The Loewenstein Rule The Increase in Ele

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Phys Chem Minerals

DOI 10.1007/s00269-013-0611-7

ORIGINAL PAPER

The Loewenstein rule: the increase in electron kinetic energy


as the reason for instability of Al–O–Al linkage
in aluminosilicate zeolites
Alexander V. Larin

Received: 7 March 2013 / Accepted: 24 June 2013


 Springer-Verlag Berlin Heidelberg 2013

Abstract Problem of Al–O–Al linkage in aluminosilicate Introduction


materials or Al-avoidance is discussed for two zeolite
structures (phillipsite and brewsterite) exchanged with The Loewenstein rule states avoidance of Al–O–Al moie-
Mg2? cations. All models are fully optimized at periodic ties in aluminosilicate crystalline materials or Al-avoidance
Hartree–Fock and hybrid density functional theory levels (Loewenstein 1954) serving till now as a landmark for fine
(the CRYSTAL code). Their properties are then calculated inorganic synthesis (Gonzalez-Gallardo et al. 2011).
at the periodic level with the same basis sets. The reasons Another relevant conclusion about the zeolite composition
for the instability of the zeolite structures including the is formulated as Dempsey rule that confirms a minimum
Al–O–Al moieties are interpreted on the basis of cell number of the –Al–O–Si–O–Al– fragments in a zeolite
energy decomposition. This destabilization comes from an (Ramdas et al. 1981; Engelhardt et al. 1981; Melchior et al.
increase in kinetic energy if the Al–O(Mg)–Al moieties are 1982). A probability of the Al–O–Al appearance in the
present in zeolites. This effect is discussed in parallel with aluminosilicates could be evaluated on the basis of exper-
already known variational evidences in favor of dominat- imental 17O determination with MAS NMR technique. It
ing role of kinetic energy for stability of molecular was tentatively estimated less than 2.0 % (Pingel et al.
systems. 1998) in low-silica A zeolites where the populations of
other Si–O–Si and Si–O–Al species were estimated (Zhao
Keywords Aluminosilicates  Loewenstein rule  et al. 2001). The absence of the Al–O–Al linkage in other
Al-avoidance  Zeolite  Periodic density functional theory  low silica X (Pingel et al. 1998; Zhao et al. 2001), analcime
Periodic Hartree–Fock  Electron kinetic energy  (Zhao et al. 2001), as well as in high-silica ZSM-5 (Pingel
Aluminosilicate glass et al. 1998) zeolites was also confirmed by MAS NMR in
opposite to low-silica glasses (Lee et al. 2000). Appearance
of Al–O–Al moieties for the latter was interpreted as a
consequence of meta-stable disorder in the course of high-
temperature glass formation (Zhao et al. 2001). Other
arguments in favor of chemical kinetic factors were for-
Electronic supplementary material The online version of this mulated by (Pelmenschikov et al. 1992). The authors
article (doi:10.1007/s00269-013-0611-7) contains supplementary evaluated that the possibility of decomposition is higher for
material, which is available to authorized users. the structure with two closest Al atoms. The values were
obtained on the basis of the HF/STO-3G energy estima-
A. V. Larin (&)
Department of Chemistry, Moscow State University, tions for framework ‘‘elements’’ such as (HO)3Si–O–
Leninskie Gory, 119991 Moscow, Russia Al(H2O)(OH)3 used to simulate reactions in zeolites. The
e-mail: nasgo@yandex.ru authors emphasized that ‘‘chemical’’ attraction is the rea-
son for the easiness of Al extraction from a framework and
A. V. Larin
OOO Plasmonika, Ural Building, 100, Novaya street, Skolkovo not the electrostatic repulsion between two Al atoms in
startup Village, Odintsovsky district, 143025 Moscow, Russia framework as proposed earlier (Dempsey 1974; Mikovsky

123
Phys Chem Minerals

and Marshall 1976; Mikovsky and Marshall 1979). Alter- Table 1 Lattice parameters a, b, c, b (Å, degrees) for MgPHI and
natively, Catlow et al. (Catlow et al. 1996) approved the MgBRE zeolites both possessing P21/m symmetry group and opti-
mized at the PDFT/B3LYP level with 85-11G*(Mg, Al)/88-31G*(Si)/
domination of thermodynamic factors also at the level of
6-31G*(O) level
isolated cluster calculations. Any microscopic reason
behind this energy gain for aluminosilicate zeolites obey- Type a b c b
ing to the Loewenstein rule (Loewenstein types or LT) was Loewenstein Mg-models
not discussed to our best knowledge. Resuming, one should PHI32 9.9413 12.1910 8.1126 128.91
notice that relative stabilities of the LT group have not PHI21 10.3958 12.9828 8.6942 128.01
been yet compared using ab initio or DFT computation PHI15 10.3965 12.9873 8.6868 127.93
with periodic boundary conditions to that of non-Loe- PHI11 10.3953 12.9812 8.6347 127.40
wenstein (NL) zeolite types containing Al–O–Al moieties. Non-Loewenstein Mg-models
One of the problems for theoretical explanation of the
PHI10 10.1619 13.8497 8.8163 128.02
Al-avoidance in zeolites is related to the absence of any
PHI26 10.1903 13.7437 8.8212 128.52
zeolite for which Al-avoidance is violated and experi-
PHI12 10.6543 13.0677 8.0525 126.87
mentally approved. Though the Al–O–Al linkage occurs in
Theorya 9.861 14.093 8.667 124.68
aluminates and aluminosilicate glasses, it is difficult to
Experimentb 9.865 14.300 8.668 124.20
propose an appropriate scheme of two similar computa-
9.879 14.139 8.693 124.81
tions, one, for example, with an aluminate and another with
Loewenstein Mg-models
a zeolite, which could demonstrate the disfavor situation
BRE6 6.8172 15.9185 7.7110 94.98
for the Al–O–Al in the zeolite.
BRE4 6.9231 15.7170 7.35823 93.41
In this work, we develop such scheme to compare
BRE5 6.6651 16.0025 7.56017 94.71
properly the stabilities of hypothetic NL zeolites with the
Non-Loewenstein Mg-models
Al–O–Al groups with those of real zeolites. For this, it is
suggested to consider at quantum chemical level the most BRE49 6.8800 17.6344 7.7662 95.40
favored electrostatic situation for stabilization of the BRE51 6.9126 18.1914 7.6348 96.83
Al–O–Al linkage by divalent compensating cation. The BRE50 6.8481 17.0317 7.5781 94.04
advantage of divalent cation as compared to monovalent Experimentc 6.793 17.573 7.759 94.54
one is connected with the absence of Coulomb repulsion a
All-siliceous model optimized with empirical force field in (Dovesi
between two monovalent cations that both should com- et al. 2006)
b
pensate the charge of the Al–O–Al moiety and should be Si/Al = 2.019 and 3.0 in the first K2Ca1.648(H2O)12Si10.7Al5.3O32
closely located one to another. Indeed, the higher fraction and second Ba2Ca0.6(H2O)12Si12Al4O32 cases, respectively, from
(Rinaldi et al. 1974)
of the Al–O–Al moieties in low silicate Ca-glasses rather c
Si/Al = 3.0 for Ba0.52Sr1.48(H2O)10Si12Al4OÅ from (Schlenker
than in Na-glasses (Cormier et al. 2003) can be interpreted et al. 1977)
as stabilizing effect of divalent cation. One of the expla-
nations of the preference of divalent cation is related to
decrease in the enthalpy of Si–Al ordering (Lee et al. exchange) with the 85-11G*(Mg, Al)/88-31G*(Si)/
2000). 6-31G*(O) basis set involving cell parameters (Table 1) with
periodic boundary conditions. For all cases, the thresholds
Computational details for the electronic integral calculations were fixed to 10-5
for the overlap Coulomb, penetration Coulomb, and over-
Two frameworks of phillipsite and brewsterite zeolites lap exchange and to 10-7 and 10-10 for the pseudo-overlap
(both possessing P21/m spatial symmetry group) have been exchange for different summation indexes. SCF FMIXING
chosen. Natural K/Ca- and Ca/Ba-phillipsites (Rinaldi et al. convergence parameter (30 %) was applied. B3LYP was
1974) and Sr/Ba-brewsterites (Schlenker et al. 1977) occur shown to be accurate for evaluation of relative stabilities of
with Si/Al modulus of 2.019 and 3.0, respectively all-siliceous zeolites (Catti et al. 2000). The B3LYP
(Table 1). The models which include Al–O–Al moieties or functional provides very similar electron density distribu-
non-Loewenstein (NL) zeolite types have been obtained by tion represented via atomic multipoles for aluminophos-
replacing heavier cations (K, Ca, Sr, Ba) with the smaller phates relative to PW91 functional, while minor
Mg ones, Si with Al atoms at T(2) or T(4) sites in PHI and differences have been observed versus the atomic multi-
T(3) site in BRE, respectively. poles calculated with PBE (Larin et al. 2005). Direct
Geometry optimization war performed using CRYS- scaling of wavefunction was performed to optimize the
TAL06 (Dovesi et al. 2006) code and hybrid density basis set in the similar way with a common a parameter for
functional theory (DFT)/B3LYP (30 % of Hartree–Fock both LT and NL zeolites. Relative energy difference

123
Phys Chem Minerals

between LT and NL models has been compared upon the regarding that hydrated zeolites with heavier cations (K,
scaling. Vibrational frequencies were calculated using the Ca, Sr, Ba) were applied for X-ray experiments (Rinaldi
finite difference method as implemented in the CRYS- et al. 1974; Schlenker et al. 1977). The agreement with
TAL06 code. Small displacements (0.00567 Å) of all experimental cell parameters is better for BRE framework.
atoms were used to estimate the numerical Hessian matrix. Surprisingly, the most accurate optimization of experi-
Analogous optimizations of the same BRE and PHI mental PHI cell sizes was earlier reached with Catlow force
models at the periodic Hartree–Fock (PHF) level have been field for all-siliceous PHI model without any cations (Larin
performed applying minimal STO-3G basis set and keeping et al. 2008). The geometry of the Mg–OAl2 groups and of
the tolerance criterions. The stabilities are then discussed in the others is given in Tables 2, 3. The Mg–O bond lengths
the same terms of the ‘‘structure–energy’’ components as (Table 2) are in the conventional range obtained with
above. Due to totally coherent results at this theory level, cluster calculations (Sykes and Kubicki 1996). The Mg
we have presented them in Supplementary Materials. cation coordination is described via all the nearest O types
in the right column of Table 2. First, in the right column of
Table 2, there are n Mg–O distances for n-coordinated Mg.
Results Then, there is a symbol containing two parts, i.e., the
coordination number n of the Mg cation and n symbols of
The geometry of the optimized LT and NL models the nearest i-oxygen atoms in the increasing |Mg–Oi| order.
and virial ratio The O symbols are partitioned into three types denoted by
plus, minus, or star (Fig. 2a–c), which relate to the Mg–
The cell parameters of the optimized LT and NL (Fig. 1) OSiAl, Mg–OSi2, or Mg–OAl2 species, respectively. In
models are shown in Table 1. This moderate difference in order to facilitate the notations in Table 2, the key com-
the cell sizes of the Mg-forms seems to be predicted binations used, i.e., Al–O–Si as ‘‘?,’’ Si–O–Si as ‘‘-,’’
Fig. 1 Two non-Loewenstein
(NL) MgPHI10 (a) and
MgBRE51 (b) type zeolites.
The color code is as follows:
O in red, Si in yellow, Al in
violet, Mg in green

123
Phys Chem Minerals

Table 2 The MgPHI and MgBRE models, positions of the Si atom 85-11G*(Mg, Al)/88-31G*(Si)/6-31G*(O) (shift by DU1 = 7253 a.u.),
replaced by Al in the initial X-ray diffraction models (Rinaldi et al. Upot/Ukin ratio, average Mg–O distance Rav (Å), Mg–O distances, and
1974; Schlenker et al. 1977), framework density (FD, g 9 cm-3), coordination type (Å)
energies U (a.u.) optimized at the PDFT/B3LYP level with
Type Al position FD |U ? DU1| -Upot/Ukin Rav Mg–O distances and coordination typea

Loewenstein (LT) models


PHI32 Al(1) 2.178 0.554083 2.003716 2.097 2 9 2.044, 2.111, 2 9 2.144, 5??-??
PHI21 Al(3) 1.802 0.561645 2.003651 2.048 2 9 2.039, 2 9 2.057, 4????
PHI15 Al(3) 1.801 0.561429 2.003651 2.048 2 9 2.039, 2 9 2.058, 4????
PHI11 Al(3) 1.800 0.565034 2.003637 2.031 2 9 1.979, 2 9 2.082, 4????
Non-Loewenstein (NL) models
PHI10 Al(4) 1.705 0.516355 2.003630 1.922 1.913, 2 9 1.927, 3*??
PHI26 Al(4) 1.724 0.516773 2.003632 1.929 1.915, 2 9 1.936, 3*??
PHI12b Al(2) 1.858 0.365101 2.003666 2.051 2 9 2.042, 2 9 2.061, 4----
Loewenstein (LT) models
BRE6 Al(1) 2.025 0.580203 2.003650 1.999 2 9 2.013, 2 9 2.038, 4????
BRE5 Al(1) 2.073 0.521973 2.003686 2.083 2 9 2.063, 2 9 2.102, 4????
BRE4 Al(1) 2.085 0.535645 2.003674 1.974 2 9 1.951, 2.021, 3??-
Non-Loewenstein (NL) models
BRE49 Al(3) 1.776 0.508078 2.003624 1.904 2 9 1.895, 1.922, 3??*
BRE50 Al(3) 1.926 0.497070 2.003645 1.925 1.906, 2 9 1.935, 3*??
BRE51 Al(3) 1.748 0.521484 2.003628 1.888 2 9 1.869, 1.927, 3??*
a
for non-Loewenstein type (NL) models see please two Al–O and one Me–OAl2 distances and the Al–O–Al angle in Table 3; the notation
‘‘?’’corresponds to the ‘‘Mg cation–oxygen atom of the Si–O–Al type,’’ while the ‘‘-’’and ‘‘*’’ are related to Mg cation coordinated with oxygen
atom in the Si–O–Si and Al–O–Al species, respectively
b
Mg is not coordinated to Al–O*–Al species, |Mg…O*| = 5.024 Å

Al–O–Al as ‘‘*,’’ are shown in Fig. 2. The 3??* (Fig. 2e) The zeolite ionicity can be crudely evaluated via the
or 3*?? types of the Mg-coordination dominate in the NL calculated Mulliken O charges. Additionally, regarding the
group. The ‘‘3??*’’ type denotes that two closest O1 and close O charges in any O?, O-, or O* positions (Fig. 2) in
O2 atoms near Mg cation are of the Si–(Mg)Oi–Al type LT and NL cases, no essentially higher Coulomb interac-
(Fig. 2a), i = 1, 2, and only the third nearest O3 atom is of tions could be predicted for NL systems as compared to
the Al–(Mg)O3–Al type (Fig. 2c). Respective distances those in LT ones. The Mulliken O charges for coordinated
satisfy the sequence |Mg–O1| \ |Mg–O2| \ |Mg–O3|, and Al–O(Mg)–Si, Al–O(Mg)–Al(Mg) and non-coordinated
the next nearest O4 atom is located outside radius of 3 Å Al–O–Si, Si–O–Si types are pretty close, being -1.153,
from the Mg atom. The ratio between the Mg–OAl2 and -1.174, -1.036, and -1.004 e, respectively, in the NL
Mg–OSiAl bond lengths in the 3??* or 3*?? groups type (BRE51). The O charges in the LT zeolite (PHI32) are
depends on each PHI or BRE model. Main geometric dif- nearly the same for coordinated Al–O(Mg)–Si, Si–O(Mg)–
ferences between the LT and NL models are as follows: Si, and non-coordinated Al–O–Si, Si–O–Si moieties, being
(a) the shorter average Mg–O bond length Rav = -1.153, -1.146, -1.029, and -1.011 e. The absolute O
(Ri=1,n|Mg–Oi|)/n and (b) the smaller coordination number charges in the Si–O–Si or Al–O–Si species thus grow
n in the NL case (n = 3) than that in the LT ones (n = 4 or similarly as much as by 0.1 e upon Mg-coordination.
5). The Rav decreases due to shortening of all the Mg–O Important consequence of the interaction between
distances including the O?, O-, and O* types. The Rav divalent cation and the T–O–T’ groups, T or T’ = Al or Si,
difference between LT and NL models is similar to the is the correlation between the Rav and the cell electron
difference of 0.09 Å between the Mg ionic radii corre- kinetic energy (Fig. 3a) and between the Rav and the Upot/
sponding to n = 4 and 5 (Larin et al. 2005). As the zeolites Ukin ratio (Fig. 3b, S1). The Upot/Ukin ratio or virial ratio
are semi-ionic frameworks, the idea of ionic radii is rele- (VR) should be equal to -2 in equilibrium system as
vant here so that the smaller NL coordination number 3 is determined by virial theorem. Respective energy should be
partly responsible for their shorter bond lengths. Unfortu- calculated at the same computational level, which is
nately, no data were shown in (Shannon 1976) for Mg if applied to fully optimize unit cell geometry. As demon-
n = 3. strated (Figs. 3b, S1), the VR for both LT and NL remains

123
Phys Chem Minerals

-2.003624, respectively. But this explanation due to


higher Rav value is not totally justified because the PHI26
model of the NL type has larger Rav of 1.929 Å and smaller
(in absolute value) VR of -2.003632. Anyway, the BRE50
is much closer to the group of the NL points (open symbols
in left upper corner in Fig. 3a) with higher kinetic energy
than strongly destabilized PHI12 model.
The six digits used above (Table 2) in the VR values
could be questioned. Do our methods allow it? In order to
answer to the question, we have transformed the reliable
accuracy (1 kcal/mol) of the CRYSTAL code to the VR
scale. Finally, this accuracy depends on the basis set, but
we have tested the basis set for MgPHI as converged one
relative to further improvement (Larin et al. 2008), so that
such error of 1 kcal/mol is rather overestimated. We accept
this error for both Upot and Ukin energies. The VR value is
the potential energy divided by the kinetic energy. For
Fig. 2 The types of O atoms: a Si–O–Al or ‘‘?,’’ b Si–O–Si or ‘‘-,’’ transformation of the energy error of 1 kcal/mol to the
c Al–O–Al or ‘‘*,’’ and two examples of the Mg connections to the scale of VR, we have considered the typical LT model
nearest O atoms d 4???? in MgBRE6 and e 3??* in MgBRE51. PHI11 with kinetic energy of 7227.2747 a.e. (Fig. 3a).
The color code is as follows: O in red, Si in yellow, Al in violet, Mg in
green Then, we have evaluated the accuracy 1 kcal/
mol = 1.594 9 10-3 a.u. in the VR scale by dividing it by
Ukin. The accuracy in the scale of VR is 1.594 9 10-3/
close to -2. The potential/kinetic ratio is, however, sys- 7227.2747 = 2.2 9 10-7. So, the variation in seventh digit
tematically lower in absolute value for NL type models in the VR value corresponds to the accuracy 1 kcal/mol
(Figs. 3b, S1). The only PHI12 model with the higher (in allowed by the CRYSTAL code. The usual differences in
absolute value) VR ratio of -2.0036661 (Table 2) the VR values between LT and NL models with B3LYP
throughout all the NL group deviates from this trend. (Table 2) range from 5 9 10-6 to 8 9 10-5 (omitting
However, it does not contain the Mg–OAl2 group, which is much less stable PHI12 model) for the PHI zeolites and
typical for all ‘‘stable’’ NL zeolites, because of the remote from 5 9 10-6 to 6 9 10-5 for the BRE ones. Both values
Mg location at 5.024 Å far from Al–O–Al oxygen are larger than the accuracy of 2.2 9 10-7 (dimension-
(Table 3). As a consequence of this long Mg–O* distance, less) = 1 kcal/mol as presented in the same VR scale.
PHI12 is less stable (Fig. 3c) by around 95 kcal/mol Hence, these small VR differences between LT and NL
compared with the other NL zeolites (PHI10 and PHI26). models are valid.
Hence, the PHI12 model is not a typical NL system, and Our result can signify that together with the energy gain
respective point should not be taken into account (open owing to higher potential energy between Mg and O*
circle with the highest Rav = 2.051 Å throughout all the atoms for NL systems, the Al–O(Mg)–Al groups lead to
NL models in Fig. 3a–c). higher average electron kinetic energy per cell. The strong
Analogous less stable NL model BRE50 also exists in Mg–O interaction appears due to the shorter Mg–O dis-
the BRE group, i.e., destabilized as much by 15.3 kcal/mol tances with all oxygen types. Resuming these variations,
relative to the most stable NL type BRE51 model (Fig. 3c), the total energy leads to less stable NL zeolites (Fig. 3c).
with the largest (in absolute value) VR = -2.003645. The The higher relative stabilities of the LT than those of the
latter is comparable with that of the LT models, for NL models hold at PHF level (see Table S3 in Supple-
example BRE6 possesses VR = -2.003650 (Table 2). The mentary Materials). Hence, such behavior is not condi-
reason seems to be related to higher Rav (1.925 Å) in the tioned by the B3LYP functional selected herein or any
BRE50 case (Table 2). The two other NL BRE models other DFT functional. Earlier (Larin et al. 2005; Larin and
with average Mg–O distances of 1.888 and 1.904 Å possess Vercauteren 2001), we have compared electron density
the smaller (in absolute value) VR of -2.003628 and distributions and relative energies of H-form zeolite mod-
els using periodic boundary conditions at the PHF/STO-3G
1
level. In opposite to the bad experience with this basis set
The VR value shifts the energy scale requiring at least 6 digits after
for isolated cluster calculations, the coherent relative
comma. This 6th digit corresponds nearly to 10-3 a.u. or 0.63 kcal/
mol. For the Utot, the third digit after comma corresponds to kcal/mol. energies have been obtained for proton positions with two
The validity of the digits of VR value is discussed below in the text. (STO-3G and ps-21G*) basis sets. Different order of the

123
Phys Chem Minerals

Fig. 3 Dependencies between (a) Rav and Ukin, (b) -Upot/Ukin and models of MgPHI (circles) and MgBRE (triangles). Correlation
Rav, (c) Upot/Ukin and Utot calculated at the (a–c) PDFT/B3LYP/85- coefficients are shown for PHI (dashed line) and BRE (dotted line) for
11G*(Mg, Al)/88-31G*(Si)/6-31G*(O) level for Loewenstein (LT, both LT and NL models and for all the models together (solid line)
closed symbols) and non-Loewenstein (NL, open symbols) type

Table 3 The parameters of non-Loewenstein (NL) models in MgPHI between the LT and NL models do not change. The poorer
and MgBRE zeolites optimized at the PDFT/B3LYP level with STO-3G basis set results in the larger VR deviation up to
85-11G*(Mg, Al)/88-31G*(Si)/6-31G*(O) level: Al–O distance (Å), -2.01 for optimized models than that obtained at the
Al–O–Al angle (degree), and Mg-OAl2 (or Mg–O*) distances (Å)
B3LYP/85-11G*(Mg, Al)/88-31G*(Si)/6-31G*(O) level, but
Type |Al–O| Al–O–Al |Mg–O*| VR is also larger (in absolute value) for LT models than for NL
PHI10 2 9 1.746 141.3 1.913 ones (Table S3, Fig. S1). Hence, the kinetic energy also larger
PHI26 2 9 1.745 142.1 1.915 for the NL type models at the PHF/STO-3G level.
PHI12 2 9 1.693 122.4 5.024
BRE50 2 9 1.739 140.1 1.906
Direct scaling of atomic wavefunction
BRE49 2 9 1.743 157.4 1.922
The minor deviation of the VR value versus -2 for the
BRE51 2 9 1.754 165.1 1.927
optimized system shows nevertheless that the basis set
could be improved looking for a proper scaling factor for
relative energies between these basis sets was obtained in all the atomic orbitals. Formally, we should have exact
one HNAT case throughout the series of five small size equation VR = -2 at the energy minimum with respective
zeolites. That is why these relative stabilities could also be optimal value of scaling a parameter. We tested some grids
discussed with STO-3G. from 5 to 7 points nearby the initial minimum position
Herein, despite some variations in the PHF optimized (a = 1) to evaluate the optimal value of scaling a param-
geometries and cation coordinations, the main differences eter. First, parabolic approximations over 5 and 7 points

123
Phys Chem Minerals

Table 4 Energy values (a.u.) and VR using different scaling of wave (in absolute value). Our NL and LT models have inverse
functions, i.e., at the minimum of parabola fitted via 5 or 7 points, ratio with lower kinetic energy for more stable LT models
along with scaling parameter a for some LT (PHI11, PHI21) and NL
(Fig. 3c). This situation was not changed herein with the
(PHI26) models (see text for details)
scaling performed above. The conservation of the higher
Model -U/-VR stability of LT models versus NL ones after scaling
7 (a = 1.00123) 5 (a = 1.00103) approves the connection proposed herein between the sta-
bility and electron kinetic energy.
PHI11 7253.563653/2.00175 7253.564003/2.00206
PHI21 7253.565200/2.00177 7253.565586/2.00207
Evaluations of entropy differences between LT and NL
PHI26 7253.521828/2.00175 7253.522196/2.00204
models

Finally, the enthalpy gain DH is larger for LT than for NL


were applied with grid step Da = 0.01 and very high models as shown above (Table 2, Table S3), i.e., the
correlations for both cases (Table 4, five points were enthalpy differences between the most stable LT and NL
obtained from seven simply omitting two extreme points). models range from 30.3 kcal/mol (between PHI11 and
The grids of 5 points led to deeper energies at parabola PHI26) to 36.8 kcal/mol (between BRE6 and BRE51). In
minimum values being close nevertheless to a = 1.001 in order to evaluate the Gibbs energy variation DG = DH
both cases (5 or 7 points). So, we have selected a grid with - TDS for LT and NL models, one needs their entropy
smaller step Da = 0.005 around a = 1.001 for total values. Two main configurational and vibrational compo-
comparison of the cell energies after rescaling. Relative to nents can be separated. It is easy to calculate that the
the grid Da = 0.01 with the average parameter error configurational entropy term –TDS in the Gibbs energy
ranging between 0.3 and 0.4 %, the new grid results in the will be larger (in absolute value) for LT models. It is a
smaller parameter errors near 0.2 %. One has to note that consequence of the wider set of all possible types of Al
the total energy difference between 5- and 7-point grids is distributions in the LT zeolite lattice compared to the NL
minor, i.e., 0.2 kcal/mol between both energies corre- models. For the case of the BRE framework with 12 Si and
sponding to the parabola minimum. 4 Al positions, this configurational entropy difference can
In such a way, we have obtained the corrected energy be easily estimated taking into account spatial P21/m
minimum for some PHI models (Table 4) with slightly symmetry group, which simplifies the task. The BRE cell
different a-coordinates for each parabolic approximation. skeleton is partitioned into two symmetric parts each
Respective energy components do not satisfy to exact containing 6 Si and 2 Al positions (Fig. S2). So, it is
equation VR = -2 at the minimum being smaller in enough to distribute the Al atoms in the one of two sym-
absolute value but nearly -2.002 (Table 4) instead of metric parts. The LT degeneracy to distribute Al atoms
-2.0037 or -2.0036 without scaling (Table 2). The energies without Al–O–Al species is the probability for the first Al
of the corrected models calculated with scaled functions atom to occupy any of 8 sites and for the second Al one to
can only be compared at the same a-value, for example at occupy any of five remaining sites around because of one
two a values fitted over 5 (a = 1.00103) or 7 occupied and two forbidden sites, i.e., 8 9 C15 = 8 9 5!/
(a = 1.00123) points (Table 4). Respective energies 4! = 40. Analogous NL degeneracy is the probability for
demonstrate that the a-scaling procedure does not invert the first Al atom to occupy any of 8 sites and for the second
the relative stabilities of LT and NL models. The VR value Al any of two remaining sites around, i.e., or 8 9 C12 = 16.
after scaling can also be compared to that obtained with the Then, the probability ratio 40/16 = 2.5 is the consequence
minimal STO-3G basis set, i.e., between -2.0115 and of the P21/m symmetry and BRE structure. Then, this
-2.0104 (Table S3) at the PHF level. We suppose that the contribution to the Gibbs energy difference between LT
VR deviation from -2 can be deleted while optimizing the and NL models is T 9 DS = T 9 k9N0 9 ln(40/16) =
zeolite models with more accurate basis set, for example, T 9 1.38 9 10-23 9 6.02 9 1023 9 0.916 = 7.61 9 T J/
using iteratively the scaled functions. (K 9 mol), where k is Boltzmann constant, N0 is Avogadro
Ideal scaling procedure formally does not allow dis- number. At room temperature, respective term gives
cussing the variations of the kinetic energies between dif- 2.44 kJ/mol = 0.58 kcal/mol in favor of the LT model.
ferent zeolites, because if exactly VR = -2, then more Hence, the total Gibbs energy will be larger in absolute
stable zeolite model has higher kinetic energy. In such a value for the LT zeolites due to the configurational entropy.
situation, the points of more stable LT models in Fig. 3a Due to the same P21/m symmetry and content (Table 1),
should be located above the NL ones, while they are now the same estimations are valid for PHI zeolite. This small
below with the exception of the most stable PHI11 model. value cannot play significant role for the total Gibbs energy
The latter has the highest Ukin according to its highest Utot difference between LT and NL models compared to the

123
Phys Chem Minerals

enthalpy difference around 30 kcal/mol or 0.05 a.u. men- 2010), considering standard entropy of zeolitic water
tioned in the beginning of this part (Table 2). Evidently, S0H2O = 12.98 cal/mol/K (Vieillard 2010). This allows the
this configurational TDS difference will grow with Si/Al direct comparison just adding to the values for dehydrated
ratio. models (column S0 in Table 5) the total entropy of twelve
The small configurational entropy contribution to the water molecules or 155.8 cal/mol/K (column S0H in
Gibbs energy -TDS shows that vibrational entropy is the Table 5). In such a way, we obtained underestimated (in
main part of the total entropy. The latter can be directly absolute value) values for both LT and NL models by
obtained with the CRYSTAL code after full optimization around 15–18 %. The underestimation is partly explained
and frequency calculation. This permits to have the -TDS by the smaller volume of dehydrated theoretical models
difference between LT and NL models. The calculated whose contraction around some percent upon dehydration
frequencies are presented for two typical LT (PHI11, is well known (Larin et al. 2003). This error of 15–18 %
PHI21) and one NL (PHI26) models in Table S4. One should be considered as satisfactory one, regarding: (1)
should note that three (PHI11) or four (PHI21 and PHI26) different cationic content in (Vieillard 2010), and herein,
imaginary frequencies are obtained. Three values for each (2) harmonic approximation applied to calculate the par-
model from 3i to 6.1i cm-1 (not shown in the Table S4) tition functions. In some sense, such harmonic approach
correspond to three translational modes of the center of applied to calculate entropy is in a contradiction with high
mass motions that can be neglected. Imaginary frequency temperature area where the entropy term can be more
of 41i cm-1 (PHI21) and 199i cm-1 (PHI26) for two important. At high temperatures, anharmonic potentials
models corresponds to the low-frequency framework should be used instead of the harmonic ones. However,
vibration (see 41i.mpg or 199i.mpg animation files in such estimation of partition functions for solids is not yet
Supplementary Materials) and formally should be recal- realized to our best knowledge requiring models that vary
culated using algorithm to improve the geometry. The with temperature for computed energy surfaces. Their
small real frequency value for respective vibration follows temperature boundaries have to be found regarding the
the large size of vibrating fragments. However, these mobilities of various zeolite groups and checking the
vibrations do not involve the geometry variation of the possible coupling between them. The difference TDS0
MgOAl2 groups that participate as whole components with obtained with harmonic approximation using the CRYS-
fixed Mg–O and Al–O distances (please, see animation TAL code between the LT and NL models does not exceed
files in Supplementary Materials). Hence, omitting one 2.5 kcal/mol so that we believe that the enthalpy is the
frequency in the whole series is not very important for the main term, which determines the favored Gibbs energy for
entropy difference between LT and NL groups (Table 5). LT species.
Additional reason to not re-calculate the vibrational profile
is related to the minor TDS0 variation (1.8 kcal/mol)
between PHI21 (-90.6 kcal/mol with all one imaginary Discussion
frequency) and PHI11 (-92.4 kcal/mol with all real fre-
quencies). That is why we did not convert two imaginary Dominating role of kinetic energy for chemical bond for-
frequencies to real values using the CRYSTAL facilities mation in molecules was first discussed by Rudenberg et al.
(they are omitted when calculating the entropy). in a series of publications beginning from 60th years
Calculated entropies S0 and the Gibbs energy part (Ruedenberg 1962; Feinberg and Ruedenberg 1971a, b;
-TDS0 at room temperature and standard pressure for LT Ruedenberg and Schmidt 2007). In his turn, Rudenberg
and NL models are shown in Table 5. They can be com- addressed to the Hellman’s model of bond formation in
pared to known experimental value for hydrated H2? ion owing to the decrease in kinetic energy upon
K1.6Na2.16Al3.76Si12.24O32 9 12H2O phillipsite (Vieillard electron orbital sharing (Hellmann 1933). The energy gain

Table 5 Cell volume V (Å3), standard entropy S0 for dehydrated values, and the errors of the TS0H calculation (in %) for LT (PHI11,
forms (in cal/mol/K), corrected entropy S0H = S0 + 12S0H2 O (S0H2 0 ¼ PHI21) and NL (PHI26) forms
12:98cal/mol/K (Vieillard 2010)) for hydrated forms, -TS0 (in kcal/mol)
Model Type V S0 S0H -TS0H Error

PHI11 LT 926.7 154.1 309.9 -92.4 15.9


PHI21 LT 924.6 148.0 303.8 -90.6 17.6
PHI26 NL 966.5 156.5 312.3 -93.1 15.3
Experiment 996.7 (Hemingway and Robie 1984) – 368.7 (Vieillard 2010) -109.9 (Vieillard 2010) –

123
Phys Chem Minerals

happens owing to a longer potential box for electrons in the energies relative to the similar values in the LT series; (3)
direction along with the molecular H2? axis (see Fig. 1 and the relative stabilities of LT and NL types do not invert
Table 1 in (Feinberg and Ruedenberg 1971a). The funda- upon scaling of wavefunction; (4) the virial ratio is close to
mental conclusion about the reason for bond formation was -2 as determined with virial theorem that was not yet
expanded from diatomic homonuclear (Feinberg and Rue- shown numerically for DFT in a periodic case as herein; (5)
denberg 1971a) to heteronuclear (Feinberg and Ruedenberg the decrease in kinetic energy was carefully analyzed to be
1971b) molecules regarding exceptionally variational the main reason for bond formation in isolated ions and
approach. All the examples in (Ruedenberg 1962; Feinberg molecules from atoms (Ruedenberg 1962; Feinberg and
and Ruedenberg 1971a, b; Ruedenberg and Schmidt 2007) Ruedenberg 1971a, b; Ruedenberg and Schmidt 2007;
were discussed for isolated molecules. The authors (Rue- Hellmann 1933). Analogous phenomenon of instability in
denberg 1962; Feinberg and Ruedenberg 1971a, b; Rue- periodic NL zeolite systems due to the increase in kinetic
denberg and Schmidt 2007) considered different trial energy is observed herein, if a special defect, i.e., the Mg–
wavefunctions and showed that decrease in kinetic energy OAl2 species with shortened inter-atomic distances, is
does not depend on the basis set. In this respect, we have presented in the NL zeolite. Finally, we conclude that bond
tested two basis sets and have obtained similar results with shortening produces the total energy loss for the NL
both of them. We have considered the case of periodic NL models. It could be a relevant problem for the formations
systems which cannot be formed due to the increase in of interstitial defects with short inter-nuclear distances in
kinetic energy, if a special defect, i.e., the Al–O(Mg)–Al solids.
species, is present in the zeolites. We suppose that the
Loewenstein rule (Loewenstein 1954) is explained at least
partly by this destabilization due to the increase in kinetic Supplementary Materials
energy in the semi-ionic systems. One of the reasons for the
increase is the shorter Mg–O bonds in the NL type zeolites. Analogous computations for the same models performed at
We suppose that the shorter bonds in solids can be a factor the periodic Hartree–Fock (PHF) level with minimal STO-
of its lower stability as could be deduced from results of 3G basis set are given in Supplementary Materials to
(Freysoldt et al. 2011). But, important contribution of illustrate the same relative behavior of the LT and NL
kinetic energy for an interstitial defect site in GaAs (Fig. 8 models at the pure Hartree–Fock level. They contain three
in ref. (Freysoldt et al. 2011) is assigned to the electrostatic Tables which are similar to the Tables 1–3 at PHF com-
energy due to self-consistent character of the calculation. putational level, one Table with calculated frequencies, and
From other theories that showed the growth of kinetic two Figures. Animation files 199-b.mpg and 41i.mpg
energy at the shorter bond length, one should mention demonstrate vibrational modes of PHI26 and PHI21 mod-
‘‘atoms in molecules (AIM)’’ (Espinosa et al. 1998). But els with the imaginary frequencies of 199i and 41i cm-1,
electron kinetic energy is related to the critical points of respectively, discussed in the part ‘‘Evaluations of entropy
hydrogen bonds so that the impact on the total energy of differences between LT and NL models’’.
whole system was not calculated.
Acknowledgments The author is indebted to Dr. V.I. Pupyshev for
the stimulating discussions and propositions, to Prof. D.P. Vercaut-
eren and Dr. A.A. Rybakov for permanent help, and to Mr. D.M.
Conclusions Kovtun for the references to the publications of K. Rudenberg et al.
and for his interest in our work. The author thanks the financial
The divalent (Mg) cationic forms were selected for the support of Ministry of Education and Science of Russian Federation
(Minobrnauka, GK No 07.514.11.4150). The author acknowledges
stabilization of the Al–O–Al groups in two different PHI the FUNDP, F.R.S.-FRFC (convention 2.4.617.07.F) for the use of the
and BRE zeolite frameworks. The energy of fully opti- Namur Interuniversity Scientific Computing Facility Centre (Bel-
mized MgPHI and MgBRE forms, which include Al–O–Al gium) and Computer Complex SKIF of Moscow State University
moieties (non-Loewenstein or NL types) or do not (tradi- ‘‘Lomonosov’’ and ‘‘Chebyshev’’ for computational time. RFFI is
deeply acknowledged for the grant 12-03-00749a.
tional Loewenstein type or LT), is calculated at the hybrid
density functional theory level with periodic boundary
conditions. The following conclusions can be formulated:
(1) the total energies of the LT types are higher in absolute References
values than that of the NL ones. This demonstrates a
thermodynamic nature of the lower stability of NL zeolites Catlow CRA, George AR, Freeman CM (1996) Chem Commun
11:1311
containing Al–O–Al moieties in agreement with the con- Catti M, Civalleri B, Ugliengo P (2000) J Phys Chem B 104:7259
clusions (Catlow et al. 1996); (2) the presence of the Mg– Cormier L, Ghaleb D, Neuville DR, Delaye J-M, Calas G (2003) J
OAl2 groups in the NL frameworks results in higher kinetic Non-Cryst Solids 332:255

123
Phys Chem Minerals

Dempsey E (1974) J Catal 33:497 Lee SK, Stebbins JF (2000) J Non-Cryst Solids 270:260
Dovesi R, Saunders VR, Roetti C, Orlando R, Zicovich-Wilson CM, Loewenstein W (1954) Am Mineral 39:92
Pascale F, Civalleri B, Doll K, Harrison NM, Bush IJ, D’Arco P, Melchior MT, Vaughan DEW, Jacobson AJ (1982) J Am Chem Soc
Llunell M (2006) CRYSTAL06 User’s Manual. University of 104:4859
Torino, Torino Mikovsky RJ, Marshall JF (1976) J Catal 96(44):170
Engelhardt G, Lohse U, Lippmaa E, Tarmak M, Mägi M (1981) Z Mikovsky RJ, Marshall JF, Burgess WP (1979) J Catal 58:489
Anorg Allg Chem 482:49 Pelmenschikov AG, Paukshtis EA, Edisherashvili MO, Zhidomirov
Espinosa E, Molins E, Lecomte C (1998) Chem Phys Lett 285:170 GM (1992) J Phys Chem 96:7052
Feinberg MJ, Ruedenberg K (1971a) J Chem Phys 54:1495 Pingel U-T, Amoureux J-P, Anupold T, Bauer F, Ernst H, Fernandez
Feinberg MJ, Ruedenberg K (1971b) J Chem Phys 55:5804 C, Freude D, Samoson A (1998) Chem Phys Lett 294:345
Freysoldt C, Neugebauer J, Van de Walle CG (2011) Phys Stat Sol B Ramdas S, Thomas JM, Klinowski J, Fyfe CA, Hartman JS (1981)
248:1067 Nature 292:228
Gonzalez-Gallardo S, Jancik V, Delgado-Robles AA, Moya-Cabrera Rinaldi R, Pluth JJ, Smith JV (1974) Acta Cryst B 30:2426
M (2011) Inorg Chem 50:4226 Ruedenberg K (1962) Rev Mod Phys 34:326
Hellmann H (1933) Z Phys 35:180 Ruedenberg K, Schmidt MW (2007) J Comput Chem 28:391
Hemingway BS, Robie RA (1984) Am Mineral 69:692 Schlenker JL, Pluth JJ, Smith JV (1977) Acta Cryst B 33:2907
Larin AV, Vercauteren DP (2001) J Mol Cat A 168:123 Shannon RD (1976) Acta Cryst A 32:751
Larin AV, Trubnikov DN, Vercauteren DP (2003) Int J Quantum Sykes D, Kubicki JD (1996) Am Mineral 81:265
Chem 92:71 Vieillard P (2010) Eur J Miner 22:823
Larin AV, Parbuzin VS, Vercauteren DP (2005) Int J Quantum Chem Zhao P, Neuhoff P, Stebbins JF (2001) Chem Phys Lett 344:325
101:807
Larin AV, Sakodynskaya IK, Trubnikov DN (2008) J Comput Chem
29:2344

123

You might also like