Chapter 3 - Basic properties of U, S, and their differentials

S
1. Energy minimum principle
 
S can be written as S(U, X ) , where X is a const. U
vector of all independent internal extensive
variable (e.g. all but one Uk, and all other Xi). U

Because S is monotonic  in U and continuous,

we can invert to U(S, X ) . This relation is fully
equivalent to the fundamental  relation.

Because of the shape of S(U, X ) or U(S, X ) , as
shown in the figure, maximizing the entropy at
constant U is equivalent to minimizing the Xik
energy at constant S. S

Postulate 2’: The internal energy of a

composite system at constant S is minimized at
equilibrium. U

This is the familiar version from mechanics, const. S

where system properties are usually formulated
in terms of energies, instead of entropies.

Xik
2. Intensive parameters: temperature

Working for now with U for a simple system, U(S, X ) we can write
⎛ ∂U ⎞ ⎛ ∂U ⎞  ∂U
dU = ⎜ ⎟ dS + ⎜  ⎟ • dX (with appropriate constraints on each derivative)
⎝ ∂ S ⎠ X ⎝ ∂ X ⎠ X ∂ Xi
  ⎛ ∂U ⎞
= TdS + I • dX where T ≡ ⎜ > 0 (by P3)
⎝ ∂ S ⎟⎠ X

By construction, T and the Is are intensive variables. For example,

⎛ ∂λU ⎞
U → λU and S → λS ⇒ T → ⎜ ⎟=T
⎝ ∂λS ⎠
We consider in detail the properties of the energy  derivative T, and then briefly by
analogy other intensive variables Ii. Let all the dX (such as dV, dni, dM, etc.) equal to
zero: no mechanical macroscopic variables are being altered except for energy. It then
follows that dU = d q because d w = 0. Therefore
d q = TdS
for small (quasistatic) charges in heat, the change in system entropy is linearly
proportional to the heat increment. Thus as we add energy to the uncontrollable degrees
of freedom of our system, entropy increases, in accord with the notion that entropy is
disorder. Furthermore, we can rewrite this as
1
dS = dq .
T
When T is larger, the entropy increases less for a given heat input.
What is this quantity T? Consider a closed composite system {S} of two subsystems
{S1} and {S2} separated by a diathermal wall. A diathermal wall allows only heat flow,
so dX = 0 again. At equilibrium,
⎛ ∂S ⎞ ⎛ ∂S ⎞
dS = 0 = ⎜ 1 ⎟ dU1 + ⎜ 2 ⎟ dU 2 according to P2,
⎝ ∂U1 ⎠ X ⎝ ∂U 2 ⎠ X
or
1 1
dS = dU1 + dU 2 .
T1 T2
But dU = 0 for a closed system by P1, from which follows that
⎛1 1⎞
dU2 = -dU1 or dS = ⎜ − ⎟dU1 .
⎝ T1 T2 ⎠
At equilibrium, dS = 0 for any variation of dU1, which can only be true if
⎛1 1⎞
⎜ − ⎟ = 0 ⇒ T1 = T2
⎝ T1 T2 ⎠

Thus, T is the quantity that is equalized between two subsystems when heat is allowed to
flow between them.

This is the most straightforward definition of temperature: the thing that becomes equal
when heat stops flowing from one place to another. We can thus identify the intensive
variable as the temperature of the system. Temperature is always guaranteed to be
positive by P3 because entropy is a monotonically increasing function of energy.
Finally, if T = (∂U/∂S)X, we can rewrite the third postulate as
lim S = 0 ,
T →0
more commonly known as the “third law of thermodynamics.” As all the energy is
removed from a system by lowering its temperature, the system becomes completely
ordered. It is worth noting that there are systems (glasses), where reaching this limit
takes an inordinate amount of time. A very general principle of quantum mechanics
guarantees that the third law holds even in those cases, if we can actually get the system
to equilibrium: a coordinate or spin Hamiltonian always has a single groundstate of A1
symmetry. This is the state any system reaches as T → 0. In practice, this state may just
not be reachable even approximately in glasses, and heuristic replacements of the third
law have been developed for this case, which is really a non-equilibrium case.

To summarize:
ΔSclosed > 0 always by postulate P2
 dq
ds = by P2 for a quasistatic process when no work is done
T
T >0 always by postulate P3
T1 = T2 for two systems in thermal equilibrium
lim S = 0 always by P3, difficult to reach even approximately in some cases
T →0

Thus T and S have all the intuitive characteristics of temperature and disorder, and we can
take them as representing temperature and disorder. The latter can be justified even more
deeply by making use of statistical mechanics in later chapters, where the second
postulate follows from microscopic properties of the system.
A note on units: T.S must have units of energy. It would be convenient to let T have
units of energy (as an “energy per unit size of the system”) and to let S be unitless, but for
historical reasons, T has arbitrary units of Kelvin and S has units of Joules/Kelvin to
compensate.

3. Other extensive-intensive variable pairs

The more complex a composite system becomes, the more extensive variables it requires
beyond U, leading to additional intensive variables. For example:

⎛ ∂U ⎞
i) V (volume) leads to an energy change dU v = ⎜ dV ≡ −PdV . The intensive
⎝ ∂V ⎟⎠ X
derivative is called the pressure of the system. PV has units of Joules, so P must
have units of Joules/m3 or N/m2. Thus P certainly has the units we normally
associate with pressure, or force per unit area. Usually ∂U / ∂V <0 because
squeezing a system increases its energy. Thus P is generally a positive quantity,
again in accord with our intuition. Note however that there is no postulate that
says P must be positive. In fact, we can bring systems to negative pressure by
pulling on the system, or putting tension on it.

Is P is in fact pressure? It is easy to see that it is, by applying the minimum

energy principle to a diathermal flexible wall, in analogy to what was done for
temperature above:
dU = 0 = dU1 + dU 2 by P1
= T1dS1 − P1dV1 + T2 dS2 − P2 dV2 by P2'
= (T1 − T2 )dS1 − (P1 − P2 )dV1
In the third line, we assume a closed system and reversible process, so dV = 0 and
dS = 0. When the energy has reached equilibrium, the equation must hold for any
small perturbation of the entropy or volume of subsystem 1, which can only be
satisfied if T1 = T 2 (again), and P1 = P2.

Thus, P is the quantity which is the same in two subsystems when they are
connected by a flexible wall.
 This is the most straightforward definition of pressure: the thing that is equalized
between two systems when the volume can change to whatever it wants. P is a
pressure not just in units, but agrees with our intuitive notion of what a pressure
should be.

⎛ ∂U ⎞
ii) A (area in surface system) => dU A = ⎜ ⎟ dA ≡ −ΓdA
⎝ ∂A ⎠ X '
Where Γ has units of (N/m) and is therefore the surface tension.

⎛ ∂U ⎞
iii) M (magnetization) => dU M = ⎜ ⎟ dM ≡ HdM where H is the magnetic field.
⎝ ∂M ⎠ X '

⎛ ∂U ⎞
iv) ni (mole number) => dU n i = ⎜ ⎟ dn1 ≡ µi dn i where µi is the chemical
⎝ ∂n i ⎠ X '
potential equalized when particles are allowed to flow.

⎛ ∂U ⎞
v) L (length) => dU L = ⎜ ⎟ dL ≡ FdL where F is the linear tension force.
⎝ ∂L ⎠ X '

Many more conjugate pars of extensive and intensive variables are possible, but this
gives the general picture. For an arbitrary variation
 in U we have
 dU = TdS + I • dX ,
where I is the vector of all intensive variables except temperature. Often, we will use
dU = TdS − PdV + µdn
as an example, when dealing with a simple 3-dimensional 1-component system.

4. First order homogeneity

 
Consider S for a closed system. Because S is extensive, S( λU, λX ) = λS(U, X ) . This
agrees with the intuitive notion that 2 identical disordered
 systems
 amount to twice as
much disorder as a single one. Similarly, U( λS, λX ) = λU(S, X ) . Differentiating both
sides with respect to λ yields

⎛ ∂U ⎞ ⎛ ∂λ S ⎞ ⎛ ∂U ⎞ ⎛ ∂λ X ⎞  ⎛ ∂U ⎞ ⎛ ∂U ⎞  
⎜⎝ ⎟⎠  ⎜⎝ ⎟⎠ + ⎜⎝ ⎟ •⎜ ⎟ = U(S, X) or ⎜ ⎟ S+⎜  ⎟ • X = U(S, X)
∂λ S X ' ∂λ ∂λ X ⎠ X ' ⎝ ∂λ ⎠ ⎝ ∂λ S ⎠ X ' ⎝ ∂λ X ⎠ X '

When λ = 1, this yields

⎛ ∂U ⎞ ⎛ ∂U ⎞   
⎜⎝ ⎟⎠  S + ⎜⎝  ⎟⎠  • X = U or U =TS +I • X .
∂S X ' ∂X X'
Thus the energy has a surprisingly simple form: it is simply a bilinear function of the
intensive and extensive parameters. It is known as the Euler form. The formula for
energy looks like the formula for d U with the dS removed. For example,
U = TS − PV + µn for a simple one-component system.
 Solving for S yields an analogous formula in the entropy representation,

⎛1⎞ ⎛I⎞  1 ⎛P⎞ µ
S = ⎜ ⎟U − ⎜ ⎟ • X (e.g. S = U + ⎜ ⎟V − n)
⎝T ⎠ ⎝T⎠ T ⎝T ⎠ T
The entropy is also a simple bilinear functions of its intensive and extensive parameters.

5. Gibbs-Duhem relation

The differential of U combined with first order homogeneity requires that not all
intensive parameters be independent. For a completely arbitrary variation of U,
   
dU = TdS + SdT + I • dX + X • dI
But we know from earlier that
   
dU = TdS + I • dX ⇒ SdT + X • dI = 0
Using this Gibbs-Duhem relation, one intensive parameter can be expressed in terms of
the others.
For example, consider a simple multicomponent system:
r r
U = TS − PV + ∑µ i n i ⇒ SdT − VdP + ∑ n i dµ i
i=1 i=1

⎛V ⎞ ⎛S⎞ r
n
⇒ dµ1 = ⎜ ⎟dP − ⎜ ⎟ dT − ∑ i dµ i
⎝ n1 ⎠ ⎝ n1 ⎠ i= 2 n1

One chemical potential change can be expressed in terms of pressure, temperature, and
the other chemical potentials. In general, an r-component simple 3-D system has only 2
+ (r-1) = r+1 degrees of freedom. This will be useful for multi-phase systems. For
example, let two phases of the same substance be at equilibrium, and particle flow is
allowed from one phase to another. Then µ1 = µ2 (or particles would flow to the phase of
lower chemical potential according to 3.), and to remain at equilibrium when the
chemical potential changes, dµ1 = dµ2 . Combining the Gibbs-Duhem relations for each
phase,
S1dT − V1dP = −dµ1
1 = dµ 2
dP ΔS12
⎯dµ
⎯⎯⎯ → (S1 − S2 )dT = (V1 − V2 )dP or = ,
S2 dT − V2 dP = −dµ 2 dT ΔV12

Thus letting dµ1 = dµ2 traces out the T, P conditions where the two phases are at
equilibrium. This is known as the Clausius equation.

6. Equation of state and fundamental relation

 
Often we do not know the fundamental equation U(S, X ) or S(U, X ); instead we know
equation involving intensive variables, known as equations of state. For example,
  ⎛ ∂U ⎞ 
U = U(S, X) ⇒ T = ⎜ ⎟ = T (S, X).
⎝ ∂ S ⎠ X

Similarly, the derivative with respect to any other X yields the corresponding equation of
state I(S, X). These are called equations of state in normal form, and express one
intensive variable in terms of all the extensive variables. There are as many equations of
state as there are extensive variables for the system (e.g. r+2 for a simple r-component
system). Note that an equation of state does not contain the same amount of information
as the original fundamental relation; it can be integrated up to a constant that depends on
all extensive variables except the one involved in the derivative, but that part of U (or S,
if we derive equations of state from S(U, X)) cannot simply be left out.
If all the equations of state in normal form are known, we can reconstruct the
fundamental relation by using the Euler form from 4
   
U = T(S, X )S + I (S, X ) • X (also solvable for S because of P3)
If they are not known in normal form, we may also be able to obtain the fundamental
relation by integrating a differential form, such as

⎛ 1⎞ ⎛I⎞ 
dS = ⎜ ⎟ dU − ⎜ ⎟ • dX .
⎝T ⎠ ⎝T ⎠
If needed, we can compute one intensive variable from the Gibbs-Durhem relation, so
we need one less equation of state (only r+1 for a simple r-component system) to evaluate
the fundamental relation. Finally, equations of state may also be substituted into one
another, yielding equations that depend on more than one intensive variable. These are
also referred to as equations of state, but they are not in normal form.
Let us consider two examples of how to determine a fundamental relation. We start
with the fundamental relation for a rubber band, where we can write down reasonable
guesses for both equations of state needed.
1 F
dU = TdS + FdL ⇒ dS = dU − dL
T T
We need equations of state so T and F can be eliminated to yield S(U,L):
a) F=c1T(L-L0); L0 is the relaxed length of the rubber band, and we are treating it
like a linear spring once stretched. An unusual feature is that F increases with T.
At higher T polymer chains wrinkle into more random coils, causing shrinkage,
and increasing the tension for the same length.

b) U=c2L0T, as long as F depends only linearly on T => F/T=F(L) only. The reason
∂ 2 S(U,L) ∂ ⎛ −F ⎞ ∂ ⎛ 1 ⎞ 1
is that = ⎜ ⎟ = ⎜ ⎟ = 0 in that case. So can be any single-
∂U∂L ∂U ⎝ T ⎠ ∂ L ⎝ T ⎠ T
valued function of U as long as it is independent of L; for simplicity we pick
U~T, as for an ideal gas.
We can now insert the two equations of state into the differential form, and integrate it
 c2 L0
dS = dU − c1 (L − L0 )dL ⇒
U
U c1
S = S0 + c2 L0 ln − (L − L0 )2
U0 2
The constant can be determined by invoking the third law. However, note that this can
lead to singularities if the equations of state themselves are not correct at low
temperature, as is the case in this example. Moreover, note that c2 most be intensive, and
c1-1 must be extensive so that S is extensive. From the fundamental relation we can
calculate any desired properties of the rubber band.
Alternatively, we could try to obtain the fundamental relation in terms of U = TS +
1 F
FL, but then we would need T(S,L) and F(S,L) instead of (U, L) and (U, L) , which
T T
were not available. Similarly, to plug into S = U/T – FL/T, we would need T(U, L) and
F(U, L); we have the former, but not the latter: the equation of state for F=c1T(L-L0) is in
terms of another intensive variables, and not in the basic form required for the Euler
form. Note that plugging T = U/c2L0 into F(T, L) to get a F(U, L) will not help either
because this does not yield an equation of state in normal form as it would have been
obtained by taking the derivative of S.
As another example, consider the fundamental relation for an ideal monatomic gas.
In this case, we will derive one of the equations of state from the others, get all three
equations of state in normal form, before inserting all three to obtain the fundamental
relation. The gas has 1 component, so we need r+1=2 equations of state to get started:
1) Pν = RT
3
2) u = RT
2
Here the two well-known equations of state for an ideal gas are written in terms of
intensive variables u = U/n and V = V /n. Again the first equation depends on two
intensive variables and is not in standard form. We can bring both equations into
standard form as follows:
R R ⎛ 2u ⎞ 2 u ⎛ ∂u ⎞
1') P = T = ⎜ ⎟ = = −⎜ ⎟
ν ν ⎝ 3R ⎠ 3 ν ⎝ ∂ν ⎠ S,n
2u ⎛ ∂ u ⎞
2 ') T = =⎜ ⎟
3R ⎝ ∂ S ⎠ V ,n
We now need µ(u,ν) as the third equation of state. Proceeding with the Gibbs-Duhem
relation,
dµ = −SdT + ν dP .
We must eliminate S since we P and T as a function of U, not S. Using the bilinear form
of S,
⎛ u Pν µ ⎞
dµ = − ⎜ + − dT + ν dP .
⎝ T T T ⎟⎠
Next we eliminate P and T by using equations 1’ and 2’:
2 du 2 2 dν
dµ = −dµ − du + µ + du − u .
3 u 3 3 ν
We then divide by u on both sides, rearrange, and integrate:
dµ du ⎛ µ⎞ du 2 dν
− µ 2 = d⎜ ⎟ = − −
u u ⎝ u⎠ u 3 ν
u 2 ν
final
⎛ µ⎞ ⎛ µ⎞ ⎛ µ⎞
∫
0
d ⎜ ⎟ = ⎜ ⎟ − ⎜ ⎟ = − ln − ln
⎝ u⎠ ⎝ u⎠ ⎝ u⎠0 u0 3 ν 0
or

u 2 ν ⎛ µ⎞
µ = −u ln − u ln + u ⎜ ⎟
u0 3 ν0 ⎝ u⎠0
This is the third equation of state, for the chemical potential. We now have all intensive
parameters as normal form equations of state, to construct the fundamental relations
s(u,ν) or u(s, v). (Of course, the homogeneous first order property means that to get S
and U, we just multiply by n.) Doing s, for example,

1 P µ
s= u+ ν−
T T T
3R 2u 3R 3R ⎧ u 2 ν ⎛ µ⎞ ⎫
= u+ ν− u ⎨− ln − ln + ⎜ ⎟ ⎬
2u 3ν 2u 2u ⎩ u0 3 ν 0 ⎝ u ⎠ 0 ⎭
5 3 ⎛ µ⎞ 3 u ν
= R − R ⎜ ⎟ + R ln + R ln
2 2 ⎝ u⎠0 2 u0 ν0
3
= R ln u + R ln ν + c
2

⎛ ∂U ⎞ 2u
Note that this equation of state violates P3: ⎜ ⎟ = T = so T → 0 ≡ˆ u → 0 ; but
⎝ ∂S ⎠V 3R
3
S→ / 0 as u → 0 , it approaches −∞ . Thus, either PV=nRT, or U = nRT , or both must
2
be high-temperature approximations that fail as T→ 0. At low T, excluded volume
effects, particle interaction, and quantum effects come into play. The ideal gas equation
would have to be replaced by a more accurate equation, such as the van der Waals
equation to satisfy the third law closer to T = 0. In that sense, thermodynamics can point
out to us when approximate equations of state break down.

7. Stability and second derivatives

The first derivatives (intensive parameters) are very useful because they correspond to
quantities that are equalized among equilibrated subsystems. However, the first order
relationship dS=0, although necessary by P2 at equilibrium, is not sufficient. The
extremum in S must be a maximum:
d 2 S < 0 (or d 2U > 0 according to 1.)
Extrema with d 2 S > 0 or d 2 S = 0 are also possible (minima, saddles, degenerate points).
However, thermodynamics cannot make statements about such points without some
further assumptions that go beyond the postulates. This suggests that the study of second
derivative will be fruitful, to ensure that one is working near a stable equilibrium point.
Three of these second derivatives encountered later are
1 ⎛ ∂V ⎞
α= ⎜ ⎟
V ⎝ ∂T ⎠ P,ni
1 ⎛ ∂V ⎞
κ =− ⎜ ⎟
V ⎝ ∂P ⎠ T
T ⎛ ∂S ⎞ ⎛ dq ⎞
cp = ⎜⎝ ⎟⎠ = ⎜⎝ ⎟
n ∂T P dT ⎠ P
For a simple system, only three second derivatives are linearly independent if we exclude
ones based on ∂ / ∂ni . The reason is that the terms in the energy U = TS – PV + … have
only three second derivatives,
⎛ ∂T ⎞ ∂ 2U
1. ⎜ ⎟ = 2
⎝ ∂S ⎠V ∂S
⎛ ∂P ⎞ ∂ 2U
2. ⎜ ⎟ =
⎝ ∂V ⎠ S ∂V 2
⎛ ∂T ⎞ ⎛ ∂P ⎞ ∂ 2U
3. but ⎜ ⎟ = −⎜ ⎟ =
⎝ ∂V ⎠ S ⎝ ∂S ⎠V ∂V∂S
or dµ is not a perfect differential. Rather than picking those three, we will usually work
with the first independent set, corresponding to quantities with more obvious physical
interpretations to chemists working at constant pressure and temperature. We consider
the corresponding fundamental relations in the next chapter.