ARTICLES

PUBLISHED ONLINE: 23 SEPTEMBER 2013 | DOI: 10.1038/NCHEM.1754

Caesium in high oxidation states and as a p -block element

Mao-sheng Miao*
The periodicity of the elements and the non-reactivity of the inner-shell electrons are two related principles of chemistry, rooted in the atomic shell structure. Within compounds, Group I elements, for example, invariably assume the 11 oxidation state, and their chemical properties differ completely from those of the p-block elements. These general rules govern our understanding of chemical structures and reactions. Here, rst-principles calculations show that, under pressure, caesium atoms can share their 5p electrons to become formally oxidized beyond the 11 state. In the presence of uorine and under pressure, the formation of CsFn (n > 1) compounds containing neutral or ionic molecules is predicted. Their geometry and bonding resemble that of isoelectronic XeFn molecules, showing a caesium atom that behaves chemically like a p-block element under these conditions. The calculated stability of the CsFn compounds shows that the inner-shell electrons can become the main components of chemical bonds.

ne of the important discoveries in chemistry of the last century was the understanding of the atomic shell structure and, in turn, of the periodic properties of elements at the level of quantum mechanics1,2. All elements, except hydrogen and helium, possess completely lled inner shells, and all but the noble gases possess a partially lled outer shell. The chemical properties of the atoms are determined by the electrons in the outermost shellthe valence electrons. The electrons in the inner shells of elementscore electronsare not involved in forming interatomic bonds. Based on the shell model, atoms form chemical bonds with other atoms by losing, gaining or sharing valence electrons, while all core electrons remain unreactive2,3. The group I elements, for example, are invariably stable in a 1 charge state and usually form ionic compounds with other elements. Their chemical properties are thus very different from those of p-block elements, which often form covalent bonds with oxidizing agents such as uorine. The reactivity of a complete outmost shell was rst demonstrated for noble gas elements by Neil Bartlett, who discovered in 1962 that xenon can react with uorine to form Xe[PtF6]2 (ref. 4). Since then, many noble gas compounds have been found514. Similar to many other p-block elements, xenon covalently bonds with uorine to form molecules or molecular crystals. Under high pressure, XeF2 molecular crystals may lose their molecular character to become, instead, extended solids, in which atoms are bonded by network of covalent bonds15. In contrast to the outmost shell of noble gases, the non-reactivity of core electrons has not yet been successfully challenged, although several attempts have been made under ambient pressures to oxidize Cs beyond the 1 state1619. It is interesting to note that, under pressure, the 5p level in Cs solid is broadened and increases in energy, and may hybridize with the states around the Fermi level2022. More recently, a computational study on Cs polyhydrides revealed a large amount of Cs 5p character in the states around the Fermi level, indicating their involvement in CsH bonding23,24. These previous experimental and computational studies suggest that high pressure can signicantly raise the 5p core levels of Cs. It is therefore conceivable that, should the pressure be high

enough, the 5p states will be readily available for oxidation by a strong oxidizing agent. In this work, computations were used to demonstrate that the 5p electrons in the inner shell of a Cs atom can become reactive under high pressure. As a result, Cs atoms can be oxidized beyond the 1 oxidation state to form a series of uorides CsFn (n 2 2 5) under pressures ranging from 10 to 200 GPa. This approach is based on accurate rst-principles calculations, which have been successfully used in numerous predictions regarding novel compounds and structures over the past few decades2530. A non-biased automatic structure search method based on the particle swarm optimization algorithm is used to search for stable CsFn structures under pressure, across the entire potential energy surface derived from density functional theory (DFT) calculations31,32. Recently, this method has been successfully applied to predict the structures of many systems29,30,33. The calculated structures are compared under a series of pressures (0, 10, 50, 100, 150 and 200 GPa), and the most stable structure is identied for each composition.

Results and discussions

For the most stable structures at each pressure, the enthalpy of formation per atom is calculated using the following formula: hf (CsFn ) = [H (CsFn ) H (CsF) (n 1)H (F2 )/2]/(n + 1) where hf is the enthalpy of formation per atom and H is the calculated enthalpy per chemical unit for each compound33,34. We chose to use the enthalpy of CsF instead of that of Cs (see the above formula) because of the substantial stability of CsF throughout the studied pressure range. Thus, changing Cs to CsF does not change the measurement of the stability of compounds with higher F compositions, although the absolute value of hf does change33. Reading the solid lines in Fig. 1a, the compounds located on the convex hull are stable against decomposition. This is because, for these compounds, decomposing into any other compositions will cost energy34. Stability. Figure 1b shows that, for n . 1, all CsFn compounds are unstable at ambient pressure, which is consistent with the

Materials Research Laboratory, University of California, Santa Barbara, California 93106-5050, USA and Beijing Computational Science Research Center, Beijing 10084, China. * e-mail: miaoms@gmail.com
846
NATURE CHEMISTRY | VOL 5 | OCTOBER 2013 | www.nature.com/naturechemistry

2013 Macmillan Publishers Limited. All rights reserved.

NATURE CHEMISTRY
a
0.2

DOI: 10.1038/NCHEM.1754

ARTICLES
because the ZPE is the ground-state energy of phonons and represents the quantum uctuations of nuclei (generally not signicant for Cs and F). The calculations of both the atomic and electronic structures of the CsFn compounds revealed that Cs is oxidized to a higher state than 1. While sharing its 5p electrons, Cs can covalently bond with F to form CsFn molecules. The bonding features of CsFn molecules are found to be analogous to those of isoelectronic XeFn molecules, suggesting that Cs behaves like a p-block element when being oxidized beyond the 1 state.

0.0

hf (eV per atom)

0.2

0.4

CsF2

0.6

0.8

0 GPa 10 GPa 50 GPa 100 GPa 150 GPa 200 GPa

CsF3

CsF6

CsF4 CsF5

1.0 0.5 0.6 0.7 0.8 0.9 1.0 Composition (% of F)

b
CsF5 Cs+[F3] CsF2 0 50 100 Pressure (GPa) 150

CsF6 ?

CsF3

200

Figure 1 | Stability of CsFn compounds under pressure. a, Enthalpies of formation of CsFn under a range of pressures. Dotted lines connect data points, and solid lines denote the convex hull. Compounds corresponding to data points located on the convex hull are stable against disproportionating into other compositions. b, Predicted stable pressure ranges for CsFn compounds (orange). The blue segment represents the pressure range where Cs[F3]2 is stable (see Fig. 2c for its structure), instead of CsF3. The enthalpies of CsF are calculated for the NaCl structure under ambient conditions and the CsCl structure at high pressures. The enthalpies of F solids are obtained for molecular crystals, which are in C12/m1 and C12/c1 space groups. Details of the stable structures for CsF and F solid are provided in Supplementary Section SIV.

common knowledge that Cs is only stable in the 1 state. However, CsF2 becomes stable at a pressure of 5 GPa. As the pressure increases, CsFn compounds with higher ratios of F are formed. As shown in Fig. 1b, CsF2 is only stable in a pressure range from 5 to 17 GPa. At higher pressures, it decomposes into CsF and CsF3. The CsF3 and CsF5 become stable at pressures of 15 GPa and 50 GPa, respectively, and both remain stable up to 200 GPa, the highest pressure examined in this work. At 30 GPa, CsF3 undergoes a rst-order transition from a structure containing F32 polyuoride anions to a structure that has no FF bond. In contrast, the calculations show that CsF4 and CsF6 are unstable at pressures below 200 GPa, although the trend in the convex hulls suggests that CsF6 may become stable at higher pressures (no similar trend appears for CsF4). The structures are all found to be dynamically stable within their stable pressure ranges. Selected phonon spectra and calculation details are provided in Supplementary Section SV. The effect of the zero point energy (ZPE) was found to be small, less than 2 meV/atom for all calculations of the enthalpy of formation. This is understandable,
NATURE CHEMISTRY | VOL 5 | OCTOBER 2013 | www.nature.com/naturechemistry

Crystal structures and their molecular nature. The structures calculated for each composition are shown in Fig. 2af. Compound CsF2 (Fig. 2a) is stable in a structure that has I4/mmm symmetry, similar to that of XeF2 (refs 5,7,9). An analysis of the CsF distances in the structure revealed the existence of linear CsF2 molecules (with a CsF distance of 2.358 at 20 GPa). This is the case throughout the pressure range (Fig. 2a,g). However, the CsF bond length in CsF2 is signicantly larger than the XeF bond reported in XeF2 (1.999 ), indicating a weaker CsF bond. Furthermore, the shortest FF distances in CsF2 are 2.215 at 20 GPa, much larger than the FF bond length of 1.44 , suggesting that no covalent bond forms between the F atoms. The predicted CsF3 adopts a C2/m space group throughout the pressure range investigated, and exhibits a distinctive crystal structure (Fig. 2b). Instead of forming a molecular crystal of CsF3 molecules, it consists of a [CsF2]F2 complex. At 100 GPa, the shortest CsF bond length is 2.015 , close to the XeF bond length previously observed for XeF2 (refs 57), whereas the next shortest CsF distance is 2.571 , which is much longer, suggesting that the compound shows a structure of distinct molecular character. As shown in Fig. 2h, this molecular character is kept throughout the pressure range in the study. Furthermore, the shortest FF distances are much larger than the FF bond length of 1.44 in F2 molecules (for example, 2.227 at 100 GPa). Interestingly, the CsF3 C2/m structure at 0 GPa relaxes to a structure containing linear F32 ions with a FF bond length of 1.739 , indicating that the compound is actually Cs[F3]2 (Fig. 2c). Linear [F3]2 has been found in an argon matrix, rst as an alkali metal complex35 and more recently as isolated anions36. The bond length of 1.725 found for the FF bond (ref. 36) through quantum chemistry calculations (using the B3LYP hybrid functional in DFT) is very close to that of CsF3 at 0 GPa, suggesting that the F32 ions maintain a similar geometry in the gas and solid phases. The Cs[F3]2 remains stable up to 30 GPa, where a transition to CsF3 occurs (Supplementary Section SIII). By contrast, CsF4 is unstable. Some of its structures (Fig. 2d) comprise CsF4 planar molecules that are similar to XeF4 (refs 6,9). One difference between the two, however, is that XeF4 is square planar, but CsF4 has an FCsF angle of 69.118 (100 GPa). The distortion from the square planar geometry is caused by the Jahn Teller effect, because CsF4 has one more electron than XeF4. In fact, while calculating the geometry of the XeF42 anion, we found the same distorted square planar structure. CsF5 was found to be stable in a structure with Fdd2 symmetry consisting of pentagonal planar molecules, again similar to its xenon counterpart XeF52 (Fig. 2e)37. At 150 GPa, the ve CsF bonds in the CsF5 molecules have lengths of 1.886 , 1.899 and 1.957 . The distance from Cs to the next neighbouring F atoms (apart from the ve F atoms just mentioned) is 2.367 , which means that the compound again shows a distinct molecular character. This characteristic is observed throughout the pressure range (Fig. 2i). The shortest FF distance is 2.050 , again much larger than the length of the FF covalent bond. CsF5 molecules are pentagonal planar, corresponding to a structure of type AX5E2 according to the valence-shell electron-pair repulsion (VSEPR) model
847

2013 Macmillan Publishers Limited. All rights reserved.

ARTICLES
a
CsF2, 20 GPa

NATURE CHEMISTRY
b
CsF3, 100 GPa

DOI: 10.1038/NCHEM.1754

CsF3, 0 GPa

CsF4, 100 GPa

CsF5, 150 GPa

CsF6, 200 GPa

g
4

CsF2

h
5

CsF3

i
4.0

CsF5

3.5 4 Distances () Distances () Distances () 3 CsF in CsF 3.0 CsF in CsF 2.5

3 CsF in CsF

2 FF in F2

2 FF in F2

2.0 FF in F2 0 50 100 150 200

1.5 1 30 40 50 0 50 100 150 200

10

20

Pressure (GPa) CsF1 CsF' FF CsCs

Pressure (GPa) CsF1 CsF2 CsF' FF CsCs

Pressure (GPa)

CsF1 CsF' FF CsCs FF

CsF'

CsF' CsF FF

CsF1

CsF' CsF2

CsF

Figure 2 | Crystal structures of CsFn compounds, and selected interatomic distances as functions of pressure. (Parameters of all structures are shown in Supplementary Tables SIIV). a, CsF2 at 20 GPa in an I4/mmm structure. b, CsF3 at 100 GPa in a C2/m structure. c, CsF3 at 0 GPa with actual formula of Cs[F3]2. (For structure information see Supplementary Section SIII.) d, CsF4 at 100 GPa in a C2/m structure. e, CsF5 at 150 GPa in an Fdd2 structure. f, CsF6 at 200 GPa in a P1 structure. Green and pink spheres represent Cs and F atoms, respectively; dark blue spheres in b and c represent isolated F2 ions in CsF3 at 100 GPa (b), which become the central atoms in F32 at 0 GPa (c). gi, Interatomic distances (between atoms shown in the structure directly underneath) in CsF2 (I4/mmm ), CsF3 (C2/m ) and CsF5 (Fdd2) as a function of external pressure. The CsF bond in CsF and FF in the F2 molecule are shown by dashed lines for comparison.

(A, central atom; X, sigma bonds between the central atoms and outside atoms; E, non-bonding electron pairs) (shown in Fig. 3). The structure differs from that of most other AB5 molecules (where A and B represent two different elements), such as BrF5 , which adopts a square pyramidal geometry. However, a similar pentagonal planar structure was found in [XeF5]2 (ref. 37), a species that is isoelectronic with the CsF5 molecule.
848

In its lowest energy phase, the CsF6 compound adopts a P1 symmetry. It consists of Cs atoms with 12 neighbouring F atoms, which form a cage-like structure encircling the Cs atom. These cages, in turn, stack to form a crystal (Fig. 2f ). Electronic structures and the nature of CsF bonds. Investigations into the electronic structures of CsFn (n . 1) also reveal that the 5p
NATURE CHEMISTRY | VOL 5 | OCTOBER 2013 | www.nature.com/naturechemistry

2013 Macmillan Publishers Limited. All rights reserved.

NATURE CHEMISTRY
a
PDOS 4 3 2 1 0 1 0 1 Cs 5p F 2p

DOI: 10.1038/NCHEM.1754

ARTICLES

CsF2, 20 GPa

g
CsF CsF'

COHP

0.80 0.65 Cs 5p F1 2p F2 2p

c
PDOS
CsF3, 100 GPa

4 3 2 1 0 2

0.50 0.35

COHP

CsF1 CsF2

0.20

0 2

e
PDOS
XeF2, 20 GPa

3 2 1 0 COHP 4 0 4 10 8 6 4 2 0 2 XeF1 XeF2 AX2E3 AX2E3 Xe 5p F 2p

k
F Cs F F Cs F F F Cs F F F F F Cs F AX5E2 F

AX4E2

Energy (eV)

Figure 3 | Electronic structures and the nature of CsF bonds. a,b, Calculated PDOS (a) and COHP (b) for CsF2 at 20 GPa. c,d, PDOS (c) and COHP (d) for CsF3 at 100 GPa. e,f, PDOS (e) and COHP (f ) for XeF2 at 20 GPa. The states are aligned at the Fermi level (vertical dashed lines). In all PDOS plots, green lines represent the 5p state of the Cs (Xe) atoms, and the pink lines represent the 2p state of the F atoms bonded with Cs. The overlap between the two indicates possible strong interaction between Cs and F atoms. The blue line in c represents the 2p state of the isolated F2 anion, showing a bonding feature very different to the F atoms covalently bonded with Cs. In the COHP plots, black and red lines represent the COPH between Cs (Xe) and its nearest neighbour F and those between Cs (Xe) and its next nearest neighbour F. They reveal that the Cs and its nearest neighbouring F form strong covalent bonds, whereas the covalent bonding between Cs and next nearest F is rather weak. gj, ELFs of CsF3 at 50 GPa [(100) plane] (g), CsF5 at 100 GPa [(310 3) plane] (h), XeF2 at 50 GPa [(1 0 0) plane] (i) and NH4XeF5 at 100 GPa [(0 0 4)] plane] (j). Red areas (ELF 0.8) around the Cs (Xe) and F atoms indicate the lone electron pairs, and green areas (ELF 0.5) between the Cs(Xe) and F atoms indicate strong covalent bonding. Blue areas in ELFs typically indicate non-covalent bonding, for example, delocalized electrons in metals. k, Bonding features of CsFn (n 25), given in VSEPR notation.

electrons of Cs participate in covalent bonds with neighbouring F atoms. We calculated the projected density of states (PDOS) by projecting the electron wavefunctions onto the atomic orbitals inside a sphere of selected radius. The PDOS shows a large overlap between the 5p states of Cs and the 2p states of the nearest F (Fig. 3a,c). A large overlap also exists between the Cs 5p states and the 2p states of the isolated F2 ions (next nearest F) in CsF3. There are large 5p components in the conducting states, indicating the depletion of 5p electrons in Cs atoms in these compounds. Furthermore, CsF3 and CsF5 appear to have sizeable gaps between the lled and empty states, whereas the CsF2 compound does not, which means it is metallic. Above 2 eV, the PDOS is found to be almost zero until 10 eV, where the Cs 6s and 5d states are located. Although there is a Cs 5d contribution to the states around the Fermi level, it is far less signicant than that of the Cs 5p and F 2p orbitals. To detect the covalent bonds in the compounds, the crystal orbit Hamiltonian population (COHP)38 can be used, which counts the population of wavefunctions on two atomic orbitals of a pair of selected atoms. The calculated COHP (Fig. 3b,d) shows that the overlap of the Cs 5p and F 2p states around 25.5 eV for CsF2 and around 29.0 eV for CsF3 are related to the covalent bonding between Cs and its nearest neighbouring F. Although the next
NATURE CHEMISTRY | VOL 5 | OCTOBER 2013 | www.nature.com/naturechemistry

nearest neighbouring Cs and F in CsF3 also show a large overlap between their 5p and 2p states in PDOS, the COHP reveals there is no covalent bonding between the two, so the interaction between them is mainly ionic. The integrated COHP (ICOHP) up to the Fermi level is 20.923 eV/pair for bonded CsF in CsF2 (10 GPa), 21.787 eV/pair for CsF in the CsF2 ions and 20.1473 eV/pair for Cs and the isolated F2 ions in CsF3 (100 GPa). The ICOHP of the neighbouring CsF pair in CsF (a typical ionic crystal) under 100 GPa is found to be 20.245 eV/pair. These results suggest that the CsF bonds in CsF2 and in CsF2 ions in CsF3 are very covalent, whereas the Cs ions and the isolated F2 ions in CsF3 interact only ionically. For comparison, the ICOHP is found to be 21.819 eV/pair for XeF2 (10 GPa), which is close to that of CsF in CsF2 ions and is about twice as large as that of the CsF bond in CsF2. The reduction of the ICOHP in the latter is caused by the partial occupation of the antibonding states. Both the PDOS and COHP of CsF2 and CsF3 resemble the major features of XeF2 (Fig. 3e,f ). XeF2 is stable at ambient pressure and has the same structure as CsF2 (I4/mmm). The molecular orbitals (Supplementary Section SVI), and therefore the bonding features of CsF2 , are analogous to those of XeF2 , which adopts an AX2E3 structure (Fig. 3k). These molecular orbitals show the hypervalent
849

2013 Macmillan Publishers Limited. All rights reserved.

ARTICLES
a 3.5

NATURE CHEMISTRY
b
6s 3.0

DOI: 10.1038/NCHEM.1754

0 5 10

Bader charge on cs

2.5

5p

Energy difference (eV)

15 20

2.0 5s 1.5 25 30 35 1.0 40 0.5 CsF CsF2 CsF3 CsF4 CsF5 CsF6 Xe0 Cs0 Xe+ Cs+ Xe2+ Cs2+ 45

d
6s 5p 5s 0.4 p level (eV)

10

0.8

AE, ps

0.0

F 2p 5 Xe 5p Cs 5p

0.4 10

Rb 4p K 3p

0.8 15 0 1 2 3 4 5 0 100 200 300

Ba 5p

400

500

r (Bohr)

Pressure (GPa)

Figure 4 | Mechanism of pressure-driven high oxidation states. a, Calculated Bader charge of Cs in CsFn at 100 GPa. b, The 6s, 5p and 5s energy levels of the Cs atom in neutral, 1 and 2 charge states (black bars). For comparison, the same levels for Xe0, Xe and Xe2 are shown by orange bars. c, Radial wavefunction of the Cs 6s (black), 5p (orange) and 5s (blue) states. Solid and dashed lines indicate the all-electron and pseudo-wavefunctions, respectively. It clearly shows that the Cs 5p wavefunction has large components outside the Cs radius of 1.81 . d, Energies of the outermost lled p levels of selected elements, including F, Cs, Rb, K and Ba, as a function of external pressure. The pressure effect is modelled by putting elements in a face-centred cubic (fcc) He matrix. An fcc supercell of 108 He (3 3 3) is used, in which one He is replaced by the tested atom. It shows that the Cs 5p energy becomes higher than the F 2p energy at pressures higher than 10 GPa, indicating that Cs can be oxidized by F beyond the 1 state under these conditions.

nature of the CsFn molecules, and are consistent with the Rundle Pimentel three-centre four-electron model. However, as revealed in a recent work, such hypervalent molecules present both covalent and ionic bonds, and the interaction between the two is important for their stability39. In comparison with XeF2 , CsF2 contains one extra electron. It lls the CsF pzpz s antibonding state, leading to weakened CsF bonds and a metallic character in the system. The [CsF2] molecular ions in CF3 are isoelectronic with a XeF2 molecule. An energy gap exists between the p antibonding and s antibonding states, and CsF3 maintains a strong molecular character and stability over a large pressure range. To further understand the bonding features of the CsFn compounds, we calculated the electron localization function (ELF)40
850

for CsF3 and CsF5 compounds (Fig. 3gj). The ELFs of XeF2 at 20 GPa and NH4XeF5 at 100 GPa were calculated and are shown for comparison (Fig. 3i,j). NH4XeF5 models XeF52 and has the same symmetry as synthesized N(CH3) 4XeF5 compound37. For both CsF2 and CsF5 molecules, the largest ELF values are located around the Cs and F atoms, showing the feature of lone electron pairs. The ELF values at areas between nearest neighbouring Cs and F atoms are also large, showing a covalent bonding character. On the other hand, at areas between the next nearest neighbouring Cs and F atoms, the ELF values vanish, a characteristic of no covalent bonding. The ELF analysis shows the same characteristics for isoelectronic Xe compounds XeF2 and NH4XeF5.
NATURE CHEMISTRY | VOL 5 | OCTOBER 2013 | www.nature.com/naturechemistry

2013 Macmillan Publishers Limited. All rights reserved.

NATURE CHEMISTRY

DOI: 10.1038/NCHEM.1754

ARTICLES
compounds, the inner-shell 5p electrons of Cs are reactive and engage with F atoms to form CsF covalent bonds. When oxidized beyond the 1 state, Cs behaves like a p-block element, and the resulting CsFn molecules are analogous to the known XeFn , blurring the difference between s- and p-block elements. These calculations show that the underlying mechanism of the activation of Cs 5p by pressure is the increase of the atomic orbital energy. Considering the pressure range in which CsFn compounds were found to be stable (5200 GPa), it is not unreasonable to suggest that the inner-shell electrons in other group IA and IIA elements may also be oxidized by strong oxidizing agents such as F or other chemical species under higher pressures.

Mechanism of pressure-driven high oxidation states. Because Cs exhibits multiple covalent and ionic bonds with neighbouring F atoms, its oxidation states are beyond 1, formally n in CsFn. To further support this, we calculated the charges using Baders quantum theory of atoms in molecules (QTAIM) analysis41 for CsF at a pressure of 100 GPa. In QTAIM analysis, the charges of an atom are calculated by integrating the charge density in the surrounding basin, partitioned by the stationary points of the charge density. As shown in Fig. 4a, the QTAIM charges of Cs increase almost linearly with an increasing number of F atoms in the chemical formula, and are seemingly larger than 1 for CsFn (n 2) compounds, which is in agreement with the oxidation state proposed. It may be emphasized that the QTAIM charges are usually notably smaller than the formal oxidation numbers, even for typical ionic crystals such as CsF (0.81 at 100 GPa), and are more so for covalent or molecular compounds. The potential for Cs to be oxidized to a high charge state can be demonstrated by the energy and geometry of the atomic orbitals. Figure 4b compares the 6s, 5p and 5s energy levels of Xe and Cs at various charge states, calculated using DFT for a single atom, with the relativistic effect fully included. The results indicate that although the 5p state of Cs0 is 5.28 eV lower than that of Xe0, the difference is reduced to only 1.66 eV for the 1 charge states and to 2.14 eV for the 2 charge states. Because xenon can be oxidized up to a 8 charge state, one might expect that Cs could also be oxidized beyond the 1 state. Pressure has multiple effects that can lead to reactivity of Cs 5p electrons. As shown in Fig. 4c, the Cs 5p orbital peaks at 0.98 and has a large component outside the Cs radius of 1.81 . Under pressure, the CsF distance may become much smaller than the sum of the Cs and F2 radii (3.0 ), leading to strong overlap of the Cs 5p and F 2p orbitals. Pressure can also elevate the energies of the atomic orbitals. Figure 4d shows the dependence of some atomic orbital levels on external pressure, as modelled by replacing the atom in a highly compressed He matrix. Although the energies of all atomic orbitals increase with pressure, the Cs 5p increases signicantly faster than F 2p, surpassing the latter at 10 GPa. In addition, the Cs 5p bands signicantly broaden under pressure, which further elevates the energy of some 5p states, inducing a sharing of the Cs 5p electron with the F atom. Inspired by the results of CsF, we also tested the pressure dependence of the outmost core states of Rb, K and Ba. We found that Rb 4p and Ba 5p orbitals may become higher in energy than F 2p around 200 GPa. This does not necessary imply that the core states are reactive at this pressure, because the chemical surroundings of Rb and Ba in corresponding RbF and BaF compounds might be very different to CsF compounds. However, it is reasonable to expect that these two elements can be oxidized beyond their usual oxidation state under reasonably high pressure. Many recent studies have shown the stabilization of compounds with unusual stoichiometry under pressure4245. However, none shows the reactivity of inner-shell electrons in forming chemical bonds, which is distinctively different to the CsF system. Although the effect of inner-shell electrons in chemical bonding and on molecular geometry has been noticed previously (for example, the linear geometry of the UO2 ion was attributed to the effect of U 6p electrons46), here, the inner-shell electrons are shown not only to affect atomic structures and properties, but also to become the main components of chemical bonds.

Methods
To obtain stable structures for CsFn , we conducted an unbiased structure prediction based on the particle swarm optimization algorithm as implemented in CALYPSO (crystal structure analysis by particle swarm optimization)31,32. In addition, more than 100 ABn structures in the Inorganic Crystal Structure Database (ICSD) were also included in the structure optimization. CALYPSO was able to nd all the stable structures, including the two (XeF2 and XeF4 structures) that exist in the ICSD database, showing its remarkable capability of predicting new structures for novel chemical systems. The structure predictions were performed using a unit cell containing up to four CsFn units and at pressures ranging from 0 to 200 GPa. The underlying ab initio structural relaxations and the electronic bandstructure calculations were performed within the framework of DFT as implemented by the VASP (Vienna Ab initio Simulation Package) code47. The generalized gradient approximation (GGA) within the framework of PerdewBurkeErnzerhof (PBE)48 was used for the exchange-correlation functional, and projector augmented wave (PAW) potentials49 were used to describe the ionic potentials. Owing to the reduced bond lengths under pressure, a hard PAW potential of the F atom, with a core radius of 0.58 , was used. In the PAW potential for Cs, the 5s, 5p and 6s orbitals were included in the valence. The accuracy of the PAW potentials was tested and compared with a full-potential method (Supplementary Section SVIII). The cutoff energy for the expansion of the wavefunction into plane waves was set at 1,200 eV, and MonkhorstPack k meshes were chosen to ensure that all enthalpy calculations converged to better than 1 meV/atom. The ELF was also calculated using VASP. As implemented in VASP, it ranges from 0 (free electron gas) to 1 (localized electrons). The COHPs were calculated using the Stuttgart TB-LMTO package.

Received 28 April 2013; accepted 9 August 2013; published online 23 September 2013

References
1. Landau, L. D. & Lifshitz, E. M. in Quantum Mechanics (Non-Relativistic Theory) Vol. 3 (Butterworth-Heinemann, 1958). 2. Pauling, L. The Nature of the Chemical Bond (Cornell Univ. Press, 1960). 3. Murrel, J. N., Kettle, S. F. A. & Tedder, J. M. in The Chemical Bond (Wiley, 1985). 4. Bartlett, N. Xenon hexauoroplatinate (V) XePtF6. Proc. Chem. Soc. Lond. (1962). 5. Agron, P. A. et al. Xenon diuoride and nature of xenonuorine bond. Science 139, 842844 (1963). 6. Siegel, S. & Gebert, E. Crystallographic studies of XeF2 and XeF4. J. Am. Chem. Soc. 85, 240 (1963). 7. Rundle, R. E. Probable structure of XeF4 and XeF2. J. Am. Chem. Soc. 85, 112 (1963). 8. Malm, J. G. et al. Xenon hexauoride. J. Am. Chem. Soc. 85, 110111 (1963). 9. Liao, M. S. & Zhang, Q. E. Chemical bonding in XeF2 , XeF4 , KrF2 , KrF4 , RnF2 , XeCl2 , and XeBr2: from the gas phase to the solid state. J. Phys. Chem. A 102, 1064710654 (1998). 10. Krouse, I. H. et al. Bonding and electronic structure of XeF3. J. Am. Chem. Soc. 129, 846852 (2007). 11. Khriachtchev, L., Pettersson, M., Runeberg, N., Lundell, J. & Rasanen, M. A stable argon compound. Nature 406, 874876 (2000). 12. Grochala, W. Atypical compounds of gases, which have been called noble. Chem. Soc. Rev. 36, 16321655 (2007). 13. Dmochowski, I. Xenon out of its shell. Nature Chem. 1, 250 (2009). 14. Brock, D. S. & Schrobilgen, G. Synthesis of the missing oxide of xenon, XeO2 , and its implications for Earths missing xenon. J. Am. Chem. Soc. 133, 62656269 (2011). 15. Kim, M., Debessai, M. & Yoo, C. S. Two- and three-dimensional extended solids and metallization of compressed XeF2. Nature Chem. 2, 784788 (2010). 16. Bode, H. Uber einige neue uoraktive Stoffe. Naturwissenschaften 37, 477 (1950). 17. Moock, K. & Seppelt, K. Indications of cesium in a higher oxidation-state. Angew. Chem. Int. Ed. Engl. 28, 16761678 (1989). 18. Asprey, L. B., Margrave, J. L. & Silverthorne, M. E. Tetrauorohalates of cesium, rubidium and potassium. J. Am. Chem. Soc. 83, 29552956 (1961).
851

Conclusions
In summary, rst-principles calculations have been used to demonstrate that Cs is oxidized to high charge states by F atoms under high-pressure conditions (pressures that are currently accessible experimentally). A series of stable CsFn compounds have been predicted, in which Cs has a formal oxidation number of n. In these
NATURE CHEMISTRY | VOL 5 | OCTOBER 2013 | www.nature.com/naturechemistry

2013 Macmillan Publishers Limited. All rights reserved.

ARTICLES
19. Jehoulet, C. & Bard, A. J. On the electrochemical oxidation of Cs and other alkali-metal ions in liquid sulfur dioxide and acetonitrile. Angew. Chem. Int. Ed. Engl. 30, 836838 (1991). 20. Schwarz, U., Takemura, K., Hanand, M. & Syassen, K. Crystal structure of cesium-V. Phys. Rev. Lett. 81, 27112714 (1998). 21. Takemura, K. et al. Phase stability of highly compressed cesium. Phys. Rev. B 61, 1439914404 (2000). 22. Takemura, K., Minomura, S. & Shimomura, O. X-ray diffraction study of electronic-transitions in cesium under high pressure. Phys. Rev. Lett. 49, 17721775 (1982). 23. Shamp, A., Hooper, J. & Zurek, E. Compressed cesium polyhydrides: Cs sublattices and H32 three-connected nets. Inorg. Chem. 51, 93339342 (2012). 24. Hooper, J. & Zurek, E. Rubidium polyhydrides under pressure: emergence of the linear H32 species. Chem. Eur. J. 18, 50135021 (2012). 25. Johnson, K. A. & Ashcroft, N. W. Structure and bandgap closure in dense hydrogen. Nature 403, 632635 (2000). 26. Belonoshko, A. B., Ahuja, R. & Johansson, B. Stability of the body-centred-cubic phase of iron in the Earths inner core. Nature 424, 10321034 (2003). 27. Oganov, A. R. et al. Ionic high-pressure form of elemental boron. Nature 457, 863867 (2009). 28. Ma, Y. et al. Transparent dense sodium. Nature 458, 182U183 (2009). 29. Wang, Y. C. et al. High pressure partially ionic phase of water ice. Nature Commun. 2, 563 (2011). 30. Zhu, L. et al. Spiral chain O4 form of dense oxygen. Proc. Natl Acad. Sci. USA 109, 751753 (2012). 31. Wang, Y. C., Lv, J., Zhu, L. & Ma, Y. M. CALYPSO: a method for crystal structure prediction. Comput. Phys. Commun. 183, 20632070 (2012). 32. Wang, Y. C., Lv, J. A., Zhu, L. & Ma, Y. M. Crystal structure prediction via particle-swarm optimization. Phys. Rev. B 82, 094116 (2010). 33. Peng, F., Miao, M. S., Wang, H., Li, Q. & Ma, Y. M. Predicted lithiumboron compounds under high pressure. J. Am. Chem. Soc. 134, 1859918605 (2012). 34. Feng, J., Hennig, R. G., Ashcroft, N. W. & Hoffmann, R. Emergent reduction of electronic state dimensionality in dense ordered LiBe alloys. Nature 451, 445448 (2008). 35. Ault, B. S. & Andrews, L. Matrix reactions of alkali-metal uoride molecules with uorineinfrared and Raman-spectra of triuoride ion in MF32-species. J. Am. Chem. Soc. 98, 15911593 (1976). 36. Riedel, S., Koechner, T., Wang, X. F. & Andrews, L. Polyuoride anions, a matrix-isolation and quantum-chemical investigation. Inorg. Chem. 49, 71567164 (2010). 37. Christe, K. O. et al. The pentauoroxenate(IV) anion, XeF52the rst example of a pentagonal planar AX5 species. J. Am. Chem. Soc. 113, 33513361 (1991). 38. Dronskowski, R. & Blochl, P. E. Crystal orbital Hamiltonian populations (COHP)energy-resolved visualization of chemical bonding in solids based on density-functional calculations. J. Phys. Chem. 97, 86178624 (1993).

NATURE CHEMISTRY

DOI: 10.1038/NCHEM.1754

39. Braida, B. & Hiberty, P. C. The essential role of charge-shift bonding in hypervalent prototype XeF2. Nature Chem. 5, 417422 (2013). 40. Silvi, B. & Savin, A. Classication of chemical bonds based on topological analysis of electron localization functions. Nature 371, 683686 (1994). 41. Bader, R. Atoms in Molecules: A Quantum Theory (Oxford Univ. Press, 1990). 42. Zurek, E., Hoffmann, R., Ashcroft, N. W., Oganov, A. R. & Lyakhov, A. O. A little bit of lithium does a lot for hydrogen. Proc. Natl Acad. Sci. USA 106, 1764017643 (2009). 43. Baettig, P. & Zurek, E. Pressure-stabilized sodium polyhydrides: NaHn (n . 1). Phys. Rev. Lett. 106, 237002 (2011). 44. Zhang, W. et al. Unexpected stable stoichiometries of sodium chlorides. Preprint at http://arxiv.org/1211.3644 (2012). 45. Zhu, Q., Oganov, A. R. & Lyakhov, A. O. Novel stable compounds in the MgO system under high pressure. Phys. Chem. Chem. Phys. 15, 76967700 (2013). 0 0 2 46. Tatsumi, K. & Hoffmann, R. Bent cis d 0 MOO2 2 vs linear trans d f UO2 a signicant role for non-valence 6p orbitals in uranyl. Inorg. Chem. 19, 26562658 (1980). 47. Kresse, G. & Furthmuller, J. Efcient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 1116911186 (1996). 48. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 38653868 (1996). 49. Blo chl, P. E. Projector augmented-wave method. Phys. Rev. B 50, 1795317979 (1994).

Acknowledgements
The author thanks R. Hoffmann at Cornell University and R. Seshadri at University of California Santa Barbara for inspiring discussions and constructive suggestions. This work was supported by the Materials Research Science and Engineering Center programme (National Science Foundation (NSF)-Division of Materials Research (DMR)1121053) and the Conversion of Energy Through Molecular Platforms-The Integrative Graduate Education and Research Traineeship programme (NSF-Division of Graduate Education (DGE)0801627). Calculations were performed on resources at the Center for Scientic Computing, supported by the California NanoSystems Institute, Materials Research Lab. and NSF (Division of Computer and Network Systems (CNS)-0960316), on NSF-funded Extreme Science and Engineering Discovery Environment resources (Teragrid (TG)-DMR130005) and on Beijing Computational Science Research Center computing resources.

Additional information
Supplementary information is available in the online version of the paper. Reprints and permissions information is available online at www.nature.com/reprints. Correspondence and requests for materials should be addressed to M.s.M.

Competing nancial interests

The authors declare no competing nancial interests.

852

NATURE CHEMISTRY | VOL 5 | OCTOBER 2013 | www.nature.com/naturechemistry

2013 Macmillan Publishers Limited. All rights reserved.