or on neighboring atoms, appear to protect the
solvent from oxidative reactions. This insight can be used to make rough
predictions of the degradation rates of different amines. Another issue is if it is the
solvent molecule itself, the carbamate form or the protonated form that is most
vulnerable to oxidation. There is to my knowledge no work in the literature that has
brought any clear insight on this issue.
I do believe that computational chemistry calculations could be used to elucidate
the reaction mechanisms and gain some insight into how the degradation rate will
vary between different solvents.
For environmental reasons solvents that are biodegradable are preferable. It is
however possible that a solvent resistant to oxygen and water at high temperatures
is not going to degrade easily in nature. There is therefore a potential conflict
between the properties desirable in a solvent in the process and after disposal.

6.4 Corrosion
Kohl and Nielsen (1997) describe corrosion as the most serious problem affecting
amine gas absorption plants. They also observe that the corrosion rates vary with
amine solvent. Why this is so, is however not well-understood. It is known that
primary amines such as ethanolamine do have a high corrosion rate. Secondary
amines such as diethanolamine have a lower corrosion rate and even lower

101
6 Other Solvent Properties

corrosion rates are observed for tertiary amines. It would seem, and is suggested by
Kohl and Nielsen (1997), that the corrosion rate depends on the concentration of
carbamate species (carbamate species often dominate in ethanolamine systems
while they are not formed by tertiary amines). Predictions of carbamate formation
can be obtained from the work in this thesis and this might serve to give a rough
estimate of the corrosion rate associated with different solvents.
The corrosion rate is probably not only a function of the solvent and operating
conditions, it can also depend on impurities in the system. Degradation products
formed may also contribute to corrosion in different ways than the solvent itself.
The extent of corrosion obviously depends not only on the solvent, but also on the
structure of the absorption equipment and the choice of materials.

6.5 Foaming
Kohl and Nielsen (1997) describe foaming as the most common problem in amine
treating plants. Foaming may result in higher loss of amine and reduced liquid-gas
interphase area, thereby reducing the efficiency of the separation.
Kohl and Nielsen (1997) attribute foaming to impurities in the system. Their
textbook, and much of the earlier literature on the amine absorption technology,
look at plants separating CO2 from natural gas. The amount and nature of
impurities is clearly different in exhaust gases, which is the application of interest
in the present work. It has been observed (Ma’mun, Svendsen, Hoff and Juliussen
2004) that 2-methylaminoethanol produces significant foam, while ethanolamine
foams much less.
Foaming is a fairly complex and an, at present, not entirely predictable
phenomenon (Morrison and Ross 2002). Foaming is caused by some components
acting as surfactants. In an aqueous solution, amines with hydrophobic
functionalities are therefore perhaps the most likely to act as surfactants and cause
foaming. Methods in computational chemistry can be used to determine the extent

102
6 Other Solvent Properties

of hydrophobic and hydrophilic groups in a molecule. From data on hydrophobic


and hydrophilic groups some estimates can perhaps be made of the likelihood of an
amine contributing to foaming.

6.6 Toxicology
CO2 absorption is a large scale industrial process and it is clearly preferable that
solvents used are not highly toxic. The European Chemical Substances Information
System (ESIS, European Chemicals Bureau 2005) provide data on many amine
compounds. The ESIS data would suggest that most amines utilized in CO2
absorption are to some extent toxic. This is a consequence of the amines being
fairly strong organic bases. Some amines may however pose additional problems.
Piperazine is for example reported by ESIS to be harmful to aquatic organisms.
Some solvents are in wide use in the industry and for these it would seem likely
that issues regarding toxicology are well known. For amines that are not widely
used the effects of long term exposure might be unknown. I am not aware of any
correlation between amine structure and toxicology. Prediction of toxicology is
likely to be extremely difficult. This is however an important issue that must be
kept in mind during screening for new solvents.

6.7 Cost
The overall cost of a solvent depends on the cost of producing it and the
degradation rate. If a solvent is to be applied in an industrial scale process it is of
great importance that the solvent cost is not too high. The cost of producing an
amine depends on the synthesis process. Determining the optimal synthesis process
is a science in itself. For the solvents that are presently utilized on an industrial
scale it would seem likely that the production method has been carefully selected
and that costs will not change too much in the future.

103
6 Other Solvent Properties

For solvents that are not presently produced on a large scale it would seem likely
that less work has gone into finding optimal synthesis routes. If demand for any
solvent was to rise, the cost might therefore go down. It is probably at present not
possible to predict the cost of producing a molecule on bulk scale. The cost of a
solvent is however likely to rise with molecular size and complexity.

6.8 Precipitations
It has been observed that in some absorption systems precipitations can be formed.
Some precipitations are probably the result of the amine itself precipitating as a
result of lowered solvability, there might also be circumstances where some form
of salt can be formed. Kumar et al. (2003) reported precipitations for CO2
absorption with amino acid salts. This is however an issue that has received little
attention in the open literature. For most solvents and under most operating
conditions this is probably not an issue.

104
7 Present and Potential Solvents

7 Present and Potential Solvents


“Tracking something,” said Winnie-the-Pooh very mysteriously.
”Tracking what?” said Piglet, coming closer
”That's just what I ask myself. I ask myself, What?”
A. A. Milne (From Winnie-the-Pooh)

7.1 Introduction
In the last two chapters the solvent properties of importance for the performance of
a CO2 capture process were reviewed. The extent to which to different properties
can be modelled and predicted was also discussed. The present chapter deals with
the likelihood of finding solvents better than the ones currently in use. The first part
of the chapter is a brief review of the most important solvents in use. In the second
part a discussion is made of what the ideal solvent properties are. Finally the
likelihood of finding a molecule with a set of properties that are ideal or at least
better than ethanolamine (which represents the benchmark to beat) is discussed.

7.2 Solvents in Use

7.2.1 Ethanolamine
Ethanolamine (MEA) is currently being provided commercially as a solvent for
exhaust gas CO2 capture. It is sold under the name Econamine by Fluor enterprises
(Reddy et al. 2003). The main problem with the ethanolamine processes is that the
energy consumption is fairly high. MEA is known to form a stable carbamate form
(da Silva and Svendsen 2005). The energy needed to free CO2 from the carbamate
form in the stripper is the cause of the high energy consumption. Ethanolamine is
highly solvable in water, not very toxic and does not cause significant foaming.

105
7 Present and Potential Solvents

The molecule is noteworthy for being the second smallest of all alkanolamine
molecules.
Ethanolamine does have a high degradation rate and can cause corrosion. In
Econamine inhibitors are added to reduce degradation rates and limit corrosion.

7.2.2 Tertiary Amines


Tertiary amines do not form carbamate. They only act as bases, contributing to the
formation of bicarbonate. The advantage of tertiary amines is that the equilibrium
is more easily reversed in the stripper. Because the amine-CO2 stoichiometry of
bicarbonate formation is 1 to 1 tertiary amines do also have the potential to absorb
large amounts of CO2. As observed in the last chapter tertiary amines do also tend
to have low degradation rates. For natural gas treatment the tertiary amine N-
methyldiethanolamine (MDEA) is widely used (de Koeijer and Solbraa 2004). In
exhaust gases the fraction of CO2 in the gas phase is however lower and MDEA is
thought to have too low reactivity to work efficiently in such a case. Tertiary
amines are often combined with promoters in order to take advantage of the
shuttle-effect (Bishnoi and Rochelle 2002 and Zhang et al. 2003).

7.2.3 Sterically Hindered Amines


Amines with one or more substituent-group on the carbon-atoms bonded to the
nitrogen atom are called sterically hindered amines (Sartori and Savage 1983).
Such molecules tend not to form stable carbamate forms. They will therefore
mainly work as bases and contribute to bicarbonate formation. Their reactivity
towards CO2 is very similar to that of the tertiary amines. The sterically hindered
amine that has seen the greatest use is probably 2-amino-2-methylpropanol (AMP).
In addition to tertiary amines and sterically hindered amines there is likely to be
other amines that form less stable carbamate forms.

106
7 Present and Potential Solvents

7.2.4 Multiple Amine Functionalities


Some research has gone into molecules with multiple amine functionalities. Of
such molecules piperazine has probably seen the greatest use. Piperazine is usually
used as a promoter (Bishnoi and Rochelle 2002, Zhang et al. 2003 and Jenab et al.
2005). A problem with piperazine is that it has fairly low solvability in water.
Recent work (Ma’mun et al. 2004 and Bonenfant et al. 2005) has suggested that N-
(2-hydroxyethyl)ethylenediamine (AEEA) is a promising diamine solvent.
Multiple amine functionalities would suggest a potential to bind more CO2 with a
single solvent molecule. Further research is probably required to determine if there
is a particular benefit in using such solvents.

7.2.5 Ionic solvents


Some research has been done on the use of solvents where the active component is
an ionic amine species. Examples of this are potassium salts of taurine and glycine
(Kumar et al. 2003). One advantage of such ionic amines is that there is very little
loss of solvent through evaporation. The solution formed is in this case highly
ionic. There is to my knowledge no data in the open literature to support any
general conclusions on the merits of such solvents.

7.2.6 Patented Solvents


There are some amine systems for which data has been presented in some context
(publications in journals, patents or advertisement brochures) but where the
composition has been kept secret. Most notable among these are the solvents KS 1-
3 (Mimura et al. 1999) developed by Mitsubishi Heavy Industries. I believe KS 1
to be a mixture of AMP with a promoter. Less is known about KS 2 and 3. The
researchers at Mitsubishi have however done work on amino-amide molecules
(Nagao et al. 1998) and these could be possible ingredients.

107
7 Present and Potential Solvents

The Canadian company Cansolv has recently filed a patent application for
absorbents for CO2 capture (Hakka and Ouimet 2004). This again appears to be a
promotor based system. In this patent tertiary amines are utilized. The promoter is
piperazine or a derivative of piperazine. The novel feature of this patent appears to
be the use of oxidation inhibitors and molecules with two tertiary amine
functionalities, such as N, N, N',N'-tetrakis(hydroxyethyl) 1,6-hexanediamine.

7.3 Ideal Solvent Properties

7.3.1 Equilibrium Constants


The chemistry of CO2 absorption is governed by the basicity and carbamate
stability of the solvent. Too high equilibrium constants will result in a too high
energy consumption, while too low equilibrium values will result in CO2 not being
absorbed to any significant degree. It is therefore not an issue of simply finding the
solvent with strongest or weakest bonding to CO2, there is rather some intermediate
values that will provide an optimum trade-off between uptake of CO2 and energy
consumption. I am however not aware of any work that provides insight into what
these optimal values are. This is not an issue that can be addressed by
computational chemistry. To answer the question of what the optimal equilibrium
values are a model of the entire absorption process is required. While such models
are available their application is not trivial and most are developed for a specific
solvent, rather than as general models. It does however seem likely that these
models, with some further development, can be utilized to answer this question.
While modelling work is required to determine optimal equilibrium constants
some suggestions can be made based on the performance of solvents presently in
use. The optimal process may utilize a single solvent or a mixture of different
solvent components. Use of solvents with multiple amine functionalities is a further
option. I will consider each option separately.

108
7 Present and Potential Solvents

Use of a single amine solvent is the simplest option. Two solvents for which the
performance is fairly well understood are MEA and MDEA. MEA forms a fairly
stable carbamate form and is regarded as having too high energy consumption.
MDEA only acts as a base and only bicarbonate is formed. This solvent is also a
relatively weak base having a pKa of only 8.5 (Perrin 1965) where as MEA has a
pKa of 9.5 (Perrin 1965). While use of MDEA results in a lower energy
consumption than is the case for MEA, it is regarded as having a too low capacity
to bind CO2 for the exhaust gas application. The ideal amine solvent should than
have base strength and carbamate stability somewhere between these two solvents.
It is not clear if a non-carbamate forming amine is suitable as a single solvent for
exhaust gas applications, if it were to work it would have to have a high base
strength. A carbamate forming molecule should probably have carbamate stability
somewhat lower than ethanolamine.
Mixtures of solvents are usually used to draw advantage of the shuttle-effect
described in Chapter 5. In such cases one amine works as a carbamate forming
promoter, while another amine mainly works to form bicarbonate. For the
bicarbonate forming molecule a stronger base than MDEA is probably preferable
(for the same reason as given for a single solvent system). The shuttle-mechanism
probably works optimally for some given level of carbamate strength. While it is
difficult to determine what that ideal carbamate stability is for this mechanism,
some tentative conclusions can perhaps be drawn from the promoters currently in
use. Both piperazine and MEA are used as promoters (Bishnoi and Rochelle 2002
and Aroonwilas and Veawab 2004). Both these molecules form relatively stable
carbamate forms (da Silva and Svendsen 2005). This would suggest that ideal
promoters are relatively strong carbamate formers. If KS1 (Mimura et al. 1999) is
indeed promoted AMP, its properties fall within the present recommendations,
AMP having a pKa of 9.7 (Perrin 1995).

109
7 Present and Potential Solvents

In addition to taking advantage of the shuttle-effect, solvent mixtures can also be


used to control the degree of CO2 capture or energy consumption. If a single
solvent does not have ideal performance, a mixture of solvents can be utilized. By
varying concentration ratios the overall performance can be controlled. In such a
case both solvents should have properties that would have made them viable in a
single component system.
Finally there is the use of solvents with multiple amine functionalities. These can
be utilized as a single solvent, or as promoter or bicarbonate former in a mixture of
solvents. The chemistry is in this case more complex than for a solvent with a
single amine functionality. The conditions given for a single amine functionality
solvent in terms of base strength and carbamate stability should also be met by a
solvent with multiple amine functionalities. I have however no basis to suggest
what the ideal equilibrium values for secondary protonation or carbamate
formation is. Some degree of chemical reactivity on all sites is probably preferable;
otherwise part of the molecule will be “dead weight”.
For the temperature dependence the issue is more straightforward; the greater the
fall in carbamate stability and base strength over temperature the better. Greater
temperature dependency increases the net amount of CO2 that can be transported
from the absorber to the stripper (the cyclical capacity). It could also be taken
advantage of to operate the stripper at lower temperature, thereby reducing the
energy consumption.

7.3.2 Other Properties


For the other properties discussed in the last chapter the ideal characteristics are
obvious. The solvent should have as high solvability in water as at all possible. It
should not foam, have low degradation rate, be inexpensive and not toxic.

110
7 Present and Potential Solvents

7.4 Comparison with Ethanolamine


From these conclusions we can then turn to the crucial issue: What is the likelihood
of finding solvent molecules that are better than MEA. Since the ideal equilibrium
constants are not known, it is difficult to draw any conclusions as to how far
ethanolamine is from being an optimal solvent and how much room there is for
improvement. I suspect that a less stable carbamate form is optimal, but how much
the energy consumption can be lowered is hard to estimate.
MEA is probably the smallest and simplest solvent molecule that can be applied
in the CO2 capture process. It is for that reason likely to remain one of cheapest
solvents to manufacture. It is also not particularly toxic. MEA does however have a
high degradation rate (Blanc et al. 1982) and many other solvents are likely to offer
improvements over ethanolamine in this respect.
Overall this analysis would suggest that it is likely that solvents better than
ethanolamine can be found. The KS1 solvent (Mimura et al. 1999) has perhaps
already achieved this, and there is no reason to believe that this can not be
improved further upon. How great improvements that can be expected in terms of
energy consumption and operating cost is however difficult to estimate.

7.5 Conclusion
I have in this work not presented a specific list of candidate molecules for the
absorption process. To the question of what the ideal solvent molecule(s) is, the
present work provides a fairly detailed blueprint of how to answer the question, but
not the answer itself.
The screening process required to come up with a range of candidate molecules
would in itself be time-consuming. At the same time there have not been resources
available at our laboratory to test new candidate molecules resulting from such a
screening process.

111
7 Present and Potential Solvents

Some general rules for design of solvent molecules can however be proposed.
The ratio of hydrophilic to hydrophobic groups must be kept high, that usually
means high number of hydroxyl-groups and low number carbon-based groups. For
a reasonable basicity to be obtained there should be at least two carbon-groups
between hydroxyl- and amine-functionalities.
In general it appears that high basicity is desirable. High carbamate stability may
be desirable for promoters, but is probably not so for the main solvent component.

112
8 Future Work

8 Future Work
“Begin at the beginning,” the King said gravely, “and go on till you come to the
end; then stop.”
Lewis Carroll (from Alice in Wondeland)

8.1 Continuation of the Present Research


There are several ways in which the present work can be expanded and improved
upon. Degradation mechanisms are one particular area where I believe
computational chemistry could be applied to obtain greater insight. Computational
chemistry methods can also be utilized together with experimental spectroscopy
work to obtain a more detailed understanding of the various species present in a
CO2 absorption process.
On the issue of liquid structure and activity coefficients the present work is quite
limited. It should with simulations be possible to form a quite detailed
understanding of the liquid structure of amine-water systems. Such work can also
contribute to the determination of activity coefficients of various species.
Computers and the methods of computational chemistry are in continuous
development, and it is almost certain that the calculations of reaction energies
presented in this work can be improved upon.
The purpose of this work has been to contribute to the development of new
solvents. The next step should be to apply this model work in the selection of new
solvents for experimental work. New experimental data can also be used to validate
the modeling work, and offer guidance for further modeling.

8.2 Other Applications of Present Work


Determining free energies of solution and reactivity of organic molecules in
aqueous solution is a major topic in computational chemistry. My own work has to

113
8 Future Work

a large extent consisted of applying models in this area to the particular issue of
CO2 capture technology. The present work should therefore also be of interest as an
exercise in applying and validating models to calculate free energies in solution.
The present work has shown that computational chemistry tools can be
successfully applied to further the understanding of the CO2 absorption process.
While the work has been focused on the treatment of exhaust gases the work could
also be used to find optimal solvents for CO2 removal from natural gas. It is also
likely that the same methods as applied in the present work can be applied to study
other issues in gas processing.

8.3 Beyond Amines


My work has focused solely on absorbing CO2 with amine solvents. There are
however other chemical reactions resulting in CO2 being bonded, and some of
these reactions might perhaps be used to capture CO2. CO2 is captured in large
amounts in the natural photosynthesis process (Lawlor 2000). Another noteworthy
process is the catalyzed formation of CO2 from bicarbonate in the human body
(Palmer and van Eldik 1983). There are also other reactions where CO2 binds to
metal-complexes or organic compounds (Halmann 1993).
Many reagents or catalysts might turn out to be too expensive to be used in the
large scale capture of CO2. It would however seem worthwhile too carry out a more
general study of the various ways CO2 can interact with other molecules to see if
any mechanism can be taken advantage of in capture technology.

114
References

References

Aboudheir, A., Tontiwachwuthikul, P., Chakma, A. and Idem, R. (2003) Kinetics


of the reactive absorption of carbon dioxide in high CO2-loaded, concentrated
aqueous monoethanolamine solutions Chem. Eng. Sci. 58, 5195-5210.

ACIA (2004) Impacts of a Warming Arctic: Arctic Climate Impact Assessment.


Cambridge University Press.

Alejandro, J.; Rivera, J. L.; Mora, M. A.; de la Garza, V. J. (2000) Force Field of
Monoethanolamine Phys. Chem. B 104, 1332-1337.

Allen, M. P. and Tildesley, D. J. (1987) Computer Simulations of Liquids, Oxford


Science Publications, UK.

Andzelm, J. Kölmer, C. and Klamt, A. (1995) Incorporation of solvent effects into


density functional calculations of molecular energies and geometries J. Chem.
Phys. 103, 9312-9320.

Aroonwilas, A. and Veawab, A. (2004) Characterization and comparison of the


CO2 absorption performance into single and blended alkanolamines in a packed
column Ind. Eng. Chem. Res. 443, 2228-2237.

Bacskay, G. B. and Reimers, J. D. (1998) In Encyclopaedia of Computational


Chemistry, Schleyer, P. von R., Ed., Wiley.

Bayly, C. I., Cieplak, P., Cornell, W. D. and Kollman, P. A. (1993) A Well-


Behaved Electrostatic Potential Based Method Using Charge Resraints for
Deriving Atomic Charges: The RESP Model J. Phys. Chem. 1993, 97, 10269-
10280.

Ben-Naim, A. (1992) Statistical Thermodynamics for Chemists and Biochemists,


Plenum.

Ben-Naim, A. and Marcus, Y. (1984) Solvation thermodynamics of nonionic


solutes J. Chem. Phys. 81, 2016-2027.

115
References

Bishnoi, S. (2000) Carbon Dioxide Absorption and Solution Equlibrium in


Piperazine Activated Methyldiethanolamine, Dissertation, University of Texas at
Austin, USA.

Bishnoi, S. and Rochelle, G. T. (2002) Absorption of Carbon Dioxide in Aqueous


Piperazine/Methyldiethanolamine AICHE J. 48, 2788-2799.

Blanc, C., Grall, M. and Demarais G. (1982) The part played by degradation
products in the corrosion of gas sweetening plants using DEA and MDEA,
Laurence Reid Gas Conditioning Conference Proceedings, Univ. of Oklahoma,
Norman, USA.

Bolland, O. (2004) CO2 Capture Technologies-an Overview. The Second


Trondheim Conference on CO2 Capture, Transport and Storage.

Bonenfant, D., Mimeault, M. and Hausler, R. (2005) Comparative analysis of the


carbon dioxide absorption and recuperation capacities in aqueous 2-(2-
aminoethylamino)ethanol (AEE) and blends of aqueous AEE and N-
methyldiethanolamine solutions Ind. Eng. Chem. Res. 44, 3720-3725.

Button, J. K., Gubbins, K. E., Tanaka, H. and Nakanishi, K. (1996) Molecular


Dynamics Simulation of Hydrogen Bonding in Monoethanolamine Fluid Phase Eq.
116, 320-325.

Cances, M. T., Mennucci, V. and Tomasi, J. (1997) A new integral equation


formalism for the polarizable continuum model: Theoretical background and
applications to isotropic and anisotropic dielectrics Chem. Phys. 107, 3032-3041.

Caplow, M. (1968) Kinetics of Carbamate Formation and Breakdown J. Am. Chem.


Soc. 90, 6795-6803.

Car, R. and Parrinello, M. (1985) Unified Approach for Molecular Dynamics and
Density-Functional Theory Phys. Rev. Lett. 55, 2471-2474.

Chakraborty, A. B., Bischoff, K. B., Astarita, G. and Damewood, J. R. (1988)


Molecular Orbital Approach to Substituent Effects in Amine-CO2 Interactions J.
Am. Chem. Soc. 110, 6947-6954.

Chambers, C. C., Hawkins, G. D., Cramer, C.J. and Truhlar, D. G. (1996) Model
for aqueous solvation based on class IV atomic charges and first solvation shell
effects J. Phys. Chem. 100, 16385-16398.

116
References

Chandler, D. and Andersen, H. C. (1972) Optimized Cluster Expansions for


Classical Fluids. II Theory of Molecular Liquids J. Chem. Phys. 57, 1930-1937.

Chapman, W. G., Gubbins, K. E., Jackson, G and Radosz, M (1989) SAFT:


Equation-of-State Solution Model for Associating Fluids Fluid Phase Eq. 52, 31-
38.

Chapman, W. G., Gubbins, K. E., Jackson, G and Radosz, M (1990) New


Reference Equation of State for Associating Liquids Ind. Eng. Chem. Res. 29,
1709-1721.

Chi, S. and Rochelle, G. T. (2002) Oxidative Degradation of Monoethanolamine


Ind. Eng. Chem. Res. 41, 4178-4186.

Cornell, W. L., Cieplak, P., Bayly, C. I., Gould, I. R., Merz, K. M., Ferguson, D.
M., Spellmeyer, D. C., Fox, T., Caldwell, J. W. and Kollman, P. A. (1995) A
Second Generation Force Field for the Simulation of Proteins, Nucleic Acids, and
Organic Molecules J. Am. Chem. Soc. 117, 5179-5197.

Cossi, M. Barone, V. Cammi, R. and Tomasi, J. (1996) Ab initio study of solvated


molecules; a new implementation of the polarizable continuum model Chem. Phys.
Lett. 255, 327-335.

Cramer, C. C. (2002) Essentials of Computational Chemistry. John Wiley & Sons,


UK.

Cramer, C. J. and Truhlar, D. G. (1999) Implicit Solvation Models: Equilibria,


Structure, and Dynamics Chem. Rev. 99, 2161-2200.

Crooks, J.E. and Donnellan, J.P. (1989) Kinetics and Mechanism of the Reaction
between Carbon Dioxide and Amines in Aqueous Solution J. Chem. Soc. Perkins
Trans., II, 331-333.

Cummins, P. L. and Gready, J. E. (1997) Coupled Semiempirical Molecular Orbital


and Molecular Mechanics Model (QM/MM) for Organic Molecules in Aqueous
Soultion J. Comp. Chem. 18, 1496-1512.

Danckwerts, P. V. (1979) The Reaction of CO2 with Ethanolamine Chem. Eng. Sci.
34, 443-446.

da Silva, E. F. (2004) Use of Free Energy Simulations to predict Infinite Dilution


Activity Coefficients Fluid Phase Eq. 221, 15-24.

117
References

da Silva, E. F., Kuznetsova, T. and Kvamme, B. (2005) Molecular Dynamics Study


of Ethanolamine as a Pure Liquid and in Aqueous Solution.

da Silva, E. F. and Svendsen, H. , F. (2003) Prediction of the pKa Values of


Amines Using ab Initio Methods and Free Energy Perturbations Ind. Eng. Chem.
Res. 42 , 4414-4421.

da Silva, E. F. and Svendsen, H. , F. (2004a) Ab Initio study of the reaction of


carbamate formation from CO2 and alkanolamines Ind. Eng. Chem. Res. 43, 3413-
3418.

da Silva, E. F. and Svendsen, H. , F. (2004b) The Chemistry of CO2 Absorption in


Amine Solutions studied by Computational Chemistry. 7th International Conference
on Greenhouse Gas Control Technologies, Vancouver, Canada.

da Silva, E. F. and Svendsen, H. , F. (2005) Study of the Carbamate Stability of


Amines Using ab Initio Methods and Free-Energy Perturbations.

da Silva, E. F., Yamazaki, T. and Hirata, F. (2005) Comparison of Solvation


Models in the Calculation of Amine Basicity.

de Koeijer, G. (2004) CCP Technologies-The Norwegian Scenario. International


Test Network for CO2 capture: Report on 6th Workshop.

de Koeijer, G. and Solbraa, E. (2004) High Pressure Gas Sweetening with Amines
for Reducing CO2 Emissions. 7th International Conference on Greenhouse Gas
Control Technologies, Vancouver, Canada.

Donaldsen, T. L. and Nguyen, Y. N. (1980) Carbon Dioxide Reaction Kinetics and


Transport in Aqueous Amine Membranes Ind. Eng. Chem. Fundam. 19, 260-266.

Duffy, E. M. and Jorgensen, W. L. (2000) Prediction of Properties from


Simulations: Free Energies of Solvation in Hexadecane, Octanol and Water J. Am.
Chem. Soc. 122, 2878-2888.

Eckart, F. and Klamt, F. (2002) Fast Solvent Screening via Quantum Chemistry:
COSMO-RS Approach AIChE J. 48, 369-385.

Eimer, D. (1994) Simultaneous Removal of waater and hydrogen sulphide from


natural gas, Dissertation, Norwegian University of Science and Technology,
Norway.

118
References

Erga, O. Juliussen, O. and Lidal, H. (1995) Carbon Dioxide Recovery by Means of


Aqueous Amines Energy Convers. Mgmt. 36, 387-392.

European Chemicals Bureau, European Chemical Substances Information System


(accessed 2005) http://ecb.jrc.it/

Fischer, J. (1983) Remarks on molecular approaches to liquid mixtures Fluid Phase


Eq. 10, 1-7.

Franckl, M. M., Chirlian, L. (2000) In Reviews in Computational Chemistry


Lipkowitz, K. B., Boyd, D. B., Ed., Volume 14, Wiley-VCH, 1-31.

Fredenslund, A., Gmehling, J. and Rasmussen, P. (1977) Vapour-liquid


equilibrium using UNIFAC, Elsevier, Amsterdam, the Netherlands.

Frenkel, D. and Smit, B. (2002) Understanding molecular simulation, Academic


Press, USA.

Frisch, M. J., Trucks, G. W. , Schlegel, H. B. , Scuseria, G. E. , Robb, M. A.,


Cheeseman, J. R. Zakrzewski, V. G. Montgomery, Jr. J. A., Stratmann, R. E.
Burant, J. C. Dapprich, S. Millam, J. M. Daniels, A. D. Kudin, K. N. Strain, M. C.
Farkas, O. Tomasi, J. Barone, V. Cossi, M. Cammi, R. Mennucci, B. Pomelli, C.
Adamo, C. Clifford, S. Ochterski, J. Petersson, G. A. Ayala, P. Y. Cui, Q.
Morokuma, K. Malick, D. K. Rabuck, A. D. Raghavachari, K. Foresman, J. B.
Cioslowski, J. Ortiz, J. V. Baboul, A. G. Stefanov, B. B. Liu, G. Liashenko, A.
Piskorz, P. Komaromi, I. Gomperts, R. Martin, R. L. Fox, D. J. Keith, T. Al-
Laham, M. A. Peng, C. Y. Nanayakkara, A. Challacombe, M. Gill, P. M. W.
Johnson, B. Chen, W. Wong, M. W. Andres, J. L. Gonzalez, C. Head-Gordon, M.
Replogle, E. S. and Pople, J. A. (1998) Gaussian 98, Revision A.9, Gaussian, Inc.,
Pittsburgh PA.

Gao, J., (1996) In Reviews in Computational Chemistry Lipkowitz, K. B., Boyd, D.


B., Ed., Volume 7, Wiley-VCH, 119-185.

Goff, G. S. and Rochelle, G. T. (2004) Monoethanolamine Degredation: O2 Mass


Transfer Effects under CO2 Capture Conditions Ind. Eng. Chem. Res. 43, 6400-
6408.

Grant, G. H. and Richards, W. G. (1995) Computational Chemistry. Oxford


University Press, UK.

119
References

Gray, C. G. and Gubbins, K. E. (1984) Theory of Molecular Fluids, Clarendon


Press, Oxford, UK.

Gubskaya, A. V. and Kusilik, P. G. (2004a) Molecular Dynamics Simulation Study


of Ethylene Glycol, Ethylenediamine, and 2-amionethanol. 1. The Local Structure
in Pure Liquids J. Phys. Chem. A 108, 7151-7164.

Gubskaya, A. V. and Kusilik, P. G. (2004b) Molecular Dynamics Simulation Study


of Ethylene Glycol, Ethylenediamine, and 2-amionethanol. 2. Structure in Aqueous
Soultions J. Phys. Chem. A 108, 7165-7178.

Hakka, L. and Ouimet, M. A. (2004) Method for recovery of CO2 from gas
streams, US patent application 2004/0253159.

Halmann, M. M. (1993) Chemical Fixation of Carbon Dioxide. CRC Press, USA.

Hansen, J. P. and McDonald, I. R. (1990) Theory of Simple Liquids, 2nd ed.,


Academic Press, London, UK.

Haug, M. (2004) IEA World Energy Investment Outlook and the Prospects for
Carbon Capture and Storage. 7th International Conference on Greenhouse Gas
Control Technologies, Vancouver, Canada.

Hepple, R. P. and Benson, S. M. (2005) Geologic storage of carbon dioxide as a


climate change mitigation strategy: performance requirements and the implications
of surface seepage Environ. Geol. 47, 576-585.

Herzog H., Eliasson, B. and Kaarstad, O. (2000) Capturing greenhouse gases


Scientific American 282, 72-79.

Hirata, F. Ed. (2003) Molecular Theory of Solution, Kluwer Academic Publishers.

Hoff, K. A. (2003) Modeling and Experimental Study of Carbon Dioxide


Absorption in a Membrane Contactor, Dissertation, Norwegian University of
Science and Technology, Norway.

Hunter, E. P. and Lias, S. G. (2003) Proton Affinity Evaluation, National Institute


of Standards and Technology, Gaithersburg, MD, http://webbook.nist.gov.

International Panel on Climate Change (2001a) Climate Change 2001: The


Scientific Basis.

120
References

International Panel on Climate Change (2001b) Climate Change 2001: Impacts,


Adaptation and Vulnerability.

International Panel on Climate Change (2001c) Climate Change 2001: Mitigation.

Jamroz, M. H., Dobrowolski, J. and Borowiak, M. (1997) Ab initio study on the


1:2 reaction of CO2 with dimethylamine J. Mol. Struct. 404, 105-111.

Jenab, M. H., Abdi, M. A., Najibi, S. H., Vahidi, M. and Matin, N. S. (2005)
Solubility of Carbon Dioxide in Aqueous Mixtures of N-Methyldiethanolamine +
Piperazine + Sulfolane J. Chem. Eng. Data 50, 583-586.

Jones, F. M. and Arnett, E. M. (1974) Thermodynamics of Ionization and Solution


of Aliphatic Amines in Water Prog. Phys. Org. Chem. 11, 263-322.

Jónsdottir, S. O., Klein, R. A. and Rasmussen, K. (1996) UNIQUAC interaction


parameters for alkane/amine systems determined by Molecular Mechanics Fluid
Phase Eq. 115, 59-72.

Jorgensen, W. L., Maxwell, D. S. and Tirado-Rives, J. (1996) Development and


Testing of the OPLS All-Atom Force Field on Conformational Energetics and
Properties of Organic Liquids J. Am. Chem. Soc. 118, 11225-11236.

Jorgensen, W. L. and Ravimohan, C. (1985) Monte Carlo simulation of differences


in free energies of hydration J. Chem. Phys. 83, 3050-3054.

Jørgensen, E. and Faurholt, C. (1954) Reactions between Carbon Dioxide and


Amino Alcohols Acta Chem. Scand. 8, 1141-1144.

Kaminski, G. A. and Jorgensen, W. L. (1998) A Quantum Mechanical and


Molecular Mechanical Method Based on CM1A Charges. Applications to Solvent
Effects on Organic Equilibria and Reactions J. Phys. Chem. B 102, 1787-1796.

Kamps, A. P.-S. and Maurer, G. (1996) Dissociation Constant of N-


Methyldiethanolamine in Aqueous Solution at Temperatures from 278 K to 368 K
J. Chem. Eng. Data 41, 1505-1513.

Kawata, M., Ten-no, S., Kato, S. and Hirata, F. (1996) Theoretical study of the
basicities of methylamines in aqueous solution: A RISM-SCF calculation of
solvation Thermodynamics Chem. Phys. 203, 53-67.

121
References

Kjellander, R. (1992) The basis of statistical thermodynamics, University of


Göteborg, Sweden.

Klamt, A. (1995) Conductor-Like Screening Model for Real Solvents: A New


Approach to the Quantitative Calculation of Solvation Phenomena J. Phys. Chem.
99, 2224-2235.

Klamt, A., Jonas, V., Bürger, T. and Lohrenz, J. C. W. (1998) Refinement and
Parameterization of COSMO-RS J. Phys. Chem. A 102, 5074-5085.

Kofke, D. (2005) Free energy methods in molecular simulation Fluids Fluid Phase
Eq. 228-229, 41-48.

Kofke, D. A. and Cummings, P. T. (1997) Quantitative comparison and


optimization of methods for evaluating the chemical potential by molecular
simulation Mol. Phys. 92, 973-996.

Kofke, D. A. and Cummings, P. T. (1998) Precision and accuracy of staged free-


energy perturbation methods for computing the chemical potential by molecular
simulation Fluid Phase Eq. 150-151, 41-49.

Kohl, A. and Nielsen, R. (1997) Gas Purification, Gulf Publishing Company,


Houston, USA.

Kollman, P. (1993) Free Energy Calculations: Applications to Chemical and


Biochemical Phenomena Chem. Rev. 93, 2395-2417.

Kumar, P. S., Hogendoorn, J. A., Versteeg, G. F. and Feron, P. H. (2003) Kinetics


of the reaction of CO2 with aqueous potassium salt of taurine and glycine AIChE J.
49, 203-213.

Kuuskra, V., DiPietro, P., Klara, S. and Forbes, S. (2004) Future U.S. Greenhouse
Gas Emission Reduction Scenarios Consistent with Atmospheric Stabilization of
Concentrations. 7th International Conference on Greenhouse Gas Control
Technologies, Vancouver, Canada.

Kvamme, B. (2002) Thermodynamic properties and dielectric constants in


water/methanol mixtures by integral equation theory and molecular dynamics
simulations Phys. Chem. Chem. Phys. 4, 942-948.

122
References

Kvamsdal, H. M., Maurstad, O., Jordal, K. and Bolland, O. (2004) Benchmark of


gas-turbine cycles with CO2 capture. 7th International Conference on Greenhouse
Gas Control Technologies, Vancouver, Canada.

Lawlor, D. W. (2000) Photosynthesis, 3rd edition, BIOS Scientific Publishers Ltd,


Oxford, UK.

Li, J., Zhu, T., Cramer, C. J. and Truhlar, D. G. (1998) New Class IV Charge
Model for Extracting Accurate Partial Charges from Wave Functions J. Phys.
Chem. A 102, 1820-1831.

Li, J., Zhu, T., Hawkins, G. D., Winget, P., Daniel A. Liotard, D. A., Cramer, C. J.,
and Truhlar, D. G. (1999) Extension of the platform of applicability of the
SM5.42R universal solvation model Theor. Chem. Acc. 103, 9-63.

Lue, L. and Blankschtein, D. (1995) Application of integral equation theories to


predict the structure, thermodynamics and phase behavior of water J. Chem. Phys.
102, 5427-5437.

Ma’mun, S. Nilsen, R. and Svendsen, H. F (2005) Solubility of Carbon Dioxide in


30 mass% Monoethanolamine and 50 mass% Methyldiethanolamine Solutions J.
Chem. Eng. Data. 50, 630-634.

Ma’mun, S., Svendsen, H. F., Hoff, K. A. and Juliussen, O. (2004) Selection of


new absorbents for carbon dioxide capture. 7th International Conference on
Greenhouse Gas Control Technologies, Vancouver, Canada.

Miertus, S., Scrocco, E, and Tomasi, J (1981) Electrostatic Interaction of a Solute


with a Continuum. A direct Utilization of Ab Initio Molecular Potentials for the
Prevision of Solvent Effects Chem. Phys. 55, 117-129.

Mimura, T., Satsumi, S., Iijima, M. and Mitsuoka, S. (1999) Development on


Energy Saving Technology for Flue Gas CO2 Recovery by the Chemical
Absorption Method in Power Plant. Greenhouse Gas Control Technol., Proc Int
Conf., 4th 71.

Morrison, I. D. and Ross, S. (2002) Colloidal Dispersions. John Wiley & Sons,
New York, USA.

Nagao, Y., Hayakawa, A., Suzuki, H., Iwaki, T., Mimura, T. and Suda, T. (1998)
Comparative study of various amines for the reversible absorption capacity of
carbon dioxide Studies in Surf. Sci. and Catalysis 114, 669-672.

123
References

Nguyen, M. T., Raspoet, G., Vanquickenborn, L. G. and Van Duijnen, P. T. (1997)


How Many Water Molecules Are Actively Involved in the Neutral Hydration of
Carbon Dioxide J. Phys. Chem. 101, 7379-7388.

Ohno, K., Inoue, Y., Yoshida, H. and Matsuura, E. (1999) Reaction of aqueous 2-
(N-methylamino)ethanol solutions with carbon dioxide. Chemical species and their
conformations studied by vibrational spectroscopy and ab initio theories J. Phys.
Chem. A 103, 4283-4292.

Ohno, K., Matsumoto, H., Yoshida, H., Matsuura, H., Iwaki, T. and Suda, T.
(1998) Vibrational spectroscopic and ab initio studies on conformations of the
chemical species in a reaction of aqueous 2-(N,N-dimethylamino)ethanol solutions
with carbon dioxide. Importance of strong NH+center dot center dot center dot O
hydrogen bonding J. Phys. Chem. A 102, 8056-8062.

Onsager, L. (1936) Electric Moments of Molecules in Liquids J. Am. Chem. Soc.


58, 1486-1493.

Orozco, M. and Javier Luque, F. (2000) Theoretical Methods for the Description of
the Solvent Effect in Biomolecular Systems Chem. Rev. 100, 4187-4225.

Oscarson, J. L., Vandam R. H., Christensen, J. J. and Izatt, R. M. (1989) Enthalpies


Of Absorption Of Carbon-Dioxide In Aqueous Diethanolamine Solutions
Thermochimica Acta 146, 107-114.

Palmer, D. A. and van Eldik, R. (1983) The Chemistry of Metal Carbonato and
Carbon Dioxide Complexes Chem. Rev. 83, 651-731.

PC Spartan (1999) Version 1.0.7, Wavefunction, Inc., 18401 Von Karmen Ave.
#370 Irvine, CA 92715, USA.

Perrin, D. D. (1965, Supplement 1972) Dissociation Constants of Organic Bases in


Aqueous Solution. Butterworths, London.

Poplsteinova, J. (2004) Absorption of Carbon Dioxide-Modeling and Experimental


Characterization, Dissertation, Norwegian University of Science and Technology,
Norway.

Prausnitz, J., Lichtentaler, R. and Gomes de Azevedo, E. (1999) Molecular


Thermodynamics of Fluid-Phase Equilibria, 3rd ed., Prentice Hall.

124
References

Rao, A. B. and Rubin, E. S. (2002) A Technical, Economic, and Environmental


Assessment of Amine-Based CO2 Capture Technology for Power Plant
Greenhouse Gas Control Environ. Sci. Technol. 36, 4467-4475.

Reddy, S., Scherffius, J., Freguia, S. and Roberts, C. (2003) Fluor’s Econamine FG
PlusSM Technology. Second International Conference on Carbon Sequestration.

Rick, S. W. and Stuart, S. J. (2002) In Reviews in Computational Chemistry,


Lipkowitz, K. B.; Boyd, D. B.; Ed.; Volume 18, Wiley-VCH, 89-146.

Rizzo, R. C. and Jorgensen, W. L. (1999) OPLS All-Atom Model for Amines:


Resolution of the Amine Hydration Problem J. Am. Chem. Soc. 121, 4827-4836.

Sartori, G. and Savage, D. W. (1983) Sterically Hindered Amines for CO2


Removal from Gases Ind. Eng. Chem. Fundam. 22, 239-249.

Sharma, M. M. and Danckwerts, P. V. (1963) Catalysis by Brønsted bases of the


reaction between CO2 and water Trans. Faraday Soc., 59, 386-395.

Singh, U. C. and Kollman, P. A. (1984) An Approach to Computing Electrostatic


Charges for Molecules J. Comp. Chem. 5, 129-145.

Smith, D. R. and Quinn, J. A. (1979) The Prediction of Facilitation Factors for


Reaction Augmented Membrane Transport AIChE J. 25, 197-200.

Solbraa, E. (2002) Equilibrium and Non-Equilibrium Thermodynamics of Natural


Gas Processing, Dissertation, Norwegian University of Science and Technology,
Norway.

Suda, T., Zhang, Y., Iwaki, T. and Nomura, M. (1998) Correlation of the Frontier
Orbital Properties of Alkanolamines with the Experimental CO2 loading Chem.
Lett. 2, 189-190.

Sum, A. K. and Sandler, S. I. (1999) Use of ab initio methods to make phase


equilibria predictions activity coefficient models Fluid Phase Eq. 158-160, 375-
380.

Thambimuthu, K. and Davidson, J. (2004) Overview of CO2 Capture. 7th


International Conference on Greenhouse Gas Control Technologies, Vancouver,
Canada.

125
References

Tobiesen, F. A., Svendsen, H. F. and Hoff, K. A. (2005) Desorber Energy


Consumption in Amine Based Absorption Plants Int. J. Green Energy, in press.

Tomasi, J. and Persico, M. (1994) Molecular Interactions in Solution: An Overview


of Methods Based on Continuous Distributions of the Solvent Chem. Rev. 94,
2027-2094.

Versteeg, G.F., van Dijck, L.A.J. and van Swaaij, W.P.M. (1996) On the Kinetics
Between CO2 and Alkanolamines both in Aqueous and Non-Aqueous Solution. An
Overview Chem. Eng. Comm. 144, 113-158.

Versteeg, G.F., Kuipers, J. A. M., Van Beckum, F. P. H. and van Swaaij, W. P. M.


(1990) Mass Transfer with Complex Reversible Chemical Reactions-II. Parallel
Reversible Chemical Reactions Chem. Engng. Sci. 45, 183-197.

Vorholz, J., Harismiadis, V. I., Panagiotopoulos, A. Z., Rumpf, B. and Maurer, G.


(2004) Molecular Simulation of the Solubility of carbon dioxide in aqueous
solutions of sodium chloride Fluid Phase Eq. 226, 237-250.

Wang, J., Wang, W., Huo, S., Lee, M. and Kollman, P. A. (2001) Solvation Model
Based on Weighted Solvent Accessible Surface Area J. Phys. Chem. B 105, 5055-
5067.

Warshel, A. and M. Levitt, M. (1976) Theoretical studies of enzymic reactions:


Dielectric, electrostatic and steric stabilization of the carbonium ion in the reaction
of lysozyme J. Mol. Biol. 103, 227-249.

Winget, P., Hawkins, G. D., Cramer, C. J. and Truhlar, D. G. (2000) Prediction of


Vapor Pressures from Self-Solvation Free Energies Calculated by the SM5 Series
of Universal Solvation Models J. Phys. Chem. B 104, 4726-4734.

Wood, R. H., Yezzdimer, E. M., Sakane, S., Barriocanal, J. A. and Doren, D. J.


(1999) Free Energies of solvation with quantum mechanical interaction energies
from classical mechanical simulations J. Chem. Phys. 1110, 1329-1337.

Wu, H. S. and Sandler, S. I. (1991) Use of ab Initio Quantum Mechanics


Calculations in Group Contribution Methods. 1. Theory and the Basis for Group
Identifications. Ind. Eng. Chem. Res. 30, 881-889.

126
References

Yagi, Y., Mimura, T., Iijima, M., Ishida, K., Yoshiyama, R., Kamijo, T. and
Yonekawa, T. (2004) Improvements of Carbon Dioxide Capture Technology. 7th
International Conference on Greenhouse Gas Control Technologies, Vancouver,
Canada.

Xidos, J.D., Li, J., Zhu, T., Hawkins, G. D., Thompson, J. D., Chuang, Y.-Y., Fast,
P. L., Liotard, D. A., Rinaldi, D., Cramer, C. J. and Truhlar, D. G. Gamesol-version
3.1, University of Minnesota, Minneapolis (2002), based on the General Atomic
and Molecular Electronic Structure System (GAMESS) as described in Schmidt,
M. W., Baldridge, K. K., Boatz, J. A., Elbert, S. T., Gordon, M. S., Jensen, J. H.,
Koseki, S., Matsunaga, N., Nguyen, K. A., Su, S. J., Windus, T. L., Dupuis, M. and
Montgomery, J. A. (1993) J. Comp. Chem. 14, 1347.

Zhang, X., Wang, J., Zhang, C. F., Yang, Y. H. and Xu, J. J. (2003) Absorption
rate into a MDEA aqueous solution blended with piperazine under a high CO2
partial pressure Ind. Eng. Chem. Res. 42, 118-122.

127
Paper I

Prediction of the pKa Values of Amines Using ab Initio Methods and

Free Energy Perturbations

Eirik Falck da Silva and Hallvard F. Svendsen

2003

Ind. Eng. Chem. Res. 42, 4414-4421.


4414 Ind. Eng. Chem. Res. 2003, 42, 4414-4421

RESEARCH NOTES

Prediction of the pKa Values of Amines Using ab Initio Methods and


Free-Energy Perturbations
Eirik F. da Silva* and Hallvard F. Svendsen
Department of Chemical Engineering, Norwegian University of Science and Technology,
N-7491 Trondheim, Norway

A computational study has been performed on predicting the pKa values for amines and
alkanolamines used in CO2 absorption processes. Gas-phase energies were calculated using
common basis sets at Hartree-Fock (HF), MP2, and B3LYP levels. Free energies of solvation
were calculated using continuum models and Monte Carlo free-energy perturbations. Results
are compared with experimental pKa data. While the continuum methods could reproduce trends
between similar molecules, they failed to predict the overall trends for series involving amines
with different numbers of amine-group hydrogens and different numbers of intramolecular
hydrogen bonds. Considerably better results were obtained using free-energy perturbations. On
the basis of calculations of molecular vibrations, good estimates were also achieved for changes
of pKa with temperature.

Introduction Several studies have been published on the basicity


of methylamines.7-9 While these molecules are not
As a measure for preventing global warming, there directly relevant to the CO2 absorption process, they
is a steadily increasing interest in methods for removing clearly represent a closely related modeling task. Inclu-
carbon dioxide from exhaust gases as well as natural sion of these molecules in this work allows a more
gas and refinery gases. Traditionally, absorption with general validation of the models being used.
alkanolamines in mixtures with promoters has been
used for this purpose. For high-pressure applications, Methods
N-methyldiethanolamine (MDEA)-based systems have
been used successfully for many years. For exhaust The dissociation of the conjugate base of the amine
gases, the most common amine has been ethanolamine can be written as
(MEA). However, high energy demand for regeneration
and high degradation rates makes this an undesirable BH+ + H2O h B + H3O+ (1)
amine to use for large fossil fuel power plants. During
the last years, new systems such as the PSR1-31 and
When the mole fraction based activity of water is
KS1-32 have been developed and promise improved
assumed to be 1 and H3O+ is written as H+, the
performance compared to the conventional MEA. The
following equilibrium constant is obtained:
interactions between amino compounds and CO2 have
been studied extensively using both experimental and
theoretical methods.3 Ka ) aBaH+/aBH+ (2)
For both CO2 reaction rate and absorption capacity,
the pKa values of the conjugate acids of the alkanol- The definition of pKa is
amines applied are important variables.4 Reasonable
prediction of pKa values would therefore be of great pKa ) -log Ka (3)
value when screening for new candidate compounds for
exhaust gas CO2 removal. The free energy of protonation in an aqueous solution
A lot of work has been done in computational chem- (∆Gps) is related to Ka by the following equation:
istry on the prediction of pKa. While this is considered
to be a difficult task, works have been published that ∆Gps ) -2.303RT log Ka (4)
show good results for some sets of compounds.5,6 The
general applicability of these methods would, however, This gives us the relation between pKa and ∆Gps.
still seem to be uncertain. The intention of this work is
to explore the application of these models to the kinds 1
of molecules used for CO2 recovery. pKa ) ∆Gps (5)
2.303RT

* To whom correspondence should be addressed. Tel.: +47 Model predictions for ∆Gps should therefore give a linear
73594125. Fax: +47 73594080. E-mail: silva@chemeng.ntnu.no. correlation with the pKa.
10.1021/ie020808n CCC: $25.00 © 2003 American Chemical Society
Published on Web 08/23/2003
Ind. Eng. Chem. Res., Vol. 42, No. 19, 2003 4415

The general approach to calculating ∆Gps is to use a explicitly include hydrogen bonding.7,9 SM continuum
thermodynamic cycle such as8 models have also been tested in this work. Calculations
based on the explicit representation of the solvent using
free-energy perturbation (FEP) simulations have also
been performed.
A wide variety of PCM models are available, and
results have been published using a number of different
basis sets. There would, however, not seem to be any
clear guidelines as to whether some specific level of
calculation would be more appropriate or, in general,
more reliable. In this work, the IEFPCM model13 has
been chosen and all calculations were performed using
The focus of this work is to reproduce trends in pKa. its default settings in Gaussian 98 with 60 tesserea per
The energy of the proton (H+) itself is constant for all atomic sphere. These calculations will simply be referred
amines. It is therefore not relevant in this context and to as PCM calculations.
not included in the present calculations. On the basis Most calculations were done as single-point energy
of the thermodynamic cycle, ∆Gps can be divided into calculations on gas-phase geometry. Solvent-phase ge-
two contributions: ometry optimizations with the PCM models have also
been performed. While the geometries did change
∆Gps ) ∆Gpg + ∆Gs (6) slightly using solvent-phase optimization, the energy
changes were relatively small. For larger basis sets, it
where ∆Gpg is was also found that solvent-phase optimization resulted
in some convergence problems in the calculations. One
∆Gpg ) Gg(B) - Gg(BH+) (7) set of results using solvent-phase geometry optimization
has been included in the Results and Discussion section.
and The solvent-phase protonation energies (∆Gps) based
on continuum models presented in the Results and
∆Gs ) ∆Gs(B) - ∆Gs(BH+) (8) Discussion section were calculated with the solvation
energy calculated on the same geometry as that used
While the gas-phase protonation energy and the free to obtain the gas-phase energy. The B3LYP/6-311++G**
energy of solvation are expected to be the main con- gas-phase results were, however, added together with
tributors to the overall solvent-phase reaction energy, PCM results calculated at the B3LYP/3-21G* level.
thermal and zero-point energy (ZPE) corrections are also While the qualitative issues regarding gas-phase models
included. and basis sets are fairly well understood, the solvation
models are, in general, semiempirical, and it is not clear
Computational Aspects how their performance changes with the level of geom-
For the modeling of gas-phase protonation energies, etry optimization. In general, it would therefore seem
standard ab initio calculations have been used. Geom- best to consider solvation energy calculation and gas-
etry optimizations were performed using a series of phase energy calculation as separate processes. While
common methods and basis sets: HF/3-21G*, MP2/6- they should be calculated on the same conformer, it is
31G*, B3LYP/3-21G*, and B3LYP/6-311++G**. All not given that they should be calculated at the same
MP2 and B3LYP calculations were done using Gaussian level of geometry optimization.
98.10 The SM models are a series of semiempirical models
The thermal corrections to the free energy, the ZPEs, to calculate the free energy of solvation. The SM5.4A11
and the entropies were all calculated at the HF/3-21G* model is parametrized to work with AM1 geometries.
level. These contributions are relatively small and are In this work, this model has been used to calculate
not expected to change significantly with the level of solvation energies using AM1 geometries in the Spartan
modeling. To be consistent with how the gas-phase program. In addition, SM5.42R/HF/6-31G* calculations
energy is calculated, the ZPE and thermal corrections were performed with Gamesol 3.1.15
are calculated as the same relative difference as that
used for the protonation energy itself (eq 7). These Simulations
calculations were also done using Gaussian 98.
Molecular mechanics (MMFF) was used to generate As an alternative to the implicit solvent models (PCM
an initial set of conformers. All conformers were then and SM), classical Monte Carlo calculations were per-
optimized at the HF/3-21G* level. Separate conformer formed. Wiberg et al.16 have obtained good results
searches were done for the amines and their protonated combining gas-phase ab initio calculations with Monte
forms. The effect of the solvation models on the relative Carlo calculations of the solvation energy to obtain free
conformer stability was explored for some of the main energy in solvation. In the present work a similar
conformers. All calculations in this work were performed procedure has been adopted. Wiberg et al.16 used
on the same set of conformers. Gas-phase conformer perturbations between the neutral forms of the mol-
search calculations were done with the Spartan pro- ecules and between the ionic forms of the molecules; i.e.,
gram.12 they only performed perturbations between molecules
For the calculation of solvation energies, several with the same charge. The molecules in the present
models are available. The continuum models such as work do, however, vary considerably in size and struc-
the PCM models are probably the most common ones,13 ture, making perturbations between them difficult.
but there are questions regarding their general ap- Instead, the FEP calculations were done by perturbating
plicability. Of particular importance is their failure to the protonated form of the amine to the amine itself,
4416 Ind. Eng. Chem. Res., Vol. 42, No. 19, 2003

Table 1. Experimental pKa Data


no. compd name typea hydrogen bondsb exptl pKa at 25 °C ref
1 NH3 0 9.3 20
2 NH2(CH3) p 0 10.657 21
3 NH(CH3)2 s 0 10.732 21
4 N(CH3)3 t 0 9.9 20
5 ethanolamine MEA p 1 9.51 21
6 1-amino-2-propanol MIPA p 1 9.47 21
7 3-amino-1-propanol MPA p 1 9.96 21
8 2-amino-2-methylpropanol AMP p 1 9.7 21
9 N-(2-hydroxyethyl)ethylenediamine AEEA p, s 1 9.82c 21
10 diethanolamine DEA s 2 8.95 21
11 diisopropanolamine DIPA s 2 8.89 21
12 morpholine s, c 0 8.7 22
13 diethylenediamine piperazine s, c 0 9.83c 21
14 N-n-butylethanolamine BEA s 1 9.9 23
15 N-methyldiethanolamine MDEA t 2 8.63 21
16 triethanolamine TEA t 2 7.78 21
a p: primary amine. s: secondary amine. t: tertiary amine. c: cyclical amine. b Number of intramolecular hydrogen bonds for the

protonated form of the amine. c The first protonation constant.

Table 2. Gas-Phase Protonation Energies (All Results in kcal/mol)


B3LYP/3-21G* MP2/6-31G* B3LYP/6-311++G**
compd ∆Gpga rel ∆Gpg ∆Gpga rel ∆Gpg ∆Gpga rel ∆Gpg rel ∆GExptlb
NH3 229.89 0.00 217.17 0.00 211.87 0.00 0.00
NH2(CH3) 236.08 6.19 225.64 8.47 221.55 9.69 10.87
NH(CH3)2 241.26 11.38 232.29 15.12 229.29 17.43 18.52
N(CH3)3 244.80 14.92 236.42 19.25 234.23 22.37 23.69
a Thermal correction and ZPE calculated at the HF/3-21G* level included the value relative to that of NH (-9.38 kcal/mol).
3
b Experimental data from Hunter and Lias.24

giving the relative difference in free energy of solvation models were used on HF/6-31G* gas-phase geometries
(∆Gs) directly. and basis sets. For the CM2 scheme, calculations were
These calculations were done with BOSS version 4.117 done with the SM5.42R solvent field, while for the MK
using procedures developed by Rizzo and Jorgensen.9 scheme, the PCM solvent field was used.
A single solute molecule was placed in a periodic cube It should be noted that these simulations have a
with 267 TIP4P water molecules at 25 °C and 1 atm in statistical uncertainty, unlike the continuum models
the NPT ensemble. Periodic boundary conditions were that are deterministic. On the basis of the batch means
applied. A number of water molecules corresponding to procedure available in BOSS, the statistical uncertain-
the number (n) of non-hydrogen atoms in the amine ties were estimated to be on the order of (2 kcal/mol.
molecule were removed, giving 267 - n water molecules. This uncertainty is something that must be kept in mind
The perturbations were carried out over five windows when comparing results obtained by FEP and con-
of double-wide sampling, giving 10 free-energy incre- tinuum models.
ments that are summed up to give the total change in
free energy of solvation. Each window had 500 000 steps Results and Discussion
for equilibration and another 500 000 for sampling.
Wiberg et al.16 used solution-phase B3P86/6-311+G** In Table 1 are shown the amines used in this study,
geometries. In the present work, similar solution-phase and in the table, it is indicated whether the amines are
B3LYP/6-311++G** geometries (optimized with PCM) primary (p), secondary (s), tertiary (t), or cyclic (c).
were used. The same conformers as those for the Experimental pKa values at 25 °C are also shown. The
continuum calculations were used, and simulations were quality of the experimental data would seem to vary
performed using rigid geometries. The intermolecular somewhat. For some molecules, different experimental
interactions between two molecules a and b were results are available and variations between these
evaluated using Coulomb and Lennard-Jones terms: suggest uncertainties in the order of (0.1 pKa unit.

{ [( ) ( ) ]}
The series NH3, NH2CH3, NH(CH3)2, and N(CH3)3 has
on aon b qiqje2 σij 12 σij 6 been the subject of several previous studies.7-9 In Table
∆Eab ) ∑i ∑j rij
+ 4ij
rij
-
rij
(9) 2, the gas-phase reaction energies for these molecules
are shown together with experimental values. The
B3LYP/6-311++G** results in particular are in good
The Lennard-Jones σ and  were taken from the OPLS- agreement with the experimental data.
All atom force field.17 Solvation energies calculated using both continuum
Wiberg et al.16 calculated atomic charges in the gas- models and FEPs are shown in Table 3. There is
phase and scaled these by a factor of 1.2 to obtain values reasonable qualitative agreement between both the gas-
appropriate for solution. In the present work, methods phase energies and the solvation energies for all of the
were chosen that directly give solution-phase atomic sets of results. In Table 4 are shown relative free
charges. Two different methods were tested for calculat- energies of protonation in the solvent (∆Gps). The
ing the charges. One was the CM2 model18 found in experimental data are based on eq 5 and data from
Gamesol, while the other method used was the Merz- Table 1. The errors in the prediction of the solvent-phase
Kollmann (MK) scheme19 available in Gaussian. Both protonation energy are larger than the errors in the gas-
Ind. Eng. Chem. Res., Vol. 42, No. 19, 2003 4417

Table 3. Solvation Energies (All Results in kcal/mol) For the neutral amines, the most stable conformers
compd PCM/MP2a PCM/B3Lb PCM/B3Lsc FEP-CM2 FEP-MK were also found to have different forms of hydrogen
bonding. Hydrogen-bonded gauche-MEA was, for ex-
NH3 75.6 76.1 76.0 76.4 74.1
NH2(CH3) 65.7 65.9 66.0 67.1 61.7
ample, found to be 4 kcal/mol more stable than the trans
NH(CH3)2 61.5 60.9 61.3 58.1 60.2 conformer (at the HF/3-21G* level).
N(CH3)3 56.8 56.0 56.2 50.6 55.4 These conformer searches have been based on gas-
a PCM/MP2/6-31G*//MP2/6-31G*. b PCM/B3LYP/3-21G*//B3LYP/ phase calculations. The intramolecular hydrogen bonds
3-21G*. c Optimization in solution: PCM/B3LYP/3-21G*//PCM/ found to dominate in the gas phase will in the solvent
B3LYP/3-21G*. compete with hydrogen bonding to water molecules. The
calculations done with solvent models (PCM and SM5.4A)
phase energies, suggesting that the solvation energies
do, however, suggest that these intramolecular hydro-
are not completely accurate.
gen bonds are favored even in the solvent, particularly
We would argue that this series of four molecules is
for the protonated forms of the amines. Molecules 9-11
too small to draw any firm conclusion as to the ap-
and 14-16 have many free dihedral angles and have
plicability of any single model. The energy differences
many potential conformers. For these molecules in
that one is attempting to reproduce are also relatively
particular, we see that the most stable conformer can
small. The FEP calculations used have an uncertainty
change with the level of modeling; the energy differences
that is too large ((2 kcal/mol) to draw conclusions for
between the conformers are, however, expected to be
this small series.
relatively small because the effects of intermolecular
Gas-phase conformer searches were done at the HF/
hydrogen bonds and intramolecular hydrogen bonds
3-21G* level and calculations on some of the more stable
cancel out.
conformers were also done with the other models used.
For the unprotonated forms of diisopropanolamine In Table 5, gas-phase energies for molecules 5-16 are
(DIPA), diethanolamine (DEA), MDEA, and N-n-butyl- shown together with experimental values. The results
ethanolamine (BEA), the most stable conformers did from the different calculations are in reasonable quali-
change with the level of modeling. In these cases, the tative agreement. When compared with the experimen-
most stable conformers found at higher level have been tal energies, the B3LYP/6-311++G** results are in good
used. The most stable conformers for molecules 5-16 quantitative agreement for 3-amino-1-propanol (MPA),
are shown in Figure 1. The protonated forms of the morpholine, and piperazine. For DEA, the protonation
amines were found to have a strong tendency to form energies are, however, overestimated by 4 kcal/mol,
hydrogen bonds between amine hydrogens and oxygen suggesting that the strength of the intramolecular
atoms on the alcohol groups. For MDEA, the most stable hydrogen bonds for the protonated form of the molecule
conformer without hydrogen bonds was 5 kcal/mol less are overestimated. For MEA, all models show too low
stable than the most stable hydrogen-bonded conformer protonation energy, suggesting that the strength of the
(at the HF/3-21G* level). For some of the smaller hydrogen bond of the protonated form of the molecule
molecules such as MEA, only a hydrogen-bonded con- in this case is underestimated. The uncertainty in the
former was found. In Table 1, the number of such bonds estimation of the hydrogen bond strength was also seen
found for each amine in the protonated form is indi- in the optimized geometries: BEA had an intramolecu-
cated. lar hydrogen bond [H(O)-N] varying between 2.0 Å
On the basis of the results of the conformer searches, (B3LYP/3-21G*) and 2.3 Å (B3LYP/6-311++G**).
a general qualitative explanation for the trends seen in No direct correlation was found between the gas-
the experimental pKa data can be suggested. The alcohol phase protonation energy (∆Gpg) and the pKa, and this
groups in the molecules are electron-withdrawing, strongly suggests that the solvation energy is crucial
destabilizing the protonated form of the amine. This for predicting the relative pKa for these compounds. In
effect is mitigated by the alcohol groups forming hydro- Table 6, the free energies of solvation calculated with
gen bonds to the amine protons. Most of the alkanol- various models are shown. The different PCM results
amines therefore have pKa values slightly lower than can be seen to be quite similar; in particular, it can be
those of methylamines of the same order. Morpholine seen that the effect of solvent-phase optimization is
has an electron-withdrawing oxygen but cannot form small. The FEP results are, however, quite different;
hydrogen bonds and has therefore a relatively low pKa. while they show some qualitative agreement with the
Triethanolamine (TEA) has three ethanol groups that PCM results, the relative differences between the vari-
together have a strong electron-withdrawing effect. In ous molecules is much larger than those calculated with
the protonated form, there is, however, only one amine PCM.
proton that these groups can bond with, limiting the Table 7 shows calculated and experimental solvent-
stabilizing effect of the hydrogen bonding. One might phase protonation energies. Experimental data are
think of the hydrogen-bonding effect as reaching a form based on eq 5 and data in Table 1. The overall quality
of saturation, giving a significant drop in the pKa. of the results with continuum models is quite disap-
Table 4. Relative Solvent-Phase Protonation Energies (All Results in kcal/mol)
gas-phasea
MP2/6-31G* B3LYP/6-311++G**
compd PCM/MP2b FEP-CM2 FEP-MK PCM/B3Lc FEP-CM2 FEP-MK exptl
NH3 0.00 0.00 0.00 0.00 0.00 0.00 0.00
NH2(CH3) -1.47 -0.77 -4.00 -0.53 0.44 -2.78 1.85
NH(CH3)2 1.07 -3.15 1.21 2.14 -0.84 3.52 1.95
N(CH3)3 0.44 -6.49 0.55 2.21 -3.37 3.67 0.82
a Thermal correction and ZPE calculated at the HF/3-21G* level included the value relative to that of NH (-9.38 kcal/mol). b PCM/
3
MP2/6-31G*//MP2/6-31G*. c PCM/B3LYP/3-21G*//B3LYP/3-21G*.
4418 Ind. Eng. Chem. Res., Vol. 42, No. 19, 2003

Figure 1. Conformers of molecules 5-16.

Table 5. Gas-Phase Protonation Energies (All Results in kcal/mol)


B3LYP/3-21G* MP2/6-31G* B3LYP/6-311++G**
no. compd ∆Gpga rel ∆Gpg ∆Gpga rel ∆Gpg ∆Gpga rel ∆Gpg rel ∆Gexptlb
5 MEA 245.43 15.55 231.96 14.79 227.55 15.69 18.59
6 MIPA 247.30 17.41 233.42 16.25 229.27 17.40
7 MPA 254.02 24.14 238.44 21.27 234.96 23.09 23.49
8 AMP 254.33 24.45 239.02 21.85 234.56 22.69
9 AEEA 264.68 34.80 246.83 29.67 240.43 28.56
10 DEA 256.53 26.65 246.73 29.56 240.25 28.39 24.14
11 DIPA 257.50 27.62 245.48 28.31 243.17 31.31
12 morpholine 240.69 10.80 231.23 14.07 228.52 16.65 17.26
13 piperazine 247.71 17.83 237.71 20.54 234.76 22.89 22.87
14 BEA 252.90 23.01 241.45 24.28 238.98 27.11
15 MDEA 257.85 27.96 249.06 31.89 243.61 31.74
16 TEA 243.54 13.65 241.89 24.72 244.92 33.05
a Thermal correction and ZPE calculated at the HF/3-21G* level included the value relative to that of NH (-9.38 kcal/mol).
3
b Experimental data from Hunter and Lias.24

pointing. At the B3LYP/6-311++G** level, one can, for mental value (TEA) in the data set has the highest value
example, see that the molecule with the lowest experi- in the calculated results.
Ind. Eng. Chem. Res., Vol. 42, No. 19, 2003 4419

Table 6. Solvation Energies (All Results in kcal/mol)


no. compd PCM/MP2a PCM/B3Lb PCM/B3Lsc FEP-CM2
5 MEA 60.5 58.2 58.9 56.7
6 MIPA 59.3 56.4 57.4 60.1
7 MPA 55.5 51.5 52.2 57.9
8 AMP 53.8 51.7 52.3 54.7
9 AEEA 52.1 48.2 50.0 44.9
10 DEA 51.4 47.9 49.8 43.1
11 DIPA 49.0 46.7 47.5 42.4
12 morpholine 63.0 62.1 62.7 54.7
13 piperazine 59.8 58.0 58.1 50.3
14 BEA 53.6 51.2 52.3 47.4
15 MDEA 47.6 46.9 47.8 38.1
16 TEA 51.0 49.9 50.2 36.6
a PCM/MP2/6-31G*//MP2/6-31G*. b PCM/B3LYP/3-21G*//B3LYP/

3-21G*. c Optimization in solution: PCM/B3LYP/3-21G*//PCM/


B3LYP/3-21G*.

As noted previously, this data set contains molecules


with very differing bonding. Given the uncertainty in Figure 2. Calculated protonation energies in solution versus
the estimation of hydrogen bonds already noted and the experimental pKa values.
uncertainty regarding the PCM model’s ability to ac-
count for hydrogen bonding with the solvent, it is of solvation energy gave no overall correlation. In Table
interest to look at how the models perform in predicting 7, however, it can be seen that the results with FEP
the relative pKa values of molecules with similar solvation energies have larger relative differences than
hydrogen bonding. To have similar bonding, the mol- those seen in the experimental data set.
ecules must have the same number of hydrogens on the While the B3LYP/6-311++G** gas-phase energies, in
nitrogen and the same number of intramolecular hy- general, seem to be fairly accurate, there is an uncer-
drogen bonds. All of the primary alkanolamines (mol- tainty in the estimation of the strength of intramolecu-
ecules 5-8) have such a similarity, and so do the two lar hydrogen bonds. Given the uncertainty in the gas-
cyclic molecules (12 and 13) and DEA and DIPA phase energies, we cannot draw any firm conclusions
(molecules 10 and 11). These groups are marked in as to the performance of the models used to calculate
Table 7. The other PCM results are also reasonably the solvation energies. The present work does, how-
good. The order between piperazine and morpholine is ever, suggest that FEP simulations are a promising
also predicted correctly by all models. For the order tool for the solvation energy calculations for these
between DEA and DIPA, all of the models are wrong; alkanolamines. More accurate free-energy calculations
the B3LYP/6-311++G** results are, for example, off by will, however, be needed to better assess the perfor-
about 2 kcal/mol. mance of these simulations.
Calculations with the SM continuum models, in
general, gave results similar to those obtained with the Temperature Effects
PCM model.
Ab initio calculations can be used to calculate the
Using FEP solvation energies gave a better overall
molecular vibration frequencies and thereby obtain gas-
trend in the results. For the full set of 16 molecules,
phase entropies. With an estimate of the entropy, the
the FEP-CM2 results in combination with MP2/6-31G*
temperature changes in the pKa can be predicted by
gas-phase energies gave an overall correlation coef-
using the following equation:25
ficient of 0.45; in combination with B3LYP/6-311++G**
energies, a correlation coefficient of 0.42 was obtained.
In Figure 2, this FEP set of results is shown together -d(pKa)/dT ) (pKa + 0.052∆S0)/T (10)
with the best set of results for the PCM model. The same
gas-phase energy (MP2/6-31G*) together with the PCM In this work the vibration frequency calculations were

Table 7. Relative Solvent-Phase Protonation Energies (All Results in kcal/mol)


gas-phasea
B3LYP/3-21G* MP2/6-31G* B3LYP/6-311++G**
no. compd PCM/B3Lb FEP-CM2 PCM/MP2c FEP-CM2 PCM/B3Lc FEP-CM2 exptl
5 MEA -2.76 -4.44 -0.67 -5.20 4.25 2.57 0.27
6 MIPA -2.53 0.91 -0.28 -0.25 -2.30 1.14 0.22
7 MPA -0.72 5.48 0.97 2.62 -1.58 4.62 0.90
8 AMP -0.87 1.84 -0.86 -0.76 -1.74 0.98 0.57
9 AEEA 5.51 1.97 4.75 -3.16 0.63 -2.91 0.71
10 DEA 0.11 -4.87 2.52 -6.49 0.14 -4.84 -0.46
11 DIPA 1.64 -2.92 2.73 -4.61 1.90 -2.66 -0.56
12 morpholine -3.63 -11.29 1.04 -8.03 2.64 -5.02 -0.82
13 piperazine -1.00 -8.89 4.10 -6.18 4.71 -3.18 0.72
14 BEA -1.67 -5.68 2.32 -4.60 2.16 -1.86 0.82
15 MDEA -0.06 -9.06 1.77 -8.42 3.87 -5.13 -1.06
16 TEA -9.98 -23.50 2.74 -12.43 6.80 -6.72 -2.10
a Thermal correction and ZPE calculated at the HF/3-21G* level included the value relative to that of NH (-9.38 kcal/mol). b PCM/
3
B3LYP/3-21G*//B3LYP/3-21G*. c PCM/MP2/6-31G*//MP2/6-31G*.
4420 Ind. Eng. Chem. Res., Vol. 42, No. 19, 2003

Table 8. Experimental pKa Data and Calculated Entropies (Data Given in cal/mol)
compd exptl pKa(20 °C)a exptl pKa(60 °C)a ∆S(20 °C)b ∆S(60 °C)b
MDEA 8.76 7.99 -3.156 -4.045
DMMEA [2-(dimethylamino)ethanol] 9.23 8.36 -0.593 -0.699
DGA [2-(2-aminoethoxy)ethanol] 9.62 8.60 1.046 0.961
DEMEA [2-(diethylamino)ethanol] 9.76 8.71 0.336 0.247
AMP 9.88 8.78 0.844 0.07
MMEA [2-(methylamino)ethanol] 9.95 8.94 -0.416 -0.501
DIPMEA [2-(diisopropylamino)ethanol] 10.14 9.13 0.330 0.269
TBAE [2-(tert-butylamino)ethanol] 10.29 9.28 -0.429 -0.521
a Experimental data from Littel et al.26 b Reaction entropies calculated at the HF/3-21G* level.

number of amine hydrogens and the same number of


intramolecular hydrogen bonds. The full set of amines
was, however, not successfully modeled with any of the
continuum models tested. Reasonable results were
obtained using FEPs to calculate the solvation energy.
It was also found that gas-phase entropy change calcu-
lations could be used to expand the data to temperatures
other than those experimentally available. We propose
that the differences in temperature dependency of
different amines can be attributed to differences in
intramolecular hydrogen bonding.

Literature Cited
(1) Chakma, A.; Tontiwachiwuthikul, P. Designer Solvents for
Efficient CO2 Separation from Flue Gas Streams. Greenhouse Gas
Control Technol., Proc. Int. Conf., 4th 1999, 35.
(2) Mimura, T.; Satsumi, S.; Iijima, M.; Mitsuoka, S. Develop-
Figure 3. Plot of experimental versus calculated pKa values at ment on Energy Saving Technology for Flue Gas CO2 Recovery
60 °C. by the Chemical Absorption Method in Power Plant. Greenhouse
Gas Control Technol., Proc. Int. Conf., 4th 1999, 71.
(3) Jamroz, M. H.; Dobrowolski, J.; Borowiak, M. Ab initio study
on the 1:2 reaction of CO2 with dimethylamine. J. Mol. Struct.
done at the HF/3-21G* level at 20 and 60 °C using the 1997, 404, 105.
Gaussian program. The same conformer search routine (4) Versteeg, G. F.; van Dijck, L. A. J.; van Swaaij, W. P. M.
as that described previously was used for these alkanol- On the kinetics between CO2 and alkanolamines both in aqueous
amines. To get a reasonable estimate of the total and nonaqueous solutions. An overview. Chem. Eng. Commun.
entropy, we calculated the entropy of reaction given by 1996, 144, 113.
eq 1 including the values for H2O and H3O+. Calculated (5) Liptak, M.; Shields, G. Accurate pKa calculations for car-
entropies together with the experimental pKa values boxylic acids using complete basis set and Gaussian-n Models
combined with CPMC continuum solvation methods. J. Am. Chem.
used are given in Table 8. The entropies were assumed Soc. 2001, 123, 7314.
to vary linearly with temperature, and eq 10 was solved (6) Schuurmann, G.; Cossi, M.; Barone, V.; Tomasi, J. Predic-
numerically, doing a stepwise calculation from 20 to tion of the pKa of carboxylic acids using the ab initio continuum-
60 °C. Figure 3 shows these estimated values at 60 °C solvation model PCM-UAHF. J. Phys. Chem. A 1998, 102,
plotted against experimental pKa values at 60 °C. 6706.
While the agreement is very good, it should be noted (7) Kawata, M.; Ten-no, S.; Kato, S.; Hirata, F. Theoretical
study for the basicities of methylamines in aqueous solution: A
that using eq 10 even without the entropy contribution RISM-SCF calculation of solvation thermodynamics. Chem. Phys.
also gives good results. The entropy term does, however, 1996, 203, 53.
improve the correlation. We believe that molecules that (8) Tunon, I.; Silla, E.; Tomasi, J. Methylamines Basicity
depend on intramolecular hydrogen bonding for their Calculations. In Vacuo and in Soultion Comparative Analysis. J.
high pKa are more sensitive to temperature variations. Phys. Chem. 1992, 96, 9043.
The data are, however, too limited to draw any firm (9) Rizzo, R. C.; Jorgensen, W. L. OPLS All-Atom Model for
Amines: Resolution of the Amine Hydration Problem. J. Am.
conclusion.
Chem. Soc. 1999, 121, 4827.
We see that the estimated pKa values are systemati- (10) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G.
cally lower than the experimental ones. This can be E.; Robb, M. A.; Cheeseman, J. R.; Zakrzewski, V. G.; Montgomery,
attributed to systematic errors in the entropy estimation J. A., Jr.; Stratmann, R. E.; Burant, J. C.; Dapprich, S.; Millam,
and/or to changes in the solvent with temperature that J. M.; Daniels, A. D.; Kudin, K. N.; Strain, M. C.; Farkas, O.;
are not accounted for in the model. Tomasi, J.; Barone, V.; Cossi, M.; Cammi, R.; Mennucci, B.;
Pomelli, C.; Adamo, C.; Clifford, S.; Ochterski, J.; Petersson, G.
A.; Ayala, P. Y.; Cui, Q.; Morokuma, K.; Malick, D. K.; Rabuck, A.
Conclusions D.; Raghavachari, K.; Foresman, J. B.; Cioslowski, J.; Ortiz, J.
V.; Baboul, A. G.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz,
Calculations have shown that intramolecular hydro- P.; Komaromi, I.; Gomperts, R.; Martin, R. L.; Fox, D. J.; Keith,
gen bonds play a crucial part in determining the pKa T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Challacombe,
values of alkanolamines. These same hydrogen bonds M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong, M. W.; Andres,
J. L.; Gonzalez, C.; Head-Gordon, M.; Replogle, E. S.; Pople, J. A.
make the accurate modeling of these molecules difficult. Gaussian 98, revision A.9; Gaussian, Inc.: Pittsburgh, PA, 1998.
Continuum solvent models were found to be useful in (11) Chambers, C. C.; Hawkins, G. D.; Cramer, C. J.; Truhlar,
predicting the relative pKa strengths between amines D. G. Model for aqueous solvation based on class IV atomic charges
with similar solvation behavior, i.e., with the same and first solvation shell effects. J. Phys. Chem. 1996, 100, 16385.
Ind. Eng. Chem. Res., Vol. 42, No. 19, 2003 4421

(12) PC SPARTAN, version 1.0.7; Wavefunction, Inc., 18401 (19) Bash, P. A.; Singh, U. C.; Langridge, R.; Kollman, P. A.
Von Karmen Ave. #370, Irvine, CA 92715. Free Energy Calculations by Computer Simulation. Science 1987,
(13) Tomasi, J.; Persico, M. Molecular Interactions in Solu- 236, 564.
tion: An overview of methods Based on Continuous Distributions (20) Pearson, R. G. Ionization Potentials and Electron Affinities
of the solvent. Chem. Rev. 1994, 94, 2027. in Aqueous Solution. J. Am. Chem. Soc. 1986, 108, 6109.
(14) Cances, M. T.; Mennucci, V.; Tomasi, J. A new integral (21) Perrin, D. D. Dissociation Constants of Organic Bases in
equation formalism for the polarizable continuum model: Theo- Aqueous Solution; Butterworths: London, 1965; Supplement,
retical background and applications to isotropic and anisotropic 1972.
dielectrics. Chem. Phys. 1997, 107, 3032. (22) Bishnoi, S. Dissertation, University of Texas, Austin, TX,
(15) Xidos, J. D.; Li, J.; Zhu, T.; Hawkins, G. D.; Thompson, J. 2000; p 100.
D.; Chuang, Y.-Y.; Fast, P. L.; Liotard, D. A.; Rinaldi, D.; Cramer, (23) Hoff, K. A. Unpublished results from work at the Depart-
C. J.; Truhlar, D. G. Gamesol, version 3.1, University of Minnesota, ment of Chemical Engineering, NTNU.
Minneapolis, MN, 2002, based on the General Atomic and Molec- (24) Hunter, E. P.; Lias, S. G. Proton Affinity Evaluation;
ular Electronic Structure System (GAMESS) as described in: National Institute of Standards and Technology, Gaithersburg,
Schmidt, M. W.; Baldridge, K. K.; Boatz, J. A.; Elbert, S. T.; MD, 2003; http://webbook.nist.gov.
Gordon, M. S.; Jensen, J. H.; Koseki, S.; Matsunaga, N.; Nguyen, (25) Perrin, D. D.; Dempsey, B.; Sejeant, E. P. pKa prediction
K. A.; Su, S. J.; Windus, T. L.; Dupuis, M.; Montgomery, J. A. J. for organic acids and bases; Chapman and Hall: London, 1981.
Comput. Chem. 1993, 14, 1347. (26) Littel, R. J.; Bos, M.; Knoop, G. J. Dissociation constants
(16) Wiberg, K. B.; Clifford, S.; Jorgensen, W. L.; Frisch, M. J. of some alkanolamines at 293, 303, 318, and 333 K. J. Chem. Eng.
Origin of the Inversion of the Acidity Order for Haloacetic Acids Data 1990, 35, 276.
on going from the Gas Phase to Solution. J. Phys. Chem. A 2000,
104, 7625. Received for review October 16, 2002
(17) Jorgensen, W. L. BOSS, version 4.3; Yale University, New Revised manuscript received July 29, 2003
Haven, CT, 1989. Accepted August 5, 2003
(18) Li, J.; Zhu, T.; Cramer, C. J.; Truhlar, D. G. A new Class
IV Charge Model for Extracting Accurate Partial Charges from
Wawe Functions. J. Phys. Chem. A 1998, 102, 1820. IE020808N
Paper II

Ab Initio study of the reaction of carbamate formation from CO2 and

alkanolamines

Eirik Falck da Silva and Hallvard F. Svendsen

2004

Ind. Eng. Chem. Res. 43, 3413-3418.


Ind. Eng. Chem. Res. 2004, 43, 3413-3418 3413

GENERAL RESEARCH

Ab Initio Study of the Reaction of Carbamate Formation from CO2


and Alkanolamines
Eirik F. da Silva* and Hallvard F. Svendsen

Department of Chemical Engineering, Norwegian University of Science and Technology,


N-7491 Trondheim, Norway

Ab initio calculations and a continuum model have been used to study the mechanism for
formation of carbamate from CO2 and alkanolamines. The molecules studied are ethanolamine
and diethanolamine. A brief review is also made of published experimental observations relevant
to the reaction mechanism. The ab inito results suggest that a single-step, third-order reaction
is the most likely. It would seem unlikely that a zwitterion intermediate with a significant lifetime
is present in the system. A single-step mechanism also seems to be in good agreement with the
experimental data.

Introduction original one used by Caplow:2

The absorption of CO2 by alkanolamines in aqueous


solutions is a well-established technology, and both the
equilibrium and kinetics of the reactions involved have
been studied extensively.1 It is also well-known that CO2
reacts to form carbamate and/or bicarbonate, with the
relative ratio of these depending on the nature of the
alkanolamine and process conditions such as tempera-
ture and CO2 loading (mol of CO2 absorbed/mol of
amine). The mechanism of the carbamate formation has,
however, been the subject of some controversy, and the
origin of the reaction barrier in carbamate formation
has not been satisfactorily explained in the literature. A key feature of eq 3 that is lost in eqs 1 and 2 is that
Caplow assumed that a hydrogen bond is formed
Versteeg et al.1 cite the following two-step reaction between the amine and a water molecule before the
for the formation of carbamate from CO2 and alkanol- amine reacts with the CO2 molecule. It would also seem
amines: that such a hydrogen bond could be formed between the
amine and any basic species.
CO2 + R1R2NH h R1R2N+HCOO- (1) An alternative mechanism was suggested by Crooks
and Donnellan:4
R1R2N+HCOO- + B h R1R2NCOO- + BH+ (2)

where B is a base, usually a second amine molecule. The


first step of the reaction is bimolecular, and the product
of the first reaction step is a zwitterion. The mechanism
is usually referred to as the zwitterion mechanism. In
cases where an overall second-order reaction is ob-
served, one would assume the first step to be rate-
determining. Versteeg cites Caplow2 as the one who
proposed this mechanism and also notes that it was
later reintroduced by Danckwerts.3 The notation used
by Versteeg is, however, somewhat different from the

* To whom correspondence should be addressed. Tel.: Again B is used to represent a base molecule. In this
+47 73594125. Fax: +47 73594080. E-mail: silva@ mechanism, the bond formation and proton transfer to
chemeng.ntnu.no. the base take place simultaneously, giving a single-step,
10.1021/ie030619k CCC: $27.50 © 2004 American Chemical Society
Published on Web 05/25/2004
3414 Ind. Eng. Chem. Res., Vol. 43, No. 13, 2004

Figure 1. Approach of CO2 to MEA. Hydrogens on carbon are


not shown.

third-order reaction. While this mechanism appears to


be quite different from mechanisms 1 and 2, it is seen
to be quite similar to eq 3 (with water as the base in eq Figure 2. Approach of CO2 to MEA in a vacuum (open circles)
3). and in a solvent field (SM, black circles).
In this work, simple ab initio calculations were done
to investigate which mechanism is the more probable tionality. In the case of a single water molecule, it was
one. Experimental results relevant to the determination placed such that it would interact as a base with the
of the reaction mechanism are also reviewed. amine functionality (water oxygen facing the amine
It would be instructive to do calculations for alkanol- hydrogen). In the case of three water molecules, two
amines that display different kinetics and attempt to molecules were placed to interact as bases with the
account for the differences between them. We have here amine nitrogen and one water molecule was placed so
chosen ethanolamine (MEA) and diethanolamine (DEA). that one hydrogen atom forms a hydrogen bond with
MEA is known to have overall second-order kinetics in the free-electron pair of the nitrogen. The hydrogen
an aqueos solution, while DEA shows overall third-order atoms on the alcohol groups can form hydrogen bonds
kinetics.1 These molecules show markedly different both with the CO2 oxygens and with the water molecule
behavior, at the same time they both are among the oxygen. In the present work, we have chosen, somewhat
more common alkanolamines used for absorption pur- arbitrarily, to use conformers of MEA and DEA with
poses. no alcohol group-CO2 interaction.
In the Results and Discussion section, one set of
Methods calculations is shown for the full transfer of the proton
to the base, giving the final products. The calculations
In this work, we have used HF/3-21G* calculations are for two MEA molecules. Modeling of the proton-
on the alkanolamines. The solvent effects were studied transfer step is somewhat difficult when using clusters
by the use of a continuum (implicit) solvation model and of molecules in a vacuum because the molecules tend
the inclusion of a small number of water molecules. A not to separate. The final configuration often shows
continuum model represents the effects of the solvent protonated molecules and carbamate bonded to each
as a dielectric continuum. It captures the overall sol- other (particularly when water is used as a base). It is,
vation effects and some of the effects of hydrogen however, known from the equilibrium states of the
bonding. Such a model cannot be used to model interac- system that the final carbamate product is more stable
tions with the solvent such as proton transfer, and it than any intermediate state, and it is this observation
only captures the average effect of the solvent rather that will be used in the discussion.
than explicit interactions between a single solvent It should be emphasized that the purpose of this work
molecule and the solute. To study such explicit interac- is to achieve a qualitative understanding of the reaction
tions, water molecules have been added. mechanism. The uncertainty in choice of conformers,
The carbon (CO2)-nitrogen (in amine) bond length level of ab initio calculation, and representation of the
was identified as the key reaction coordinate. The solvent all contribute to making it difficult to draw any
reactions were, therefore, modeled by doing a series of quantitative conclusions from this work. The need to
calculations where this bond length was kept constant have a consistent and accurate representation of both
while the rest of the geometry was optimized. This is intermolecular and intramolecular hydrogen bonds for
shown in Figure 1. For the stable configurations identi- these molecules represents a particularly difficult issue.7
fied, calculations were done without any restraints on We do, however, believe that the present level of
the geometry. modeling is adequate to identify stable intermediates
All calculations were initially done as calculations on and reaction barriers between the unreacted amine,
the explicit molecules at the HF/3-21G* level in a CO2, and the final reaction product.
vacuum. Single-point calculations with the SM 5.4A5
solvation model were done on the optimized gas-phase
Results and Discussion
configurations. The total energy calculated in these
cases, referred to as SM in the figures, is then the sum Optimizations were initially done for the unreacted
of the HF/3-21G* gas-phase energy and SM 5.4A energy molecules placed next to each other. The energies
(SM 5.4A//HF/3-21G*), with the total energy shown then obtained for these starting configurations were defined
being the free energy in the solution. All calculations to be zero for all calculations, and all other energies are
were done using PC Spartan Pro, version 1.0.7.6 given relative to these. The plots in Figures 2-4, 6, and
Calculations were done with one and three water 7 all show the energy as a function of the carbon-
molecules and with a second alkanolamine molecule nitrogen bond length. The energy is plotted from the
acting as a base. The water molecules were placed in starting configuration to the energy minimum (or
positions where they would constitute a representative slightly longer) along the reaction coordinate. In Figure
model of the solvent interactions with the amine func- 8, the energy is plotted as a function of the distance
Ind. Eng. Chem. Res., Vol. 43, No. 13, 2004 3415

Figure 3. Approach of CO2 to MEA and one water molecule in a Figure 7. Approach of CO2 to two MEA molecules in a vacuum
vacuum (open circles) and in a solvent field (SM, black circles). (open circles) and in a solvent field (SM, black circles).

Figure 4. Approach of CO2 to MEA and three water molecules Figure 8. Transfer of proton between two MEA molecules to form
in a vacuum (open circles) and in a solvent field (SM, black circles). carbamate and protonated MEA. Open circles are results in a
vacuum, and black circles are results in a solvent field (SM).

are shown, and Figure 4 shows the results with MEA


in the presence of three water molecules. As in Figure
2, energies both in a vacuum and in the solvent field
are shown. It can be seen that in the presence of one
water molecule in a vacuum the reaction proceeds
without a barrier. This suggests that the reaction has
no intrinsic barrier. In the case of three water molecules,
one of them is hydrogen bonded to the free electron pair
on the nitrogen. For the CO2 to bond, this hydrogen bond
must first be displaced, and this gives rise to a reaction
Figure 5. Approach of CO2 to DEA in the presence of three water barrier. The same kind of barrier can be seen with the
molecules. Hydrogens on carbon are not shown. solvent field calculations (Figures 2 and 3), suggesting
that with both presentations of the solvent the reaction
barrier is caused by the CO2 molecule having to displace
the solvation shell of the amine functionality.
When the energies of the products in Figures 2-4 are
compared, it can be seen that, in the cases where base
(water) molecules are included explicitly in the calcula-
tions, the formed product is much more stable. This
strongly suggests that the presence of the base is
necessary for the reaction to proceed. In all cases it was
also found that the bond lengths of the amine hydro-
gen(s) changed during the reaction. In the case of the
MEA reaction with CO2 in the presence of a single water
molecule, the amine hydrogen had an initial H-N bond
Figure 6. Approach of CO2 to DEA and three water molecules in length of 1.008 Å and was at a distance of 2.006 Å from
a vacuum (open circles) and in a solvent field (SM, black circles). the oxygen in the water molecule. Once the CO2 had
bonded to the amine, the H-N bond length increased
between a hydrogen atom on the nitrogen and the base to 1.026 Å, while the hydrogen’s distance from the
atom it is being transferred to. oxygen decreased to 1.765 Å. These changes in bond
Figure 2 shows the results for the approach of CO2 to lengths suggest a gradual proton transfer as the CO2
a single MEA molecule in a vacuum and in the con- reacts with the amine molecule.
tinuum solvation model (SM). In a vacuum, no stable In Figure 6, the results for CO2 approaching DEA in
product is formed, while in the solvent field, a stable the presence of three water molecules are shown (the
configuration is produced. In Figure 3, the results of CO2 configuration of this approach at a C-N distance of 3.2
approaching MEA in the presence of one water molecule Å is shown in Figure 5). In this case there are appar-
3416 Ind. Eng. Chem. Res., Vol. 43, No. 13, 2004

ently two reaction barriers. The first one (at around 3.2 anism is in any case the most suited to conveying the
Å) is caused by the displacement of the water molecules nature of the reaction taking place.
around the amine functionality and is of the same
nature as the barrier identified for MEA. When the CO2
molecule approaches the amine functionality, the water Review of the Experimental Data
molecules reorganize. In this set of DEA calculations,
the water molecule hydrogen bonded as a base to the As shown in the Introduction to this paper, there has
DEA is displaced while the water molecules reorganize, been considerable discussion in the literature regarding
and it this displacement that results in the second the reaction of carbamate formation. Most of the discus-
barrier seen. However, in these calculations we were sion has been based on the analysis of experimental
only looking at single set of configurations of the DEA kinetic data. For the zwitterion mechanism, Versteeg
molecule and the water molecules. The second barrier et al.1 have presented the following expression for the
is probably a result of the particular set of configura- reaction rate:
tions that we have chosen and is probably not repre-
sentative of the DEA formation reaction barrier. It does RCO2 ) -k2[CO2][R1R2NH]/(1 + k-1/ ∑kB[B]) (6)
however (again) show that the presence of a base
molecule is necessary for a stable product to be formed. If the zwitterion formation is rate-determining, the
While not shown, calculations with CO2 approaching expression reduces to
DEA without the presence of water molecules and CO2
approaching DEA in the presence of one water molecule RCO2 ) -k2[CO2][R1R2NH] (7)
have been performed. These calculations show a quali-
tative agreement with the results shown for MEA. While This expression can explain the first-order kinetics with
the present calculations are not accurate enough to respect to amine observed for aqueous MEA. For second-
make any quantitative comparison between MEA and order kinetics, as for cases with DEA, one needs to
DEA, the results do suggest that their reaction with CO2 assume that the deprotonation is rate-determining,
proceeds along the same reaction path. leading to the following expression:
In Figure 7 is shown the approach of CO2 to MEA,
with a second MEA molecule working as the base. The
RCO2 )
-k2 ∑kB[B][CO ][R R NH] (8)
two MEA molecules were oriented toward each other k-1 2 1 2
so they only interact through the amine functionality.
Because of sterical repulsion, the amine functionalities Here B is any base in the system. With amine, water,
maintain an N-N distance of around 3 Å. This limits and other bases B, eq 8 can be written as
the degree of hydrogen bonding between them, which
results in no stable intermediate being formed in the -k2
gas phase. In Figure 8, the transfer of the proton from RCO2 ) (k [R R NH] + kH2O[H2O] +
the carbamate to the base is shown. In this figure, the k-1 am 1 2
x axis shows the distance between the proton being
transferred and the amine group (nitrogen atom) it is
∑kB[B])[CO2][R1R2NH] (9)

being transferred to; the starting configuration is iden- If the amine deprotonation term dominates in eq 9, this
tical with the final products in Figure 7. From this equation becomes
figure, it is seen that such a proton transfer can take
place without any reaction barrier. -k2
These results strongly suggest that, for the CO2 to RCO2 ) k [CO2][R1R2NH]2 (10)
k-1 am
react with an amine center, basic solvent molecules
must be present. On the basis of this observation, we This gives a reaction order of 2 with regard to the amine.
can see two possible reaction mechanisms. Reaction orders between 1 and 2 are explained by a mix
One possibility is the mechanism proposed by Crooks between the two cases. Arrhenius expressions based on
and Donnellan (eq 4), i.e., amine bonding to CO2 and this are in the literature1,8 presented not only for k2 but
proton transfer taking place simultaneously. In a gen- also for k2kam/k-1 and k2kH2O/k-1. The deprotonation
eral way, this can be written as constants kam and kw, however, cannot be obtained alone
but only in the groups given.
CO2 + R1R2NH‚‚‚B h R1R2NCOO-‚‚‚BH+ (5) The consequence of the direct third-order formulation
is that water, amine, and other bases can influence the
reaction in parallel. With water and amine as the
A second alternative is that the CO2 bonds to the amine, dominating bases, we get
with the solvent molecules stabilizing the zwitterion-
like intermediate with hydrogen bonds. Such an inter- RCO2 ) -(kam H 2O
mediate is, however, likely to have a very short lifetime. 3 [R1R2NH] + k3 [H2O])[CO2][R1R2NH]

If a strong base molecule (usually an amine molecule) (11)


appears in the vicinity, the proton would be transferred
to the base and carbamate is formed; if the reaction to Here kam 3 and k3
H2O
are real third-order rate constants.
carbamate is not completed, the zwitterion-like inter- However, for practical purposes, eq 11 is the analogue
mediate is likely to revert back to free CO2 and amine. to eq 9 with k2kam/k-1 and k2kH2O/k-1 equal to kam 3 and
While this second alternative is a form of the zwit- kH
3
2O
, respectively.
terion mechanism, it must be emphasized that any From eq 11, two extreme cases can be defined. If
zwitterion-like intermediate would have a very short kH 2O am
3 cH2O . k3 cam, then water is the dominating base.
lifetime and we would argue that the single-step mech- Because water is the solvent, this will be observed as a
Ind. Eng. Chem. Res., Vol. 43, No. 13, 2004 3417

first-order reaction in amine: observed for MEA (and other alkanolamines) when
solvents such as methanol and ethanol are changed.1
RCO2 ) -kH 2O
3 [H2O][CO2][R1R2NH] )
We are not aware of any physical explanation having
been offered in the framework of the zwitterion mech-
-2k2[CO2][R1R2NH] (12) anism to account for the different alkanolamines dis-
playing different reaction order.
If kam H2O
3 cam . k3 cH2O, then the amine itself is the For the single-step, third-order reaction, an explana-
dominating base and the reaction is second-order with tion will presently be suggested. Sartori and Savage11
respect to the amine, as observed for DEA. reported in an NMR study that MEA forms a more
stable carbamate than DEA, a conclusion that can also
RCO2 ) -k3[CO2][R1R2NH]2 (13) be inferred from kinetic data. Because MEA has a more
stable carbamate form, even a weak base, such as water,
This is identical with the third-order reaction rate might be enough to drive the reaction forward. For DEA,
expression for the zwitterion mechanism, eq 10. As a stronger base might be required, giving an overall
already seen, eq 12 is the same reaction rate equation third-order reaction. This would suggest that amines
as that for the second-order zwitterion mechanism, eq that form stable carbamates, in general, will have lower
7. The two reaction mechanisms can, therefore, be seen reaction order than amines with less stable carbamate
to give identical rate functions, and any set of experi- forms. Only limited data have been published for
mental kinetic data can be fitted to either mechanism. carbamate stability, but it would seem reasonable to use
The same observation has also been made by Kumar et the reaction rate as a proxy parameter for carbamate
al.9 This clearly means that it is very difficult to deduce stablity; i.e., a high reaction rate indicates strong
the reaction mechanism from kinetic studies. carbamate formation. In their review, Versteeg et al.1
Some arguments have however been made in the report a high reaction rate and an overall second-order
literature on the nature of the reaction based on the reaction for diglycolamine, 1-amino-2-propanol, 2-(meth-
kinetics, and these will be discussed presently. ylamino)ethanol, and 3-amino-1-propanol. For di-2-
Kumar et al.9 observed that in some systems, such propanolamine, the same authors report a low reaction
as DEA in water, the reaction rate can change with the rate and an overall third-order reaction. These observa-
amine concentration in the system.10 In the case of DEA, tions are all consistent with the present explanation.
the reaction order apparently falls at low concentration. While a solvent such as water might be a strong
They claimed that this change in reaction order could enough base to (locally) allow the formation of carbam-
only be accounted for with the zwitterion mechanism. ate for amines such as MEA, it is also clear that the
We propose an alternative explanation for this observa- presence of stronger bases in the system (usually other
tion. Starting with eq 11, an expression for the apparent amine molecules) is required to shift the equilibrium
reaction order with respect to amine can be derived by toward the formation of carbamate. Water, or other
differentiating the equation solvents, might be thought of as transporting protons
to bases not placed immediately next to the reacting
∂ ln rCO2 kam
3 cam amine group.
n) )1+ (14)
∂ ln cam kam H 2O In a recent paper, Aboudheir et al.12 argued that a
3 cam + k3 cH2O
single-step, third-order mechanism is best suited to
explain all observed kinetic phenomena. They argued
At high DEA concentrations, the effect of the amine as that when data are fitted to the expression for the
the base will be strong, and kam H2O
3 cam . k3 cH2O in eq 14. zwitterion mechanism, some parameters take on un-
This is seen to lead to a reaction order of n ) 2 in eq physical values.
12. When the DEA concentration is lowered, the two Spectroscopic techniques and NMR would seem to
terms in the denominator become more equal. At very offer the most direct experimental insight into what
low concentrations, then kam H2O
3 cam , k3 cH2O and an species are present in a given system. Ohno et al.13 have
order of n ) 1 is predicted. reported a detailed spectroscopic study of 2-(N-methyl-
It would also seem that all amines can contribute to amino)ethanol and its reaction with CO2 in an aqueous
base catalysis of bicarbonate formation. This reaction solution. Looking specifically at the issue of the zwit-
is first-order in amine concentration, and while it is terion formation, they found no evidence of its existence
slower than carbamate formation, the reaction might and suggested that only the carbamate and base forms
be significant at low amine concentrations. If significant,
of the alkonolamine were formed. This observation is
the bicarbonate formation would contribute to a lower-
consistent with a single-step mechanism.
ing of the observed overall reaction order.
It has previously been argued that a single-step, third- On the basis of the present conclusions, we can
order mechanism cannot explain the broken-order ki- attempt to identify the origin of the reaction barriers.
netics observed for some systems. When eq 11 is looked In the case of MEA and other molecules showing second-
at, it is, however, clear that this is not the case. The order kinetics in water, it would seem that the only
extent to which water, or another solvent, works as a barrier comes from the CO2 molecule having to dislodge
base will vary (as explained above), and this can directly the solvent molecules in the solvation shell of the amine
account for the broken-order kinetics observed. groups. For the systems where stronger bases are
Equation 11 also suggests that when the solvent is required, the need for a base molecule to approach the
changed to a weaker base, the reaction order in the amine functionality clearly represents a barrier to the
amine will increase. This depends on the autoprotolytic reaction taking place. We believe this effect to be
constant of the solvent. For water, pKw is 14, whereas dominant for molecules displaying an overall third-order
for ethanol, pKEtOH is 19, and this effect is indeed reaction.
3418 Ind. Eng. Chem. Res., Vol. 43, No. 13, 2004

Conclusion (2) Caplow, M. Kinetics of Carbamate Formation and Break-


down. J. Am. Chem. Soc. 1968, 90, 6795.
In studying the formation of carbamate from CO2 and (3) Danckwerts, P. V. The Reaction of CO2 with Ethanolamines.
alkanolamines in solution, we find a single-step, third- Chem. Eng. Sci. 1979, 34, 443.
order reaction mechanism to be the most likely. Such a (4) Crooks, J. E.; Donnellan, J. P. Kinetics and Mechanism of
mechanism is consistent with both ab initio calculations the Reaction beween Carbon Dioxide and Amines in Aqueous
and experimental observations on these systems. Solution. J. Chem. Soc., Perkins Trans. 2 1989, 331.
The apparent second-order mechanism for MEA in (5) Chambers, C. C.; Hawkins, G. D.; Cramer, C. J.; Truhlar,
water can be explained by water acting as a base. The D. G. Model for aqueous solvation based on class IV atomic charges
and first solvation shell effects. J. Phys. Chem. 1996, 100, 16385.
varying extent as to which the solvent can act as a base
(6) PC SPARTAN, version 1.0.7; Wavefuncion, Inc.: Irvine, CA,
would also seem to account for the broken-order kinetics 2001.
observed for this reaction. Overall third-order kinetics (7) da Silva, E. F.; Svendsen, H. F. Prediction of the pKa Values
is most likely to be displayed by amines having less of Amines Using ab Initio Methods and Free-Energy Perturba-
stable carbamate forms. tions. Ind. Eng. Chem. Res. 2003, 42, 4414.
The barriers for this reaction could either originate (8) Xu, S.; Wang, Y.-W.; Otto, F. D.; Mather, A. E. Chem. Eng.
from the CO2 molecule having to displace the solvation Sci. 1996, 51, 841.
shell around the amine group before reacting with the (9) Kumar, P. S.; Hogendoorn, J. A.; Versteeg, G. F.; Feron, P.
amine functionality and/or from the need for a base H. Kinetics of the reaction of CO2 with aqueous potassium salt of
taurine and glycine. AIChE J. 2003, 49, 203.
molecule to approach the amine functionality before
(10) Versteeg, G. F.; Oyevaar, M. H. The reaction between CO2
reaction can take place. and Diethanolamine at 298 K. Chem. Eng Sci. 1989, 44, 1264.
Further studies are needed before any quantitative (11) Sartori, G.; Savage, D. W. Sterically Hindered Amines for
estimates of the various aspects of the reaction can be CO2 Removal from Gases. Ind. Eng. Chem. Fundam. 1983, 22,
made. More advanced modeling of solvent effects for this 239.
reaction will then be needed. (12) Aboudheir, A.; Tontiwachwuthikul, P.; Chakma, A.; Idem,
R. Chem. Eng. Sci. 2003, 58, 5195.
Acknowledgment (13) Ohno, K.; Matsumoto, H.; Yoshida, H.; Matsuura, H.;
Iwaki, T.; Suda, T. Reaction of Aqueous 2-(N-Methylamino)ethanol
Financial support for this work by the Norwegian Solutions with Carbon Dioxide. Chemical Species and Their
Research Council is greatly appreciated. Gratitude is Conformers Studied by Vibrational Spectroscopy and ab initio
also expressed to Dr. Karl Anders Hoff for helpful Theories. J. Phys. Chem. A 1998, 102, 8056.
discussions.

Literature Cited Received for review July 29, 2003


Revised manuscript received March 19, 2004
(1) Versteeg, G. F.; van Dijck, L. A. J.; van Swaaij, W. P. M. Accepted March 30, 2004
On the Kinetics Between CO2 and Alkanolamines both in Aqueous
and Non-Aqueous Solution. An Overview. Chem. Eng. Commun.
1996, 144, 113. IE030619K
Paper III

Use of Free Energy Simulations to predict Infinite Dilution Activity

Coefficients

Eirik Falck da Silva

2004

Fluid Phase Equilibria, 221, 15-24 and


Erratum Fluid Phase Equilibria 231, 252-253.
Use of Free Energy Simulations to Predict Infinite Dilution Activity Coefficients
Fluid Phase Equilibria, 221, 15-24 and Erratum Fluid Phase Equilibria 231, 252-253.

Eirik F. da Silva*

Department of Chemical Engineering, Norwegian University of Science and Technology,

N-7491 Trondheim, Norway

Abstract
An important challenge in applied thermodynamics is the prediction of mixture
phase behavior without the use of experimental data. Current group contribution methods
are sometimes, but not always successful in this regard. In the present work Monte Carlo
free energy perturbations are used in calculating the free energies of solvation for pure
component and infinite dilution using the OPLS force field. Infinite dilution activity
coefficients are calculated by pure simulation and simulations used in combination with
experimental vapour-pressures. The activity coefficients are then used to fit the parameters
in Wilson’s equation thereby giving overall predictions of activity. Results are compared
with free energies and activity coefficients based on experimental values. The systems
studied are methanol + water, ethanol + water, acetonitrile + water, formic acid + water
and ethanolamine (MEA) + water.

Keywords: Vapour-liquid equilibria; Activity coefficient; Molecular Simulation;


Ethanolamine

1
Introduction
Knowledge of chemical activity is an important component in understanding
various chemical processes, chemical gas-absorption being one example of this [1]. Infinite
dilution activity coefficients can for example be used to make overall predictions of
activity and approximate vapour-liquid equilibrium curve.
Traditionally Gibbs excess free energy models have been used to model activity
coefficients. The only predictive models have been group contribution models such as
UNIFAC [2]. These models are to some extent limited by access to experimental data. In
systems with a large number of chemical species and limited experimental data such as in
gas-absorption the use of such models becomes difficult.
The value of models that can predict activity coefficients is therefore clear. And it
should be emphasized that even models that are only qualitative can in some cases be a
useful tool, particularly in understanding complex systems.
The ideal would be some form of ab initio (quantum mechanical) model without
any need of parameterization to experimental data. It is however clear that treating the
condensed phase at a quantum mechanical level is computationally very expensive. Guillot
[3] reports a vaporization enthalpy for water of 7.38 kcal/mol calculated using ab initio
molecular dynamics (Car-Parrinello), while the experimental vaporization enthalpy for
water is around 11 kcal/mol [3]. While this result is encouraging it is clear that many
issues remains before ab initio molecular dynamics methods can be used for quantitative
predictions of solvation.
In a recent work Sum and Sandler [4] use ab initio calculations on clusters of eight
molecules to calculate parameters for an excess Gibbs energy model (UNIQUAC). The
results presented in the article show very good agreement with experimental data. The
procedure does however involve a number of approximations and uncertainties that are not
fully acknowledged in the article. The ab initio method used is Hartree-Fock, involving
approximations that usually makes it unsuited for obtaining absolute bonding energies [5].
It is also assumed that the interactions of a cluster of eight molecules is representative of
the interactions in a liquid. The size of the cluster at the same time also results in a very
small number of data for the statistical averaging that is done. Finally there is the
approximation in the use of the UNIQUAC equation itself which is known not to be
generally accurate [6,7]. It is therefore difficult to reconcile the large uncertainties and

2
numerous approximations in the method with the apparent high quality in the results. This
work does not in any case represent any true ab initio modeling of the condensed phase.
Classical simulations have been central in developing the understanding of the
condensed phase. Despite their fairly long history it has been observed [8] that their use for
the practical prediction of chemical activity is at an early stage. There are two main issues
regarding use of simulations:
One is the representation of the molecule used. Some models such as OPLS
(Optimized Parameters for Liquid Simulation) [9] include polarization only implicitly,
while some other force fields are polarizable. There are also differences in the flexibility of
molecules and the number of sites used. Force fields such as OPLS can be thought of as
offering a molecular level group contribution method [10].
There are numerous choices in how to optimize a force field model. There are also
many ways ab initio calculations can be used to parameterize force fields. Examples are
determination of torsional potential parameters [11] and atomic charges [12].
The second issue is how accurate data for the free energy can be achieved for a
given molecular representation. There are a number of calculation schemes based on forms
of thermodynamic integration and free energy perturbation [10,13]. For each scheme there
is usually also a number of simulation parameters to set. There would not seem to be any
general agreement on which methods are superior. Particularly if one considers the
tradeoffs between accuracy and simulation time it is difficult to know what methods to
choose.
In this work Monte Carlo free energy perturbations are used with OPLS force field
representation of the molecules. The simulation method is one of the most widely used, its
main appeal being that it is relatively inexpensive in terms of computation time.
The focus of this work is on binary systems with water. Methanol, ethanol,
acetonitrile and formic acid are chosen because parameters for them are available in the
OPLS force field and because experimental data is available for binary systems with water.
Ethanolamine, also known as MEA, is an important molecule in gas-purification
[1,14]. For ethanolamine itself considerable experimental data is available, for other
alkanolamines the data is however often very limited. If a molecular force field (such as
OPLS) was found to offer parameters that could be transferred to alkanolamines it could be
used in modeling of a number of systems of importance in gas-purification.

3
Theory
The difference between the chemical potential in a mixture and the ideal gas-phase
is called the residual chemical potential and at infinite dilution is directly related to the
activity coefficient [10] by the following equation:

res res res


kT ln γ1∞ = (µ12 −µ12 ) = (µ12 −µ 22 ) + (µ 22 −µ11 ) (1)

Where µ12 is the chemical potential of a molecule of type 1 in a pure solvent of type 2. The

equation is valid for constant temperature (T) and pressure, k is the Boltzmanns constant.
There are several ways in which data can be obtained to solve Eq. (1). One can
directly calculate the free energy of a single particle by insertion or removal. This can for
example be done stepwise:

res res
res
µ 21 = (µ 21 −µ11 ) − (µ11 −µ 01 ) (2)

The subscript 0 is used to indicate that a particle does not exist in the system, it has either
been removed or will be inserted. The experimental vapour-pressures can also be used to
calculate the difference in the residual free energy of the pure components by use of the
following equation [10]:

⎛ f10 ⎞⎟ ⎛ P1sat (T ) ⎞⎟
(µ 22 −µ11 ) = kT ln ⎜⎜⎜ ⎜⎜ sat
res
0⎟
⎟ ≈ kT ln ⎟⎟ (3)
⎝⎜ f 2 ⎠⎟ ⎝⎜ P2 (T ) ⎠⎟

where f is the fugacity and P sat is the saturation pressure. Lazaridis and Paulaitis [15]
used Eq. (3) in combination with equation 1 to obtain activity coefficients at infinite
dilution.
In the present work the free energy of solvation will be calculated. While
differences in the residual chemical potential and the free energy of solvation are the same
at infinite dilution their relative values for pure components is different. The free energy
of solvation ( ∆µ s ) has the following relationship to the saturation pressure[16]:

∆µs = kT ln( P sat M s / RTd s ) (4)

4
where Ms is molar mass and ds is density. To obtain the difference in residual free energies
for the pure components from solvation energies the Ms/ds ratio for each component must
be subtracted.
For the pure component there are also other techniques available for the calculation
of the free energy. Hermans et al. [17] and Mezei [18] have presented results based on
thermodynamic integration of the entire ensemble.
Once both infinite dilution activity coefficients in a binary system are estimated the
data can be fit to any two-parameter Gibbs excess model such as Wilson’s or UNIQUAC
to get an estimate of activity coefficients for all compositions, a possibility mentioned by
Haile [19]. In this work the Wilson equation has been chosen [6]:

⎛ Λ12 Λ 21 ⎞⎟
ln( γ1 ) = − ln ( x1 + Λ12 x2 ) + x2 ⎜⎜ + ⎟⎟ (5)
⎝⎜ x1 + Λ12 x2 x2 + Λ 21 x1 ⎠⎟

5
Methods
The free energy perturbation (FEP) approach centers on the relationship [20]:

A1 − A0 = − kT ln exp[− β (E1 − E 0 )] 0 (6)

The equation expresses the free energy difference between systems 0 and 1 by an average
of a function of their energy difference that is evaluated by sampling based on E0. This
calculation works if the two systems are similar to each other. A coupling parameter is
introduced to allow gradual interconversion of the potential functions and geometries. This
interconversion can be represented by the following equation [20]:
ξ (λ ) = ξ0 + λ (ξ1 − ξ0 ) (7)

Where is any feature of the system. In the present work the “double-wide” sampling
technique described by Jorgensen and Ravimohan [20] will be used. This is a so-called
single-topology method.
The calculation of absolute free-energy is somewhat more difficult than calculating
relative free-energies. In obtaining absolute free energies insertion schemes are considered
superior to deletion schemes[21]. Standard double-wide calculations involve both deletion
and insertion steps and are their use in this context is therefore questionable. In this work
staged insertion of argonlike particles will therefore be used to obtain the absolute free-
energy of solvation. A very similar approach has been used previously by Jorgensen et al.
[22] in obtaining the absolute free energy of solvation of TIP4P water, the method was in
that work called single solvent perturbation.
As an alternative approach the free-energies of the pure components can be
determined from experimental vapour-pressure (Eq. 3). The simulations are then only used
to obtain relative free-energies of solvation.
In BOSS [23] the potential energy is expressed as the sum of Lennard-Jones and
Coulomb potential functions:

Ai A j Ci C j qi q j
U =∑ −∑ +∑ (8)
i< j rij12 i< j rij6 i< j rij

6
where the sums are over all pairs of interaction sites, Ai=(4 εi σi 12)1/2 ,Ci=(4 εi σi 6)1/2 , qi is

the partial electric charge of interaction site i and rij is the separation between interaction
sites.
The interaction sites are usually atoms; however, non-polar hydrogens bonded to carbon
atoms can be combined with the carbon atom to form a single «united» atom (UA).
For the water molecule two representations were used: the TIP3P model [24] for
water as solute and the TIP4P for water as solvent. The TIP4P model has a total of four
interaction sites: the oxygen atom, the two hydrogen atoms, and a center for the negative
charge located along the dipole vector. The TIP3P model was chosen as solute because the
three-center model can more easily be used in perturbations. Calculations have also been
done on TIP4P as solute to offer comparison with data reported in the literature.
Internal rotations were included for ethanol, formic acid and ethanolamine. The
rotational potential is represented by the Fourier series in Eq. (9):

1 1 1
V (φ) = V1 (1 + cos (φ)) + V2 (1− cos (2φ)) + V2 (1 + cos (3φ)) (9)
2 2 2

where V is a coefficient and φ is the torsional angle. Following Jorgensen et al. [11] the
intramolecular energy interactions between molecules separated by 3 bonds (1-4
interactions) were scaled down by a factor of 2. Coefficients for internal rotations are
given in Table 1.

Table 1 Fourier Coefficients (in kcal/mol)

Molecule Dihedral Angle V1 V2 V3


ethanol H-O-CH2-CH2 0.834 -0.116 0.747
formic acid H-O(H)-C-O 0.000 0.000 0.000
ethanolamine H-O-CH2-CH2 0.834 -0.116 0.747
N-CH2-CH2-O 9.900 0.000 0.000
H(N)-N- CH2-CH2 -0.190 -0.417 0.418

OPLS geometries published by Jorgensen et al. were used for TIP4P [24], TIP3P [24], UA
methanol [25], UA ethanol [25] and UA acetonitrile [26]. For formic acid unpublished
geometry included in the BOSS program distribution [23] was used. Bond lengths for
formic acid were 1.2258 Å (C=O), 1.356851 (C-O(H)), 1.09 (C-H) and 0.943611 (O(H)-

7
H). Angles are 119.277520 ( O(H)-C-O), 108.159159 (H(O)-O(H)-C) and 124.451989
(H(C)-C-O). It should be noted the formic acid parameters in BOSS were for a fully
flexible molecule, in the present work the parameters are used with rigid bond-lengths and
angles.
For ethanolamine no OPLS parameterization has been published. The amine
parameters are taken from Rizzo and Jorgensen [9] and for the alcohol functionality the
UA ethanol parameters are used. The CH2 group neighboring the amine group is given the
same Lennard-Jones parameters as the ethanol functionality CH2 group, finally the charge
of this group is chosen to give the molecule a total charge of zero. Geometry parameters
are based on mixture of HF/6-31G* ab initio geometry and AM1 semiemprical geometry.
Bond-lengths are set to 1.012 Å (N-H), 1.45 Å (N-CH2), 1.53 Å (CH2-CH2), 1.43 Å (CH2-
O) and 0.945 (O-H). Angles are 108.5 (H-CH2-O), 112 (CH2-CH2-O) 108.46( N-CH2-CH2)
and 109.69 (H-N-CH2). Ethanolamine has four dihedral angles. The relative positions of
the amine-group hydrogens are fixed by setting the dihedral angle H1(N)-N-CH2-H2(N) to
121.68. The three other dihedral angles are chosen to be flexible. Using HF/6-31G* ab
initio calculations on a single ethanolamine molecule the most stable gauche conformer
was found to be 2.81 kcal/mol more stable than the most stable trans conformer. Following
Jorgensen et al. [11] the coefficients for the rotations are set to reproduce this conformer
energy difference. For the alcohol-group the same coefficients as used for ethanol are
chosen. The coefficients are given in Table 1. All Lennard-Jones and charge parameters for
the molecules studied are listed in Table 2.

8
Table 2 Force Field Parameters
Molecule United Q
Atom (Å) (kcal/mol) (Electron units)
argon 3.401 0.2339 0.000
water O 3.15365 0.155 0.000
(TIP4P) H 0.000 0.000 0.520
M 0.000 0.000 -1.040
water O 3.15061 0.1521 -0.834
(TIP3P) H 0.000 0.000 0.417
methanol O 3.070 0.170 -0.700
CH3 3.775 0.207 0.265
H 0.000 0.000 0.435
ethanol CH3 3.905 0.175 0.000
CH2 3.905 0.118 0.265
O 3.070 0.170 -0.700
H 0.000 0.000 0.435
acetonitrile C 3.650 0.150 0.280
CH3 3.775 0.207 0.150
N 3.200 0.170 -0.430
formic acid C 3.750 0.105 0.520
O 2.960 0.210 -0.440
H(C) 2.420 0.015 0.000
O(H) 3.000 0.170 -0.530
H(O) 0.000 0.000 0.450
ethanolamine N 3.300 0.170 -0.900
H(N) 0.000 0.000 0.360
CH2(N) 3.905 0.118 0.180
CH2(O) 3.905 0.118 0.265
O 3.070 0.170 -0.700
H(O) 0.000 0.000 0.435

Formic acid and ethanolamine have an intramolecular contribution to the free energy. To
obtain the free energy of solution the intramolecular free energy in the gas-phase must be
subtracted from the value in solution. Results in this work will be presented with the gas-
phase contribution subtracted.
The OPLS force field is intended to offer transferable parameters for molecules
both as solute and solvent, they are usually fitted to density and vaporization enthalpy. The
representations for water (TIP4P), UA methanol, UA ethanol and UA acetonitrile are all
reported to reproduce fairly accurately densities and heats of vaporization for pure
components (see Table 3).
The particle insertion simulations were done by growing a uncharged (UC) TIP4P
molecule in the ensamble. The particle was not grown from zero but from a small particle
with negligible free-energy. This particle had a Lennard-Jones σ =0.01 Å and a ε =0.015
kcal/mol. Most of the other simulations were done directly between the solutes and a water
molecule. The size difference between the ethanolamine molecule and water is however

9
very large, in this case ethanolamine was first perturbed to an argonlike uncharged (UC)
ethanolamine molecule. This uncharged molecule was then perturbed to UC TIP4P.
Boxes with 267 solvent molecules were used. All solvent boxes were equilibrated
for at least 10 million configurations before use. Double-wide perturbations were done
with ∆λ = +/-0.05, with λ =0.05, 0.15, 0.25, 0.35,0.45,0.55,0.65,0.75,0.85 and 0.95. Each
step had 500 000 configurations of equilibration and 500 000 configurations of averaging.
The particle insertions were done with ∆λ = +0.05 and same amount of sampling and
equilibration for each step as for the double-wide perturbations.
Periodical boundary conditions were used. Preferential sampling of the solvent near
the solute [20] was used for the double-wide sampling calculations based on the following
weighting:
1
(10)
(r + wkc)
2

Following recommendations in the BOSS documentation [23] wkc values between 200 and
300 were used. This procedure was however not used for the particle insertion
calculations. All simulations were done in a NPT ensemble at a pressure of 1 atmosphere,
the ethanolamine simulations were done at 60 ° C, while the other simulations were done at
25 ° C. Attempts to change the volume of the system were done every 700-1625
configurations. Lowest frequency of volume changes was for ethanolamine while the
highest was for TIP4P. Each volume change was 130 Å3. New configurations are
generated by selecting a molecule, translating it randomly in all three Cartesian directions,
rotating it randomly about a randomly selected axis and performing any internal rotations.
Ranges for translations are set at 0.15 Å while ranges for angular rotations and dihedral
rotations are set at 15 degrees.
Interactions were cut off with a quadratic “switching” function. The same cutoffs as
previously used by Jorgensen et al. were used: 8.5 Å for water (TIP4P) [22], 9.5 Å for
methanol [25], 11.0 for ethanol [25], 10 Å for acetonitrile [26] and 12 Å for formic acid
(value used for acetic acid [27]). Ethanolamine is a larger molecule and in this case a
cutoff of 14 Å was chosen. The same cutoffs were chosen for solvent-solvent interactions
and solute-solvent interactions, except in water where a 10 Å solute-solvent interaction
was chosen. The statistical uncertainty was estimated by the batch means procedure [9]
and standard deviations from the total energy of separate runs using different starting
configurations. For each free energy difference three separate simulations from different

10
staring configurations were used. Each free energy calculation took between 2 and 12
hours on a single processor PC.
The heat of vaporization to the ideal gas has been calculated to make comparison
with experimental data and with previous work by Jorgensen et al. (Table 3). It is
calculated according to the following equation [27]:

∆H vap = Eintra ( g ) − ( Ei (l ) + Eintra (l )) + RT (11)

where Eintra ( g ) and Eintra (l ) are the intramolecular rotation energies for the gas and the

liquid, Ei (l ) is the intermolecular energy for the liquid. There has apparently not been

published any estimates for the enthalpy of vaporization of ethanolamine except at the
boiling temperature. Antoaine equations have however been published and in the present
work this will be used to estimate the experimental vaporization enthalpy using the
Clapeyron equation:

dP ∆H vap
= (12)
dT T ∆Vvap

where P is the pressure and ∆Vvap is the molar volume change upon vaporization. The

calculation will be done assuming ideal gas and neglecting the molar volume in the liquid
phase.

11
Results and Discussion
In Table 3 densities and vaporization enthalpies for the solvents studied are shown.
While some of the experimental vaporization enthalpies are not for ideal gas, the energy
differences between ideal gas and real gas are in these cases small.

Table 3 Densities and Energy Results

Solvent T d [g/cm3] ∆ Hvap [kcal/mol]


[ ° C]
Calcd Calcd Exptl Calcd Calcd Exptl
water (TIP4P) 25 1.012 0.999 [24] 0.997 [28] 10.60a 10.66 [24] a 10.51 [24]
methanol 25 0.720 0.759 [25] 0.786 [28] 8.97 a 9.05 [25] a 8.94 [25]
ethanol 25 0.741 0.748 [25] 0.785 [28] 9.90 a 9.99 [25] a 10.11 [25]
acetonitrile 25 0.750 0.765 [26] 0.776[28] 7.57 a 8.03 [26] a 8.01 [31]a
formic acid 25 1.236 1.214 [28] 12.39 a 11.03 [31]a
ethanolamine 25 1.056 1.012 [29] 14.59 a
ethanolamine 60 1.029 0.984 [30] 13.59 a 13.79b
a
: Heat of vaporization to ideal gas. b: Obtained by using the Clapeyron equation (Eq. 12) with
Antoaine equation data given in [32].

For TIP4P, methanol, ethanol and acetonitrile the simulation results are in good agreement
with the experimental values, reflecting the fact that the OPLS force field has been
parameterized to reproduce these properties. The results are slightly different from those
obtained for the same systems by Jorgensen et al. The number of molecules in the
simulation boxes is in the present study different (in all cases larger) than used in the
original studies and this probably explains part of the discrepancy. It should also be noted
that the present simulations with use of preferential sampling are not optimal for estimating
overall solvent properties.
The agreement for formic acid is slightly worse reflecting that this model has not
been parameterized in the same way.
For ethanolamine the parameters are taken from other molecules, it is therefore
very encouraging to see the good agreement with the experimental data. The densities are
overestimated by around 4.5% at both temperatures. The agreement with the vaporization
enthalpy is even better. These results suggest that the OPLS force field has reasonable
transferability to alkanolamines.
In Table 4 the results for the separate free energy calculations based on double-
wide sampling are shown. For most calculations free energy differences have been
calculated using three simulations based on different starting configurations, the result
shown is the average of all simulations. The uncertainty given with the result is the
standard deviation over subsets in the calculations, again the average for all runs is shown

12
(the standard deviation not varying that much from run to run). The standard deviation of
the total free energy from repeated runs is shown in a separate column.

Table 4 Free Energy Differences (Results in kcal/mol)


Solvent T [ ° C] ∆ Gcalca S.D.b
methanol → waterc waterd 25 -1.97 +/-0.21 0.26
waterd → UC water waterd 25 8.25 +/-0.15 0.54
waterc → UC water waterd 25 8.93 +/-0.18 0.30
methanol → waterc methanol 25 -0.53 +/-0.16 0.31
waterc → UC water methanol 25 6.43 +/-0.19 0.15
ethanol → waterc waterd 25 -1.97 +/-0.37 0.38
ethanol → waterc ethanol 25 0.36 +/-0.25 0.44
waterc → UC water ethanol 25 5.64 +/-0.17 0.76
acetonitrile → waterc waterd 25 -3.49 +/-0.46 0.49
acetonitrile → waterc acetonitrile 25 1.07 +/-0.15 0.12
waterc → UC water acetonitrile 25 4.57 +/-0.06 0.12
formic acid → waterc waterd 25 -0.22 +/- 0.51 0.23
formic acid → waterc formic acid 25 0.59 +/-0.33 1.52
waterc → UC water formic acid 25 10.26 +/-0.13 1.11
MEAe → UC MEAe waterd 60 11.40 +/-0.20 0.40
MEAe → UC water waterd 60 0.39 +/-0.16 f
waterc → UC water waterd 60 8.82 +/-0.14 0.47
MEAe → UC MEAe MEAe 60 7.90 +/-0.20 0.61
UC MEAe → UC water MEAe 60 0.57 +/-0.33 f
waterc → UC water MEAe 60 7.07 +/-0.17 0.76
a
Standard deviation from batch means procedure. b Standard deviation of total free energy for three
separate simulations. c TIP3P.d TIP4P. e ethanolamine. f Single simulation.

For the methanol to water (TIP3P) in water Lazaridis and Paulaitis [15] obtained a value of
-1.77 kcal/mol using almost identical method. While the difference in the results is
significant it is in fact within the uncertainties observed both in the work by Lazaridis and
Paulaitis and in the present one. The present result is based on greater sampling and is
probably the more accurate value. Using slightly different molecular representation
Slusher [33] obtained a value of -1.58 kcal/mol for the same energy difference.
For the methanol to water free energy difference in methanol Slusher [33] obtained
a value of -1.65 kcal/mol using somewhat different molecular representation. The present
result is –0.53 kcal/mol and the difference is clearly greater then can be accounted for by
the uncertainties in the calculations.
The standard deviations based on calculations from repeated runs are in general
somewhat higher then those based on the subsets of a single run.
In Table 5 the results of the particle insertion calculations are shown. In the
appendix underlying simulation results are shown for the insertion calculations together

13
with results based on deletion. Results from the present work do not give grounds to draw
general conclusions regarding insertion and deletion. As noted in the Methods section
double-wide sampling involves both deletion and insertion steps and should only be used
when deletion and insertion are equivalent. Further work might therefore be required to
clarify under which conditions double-wide sampling and insertion are equivalent.

Table 5 Free Energy of Insertion (Results in kcal/mol)


Solvent T [ ° C] ∆ Gcalca S.D.b
0 → argon waterc 25 2.44 +/-0.19 0.13
0 → UC water waterc 25 2.51 +/-0.18 0.30
0 → UC water methanol 25 1.1 +/-0.10 0.09
0 → UC water ethanol 25 0.95 +/-0.10 0.27
0 → UC water acetonitrile 25 1.29 +/-0.12 0.15
0 → UC water formic acid 25 2.93 +/-0.19 0.77
0 → UC water waterc 60 2.98 +/-0.17 0.36
0 → UC water ethanolamine 60 2.83 +/-0.18 0.49
a
Standard deviation from batch means procedure. b Standard deviation of total free energy for three
separate simulations. c TIP4P.

Adding together the free energy differences from Table 4 and Table 5 the total free energy
of solvation at infinite dilution, or for pure component can be calculated. In Table 6 total
simulation free energies are compared with values estimated from experimental data. The
present result for water (TIP4P) is in good agreement with a Helmoltz free energy of
TIP4P obtained by Thermodynamic Integration reported by Hermans et al. [17]. They
reported a Helmoltz free energy of –5.3 kcal/mol that translates to a Gibbs free energy of
–5.9 kcal/mol, while the present result is –5.74 +/-0.24. Using a method very similar to the
one used in the present work Jorgensen [22] obtained a value of –6.06 kcal/mol for the free
energy of solvation of TIP4P.

14
Table 6 Free Energy of Solvation (Results in kcal/mol)
Solute Solvent T [ ° C] ∆ Gs,calca ∆ Gs,expb γ∞
calc
c
γ∞
calc
d
γ∞
exp

argon watere 25 2.44+/-0.19 2.002


e e
water water 25 -5.74 +/-0.24 -6.324
waterf watere 25 -6.42 +/-0.26
methanol watere 25 -4.45 +/-0.33 -5.1 4.0 5.2 1.74g
f
water methanol 25 -5.33 +/-0.21 2.8 2.2 1.57h
methanol methanol 25 -4.80 +/-0.26 -4.859
ethanol waterb 25 -4.45 +/-0.46 -5.05 9.1 11.3 3.91g
waterf ethanol 25 -4.69 +/-0.19 5.7 4.6
ethanol ethanol 25 -5.06 +/-0.32 -5.079
acetonitrile watere 25 -3.78 +/-0.52 7.8 1758 11.1g
f
water acetonitrile 25 -3.28 +/-0.16 68.3 29
acetonitrile acetonitrile 25 -4.35 +/-0.31
formic acid watere 25 -7.02 +/-0.42 9.6 0.1 0.64g
waterf formic acid 25 -7.33 +/-0.23 0.1 10.1
formic acid formic acid 25 -7.92 +/-0.41 -5.538
waterf watere 60 -5.85 +/-0.22
ethanolamine watere 60 -8.81 +/-0.31 0.02 0.0002 0.27i
waterf ethanolamine 60 -4.24 +/-0.25 4.4 333.6 0.51i
ethanolamine ethanolamine 60 -5.65 +/-0.43
a
Standard deviation from batch means procedure. b Ref. [16]. c Pure simulation activity coefficients
(sim).d Activity coefficients based on simulations and experimental vapour pressures (sim-P).
e
TIP4P. f TIP3P. g Data from Kojima et al. [34]. h Data from Slusher [33]. i Data from fitting of
vapour-liquid equilibria to UNIQUAC [35].

The overall agreement between FEP results and experimental free energies is quite good.
Only in the case of formic acid does the energy deviate by more than 1 kcal/mol. This
overestimation of the free energy of formic acid is probably due to the use of a molecular
representation that gives too high vaporization enthalpy (Table 3).
Comparing infinite dilution activity coefficients calculated from the simulations
with experimental values good overall agreement is found in most cases. In the case of
ethanolamine the agreement is however poor. The results suggest that the simulated
solvation energies in pure ethanolamine are too low both for water and ethanolamine itself,
leading to the activity coefficient of water being too high in ethanolamine and at the same
time giving too low activity coefficient for ethanolamine in water.
Figures 1-4 show activity coefficients based on simulations together with
experimental activity coefficients. Two different estimates of the activity coefficients are
shown.
Using the data in Table 4 and Table 5 with Eq. (1) and the molar mass-density
correction from Eq. (4) the infinite dilution activity coefficients are obtained. Density data

15
are from references given in Table 3. Using these values one can determine the parameters
for Wilson’s Eq. (4) and thereby obtain a general prediction for the activity coefficients.
Results based on this calculation are referred to as simulation (sim) in Figures 1 to 4.
The other approach is to use Eq. (3) with experimental data for the saturation
vapour pressure [35] to obtain the free energies for the pure components and only use the
res
simulations for the free energy differences in the same solvent (µ 21 −µ11 ) . This gives the

activity coefficients at infinite dilution, which again can be used to fit the parameters in
Wilson’s equation. Results based on this are shown as sim-P in Figures 1-4. The infinite
dilution activity coefficients obtained using this method are shown in Table 6.
With experimental liquid composition-vapour pressure data and liquid
composition-vapour composition data a set of equations is obtained that can be solved to
obtain the activity coefficients. The experimental results shown in figures 1-4 are based on
this approach. Data from Gmehling et al. [35] has been used. Experimental methanol +
water and ethanol + water systems were at 25 C, while acetonitrile + water and formic acid
+ water systems were at 30 C. To obtain the activity coefficients the data for formic acid +
water were corrected for vapour phase association of formic acid following Gmehling et al.
[35].
For ethanolamine isothermal vapour-composition data are not available. In this case
only the vapour pressure is shown. Here the activity coefficients from the simulations have
been used together with experimental vapour pressures for pure components to obtain a
vapour-liquid equilibrium curve on the same form as the experimental data. Experimental
vapour pressures are from Nath and Bender [29] (same data given in Gmehling et al. [35] ),
while vapour pressures for pure components are from Antoaine equations given in
Gmehling et al. [35]. Plots are shown in figure 5.
Experimental activity coefficients are obtained from liquid composition-vapour pressure
data and vapour composition data.

16
Figure 1. Activity coefficients for methanol + water at 25 C. Points are experimental data. Dashed
lines are activity coefficients based purely on simulations (sim). Solid lines are based on
simulations and experimental vapour pressures (sim-P).

Figure 2. Activity coefficients for ethanol + water at 25 C. Points are experimental data. Dashed
lines are activity coefficients based purely on simulations (sim). Solid lines are based on
simulations and experimental vapour pressures (sim-P).

17
Figure 3. Activity coefficients for acetonitrile + water at 30 C. Points are experimental data.
Dashed lines are activity coefficients based purely on simulations (sim). Solid lines are based on
simulations and experimental vapour pressures (sim-P).

Figure 4. Activity coefficients for formic acid + water at 30 C. Points are experimental data.
Dashed lines are activity coefficients based purely on simulations (sim). Solid lines are based on
simulations and experimental vapour pressures (sim-P).

18
Figure 5. Vapour pressure for ethanolamine + water at 60 C. Points are experimental data.
Dashed line is from activity coefficients based purely on simulations (sim).

In the fitting process it was found that not all sets of infinite dilution activity coefficients
could be fit easily to Wilson’s equation. Particularly in the fitting of the ethanolamine
system extreme parameters (10-74) were needed, and even this does not produce a perfect
fit with the activity coefficients from the simulations. This problem arises because the
parameters in Wilson’s equation are strongly coupled and the equation is inherently better
in fitting to some forms of data.
The pure simulation (sim) and simulations in combination with experimental
vapour-pressures (sim-P) are quite good for methanol + water, ethanol + water and
acetonitrile + water. For the formic acid + water system the pure simulation results (sim)
are wrong while the sim-P results are inconclusive. The difference between the sim and
sim-P results are for formic acid dramatic, but this can be easily understood in terms of the
uncertainties in the simulations and the use of Wilson’s equation to fit the data. Formic
acid + water is also a highly non-ideal system and data obtained at infinite dilution might
be insufficient to describe the system at all compositions.
For ethanolamine +water there is only limited agreement. Part of the problem might
be the isssue regarding use of Wilson’s equation mentioned previously but the agreement
with experimental data at infinite dilution is in this case also poor (Table 6). Further
studies are needed for this system.

19
The methanol + water system has also been studied by Crozier and Rowley [36]
using different molecular representation, the results from the present work would appear to
be in better agreement with experimental data.
In general the results can be said to be fairly good suggesting that the OPLS force
field based on parameterization to density and enthalpy of vaporization is useful in
predictions of activity coefficients.
The present simulations do not have the same precision as work by Slusher [33]
and by Crozier and Rowley [36], but when comparing the quality of the results it must be
kept in mind that the simulations used in the present work are relatively inexpensive in
terms of simulation time. Further studies are needed to determine which method can
deliver the most precise results for a given amount of CPU time.
For highly non-ideal systems such as formic acid + water it would appear that data
at infinite dilution and for pure component are insufficient basis for predicting the full
equilibrium. One possibility along the lines of the present work would be to also do free
energy perturbations in mixtures of the solvents.
As noted previously there are a number of ways in which a force field can be
optimized. While the present work uses TIP4P for water and UA OPLS for methanol,
Slusher [8] uses SPC water and a different UA methanol model. The results of the present
work suggest that the difference between the models is significant. It is however not clear
if any type of force field parameterization is in general superior with respect to obtaining
free energies.
It should also be noted that systems involving hydrogen bond forming molecules
and water are among the more difficult to model. In work on binary systems with osmotic
molecular dynamics Crozier and Rowley [36] in general obtained better results for systems
involving alkanes than for systems involving molecules such as methanol and water.
At present it would seem that it is not possible to a priori predict how well a given
force field will reproduce experimental data. A semi-qualitative reproduction of
experimental data is perhaps what can be expected.

20
Conclusion
In this paper free energy perturbations have been used to calculate free energies of
solvation for pure components and solutes at infinite dilution. The results were used to fit
the parameters in Wilson’s equation to give overall predictions of activity coefficients for
mixtures. Results are shown both for pure simulation results and for simulations used in
combination with vapour pressure over pure component. For the systems methanol +
water, ethanol +water and acetonitrile + water reasonable agreement was found with
experimental data. For formic acid + water the comparison with experimentally based data
is ambiguous. For the system ethanolamine + water only partial agreement with
experimentally based data was obtained. Apparently the ethanolamine representation used
gives too low solvation energies. The present work suggests that the OPLS force field is
sufficiently accurate to give useful predictions of overall vapour-liquid equilibrium.

21
References

[1] D. M. Austgen, G. T. Rochelle, Ind. Eng. Chem. Res., 28 (1989) 1061.


[2] A. Fredenslund, J. Gmehling, P. Rasmussen, Vapour-liquid equilibrium using UNIFAC,
Elsevier, Amsterdam, 1977.
[3] B. Guillot, J. Mol. Liq., 101 (2002) 219.
[4] A. K. Sum, S. I. Sandler, Fluid Phase Equilibria, 158-160 (1999) 375.
[5] C. J. Cramer, Essentials of Computational Chemistry, John Wiley & Sons, 2002.
[6] J. Prausnitz, R. Lichtentaler, E. Gomes de Azevedo, Molecular Thermodynamics of Fluid-
Phase Equilibria, 3rd ed., Prentice Hall 1999.
[7] J. Fischer, Fluid Phase Equilibria, 10 (1983) 1.
[8] J. T. Slusher, J. Phys. Chem. B, 103 (1999) 6075.
[9] R. C. Rizzo, W. L. Jorgensen, J. Am. Chem. Soc., 121 (1999) 4827.
[10] J. T. Slusher, Fluid Phase Equilibria, 153 (1998) 45.
[11] W. L. Jorgensen, D. S. Maxwell, J. Tirado-Rives, J. Am. Chem. Soc., 118 (1996) 11225.
[12] A. Jakalian, D. B. Jack, C. I. Bayly, J. Comp. Chem., 23 (2002) 1623.
[13] P. Kollman, Chem. Rev., 93 (1993) 2395.
[14] G. F. Versteeg, L. A. J. van Dijck, W. P. M. van Swaaij, Chem. Eng. Commun., 144 (1996)
113.
[15] T. Lazaridis, M. E. Paulaitis, AIChE J., 39 (1993) 1051.
[16] A. Ben-Naim, Y. Marcus, J. Chem. Phys., 81 (1984) 2016.
[17] J. Hermans, A. Pathiaseril, A. Anderson, J. Am. Chem. Soc., 110 (1988) 5982.
[18] M. Mezei, J. Comp. Chem., 13 (1992) 651.
[19] J. M. Haile, Fluid Phase Equilibria, 26 (1986) 103.
[20] W. L. Jorgensen, C. Ravimohan, J. Chem. Phys., 83 (1985) 3050.
[21] D. A. Kofke, P. T. Cummings, Molecular Physics, 92 (1997) 973.
[22] W. L. Jorgensen, J. F. Blake, J. K. Buckner, Chemical Physics, 129 (1989) 193.
[23] W. L. Jorgensen, BOSS version 4.3, Yale University, New Haven, CT(1989a)
[24] W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W Impey, M. L. Klein, J. Chem.
Phys., 79 (1983) 926.
[25] W. L. Jorgensen, J. Phys Chem., 90 (1986) 1276.
[26] W. L. Jorgensen, J. M. Briggs, Molecular Physics, 63 (1988) 547.
[27] J. M. Briggs, T. B. Nguyen., W. L. Jorgensen, J. Phys. Chem., 95 (191) 3315.
[28] J. A. Riddick, W. B. Bunger, T. K. Sakano, Organic Solvents, 4th ed., Wiley, New York,
1986, Vol, II, 360.
[29] A. Nath, E. Bender, J. Chem. Eng. Data, 28 (1983) 370.

22
[30] R. M. DiGullio, R.-J. Lee S. T. Schaeffer, L. L. Brasher, A. S. Teja, J. Chem. Eng. Data, 37
(1992) 239.
[31] D. D. Wagman, W. H. Evans, V. B. Parker, R. H. Schumm, I. Halow, S. M. Bailey, K. L.
Churney, R. L. Nuttall, J. Phys. Chem. Ref. Data, Suppl. 2 (1982) 11.
[32] S.-B. Park, H. Lee, Korean J. Chem. Eng., 14 (1997) 146.
[33] J. T. Slusher, Fluid Phase Equilibria, 154 (1999) 181.
[34] K. Kojima, S. J. Zhang, T. Hiaki, Fluid Phase Equilibria, 131 (1997) 145.
[35] J. Gmehling, U. Onken, W. Arlt, Vapour-Liquid Equilibrium Data Collection, DECHMA,
Frankfurt, 1977 and onwards.
[36] P. S. Crozier, R. L. Rowley, Fluid Phase Equilibria, 193 (2002) 53.

23
24
Appendix

In the table data are shown for simulations with overlapping double-wide sampling.
In such a simulation each perturbation step is covered in both (insertion and deletion)
directions. From this total free energies are obtained for insertion and deletion, the average
of these two values would correspond to two double-wide sampling calculations. Some of
the data are the basis for the averaged results shown in table 5. Results are shown for singe
simulations.

Table 7 Free Energy of Insertion and Deletion (Results in kcal/mol)


Solvent T [ ° C] Insertion Deletion
a
water → methanol(1) waterb 25 2.49 +/-0.21 2.54 +/-0.25
watera → methanol(2) waterb 25 2.11 +/-0.19 2.11 +/-0.22
watera → methanol(3) waterb 25 2.12 +/-0.16 2.02 +/-0.15
watera → methanol(1) methanol 25 0.32 +/-0.15 0.23 +/-0.16
watera → methanol(2) methanol 25 0.70 +/-0.12 0.54 +/-0.10
watera → methanol(3) methanol 25 0.82 +/-0.15 0.77 +/-0.16
0 → UC water(1) waterb 25 2.22 +/-0.16 2.08 +/-0.17
0 → UC water(2) waterb 25 2.73 +/-0.24 2.72 +/-0.22
0 → UC water(3) waterb 25 2.57 +/-0.17 2.43 +/-0.16
0 → UC water(1) methanol 25 1.09 +/-0.09 1.11 +/-0.11
0 → UC water(2) methanol 25 1.19 +/-0.10 1.17 +/-0.11
0 → UC water(3) methanol 25 1.02 +/-0.10 0.89 +/-0.10
0 → UC water(1) ethanol 25 1.20 +/-0.16 1.18 +/-0.15
0 → UC water(2) ethanol 25 0.99 +/-0.08 0.86 +/-0.10
0 → UC water(3) ethanol 25 0.66 +/-0.06 0.66 +/-0.08
0 → UC water(1) acetonitrile 25 1.31 +/-0.12 1.18 +/-0.12
0 → UC water(2) acetonitrile 25 1.43 +/-0.14 1.24 +/-0.12
0 → UC water(3) acetonitrile 25 1.12 +/-0.11 1.07 +/-0.12
0 → UC water(1) formic acid 25 2.04 +/-0.14 1.60 +/-0.12
0 → UC water(2) formic acid 25 3.45 +/-0.18 2.91 +/-0.15
0 → UC water(3) formic acid 25 3.30 +/-0.25 2.84 +/-0.18
0 → UC water(1) ethanolamine 60 2.46 +/-0.12 2.10 +/-0.14
0 → UC water(2) ethanolamine 60 3.38 +/-0.22 2.85 +/-0.21
0 → UC water(3) ethanolamine 60 2.64 +/-0.18 2.61 +/-0.15
0 → UC water(1) waterb 60 2.71 +/-0.13 2.59 +/-0.13
0 → UC water(2) waterb 60 3.38 +/-0.22 2.85 +/-0.21
0 → UC water(3) waterb 60 2.84 +/-0.15 2.79 +/-0.16
a
TIP3P. b TIP4P.

25
Paper IV

Comparison of Quantum Mechanical and Experimental Gas Phase

Basicities of Amines

Eirik Falck da Silva

2005

J. Phys. Chem. A 109, 1603-1607.


J. Phys. Chem. A 2005, 109, 1603-1607 1603

Comparison of Quantum Mechanical and Experimental Gas-Phase Basicities of Amines and


Alcohols

Eirik F. da Silva*
Department of Chemical Engineering, Norwegian UniVersity of Science and Technology,
N-7491 Trondheim, Norway
ReceiVed: October 19, 2004; In Final Form: NoVember 25, 2004

A comparison was made between the experimental and B3LYP relative gas-phase basicities and proton affinities
of a series of 9 amine, 3 alcohol, and 3 alkanolamine molecules. While agreement is good for most of the
species studied, it is poor for the alkanolamines and 1,2-ethanediol. A series of calculations were undertaken
at the B3LYP and MP2 levels using various basis sets to see if the uncertainties in the calculations can
account for the discrepencies. The results suggest that this is unlikely and that the theoretical values are
likely to be reasonably accurate. Calculations are also presented for the dimer formation energies of
alkanolamine molecules, diamine molecules, and 1,2-ethanediol. These calculations suggest that all of these
species can form proton-bound dimers. The alkanolamines and 1,2-ethanediol also appear to have relatively
high formation energies for neutral dimers.

Introduction of diffuse basis functions.10 In this work, relatively large basis


sets will be used to limit the effect of BSSE and the results
Gas-phase basicities and proton affinities have been the from different basis sets will be compared.
subject of substantial experimental1 and theoretical2-4 work.
Experimental data are at 298 K and zero-point energies, and
These properties offer a useful benchmark for quantum me-
therefore, thermal corrections should be added to the calculated
chanical calculations, and high-level calculations have been
values. They are calculated at the HF/6-31G(d) level.
successful in reproducing experimental data. Calculations have
also been used to interpret the experimental data.5 In addition Calculations are also carried out to determine the likelihood
to the inherent interest in gas-phase basicities, they are also of the alkanolamines forming dimers in the gas phase. They
important when studying properties in solution. If gas-phase are carried out at the HF/6-311++G(d,p) level. They are
basicity, solution-phase basicity, and the free energy of solvation intended to give a quantitative picture of dimer formation. The
of the neutral solute are known, then the free energies of omission of electron correlation in the HF calculations means
solvation of the ionic form can be derived. This quantity is that the energies calculated with this method are less accurate
otherwise difficult to estimate.6 Accurate quantum mechanical than the gas-basicity calculations. For the ethanolamine, a dimer
calculation of the gas-phase basicity is also required for calculation was also carried out at the MP2/6-311++G(2d,2p)//
predicting basicities in solution. HF/6-311++G(2d,2p) level.
Amines are organic bases of importance in many contexts. All calculations were carried out in Gaussian 98.11
Our interest lies in the application of amines for the removal of
CO2 from exhaust gases. Alkanolamines are of particular interest Results and Discussion
in this context.7 In a previous study,8 poor agreement was found
Figure 1 shows the conformers that were identified as the
between the experimental and calculated gas-phase basicities
most stable at the B3LYP/6-311++G(d,p) level for the alkanol-
for some alkanolamine molecules; the present work is intended
amines, diamines, 1,2-ethanediol, and their protonated forms.
to further study the accuracy of the calculated and experimental
For the neutral forms of the three alkanolamines, the conformers
values.
are characterized by hydrogen bonding between the alcohol and
amine functionalities. The H(O)-N bond seems to be the most
Methods
energetically favored. For the protonated forms, only H(N)-O
Calculations for the gas-phase basicity have been carried out hydrogen bonds are found. The conformers of the diamines and
at the B3LYP and MP2 levels. The propensity to form 1,2-ethanediol are also characterized by intramolecular hydrogen
intramolecular hydrogen bonds8 is an important feature of the bonds. The lengths of these bonds are given in Table 1.
alkanolamines. The accurate calculation of the energies of The selected conformers are drawn from a conformer search
species containing hydrogen bonds is not trivial; one of the at the HF/3-21G(d) level.8 Some of the most stable conformers
difficulties is the basis-set superpositon error (BSSE). While identified at that level have also been studied at the B3LYP/
the counterpoise correction can be applied for bonding between 6-311++G(d,p) level, and it seems likely that these are, in fact,
different molecules, it cannot easily be applied to intramolecular the most stable conformers at this level of theory. It should,
bonds.9,10 The effect of BSSE is, however, expected to decrease however, be noted that a full study of the conformers at the
with the increasing size of the basis set and with the inclusion B3LYP/6-311++G(d,p) level has not been undertaken. Diethanol-
amine has a large number of potential conformers, and therefore,
* E-mail: silva@chemeng.ntnu.no. for the neutral form of this molecule, there is less confidence
10.1021/jp0452251 CCC: $30.25 © 2005 American Chemical Society
Published on Web 02/04/2005
1604 J. Phys. Chem. A, Vol. 109, No. 8, 2005 da Silva

Figure 1. Stable conformers of alkanolamines, diamines, and 1,2-ethanediol. Dashed lines indicate hydrogen bonds.

TABLE 1: Hydrogen Bond Lengths TABLE 2: Relative Gas-Phase Basicities and Proton
Affinitiesa
molecule bond length (Å)a
gas basicity proton affinity
ethanolamine H(O)-N 2.273
ethanolamine(H+) H1(N)-O 2.034 molecule theoreticalb experimentalc theoreticalb experimentalc
H(O1)-N 2.296 ammonia 0.0 0.0 0.0 0.0
diethanolamine
H(N)-O2 2.433 ethanolamine 16.2 18.6d 16.2 18.3d
H1(N)-O1 2.084
diethanolamine(H+) diethanolamine 28.2 24.1e 28.4 23.8e
H2(N)-O2 2.084 3-amino-1- 23.4 23.5d 23.6 26.0d
3-amino-1-propanol N-H(O) 2.033
propanol
3-amino-1-propanol(H+) H1(N)-O 1.760
1,2-ethylene- 22.8 22.3d 23.2 23.4d
1,2-ethylenediamine H1(N2)-N1 2.508
diamine
1,2-ethylenediamine(H+) H1(N1)-N2 1.906
1,3-propane- 29.9 28.9d 30.6 31.9d
1,3-propanediamine H1(N1)-N2 2.305
diamine
1,3-propanediamine(H+) H1(N1)-N2 1.684
methylamine 10.2 10.9 11.0 10.9
1,2-ethanediol H(O2)-O1 2.303
ethylamine 14.3 14.1 14.4 14.0
1,2-ethanediol(H+) H1(O1)-O2 1.642
dimethylamine 17.9 18.5 18.0 18.1
a trimethylamine 22.2 23.7 22.3 22.8
B3LYP/6-311++G(d,p) geometry.
piperidine 24.2 24.4 24.3 24.0
in the determination of the most stable conformer. The conform- piperazine 23.3 22.9 23.5 21.5
ers for neutral ethanolamine are consistent with those found, morpholine 17.0 17.3 17.2 16.9
both theoretically and experimentally, by other authors (Tuber- pyrrolidine 23.6 23.0 23.7 22.6
methanol -23.1 -22.6f -23.2 -23.7f
gen et al.13 and Vorobyov et al.12 and the references therein). ethanol -17.0 -17.4 -17.3 -18.5
The conformer identified for 3-amino-1-propanol is in agreement 1,2-ethanediol -13.1 -10.9g -12.6 -9.0g
with theoretical work by Kelterer and Ramak.14 a Results in kcal/mol. b B3LYP/6-311++G(d,p) energy with thermal
In Table 2 the relative basicity of these amines together with correction and zero-point energy calculated at the HF/6-31G(d) level.
other amine and alcohol molecules calculated at the B3LYP/ c Data from Hunter and Lias,1 also available at webook.nist.gov.15
6-311++G(d,p) level is shown. Data are given relative to Original papers indicated for alkanolamines, diamines, and 1,2-
ammonia. For piperazine, calculations were done on the chair ethanediol. d Data from Meot-Ner et al.16 e Data from Sunner et al.17
f Value is theoretical. g Data from Chen and Stone.18
conformer. The conformers of ethanol are shown in the
Supporting Information, and the other molecules have only one
conformer form. for diethanolamine. The relative basicity for these two alkanol-
The data in Table 2 show good agreement for most of the amines differs by 6.5 kcal/mol. A similar trend can be seen for
molecules. For the alkanolamines, there is, however, a significant the proton affinities of these molecules. For 3-amino-1-propanol,
disagreement between the experimental data and the calculated the final alkanolamine in this study, there is also a considerable
results. The experimental and theoretical gas-phase basicities difference between the experimental and theoretical proton
differ by 2.4 kcal/mol for ethanolamine and by 4.1 kcal/mol affinities. However, the gas-phase basicities, in this case, are
Gas-Phase Basicities of Amines and Alcohols J. Phys. Chem. A, Vol. 109, No. 8, 2005 1605

in better agreement. For 1,2-ethanediol, significant differences


can also be seen between the experimental and theoretical
values.
As the largest deviations occur for molecules displaying
intramoleculer hydrogen bonds, this would suggest that there
might be errors in the calculation of strengths of these bonds.
To explore the method and basis-set dependency of the
results, the basicities of the alkanolamines, 1,2-ethanediol, and
ammonia were calculated with different basis sets and using
both the MP2 and B3LYP level of theory. The results are shown
in Table 3.
From the values in Table 3, one can see that there is some
variation in the results with variation in the level of theory and
size of the basis set. However, the B3LYP/6-311++G(d,p)
results are in fairly good agreement with MP2 calculations with
larger basis sets. More importantly, the relative basicity of the
alkanolamines and 1,2-ethanediol remains almost unchanged.
Therefore, it seems unlikely that there is any level of quantum
mechanical calculation that will be in agreement with the
experimental data. The MP2/Aug-cc-pVTZ//B3LYP/6-311++G-
(d,p) calculations use the largest basis set, and these results are
probably the most accurate.
Gas-Phase Dimer Formation. Quantum mechanical calcula-
tions were performed to investigate the likelihood of the
alkanolamines, diamines, and 1,2-ethanediol forming dimers.
Calculations are carried out both for proton-bound dimers and
neutral dimers.
The initial geometries were based on conformers that would
give the largest number of hydrogen bonds, in particular the
H(O)-N-type bonds that appear to be the most energetically
favored for alkanolamines. The determination of conformers was
not based on any rigorous exploration of the potential dimers
of these molecules; for diethanolamine, in particular, there are
a large number of potential dimers. The ethanolamine dimer is
the same as that reported to be the most stable in calculations
by Vorobyov et al.12 The 1,2-ethanediol dimer geometry is from Figure 2. Dimers optimized at the HF/6-311++G(d,p) level. Dashed
lines indicate hydrogen bonds.
work by Bako et al.19
Calculations are carried out at the HF/6-311++G(d,p) level, that 1,2-ethanediol had a stronger propensity to form proton-
and only the binding energy of the dimer is calculated. This is bound dimers than diols with longer carbon chains. A similar
the energy difference between the dimers and the monomers trend can be seen in Table 5; the results suggest that 1,3-
calculated at the same level. The dimers are shown in Figure 2 propanediamine forms a weaker proton-bound dimer than 1,2-
and Figure 3, and the hydrogen bond lengths are given in Table ethanediamine. Apparently, 3-amino-1-propanol also forms a
4. The dimer formation energy is given in Table 5. less stable proton-bound dimer than ethanolamine. It appears
The geometries and the data in Table 5 suggest that all of that molecules with longer carbon chains can form less strained
these molecules form stable proton-bound dimers. They, ap- intramolecular hydrogen bonds, and therefore, the additional
parently, can act as bidentate ligands to a protonated molecule. stability gained by bonding to a second molecule is less.
Proton-bound dimers have also been observed experimentally At the MP2/6-311++G(2d,2p)//HF/6-311++G(2d,2p) level,
for diethanolamine17 and 1,2-ethanediol.18 In the experimental the dimer binding energy of ethanolamine is calculated to be
work on 1,2-ethanediol,18 it was also proposed that one molecule -11.00 kcal/mol. As a comparison, the dimer formation energy
acts as a bidentate ligand. In the same paper, it was observed of water has been calculated to be -5.18 kcal/mol using a

TABLE 3: Gas-Phase Basicities and Proton Affinities Relative to Ammoniaa


ethanolamine diethanolamine 3-amino-1-propanol 1,2-ethanediol
B3LYP/6-311++G(d,p)//b 16.2 16.2 28.2 28.4 23.4 23.6 -13.1 -12.6
B3LYP/6-311++G(d,p)
B3LYP/6-311++G(3df,2p)// 16.8 16.8 28.7 28.9 23.8 24.0 -12.3 -11.8
B3LYP/6-311++G(3df,2p)
MP2/6-311++G(d,p)// 14.8 14.8 26.7 26.9 21.9 22.0 -14.6 -14.1
B3LYP/6-311++G(d,p)
MP2/6-311++G(2d,2p)// 15.9 15.9 27.7 27.9 22.9 23.0 -13.9 -13.4
B3LYP/6-311++G(d,p)
MP2/Aug-cc-pVTZ// 16.1 16.1 28.0 28.2 23.0 23.2 -12.9 -12.5
B3LYP/6-311++G(d,p)
experimental 18.6 18.3 24.1 23.8 23.5 26.0 -10.9 -9.0
a Results in kcal/mol. b Same data as in Table 1.
1606 J. Phys. Chem. A, Vol. 109, No. 8, 2005 da Silva

TABLE 4: Dimer Hydrogen Bond Lengths


dimer bond length (Å) dimer(H+)a bond length (Å)
ethanolamine N1-H(O2) 2.119 ethanolamine O1-H1(N1) 2.249
H(O1)-N2 2.118 H2(N1)-N2 1.913
H1(N1)-O1 2.628 H3(N1)-O2 2.092
H2(N2)-O2 2.627 diethanolamine H1(N1)-O4 2.321
diethanolamine H(N1)-O4 2.361 H2(N1)-N2 2.112
N1-H(O3) 2.017 H2(N1)-O3 2.319
H(O1)-O3 2.094 O1-H(O3) 2.505
H(O2)-N2 2.100 H(O2)-O4 2.109
H(O4)-O2 2.065 3-amino-1-propanol O1-H3(N1) 1.989
3-amino-1-propanol N1-H(O2) 2.080 H1(N1)-N2 1.965
H(O1)-N2 2.069 H2(N1)-O2 1.993
H1(N1)-O1 2.252 1,2-ethanediamine N1-H1(N2) 2.285
H2(N2)-O2 2.295 H2(N2)-N3 1.999
1,2-ethanediamine N1-H3(N3) 2.560 H3(N2)-N4 2.041
H1(N1)-N2 2.625 1,3-propanediamine N1-H1(N2) 2.069
H2(N2)-N4 2.544 H2(N2)-N3 1.967
H4(N4)-N3 2.776 H3(N2)-N4 2.003
1,3-propanediamine N1-H3(N3) 2.405 1,2-ethanediamine O1-H1(O2) 2.286
H1(N1)-N2 2.399 H1(O2)-O3 1.789
H2(N2)-N4 2.447 H2(O3)-O4 1.632
H4(N4)-N3 2.336
1,2-ethanediamine H(O1)-O4 2.051
O1-H(O3) 2.155
H(O1)-O2 2.699
O2-H(O4) 2.160
H(O3)-O4 2.748
a
Proton-bound dimer.

effects may have affected the experimental results for all of


these molecules. The relatively low volatility of these alkanol-
amines may also make the experimental determination of their
gas-phase basicities more difficult. Therefore, it appears that
there is considerable uncertainty regarding the experimental data
for alkanolamines and 1,2-ethanediol.

Figure 3. 1,2-Ethanediol dimers optimized at the HF/6-311++G(d,p) Conclusion


level. Dashed lines indicate hydrogen bonds. Gas-phase basicities from quantum mechanical calculations
TABLE 5: Dimer Binding Energiesa are generally shown, in this article and work of other authors,
to be in good agreement with experimental data. However, poor
neutral dimer proton-bound dimer
agreement was observed for a series of alkanolamines and 1,2-
ethanolamine -5.9 -25.2 ethanediol. Calculations using the MP2 and B3LYP methods
diethanolamine -5.3 -14.8 with different basis sets revealed some method and basis-set
3-amino-1-propanol -4.7 -23.2
1,2-ethanediamine -2.3 -26.2 dependency in the results, but these uncertainties cannot account
1,3-propanediamine -2.9 -23.2 for the differences between the calculated results and the
1,2-ethanediol -5.4 -32.2 experimental data. Therefore, the current MP2/Aug-cc-pVTZ
a
Results in kcal/mol.
results are probably the most reliable estimate of the basicity
of these molecules.
Calculations on dimer forms suggest that the alkanolamines,
similar level of theory.20 This would suggest that the alkanol-
diamines, and 1,2-ethanediol form stable proton-bound dimers.
amines and 1,2-ethanediol have relatively stable dimers. The
Less stable neutral dimers can also be formed. Dimer effects
diamines appear to form weaker dimers.
and low volatility suggest high uncertainty in the experimental
Experimental Values basicities for the alkanolamines and 1,2-ethanediol.

The experimental basicity and proton affinity data for Acknowledgment. This work has received support from The
ethanolamine, 3-amino-1-propanol, 1,2-ethanediamine, and 1,3- Research Council of Norway (Program for Supercomputing)
propanediamine are from pulsed high-pressure mass spectrom- through a grant of computing time.
etry work by Meot-Ner et al.16 The diethanolamine data are
from fast atom bombardment mass spectroscopy work by Sunner Supporting Information Available: Underlying values for
et al.,17 and the 1,2-ethanediol data are from pulsed electron data in Table 2 and Table 3, geometric parameters and Cartesian
beam high-pressure mass spectrometry experiments.18 It was coordinates of the alkanolamines, and illustrations of other
noted in the work on 1,2-ethanediol that the presence of proton- conformers of diethanolamine. This material is available free
bound dimers created problems in determining the proton of charge via the Internet at http://pubs.acs.org.
affinity and basicity of this molecule. The present work suggests
that all of the alkanolamines have a comparable propensity to References and Notes
form dimers. While the basicity of these molecules was (1) Hunter, E. P. L.; Lias, S. G. J. Phys. Chem. Ref. Data 1998, 27,
determined with different experiments, it seems that dimer 413.
Gas-Phase Basicities of Amines and Alcohols J. Phys. Chem. A, Vol. 109, No. 8, 2005 1607

(2) Smith, B. J.; Radom, L. J. Am. Chem. Soc. 1993, 115, 4885. Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Challacombe, M.; Gill, P.
(3) East, A. L. L.; Smith, B. J.; Radom, L. J. Am. Chem. Soc. 1997, M. W.; Johnson, B.; Chen, W.; Wong, M. W.; Andres, J. L.; Gonzalez, C.;
119, 9014. Head-Gordon, M.; Replogle, E. S.; Pople, J. A. Gaussian 98, revision A.9;
(4) Pokon, E. K.; Liptak, M. D.; Feldgus, S.; Shields, G. C. J. Phys. Gaussian, Inc.: Pittsburgh, PA, 1998.
Chem. A 2001, 105, 10483. (12) Vorobyov, I.; Yappert, M. C.; DuPre, D. B. J. Phys. Chem. A 2002,
(5) Pople, J. A.; Curtiss, L. A. J. Phys. Chem. 1987, 91, 155. 106, 668.
(6) Pearson, R. G. J. Am. Chem. Soc. 1986, 108, 6109. (13) Tubergen, M. J.; Torok, C. R.; Lavrich, R. J. J. Chem. Phys. 2003,
(7) Versteeg, G. F.; van Dijck, L. A. J.; van Swaaij, W. P. M. Chem. 119, 8397.
Eng. Commun. 1996, 144, 113. (14) Kelterer, A. M.; Ramak, M. THEOCHEM 1991, 232, 189.
(8) da Silva, E. F.; Svendsen, H. F. Ind. Eng. Chem. Res. 2003, 42, (15) NIST Chemistry WebBook, NIST Standard Reference Database
4414. Number 69; Linstrom, P. J., Mallard, W. G., Eds.; National Institute of
(9) Reiling, S.; Brickmann, J.; Schlenkrich, M.; Bopp, P. A. J. Comput. Standards and Technology: Gaithersburg, MD, http://webbook.nist.gov,
Chem. 1996, 17, 133. March 2003.
(10) Lii, J. H.; Ma, B.; Allinger, N. L. J. Comput. Chem. 1999, 20,
1593. (16) Meot-Ner(Mautner), M.; Hunter, E. P.; Hamlet, P.; Field, F. H.
(11) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, Proceedings of the 28th Annual Conference on Mass Spectrometry and
M. A.; Cheeseman, J. R.; Zakrzewski, V. G.; Montgomery, J. A., Jr.; Allied Topics, May 25-30, 1980, 233.
Stratmann, R. E.; Burant, J. C.; Dapprich, S.; Millam, J. M.; Daniels, A. (17) Sunner, J. A.; Kulatunga, R.; Kebarle, P. Anal. Chem. 1986, 58,
D.; Kudin, K. N.; Strain, M. C.; Farkas, O.; Tomasi, J.; Barone, V.; Cossi, 1312.
M.; Cammi, R.; Mennucci, B.; Pomelli, C.; Adamo, C.; Clifford, S.; (18) Chen, Q. F.; Stone, J. A. J. Phys. Chem. 1995, 99, 1442.
Ochterski, J.; Petersson, G. A.; Ayala, P. Y.; Cui, Q.; Morokuma, K.; Malick, (19) Bako, I.; Grosz, T.; Palinkas, G.; Bellissent-Funel, M. C. J. Chem.
D. K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Cioslowski, J.; Phys. 2003, 118, 3215.
Ortiz, J. V.; Baboul, A. G.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz, (20) Frisch, M. J.; Del Bene, J. E.; Binkley, J. S.; Schaefer, H. F., III.
P.; Komaromi, I.; Gomperts, R.; Martin, R. L.; Fox, D. J.; Keith, T.; Al- J. Chem. Phys. 1986, 84, 2279.
Paper V

Study of the Carbamate Stability of Amines Using ab Initio Methods

and Free-Energy Perturbations

Eirik Falck da Silva and Hallvard F. Svendsen

2005

Accepted in Ind. Eng. Chem. Res.


Study of the Carbamate Stability of Amines Using ab Initio Methods and Free Energy
Perturbations

Eirik F. da Silva* and Hallvard F. Svendsen

Department of Chemical Engineering, Norwegian University of Science and Technology, N-


7491 Trondheim, Norway

Abstract

The relative carbamate stability of a series of amines used in CO2 absorption


processes have been studied with different solvation models and gas phase energies
calculated with the B3LYP density functional method. The solvation energies were
calculated with Monte Carlo free energy perturbations and continuum models.
Comparison between calculated energies and experimental NMR and kinetic data
shows reasonable agreement. The trends in carbamate stability can apparently not be
explained in terms of any single molecular characteristic.

Keywords: ab initio calculations, solvation energy, carbamate, amines

Introduction
As a measure for preventing global warming there is a steadily increasing interest in
methods for removing carbon dioxide from exhaust gases. Absorption with
alkanolamines in mixtures with promoters has traditionally been used for the removal
of carbon dioxide from natural gas and refinery gas and the same technology is an
option for the treatment of exhaust gases. For high-pressure applications, N-
methyldiethanolamine (MDEA) based systems have been used successfully for many
years. For exhaust gases, the most common amine has been ethanolamine (MEA).
The high energy demand for regeneration and relatively high degradation rates for this
amine are however unfavorable for large fossil fuel power plants. During the last
years, new systems like the PSR 1-31 and Mitsubishi KS 1-32 have been developed
and promise improved performance compared to the conventional MEA. At the same
time it should also be noted that improvements in performance have been reported for
MEA.3

*Tel.: +47 73594125, Fax.: +47 73594080 e-mail: silva@chemeng.ntnu.no


1
In the chemical absorption of CO2 in aqueous amine systems the CO2 is bound
as either bicarbonate or carbamate.4 If the equilibrium constants governing the
formation of these species is known or can be predicted the overall performance of a
solvent can to a large extent be predicted.
For the formation of carbamate there is the following reaction:4
CO2 + R2 NH + B U R2 NCO2− + BH + (1)

where B indicates a base molecule and R2 NH is any primary or secondary amine


molecule. For bicarbonate formation there is the following reaction:
CO2 + 2 H 2O U HCO3− + H 3O + (2)
While no amine molecule appears in equation 2, the extent to which this reaction will
proceed is in fact governed by the strength of the amine as a base:
BH + + H 2O U B + H 3O + (3)

The present work will deal with the modeling of carbamate stability. A number of
models are available for the base strength. If the carbamate stability can be modeled
in similar fashion the performance of different amine solvents can be predicted.
For the carbamate formation an alternative equilibrium can be set up that does
not involve the base molecule:
HCO3− + R2 NH U R2 NCO2− + H 2O (4)
If the mole fraction based activity of water is assumed to be 1 and if H3O+ is written
as H+ the following equilibrium constant is obtained for equation 3:
aB a H +
Ka = (5)
aBH +

Similarly, the following equilibrium constant is obtained for carbamate formation


(equation 4):
aR NCO−
Kc = 2 2
(6)
aR2 NH aHCO−
3

The carbamate equilibrium constant from equation 1 (Kc2) can be expressed as a


product of the equilibrium constants of equation 2,3 and 4:
Kc K2
Kc2 = (7)
Ka

2
where K2 is the equilibrium constant of equation 2. The interactions between an amine
species and CO2 in solution can therefore be described by two equilibrium constants:
Ka and Kc. While there are also other reactions that take place4 these are independent
of the amine present in the system.
From knowledge of these two equilibrium constants the amount of CO2
captured and energy consumption of the process can be estimated. As will be shown
in the present work knowledge of the equilibrium constants can also be used to predict
the reaction rates. Models that can predict these equilibrium constants are therefore
likely to be very useful in efforts to improve the absorption process in terms of energy
consumption and overall efficiency.
The equilibrium constant Ka is usually presented in the form of the pKa
( pK a = − log K a ). Well-established methods exist for the experimental determination
of the pKa and the determination can be done with a fairly high degree of accuracy.
Substantial data on the pKa for alkanolamines and other organic bases are also
available in the literature.5
For the carbamate stability (Kc) the situation is however very different. The
carbamate species only appear in significant concentrations in systems where
several other species are also present in significant concentrations. The carbamate
species is also in equilibrium with solvated CO2. Because the solvated CO2 is only
present in small quantities its concentration can be difficult to establish. The only
direct experimental route to obtaining this carbamate equilibrium constant appears to
be NMR-techniques. The NMR-experiments are however very demanding and data
has only been published for a small number of molecules.6-8 The uncertainty in the
results are also much larger than for pKa measurement. To compare NMR results
obtained under different conditions, estimates for activity coefficients are also
required. As already noted these systems have many components, several of them
ionic, making the estimation of activity coefficients challenging. In summary it can be
observed that the experimental determination of carbamate stability is a more
demanding task than the determination of the pKa and the quality of the results that
can be obtained is inferior.
Although the experimental tasks of determining pKa and carbamate stability
are very different, the modeling tasks are very similar. In both cases one is attempting
to calculate the free energy difference in aqueous solution between an ionic species

3
and a closely related neutral species. The present authors have used quantum
mechanical calculations and free energy perturbations to estimate trends in the pKa. 9
In the present work we will use essentially the same models to predict trends in
carbamate stability. While some theoretical work has been published on carbamate
species10-13 the present work is to our knowledge the first modeling work on relative
carbamate stability.

4
Methods
For equation 4 the following thermodynamic cycle can be set up:

On the basis of the thermodynamic cycle the reaction energy in solution ∆Gcs can be

divided into two contributions.


∆Gcs = ∆Gcg + ∆∆Gs (8)

where ∆Gcg is:

∆Gcg = Gg ( R2 NCO2− ) + Gg ( H 2O) − Gg ( R2 NH g ) − Gg ( HCO3− ) (9)

and
∆∆Gs = ∆Gs ( R2 NCO2− ) + ∆Gs ( H 2O ) −∆Gs ( R2 NH g ) −∆Gs ( HCO3− ) (10)

Molecular mechanics (MMFF) was used to generate an initial set of conformers. All
conformers were then optimized at HF/6-31G(d) level. Separate conformer searches
were done for the amines and the carbamate forms. The conformers found to be the
most stable in the gas phase were also assumed to be the most stable in solution and
all calculations in this work were performed on the same set of conformers. Gas phase
conformer search calculations were performed with PC Spartan Pro version 1.0.7.16
Calculation of gas phase reaction energies were carried out at B3LYP/6-
311++G(d,p) and MP2/6-31G(d) levels. Our previous work with pKa9 suggest that the
B3LYP/6-311++G(d,p) method produces results in better agreement with
experimental data. It has also been noted in the literature14 that this level of theory
reproduces experimental properties quite well. The presence of intramolecular
hydrogen bonds in the amines identified in our previous work means that they can not
reliably be modeled with the small basis-set in the MP2 calculation. B3LYP and MP2
calculations were performed with their implementations in Gaussian 98.15

5
Thermal corrections and zero-point energies have been calculated at HF/3-
21G(d) level. While this method is not very accurate it should be noted that these
contributions to the gas phase energy are relatively small. These calculations were
also performed with Gaussian 9815 implementations.
In the calculation of solvation energy three different models are used. One is a
free energy perturbation scheme while the two others are widely used continuum
models.
The free energy perturbation scheme is similar to what we have previously
used in the modeling of pKa. A approach that was modeled on work by Wiberg et
al.17 In the present work free energy perturbations have been performed as two
separate series, one between the neutral form of the amines, the second between the
carbamate forms of the molecules. All perturbations were therefore between species
of the same charge, this is the same form as used by Wiberg et al.17 and more accurate
than the form used in our previous work on pKa.
In the present work simulations were carried out on rigid gas phase B3LYP/6-
311++G(d,p) geometries. The intermolecular interactions between two molecules a
and b were evaluated using Coulomb and Lennard-Jones terms:

⎪ q q e2 ⎡ 12
⎛ σij ⎞⎟ ⎥⎤⎪⎫
6


on a on b
⎪ ⎢⎜⎛ σij ⎞⎟⎟ ⎜
∆Eab = ∑∑ ⎨ + 4εij ⎢⎜⎜ ⎟ − ⎜⎜ ⎟⎟ ⎥⎪⎬
i j
(11)
⎪ rij
j ⎪ ⎢⎝⎜ rij ⎠⎟ ⎝⎜ rij ⎠⎟ ⎥⎪
⎦⎪
i

⎩ ⎣ ⎪

The Lennard-Jones σ and ε for alcohol, alkane and amine groups were from the
OPLS-All atom force field.18, 19 For CO2 the EPM220 parameters were chosen: σ (C)=
2.757 Å, σ (O)= 3.033 Å, ε (C)= 0.066 [kcal/mol] and ε (O)= 0.170 [kcal/mol].
Partial atomic charges were determined by using solvent phase (SM 5.42R21 ) CM222
charges implemented in the Gamesol23 program. These charges were calculated on
HF/6-31G(d) geometries. It should be noted that atomic charges are not uniquely
defined and a number of schemes for their calculation are available.24, 25
The
calculation of charges is one of the main uncertainties when using simulations to
determine free energies.
The perturbations were carried out between the amines closest in size. To
allow smooth transitions no more than one non-hydrogen atom was deleted in a single
simulation. Series of up to three simulations were therefore used to transform larger
solutes to smaller ones.

6
These calculations were performed with BOSS version 4.126 using procedures
developed by Jorgensen et al.27 A single solute molecule was placed in a periodic
cube with 267 TIP4P water molecules at 25°C and 1 atmosphere in the NPT
ensemble. A number of water molecules corresponding to the number (n) of non-
hydrogen atoms in the amine molecule were removed, giving 267 - n water
molecules. The perturbations were carried out over ten windows of double-wide
sampling giving 20 free energy increments that are summed up to give the total
change in free energy of solvation. Each window had 500000 steps for equilibration
and another 500000 for sampling. The present free energy perturbations have a
statistical uncertainty of around ±1 kcal/mol.
This combination of B3LYP gas phase energies and solvation energies from
free energy perturbations is essentially the same as the one presented in our work on
pKa. It should be emphasized that the model involves no form of adjustment to obtain
agreement with experimental data.
The first of the two continuum models used to calculate the solvation energies
is the IEFPCM model.28 Calculations were carried out with 60 tesserae per atomic
sphere and other default settings in Gaussian 98. Calculations were performed as
single point calculations on gas phase MP2/6-31G(d) geometries (IEFPCM/ MP2/6-
31G(d)// MP2/6-31G(d)). This model was also utilized in our work on pKa modeling.9
The other continuum model is the SM 5.42R21 model, in this case single-point
calculations on the HF/6-31G(d) level were carried out (SM 5.42R/ HF/6-31G(d)//
HF/6-31G(d)). The IEFPCM and SM 5.42R calculations were carried out in Gaussian
9815 and Gamesol23 respectively. The IEFPCM model has also been used to determine
the solvation energies of HCO3− , CO2 and H 2O .

Results
In Table 1 are listed the molecules studied. In the table it is indicated weather amines
are primary or secondary. Tertiary amines do not form carbamate species and are
therefore not included in the present study. It is also indicated in the table if the
molecule has any substituent-groups on the carbon(s) adjacent to the amine. This
particular characteristic has been used to rationalize carbamate-stability and we will
look at this characteristic in the discussion. Experimental pKa’s are also given in
Table 1, they will be used to convert data from Kc to Kc2 and they are also relevant to

7
the discussion of explanations given for carbamate stability. Finally we have given
some references to the amines in the CO2 absorption literature and included references
to NMR-results when these are available.

Table 1. Experimental pKa data

No. Compd Name Typea Exptl pKa Kinetics/NMR


(25 C) references

1 Ethanolamine MEA p 9.51b 4, 31, 7,e 8e


2 3-Amino-1-Propanol MPA p 9.96 b 4
3 2-(2-Aminoethoxy)Ethanol DGA p 9.46c 4
b, d
4 Ethylendiamine EDA p 9.92 31
b
5 1-Amino-2-Propanol MIPA p 9.46 4

6 N-(2-Hydroxyethyl)Ethylenediamine AEEA p, s 9.82 b, d


7 2-Amino-2-Methylpropanol AMP p* 9.7 b 31, 7e
8 2-Amino-2-Methyl-1,3-Propanediol AMPD p* 8.8 b 31, 6e, 7e
9 2-Amino-2-Ethyl-1,3-Propanediol AEPD p* 8.8 b 7e
10 Diethanolamine DEA s 8.95 b 4, 6e

11 Diisopropanolamine DIPA s 8.89 b 4


b
12 2-(Methylamino)Ethanol MMEA s 9.77 4, 8e
13 Morpholine s, c 8.49 b 31
14 Piperazine s, c 9.83 b, d 30

15 Piperidine s, c 11.12 b 31

a: p: primary amine, s: secondary amine, c: cyclical amine, *: subsituent group on atom neighboring
the amine functionality. b: Data from Perrin.5 .c: Data from Littel et al.29 (extrapolated from data at
20C). d: The first protonation constant. e: Reference to NMR data.

The compounds chosen for this study covers a fairly large part of the amines
commonly used in CO2 absorption. Almost all amines covered in a review by
Versteeg et al.4 are included. In addition we have attempted to cover amines in the
literature that seem promising or that display particularly interesting behavior.
The geometry of the most stable conformers found for the carbamate forms of
MEA, 2-amino-2-methylpropanol (AMP) and piperidine are shown in Figure 1. The
most stable carbamate conformers of the other amine molecules are shown in Figure
2. For N-(2-hydroxyethyl)ethylenediamine (AEEA) conformers are shown for
bonding to both amine functionalities. The conformers of the amines themselves are

8
shown in the supporting information, most of these conformers were also presented in
our previous work on pKa.9

Figure 1. Carbamate forms of ethanolamine (MEA), 2-amino-2-methylpropanol (AMP) and piperidine.


Stippled lines indicate hydrogen bonds.

Figure 2. Carbamate forms of amines. Stippled lines indicate hydrogen bonds.

9
To look at the most likely form of intramolecular hydrogen bonding for carbamate
species the potential conformer forms of 3-amino-1-propanol (MPA) have been
studied in greater detail. There can be conformers with hydrogen bonding between
alcohol-group and carbamate oxygens, between alcohol-group and nitrogen atom and
finally there are conformers without any hydrogen bonds. In Figure 3 the conformers
identified as the most stable of each type in the gas phase are shown and energies for
these conformers are shown in the tables. In Table 4 reaction energies are shown for
these different carbamate conformers of MPA (all calculated relative to the same
conformer of MPA itself). Results with different solvation models all suggest that the
carbamate conformer found to be most stable in gas phase (Table 2) remains the most
stable in solution.

Figure 3. Conformers of 3-amino-1-propanol (MPA) displaying different intramolecular hydrogen bonds.

In Figure 2 it can be seen that the carbamate molecules tend to form intramolecular
hydrogen bonds between alcohol-group hydrogen atoms and CO2-group oxygens. For
the amines themselves we found intramolecular hydrogen bonding between the
amine-groups and the alcohol-groups.
In the present work we assumed that these conformers with intramolecular
hydrogen bonds also dominate in solution. It is however clear that in solution this

10
intramolecular hydrogen bonding competes with hydrogen bonding to the solution
water-molecules.
Modeling32 and experimental work33, 34
for ethanolamine suggest that in the
liquid form or aqueous solution the conformers change compared to the gas phase, the
geometry changes to allow greater formation of intermolecular hydrogen bonds. The
results do however also suggest that intramolecular hydrogen bonding remains. Work
on 2-(methylamino)ethanol12 (MMEA) suggest that this molecule also has some
degree of intramolecular hydrogen bonding in solution. The same work also finds
intramolecular hydrogen bonds in solution for the carbamate species. This is
consistent with the present model results for MPA. In summary this would suggest
some degree of conformer change for neutral amines in aqueous solution compared to
the gas phase, less so for the carbamate species. Using conformers identified as the
most stable in the gas phase is therefore an approximation, but since the conformers
are likely to be close in energy this should not have a too large effect on calculated
energies.
In Table 2 are shown the results for gas phase reaction energies ( ∆Gcg )

calculated at B3LYP and MP2 level. Considerable differences can be seen MP2 and
B3LYP gas phase results. The differences are however comparable to what we found
using the same levels of theory in our work on base-strength.9 As noted in the
Methods Section we believe the B3LYP results to be the more accurate.

11
Table 2. Calculated Gas Phase Reaction Energies

No. Name Gcg

MP2a B3LYPb
1 MEA -7.79 -6.26

2 MPA(1)c -8.50 -5.83


MPA(2)c -2.98 -0.90
MPA(3)c 6.19 3.31

3 DGA -5.60 -3.55

4 EDA -2.60 0.28

5 MIPA -7.99 -5.11

6 AEEA(p) -8.12 -4.56


AEEA(s) -13.41 -6.87

7 AMP -4.09 -1.13

8 AMPD -12.18 -8.39

9 AEPD -13.13 -8.54

10 DEA -19.66 -12.99

11 DIPA -20.55 -13.10

12 MMEA -9.14 -5.46

13 Morpholine -4.52 -2.26

14 Piperazine -3.20 -0.53

15 Piperidine -2.26 2.14

a: MP2/6-31G(d) with thermal correction and zero-point energy at HF/3-21G(d) level. b: B3LYP/6-
311++G(d,p) level with thermal correction and zero-point energy at HF/3-21G(d) level. c: With
different carbamate conformers as shown in Figure 3.

In Table 3 are shown the solvation energies calculated with free energy
perturbations and the continuum models. The continuum and free energy perturbation
solvation energies are mostly in reasonable agreement, but do differ significantly in
some cases. Finally in Table 4 is shown the solvation phase energy for equation 1 and
equation 4. The energies are based on the B3LYP gas phase energies and separate sets
of results are shown based on the different solvation models. To calculate the energy
for Kc2 (equation 1) experimental pKa data were used together with model data for
equation 2 (given in the supporting information).

12
Table 3. Solvation Energies (All Results in kcal/mol)

No. Name FEPa PCMb SMc

Am AmCO2- Am AmCO2- Am AmCO2-


1 MEA -8.9 -67.8 -8.9 -67.8 -9.0 -72.3
d
2 MPA(1) -8.7 -64.1 -8.6 -66.8 -9.2 -70.7
MPA(2)d -63.0 -68.0 -71.8
MPA(3)d -70.5 -75.9 -78.1
3 DGA -12.9 -68.9 -10.5 -63.9 -11.3 -68.9
4 EDA -11.8 -85.7 -9.1 -69.2 -9.3 -74.3
5 MIPA -10.2 -68.9 -8.1 -66.0 -8.4 -70.7
6 AEEA(p) -10.3 -46.3 -11.5 -64.4 -11.0 -71.4
AEEA(s) -54.0 -62.3 -67.4
7 AMP -6.8 -65.0 -7.3 -64.4 -6.8 -69.1
8 AMPD -7.7 -59.9 -8.7 -62.5 -9.9 -68.3
9 AEPD -6.9 -58.4 -7.7 -60.9 -9.3 -67.2
10 DEA -11.2 -60.6 -9.8 -60.0 -12.9 -66.0
11 DIPA -8.1 -58.3 -10.5 -57.3 -10.6 -64.0
12 MMEA -5.7 -68.8 -8.1 -64.3 -7.4 -69.6
13 Morpholine -4.4 -76.7 -9.1 -70.2 -7.2 -72.3
14 Piperazine -5.7 -86.4 -11.0 -72.7 -9.1 -75.3
15 Piperidine -2.7 -81.8 -5.4 -68.5 -4.3 -71.4

a: Free Energy Perturbations with CM2 charges on B3LYP/6-311++G(d,p) geometry. b:


IEFPCM/MP2/6-31G(d)//MP2/6-31G(d). c: SM 5.42R/HF/6-31G(d)// HF/6-31G(d). d: With different
carbamate conformers as shown in Figure 3.

13
Table 4. Free Energies of Reaction in solution (All Results in kcal/mol)

No. Name ∆Gcs a, b ∆Gcs 2 a, c, d


FEP PCM SM FEP PCM SM

1 MEA 1.9 1.9 -2.5 13.6 13.6 9.2


e
2 MPA(1) 5.9 3.1 -0.2 17.0 14.2 10.9
e
MPA(2) 11.9 6.9 3.6 23.0 18.0 14.7
MPA(3)e 8.7 3.2 1.4 19.8 14.3 12.5
3 DGA 7.5 10.2 6.0 19.3 21.9 17.8
4 EDA -6.5 7.3 2.4 4.7 18.5 13.5
5 MIPA 3.3 4.0 -0.3 15.1 15.8 11.5
6 AEEA(p) 26.5 9.7 2.1 37.8 21.0 13.4
AEEA(s) 16.5 9.4 3.8 27.8 27.6 22.0
7 AMP 7.8 8.8 3.6 19.3 20.2 15.0
8 AMPD 6.6 4.8 -2.3 19.2 17.5 10.4
9 AEPD 7.1 5.4 0.7 19.7 18.1 13.4
10 DEA 4.7 3.9 1.0 17.1 16.4 13.5
11 DIPA 3.8 7.3 0.6 16.4 19.8 13.2
12 MMEA -1.4 5.5 -0.6 9.9 16.8 10.8
13 Morpholine -7.5 3.7 -0.2 5.6 16.8 12.9
14 Piperazine -12.5 4.8 0.3 -1.2 16.1 11.6
15 Piperidine -11.4 6.1 2.2 -1.9 15.7 11.7

a: Gas phase energies are B3LYP/6-311++G(d,p) level with thermal correction and zero-point energy
at HF/3-21G(d) level. b: Reaction energy for eq 4. c: Reaction energy for eq 1. d: Experimental pKa
data used in calculation. e: With different carbamate conformers as shown in Figure 3.

AEEA is special among the amines in present study in having two non-equivalent
amine-functionalities. In Table 4 equilibrium constants are calculated based on
bonding to both amine functionalities. It can be seen from the table that while the free
energy perturbation results suggest that CO2 bonded to the secondary amine
functionality produces the more stable carbamate form, the SM results suggest that
primary carbamate is more stable. All the models suggest that AEEA has a lower
equilibrium constant for carbamate formation than MEA. The models do however
differ on how great this difference is.

14
It can also be seen from Table 4 that different solvation models give different
relative energies for many of the other amines. The assessments of the accuracy of the
different models will be based on comparison with experimental data.
Before turning to comparison with experimental data some general comments
should be made on the quality of the quantum mechanical calculations and solvation
energies. In our study of amine pKa values9 we obtained results that can perhaps best
be summarized as semi-quantitative. In the present study the method used is almost
identical. The carbamate formation reaction is however between an anionic species
and a neutral species, while the pKa study involved cationic molecules. There are
some reasons to believe that the carbamate stability calculations will be at least as
accurate as the pKa calculations: the CO2-group does not vary that much in nature
between the different carbamate molecules, and the intramolecular hydrogen bonds
seem to play a lesser role than is the case for the protonated amines in the pKa study.
It should be noted that we do not believe the absolute energies calculated to be
reliable, the present level of modeling should however be sufficient to reasonably
predict the relative stability of carbamates formed from different amines.

Comparison with Experimental Data


While tertiary amines are not included in the present study, some calculations were
performed on the stability of carbamate-like species from the tertiary amines
triethanolamine and MDEA. No stable species involving CO2 bonding to the amine
functionality were found in these calculations. This is in agreement with the
established knowledge that tertiary amines do not form carbamate species.8
The calculations were performed on single solute molecules with solvation
models that produce results corresponding to infinite dilution in aqueous solution.
As noted in the introduction the carbamate is in fact only formed at detectable levels
in systems where several other species are also present in significant concentrations,
and most experimental measurements are performed in conditions far from infinite
dilution. In comparing the calculated results with experimental data we will be
ignoring concentration effects, this is an approximation that should be kept in mind
when studying the results.
As already noted, NMR-experiments offer the only direct route to finding
carbamate stabilities. Sartori and Savage6 reported the following order for Kc: MEA <
diethanolamine (DEA)< AMP. The results with free energy perturbations in Table 4

15
reproduce this trend and so do the continuum models. Suda et al.8 gave the following
order in Kc2: MEA<MMEA< MDEA. The results with free energy perturbations give
MMEA<MEA<MDEA, while both the continuum models produce the same trend as
the experimental data.
Data from Yoon and Lee7 give the following order in Kc2: MEA<2-amino-2-
methyl-1,3-propanediol (AMPD)<2-amino-2-ethyl-1,3-propanediol (AEPD) <AMP.
The free energy perturbation data in Table 4 give the following order:
MEA<AMP ≈ AMPD<AEPD. The continuum models produce results in full
agreement with the experimental trend.
In summary it can be observed that the continuum model based results give
the same relative carbamate stabilities as the NMR-data, while the free energy
perturbations errs in some cases.
While the NMR-methods offer the only direct method for estimating
carbamate stability there are other ways to infer carbamate stability from experimental
data. Carbamate formation (eq 1) is a much faster reaction than bicarbonate
formation, high reaction-rates is therefore evidence of carbamate formation, and it
would seem reasonable to expect a correlation between the reaction-rate and the
stability of the carbamate formed (i.e. the reaction energy). Available kinetic data are
from experiments on a single amine species in aqueous solution. In this case eq 1
represents the reaction that takes place. For molecules that undergo the same reaction
we can assume that there is a linear relation between the logarithm of the rate of
reaction and the reaction energy. In Figure 4 calculated ∆Gcs 2 from Table 4 is plotted

against the logarithm of experimental kinetic parameters for second order reaction (k2
, m3mol-1s-1). Experimental data are from the references given in Table 1 and the
chosen values are given in the Supporting Information. The data are at 25°C, where
several values are available the most representative values have been chosen. It
should be noted that there are inconsistencies and uncertainties in experimental
kinetics data, but these should not affect the overall trends in the results. The
experimental value for AMPD was very low and was considered to be relatively
uncertain, this value was therefore omitted from the plot.

16
Figure 4. Plot of logarithm of experimental reaction rate versus calculated energies. Stippled lines
indicate the set of results for a molecule.

Figure 4 shows good overall correlation between calculated reaction energies with
free energy perturbation solvation energies and the kinetic data, the trend line might
however overstate the relative energy differences. The outlying points in the free
energy perturbation based plot in Figure 4 are for MPA and diglycolamine (DGA).
The results with the continuum solvation models in this case show somewhat poorer
correlation with experimental data. It would appear that both the continuum models
underestimate the solvation energy of the carbamate-forms of the cyclical amines.
This suggests that while the continuum models are more reliable for molecules with
similar structures they are not as good as the free energy perturbations in predicting
the relative solvation energies of species with different structures.
In general the calculations appear to be in reasonable good agreement with
trends in experimental data. It is however clear that the models are not completely
accurate, and the solvation energies appear to be the main source of uncertainty.

17
A final possibility to extract information on relative carbamate stability from
experimental data is to look at the amount of CO2 absorbed in the system. This is
usually measured as the loading: mol CO2/mol amine in solution. Carbamate
formation consumes amine and CO2 at a stoichiometry of 2:1, while bicarbonate
formation has a stoichiometry of 1:1. Systems where carbamate formation dominate
will therefore have loadings not much higher then 0.5, while systems with only
bicarbonate formation can have loadings close to 1. Such an analysis can however be
ambiguous as a low loading can indicate either carbamate formation or a low overall
equilibrium. While comparison which such data is not included in the present work, it
is an option for systems where other experimental data is not available.

Contributions to Carbamate Stability


Some efforts have been made to rationalize observed trends in carbamate stability. It
is of interest to see if the results from the present work validate such rationalizations
or if any new general correlations can be identified.
Versteeg et al.4 observed a correlation between the pKa and the reaction rate
for a series of amines. A similar plot for the molecules in Figure 4 is given in the
Supporting Information. Little overall correlation is found in this plot. The correlation
observed by Versteeg et al.4 can be seen for a group of nine molecules, the data for
piperidine, morpholine and AMP do however not fit with this correlation. That the
previously observed correlation does not hold for all primary and secondary
molecules is not at all surprising. The base strength and carbamate stability are two
separate equilibrium constants representing the stability ratio between different
chemical species, and it is not at all a priori evident, but rather surprising, that there
should be any such correlation at all. While the correlation observed by Versteeg et
al.4 is of interest it would therefore not appear likely that it has predictive value. We
see no apparent explanation for the correlation observed for some amines. It might be
that some molecular characteristics is favorable both to the formation of anionic and
cationic species.
Sartori and Savage6 showed that a series of amines with substituent groups on
the carbon bonded to the amine functionality formed carbamate to a lesser extent than
other primary and secondary amines. They referred to these amines as “sterically
hindered”. While these amines do have some degree of steric hindrance, we would

18
caution that the name conveys an overly simple physical interpretation of carbamate
stability.
The effect of a substituent group on the stability of a species can take several
forms. There are the effects of donating or withdrawing electrons through bonds, there
can be energetically favorable or unfavorable interactions to groups to which the
substituent is not directly bonded. Steric hinderance refer to the latter of these
interactions. In addition a substituent group can also affect the accessibility of the
solvent to various parts of the molecular surface, thereby changing the solvation
energy.
Looking at the geometry of AMP-carbamate (Figure 1 and Supporting
Information) it can be seen that there is some steric interaction between a methyl
substituent and the CO2 carbon. In particular the N-C-C(OH) angle tightens from
114.53 in MEA-carbamate to 111.38 in AMP-carbamate, suggesting that the nitrogen
atom together with the carbamate functionality is forced away from one of the methyl
groups. While this effect is significant for AMP it will however not always
necessarily be the dominant factor. When estimating the reactivity of an amine all of
these effects must be considered together, AMPD and AEPD are for example clearly
no less sterically hindered than AMP but nevertheless the experimental results
strongly suggest that they have more stable carbamate forms.
One group of amines that stand out in terms of rate of carbamate formation is
the cyclical amines. There would appear to be two factors that can account for these
amines having a strong tendency to form carbamate species. The carbamate group on
the cyclical molecules are completely accessible to solvent, leading to high solvation
energies for the carbamate form. The solvation energy of the neutral amines
themselves is also relatively low, these two solvent effects together contributes to the
carbamate formation being favored. These effects will however vary with the
structure of the cyclical amine, and this should not be thought of as some form of
general rule.
It would therefore not appear that the carbamate stability can be explained in
terms of single molecular property. While steric effects do play a role, other factors
such as intramolecular hydrogen bonding and variations in solvation energy can
dominate.

19
Temperature Effects
In the earlier work on pKa calculations9 it was found that entropies from quantum
mechanical calculations together with pKa values can be used to predict changes in
pKa with temperature. The equation presented for pKa in that work can be written in a
more general form:
d (ln K ) / dT = (∆S reaction / R − ln K ) / T (12)

With this equation it should also be possible to estimate how the carbamate stability
changes over temperature. The lack of accurate carbamate equilibrium constants even
at room temperature means that the same level of accuracy in prediction can not be
achieved. There is also very little experimental data for carbamate to validate model
results.

Solvation Energy
In this work free energies of solvation has been used to calculate equilibrium
constants, but they are also of intrinsic interest. The free energy of solvation
represents the partitioning of a species between the gas and liquid phase and it is
directly related to the partial pressure of a component in the gas phase.35 Negative
values indicate a preference for the solute to stay in the liquid phase. For the amines
used in aqueous solution it is preferable if they are soluble in large quantities and do
not evaporate. The solubility is one of the factors that must be considered when
selecting solvents for industrial application. High solubility is perhaps the main reason
why alkanolamines are usually chosen for CO2 absorption processes.
The solvation models used in this work can give indications of which amines
have high solubility. Looking at the data for neutral amines in Table 3 it can be seen
that piperidine has low solubility and that AMP has relatively low solubility. This is
consistent with what is known about these amines experimentally. New solvents for
this process should ideally have solvation energies comparable to MEA (Table 3).

Conclusions
In the present work the carbamate stability of a series of amines have been calculated
with quantum mechanical gas phase energies and various solvation models. The
results are in good overall agreement with trends in available experimental data.

20
This shows that carbamate stability can be predicted with a reasonable
accuracy. Together with similar models for calculating amine basicity the present
work can be utilized to predict the chemistry and overall performance of different
amine solvents in CO2 absorption.
Different solvent models produced results in reasonable overall agreement.
The results obtained from different models did however vary and none of the models
are completely accurate.
The carbamate species appear to form hydrogen bonds between alcohol groups
and carbamate functionality oxygen atoms.
The model results suggest that steric hinderance is only one of several effects
contributing to relative carbamate stability. The high carbamate stability of some
cyclical amines appears to be caused by the high solubility of the carbamate
functionality.

Acknowledgement
This work has received support from the Research Council of Norway through a grant
of computing time.

Supporting Information Available:


Underlying values for data in Table 2, conformers of amine molecules, details of
MEA and AMP carbamate geometry, rate constants for Figure 4 and a plot of
experimental pKa versus reaction rate.

References
(1) Chakma, A.; Tontiwachiwuthikul P. Designer Solvents for Efficient CO2 Separation from
Flue Gas Streams. Greenhouse Gas Control Technol., Proc Int Conf., 4th 1999, 35.
(2) Mimura, T.; Satsumi, S.; Iijima, M.; Mitsuoka, S. Development on Energy Saving
Technology for Flue Gas CO2 Recovery by the Chemical Absorption Method in Power Plant.
Greenhouse Gas Control Technol., Proc Int Conf., 4th 1999, 71.
(3) Reddy, S.; Scherffius, J.; Freguia, S.; Roberts, C. Fluor’s Econamine FG PlusSM
Technology. Presented at the Second International Conference on Carbon Sequestration, U.S.
Department of Energy (DOE), Alexandria, VA, 2003.

21
(4) Versteeg, G. F.; van Dijck, L.A.J.; van Swaaij, W.P.M. On the kinetics between CO2 and
alkanolamines both in aqueous and non-aqueous solutions. An overview. Chem. Eng. Comm.
1996, 144, 113.
(5) Perrin, D. D. Dissociation Constants of Organic Bases in Aqueous Solution.
Butterworths, London, 1965; Supplement, 1972.
(6) Sartori, G.; Savage, D. W. Sterically Hindered Amines for CO2
Removal from Gases. Ind. Eng. Chem. Fundam., 1983, 22, 239.
(7) Yoon, S. J.; Lee, H. Substituent Effect in Amine-CO2 interaction investigated by NMR
and IR Spectroscopies. Chem. Lett., 2003, 32, 344.
(8) Suda, T.; Iwaki, T.; Mimura, T. Facile Determination of dissolved species in CO2-amine-
H2O-system by NMR spectroscopy. Chem. Lett., 1996, 25, 777.
(9) da Silva, E. F.; Svendsen, H. F. Prediction of the pKa Values of Amines Using ab Initio
Methods and Free-Energy Perturbations. Ind. Eng. Chem. Res., 2003, 42, 4414.
(10) Jamroz, M. H.; Dobrowolski, J.; Borowiak, M. Ab initio study on the 1:2 reaction of
CO2 with dimethylamine. J. Mol. Struct. 1997, 404, 105.
(11) Chakraborty, A. B.; Bischoff, K. B.; Astarita, G.; Damewood, J. R. Molecular Orbital
Approach to Substituent Effects in Amine-CO2 Interactions. J. Am. Chem. Soc. 1988, 110,
6947.
(12) Ohno, K.; Inoue, Y.; Yoshida, H.; Matsuura, H. Reaction of Aqueous 2-(N-
Methylamino)ethanol Solutions with Carbon Dioxide. Chemical Species and Their
Conformations Studied by Vibrational Spectroscopy and ab Initio Theories. J. Phys. Chem. A
1999, 103, 4283.
(13) da Silva, E. F.; Svendsen, H. F. Ab Initio Study of the Reaction of Carbamate
Formation from CO2 and Alkanolamines. Ind. Eng. Chem. Res., 2004, 43, 3413.
(14) Lukin, O.; Leszczynski, J. Rationalizing the Strength of Hydrogen –Bonded
Complexes. Ab Initio HF and DFT Studies. J. Phys. Chem. A, 2002, 106, 6775-6782.
(15) M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R.
Cheeseman, V. G. Zakrzewski, J. A. Montgomery, Jr., R. E. Stratmann, J. C. Burant, S.
Dapprich, J. M. Millam, A. D. Daniels, K. N. Kudin, M. C. Strain, O. Farkas, J. Tomasi, V.
Barone, M. Cossi, R. Cammi, B. Mennucci, C. Pomelli, C. Adamo, S. Clifford, J. Ochterski,
G. A. Petersson, P. Y. Ayala, Q. Cui, K. Morokuma, D. K. Malick, A. D. Rabuck, K.
Raghavachari, J. B. Foresman, J. Cioslowski, J. V. Ortiz, A. G. Baboul, B. B. Stefanov, G.
Liu, A. Liashenko, P. Piskorz, I. Komaromi, R. Gomperts, R. L. Martin, D. J. Fox, T. Keith,
M. A. Al-Laham, C. Y. Peng, A. Nanayakkara, M. Challacombe, P. M. W. Gill, B. Johnson,
W. Chen, M. W. Wong, J. L. Andres, C. Gonzalez, M. Head-Gordon, E. S. Replogle, and J.
A. Pople, Gaussian 98, Revision A.9, Gaussian, Inc., Pittsburgh PA, 1998.

22
(16) PC SPARTAN Version 1.0.7, Wavefunction, Inc., 18401 Von Karmen Ave. #370
Irvine, CA 92715, USA.
(17) Wiberg, K. B.; Clifford, S.; Jorgensen, W. L.; Frisch, M. J.; Origin of the Inversion of
the Acidity Order for Haloacetic Acids on going from the Gas phase to Solution. J. Phys.
Chem. A 2000, 104, 7625.
(18) Rizzo, R. C.; Jorgensen, W. L., OPLS All-Atom Model for Amines: Resolution of the
Amine Hydration Problem. J. Am. Chem. Soc. 1999, 121, 4827.
(19) Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. Development and Testing of the
OPLS All-Atom Force Field on Conformational Energetics and Properties of Organic Liquids
J. Am. Chem. Soc. 1996, 118, 11225.
(20) Harris, J. G.; Yung, K. H. Carbon dioxide’s liquid-vapor coexistence curve and
critical properties as predicted by a simple molecular model. J. Phys. Chem. 1995, 99, 12021.
(21) Hawkins, G. D.; Zhu, T.; Li, J.; Chambers, C. C.; Giesen, D. J.; Liotard, D. A.;
Cramer, C. J.; Truhlar, D. G. Universal Solvation Models in Combined Quantum Mechanical
and Molecular Mechanical Methods, Gao, J.; Thompson, M. A. Eds. American Chemical
Society: Washington DC, 1998, 201-219.
(22) Li, J.; Zhu, T.; Cramer, C. J.; Truhlar, D. G. New Class IV Charge Model for
Extracting Accurate Partial Charges from Wave Functions J. Phys. Chem. A 1998, 102, 1820-
1831.
(23) Xidos, J.D.; Li, J.; Zhu, T.; Hawkins, G. D.; Thompson, J. D.; Chuang, Y.-Y.; Fast, P.
L.; Liotard, D. A.; Rinaldi, D.; Cramer, C. J.; Truhlar, D. G. Gamesol-version 3.1, University
of Minnesota, Minneapolis 2002, based on the General Atomic and Molecular Electronic
Structure System (GAMESS) as described in Schmidt, M. W.; Baldridge, K. K.; Boatz, J. A.;
Elbert, S. T.; Gordon, M. S.; Jensen, J. H.; Koseki, S.; Matsunaga, N.; Nguyen, K. A.; Su, S.
J.; Windus, T. L.; Dupuis, M.; Montgomery, J. A. J. Comp. Chem. 1993, 14, 1347.
(24) Cramer, C. J. Essentials of Computational Chemistry, John Wiley & Sons, 2002.
(25) Sigfridsson, E.; Ryde, U. Comparison of Methods for Derving Atomic Charges from
the Electrostatic Potential and Monomers J. Comp. Chem. 1998, 19, 377.
(26) Jorgensen, W.L. BOSS version 4.3, Yale University, New Haven, CT(1989).
(27) Jorgensen, W. L.; Ravimohan, C. Monte Carlo simulation of differences in free
energies of hydration J. Chem. Phys. 1985, 83, 3050.
(28) Cances, M. T.; Mennucci, V.; Tomasi, J. A new integral equation formalism for the
polarizable continuum model: Theoretical background and applications to isotropic and
anisotropic dielectrics. Chem. Phys. 1997, 107, 3032.
(29) Littel, R.J.; Bos, M.; Knoop, G.J. Dissociation constants of some alkanolamines at
293, 303, 318, and 333K. J. Chem. Eng. Data 1990, 35, 276.

23
(30) Bishnoi, S. Dissertation, University of Texas, 2000.
(31) Sharma, M. M., Kinetics of Reactions of Carbonyl Sulphide and Carbon Dioxide with
Amines. Trans. Faraday Soc., 1965, 61, 681.
(32) Alejandro, J.; Rivera, J. L.; Mora, M. A.; Garza V. d. l. Force Field of
Monoethanolamine J. Phys. Chem. B 2000, 104, 1332.
(33) Tubergen, M. J.; Torok, C. R.; Lavrich, R. J. Effect of solvent on molecular
conformation: Microwave spectra and structures of 2-aminoethanol van der Waals complexes.
J. Chem. Phys. 2003, 119, 8397.
(34) Silva, C. F. P.; Duarte, M. L. T. S.; Fausto, R. A concerted SCF-MO ab initio and
vibrational spectroscopic study of the conformational isomerism in 2-aminoethanol. J. Mol.
Struct. 1999, 483, 591.
(35) Winget, P.; Hawkins, G. D.; Cramer, C. J.; Truhlar, D. G. Prediction of Vapor
Pressures from Selv-Solvation Free Energies Calculated by the SM5 Series of Universal
Solvation Models, J. Phys. Chem. B, 2000, 104, 4726.

24
Supporting Information for:
Study of the Carbamate Stability of Amines Using ab Initio Methods and
Free Energy Perturbations

Eirik F. da Silva and Hallvard F. Svendsen

Supporting Information is 6 pages.

Tables:
S1. Key B3LYP/6-311++G(d,p) geometry values of MEA carbamate
S2. Key B3LYP/6-311++G(d,p) geometry values of AMP carbamate
S4. Energy data for Table 2 in Manuscript

S5. Energy Data to Calculate Reaction Energy for Eq 2 in Manuscript

S6. Rate constants for various amines at 25ºC

Figures:
S3. Amine molecule conformers
S7. Plot of experimental pKa versus logarithm of experimental reaction rate

1
S1. Key B3LYP/6-311++G(d,p) Geometry Values of MEA Carbamate
Ethanolamine
Bond lengths [Å]
N-C 1.459
C-C 1.535
C-O 1.409
N-C(CO2) 1.444
H1(N)-N 1.010
H(O)-O 1.000
O1(CO2)-C(CO2) 1.267
O2(CO2)-C(CO2) 1.245
Angles
N-C-C 114.53
C-C-O 113.99
H(N)-N-C 114.33
C(CO2)-N-C 121.49
H(O)-O-C 105.52
O1(CO2)-C(CO2)-N 115.65
O2(CO2)-C(CO2)-N 115.50
Dihedral angles
N-C-C-O -75.63
H(N)-N-C-C -144.87
H(O)-O-C-C 32.11
C(CO2)-N-C-C 79.56
O1(CO2)-C(CO2)-N-C -27.43
O2(CO2)-C(CO2)-N-C 153.65
Hydrogen Bonds [Å]
H(O)-O1(CO2) 1.650

2
S2. Key B3LYP/6-311++G(d,p) Geometry Values of AMP Carbamate
Ethanolamine
Bond lengths [Å]
N-C1 1.473
C1-C2 1.555
C2-O 1.406
N-C(CO2) 1.445
H1(N)-N 1.011
H(O)-O 1.004
O1(CO2)-C(CO2) 1.267
O2(CO2)-C(CO2) 1.246
C3-C1 1.544
C4-C1 1.539
Angles
N-C1-C2 111.38
N-C1-C3 112.25
N-C1-C4 106.08
C1-C2-O 115.50
C2-O-H(O) 104.76
H(N)-N-C1 113.08
C(CO2)-N-C1 125.37
O1(CO2)-C(CO2)-N 116.91
O2(CO2)-C(CO2)-N 114.86
Dihedral angles
N-C1-C2-O 76.91
H(N)-N-C1-C2 153.24
H(O)-O-C2-C1 -44.22
C(CO2)-N-C1-C2 -70.07
O1(CO2)-C(CO2)-N-C1 28.46
O2(CO2)-C(CO2)-N-C1 -153.06
C3-C1-C2-O -48.49
C4-C1-C2-O -167.66
Hydrogen Bonds [Å]
H(O)-O1(CO2) 1.608
C1 is bonded to Nitrogen. C2 is bonded to C1 and alcohol-group oxygen. C3 and C4
are methyl group carbons bonded to C1.

3
S3. Amine molecule conformers. Stippled lines indicate hydrogen bonds.

4
S4. Energy Data for Table 2 in Manuscript (Data in Atomic Units)
No. Name MP2/6-31G(d) B3LYP/ Thermal correction
6-311++G(d,p) and ZPE
Am AmCO2- Am AmCO2- Am AmCO2-

1 MEA -209.704 -397.257 -210.464 -398.567 0.07819 0.07652

2 MPA -248.872 -436.427 -249.790 -437.893 0.10779 0.10689

3 DGA -363.057 -550.608 -364.342 -552.442 0.14041 0.14061

4 EDA -189.858 -377.403 -190.588 -378.68 0.09103 0.08933

5 MIPA -248.877 -436.431 -249.794 -437.896 0.10621 0.10508

6 AEEA(p) -343.215 -530.771 -344.473 -532.575 0.15237 0.15303

AEEA(s) -530.779 -532.578 0.15191

7 AMP -288.048 -475.601 -289.118 -477.218 0.13059 0.13449

8 AMPD -363.076 -550.638 -364.358 -552.467 0.14008 0.14042

9 AEPD -402.24 -589.804 -403.679 -591.788 0.16889 0.16954

10 DEA -363.056 -550.630 -364.346 -552.462 0.13860 0.13889

11 DIPA -441.402 -628.978 -443.004 -631.12 0.19498 0.19564

12 MMEA -248.862 -436.417 -249.779 -437.88 0.10629 0.10476

13 Morpholine -286.854 -474.4 -287.882 -475.977 0.11660 0.11346

14 Piperazine -267.012 -454.557 -268.011 -456.104 0.12959 0.12674

15 Piperidine -251.003 -438.546 -251.979 -440.067 0.14202 0.13866

S5. Energy Data to Calculate Reaction Energy for Eq 2 in Manuscript

[Data in Atomic Units].


Species B3LYP/ ∆Gs b Thermal
6-311++G(d,p) a correction
and ZPE c
H2O -76.4585 -0.01027 0.004129
HCO3- -264.551 -0.1172 0.002176
CO2 -188.647 0.001066 -0.00882
H3O+ -76.7295 -0.16665 0.017562
a: Gas phase energy. b: Solvation energy (IEFPCM/ MP2/6-31G(d)// MP2/6-31G(d).
c: Calculated at HF/3-21G(d) level.

5
S6. Rate Constants for Various Amines at 25ºC.
k2 [m3mol-1s-1] Reference
Ethanolamine ⎛ 5400 ⎞⎟ a
4.4×108 exp ⎜⎜− ⎟
⎜⎝ T ⎠⎟
3-Amino-1-Propanol ⎛ 5605 ⎞⎟ a
1.3×109 exp ⎜⎜−
⎜⎝ T ⎠⎟⎟
2-(2-aminoethoxy)ethanol 3.99 a

Ethylendiamine 15.1 b

1-Amino-2-Propanol ⎛ 5033 ⎞⎟ a
9.11×107 exp ⎜⎜
⎜⎝ T ⎠⎟⎟
2-amino-2-methylpropanol 1.05 b

2-Amino-2-methyl-1,3-propanediold 0.04 b

Diethanolamine 3.24 a

Diisopropanolamine 2.70 a

2-(methylamino)ethanol ⎛ 3532 ⎞⎟ a
9.63×105 exp ⎜⎜− ⎟
⎜⎝ T ⎠⎟

Morpholine 20.0 b

Diethylenediamine 53.7 c

Piperidine 60.3 b

a: Versteeg, G. F.; van Dijck, L.A.J.; van Swaaij, W.P.M. On the kinetics between CO2 and
alkanolamines both in aqueous and non-aqueous solutions. An overview. Chem. Eng. Comm.
1996, 144, 113. b: Sharma, M. M., Kinetics of Reactions of Carbonyl Sulphide and Carbon
Dioxide with Amines. Trans. Faraday Soc., 1965, 61, 681. c: Bishnoi, S. Dissertation,
University of Texas, 2000. d: Not utilized in plot as number would seem relatively uncertain.

6
S7. Plot of experimental pKa versus logarithm of experimental reaction rate. a
Versteeg, G. F.; van Dijck, L.A.J.; van Swaaij, W.P.M. On the kinetics between CO2
and alkanolamines both in aqueous and non-aqueous solutions. An overview. Chem.
Eng. Comm. 1996, 144, 113.

7
Paper VI

Molecular Dynamics Study of Ethanolamine as a Pure Liquid and in

Aqueous Solution

Eirik Falck da Silva, Tatyana Kuznetsova and Bjørn Kvamme

2005
Molecular Dynamics Study of Ethanolamine as a Pure

Liquid and in Aqueous Solution

Eirik F. da Silva,*a Tatyana Kuznetsovab and Bjørn Kvammeb

a
Department of Chemical Engineering, Norwegian University of Science and Technology, N-7491

Trondheim, Norway.

b
Department of Physics, University of Bergen, N-5007 Bergen, Norway.

silva@chemeng.ntnu.no

Molecular dynamics simulations have been carried out for ethanolamine as a pure liquid and in aqueous

solution at 298.15K and 333K. Two force field representations were utilized for ethanolamine. Results

are presented for density, enthalpy of vaporization, radial distribution functions and conformer

distributions. The results strongly suggest that the main (O-C-C-N) dihedral tends to stay in its gauche

conformers in solution and that the ethanolamine molecules populate conformers with a considerable

degree of intramolecular hydrogen bonds. Aqueous ethanolamine shows a preference to be solvated by

water molecules, resulting in a liquid structure that is fairly homogeneous at the molecular level. A

simulation was also carried out with a CO2 molecule in an aqueous ethanolamine system, the results

suggesting that CO2 is almost equally attracted to ethanolamine and water.

Introduction

Mixtures of Alkanolamines and water are commonly used to absorb carbon dioxide from natural gas

and exhaust gases.1 This is currently one of the most viable among the available technologies to capture
1
2
carbon dioxide. While alkanolamine based CO2 capture has received considerable experimental

attention little work has been done on the understanding of these systems at the molecular level.

Ethanolamine is the simplest of the alkanolamine molecules and some simulation work has been done in

recent years to ascertain its liquid structure and properties as well as its behavior in aqueous mixtures.3 -7

Besides being important in gas sweetening and other industrial processes, ethanolamine and other

alkanolamines are also of interest for other reasons. Together with 1-2 ethanediol, ethanolamine

represents one of the simplest molecules able to form intramolecular hydrogen bonds. Both in pure

liquids and aqueous solutions intramolecular hydrogen bonds will compete with the formation of

intermolecular hydrogen bonds. This potential for hydrogen bonding also raises questions about the

structure of ethanolamine both as a pure liquid and in aqueous solution. The modeling of such small

organic molecules is clearly also of relevance to the modeling of more complex biomolecules.

In the parameterization of molecular force fields, ethanolamine and other alkanolamines present an

interesting challenge. Two main approaches for parameterization of force fields exist in the open

literature. For water and small organic molecules, such as methanol, a number of specially fitted force

fields have been presented (Guillot8 and Walser et al.9 and references therein). These are typically based

on reproducing certain experimental properties, such as density, radial distribution functions and

enthalpy of vaporization, and each parameter in the force field is often fine-tuned. This approach to

force field fitting is not viable in case of larger organic molecules with many atomic sites. For such

molecules general transferable force fields such as OPLS10 have been developed. Ethanolamine is a

relatively small molecule, but the number of parameters in a force field will tend to be much higher than

for water, and the molecule has not been as rigorously studied experimentally as water or the simple

mono-alcohols. There would however appear to be room for more detailed force field parameterization

than usually seen for biomolecules.

The intention of the present work has been to look in greater detail at force field parameterization,

conformer distribution and other aspects of the system of particular relevance to the gas absorption

2
process. We were especially interested in understanding the carbamate formation mechanism, which

according to da Silva and Svendsen11 can be written in the following way:

CO2 + R1R2 NH " B U R1 R2 NCOO− " BH + (1)

Where B is a base molecule (usually a second ethanolamine molecule) and R1 R2 NH is an

ethanolamine molecule. The stippled line indicates a hydrogen bond.

Force Field

Two force field representations of ethanolamine were studied in the present work. One is a somewhat

adjusted version of the force field presented by Alejandro et al.4 while the second force field was based

on our own parameterization and analysis. A united atom approach in which methyl group hydrogens

are not explicitly represented is utilized for both representations of ethanolamine.

Bond angles are handled by harmonic type potentials:


2
U (θ) = kθ (θ −θ 0 ) (2)

where θ is the bond angle and the subscript 0 denotes the equilbrium value. kθ is the spring constant.

Dihedral angle energies around bonds are given by:


5
U (φ) = ∑ Ci cos (φ)
i
(3)
i =1

where φ is the dihedral angle and the Ci are constants. In this potential φ = 0 corresponds to the trans

form and φ = 180 to the cis form. The potential energy is given by the standard combination of

Lennard-Jones and Coulomb potentials:

⎡⎛ ⎞12 ⎛ ⎞6 ⎤
⎢ σ⎟ σ ⎟ ⎥ qi q j
U (rij ) = 4ε ⎢⎜⎜⎜ ⎟⎟ − ⎜⎜⎜ ⎟⎟ ⎥ + (4)
⎢⎜⎝ rij ⎠⎟ ⎝⎜ rij ⎠⎟ ⎥ rij
⎣ ⎦

where q are atomic charges, σ and ε are the Lennard-Jones parameters and i and j are any atomic

sites. Only intermolecular potential energy is calculated in both the force fields, the intramolecular

3
potential is represented solely by the dihedral and bond energies. Standard Lorenz-Berthelot mixing

rules were applied.

The new representation was based on quantum mechanical calculations to determine geometry and

intramolecular potential together with standard Lennard-Jones parameters from the united12 and all
10, 13
atom OPLS force field. Atomic charges were fitted to the electrostatic potential from quantum

mechanical calculations. The choice of parameters was also guided by our desire to reproduce the

experimental enthalpy of vaporization and density. This parameterization combines elements of the

approach taken by Jorgensen and coworkers10 and Cornell and coworkers14 in their respective force field

development.

Conformer notation has been adopted from work by Vorobyov et al.15 A conformer is represented as

xYz. Where x designates the C-C-N- lpN dihedral angle, Y is the O-C-C-N dihedral angle, and z the C-

C-O-H dihedral angle. lpN denotes the lone pair on the nitrogen atom. G or g indicates gauche(+), G’ or

g’ indicates gauche(-) and T or t indicates trans.

Vorobyov et al.15 have published results of quantum mechanical calculations on the ethanolamine

geometry more advanced than those presented by Alejandro et al.4 In the present work we have replaced
4
the force field geometry presented by Alejandro et al. with the B3LYP/6-311++G(2d,2p) g’Gg’

geometry from this new paper.15 The resulting force field will be referred to as MEAa, MEA

(monoethanolamine) being the common acronym for ethanolamine in chemical engineering.

Gubskaya and Kusilik6 observed that the Alejandro et al.4 force field has a enthalpy of vaporization

significantly lower than the experimental value. In the present work we will use a parametrization

procedure similar to that utilized by Alejandro et al.4, but making sure that the experimental heat of

vaporization is reproduced reasonably well. One of the purposes of the present work is to study the

conformer distribution of ethanolamine in greater detail, and for this purpose we performed a more

detailed analysis of the intramolecular potential.

For the new force field we again chose geometry from recent quantum mechanical calculations15 and

OPLS Lennard Jones parameters similar to the ones selected by Alejandro et al.4 The bond angle

4
4
potential was adopted from Alejandro et al. The approach of deriving the atomic charges from the

electrostatic potential taken by Alejandro et al.4 seems to us a sensible one, but as their model produces

to low heat of vaporization we carried out more detailed analysis of the atomic charges. To calculate the

atomic charges from the electrostatic potential the Merz-Kollman16 (MK) scheme implemented in
17
Gaussian 98 was utilized. Calculation on a single molecule in gas-phase will yield only gas-phase

charges, while molecules in solution are polarized by the environment and tend to have higher charges.

Calculations with a continuum solvent model were carried out to obtain solution phase charges.

Different conformer geometries were optimized at the B3LYP/6-311++G(d,p) level and continuum

model calculations were carried out as single-point IEFPCM18 calculations (IEFPCM-B3LYP/6-

311++G(d,p)// B3LYP/6-311++G(d,p)). The methyl group charges were added together, and the total

charge of these groups were distributed evenly between the two carbon atoms. The amine group

hydrogen atom charges were also averaged.

A modified version of the EMP2 model19 has been utilized to simulate CO2. The modification consists

of adding a flexible bond angle. The spring constant was determined by use of ab initio calculation.

B3LYP/6-311++G(d,p) calculations were carried out for unconstrained CO2 and CO2 with the bond

angle set to 150º. The energy difference between these forms was determined to be 81 kJ/mole, and the

spring constant was set to reproduce this energy difference. The model details are given in the

supporting information.

Intramolecular Potential

Some simple spread-sheet calculations were performed to study the intramolecular potential for

different conformers. The intramolecular distances can be determined from the geometry data for
15
different conformers and the energy of a conformer can be calculated directly for a given force field.

The angle energies were neglected in these calculations. Energies were calculated both with ab initio

dihedral angles and dihedral values found to be optimal in simulations. Such approximate spread-sheet

calculations can provide an immediate picture of relative conformer energies.

5
Simulations Details

The SPC model20 was chosen to represent water. Simulations were carried out using the constant-

temperature, constant-pressure algorithm (nPT) in the MdynaMix package written by Lyubartsev and
21
Laaksonen. Long-range electrostatic interactions were handled by means of Ewald summation.

Lennard-Jones forces and real-space electrostatics were cut off at 11.5Å. A dual time step algorithm

was used, with all the forces in the system divided into fast and slow ones. Fast forces included

interactions arising from bonds, constrained angles and dihedral angles, as well as Lennard-Jones and

real-space electrostatic forces within a cutoff of 5Å. They were recalculated each short time step (0.05

fs), with the rest of the forces recalculated once every 10 short steps. Simulations were carried out with

512 molecules at 298.15 K and 333K and 1 atmosphere of pressure. Two different compositions were

studied, one with pure ethanolamine and one with 461 water molecules and 51 ethanolamine molecules.

The latter corresponds to 10 mol percent ethanolamine, representative of the composition often used in

industrial applications. For comparison purposes simulations we also carried out pure water simulation.

Each system was equilibrated over at least 200000 steps (100ps), and sampling was done over 100000

steps (50ps).

The single-molecule simulations (gas-phase) were run for over 20 million steps (20 ns). For these

simulations, we followed the approach of Jellinek and Li22 and modified the original MdynaMix

package to implement separate temperature controls for each kinetic mode (translation, rotation, and

internal).

6
Results

The relative conformer energies of the MEAa force field are shown together with the relative energies

from quantum mechanical calculations in Table 1. Energies for all 14 nonequivalent conformers are

shown.15 The force field results are presented both for quantum mechanical dihedral angles determined
15
by Vorobyov et al. and for the average dihedral angles for each conformer determined from liquid-

phase simulations. These were estimated to be 77 for O-C-C-N gauche and 180 for trans, 45 for C-C-

O-H gauche and 182 for trans and finally 58 for C-C-N- lpN gauche and 180 for trans.

The most stable gas-phase conformer (g’Gg’) is shown in Figure 1.

Table 1. Relative conformer energies. Values in [kJ/mole].

B3LYPa MEAab MEAac


g'Gg' 0 0 0.0
gGg' 5.5 8.5 0.0
gGt 5.3 5.0 2.6
tGt 5.7 17.0 5.3
tGg 6.6 11.5 2.7
gGg 6.9 -1.4 0.0
tGg’ 8.0 9.2 2.7
tTt 9.5 18.2 6.2
tTg 9.7 9.6 3.6
gTt 10.0 9.1 3.5
gTg 10.5 0.5 0.9
g’Tg 10.6 -0.1 0.9
g’Gt 16.2 6.9 2.6
g’Gg No minima -1.7 0.0
a
: Sum of electronic and zero-point energies at B3LYP/6-311++G(2d,2p) level from Vorbyov et al.15 b:
c
Conformer energies calculated with dihedral angles determined from ab initio calculations. : Conformer
energies calculated with optimal dihedral angles for force field.

7
Figure 1. g’Gg’ conformer of ethanolamine

The MEAa force field can be seen from the data in Table 1 to be in mostly reasonable agreement with

the quantum mechanical results. Some conformers that are not equivalent in the ab initio calculations do

however become equivalent in the force field. This is a direct consequence of using a force field with an

intramolecular energy only represented by uncoupled dihedral energies. Thus g'Gg', gGg', gGg and g'Gg

become equivalent in terms of energy, the same holds for the g'Gt and gGt pair and gTg and g’Tg as

well.

A number of different force fields of ethanolamine were studied in a recent paper by Gubskaya and

Kusalik.6 They apparently treated the intramolecular potential as a parameter to be set in the overall

fitting of liquid properties. It is not clear from their paper if the representations chosen reproduce

relative conformer energy differences from quantum mechanical calculations. We believe a more

realistic solvent representation is obtained by determining the intramolecular potential from quantum

mechanical potential, and that this part of the potential should not be arbitrarily fitted to experimental
10
data. In this we follow the approach taken by Jorgensen et al.

Gubskaya and Kusalik6 observed, as previously noted, that the MEAa force field had a too low

enthalpy of vaporization. Results from our own calculations shown in Table 2 are in agreement with

that observation. It should be noted that our density for MEAa is higher than reported by Alejandro et

al.4 The discrepancy is most probably caused by different ensambles utilized and that, unlike Alejandro

et al.4, we did not shift or truncate the Lennard-Jones potential.

8
Table 2. Densities and Heats of Vaporization for ethanolamine.

MEAa MEAb Experimental

ρ (298.15K ) 1.032 1.060 1.012d, 1.008e


a

ρ (333K ) 1.003 1.037 0.984d,0.984e


a

U liq (298.15 K ) a -43.04 -62.34

U gas (298.15 K )
a
1.90 -1.97

∆H vap (298.15 K )
b
47.7 63.13

U liq (333K )
a
-37.45 -55.34

U gas (333K )
a
2.59 -1.32

∆H vap (333K ) 57.7d


b
42.8 56.80

a
Density in g/cm3. b Energy in kJ/mole. c Enthalpy of vaporization in kJ/mole. d Data from da Silva5
e 27
and references therein. Data from Cheng et al.

Alejandro et al.4 reported a single set of atomic charges for the fitting of charges to the electrostatic

potential. Such procedures to fit charges are however known to be ambiguous,23 and the charges may
24
depend on the conformer form of the molecule. In Table 3 results of fitting of the charges to the

electrostatic potential in a solvent field (IEFPCM model) for a number of conformer forms are shown.

9
Table 3. Dipole Moments and Atomic Charges from Fitting the Electrostatic potential.a

g’Gg’ tGt gGt tTt gTt

Dipole Moment: 4.110 2.360 1.329 3.607 0.998

N -1.035 -1.162 -1.163 -1.224 -1.256

H1(N) 0.397 0.423 0.415 0.427 0.436

H2(N) 0.399 0.423 0.436 0.425 0.450

C(N) 0.274 0.402 0.367 0.490 0.448

H1(C(N)) -0.036 -0.016 0.003 -0.008 0.014

H2(C(N)) 0.048 0.026 -0.016 -0.004 -0.043

C(O) 0.189 0.161 0.238 0.317 0.312

H1(C(O)) 0.021 -0.011 -0.027 -0.028 -0.011

H2(C(O)) 0.046 0.034 0.026 -0.030 0.009

O -0.705 -0.767 -0.756 -0.891 -0.864

H(O) 0.402 0.484 0.477 0.526 0.504

a
MK charges at IEFPCM-B3LYP/6-311++G(d,p)//B3LYP/6-311++G(d,p) level.

It can be seen from Table 3 that there are significant differences in the charges calculated for different

conformers. Both the magnitude of charges and (to a lesser extent) their relative distribution between

the hydroxyl functional group and the amino functional group vary. The present charges calculated in a

solvent field for the g’Gg’ conformer are as might be expected higher than the gas-phase charges
4
utilized by Alejandro et al. It can also be noted that the g’Gg’ conformer has the lowest charges of all

these conformers. Alejandro et al.4 suggested that their choice of charges was validated by the

agreement with experimentally-determined dipole moment. The experimental dipole moment is

however an average over the conformer population, and as the data in Table 3 suggest that the

conformers have widely varying dipole moments, it becomes apparent that this comparison with the

experimental value is inconclusive.

We looked at which set of charges that would produce best agreement with the experimental heat of

vaporization. A force field based on the g’Gg’ charges gave good agreement with the experimental heat
10
of vaporization and these charges have been adopted. This agreement must be viewed as fortuitous and

can not be used to draw conclusions regarding the conformers in solution.

While there is some ambiguity in which charges to choose, the MEAa and MEAb representation are

in reasonable qualitative agreement and do not differ too much from the values for ethanolamine in the
3, 5
semi-empirical OPLS force field.

The MEAb model utilized the same intramolecular potential as in MEAa. In our work we did attempt

to apply different intramolecular potentials to get the best possible agreement with the relative

conformer energies determined from quantum mechanical calculations. The presence of intramolecular

hydrogen bonds in some conformers does mean that the energy dependency of the different dihedral

angles is strongly coupled. We have looked into adding intramolecular Coulomb and Lennard-Jones

forces interactions as a means to capture this coupling. In such representations there was unfortunately a

tendency for the strength of the hydrogen bonds to be overestimated and we were unable to come up

with a set of parameters that produced a representation better than the MEAa force field.

The MEAb charges are higher than those used by Alejandro et al.4, combined with an otherwise

identical force field, this results in increased density. To keep the density from deviating too much from

the experimental values, we looked at the possible changes that could be made to the parameters

without deviating from geometries derived from quantum mechanical calculations. Alejandro et al.4

used the g’Gg’ conformer to determine the bond lengths and angles. It is however not clear if this is the

best choice when representing a liquid phase were a number of conformer forms are populated.
15
Conformer geometries reported by Vorobyov et al. show that while most bond-lengths change little

with conformer, most conformer have a C-O bond of around 1.43Å, longer than for the g’Gg’

conformer (1.42 Å). For MEAb the value of 1.43 Å is therefore chosen. The g’Gg’ values are clearly

not representative for some of the angles either. We set the C-O-H(O) angle to 108.7 in MEAb, while

both the C-C-N and C-C-O angles were set to 112. We also chose to increase the Lennard-Jones σ for

Nitrogen to 3.3Å, which is the standard in the all atom OPLS force field.13 The MEAb force field

parameters are shown in Table 4, the MEAa parameters are given in the Supporting Information.

11
Table 4. Potential Parameters of MEAb

Bond r0 [Å]

N-H 1.012
N-C 1.471
C-C 1.524
C-O 1.43
O-H 0.966
Angle θ0 kθ [J/mole]

H-N-H 107.4 269.39


H-N-C 111.4 316.65
N-C-C 112 506.21
C-C-O 112 546.91
C-O-H 108.8 298.79
Dihedral C1 a C2 a C3 a C4 a C5 a
H-N-C-C -11.05 -2.37 25.30 0.43 -2.02
O-C-C-N -17.78 23.60 36.81 -16.64 -13.32
C-C-O-H -19.15 -2.79 13.61 -0.12 1.55
Site σ [Å] ε [KJ/mole] q

H(N) 0 0 0.40
N 3.3 0.7108 -1.035
CH2 3.905 0.494 0.27
O 3.07 0.7108 -0.705
H(O) 0 0 0.40
a
In kJ/mole.

The density and heat of vaporization of MEAb is shown in Table 1. The ethanolamine force field has

a large number of parameters and one could certainly refine the parameters to improve the fit with

experimental data. It is however not clear to us at present what kind of parameter adjustment is the most

12
reasonable, and we have chosen not to deviate from the quantum mechanical representation of the

molecule.

The simulated and experimental densities of both MEAa and MEAb as pure component and in

mixture with water at 298.15K and 333K are given in Table 5. MEAa reproduces the experimental

values quite well, the deviation from experimental values being small and fairly constant. The MEAb is

in somewhat worse, but still reasonable, agreement with experimental data. The densities are in general

somewhat high, even for SPC water, this would appear to be a result of the type of ensemble and

simulation algorithm employed.

Table 5. Densities of ethanolamine-water systems

b
MEA 10% MEA H2O

ρ (298.15K ) -Exp. 1.008 1.009 0.997


a

ρ (298.15K ) -MEAa 1.032 1.040 1.015


a

ρ (298.15K ) -MEAb 1.060 1.045


a

ρ (333K ) -Exp. 0.984 0.992 0.983


a

ρ (333K ) -MEAa 1.003 1.011 0.994


a

ρ (333K ) -MEAb 1.037 1.019


a

a
Density in g/cm3, experimental data are from Cheng et al.27 b 10 mol percent ethanolamine.

Some of the intermolecular radial distributions for pure ethanolamine at 298.15K and 333K are

presented in Figure 2. The MEAa results are mostly in good agreement with those presented by

Alejandro et al.4 The O-H(N) curve does however show a significantly higher peak in the present

calculations. Despite the differences in the force fields utilized most of the radial distribution functions

are also in good agreement with results obtained by Gubskaya and Kusilik.6

13
Figure 2. Radial distribution functions for pure ethanolamine. Gray lines are MEAa and black lines are

MEAb. Solid lines are for 298.15K while dashed lines are for 333K.

MEAb produces somewhat higher peaks in the radial functions than MEAa, this is consistent with

MEAb being the force field with higher atomic charges. The force fields do however appear to produce

very similar liquid structures, suggesting that results can be viewed with a high degree of confidence.

The results suggest that the strongest bonding takes place between hydroxyl-groups. The second

strongest feature is bonding between hydroxyl-group oxygen atoms and amino-group hydrogens.

Bonding between the amino groups appear to be somewhat weaker. Both the N-H(O) and N-H(N)

curves have second peaks that are higher then first, these reflect bonding taking place at the hydroxyl-

functionality of the molecule. The liquid ethanolamine structure is clearly dominated by the hydrogen

bonding features. Neither radial distribution functions nor visual inspection of the ensemble suggested

any ordered structure.

14
At 333K the radial distribution functions become somewhat less pronounced, a trend which is

common and expected in most liquids. The changes in structure do however appear to be quite small for

the temperature range studied.

Radial distribution functions for 10 mol percent MEAa and MEAb in aqueous solution are presented

in Figures 3 and 4. Once again we have chosen to display the radial distribution functions that convey

information about the hydrogen bonding. In Figure 3 the same ethanolamine-ethanolamine radial

distributions as plotted for pure ethanolamine (Figure 2) are shown. All the peaks are lower than in the

pure liquid suggesting that ethanolamine molecules have a preference to be surrounded by water

molecules, bonding between ethanolamine molecules does however persist. The amino-amino bonding

becomes significantly less frequent in the aqueous solution, while the hydroxyl-group interactions

change less in their relative prevalence. A fairly homogeneous and random molecular-level structure is

suggested by the radial distribution functions and visual inspection of the structure.

15
Figure 3. Radial distribution functions for 10 mol percent ethanolamine in water. Gray lines are MEAa

and black lines are MEAb. Solid lines are for 298.15K while dashed lines are for 333K.

16
Figure 4. Radial distribution functions for 10 mol percent ethanolamine in water. Gray lines are MEAa

and black lines are MEAb. Solid lines are for 298.15K while dashed lines are for 333K.

Radial distributions for the water molecules are shown in Figure 5. The results are for 10 mole

percent aqueous ethanolamine and pure water. While the changes are quite small it can be observed that

first peaks in the radial distribution functions become somewhat higher in the aqueous ethanolamine

system.

17
Figure 5. Water-water radial distribution functions at 333K. Stippled lines are from 10 mole percent

mixture with MEAb, while solid lines are from pure water simulation.

It has been suggested that ethanolamine forms dimers in aqueous solution.25 The liquid structure and

the nature of the bonding in the simulated system is however such that stable dimers would seem

unlikely to be formed. The force fields in the present work have however not been parameterized to

reproduce dimer formation energies, and more careful studies would be required to draw confident

conclusions on this point.

The populations of the various conformers for pure ethanolamine and 10 mol percent aqueous

solution of ethanolamine are shown in Table 6. Most of the data pertains to 333K, pure MEAa and

MEAb data at 298K are also included.

18
Table 6. Conformer Populations.

a a, c a a, c b b, c b, b, c d d, e e,
Ma Ma Ma Ma Mb Mb Mb Mb aH2O aH2O bH2O bH2O
c c

333K 333K 298K 298K 333K 333K 298K 298K 333K 333K
333K 333K

g'Gg' 3.3 13 3.4 11 1.2 1 1.2 9 3.9 27 1 12


gGg' 34.0 43 35.6 37 21 6 22.6 54 22.7 28 14.4 54
gGt 0.1 0 0.1 4 0 0 0 0 0.1 0 0 0
tGt 0.0 0 0.0 0 0 0 0 0 0.0 0 0 0
tGg 0.1 4 0.3 0 1.3 8 0 0 0.0 0 0 0
gGg 40.0 16 37.3 12 16.7 1 16.9 13 55.7 22 20 23
tGg’ 0.7 11 1.4 20 16.4 63 0.4 13 0.0 0 0 0
tTt 0.0 0 0.0 0 0 0 0 0 0.0 0 0 0
tTg 0.1 0 0.2 5 2.3 20 0.2 0 0.0 0 0 0
gTt 0.0 0 0.1 0 0.0 0 0 0 0.0 0 0 0
gTg 5.7 3 5.4 2 3.1 0 3 3 3.5 2 0.8 1
g’Tg 13.8 7 13.7 6 6 1 7.4 7 0.1 0 2 3
g’Gt 2.3 3 2.5 3 0.5 0 0.3 1 14.1 21 1.5 7
g’Gg 0.0 0 0.0 11 0 0 0 0 0.0 0 0 0
a
MEAa.b MEAb. c Values corrected for errors in relative conformer energies of force field. d10 weight
e
percent MEAa in water. 10 weight percent MEAb in water.

We have analyzed the relative conformer populations of both liquid and aqueous MEA at the two

temperatures. The dihedral angle distributions for MEAa at 298K are plotted in Figure 6, with the data

for the other systems given in the supporting information section. The peaks of MEAa dihedral

distributions are located at 77 (gauche) and 180 (trans) in case of the O-C-C-N torsional, 45 (gauche)

and 180 (trans) for the C-C-O-H torsional, and finally 58 (gauche) and 180 (trans) for C-C-N-lpN

angle. A molecule was classified as a given conformer if its dihedral angles lay within 40 degrees (45

for MEA-water systems) of the corresponding dihedral peaks. Varying the 40 degrees limit did not

19
significantly alter the relative conformer populations. With this kind of definition a part of the

molecules will not be counted as occupying any conformer form, this fraction amounted to between 30

and 60 percent of the ethanolamine molecules. The statistics was worse in case of aqueous solutions,

resulting in significantly larger uncertainty in the relative conformer populations.

Figure 6. Dihedral angle distribution for pure MEAa at 298K.

The ethanolamine force fields do not fully reproduce the relative conformer energies from ab initio

calculations (Table 1). We have therefore done calculations to attempt to correct for this error and give

a picture of what the conformer populations would have been if the force field intramolecular potential

had accurately reproduced the ab initio energies. The conformer populations were first converted to

relative energies. The relative energies were then corrected by the difference between ab initio and force

field energies (data in Table 1, second and fourth column). The relative energies were then finally

converted back to conformer populations. The results with these corrections provide an approximate

estimate of what the conformer population would have been for a more accurate intramolecular

potential. Such a correction will however produce uncertain results in case of sparsely populated

conformers, therefore no correction was applied and the corrected population values were set to 0 for

conformers accounting for less than 0.5 percent of all molecules.

20
The g’Gg’ conformer has a small population in solution, despite being the most stable in the gas-

phase. This conclusion holds for both force fields and in aqueous solution as well as in the pure liquid.

It can also be seen that the total population for O-C-C-N trans conformers is quite small. While the

corrected values for MEAb in aqueous solution in one case shows a significant trans conformer

population, the correction was in this case being made to a very small population number and is highly

uncertain. Several of the trans conformers in the force field have intramolecular energy differences

relative to the g’Gg’ conformer that are equal or smaller to the ab initio differences. This would suggest

that the force fields overestimate the trans populations, and the total trans (O-C-C-N) population would

therefore in general seem to be less than 20 percent. These conclusions are in good agreement with

those drawn from experimental work on pure ethanolamine.26 These results strongly suggest that

ethanolamine forms a significant amount of intramolecular hydrogen bonds in solution.

In the case of pure ethanolamine the most populated conformers are for MEAa gGg’ and gGg, and

tGg’, gGg’ and tTg’ for MEAb. Looking at the values corrected for the intramolecular potential there

can however be seen to be significant changes. There are also some significant difference between the

two force fields. An interesting feature of the MEAb force field is the high tGg’ populations for pure

component at 333K, at 298K and in aqueous solution this conformer has a much smaller population.

The MEAa force field has a very small population for the same conformer. Conformers with the C-C-N-

lpN dihedral in a trans form can more readily form intermolecular hydrogen bonds to nitrogen atom.

Stronger hydrogen bonds to the nitrogen atom in the MEAb force field are therefore the most likely

explanation for the high tGg’population.

Silva et al.26 reported the gGt and tGt conformers to be the most common for pure ethanolamine

solution. The main difference between their findings and the simulation results would appear to be in

the C-C-O-H dihedral angle. All the simulation results (see Figure 6, supporting information and Table

5) suggest that this dihedral remains mainly in the gauche conformers. This might suggest that the

atomic charge on the hydroxyl-group hydrogen atom is too low, thereby underestimating the strength of

21
intermolecular hydrogen bonds formed with the hydroxyl-group hydrogen atom. This observation could

perhaps in future work be utilized for further refinement of the force field.

Radial distribution functions for system of a CO2 molecule in a 10 mole percent ethanolamine

(MEAb) in aqueous solution at 333K are shown in Figure 7. The radial distributions shown are for the

interactions that are expected to dominate the bonding of CO2. The results suggest that the CO2

molecule is somewhat more attracted to the ethanolamine amino-functionality than to the hydroxyl

group. This interaction with the amino group is however somewhat less favored than with the water

molecules.

Figure 7. CO2-ethanolamine-H2O radial distribution functions at 333K for 10 mol percent

ethanolamine (MEAb) in water. Lines in left plot: Solid line is C(CO2)-N(MEA), dashed line is C(CO2)-

O(MEA) and dotted line is C(CO2)-O(H2O). Lines in right plot: Solid line is O(CO2)-H(N)(MEA),

dashed line is O(CO2)-H(O)(MEA) and dotted line is O(CO2)-H(H2O).

CO2 is known to react with ethanolamine according to equation 1 in an aqueous alkanolamine system.

da Silva and Svendsen10 concluded that the reaction mechanism had no intrinsic barrier. They also

suggested that the barrier could either arise from the CO2 molecule needing to displace the water

molecules around the amino group before reacting or from the need for two amine molecules to

22
approach each other sufficiently for a proton transfer to take place. Such a reaction can not readily be

studied with the classical simulations in the present work. The picture of the liquid structure from the

present work can nevertheless further the understanding of the reaction mechanism. Since the CO2

molecule bonds readily to the amino group it would appear that there is not a significant barrier to the

approach of the ethanolamine amino group and CO2. The low degree of direct interaction between

amino-functionalities on different molecules would on the other hand suggest that the low likelihood of

such interactions may be a significant barrier to reaction.

Conclusions

Two different force fields were used to perform simulations of ethanolamine. The general agreement

in results between the two parameterization suggest that the results can be viewed with a high degree of

confidence. The results suggest that the ethanolamine O-C-C-N dihedral tends to stay in a gauche

conformer and that there is a significant degree of intramolecular hydrogen bonding. These findings

holds for ethanolamine as a pure liquid as well as in aqueous solution. While the simulation results are

in broad general agreement with conclusions drawn from experimental work on conformer populations,

it would appear that the force fields are not sufficiently accurate to predict relative conformer energies

in solution. In aqueous solution ethanolamine is preferentially solvated by water molecules, producing

an aqueous solution that is homogeneous on the molecular level.

Simulations in aqueous solution suggest that CO2 has a comparable level of affinity to ethanolamine

molecules and water.

Supporting Information Available: Force field paramaters for MEAa and CO2 and dihedral angle

distribution figures.

23
REFERENCES

(1) Versteeg, G. F.; Van Dijck, L. A. J.; Van Swaaij, W. P. M. Chem. Eng. Comm. 1996, 144, 113.

(2) Rao, A. B.; Rubin, E. S. Environ. Sci. Technol. 2002, 36, 4467.

(3) Button, J. K.; Gubbins, K. E.; Tanaka, H.; Nakanishi, K. Fluid Phase Equilib. 1996, 116, 320.

(4) Alejandro, J.; Rivera, J. L.; Mora, M. A.; de la Garza, V. J. Phys. Chem. B 2000, 104, 1332.

(5) da Silva, E. F. Fluid Phase Equilib. 2004, 220, 239.

(6) Gubskaya, A. V.; Kusilik, P. G. J. Phys. Chem. A 2004, 108, 7151.

(7) Gubskaya, A. V.; Kusilik, P. G. J. Phys. Chem. A 2004, 108, 7165.

(8) Guillot, B. J. Mol. Liq. 2002, 101, 219.

(9) Walser, R.; Mark, A. E.; van Gunsteren, W. F. J. Chem. Phys. 2000, 112, 10450.

(10) Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. J. Am. Chem. Soc. 1996, 118, 11225.

(11) da Silva, E. F.; Svendsen, H. F. Ind. Eng. Chem. Res. 2004, 43, 3413.

(12) Jorgensen, W. L.; Tirado-Rives, J. J. Am. Chem. Soc. 1988, 110, 1657.

(13) Jorgensen, W. L.; Rizzo, R. C. J. Am. Chem. Soc. 1999, 121, 4827.

(14) Cornell, W. L.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Merz, K. M.; Ferguson, D. M.;

Spellmeyer, D. C.; Fox, T.; Caldwell, J. W.; Kollman, P. A. J. Am. Chem. Soc. 1995, 117,

5179.

(15) Vorobyov, I.; Yappert, M. C.; DuPre, D. B. J. Phys. Chem. A, 2002, 106, 668.

(16) Singh, U. C.; Kollman, P. A. J. Comp. Chem. 1984, 5, 129.

24
(17) M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R.

Cheeseman, V. G. Zakrzewski, J. A. Montgomery, Jr., R. E. Stratmann, J. C. Burant, S.

Dapprich, J. M. Millam, A. D. Daniels, K. N. Kudin, M. C. Strain, O. Farkas, J. Tomasi, V.

Barone, M. Cossi, R. Cammi, B. Mennucci, C. Pomelli, C. Adamo, S. Clifford, J. Ochterski, G.

A. Petersson, P. Y. Ayala, Q. Cui, K. Morokuma, D. K. Malick, A. D. Rabuck, K. Raghavachari,

J. B. Foresman, J. Cioslowski, J. V. Ortiz, A. G. Baboul, B. B. Stefanov, G. Liu, A. Liashenko,

P. Piskorz, I. Komaromi, R. Gomperts, R. L. Martin, D. J. Fox, T. Keith, M. A. Al-Laham, C. Y.

Peng, A. Nanayakkara, M. Challacombe, P. M. W. Gill, B. Johnson, W. Chen, M. W. Wong, J.

L. Andres, C. Gonzalez, M. Head-Gordon, E. S. Replogle, and J. A. Pople, Gaussian 98,

Revision A.9, Gaussian, Inc., Pittsburgh PA, 1998.

(18) Cances, M. T.; Mennucci, V.; Tomasi, J. Chem. Phys. 1997, 107, 3032.

(19) Harris, J. G.; Yung, K. H.; J. Phys. Chem. 1995, 99, 12021.

(20) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, Klein, M. L. J. Chem. Phys.

1983, 79, 926.

(21) Lyubartsev, A. P.; Laaksonen, A. Comput. Phys. Commun. 2000, 128, 565.

(22) Jellinek, J.; Li, D. H. Phys. Rev. Let. 1989, 62, 241.

(23) Franckl, M. M.; Chirlian, L. In Reviews in Computational Chemistry Lipkowitz, K. B.;

Boyd, D. B.; Ed.; Volume 14, Wiley-VCH, 2000, 1-31.

(24) Bayly, C. I.; Cieplak, P.; Cornell, W. D.; Kollman, P. A. J. Phys. Chem. 1993, 97, 10269.

(25) Mate-Divo, M.; Barcza, L. Zeit. Physicalische Chemie 1995, 190, 223.

(26) Silva, C. F. P.; Duarte, M. L. T. S.; Fausto, R. J. Mol. Struct. 1999, 482-483, 591.

(27) Cheng, S.; Meisen, A., Chakma, A. Hydrocarbon Processing, 1996, February, 81.

25
26
Supporting Information for:
Molecular Dynamics Study of Ethanolamine as a Pure Liquid and in Aqueous
Solution
Eirik F. da Silva, Tatyana Kuznetsova and Bjørn Kvamme

Potential Parameters of MEAa

Bond r0 [Å]

N-H 1.012
N-C 1.471
C-C 1.524
C-O 1.419
O-H 0.966
Angle θ0 kθ [J/mole]

H-N-H 107.4 269.39


H-N-C 111.4 316.65
N-C-C 109 506.21
C-C-O 111.4 546.91
C-O-H 105.8 298.79

Dihedral C1a C2 a C3 a C4 a C5 a

H-N-C-C -11.05 -2.37 25.30 0.43 -2.02


N-C-C-O -17.78 23.60 36.81 -16.64 -13.32
C-C-O-H -19.15 -2.79 13.61 -0.12 1.55
Site σ [Å] ε [J/mole] q

H(N) 0 0 0.347
N 3.25 0.7108 -0.938
CH2 3.905 0.494 0.257
O 3.07 0.7108 -0.644
H(O) 0 0 0.374
a
In KJ/mole.
Potential Parameters of CO2

Bond r0 [Å]

C-O 1.149
Angle θ0 kθ [J/mole]

O-C-O 180 236.3


Site σ [Å] ε [J/mole] q

C 2.757 0.2334 0.6512


O 3.033 0.6694 -0.3256
a
In kJ/mole.
Dihedral angle distribution for pure MEAa at 333K.
Dihedral angle distribution for pure MEAb at 333K.
Dihedral angle distribution for pure MEAb at 298K.
Dihedral angle distribution for 10 mol percent MEAa in aqueous solution at 333K.
Dihedral angle distribution for 10 mol percent MEAb in aqueous solution at 333K.
Paper VII

Comparison of Solvation Models in the Calculation of Amine Basicity

Eirik Falck da Silva, Takeshi Yamazaki and Fumio Hirata

2005
Theoretical Study of Amine Basicity in Solution with RISM-SCF Calculations and Free Energy

Perturbations

Eirik F. da Silva*, Takeshi Yamazaki and Fumio Hirata

Contribution from the Department of Chemical Engineering, Norwegian University of Science and

Technology, Trondheim, Norway and the Institute of Molecular Science, Okazaki, Japan

silva@chemeng.ntnu.no

Abstract

The relative basicity of a series of 10 amines in aqueous solution was calculated with three separate sets

of solvation energies and gas phase basicities calculated at the B3LYP/6-311++G(d,p) level. The

solvation energies were calculated with RISM-SCF and two sets of Monte Carlo simulations. RISM-

SCF is a method combining ab initio solute description with the reference interaction site method in

statistical mechanics to describe solvation. The simulations were free energy perturbations with

classical force fields and charges derived from ab inito calculations. Two different types of charges

were studied. Both the RISM-SCF calculations and simulations produced results in reasonable

agreement with experimental data. The level of agreement between RISM-SCF and free energy

perturbations is consistent with the similarity in solute and solvent representation. Explicit solute

polarization, included in RISM-SCF but not simulations, did have a significant effect on relative

solvation energies. Solvation energies were found to be sensitive to the scheme utilized to determine

atomic charges.

1
Introduction

Many important chemical processes take place in solution and the effects of the solvent are often of

crucial importance. Solution phenomena are therefore of interest in many fields of chemistry, from

chemical engineering to biochemistry. Considerable effort has also gone into the study of solution

phenomena in computational chemistry, and while various methods in computational chemistry have

contributed to the understanding of solvation phenomena it is also clear that the modeling of solvation

effects remains challenging.

One of the solvation properties of greatest interest is the free energy of solvation. It is key both to

modeling chemical equilibrium in the solution and vapor-gas equilibrium. A fairly large number of

schemes have been developed to calculate the free energy of solvation. The most common forms of

solvation models in the context of computational chemistry are simulations and continuum models.

Some reviews offer general coverage of most schemes1-4 while others focus on continuum models. 5-6

In the present work the RISM-SCF method and simulations were used for the calculation of solvation

energy. RISM-SCF7-8 is a method developed from the statistical mechanics of molecular liquids. While
9
various forms of RISM have successfully been applied to a number of issues in the modeling of liquids

it has still not received much attention as a general solvation model. The simulations were Monte Carlo

free energy perturbations (FEP). The RISM-SCF calculations and FEP were carried out with similar,

but different, solute and solvent representation. Comparison of the results serves to validate the

methodologies, and at the same time this can give insight into what effect differences in solute and

solvent representation have on calculated solvation energies.

The solvation energies obtained from RISM-SCF and simulations were used to calculate basicities of

a series of amine molecules. While estimates of free energy of solution can be made directly from

experimental data for neutral species, this is much more difficult for ionic species. Calculation of base,

or acid, strength offers an indirect route to testing the ability of a method to calculate the energy of both

2
neutral and ionic species. For base strength fairly accurate experimental data are available for a large

number of molecules.

It should be noted that both RISM-SCF and simulations can be used to acquire a microscopic picture

of solvation effects, for example in terms of radial distribution functions. In the present work we will

however focus on solvation energies and solute atomic charges.

The dissociation of the conjugate base of a molecule can be written as

BH + + H 2O U B + H 3O + (1)

assuming the molfraction based activity of water to be 1 and writing H3O+ as H+ the following

equilibrium constant is obtained:

aB aH +
Ka = (2)
aBH +

The definition of pKa is:

pK a = − log K a (3)

The free energy of protonation in aqueous solution ( ∆G ps ) is related to Ka by the following equation:

∆G ps = −2.303RT log K a (4)

This gives us the relation between pKa and ∆G ps .

1
pK a = ∆G ps (5)
2.303RT
3
Model predictions for ∆G ps should therefore give a linear correlation with the pKa.

The general approach for calculating ∆G ps is to use a thermodynamic cycle:10

where ∆G pg is the gas phase basicity and the ∆Gs ’s are the solvation energies of the different species

involved.

In the present work the relative basicity in solution, will be studied. The relative basicity is defined as:

∆∆G ps = ∆Gg ( B ) −∆Gg ( B ref ) (6)

Ammonia will be used as reference base ( B ref ). The energy of the proton itself (H ) is not needed to
+

determine relative basicity and is not included in the calculations. Based on the thermodynamic cycle,

∆∆G ps can be divided into the relative gas phase basicity ( ∆∆G pg ) and contributions from solvation:

∆G pg = ⎡⎢⎣Gg ( B ) − Gg ( NH 3 )⎤⎦⎥ − ⎡⎢Gg ( BH + ) − Gg ( NH 4+ )⎤⎥ (7)


⎣ ⎦

and

∆∆G ps = ∆∆G pg ( B ) + [∆Gs ( B ) −∆Gs ( NH 3 ) ] − ⎡⎢⎣∆Gs ( BH + ) −∆Gs ( NH 4+ )⎤⎥⎦ (8)

This can be written as:

∆∆G ps = ∆∆G pg ( B ) + ∆∆Gs ( B ) −∆∆Gs ( BH + ) (9)

Given a reasonably accurate model for the calculation of relative gas phase basicity the challenge is

reduced to calculating the relative solvation energies.

4
Methods

Gas Phase Basicity

The theoretical basicity of a series of amines at B3LYP/6-311++G(d,p) level with thermal

corrections to the free energy and the zero point energies (ZPE) calculated at HF/6-31G(d) level have

recently been published.11 It was in the same publication shown that basicities calculated at this level are

in good agreement with experimental values. For 2,2,6,6-Tetramethyl-4-piperidinol (TMP) we have

performed calculations at the same theoretical level as these published values. Gas phase energy

calculations were performed with Gaussian 98.12

Solvent Phase Geometry

Amine geometries were optimized in gas phase and the same geometries were used for all

calculations in solution. The RISM-SCF calculations were performed on gas phase HF/6-31G(d,p)

geometry, similar to what was utilized in a previous study.13 Simulations were carried out on HF/6-

31G(d) gas phase geometries. The RISM-SCF calculations were performed without optimization of

solute geometry. Optimization of solute geometry is in general expected to have a limited effect on
14,15
calculated energies, something that should hold true for the fairly rigid molecules studied in the

present work. We have chosen to carry out calculations on a widely used and not particularly large

basis set. In this work we chose to calculate gas phase energies and geometry at a different level from

the solution phase calculations. While it might seem more consistent to use the same level of theory for

calculations in gas phase and in solution, it must be noted that the relationship between level of theory

and quality of results is not the same in the two phases. In gas phase a level of theory that is accurate in

calculating basicities is desired. For the solvation energy calculations partial atomic charges and

geometry are required as input, and the level of theory should be selected keeping these properties in

mind.

5
RISM-SCF

The Reference Interaction Site Model (RISM) is a method originally developed by Chandler and
16 7,8,
Anderson in statistical mechanics for the description of interactions in molecular liquids. RISM-SCF
17
(Self-Consistent Field) is an extension that allows the simultaneous calculation of distribution of

solvent molecules and the electronic structure of the solute molecule. For the calculation of free energy

of solvation, the equation derived by Singer and Chandler is employed.18 In the present work the focus

will be on the molecular representation, for description of theory and method we refer the reader to

previous work.9

In the RISM-SCF framework, the solute molecule is described by ab initio electronic structure theory

and the solvent is represented by a classical force field. The interaction between the solute and solvent

is described as a sum of classical Coulomb and Lennard-Jones potentials. The Coulomb potential is the

potential between the partial charges of the solvent and partial charges of solute sites derived from the

electronic wave function of the solute molecule. In the present RISM-SCF calculations Hartree-Fock

theory is utilized. The contribution of the solvent reaction-field is introduced in the Fock operator of the
8, 17
solute:

Fi solv = Fi vacuum − f i ∑ Vλ bλ (9)


λ∈u

where Fi vacuum is the Fock operator in vacuum, fi is the occupation number of orbital i and bλ is the

population operator that determines the partial charge on site λ in the solute molecule (subscript u ).

Vλ is obtained as the electrostatic potential on site λ in the solute molecule generated by the atomic

charges on the solvent molecules (subscript v ):



hλα (r )
Vλ∈u = ρ∑ qα ∫ 4πr 2 dr (10)
α∈v 0
r

6
where hλα is the pair correlation function between site λ on the solute molecule and site α on the

solvent molecule. The pair correlation function is determined from the RISM equations.19 The solvent

effect is in this calculation represented by the microscopic distribution of the charges on the solvent

molecules. A RISM-SCF calculation begins with deriving solute partial atomic charges from the gas

phase electronic structure in a quantum mechanical calculation. From these partial atomic charges the

solute-solvent interactions are tabulated. The solvent structure around the solute is then calculated using

the RISM equation. The solute electronic structure and partial atomic charges are then recalculated from

the solvated Fock operator. These calculations proceed in an iterative cycle until the electronic structure

of the solute and solvent distribution converge.

Atomic charges are in the present RISM-SCF calculations represented with MK* charges described in

the following section. The solute Lennard-Jones parameters are from the all-atom OPLS force field.20,21

The choice was made to use the same Lennard-Jones parameters for the neutral and protonated forms of

the amines. The solvent was represented with the TIP3P22 water model. Hydrogen atom sites with zero

Lennard-Jones parameters in the OPLS force field are augmented with a small core to facilitate the

RISM calculations.8 Calculations were performed at 298.15K and a density of 0.997 g/cm3 (0.03334
23
molecules/Å). To improve convergence the modified-DIIS method for RISM was utilized. In the

present work the RISM equations were solved with the hyper-netted chain (HNC) approximation. 24 It

should be noted that the use of RISM with the HNC closure produces results that are known to
25
overestimate the effect of solute size on solvation energy. In the present work where the focus is on

calculating the relative solvation energies this effect is expected to cancel out.

Simulations

Monte Carlo FEP simulations were carried out in a NPT ensemble at 298.15K and 1 atmosphere

pressure. These calculations were performed with BOSS version 4.126 using procedures developed by

Jorgensen et al.20, 27 A single solute molecule was placed in a periodic cube with 267 TIP4P22 water

molecules. Periodic boundary conditions were applied. A number of water molecules corresponding to

7
the number (n) of non-hydrogen atoms in the amine molecule were removed, giving 267 - n water

molecules. The perturbations were carried out over 10 windows of double-wide sampling giving 20 free

energy increments that are summed up to give the total change in the free energy of solvation. Each

window had 500000 steps for equilibration and another 500000 for sampling. FEPs were used to

calculate the relative free energies between neutral species and in a separate series the relative free

energies of the protonated forms. Simulations were carried out between the amines closest in size.

The Lennard-Jones potential parameters from the all-atom OPLS force field20,21 ( same as used for the

RISM-SCF calculations) were utilized. These were used together with two different types of charges

derived from quantum mechanical calculations. The description of the charges is given in the following

section. These simulations are very similar in form to what has been presented by Wiberg et al.,28 the

main difference being the choice of routines to calculate the atomic charges.

The simulations, unlike the RISM-SCF calculations, are not deterministic and each perturbation has a

statistical uncertainty. This was by the batch means procedure20 determined to be between 0.1 to 0.5

kcal/mol for each simulation. While this uncertainty is significant it will be seen in the results that

energy differences are in general larger than these uncertainties.

In the present study the focus is on the description of solute partial atomic charges. The main

difference between the RISM-SCF and simulations lies in the schemes to calculate the atomic charges

and the use of a polarizable solute representation, in which the energy to polarize the solute is also

accounted for, in the RISM-SCF calculations.

Atomic Charges

In the RISM-SCF and the simulations atomic charges are used together with a Lennard-Jones

potential both in the solute and solvent representation. Intermolecular energies between molecules a and

b are then written in the following well-known form:

8

⎪ ⎡⎛ ⎞12 ⎛ ⎞6 ⎤⎫
⎪ qi q j e 2
on a on b
⎢ σij ⎟ σij ⎟ ⎥⎪ ⎪
∆Eab = ∑∑ ⎨⎪ + 4εij ⎢⎜⎜⎜ ⎟⎟ − ⎜⎜⎜ ⎟⎟ ⎥⎪⎬ (11)
⎪ rij
j ⎪ ⎢⎜⎝ rij ⎠⎟ ⎝⎜ rij ⎠⎟ ⎥⎪
⎦⎪
i

⎩ ⎣ ⎪

Where q is the atomic charge, ε and σ are the Lennard-Jones potential parameters. In both RISM-

SCF calculations and simulation, solute partial atomic charges derived from quantum mechanical

calculations were utilized. Atomic charges are not uniquely defined and a number of different schemes

for their calculation have been proposed. These are often based on reproducing some form of

observable quantities or properties from quantum mechanical calculations. One of the more common

schemes is to reproduce the electrostatic potential around the solute. Even for this approach there are

however a number of different implementations. 29-33 Singh and Kollmann29 developed a procedure based

on reproducing the electrostatic potential on gridpoints distributed spherically around each solute atom

center, outside the van der Waals volume of the solute. This scheme will in the present work be referred

to by its common acronym MK. In the RISM-SCF calculations a procedure similar to the one proposed

by Singh and Kollmann is utilized. A grid was set up with nine spherical layers around each atom

center. Layers with equal thickness were utilized, the inner having a radius of 1.5 Å and the outer a

radius of 2.4 Å. For each layer 36 gridpoints were evenly distributed. Any gridpoints within the van der

Waals radii of the solute atoms were disregarded. The atomic charges were then fitted to optimize the

agreement, as measured by the mean square error, with the electrostatic potential at these gridpoints.

Charges calculated by this procedure will be referred to as MK*. In the first cycle of the RISM-SCF

calculation the gas phase charges are calculated, the (converged) solution phase charges are obtained in

the final step of the iterative loop.


34
It has been observed by Besler, Merz and Kollman that the Hartree-Fock 6-31G(d) procedure

overestimate gas phase dipole moments by 10-20%. The same authors suggest that charges calculated at

this level therefore implicitly account for solvation effects. This approach has later been implemented

by Cornell et al. 35 in a simulation force field. In the RISM-SCF calculation polarization is added to such

charges, and there might be a risk of overestimation of polarization in solution.

9
Simulations were carried out with charges derived from two different schemes. One scheme is

reproduction of the electrostatic potential in the gas phase. Here the standard Gaussian 9811 version of

the MK29 scheme was utilized. Charges were calculated at the HF/6-31G(d) level. Here gas phase values

were chosen. As noted above gas phase Hartree-Fock level charges are believed to be inherently high,
34
and we adopt the approach suggested by Besler et al. of using these gas phase charges without

modification in solution. Free energy perturbations carried out with the combination of these charges

and the OPLS force field will be referred to as FEP-MK.

The second type of charges utilized in the simulations is CM236 charges calculated at the HF/6-31G(d)

level. While this model is also based on the quantum mechanical description of the solute, it involves

semi-empirical corrections to reproduce experimental gas phase dipole moments. With the CM2 charges

there is no implicit polarization present, and the charges should be adjusted to the levels one would

expect in solution. Here we chose to do the calculations with a continuum solvent model to obtain

solution phase charges. CM2 charges were therefore calculated with the SM 5.42R solvent field model37

in Gamesol.38 Free energy perturbations carried out with the combination of these charges and the OPLS

force field will be referred to as FEP-CM2.

While atomic charges are not uniquely defined some criteria have been presented for when they can
32, 33
be considered reasonable. In the present work the purpose is to calculate solvation energies and in

this context we argue that the best charges are those that produce free energies of solvation closest to

experimental values. Charges should also take on values that are not unreasonably large and should

reflect molecular symmetry.

Molecules

In Figure 1 are shown the amine molecules studied. The purpose of the present study was to look at

how RISM-SCF and simulations perform in general prediction of solvation energies. The set of amines

was therefore chosen to include varied forms of molecular geometry. A second consideration in

selecting the amines has been to use molecules for which experimental gas phase and solution phase

10
basicity data are available. No gas phase value has however been reported for 2,2,6,6-Tetramethyl-4-

piperidinol (TMP). Finally the set has been limited to molecules with no conformers or known

conformers (piperazine). For piperazine calculations have been done on the chair-conformer.

Figure 1. Amine molecules studied.

Results

In Table 1 calculated and experimental gas phase basicity for the amines are shown. All data are

given relative to ammonia, a convention that will be used for all basicity results presented. While

neither experimental nor calculated gas phase basicites are likely to be completely accurate, the

agreement between the data offers a level of mutual validation. While no experimental data is available

for TMP, the agreement between theoretical and experimental results for the other molecules would

suggest that the level of theory is adequate for this molecule too. It should be noted that the present

values differ significantly from what have been used in previous studies13,14 on ammonia, methylamine,

dimethylamine and trimethylamine. The present values are in better agreement with currently accepted

experimental values (Table 1) and are likely to be the more accurate.

11
Table 1. Relative Gas Phase Protonation Energies. Data in [kcal/mol].

Molecule Theoreticala Experimentalb

Ammonia 0.0 0.0


Methylamine 10.2 10.9
Ethylamine 14.3 14.1
Dimethylamine 17.9 18.5
Trimethylamine 22.2 23.7
Piperidine 24.2 24.4
Piperazine 23.3 22.9
Morpholine 17.0 17.3
Pyrrolidine 23.6 23.0
2,2,6,6-Tetramethyl-4-piperidinol (TMP) 29.7
a
B3LYP/6-311++G(d,p) energy with thermal correction and ZPE calculated at HF/6-31G(d), TMP
data are from present work, other data from da Silva.11 b Data from Hunter and Lias.39

In Table 2 the atomic charges of the amine functionalities for the neutral amines are given and in

Table 3 data for the protonated forms are given. MK* charges are given both for gas phase and solution

(from RISM-SCF calculation). MK charges are in gas phase and CM2 charges are in solution (SM

5.42R solvation model). These are the same form of the charges as utilized in the energy calculations.

The MK* gas phase charges are however only used as a starting point for the RISM-SCF calculations.

12
Table 2. Partial Atomic Charges of Neutral Amines.

Molecule MK* MK* MK CM2


a b
gas phase solution gas phase solution
c c c c
N H N H N H N H

Ammonia -1.02 0.34 -1.28 0.43 -1.11 0.37 -0.86 0.29


Methylamine -1.01 0.35 -1.18 0.45 -1.03 0.38 -0.75 0.33
Ethylamine -1.02 0.35 -1.33 0.46 -1.06 0.37 -0.72 0.32
Dimethylamine -0.75 0.35 -0.96 0.44 -0.72 0.38 -0.55 0.28
Trimethylamine -0.41 ---- -0.55 ---- -0.23 ---- -0.41 ----
Piperidine -0.83 0.37 -1.17 0.52 -0.80 0.36 -0.54 0.28
Piperazine(1) -0.73 0.34 -0.94 0.43 -0.74 0.37 -0.53 0.27
Piperazine(2) -0.80 0.36 -1.07 0.47 -0.75 0.39 -0.54 0.28
Morpholine -0.82 0.38 -1.16 0.53 -0.77 0.37 -0.54 0.28
Pyrrolidine -0.81 0.30 -0.98 0.45 -0.75 0.35 -0.53 0.33
TMP -1.14 0.37 -1.67 0.52 -1.06 0.38 -0.52 0.35
a b c
RISM-SCF method. SM 5.42R solvation model. Amine functionality hydrogen atoms, average
value for amine group with more than one hydrogen atom.

13
Table 3. Partial Atomic Charges of Protonated Amines.

Molecule MK* MK* MK CM2


a b
gas phase solution gas phase solution
c c c c
N H N H N H N H

Ammonia -0.78 0.44 -0.81 0.45 -0.87 0.47 -0.52 0.38


Methylamine -0.21 0.31 -0.26 0.33 -0.30 0.32 -0.44 0.37
Ethylamine -0.65 0.39 -0.72 0.43 -0.69 0.40 -0.69 0.40
Dimethylamine -0.05 0.29 -0.06 0.33 -0.05 0.30 -0.34 0.35
Trimethylamine 0.17 0.31 0.14 0.37 0.08 0.34 -0.25 0.34
Piperidine -0.25 0.31 -0.29 0.35 -0.28 0.32 -0.32 0.35
Piperazine(1)d -0.23 0.33 -0.32 0.39 -0.22 0.33 -0.32 0.35
Piperazine(2) -0.83 0.42 -1.09 0.52 -0.81 0.42 -0.53 0.29
Morpholine -0.24 0.32 -0.33 0.38 -0.32 0.36 -0.32 0.36
Pyrrolidine -0.23 0.30 -0.28 0.35 -0.34 0.33 -0.53 0.33
TMP -0.13 0.22 -0.31 0.31 -0.93 0.45 -0.34 0.34
a b c
RISM-SCF method. SM 5.42R solvation model. Amine functionality hydrogen atoms, average
d
value for amine group with more than one hydrogen atom. Protonated site.

It can be observed that in the RISM-SCF results polarization is significantly higher for the neutral

(unprotonated) amines than for the protonated forms. For the neutral species charges on the amine

functionalities can be seen to increase with 20-40% over the gas phase values, while the increase for the
13
protonated forms are around 10-20%.The same observation was made in previous RISM-SCF work. It

was in that work suggested that this was caused by the (uncharged) amines producing an anisotropic

field in the solvent, which again causes an anisotropic reaction field on the solute giving additional

electrical polarization. The protonated amines will on the other hand produce a more isotropic field and

the same effect will not be seen. Most of the amines show comparable level of increases in charges in

soultion, but particularly large increases can be observed for neutral (uncharged) TMP, piperidine and

morpholine. For TMP the value would appear to be greater than what can be considered physically

reasonable.

14
The amine functionalities and their representation are expected to be the most important in

determining the solvation behavior of the molecules. Some observations will however be made on the

other charges, the full set of data is given in the supporting information.

In most cases neutral amine gas phase MK* and MK were in reasonable agreement. Carbon charges

were mostly in a range between 0 to 0.4, while hydrogen charges were close to zero. For trimethylamine

the average MK carbon charge was however –0.29 while the average MK* value was 0.07. In some

cases MK and MK* charges took on unreasonable high values. For TMP the MK carbon charges varied

between 0.87 and –0.53, while the MK* charges varied between 0.88 and -0.41. In most cases MK and

MK* charges were quite similar for atoms in symmetrically equivalent positions. Both MK and MK*

charges were however somewhat different for carbons in equivalent positions in morpholine.

The CM2 charges for the alkane functionalities were much more stable, reflecting the different nature

of the model. Carbon charges stayed mostly in a range of 0 to –0.3 for neutral amines. Alkane-

hydrogens mostly had charges around 0.1. The CM2 charges did in general display values that would

appear not to be unreasonable in terms of the dimension of the charges.

In summary it can be observed that MK* and MK charges are quite similar, which is reassuring since

they are only variations of fittings to the same electrostatic potential. The trends in charges for different

molecules would seem to be similar for the MK and CM2 charges. The CM2 charges are however

systematically lower than the MK charges, despite the fact that CM2 charges have been scaled up from

their gas phase values with the use of a continuum model. The MK/MK* charges also tend to fluctuate

more than the CM2 charges, this being particularly true for sp3 carbon atoms.

In Table 4 the relative solvation energies are shown, while in Table 5 the experimental pKa values and

relative basicity in solution calculated from these are shown together with the basicity in solution

derived from RISM-SCF and simulations.

15
Table 4. Total Relative Solvation Energy Contribution. Data in [kcal/mol].

Molecule ∆∆Gs ( B) −∆∆Gs ( BH + )

RISM-SCF FEP-MK FEP-CM2

Ammonia 0 0.0 0.0


Methylamine -8.7 -10.4 -8.2
Ethylamine -9.3 -12.0 -9.1
Dimethylamine -14.0 -14.6 -14.8
Trimethylamine -17.4 -18.3 -22.4
Piperidine -23.7 -23.0 -17.9
Piperazine -19.3 -21.6 -17.2
Morpholine -19.3 -15.2 -14.4
Pyrrolidine -18.5 -18.4 -20.9
TMP -33.2 -21.1 -24.0

Table 5. Relative Basicity in solution. Data in [kcal/mol].

∆∆G ps
a
Molecule exptl pKa

exptl RISM-SCF FEP-MK FEP-CM2


b
Ammonia 9.24 0.00 0.0 0.0 0.0
b
Methylamine 10.65 1.92 1.5 -0.2 2.1
b
Ethylamine 10.78 2.10 5.0 2.3 5.3
b
Dimethylamine 10.8 2.13 3.9 3.3 3.1
b
Trimethylamine 9.80 0.90 4.8 3.9 -0.2
c
Piperidine 11.12 2.56 0.5 1.2 6.3
c
Piperazine 9.83 0.80 4.0 1.7 6.1
c
Morpholine 8.49 -1.02 -2.3 1.7 2.6
c
Pyrrolidine 11.30 2.81 5.1 5.2 2.7
c
TMP 10.05 1.10 -3.5 8.6 5.7
a b 10 c 40
Energies relative to ammonia. Data from Jones and Arnett. Data from Perrin.

16
In Figure 2 the calculated gas phase basicities are plotted against the experimental pKa of the amines,

the low correlation in this plot shows how important the solvation energies are in determining basicity

in solution. In Figure 3 the relative pKa values calculated from RISM-SCF and FEP-MK are plotted

against the experimental pKa. In Figure 4 the pKa values determined from FEP-MK and FEP-CM2 are

plotted against experimental pKa. The stippled line in the figures indicates the theoretical ratio between

pKa and protonation energy from equation 5. All data are relative to ammonia.

Figure 2. Calculated gas phase basicity versus experimental pKa. The stippled line indicates the

theoretical trend relative to ammonia.

17
Figure 3. Calculated pKa versus experimental pKa. Crosshairs are RISM-SCF and Open circles are FEP-

MK. The stippled line indicates the theoretical trend relative to ammonia.

Figure 4. Calculated pKa versus experimental pKa. Open circles are FEP-MK and black circles are FEP-

CM2. The stippled line indicates the theoretical trend relative to ammonia.
18
In Figure 5 the contributions to the RISM-SCF solvation energy are shown together with the FEP-MK

solvation energies. The contributions to the RISM-SCF solvation energy are the solute polarization and

excess chemical potential from interactions with the solvent. The plot shows the solute polarization to

have a relatively small, but significant, effect on relative solvation energies. The figure also illustrates

the good overall agreement between RISM-SCF and FEP-MK solvation energies.

Figure 5. Contributions to RISM-SCF solvation energies. RISM-SCF (Electronic) refers to the solute

polarization energy. RISM-SCF (Excess) refers to the excess chemical potential from interactions with

the solvent.

19
Discussion

Basicity

Looking at Figures 3 and 4 it can be seen that neither RISM-SCF nor simulations produce full

quantitative agreement with experimental data. Comparing the solvation models results with the gas

phase energies in Figure 2 it can however be seen that the large gap between relative energies in the gas

phase and solution is closed. It must also be kept in mind that this set of amines contains large

variations in geometry and that the experimental energy differences to be reproduced are relatively

small.

The RISM-SCF results are mostly in reasonable agreement with the experimental data. The largest

error was for TMP. As noted in the Results section the MK* charges for this molecule were very high.

These charges are most likely the cause of the inaccurate solvation energies.

The present results differ somewhat from previous RISM-SCF results13 for ammonia, methylamine,

dimethylamine and trimethylamine. This is due to changes in the scheme to calculate atomic charges,

illustrating the sensitivity of the results to the choice of scheme.

The FEP-MK simulations and RISM-SCF calculations utilize similar charges, same Lennard-Jones

potential and similar solvent representation (TIP4P and TIP3P respectively). The agreement between

the results (Figure 3 and Figure 5) is consistent with the underlying similarities in solute and solvent

representation. For TMP the FEP-MK results and RISM-SCF results differ strongly from each other and

neither is close to the experimental value. In this case the MK and MK* charges were quite different,

both taking on unreasonable values. For morpholine the RISM-SCF result was in better agreement with

the experimental data than the FEP-MK result. This difference is probably due to the increase of solute

charges from solvent polarization in the RISM-SCF model. In this particular case this contribution

would appear to be necessary to predict the basicity relative to ammonia.

The FEP-MK and FEP-CM2 simulations only differ in the scheme to calculate atomic charges.

Comparing the two sets of results in Figure 4 one can see that the choice of scheme for calculating

atomic charges has a considerable effect on the solvation energies.

20
13,14
Some previous studies have looked at the irregular trend in basicity for the series ammonia,

methylamine, dimethylamine and trimethylamine. Both the RISM-SCF and simulations give a

decreasing trend in solvation energy of the protonated forms of these amines and stable values for the

neutral forms. If this trend in the solvation energies is taken together with the increasing trend in gas

phase basicities the irregular order in basicity is readily accounted for. The models can therefore be said

to capture this in a qualitative sense. At a quantitative level none of the models is however accurate

enough to confidently reproduce the small energy differences involved.

Atomic Charges

It can clearly be seen from the results, both for RISM-SCF and simulations, that the free energies are

sensitive to the scheme for calculating atomic charges. Some examples are also seen in the present study

of MK and MK* charges taking on unphysical values resulting in unreasonable solvation energies. This

is related to a known issue41 with fitting charges to the electrostatic potential: if an atomic center is far

away from the grid points in the fitting its value becomes ill-determined and the charges can fluctuate

dramatically without significantly altering the quality of the fit. This is often the case for sp3-carbons, in

the fitting they can become buried behind the atoms they are bonded to. In the present work the problem

was particularly large for TMP. This is not surprising given the structure of the molecule. Because of

the methyl substituents there are few gridpoints close to the amine functionality. Several atoms are

almost completely buried. This can result in charges that fluctuate widely. In the RISM-SCF

calculations polarization effects are added and ill-determined charges can fluctuate even more.

It has been suggested that this problem can be handled by adding constraints on the fitting. In the

RESP41 procedure the carbon-charges are constrained. This approach should resolve some of the

problems seen in the present work, but further work is perhaps needed to determine what are the best

constraining conditions for free energy calculations. A hybridization between fitting to the electrostatic

potential and semi-empirical corrections to reproduce different properties has also been proposed.42

Another modification that might improve the fitting is to include Boltzmann-weighting of the points

21
32
being fitted. It is clear that fitting atomic charges to the electrostatic potential is difficult and with the

scheme used in the present work not entirely reliable. The proposals to improve the fitting schemes do

however suggest that this approach to determine atomic charges can be developed further.

As noted in the Methods section the addition of solvent polarization to already high Hartree-Fock

level charges in the RISM-SCF calculations might result in too high charges. In future work the RISM-

SCF calculations should perhaps be performed at another level of theory or with some form of scaling

of charges to correct for the overestimation.

RISM-SCF

Improvement of the schemes to calculate the atomic charges are likely to lead to a more accurate

RISM-SCF model. Work has also been done on directly calculating the interaction between solvent
42
molecules and the electrostatic field of the solute, an approach that eliminates the task of determining

atomic charges.

One important approximation in the present work, and most simulations, is the use of a solvent model

with fixed charges. It has been suggested44 that introduction of polarizable solvent models have a

significant effect on solvation energies. This is an issue that should be explored further.

All together this would suggest that there is room for further refinement of the RISM-SCF method to

produce a general and accurate solvation model. It should also be noted that the present implementation

of RISM-SCF is fairly robust in terms of convergence.

Simulations

As with the RISM-SCF calculations the main issue with the simulations is the calculation of atomic

charges. In the simulations it is however less clear how polarization effects should be incorporated. In

the present work a continuum model and inherently high Hartree-Fock level charges have been used to

obtain charges in solution. In the literature other schemes can also be found such as scaling up gas phase

values and calculating the polarization from the average electrostatic potential of the solvent in

22
45
QM/MM simulations. Charges derived from RISM-SCF calculations could also be an interesting

option in simulations, particularly as they are derived in the context of the same solvent representation.

The choice of scheme to calculate charges is clearly of great importance and is likely to remain an area

of active research.

In RISM-SCF the charges include polarization effects and the energy to polarize the solute is included

when determining the solvation. It remains to be determined if and how this contribution should be

accounted for in simulations.

Conclusions

In the present work RISM-SCF and simulations have been used to calculate the relative basicity of a

series of 10 amine molecules. Results showed mostly reasonable agreement between experimental data

and calculated values. The results were however found to be sensitive to the scheme for calculation of

atomic charges. None of the methods used for determining charges in the present work were found to be

completely satisfactory. There would however appear to be a number of ways in which such schemes

can be improved upon. Comparison between results with (RISM-SCF) and without (simulations)

polarizable solute representation suggests that polarization can have a significant effect on relative

solvation energies.

Acknowledgment

Gratitude is expressed to the Japan-Norway Sasakawa foundation for supporting a visit by Eirik F. da

Silva to the Institute for Molecular Science.

23
Supporting Information Available:

Atomic charges for alkane and alcohol groups, Lennard-Jones force field parameters and data utlized

in Figure 5 are given in the supporting information. This material is available free of charge via the

Internet at http://pubs.acs.org.

24
References

1 Cramer, C. J. Essentials of Computational Chemistry, John Wiley & Sons, 2002.

2 Orozco, M.; Javier Luque, F. Chem. Rev. 2000, 100, 4187.

3 Tomasi. J.; Menucci, B.; Cammi, R. In Handbook of Molecular Physics and Quantum

chemistry Wilson, S. Ed.; Volume 3, John Wiley & Sons, 2002, 299.

4 Bacskay, G. B.; Reimers, J. R. In Encyclopedia of computational chemistry, Schleyer, P. R.

Ed.; John Wiley & Sons, 1998, 2620.

5 Cramer, C. J.; Truhlar, D. G. Chem. Rev. 1999, 99, 2161.

6 Tomasi, J.; Persico, M. Chem. Rev. 1994, 94, 2027.

7 Ten-no, S.; Hirata, F.; Kato, S. Chem. Phys Lett., 1993, 214, 391.

8 Ten-no, S.; Hirata, F.; Kato, S. J. Chem. Phys. 1994, 100, 7443.

9 Hirata, F. Ed. Molecular Theory of Solution, Kluwer Academic Publishers, 2003.

10 Jones, F. M.; Arnett, E. M. Prog. Phys. Org. Chem. 1974, 11, 263.

11 da Silva, E. F. J. Phys. Chem. A, 2005, 109, 1603.

12 M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, V.

G. Zakrzewski, J. A. Montgomery, Jr., R. E. Stratmann, J. C. Burant, S. Dapprich, J. M.

Millam, A. D. Daniels, K. N. Kudin, M. C. Strain, O. Farkas, J. Tomasi, V. Barone, M. Cossi,

R. Cammi, B. Mennucci, C. Pomelli, C. Adamo, S. Clifford, J. Ochterski, G. A. Petersson, P.

Y. Ayala, Q. Cui, K. Morokuma, D. K. Malick, A. D. Rabuck, K. Raghavachari, J. B.

Foresman, J. Cioslowski, J. V. Ortiz, A. G. Baboul, B. B. Stefanov, G. Liu, A. Liashenko, P.

Piskorz, I. Komaromi, R. Gomperts, R. L. Martin, D. J. Fox, T. Keith, M. A. Al-Laham, C. Y.

Peng, A. Nanayakkara, M. Challacombe, P. M. W. Gill, B. Johnson, W. Chen, M. W. Wong, J.

25
L. Andres, C. Gonzalez, M. Head-Gordon, E. S. Replogle, and J. A. Pople, Gaussian 98,

Revision A.9, Gaussian, Inc., Pittsburgh PA, 1998.

13 Kawata, M.; Ten-no, S.; Kato, S.; Hirata, F. Chem. Phys., 1996, 203, 53.

14 Tunon, I.; Silla, E. Tomasi, J. J. Phys. Chem., 1992, 96, 9043.

15 Liptak, M. D.; Gross, K. C.; Seybold, P. G.; Feldgus, S.; Shields, G. C. J. Am. Chem. Soc.

2002, 124, 6421.

16 Chandler, D.; Andersen, H. C. J. Chem. Phys. 1972, 57, 1930.

17 Sato, H.; Hirata, F.; Kato, S. J. Chem. Phys. 1996, 105, 1546.

18 Singer, S. J.; Chandler, D. Mol. Phys. 1985, 55, 621.

19 Hirata, F.; Rossky, P. J.; Pettitt, B. M. J. Chem. Phys. 1983, 78, 4133.

20 Rizzo, R. C.; Jorgensen, W. L. J. Am. Chem. Soc. 1999, 121, 4827.

21 Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. J. Am. Chem. Soc. 1996 , 118, 11225.

22 Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein, M. L. J. Chem.

Phys. 1983 79, 926.

23 Kovalenko, A. Ten-No, F.; Hirata, F. J. Comp. Chem. 1999, 20, 928.

24 Hansen, J. P.; McDonald, I. R. Theory of Simple Liquids, AP Academic Press, 1986.

25 Kovalenko, A.; Hirata, F. J. Chem. Phys. 2000, 113, 2793.

26 Jorgensen, W.L. BOSS version 4.3, Yale University, New Haven, CT(1989a)

27 Jorgensen, W. L.; Ravimohan, C. J. Chem. Phys. 1985, 83, 3050.

28 Wiberg, K. B.; Clifford, S.; Jorgensen, W. L.; Frisch, M. J. J. Phys. Chem. A 2000, 104, 7625.

26
29 Singh, U. C.; Kollman, P. A. J. Comp. Chem. 1984, 5, 129.

30 Chirlian, L. E.; Franckl, M. M. J. Comp Chem. 1987, 8, 894.

31 Breneman, C. M.; Wiberg, K. B. J. Comp. Chem. 1990, 11, 361.

32 Sigfridsson, E.; Ryde, U. J. Comp. Chem. 1998, 19, 377.

33 Franckl, M. M.; Chirlian, L. In Reviews in Computational Chemistry Lipkowitz, K. B.; Boyd,

D. B.; Ed.; Volume 14, Wiley-VCH, 2000, 1.

34 Besler, B. H.; Merz, K. M.; Kollman, P. A. J. Comp. Chem. 1990, 11, 431.

35 Cornell, W. L.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Merz, K. M.; Ferguson, D. M.;

Spellmeyer, D. C.; Fox, T.; Caldwell, J. W.; Kollman, P. A. J. Am. Chem. Soc. 1995, 117,

5179.

36 Li, J.; Zhu, T.; Cramer, C. J.; Truhlar, D. G. J. Phys. Chem. A 1998, 102, 1820.

37 Hawkins, G. D.; Zhu, T.; Li, J.; Chambers, C. C.; Giesen, D. J.; Liotard, D. A.; Cramer, C. J.;

Truhlar, D. G. Universal Solvation Models in Combined Quantum Mechanical and Molecular

Mechanical Methods, Gao, J.; Thompson, M. A. Eds. American Chemical Society:

Washington DC, 1998, 201.

38 Xidos, J.D.; Li, J.; Zhu, T.; Hawkins, G. D.; Thompson, J. D.; Chuang, Y.-Y.; Fast, P. L.;

Liotard, D. A.; Rinaldi, D.; Cramer, C. J.; Truhlar, D. G. Gamesol-version 3.1, University of

Minnesota, Minneapolis 2002, based on the General Atomic and Molecular Electronic

Structure System (GAMESS) as described in Schmidt, M. W.; Baldridge, K. K.; Boatz, J. A.;

Elbert, S. T.; Gordon, M. S.; Jensen, J. H.; Koseki, S.; Matsunaga, N.; Nguyen, K. A.; Su, S.

J.; Windus, T. L.; Dupuis, M.; Montgomery, J. A. J. Comp. Chem. 1993, 14, 1347.

39 Hunter, E. P.; Lias, S.G.; Proton Affinity Evaluation, National Institute of Standards and

Technology, Gaithersburg, MD, 2003, http://webbook.nist.gov.


27
40 Perrin, D. D. Dissociation Constants of Organic Bases in Aqueous Solution. Butterworths,

London, 1965; Supplement, 1972.

41 Bayly, C. I.; Cieplak, P.; Cornell, W. D.; Kollman, P. A. J. Phys. Chem. 1993, 97, 10269.

42 Jakalian, A.; Jack, D. B.; Bayly, C. I. J. Comp. Chem. 2002, 23, 1623.

43 Sato, H.; Kovalenko, A.; Hirata, F. J. Chem. Phys. 2000, 112, 9463.

44 Meng, E. C.; Cieplak, P.; Caldwell, J. W.; Kollman, P. A. J. Am. Chem. Soc.1994, 116, 12061.

45 Gao, J.; Luque, F. J.; Orozco, M. J. Chem. Phys. 1993, 98, 2975.

28
Appendix 1

OPLS Lennard Jones Parameters

[Å] [kcal/mol] electron units

N-Amine 3.3 0.17


N-Ammonia 3.42 0.17
H(N)-Amine 0.0 (1.0)a 0.0 (0.056) a
C-Carbon 3.5 0.066
H(C(N))-Hydrogen on carbon bonded to amine 2.5 0.015
H(C) Alkane 2.5 0.030
O-Alcohol 3.12 0.17
H(O) Alcohol 0.0 (1.0) a 0.0 (0.056) a
O-Morpholine 2.9 0.14
O-TIP3P 3.15061 0.1521 -0.834
a a
H-TIP3P 0.0 (1.0) 0.0 (0.015) 0.417
O-TIP4P 3.15365 0.155 0.0
H-TIP4P 0.0 0.0 0.52
M-TIP4P 0.0 0.0 -1.040
a
Parameters utilized in RISM-SCF calculations that differ from the standard values are given in
parenthesis.

29
Appendix 2

Partial Atomic Charges of Alkane and Oxygen/Alcohol in Neutral Amines

Molecule sitea MK* MK* MK CM2


b c
gas phase solution gas phase solution
d d d d d d d d
C/O H C/O H C/O H C/O H

Methylamine-C 0.58 -0.09 0.69 -0.10 0.38 -0.04 -0.35 0.15


Ethylamine-C1 0.58 -0.07 0.69 -0.07 0.54 -0.06 -0.22 0.14
Ethylamine-C2 -0.27 0.07 -0.36 0.08 -0.33 0.08 -0.43 0.15
Dimethylamine-C 0.23 -0.02 0.26 0.00 0.05 0.04 -0.10 0.08
Trimethylamine-C 0.07 0.03 0.07 0.04 -0.29 0.12 -0.10 0.08
Piperidine-C1,C5 0.24 0.01 0.27 0.04 0.20 0.01 -0.01 0.09
Piperidine-C2, C4 -0.12 0.03 -0.23 0.07 -0.13 0.01 -0.16 0.09
Piperidine-C3 0.12 -0.01 0.13 0.00 0.11 0.02 -0.16 0.09
Piperazine-C2, C5 0.21 -0.02 0.23 0.00 0.00 0.06 -0.19 0.08
Piperazine-C3, C4 0.27 -0.03 0.35 -0.02 0.12 0.06 -0.20 0.08
Morpholine-C2 0.21 0.00 0.25 0.04 0.12 0.04 -0.02 0.08
Morpholine-C5 0.27 -0.01 0.33 0.02 0.20 0.02 -0.02 0.08
Morpholine- C3, C4 0.10 0.04 0.07 0.07 0.06 0.07 0.00 0.08
Morpholine-O -0.41 -0.52 -0.40 -0.33
Pyrrolidine-C1 0.26 -0.03 0.25 0.00 0.16 0.01 -0.20 0.14
Pyrrolidine-C4 0.26 -0.03 0.25 0.00 0.22 0.00 -0.20 0.14
Pyrrolidine-C2 0.03 -0.01 0.01 0.00 -0.04 0.03 -0.28 0.15
Pyrrolidine-C3 0.03 -0.01 0.01 0.00 -0.08 0.03 -0.28 0.15
TMP-C1,C5 0.88 1.15 0.82 0.13
TMP-C2,C4 -0.37 0.06 -0.39 0.05 -0.44 0.11 -0.16 0.08
TMP-C3 0.69 -0.13 0.72 -0.11 0.59 0.02 0.07 0.08
TMP-CH3 -0.41 0.08 -0.42 0.07 -0.53 0.11 -0.25 0.08
TMP-O -0.70 0.42 -0.90 0.51 -0.75 0.43 -0.49 0.27
a
For symmetric sites average charges are given except when they diverge significantly. The carbon
given the number 1 is bonded to the amine, the carbon bonded to C1 is given the number C2 and so on.
b c d
RISM-SCF method. SM 5.42R solvation model. Average value in case of more than atom.

30
Partial Atomic Charges of Alkane and Oxygen/Alcohol in Protonated Amines

Moleculea MK* MK* MK CM2


b c
gas phase solution gas phase solution
d d d d d d d d
C/O H C/O H C/O H C/O H

Methylamine-C -0.14 0.14 -0.13 0.13 -0.02 0.11 -0.09 0.14


Ethylamine-C1 0.42 0.06 0.57 0.02 -0.22 0.26 -0.19 0.20
Ethylamine-C2 -0.84 0.26 -1.07 0.30 -0.52 0.22 -0.43 0.17
Dimethylamine-C -0.23 0.15 -0.23 0.14 -0.37 0.26 -0.09 0.20
Trimethylamine-C -0.32 0.17 -0.33 0.16 -0.35 0.25 -0.10 0.13
Piperidine-C1,C5 0.04 0.08 0.06 0.08 0.01 0.10 -0.01 0.11
Piperidine-C2, C4 -0.03 0.05 -0.03 0.04 -0.08 0.07 -0.01 0.11
Piperidine-C3 0.01 0.03 0.02 0.02 -0.04 0.06 -0.01 0.13
Piperazine-C2, C5 -0.12 0.13 -0.14 0.14 -0.13 0.14 -0.16 0.16
Piperazine-C3, C4 0.22 0.05 0.28 0.06 0.19 0.07 -0.15 0.15
Morpholine-C2 -0.07 0.12 -0.01 0.11 0.12 0.14 -0.02 0.14
Morpholine-C5 -0.07 0.12 -0.01 0.11 -0.11 0.10 -0.02 0.14
Morpholine-C3 0.19 0.07 0.22 0.07 -0.08 0.13 -0.01 0.10
Morpholine-C4 0.19 0.07 0.22 0.07 0.14 0.10 -0.01 0.10
Morpholine-O -0.41 -0.53 -0.41 -0.33
Pyrrolidine-C1,C4 0.06 0.08 0.09 0.08 0.08 0.08 -0.17 0.26
Pyrrolidine-C2,C3 -0.04 0.07 -0.07 0.06 -0.03 0.06 -0.27 0.23
TMP-C1,C5 0.45 0.61 0.66 0.14
TMP-C2,C4 -0.34 0.11 -0.39 0.11 -0.50 0.16 -0.16 0.10
TMP-C3 0.50 -0.04 0.61 -0.04 0.58 0.06 0.06 0.09
TMP-CH3 -0.44 0.13 -0.50 0.13 -0.56 0.16 -0.26 0.11
TMP-O -0.69 0.44 -0.85 0.51 -0.71 0.45 -0.48 0.35
a
For symmetric sites average charges are given except when they diverge significantly. The carbon
given the number 1 is bonded to the amine, the carbon bonded to C1 is given the number C2 and so on.
b
RISM-SCF method. c SM 5.42R solvation model. d Average value in case of more than atom.

31
Appendix 3

RISM-SCF Relative Solvation Energy Contributions. Data in [kcal/mol].

excess chemical solute polarization


potential energy
Ammonia 0.00 0.00

Methylamine -8.46 -0.21

Ethylamine -7.94 -1.38

Dimethylamine -13.14 -0.85

Trimethylamine -15.53 -1.82

Piperidine -25.84 2.19

Piperazine -17.90 -1.39

Morpholine -19.29 -0.04

Pyrrolidine -17.65 -0.84

TMP -35.52 2.30

32

You might also like