THE JOURNAL OF CHEMICAL PHYSICS 130, 134716 共2009兲

Density functional theory, molecular dynamics, and differential scanning

calorimetry study of the RbF–CsF phase diagram
O. Beneš,1,2,a兲 Ph. Zeller,2 M. Salanne,3,4 and R. J. M. Konings1
1
European Commission, Joint Research Centre, Institute for Transuranium Elements (ITU),
P.O. Box 2340, 76125 Karlsruhe, Germany
2
Department of Physics and Chemistry, CEA/Saclay, F-91191 Gif-sur-Yvette, France
3
UPMC Univ Paris 06, UMR 7612, LI2C, F-75005 Paris, France
4
CNRS, F-75005 Paris, France
共Received 27 November 2008; accepted 18 February 2009; published online 7 April 2009兲

A multiscale modeling approach is developed to compute the phase diagram of the RbF–CsF binary
system. The mixing enthalpies of the 共Rb,Cs兲F solid and liquid solutions are evaluated using density
functional theory and classical molecular dynamics calculations, respectively. For the solid solution,
18 different configurations are studied with density functional theory and the surrounded atom
model is applied in order to compute the configurational partition function. We also measure the
solidus and liquidus equilibria using differential scanning calorimetry. Finally the RbF–CsF phase
diagram is constructed using the calculated excess free enthalpies of the solid and liquid solutions
and a very good agreement with our experimental data is found. © 2009 American Institute of
Physics. 关DOI: 10.1063/1.3097550兴

I. INTRODUCTION
The design of tomorrow’s nuclear reactors involves the
development of new materials. Among them, in the molten
salt reactor 共MSR兲 the fuel consists of a molten fluoride
matrix in which the actinides are dissolved. Several criteria
must be fulfilled by MSR fuels, thus limiting the selection
of the salt. In the Russian concept of the nonmoderated
MSR 共MOSART concept兲1 the fuel is based on the
LiF– NaF– BeF2 matrix with 1.3 mol % of AnF3, mostly rep-
resented by 239PuF3, added as a fissile material. In a previous
study2 five fuel alternatives based on the MOSART concept
have been presented and one of them is a system based on
the LiF–NaF–RbF matrix. To predict the fuel behavior in
terms of its melting temperature, vapor pressure, or actinide
solubility, a thermodynamic description of the system
is needed. Moreover to investigate the thermodynamic
behavior of the fuel contaminated with some amount of
fission products that are difficult to extract during the
MSR clean-up process one must consider the presence of the
CsF compound. The thermodynamic assessment of the
LiF– NaF– RbF– CsF– PuF3 system has therefore been made
as described in this previous study.2 It has, however, been
noted in Ref. 3 that some additional data would be necessary
in order to describe the RbF–CsF subsystem more precisely.
This paper presents a thermodynamic assessment of the
RbF–CsF system. At the start of this work it was already
known that no intermediate compound exists in this system.
Both end members crystallize in the same fcc structure, the
solution is substitutional, and there is extended solubility
both in the solid and in the liquid states. Also, the thermo-
dynamic functions of both end members are well known.
Hence what was missing was the exact boundaries of the
a兲
Electronic mail: ondrej.benes@cea.fr.
liquid+ solid two-phase field, i.e., the liquidus and the soli-
dus, as well as the excess contributions to the thermody-
namic potentials. In this work we measure the solid-liquid
equilibria using differential scanning calorimetry 共DSC兲
共Sec. II兲. We also compare them to a phase diagram that we
build using the excess Gibbs energy of the 共Rb,Cs兲F solid
solution and that of the 共Rb,Cs兲F liquid solution, both calcu-
lated here. These latter calculations involve two distinct ap-
proaches, which have in common their first-principles basis,
for the solid and liquid phases.
The solid solution is treated in Sec. III. First some con-
figuration energies are calculated using the density functional
theory 共DFT兲.4 Next the relevance of the ”bond model”
共BM兲 and the ”surrounded atom model” 共SAM兲 共Ref. 5兲 is
discussed to extrapolate the energy to any configuration of
the solid solution. Then we use either of two approximations,
those of Bragg and Williams6 and Guggenheim,7 to compute
the partition function ⍀ and hence the free energy. Although
historically the Bragg–Williams 共BW兲 and Guggenheim ap-
proximations were presented within the frame of the BM
whereas the SAM was introduced combined with the
Guggenheim approximation, we distinguish in the present
paper the configurational energy models 共Sec. III B兲 from the
approximations made in the calculation of ⍀ 共Sec. III C兲.
The liquid solution is treated by classical molecular dy-
namics 共MD兲 共Sec. IV兲 with potentials built exclusively from
DFT data, as opposed to empirical or semiempirical poten-
tials. This technique has already proven to be successful for
the determination of a wide range of physicochemical prop-
erties of several molten salts: Dynamic 共diffusion coeffi-
cients, electrical conductivity, viscosity兲8 as well as thermo-
dynamic 共heat capacities, thermal expansions, activity
coefficients兲9,10 and structural quantities are well described
provided that the interaction potentials between the involved
species are well chosen.

0021-9606/2009/130共13兲/134716/13/$25.00 130, 134716-1 © 2009 American Institute of Physics

134716-2 Beneš et al. J. Chem. Phys. 130, 134716 共2009兲

Many of the techniques and tools used here 共i.e., DSC, TABLE I. Solidus and liquidus temperatures of the RbF–CsF system ob-
tained by DSC in this study.
first-principles calculations, MD, calculation of phase dia-
grams兲 are fairly standard and have already been thoroughly Tsolidus Tliquidus
described elsewhere–with the possible exception of our treat- xCsF 共K兲 共K兲
ment of the solid solution. This is why more emphasis is
given on the corresponding part of the paper 共Sec. III兲. 0 ¯ 1060.9
0.069 1022.6 1040.6
0.088 958.9 978
0.141 1015.6 1033.1
II. EXPERIMENTAL 0.236 985.9 1010.7
0.297 973.7 999.9
The RbF–CsF binary system has been investigated by 0.396 961.7 983.8
Samuseva and Plyushchev,11 who used thermal analysis to 0.506 962.1 978.2
determine the solidus and liquidus points. Their reported 0.597 947.1 967.9
phase diagram, however, shows a temperature minimum on 0.694 953.1 961.3
0.788 941.4 953.2
the solidus curve without a corresponding minimum on the
0.839 944.4 954.6
liquidus curve which results into a binary phase field which
0.858 960.3 999.4
is defined by two different equilibria 共solid solution–liquid 1 ¯ 959.4
solution and solid solution–solid solution兲, thus violating the
phase rule. Therefore new measurements of the solidus and
liquidus temperatures of the RbF–CsF system were made III. SOLID SOLUTION
in the present study. Our measurements are performed
using DSC performed on a SETARAM multidetector high We consider here a macroscopic sample made of NRb
temperature calorimeter. rubidium atoms, NCs cesium atoms, and N = NRb + NCs fluo-
The samples were prepared by mixing the initial pure rine atoms. Molar fractions are denoted
RbF 共Alfa Aesar, 99.975% metallic purity兲 and CsF
共Alfa Aesar 99.99% metallic purity兲 compounds. These NRb NCs
xRbF = and xCsF = . 共1兲
fluorides are very hygroscopic and therefore prior to the N N
sample preparation they were dried in a powder form under All the calculations performed in this work are done at zero
vacuum at T = 453 K for 4 h. Thus obtained dry precursors pressure so that the pV term is equal to zero, enthalpy is
were stored in a glove box which was kept under constant equal to internal energy 共H = U兲, and Helmholtz energy is
argon gas flow of 6.0 grade 共measured long time content of equal to Gibbs energy 共F = G兲. It must be noted that although
the moisture in the glove box is 28 ppm兲. In order to mini- for ambient conditions p = 1 bar, due to the fact that we are
mize the weighing error, all samples were weighed directly dealing with condensed phases the pV term is negligible
in the stainless steel, gas tight DSC crucibles specially de- throughout. Statistical thermodynamics in the canonical en-
veloped at ITU.12 Boron nitride 共oxygen-free, AX05 grade兲 semble at absolute temperature T relates the configurational
liner was inserted into the crucible to avoid direct contact part of the free energy to the partition function ⍀ through the
between the steel and the fluoride salts which are corrosive at following formulas:
high temperatures.
Every DSC measurement consisted of three heating and F = − kBT ln ⍀ 共2兲
cooling runs. The first run was always used to achieve good
and
mixing between the RbF and CsF compounds. During this
run the salt was kept for 1 h at T = 1173 K, well above the
melting temperature of both components. The data from this ⍀ = 兺 exp −
c
冉 ⌽共c兲
k BT
冊
, 共3兲
run have never been considered for the analysis. The two
other runs were monitored and consisted of the heating and where c represents a configuration, i.e., an arrangement of
cooling sequences, both at temperature rate of 3 K/min. The the set of Rb and Cs atoms on the cation sublattice, ⌽共c兲 is
peak temperature was 1173 K as in the case of the prerun. its energy, and kB is the Boltzmann constant. Notations are
Since the melting temperatures of the pure compounds were those of Mathieu et al.5 We assume that disorder does not
well reproduced from the heating and cooling curves, it has mix cations with anions since it is well established that an-
been assumed that there is no supercooling effect in this tisite defects in these strongly ionic compounds have a very
system. This observation was also supported by the fact that large positive energy. Accordingly, the configurational en-
RbF and CsF are highly ionic compounds and thus their ergy is attributed to cations only and from here on we do not
crystallization is very fast. Therefore it was possible to ana- mention F− anions any more in this section. In contrast to a
lyze the solidus points as the onset points of the DSC peaks “configuration,” a solid solution is a macroscopic concept
from the heating curves and the liquidus points as the onset encompassing all these microscopic configurations. The
points of the DSC peaks from the cooling curves. The results weight of a configuration in a given solid solution depends
are reported in Table I and included in Figs. 8 and 9 共in Sec. on external constraints as described by thermodynamic vari-
V兲 in which a comparison to the calculated phase diagrams is ables 共e.g., p, T, U, S, xRbF, xCsF, ␮Rb, and ␮Cs兲, out of which
made. we choose here p = 0, T, and xRbF. Mixing quantities obey
134716-3 Investigation of the RbF–CsF phase diagram J. Chem. Phys. 130, 134716 共2009兲

similar relationships as shown in Eqs. 共2兲 and 共3兲. In this TABLE II. Atomization energies and lattice parameters of some chosen
reference compounds.
study it is assumed that configurational energies do not de-
pend on temperature and they are computed for T = 0 K. In Lattice parametera Atomization energyb
other words the vibrational contributions to the excess free 共Å兲 共kJ mol−1兲
energy are neglected. This includes the so-called zero-point
energy, i.e., the phonon energy at T = 0 K, since we do not Compound Experimentalc CASTEP Experimentald CASTEP

compute any phonon spectrum. Rb metal 5.70 5.70 82.192 75.4

A. DFT calculation
1. CASTEP parameters
Our calculations are performed using the plane-
wave pseudopotential code CASTEP which implements
DFT and has been extensively described elsewhere.13 We
use the generalized gradient approximation of Perdew-
Burke-Ernzerhof 共GGA-PBE兲 approximation14 for the ex-
change correlation functional. The first Brillouin zone is
sampled with the Monkhorst–Pack scheme15 with intervals
no larger than 0.040 Å−1 on all axes. We use the MATERIALS
STUDIO 共Ref. 16兲 interface for launching the calculation and
analyzing the results as well as various interfaces and tools
within the CYGWIN/X package.17 The pseudopotentials are
also taken from the MATERIALS STUDIO library. To identify
the best cut-off energy value, several calculations of tetrag-
4
onal RbCsF2 共P m mm space group兲 have been performed
with cut-off energies ranging from 350 to 500 eV by 50 eV
steps. A satisfactory accuracy is achieved for a cut-off energy
of 400 eV and this value has been selected for all our calcu-
lations. All elements are described using Vanderbilt scheme
ultrasoft pseudopotentials18 with configurations 关He兴2s22p5
for F, 关Ca 3d10兴4s24p65s1 for Rb, and 关Sr 4d10兴5s25p66s1 for
Cs. For each configuration a “geometry optimization” is per-
formed, i.e., ionic positions and cell parameters are varied so
as to minimize the total energy.
Pseudopotentials from the MATERIALS STUDIO library are
provided with several convergence and validity tests. In or-
der to assess the validity and the performance of our chosen
set of parameters–including these pseudopotentials–we have
complemented those tests by computing equilibrium geom-
etries and atomization energies of pure Rb, Cs, RbF, and CsF
crystals and comparing them to experimental values taken
from the literature.19,20 For atomization energies we have in-
cluded in the calculated data the zero-point energies com-
puted with CASTEP 共all of them in the 1 kJ/mol atom order of
magnitude兲 and in the experimental data all the corrections
describing the transition from the standard state at 298 K to
the zero pressure and zero temperature conditions. The re-
sults are summarized in Table II. All lattice parameters agree
to within 1%, which is a good result consistent
with CASTEP standard performances for these settings,21 con-
firming the quality of the pseudopotentials. All calculated
atomization energies happen to be underestimated by
7 ⫾ 1 kJ/ mol atom. This similarity is of course fortuitous,
and the relative differences are 10% for the pure metals and
2% for RbF and CsF. The discrepancy for the pure metals is
in the usual range, but it may also reflect the fact that for
those simple metals the GGA usually performs worse than
the local density approximation. Since this paper focuses on
ionic and partly covalent materials GGA is arguably the best
Cs metal 6.14 6.14 78.014 70.0
RbF 5.64共2兲 5.710 718.180 704.5
CsF 6.03 6.100 709.598 693.0
a
All these compounds crystallize in a cubic structure.
b
At 0 K and 0 bar.
c
Reference 20.
d
References 19 and 42.
choice. For RbF and CsF these energy differences are accept-
able given the purpose of this work. In fact although errors
on energies of the order of 7 kJ/mol atom would in general
preclude the computation of a phase diagram completely
ab initio, relative errors of a few percentage on the excess
quantities are quite satisfactory and we may in fact expect
small relative errors on the excess quantities due to well-
known error cancellation properties in first-principles
calculations.4,21
2. Excess configurational energies
The energies of 18 different configurations of the
共Rb,Cs兲F solid solution have been calculated using CASTEP.
Energies of the pure RbF and CsF end members have then
been proportionally subtracted from the energy of each
共Rb,Cs兲F solid solution configuration in order to obtain the
excess configurational energy as
⌽xs共RbxCsyFx+y兲 = ⌽共RbxCsyFx+y兲 − x⌽共RbF兲
− y⌽共CsF兲. 共4兲
The choice of the various configurations was made by arbi-
trary displacement of the Rb and Cs atoms on the cation
sublattice in order to cover a large array of nearest neighbor
environments. The calculated excess configurational energies
are shown in Fig. 1 and reported in Table III. Although some
of the configurations have the same crystallographic space
group and contain the same amounts of Rb and Cs cations
per unit cell 共e.g., case of configurations 6, 8, and 9兲 the
arrangements of the cations within the cation sublattice are
different, as well as their cell parameters, so that they corre-
spond to different configurations and are in general associ-
ated with different energies.
B. Configurational energy models
In order to extrapolate ⌽ 关see Eq. 共3兲兴 from our CASTEP
data set to any configuration, several models can be consid-
ered. A very common one, used by Guggenheim in his treat-
ment of mixtures,7 can be called the bond model 共BM兲. It
distributes the configurational energy among the first neigh-
bor bonds, that is, in the present case, among the cation-
cation bonds since we neglect F− anions. Three parameters
are introduced: ␧RbRb, ␧CsCs, and ␧RbCs = ␧CsRb which describe
the bond energies according to their type. The energy of
134716-4 Beneš et al. J. Chem. Phys. 130, 134716 共2009兲

⌽SAM
xs
共c兲 = ⌽SAM
xs
共共M j兲,共Pi兲兲
= ⌽SAM
xs
共M 1, . . . ,M 12, P1, . . . , P12兲
12 12
=兺 M j共ERb
j
− 0
ERb 兲 + 兺 Pi共ECs
i 0
− ECs 兲. 共7兲
j=1 i=1

Fitting Eq. 共7兲 to CASTEP data requires optimization of

2z = 24 parameters 共12 ERbj 0
− ERb i
and 12 ECs 0
− ECs excess en-
ergies兲. Beyond the fact that this is impossible based on the
knowledge of “only” 18 configurations calculated in this
work, it would not make much physical sense to fit these
parameters completely independently of one another by re-
curring to, say, 26 configurations instead of 18. Nevertheless,
in the same way as in the original work by Mathieu et al.,22
we observe that a three-parameter law of variation of these
FIG. 1. Excess configurational energy ⌽xs of the various configurations of excess energies as a function of the atom surroundings is
the 共Rb,Cs兲F solid solution calculated with CASTEP sufficient in order to obtain a good agreement between the
共⽧ symbol兲 and with the SAM 共䊊 symbol, Sec. III B兲. 共Solid curve兲 Excess SAM and CASTEP results. Introducing the molar excess
enthalpy Hxs of the solid solution obtained with the BW thermodynamics
energies
model applied to the SAM data 共Sec. III C 1兲.

U j = NA共ERb
j 0
− ERb 兲 and Vi = NA共ECs
i 0
− ECs 兲, 共8兲
configuration c and the associated excess energy are
where NA is Avogadro’s number, the parabolic laws used in
z this study are
⌽共c兲 = 共NRb␧RbRb + NCs␧CsCs兲 + W␻
2
U j = ␣1 j + ␣2 j 2 共9兲
共5兲
and ⌽ 共c兲 = W␻ ,
xs
and

where z is the lattice dependent coordination parameter V i = ␤ 1i + ␤ 2i 2 , 共10兲

which in case of the 共Rb,Cs兲F fcc lattice is equal to 12, W is
the number of Rb–Cs nearest neighbor pairs in the sample, where ␣1 = 0 共Ref. 23兲 and ␣2, ␤1, and ␤2 are the three pa-
and ␻, the single parameter of the model for excess energy, rameters optimized by least squares fit to
has the dimension of an energy and is defined as
␣2 = + 76.4 J/mol cation, 共11兲
␧RbRb + ␧CsCs
␻ = ␧RbCs − . 共6兲
2 ␤1 = − 550.9 J/mol cation, 共12兲
The BM cannot be satisfactorily fitted to our CASTEP data set
as it leads to a rms deviation of the order of the excess ␤2 = + 85.2 J/mol cation. 共13兲
energies themselves. Therefore we switched to the more Excess configurational energies obtained from the SAM
elaborate treatment provided by the SAM. are plotted in Fig. 1 together with CASTEP values and a
The SAM has been presented by Mathieu et al.5,22 to very good data reproduction is evident. These results are also
describe binary alloys. We apply it here to the pseudobinary reported in Table III as well as the difference between
mixture by treating RbF and CsF as two atomic species. In CASTEP and the SAM giving a maximum absolute error of
the following we briefly summarize the original derivation 0.141 J/mol.
by Mathieu et al. using as much as possible the notations of We conclude this section with a few relationships5 which
their work.5 The main characteristic of the SAM is the el- will be needed in the next section.
ementary configurational support as a central atom 共in our
case a cation兲 in the force field of its z nearest neighbors on 12 12
the cation sublattice. This replaces the partitioning based on 兺 M j = NRb, 兺 P j = NCs 共14兲
pair interactions used in the BM described above. A configu- j=0 j=0
ration is characterized by a set of 2共z + 1兲 = 26 numbers de-
noted 共M 0 , M 1 , . . . , M 12 , P0 , P1 , . . . P12兲 where M j is the num- and
ber of Rb atoms having exactly j Cs first neighbors and Pi is 12 12
the number of Cs atoms having exactly i Rb first neighbors.
The corresponding individual atom energies are denoted ERb j 兺
j=0
jM j = 兺 jP j = W,
j=0
共15兲
i
and ECs according to the atom type and its surroundings. The
excess configurational energy of configuration c is therefore where W is the number of Rb–Cs nearest neighbor pairs, are
simply equal to derived from obvious conservation laws.
134716-5 Investigation of the RbF–CsF phase diagram J. Chem. Phys. 130, 134716 共2009兲

TABLE III. Calculated excess configurational energy of some configurations of the 共Rb,Cs兲F solid solution.

No. Compositiona Structure Space group 共IT number兲 ⌽CASTEP

xs b
⌽SAM
xs bc
, ⌽SAM
xs
− ⌽CASTEP
xs bc
,

1 RbF Cubic Fm3̄m 共225兲 0 0 0

2 Rb7CsF8 Cubic Fm3̄m 共225兲 0.935 0.937 0.002
4
3 Rb7CsF8 Tetragonal P m mm 共123兲 1.154 1.013 ⫺0.141
4 Rb3CsF4 Orthorhombic Cmmm 共65兲 1.401 1.441 0.040
2
5 Rb5Cs3F8 Monoclinic P m 共10兲 2.207 2.107 ⫺0.100
6 Rb20Cs12F32 Monoclinic Pm 共6兲 2.263 2.199 ⫺0.064
2
7 Rb20Cs12F32 Monoclinic P m 共10兲 1.499 1.635 0.136
8 Rb20Cs12F32 Monoclinic Pm 共6兲 1.816 1.877 0.061
9 Rb20Cs12F32 Monoclinic Pm 共6兲 2.319 2.431 0.112
4
10 RbCsF2 Tetragonal P m mm 共123兲 2.946 2.968 0.022
11 Rb2Cs2F4 Orthorhombic Pmmn 共59兲 1.319 1.256 ⫺0.062
12 Rb3Cs3F6 Hexagonal R3̄m 共166兲 1.256 1.256 0.000
13 Rb4Cs4F8 Tetragonal Amm2 共38兲 2.142 2.112 ⫺0.030
14 Rb4Cs4F8 Tetragonal Pmma 共51兲 2.142 2.112 ⫺0.030
15 Rb12Cs20F32 Monoclinic Pm 共6兲 1.663 1.654 ⫺0.010
2
16 Rb3Cs5F8 Monoclinic P m 共10兲 2.081 2.030 ⫺0.050
17 RbCs3F4 Orthorhombic Cmmm 共65兲 1.270 1.299 0.029
18 RbCs7F8 Cubic Fm3̄m 共225兲 0.796 0.804 0.009
4
19 RbCs7F8 Tetragonal P m mm 共123兲 0.845 0.890 0.044
20 CsF Cubic Fm3̄m 共225兲 0 0 0
a
The indicated composition corresponds to the number of atoms per unit cell.
b
All units are in kJ/mol of one RbxCs1−xF unit formula.
c
SAM stands for surrounded atom model 共Sec. III B兲.

Mj Pi N!
mj = and pi = , 共16兲 = 兺 g共a兲. 共18兲
NCzj NCzi NRb ! NCs! a=all A

where Csr = 共 rs 兲 = s ! / 共r ! 共s − r兲!兲 is the usual binomial coeffi- Given that g共a兲 is always positive, there exists a value aញ such
cient, define useful auxiliary variables m j and pi. that Eqs. 共17兲 and 共18兲 combine into

⍀=
N!
NRb ! NCs!
exp − 冉
⌽共aញ 兲
k BT
, 冊 共19兲
C. Calculation of the partition function
At this point all the terms of the sum in Eq. 共3兲 are where aញ is in general a function of T, NRb, and NCs and
ញ = ⌽共aញ 兲 is a weighed average of the configurational energy.
⌽
known, at least formally, from Eqs. 共7兲–共13兲. To calculate the
sum two basic models may be applied, the BW model6 and The free energy of mixing is obtained from ⍀ using Eq. 共2兲
the Guggenheim model7 also known as the quasichemical
model. Fm = ⌽共aញ 兲 + kBT共NRb ln xRbF + NCsF ln xCsF兲, 共20兲
We have just seen that the configurations can be sorted
and the Boltzmann law for the probability of an arbitrary
using an auxiliary parameter, say, A, such that all configura-
configuration c to occur becomes
tions associated with A = a have the same energy E共a兲. The
number of configurations having A = a is the degeneracy de-
noted g共a兲. In the BM A is simply equal to W, the number of
Rb–Cs nearest neighbors pairs, while in the SAM A is an
prob共c兲 =
NRb ! NCs!
N!
exp 冉
⌽共aញ 兲 − ⌽共A共c兲兲
k BT
. 冊 共21兲

array of 26 numbers 共M 0 , M 1 , . . . , M 12 , P0 , P1 , . . . , P12兲 关see

Eq. 共7兲兴. What matters here is that any single configuration c
can be given a specific value a = A共c兲, be it in the BM or in
1. Bragg–Williams model
the SAM. After having grouped all the configurations which
have the same value of A, and therefore the same energy The BW model considers a completely disordered solu-
denoted from now on ⌽共a兲, Eq. 共3兲 becomes tion. From the point of view of configurational arrangements
this assumption implies that the probability for any given

⍀= 兺 g共a兲exp
a=all A
冉 −
⌽共a兲
k BT
冊 共17兲
cation site to be occupied by a Cs 共resp. Rb兲 cation is equal
to xCsF 共resp. xRbF兲 whatever its surroundings. Therefore the
probability for a given Rb atom to have exactly j Cs atoms
and z − j Rb atoms among its 12 nearest neighbors is given
and the total number of configurations is by the binomial distribution as
134716-6 Beneš et al. J. Chem. Phys. 130, 134716 共2009兲

M ⴱj clear departure from the BM and the solution is asymmetric.

= CzjxCsF
j z−j
xRbF , 共22兲 This has to do with the fact that the substitution of a Cs atom
NRb
into a RbF crystal is energetically different from the opposite
from which we get case of substituting a Rb atom into a CsF crystal. The main
mⴱj = xCsF
j z−j+1
xRbF and symmetrically pⴱj = xRbF
j z−j+1
xCsF . characteristics of the BW model are the temperature indepen-
dence of Hxs and the absence of any excess entropy. Its as-
共23兲 sumption of complete disorder is best justified in the limiting
In this and the following sections all SAM quantities relative case of infinite temperature. It is interesting to note that as
to the BW solution are denoted with a star superscript. These illustrated by Fig. 1 any single one among the configurations
expressions together with Eq. 共15兲 yield the number of near- studied with CASTEP would in itself generally give a poor
est neighbor Rb–Cs bonds, estimate of the excess internal energy/enthalpy of the solu-
tion. In fact even a zero excess configurational energy is
Wⴱ = NzxCsFxRbF . 共24兲 possible for any concentration—although it is not displayed
From a thermodynamics point of view the BW assumption of on the graph. This zero energy is obtained for a single con-
complete disorder implies that the entropy of mixing is equal figuration, that with complete spatial separation of both com-
to that of ideal mixing, ponents. The excess internal energy of the solid solution, as a
thermodynamic potential, takes into account the fact that in
m
SBW = − NkB共xRbF ln xRbF + xCsF ln xCsF兲. 共25兲 the real crystal disorder may favor higher configurational en-
A comparison of this expression with Eq. 共20兲 shows imme- ergies due to their higher statistical weight.
diately that the internal energy of mixing is
m
UBW m
= FBW m
+ TSBW = ⌽共aញ 兲 共26兲
2. The quasichemical “Guggenheim” model
and due to the general law 共⳵U / ⳵T兲 p=0,xCsF = T共⳵S / ⳵T兲 p=0,xCsF
it appears that ⌽共aញ 兲, like SBW
m
, is temperature independent. Equation 共21兲 shows that if two configurations have a
Evaluating it as an average configurational energy in the large energy difference the one with the lowest energy has a
limit of infinite temperature 共so that all configurations have significantly larger probability of occurrence. This fact is ne-
the same probability of occurrence兲, using the linearity of glected by the BW model but it is taken into account by the
Eq. 共7兲 with respect to M j and Pi and using the fact that M ⴱj Guggenheim model which we now briefly recall. A crucial
and Pⴱi given by Eqs. 共22兲 and 共23兲 are also configurational step in this model is to write g共a兲 as
averages, we obtain g共a兲 = h共NRb,NCs兲u共a兲, 共31兲
12
1 where u共a兲 is some known expression 共which will be made
m
UBW = lim
⍀
兺 ⌽共c兲 = 兺
x→⬁ all c j=1
共M ⴱj U j + Pⴱj V j兲 = ⌽共aⴱ兲.
explicit later兲 that makes up a fair approximation to g but
also needs to be complemented by h, called the “normaliza-
共27兲 tion factor,” which only depends on NRb and NCs and is not
This derivation, in the course of which all exponential Bolt- known explicitly. The Guggenheim approximation consists
zmann terms of Eq. 共21兲 are taken equal to 1, clearly shows in making use of the theorem of the dominant distribution
where the approximation lies and that the BW model is only which states that the sum in Eq. 共17兲 关Eq. 共18兲兴 can be re-
justified at high enough temperatures. Using Eqs. 共23兲 and placed by its maximum term. Forgetting momentarily that
共9兲–共13兲 and summing over index j leads to the following the starred quantities have already been defined above within
expression for the molar excess energy and enthalpy 共equal the frame of the BW model, we define here aⴱ such that
in the present case兲: g共aⴱ兲 is the maximum term of the sum in Eq. 共18兲. We thus
obtain aⴱ, which is a function of NRb and NCs but is indepen-
xs
Hm,BW ⬅ Um,BW
xs
= xRbF共1 − xRbF兲共L11 + L21xRbF兲, 共28兲 dent of temperature, by solving
where
L11 = z共␣1 + ␤1 + z␣2 + ␤2兲 = 5413 J/mol 共29兲
冏 冏
du
da a=aⴱ
= 0. 共32兲

and Equation 共18兲 becomes

L21 = z共z − 1兲共␤2 − ␣2兲 = 1162 J/mol. 共30兲 N!
= g共aⴱ兲, 共33兲
The resulting molar excess enthalpy is shown in Fig. 1 NRb ! NCs!
together with the calculated molar configurational en-
ergies. The peak of the curve corresponds to from which we get h共NRb , NCs兲 as
xs
Hm,BW = 1.502 kJ/ mol and xCsF = 0.48 so that a slight asym- N! 1
metry toward the RbF side is found. As shown by Mathieu h共NRb,NCs兲 = . 共34兲
NRb ! NCs! u共aⴱ兲
et al.22 if U j and Vi in Eqs. 共9兲 and 共10兲 followed linear laws
共␣2 = ␤2 = 0兲 the SAM would be completely equivalent to the In a similar way the value of a that maximizes the running
BM with ␻ = 共␣1 + ␤1兲 / NA and it would describe a symmetric term in Eq. 共17兲, denoted ā, is obtained as a function of NRb,
“regular” solution. Since a quadratic law is obeyed there is a NCs, and temperature T by solving
134716-7 Investigation of the RbF–CsF phase diagram J. Chem. Phys. 130, 134716 共2009兲

冏 冉
d
da
u共a兲exp −
⌽共a兲
k BT
冉 冊冊冏 a=ā
= 0. 共35兲 u=
NRb ! NCs!
j
Nm0 ! 共Nm1!兲z ¯ 共Nm j!兲Cz ¯ Np0 ! ¯ 共Npi!兲Cz ¯
i

The Guggenheim approximation leads to 共44兲

冉 冊
and ln共u兲 is well approximated by
⌽共ā兲
⍀ = g共ā兲exp − 共36兲 12

ln共u兲 = − N 兺 Czj共m j ln m j + p j ln p j兲.

k BT
共45兲
j=0
and using Eq. 共34兲 to eliminate h共NRb , NCs兲 we obtain the
final expression of ⍀ as The starred quantities 关aⴱ, i.e., the set of M ⴱj and Pⴱi that
cancel the differential du in Eq. 共32兲兴 end up having the same
⍀=
N! u共ā兲
ⴱ exp −
NRb ! NCs! u共a 兲
⌽共ā兲
k BT
. 冉 冊 共37兲
expressions as those already given within the BW model
关Eqs. 共22兲 and 共23兲兴: Maximization of g in Eq. 共18兲 is
equivalent to the hypothesis of total disorder and there is no
The free energy of mixing is obtained from ⍀ using Eq. 共2兲 ambiguity in our notations. The barred quantities are equal

冉 冊
to25

冉 冊
u共ā兲
Fm = ⌽共ā兲 + kBT NRb ln xRbF + NCsF ln xCsF − ln . Uj
u共aⴱ兲 m̄ j = m̄0e j␭ exp − , 共46兲
RT
共38兲

As shown in the original paper5 the following treacherously

simple expressions for the molar excess internal energy,
p̄i = p̄0e−i␭ exp − 冉 冊
Vi
RT
, 共47兲

enthalpy, and entropy do hold:24

where
12
NA xRbF
Hxs ⬅ Uxs = ⌽共ā兲 = 兺 Czj共m̄ jU j + p̄ jV j兲 共39兲 m̄0 = , 共48兲
N j=1
Z共␭兲

and 1 − xRbF
p̄0 = , 共49兲
Q共␭兲
R u共ā兲
Sxs = 共40兲
冉 冊
ln ,
N u共aⴱ兲 z
Uj
Z共␭兲 = 兺 Czje j␭ exp − , 共50兲
where R is the perfect gas constant. Let us now specify ex- 0 RT
plicitly function u共a兲.

冉 冊
z
In the BM, which we recall for illustration purposes, a is Vi
the number W of Rb–Cs nearest neighbor pairs in the sample Q共␭兲 = 兺 Czie−i␭ exp − , 共51兲
0 RT
and u共a兲 is defined as

冉 冊
and ␭ is such that e␭ is the single real positive root of the
z polynomial of degree 2z = 24 in e␭ defined by
N !

冋 册
2

冉 冊冉 冊冉冊冉冊
u共a兲 = . 共41兲 ⳵Z ⳵Q
z a z a a a R共e␭兲 = ez␭ xRbFQ + 共1 − xRbF兲Z . 共52兲
NRb − ! NCs − ! ! ! ⳵␭ ⳵␭
2 2 2 2 2 2
These equations, which complete the presentation of the
Therefore aⴱ, determined by du / da = 0 using Stirling’s for- Guggenheim approximation applied to the SAM, have no
mula, is equal to simple analytical solution in the general case. A numerical
solution is conveniently obtained using SCILAB.26
aⴱ = Wⴱ = N z xRbF xCsF 共42兲 The BW model of Sec. III C 1 can be derived using the
present Guggenheim method by just replacing the definition
and ā is the solution of Eq. 共35兲 which reads
of ā in Eq. 共35兲 by ā = aⴱ, where aⴱ is given by Eq. 共32兲. Thus
ā2
共zNRb − ā兲共zNCs − ā兲
= exp − 冉 冊 2␻
k BT
共43兲
Eqs. 共37兲 and 共38兲 become

⍀BW =
N!
exp −冉⌽共aⴱ兲
冊 共53兲
NRb ! NCs! k BT
and gives its name to the “quasichemical” model due to the
resemblance with the mass-action law. However, as stated in and
Sec. III B, in the present case the BM is not satisfactory. m
FBW = ⌽共aⴱ兲 + NkBT共xRbF ln xRbF + xCsF ln xCsF兲, 共54兲
In the SAM the expressions are much more complicated
due to the fact that a is an array of 26 parameters. Function from which Eqs. 共27兲 and 共25兲 can be deduced. This deriva-
u is equal to tion clearly shows why the BW model is also called “zero-
134716-8 Beneš et al. J. Chem. Phys. 130, 134716 共2009兲

FIG. 3. 共Color online兲 Excess enthalpy of the solid solution calculated for
FIG. 2. 共Color online兲 Excess Gibbs energy of the 共Rb,Cs兲F solid solution.
different values of SRO with fixed T = 1050 K, បCs xs,⬁
= 5417 J / mol, and
The solid curves, obtained with the Guggenheim approximation, are iso-
បRb
xs,⬁
= 6580 J / mol. Dashed 共SRO⬍ 0兲 and solid lines 共SRO⬎ 0兲 are
therms between 300 and 900 K with a 100 K step. The dashed line is
obtained with the Guggenheim approximation and the thick line
obtained with the BW approximation.
共SRO= 0.019兲 corresponds to CASTEP results. The dotted line is obtained
with the BW approximation and is insensitive to SRO.
order approximation” while the Guggenheim model is called
“first order approximation.” It also shows that the difference
perature. The third relationship, written here in the limiting
between both models can be quantified by ā − aⴱ.
case of a cesium-rich solution, reads
Going back to the approach behind Eq. 共31兲, a careful
examination of Eq. 共41兲 reveals that u共a兲 does not qualify as
the number of configurations having a number of Rb–Cs
bonds equal to a because of interferences between the differ- SRO = 1 −
P̄2
Pⴱ2
= 1 − exp −冉V2 − 2V1
RT
冊
= 1 − exp −
2␤2
RT
冉 冊
ent types of pairs and of double counting of oriented Rb–Cs
bonds. Similarly u共a兲 in Eq. 共44兲 is not quite equal to the 共57兲
number of configurations having A = a because a given atom
belongs to the surroundings of several other atoms and this and defines a parameter that we call SRO for “short-range
kind of interference is not taken into account.27 Therefore the order.” Recalling that P2 is the proportion of cesium atoms
sum of u over all values of a allowed by the model is never that have two rubidium first neighbors, SRO quantifies a dif-
exactly equal to the total number of configurations. All these ference between the BW 共Pⴱ2兲 and the Guggenheim models
effects would be very difficult to take into account properly; 共P̄2兲 in terms of cation surroundings in the first coordination
hence the correction introduced with function h partly shell, i.e., in fact some kind of short-range order associated
removes these defects and leads to the correct orders of with the Guggenheim model since the BW model is based on
magnitude. an assumption of complete disorder. Now by inverting Eqs.
The excess Gibbs energy as a function of T and xCsF is 共55兲–共57兲 the individual excess cation energies U j and Vi in
reported in Fig. 2 where a comparison between the Guggen- terms of which the SAM describes the solid solution can be
heim and the BW models is made. expressed using បRb xs,⬁
, បCs
xs,⬁
, and SRO. For given បRb xs,⬁
and
The solid solution is described in the present model us- បCs the BW model is insensitive to the third parameter,
xs,⬁

ing the three fitting parameters denoted ␣2, ␤1, and ␤2 and SRO, whereas the results of the Guggenheim model depend
given by Eqs. 共11兲–共13兲. A more physically intuitive interpre- on it. This is illustrated in Fig. 3 where the excess enthalpy is
tation of these three parameters will now be given by resort- plotted for បCsxs,⬁
and បRb xs,⬁
fixed at the values cited above for
ing to three relationships given in Refs. 5 and 22. Two of a constant temperature T = 1050 K and for a range of SRO
them pertain to the excess partial molar enthalpy of CsF in a values. By construction all the curves merge at the end
rubidium-rich solution 共xRbF close to 1兲, approximately given points 共xRbF ⬃ 0 and xRbF ⬃ 1兲 and they mostly differ in the
by the following limiting value: middle range of xCsF. This figure also shows that according
to CASTEP and the SAM, i.e., ␤2 given by Eq. 共13兲 or
បCs
xs,⬁
= Vz + zU1 = z共␣1 + ␤1 + ␣2 + z␤2兲, 共55兲
SRO= 0.019, short-range ordering occurs indeed in the
and symmetrically to the excess partial molar enthalpy of 共Rb,Cs兲F solid solution but its energetic influence remains
RbF in a cesium-rich solution quite limited.
Let us now relate this to a more commonly used defini-
បRb
xs,⬁
= Uz + zV1 = z共␣1 + ␤1 + z␣2 + ␤2兲. 共56兲
tion of short-range order. In a given configuration the aver-
In the present case we find បCs
xs,⬁
J / mol and
= 5417 age surroundings of a Rb cation, i.e., the average number of
បRb
xs,⬁
= 6580 J / mol. These values and the expressions for Cs cations around it, is equal to 兺zj=0 jM j / 兺zj=0M j. Hence,
these limiting partial molar enthalpies are valid for both the given a Rb cation and a given site s among its nearest neigh-
BW and the Guggenheim model: They are insensitive to tem- bors, the probability prob共Cs兩 Rb兲 that site s is occupied by a
134716-9 Investigation of the RbF–CsF phase diagram J. Chem. Phys. 130, 134716 共2009兲

IV. LIQUID SOLUTION: MOLECULAR DYNAMICS

The 共Rb,Cs兲F liquid solution has been studied by MD

simulations. In the case of molten fluorides, the potential is
best described as the sum of four different components:
Charge-charge, dispersion, overlap repulsion, and
polarization.30 The first three components are purely pairwise
additive; first the charge-charge term is
q iq j
Vq−q = 兺 , 共59兲
i⬍j rij

where qi is the charge on ion i, and formal charges are used

throughout. The dispersion component includes dipole-
dipole and dipole-quadrupole terms,

FIG. 4. 共Color online兲 The correlation between short-range order parameters

SRO and ␹ at different values of xRbF. All curves are calculated with fixed i⬍j
冉
Vdisp = − 兺 共1 − f 6共bijrij兲兲
C6,ij
共rij兲 6 + 共1 − f 8共bijrij兲兲
C8,ij
冊
共rij兲8
,
T = 1050 K, បCs
xs,⬁
= 5417 J / mol, and បRb
xs,⬁
= 6580 J / mol.
共60兲

Cs cation is equal to 1 / 共zNRb兲兺zj=0 jM j and one wants to com- where C6,ij 共C8,ij兲 is the dipole-dipole 共dipole-quadrupole兲
pare it to xCsF. This suggests to define a short-range order dispersion coefficient, and 共1 − f n兲 are Tang–Toennies disper-
parameter ␹ by sion damping functions31 describing the short-range penetra-
tion correction to the asymptotic multipole expansion of
dispersion32 关f n共0兲 = 1 and f n共⬁兲 = 0兴. These functions take
z the form
prob共Cs兩Rb兲 1 W̄
␹=1− =1− 兺 jM̄ j = 1 − Wⴱ , n
xk
f n共x兲 = e−x 兺
xCsF NzxRbFxCsF j=0
共61兲
共58兲 k=0 k!

and the parameter bij represents the distance at which the

correction begins to be taken into account. The third term of
where W̄ is obtained from Eqs. 共15兲, 共16兲, 共46兲, and 共47兲 and
the interaction potential, the repulsion overlap component, is
computed numerically together with the solution of the
given by
Guggenheim model. This is in fact the definition of the
Cowley order parameter for the first coordination sphere.28 If Vrep = 兺 Bije−aijrij . 共62兲
the solution is completely disordered then ␹ = 0. If ␹ ⬍ 0 the i⬍j
solution shows a trend to intermediate compound formation.
The polarization part of the potential includes charge-
If ␹ ⬎ 0 the trend is toward segregation or phase separation.
dipole and dipole-dipole terms
Here ␹ is a function of T and xCsF and it also depends on the
SAM parameters បCs , បRb , and SRO. In Fig. 4 we plot ␹ Vpol = − 兺 qi␮␣,j共1 − c ji f 4共bD,ijrij兲兲T␣共1兲
xs,⬁ xs,⬁

versus SRO for several values of xRbF while បCs xs,⬁

, បRb
xs,⬁
, and i,j
T remain fixed at the same values as for Fig. 3. The figure
shows that SRO is in good qualitative agreement with the + 兺 q j␮␣,i共1 − cij f 4共bD,ijrij兲兲T␣共1兲
i,j
more standard Cowley definition and that it has the same
order of magnitude. A plot of ␹ versus SRO for a different 1
− 兺 ␮␣,i␮␤,jT␣␤
共2兲
+兺 ជ i兩 2 .
兩␮ 共63兲
temperature would also yield an almost linear dependence, i,j i 2␣i
only with a different slope. It is interesting to note that SRO
is explicitly based on correlations in the first coordination Here T␣共1兲 and T␣␤共2兲
are the charge-dipole and dipole-dipole
sphere whereas ␹ only reflects the average surroundings, and interaction tensors while ␣i is the polarizability of ion i.
this difference appears most strikingly in highly diluted so- Again, Tang–Toennies functions are included to account for
lutions where ␹ becomes worthless since “at first order” such the short-range effects. The set of induced dipoles 兵␮ ជ i其i苸N is
a solution is always completely disordered. Besides, the fact treated as 3N additional degrees of freedom of the system.
that ␹ and SRO are so closely linked is due to the shape The dipoles are determined at each time step by minimiza-
functions we chose for the surrounding atom energies 关Eqs. tion of the total polarization energy and they depend on the
共9兲 and 共10兲兴, whereas with arbitrary U j and Vi there would positions of all the atoms at the corresponding time; there-
not necessarily be such similarity between them.29 The big fore the polarization part of the potential is considered to be
merit of SRO is that it provides us with a clear and simple a many body term.
physical interpretation for the third parameter of the SAM, All the parameters necessary to simulate RbF–CsF mix-
whereas the Guggenheim model, and therefore W or ␹ as tures have been determined from a recently developed first-
well, has no analytical solution in the general case. principles procedure.33,34 The pair parameters are summa-
134716-10 Beneš et al. J. Chem. Phys. 130, 134716 共2009兲

TABLE IV. Parameters of the interaction potential 共a.u.兲.

Ion pair ij Bij aij C6,ij C8,ij bij bD,ij cij c ji

− −
F –F 282.3 2.444 15.0 150.0 1.9 1.0 0.0 0.0
F− – Rb+ 150.96 1.961 0.001 0.001 1.9 1.822 3.463 ⫺0.440
F− – Cs+ 151.12 1.874 10.95 109.5 1.9 1.930 3.391 0.485
Rb+ – Rb+ 1.0 5.0 0.001 0.001 1.9 1.0 0.0 0.0
Rb+ – Cs+ 5000.0 3.0 0.001 0.001 1.9 1.0 0.0 0.0
Cs+ – Cs+ 5000.0 3.0 8.0 80.0 1.9 1.0 0.0 0.0

rized in Table IV. The polarizabilities were, respectively, of
7.9, 8.4, and 14.8 a.u. for F−, Rb+, and Cs+ ions.
The MD simulations were performed on 11 molten salt
compositions ranging from pure RbF to pure CsF. All the
corresponding simulation cells contained 432 ionic pairs.
The mixtures were simulated in the NPT ensemble following
the method described by Martyna et al.,35 with a pressure
fixed at 0 GPa and a temperature of 1200 K. We chose a time
step of 0.5 fs and after 100 ps of equilibration, production
runs of 200 ps were conducted for each composition.
Enthalpy of the systems was sampled each 50 fs during
the production runs. This quantity was then averaged over
the full simulation, and its composition dependence is dis-
played in Fig. 5. The error bars were determined by block
averaging. We could then extract the enthalpy of mixing for
each composition, which corresponds to the difference be-
tween the obtained enthalpy and the value corresponding to
ideal mixing between the two components according to
Eq. 共4兲.
The variation of the enthalpy of mixing with composi-
tion is displayed in Fig. 6. It does not exceed 140 J mol−1,
which is a very small value in this type of system. The error
bars have the same amplitude as in Fig. 5, but because of the
small quantities involved, they now have the same order of
magnitude as the enthalpies of mixing themselves. As a com-
parison, in molten fluoride systems which involve the mixing
of very different cations, such as Li+ and Cs+ cations, the
excess enthalpy reaches a value of −3330 J mol−1, with a
similar error bar 共and at the same temperature兲, as calculated
within the frame of this study. This means that the RbF and
CsF systems can be considered to mix almost ideally.
The mixing properties in a liquid depend a lot on the
structural properties of its various components.36,37 In pure
molten salts, the first-shell structure of a given cation varies
with its size and valence charge. Longer range structure is
mainly understood on the basis of packing and Coulomb
ordering. When adding a second component, this structure is
broken to an extent which depends on the compatibility of
the two materials. For example, one can distinguish the net-
work formers such as BeF2 共Refs. 8 and 38兲 from the net-
work breakers such as CsF. In the case of the RbF–CsF mix-
ture, the two cations have some very similar first-shell
structures. In Fig. 7, we have plotted the Rb–F, Cs–F, and
F–F radial distribution functions for three different composi-
tions. First, it appears that the first maxima of the cation-
anion functions correspond to distances of, respectively, 2.6
and 2.7 Å for Rb+ and Cs+. Second, all the radial distribution
functions change only slightly upon mixing, which means
that all the species have similar environments in the pure
liquids and in the mixture. This is of course consistent with
the variations obtained for the enthalpies of mixing, and it
confirms the nearly ideal behavior of the mixture.
V. THERMODYNAMIC ASSESSMENT OF THE RbF–CsF
PHASE DIAGRAM
As described above, two models, the BW and the
Guggenheim models, have been used to interpret the DFT
data relative to the solid solution. In order to simplify and to
keep a compatibility with our developed database,2 the re-
sults from these models have not been directly used for the
assessment of the RbF–CsF phase diagram. For the excess

FIG. 6. 共䊏兲 Enthalpy of mixing and their error bars for all considered
compositions calculated for T = 1200 K. 共Solid line兲 Fit using the modified
FIG. 5. Computed mean enthalpies for all the compositions. quasichemical model 共Sec. V兲.
134716-11 Investigation of the RbF–CsF phase diagram J. Chem. Phys. 130, 134716 共2009兲

xs
Gm,Gug 冉
= xRbF共1 − xRbF兲 5423 −
544 000
T共K兲
冊
2
+ xRbF 共1 − xRbF兲1500 J mol−1 . 共66兲
For a description of the excess Gibbs energy of the
共Rb,Cs兲F liquid solution a modified quasichemical model
based on the quadruplet approximations proposed by Pelton
and co-workers39,40 which is well suited for ionic liquids has
been used. The optimization of the excess parameters was
based on the results of the MD study 共see Sec. IV兲 which
suggested slight positive excess enthalpy and little or no ex-
cess entropy. We choose to describe the excess Gibbs energy
of the 共Rb,Cs兲F liquid solution by a single, temperature in-
dependent parameter, optimized in this study to the value

⌬gRbCs/FF = 188.3 J mol−1 . 共67兲

FIG. 7. Radial distribution functions for the xCsF = 0.0, 0.5, and 1.0
compositions.
Gibbs energy description of the 共Rb,Cs兲F solid solution the
classical polynomial formalism generally defined as
The same notation as proposed by Chartrand and
Pelton41 is kept here. A comparison between the mixing
enthalpy obtained from the MD study and the calculated data
based on our thermodynamic assessment are given in Fig. 6
and show an excellent agreement. The ⌬gRbCs/FF term which
is the matter of optimization when using the modified qua-
sichemical model is in fact equal to the Gibbs energy differ-
ence of the exchange reaction of the Rb and Cs atoms on the
cation sublattice as shown by
共Rb – F – Rb兲 + 共Cs – F – Cs兲

n
↔ 2共Rb – F – Cs兲 ⌬gRbCs/FF . 共68兲
xs
G = 兺
i,j=1
xi1x2j Lij 共64兲
Hence it is related to the short-range ordering of the
共Rb,Cs兲F liquid solution. In other words only when
has been used, where Lij are the parameters to be optimized ⌬gRbCs/FF = 0 is the solution characterized by a random mix-
and are written in a polynomial form. In the case of the BW ing and the configurational entropy is equal to the ideal mix-
results the expression is exact and the Lij coefficients were ing term 共−xRbFR ln xRbF − xCsFR ln xCsF兲. Any nonzero value
obtained analytically according to Eqs. 共28兲–共30兲 using the always results into ordering of the solution. However, since
␣2, ␤1, and ␤2 parameters from Eqs. 共11兲–共13兲 and the cor- the ⌬gRbCs/FF value obtained in our study is very small 关see
responding excess Gibbs energy is Eq. 共67兲兴 the “ordering” effect is negligible and the configu-
rational entropy is close to ideal. For example, the calculated
xs
Gm,BW = xRbF共1 − xRbF兲5413 mixing entropy at xRbF = 0.5 and T = 1000 K deviates from
the ideal behavior only by 0.0003 J mol−1 K−1.
2
+ xRbF 共1 − xRbF兲1162 J mol−1 . 共65兲 Furthermore the modified quasichemical model re-
quires the definition of the coordination numbers of the
The excess Gibbs energy function based on the Guggenheim Rb and Cs atoms in the unary and binary interactions.
model was obtained by the least squares fit of its results and They were taken from our previous study3 and are set to
Rb Cs Rb Cs
the final equation, valid for the temperature range 300–1200 ZRbRb = ZCsCs = ZRbCs = ZRbCs = 6.
K, is given below The thermodynamic data of the solid and liquid phases

TABLE V. Gibbs energy equations of the solid and liquid phases of the pure RbF and CsF compounds.
Coefficients are in J mol−1. All data are valid for temperature range 298–2000 K.

Phase a bT cT ln T dT2 eT−1

RbF 共s兲 ⫺573 639 213.919 ⫺42.343 −1.3012⫻ 10−02 23 450

RbF 共l兲 ⫺563 539 390.427 ⫺71 ¯ ¯
CsF 共s兲 ⫺569 439 230.509 ⫺46.811 −8.7950⫻ 10−03 4 111
CsF 共l兲 ⫺565 798 405.400 ⫺74.057 ¯ ¯
134716-12 Beneš et al.
FIG. 8. RbF–CsF phase diagram assessed using the BW model 共solid lines兲
and comparison to the solidus and liquidus data measured in this study by
DSC. 共쎲兲 Solidus data determined from the heating run; 共䊊兲 liquidus data
determined from the cooling curves.
of pure RbF were taken from JANAF tables42 as well as the
data for the solid phase of CsF. However, the data of the CsF
liquid phase had to be modified in order to reproduce its
melting temperature, T = 961 K 共compared to 975 K accord-
ing to Ref. 42兲, measured in this study. The Gibbs energies of
all four phases as a function of temperature G共T兲 = a + bT
+ cT ln T + dT2 + eT−1 are given in Table V.
The RbF–CsF phase diagrams have been calculated by
minimization of the Gibbs energy with respect to the propor-
tion of the solid and liquid phases. This procedure yields the
solidus and liquidus curves and it is carried out using the
43
FACTSAGE software. The resulting phase diagram based on
the BW approximation is shown in Fig. 8 and that based on
the Guggenheim approximation is shown in Fig. 9. Both
phase diagrams are characterized by a minimum on the liqui-
dus curve found at similar temperatures and compositions. In
the case of the BW approximation the minimum is at
T = 947 K and xRbF = 0.274, whereas with the Guggenheim
approximation it is located at T = 951 K and xRbF = 0.250.
However, from a comparison of both figures it is evident that
in the case of the BW approximation a miscibility gap with a
critical temperature T = 368 K at xRbF = 0.575 appears,
whereas in the case of the Guggenheim model no immiscible
region in the solid state has been found. This is due to the
fact that in the former case the excess Gibbs energy is as-
sumed to be temperature independent and thus has always
the same value for a given composition. Consequently at low
temperatures where the ideal mixing term 共xRbFRT ln xRbF
+ xCsFRT ln xCsF兲 has relatively small influence, the positive
excess term is dominant and phase separation occurs. This is
not the case with the Guggenheim approximation which does
evolve with temperature. Hence the positive excess param-
eters are small at low temperatures while they increase with
increasing temperature and reproduce nearly the same mini-
mum on the liquidus curve as found from the assessment
based on the BW model.
Due to the lack of experimental data at low temperature
it is very difficult to judge which phase diagram is closer to
reality; however, since the Rb+ and Cs+ cations are highly
ionic and their sizes are very similar it is not expected that a
J. Chem. Phys. 130, 134716 共2009兲
FIG. 9. RbF–CsF phase diagram assessed using the Guggenheim model
共solid lines兲 and comparison to the solidus and liquidus data measured in
this study by DSC. 共쎲兲 Solidus data determined from the heating run; 共䊊兲
liquidus data determined from the cooling curves.
miscibility gap would appear in the solid state. Thus the
phase diagram optimized based on the data of the Guggen-
heim approximation as shown in Fig. 9 is to be preferred.
VI. CONCLUSIONS
The RbF–CsF pseudobinary phase diagram we have
measured using DSC is in very good agreement with the one
we have calculated using our multiscale approach involving
exclusively, for the smaller scale of description, DFT. The
calculated excess enthalpy reveals an almost ideal behavior
of the liquid phase whereas a small but significant positive
excess enthalpy occurs in the solid phase. This is a common
feature in systems with extended solubilities and it illustrates
the similarity of the end members. As for the solid solution
the calculation reveals that some short-range order is present,
quantified by the difference between the zero-order BW and
the first order Guggenheim approximations, and that its ne-
glect would lead to the appearance of a miscibility gap in the
solid phase below 367 K in the calculated phase diagram.
Beyond this would-be miscibility gap, which is not amenable
to experimental confirmation, the short-range order does not
have any significant influence on the melting transition
curves.
From a more fundamental point of view we have dem-
onstrated the applicability of a DFT-based approach to cal-
culate the relatively small excess free energies that determine
the phase diagram of a pseudobinary fluoride system. The
SAM proves to be both flexible enough to accommodate the
subtlety of interactions in the solid phase and quite efficient
in terms of the amount of input DFT data it requires. Accord-
ing to the literature the SAM has been used extensively in
the past to rationalize measured macroscopic thermodynamic
data. We show here that SAM and DFT make up an ideal
combination since they are both based on configurational
energies—which are not accessible experimentally. As for
DFT-based MD with polarizable ions, the technique used
here had already been validated in several instances and the
present work shows that it has become a routine and reliable
technique.
134716-13 Investigation of the RbF–CsF phase diagram J. Chem. Phys. 130, 134716 共2009兲

22
ACKNOWLEDGMENTS J. C. Mathieu, F. Durand, and E. Bonnier, J. Chim. Phys. Phys.-Chim.
Biol. 62, 1297 共1965兲.
23
O.B. acknowledges the European Commission for sup- It can be shown that the configurational energies are only sensitive to the
port given in the frame of the program “Training and Mobil- sum ␣1 + ␤1. We may thus arbitrarily set ␣1 = 0 but we keep it in the
algebraic formulas to preserve the symmetry.
ity of Researchers.” The ACTINET network for actinide sci- 24
The proof in Ref. 5 is a bit difficult to follow but can also be obtained
ences, supported by the European Commission in the 6th simply as follows. Considering ā in Eq. 共38兲 as an independent 共set of兲
Framework Programme 共Contract No. FP6-508836兲, is variable共s兲 and writing down formally the differential of F in variables ā
thanked for the fellowship provided in order to perform this and T, it is easy to recognize that due to the Guggenheim approximation
leading to Eq. 共36兲 the term 共⳵F / ⳵ā兲dā in dF includes the logarithm of
study 共JRP 06-020兲. M.S. acknowledges financial support of
the left side of Eq. 共35兲, so that this part of the differential vanishes
PCR-RSF 共Programme concerté de recherches—Réacteur à identically. Therefore the entropy S = −⳵F / ⳵T can indeed be computed
sels fondus兲. from Eq. 共38兲 as if ā did not depend on T.
25
The polynomial denoted S共␭兲 by Mathieu et al. 共see Ref. 5兲 is denoted
1
A. L. Zherebtsov and V. V. Ignatiev, “Experimental mock-up of here by Z共␭兲 to avoid confusion with entropy.
26
accelerator-based facility for transmutation of radioactive waste and con- SCILAB, an open source platform for numerical computation 共see http://
version of military plutonium,” Russsian Research Center Technical Re- www.scilab.org兲.
27
port No. 1606, 2006兲. As an example, let us consider a configuration with NRb = NCs and M 12
2
O. Beneš and R. J. M. Konings, J. Nucl. Mater. 377, 449 共2008兲. = NRb. Then on the model side Eqs. 共14兲 and 共15兲 are compatible with
3
O. Beneš and R. J. M. Konings, CALPHAD: Comput. Coupling Phase P12 = NCs, but on the other side reality is not: Some Cs cations, due to the
Diagrams Thermochem. 32, 121 共2008兲. fact that they are near neighbors of an Rb cation fully surrounded by Cs,
4
R. M. Martin, Electronic Structure Basic Theory and Practical Methods have at least three Cs near neighbors—which implies P3 ⫽ 0. This leads
共Cambridge University Press, Cambridge, 2004兲. to the contradiction P3 + P12 ⬎ NCs.
28
5
J. C. Mathieu, F. Durand, and E. Bonnier, J. Chim. Phys. Phys.-Chim. J. M. Cowley, Phys. Rev. 138, A1384 共1965兲.
Biol. 62, 1289 共1965兲.
29
When SRO= 0, i.e., ␤2 = 0, there remains a slight short-range order ac-
6
W. L. Bragg and E. J. Williams, Proc. R. Soc. London, Ser. A 145, 699 cording to ␹. The explanation of this paradox is that ␣2 is then slightly
共1934兲. different from zero and in fact both ␣2 and ␤2 play a role in short-range
7
E. A. Guggenheim, Mixtures 共Clarendon, Oxford, 1952兲. order. ␣2 and ␤2 would cancel simultaneously only if the solution was
8 symmetric but this is not quite the case here, and on the other hand if
M. Salanne, C. Simon, P. Turq, R. J. Heaton, and P. A. Madden, J. Phys.
Chem. B 110, 11461 共2006兲. there was a large difference between ␣2 and ␤2 we would even have to
9
M. Salanne, C. Simon, P. Turq, and P. A. Madden, J. Phys. Chem. B 112, change the definition of SRO to reliably account for the short-range
1177 共2008兲. order.
30
10
M. Salanne, C. Simon, P. Turq, and P. A. Madden, J. Fluorine Chem. P. A. Madden and M. Wilson, Chem. Soc. Rev. 25, 339 共1996兲.
31
130, 38 共2009兲. K. T. Tang and J. P. Toennies, J. Chem. Phys. 80, 3726 共1984兲.
32
11
R. G. Samuseva and V. E. Plyushchev, Russ. J. Inorg. Chem. 10, 688 A. J. Stone, Theory of Intermolecular Forces 共Oxford University Press,
共1965兲. Oxford, 1996兲.
12 33
O. Beneš, R. J. M. Konings, “A DSC study of the KNO3-NaNO3 system A. Aguado, L. Bernasconi, S. Jahn, and P. A. Madden, Faraday Discuss.
using a new encapsulation technique,” Thermochimica Acta 共submitted兲. 124, 171 共2003兲.
13 34
S. J. Clark, M. D. Segall, C. J. Pickard, P. J. Hasnip, M. J. Probert, K. R. J. Heaton, R. Brookes, P. A. Madden, M. Salanne, C. Simon, and P.
Refson, and M. C. Payne, Z. Kristallogr. 220, 567 共2005兲. Turq, J. Phys. Chem. B 110, 11454 共2006兲.
14 35
J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 G. J. Martyna, D. Tobias, and M. L. Klein, J. Chem. Phys. 101, 4177
共1996兲. 共1994兲.
15
H. J. Monkhorst and J. O. Pack, Phys. Rev. B 13, 5188 共1976兲. 36
J. Kieffer and C. A. Angell, J. Chem. Phys. 90, 4982 共1989兲.
16
MATERIALS STUDIO v. 4.3, Accelrys Software Inc. 共http:// 37
M. Salanne, C. Simon, P. Turq, and P. A. Madden, J. Phys.: Condens.
www.accelrys.com兲. Matter 20, 332101 共2008兲.
17
CYGWIN, a Linux-like environment for Windows 共see http://
38
M. Salanne, C. Simon, P. Turq, and P. A. Madden, J. Phys. Chem. B 111,
www.cygwin.com兲. 4678 共2007兲.
18
D. Vanderbilt, Phys. Rev. B 41, 7892 共1990兲. 39
A. D. Pelton, P. Chartrand, and G. Eriksson, Metall. Mater. Trans. A 32,
19
J. D. Cox, D. D. Wagman, and V. A. Medvedev, CODATA Key Values for 1409 共2001兲.
Thermodynamics 共Hemisphere, New York, 1989兲. 40
P. Chartrand and A. D. Pelton, Metall. Mater. Trans. A 32, 1397 共2001兲.
20
Inorganic Crystal Structure Database 共ICSD兲, Fachinformationszentrum 41
P. Chartrand and A. D. Pelton, Metall. Mater. Trans. A 32, 1385 共2001兲.
共FIZ兲, Karlsruhe. 42
M. W. Chase, Jr., J. Phys. Chem. Ref. Data Monogr. 9, 1 共1998兲.
21 43
V. Milman, B. Winkler, J. A. White, C. J. Pickard, M. C. Payne, E. V. FACTSAGE 5.3, software for thermochemical calculations 共see http://
Akhmatskaya, and R. H. Nobes, Int. J. Quantum Chem. 77, 895 共2000兲. www.factsage.com兲.