Daniil
Itkis
*ab,
Luigi
Cavallo
*c,
Lada V.
Yashina
*ba and
Yury
Minenkov
*ad
aN.N. Semenov Federal Research Center for Chemical Physics RAS, Kosygina Street 4, 119991 Moscow, Russia. E-mail: D.Itkis@chph.ras.ru; Yury.Minenkov@chph.ras.ru
bLomonosov Moscow State University, Leninskie Gory 1, Bld. 3, 119991 Moscow, Russia. E-mail: yashina@inorg.chem.msu.ru
cKAUST Catalysis Center (KCC), King Abdullah University of Science and Technology, Thuwal-23955-6900, Saudi Arabia. E-mail: Luigi.Cavallo@kaust.edu.sa
dJoint Institute for High Temperatures, Russian Academy of Sciences, 13-2 Izhorskaya Street, Moscow 125412, Russia
First published on 2nd July 2021
Gibbs free energies for Li+ solvation in water, methanol, acetonitrile, DMSO, dimethylacetamide, dimethoxyethane, dimethylformamide, gamma-butyrolactone, pyridine, and sulfolane have been calculated using the cluster-continuum quasichemical theory. With n independent solvent molecules S initial state forming the “monomer” thermodynamic cycle, Li+ solvation free energies are found to be on average 14 kcal mol−1 more positive compared to those from the “cluster” thermodynamic cycle where the initial state is the cluster Sn. We ascribe the inconsistency between the “monomer” and “cluster” cycles mainly to the incorrectly predicted solvation free energies of solvent clusters Sn from the SMD and CPCM continuum solvation models, which is in line with the earlier study of Bryantsev et al., J. Phys. Chem. B, 2008, 112, 9709–9719. When experimental-based solvation free energies of individual solvent molecules and solvent clusters are employed, the “monomer” and “cluster” cycles result in identical numbers. The best overall agreement with experimental-based “bulk” scale lithium cation solvation free energies was obtained for the “monomer” scale, and we recommend this set of values. We expect that further progress in the field is possible if (i) consensus on the accuracy of experimental reference values is achieved; (ii) the most recent continuum solvation models are properly parameterized for all solute–solvent combinations and become widely accessible for testing.
For rationalization and careful tuning of these industry-appreciated processes, the quantitative measure of the ionic solute–solvent interplay, the solvation Gibbs free energy of ions, ΔGsolv, is introduced. In contrast to the solvation thermodynamic functions of the neutral solutes accessible via a combination of the sublimation (vaporization) and solubility equilibrium constants from mass-spectrometric, calorimetric and chemical analysis studies, these characteristics of ionic solutes are challenging to obtain.7 This is due to the simultaneous presence of both ions and counter-ions in electrolyte solutions.
An indirect way to retrieve the experimental solvation Gibbs free energy of the ion in non-aqueous solvents suggests taking the sum over the following values: (a) Gibbs free energy of ion transfer from water to non-aqueous solution ΔGtr (W → S); (b) conventional hydration Gibbs free energy of the ion relative to that of the proton for which ΔGconvhyd is set to 0 at all temperatures; (c) absolute hydration Gibbs free energy of the proton, ΔGhyd (H+). The ΔGconvhyd of anions are precisely measured for equilibria involving strong acids and further utilized to extract these quantities for cations via the solubility products of soluble salts. The ΔGtr (W → S) and absolute ΔGhyd (H+) require a number of extra approximations to be extracted from the experimental measurements.
Marcus employed the tetraphenyl arsonium tetraphenyl borate (TATB) assumption8–10 that assigns the same solvation energies to large spherical ions TA+ and TB− to derive both the ΔGtr (W → S) and the absolute proton hydration free energy of −254.3 kcal mol−1 (TATB scale). As the soundness of the TATB assumption is questioned,11–15 the resulting estimates of Marcus might be insufficiently accurate.
Tissandier and co-workers16 utilized the so-called cluster pair approximation (CPA),16,17 claiming that the difference between the solvation free energy of water clusters containing either a single cation or anion disappears as the cluster grows up. This resulted in an absolute hydration free energy of the proton of −266.0 kcal mol−1. Kelly and co-workers18 using the CPA approximation arrived at −266.1 kcal mol−1 supporting the data of Tissandier et al.16 The difference of ca. 12 kcal mol−1 between the absolute proton hydration free energy derived with the TATB and CPA approximations is ascribed to the surface potential between gas and liquid. The surface potential exists in the CPA approximation, leading to the “real scale”, and is missing in the TATB assumption, leading to the “bulk scale”. The difference of a few kcal mol−1 between the Gibbs free energy of ion transfer from water to non-aqueous solutions derived by Marcus8 on the one hand and Kelly et al.14 on the other hand is much harder to explain, highlighting the large uncertainties in the available experimental data on solvation free energies of ions.
Approaching the solvation free energies from the theory side has also appeared to be troublesome. Despite the fact that explicit consideration of ion solvation through molecular dynamics simulations in conjunction with free energy methods is scientifically worth,20–22 this scheme is not a black-box and poses some technical and practical problems. In particular, for reliable solvation free energies, quite accurate long trajectories on large molecular ensembles are mandatory. The standard non-polarizable force fields are not accurate for this purpose in general as they cannot describe charge-transfer interactions properly.23,24 Custom-built force fields designed for specific solvent–solute combinations are an option; however, their speed is undermined by too long development time. The density functional theory (DFT) and non-empirical ab initio methods are prohibitively expensive and applicable to study only a limited combination of ionic solute–solvent combinations by a few scientific groups with an exceptional computational budget.
Solving the stationary Schrodinger equation for large gas-phase clusters of solvent molecules with and without the incorporated ionic solute is another option to derive the solvation free energy. Assuming that the DFT optimized geometries and frequencies are of good quality,25 very accurate yet computationally efficient local coupled cluster methods26,27 can be applied in a single-point (SP) fashion to study the clusters of a few thousand atoms. However, identification of either global or the most stable local minima of the solute–solvent or especially solvent clusters is an unworkable task.
The above-mentioned obstacles in explicit theoretical modeling of solvation resulted in the development of implicit solvation models in which the solvent is described as a dielectric medium. Different computational recipes for the calculation of electrostatic and non-electrostatic contributions to solvation open a series of dielectric continuum methods28–30 such as various implementations of the polarizable continuum model (PCM),31,32 conductor-like screening model (COSMO),33–35 solvation model based on density (SMD)36 and its reparameterization,37 generalized Born solvation model SM12,38 Poisson–Boltzmann solvation model,39 composite method for implicit representation of solvent (CMIRS),40–43 self-consistent continuum solvation (SCCS),44,45 easy solvation energy (ESE) approach,46 extended easy solvation estimation (xESE)47 and universal easy solvation evaluation (uESE),48 to name a few. Despite the fact that these models were shown to perform reasonably accurate for neutral solutes,29,49–55 large failures were recorded for solvation of ionic solutes.56–59
Finally, the performance of implicit solvation models can be further improved (albeit at substantially higher computational costs) via using the so-called hybrid approach which is often referred to as cluster-continuum quasichemical theory or mixed cluster/continuum model and detailed elsewhere.11,56,60–66 At first, the model introduces a few explicit solvent molecules around the solute, forming a cluster and mimicking the solute–solvent specific interactions. Then, the whole cluster is immersed in a dielectric continuum to imitate the non-specific solute–solvent interactions. As there are only a few solvent molecules, the conformational issues are not severe, and all essential local minima can be identified. Moreover, reliable ab initio methods can be employed to calculate the cluster formation energies.
Despite the fact that important findings were obtained with the hybrid approach of solvation,56–58,64,67–73 the results were shown to be considerably affected by the choice of (a) initial state of the solvent (separate molecules or clusters);60,61,66 (b) dielectric continuum model;61,66 (c) electronic structure theory method.62 In this work using the cluster/continuum framework, we systematically assess the influence of factors (a)–(c) on the resulting solvation free energies of the lithium cation in a number of protic (water, methanol) and aprotic (acetonitrile, DMSO, dimethylacetamide, dimethoxyethane, dimethylformamide, gamma-butyrolactone, pyridine, sulfolane) solvents. Keen attention was paid to the conformational issues. The obtained results are influential for further development of the cluster continuum solvation models as well as for unraveling the mechanisms of fundamental processes in electrochemical energy storage devices based on lithium.2,74,75
Then, according to the thermodynamic (“monomer”) cycle depicted in Scheme 1, the solvation free energy of an ion forming a strong bond with the surrounding solvent molecules is given via the following equation:
![]() | (1) |
In this formula, the term is the gas phase Gibbs free energy change of the A±(S)n cluster formation from n separate molecules of solvent S and ion A± with all components in 1 mol L−1 standard state. The 1 mol L−1
can be obtained from
1 atm standard state minus ΔG°→* correction of 1.89 kcal mol−1 (T = 298.15 K) multiplied n times. The terms
and
are solvation Gibbs free energies of the solute–solvent cluster and individual solvent molecule as calculated via the continuum approach. The last term, nRT
ln[S], contains the solvent density number [S] given in Table 1 for water (H2O), acetonitrile (MeCN), dimethyl sulfoxide (DMSO), dimethylacetamide (DMA), dimethoxyethane (DME), dimethylformamide (DMF), gamma-butyrolactone (GBL), methanol (MeOH), pyridine (Py), and sulfolane (TMS). The exact number of solvent molecules can be refined variationally.56
Solvent | [S] |
---|---|
H2O | 55.34 |
MeCN | 19.12 |
DMSO | 14.08 |
DMA | 10.76 |
DME | 9.64 |
DMF | 12.96 |
GBL | 185.73 |
MeOH | 24.72 |
Py | 12.42 |
TMS | 10.49 |
Opposite to the concept of Pliego and co-workers,56,60,62 Bryantsev and colleagues59,61 advocated for using the cluster of solvent molecules as the solvent resting state. Then, according to the thermodynamic (“cluster”) cycle given in Scheme 2, the solvation free energy of the A± ion can be expressed via the following equation:
![]() | (2) |
The 1 atm can be converted to 1 mol L−1 standard state
by subtracting from the former the ΔG°→* correction of 1.89 kcal mol−1 (T = 298.15 K).
The term stands for the solvation free energy of the cluster formed by n molecules of the solvent. It is claimed that the resulting solvation free energies of ions viaeqn (2) should converge with the growth of the number n of explicitly considered solvent molecules. The surface potential is not developed on the small solvent and solute–solvent clusters.19,76 Nevertheless, the solvation free energies predicted viaeqn (1) are systematically more positive and compared11,56,60,62,65 with the “bulk” scale values of Marcus,8 and the ΔGsolv calculated viaeqn (2) are more negative and gauged59,61 against the “real” scale18,19 quantities, presumably due to the surface potential.
For all Li+(S)n and (S)n clusters and each number of solute molecules n identified in Section 2.2, we performed the 20000-step-long non-periodic scalar relativistic PBE/λ1 molecular dynamic simulations with the isokinetic thermostat at T = 298.15 K. From each trajectory we identified 20 most stable, and energetically and structurally diverse structures. Each of 20 structures of Li+(S)n and (S)n clusters was optimized with the PBE-D3 method,79,80,83,84 def2-sv(p) basis set85,86 and with/without the SMD continuum solvation model as implemented in the ORCA 4.2 suite of programs.87 Default values were adopted for the self-consistent-field (SCF) and geometry optimization convergence criteria. Numerical integration of the exchange–correlation (XC) terms was performed with the “Grid5 FinalGrid6” option. The resolution-of-identity (RI) approximation was turned on for the sake of computational efficiency. The 20 gas-phase optimized geometries formed the SET_GAS set. Another 20 geometries optimized with the SMD solvation model formed the SET_SMD set. The spatial structures of the most energetically stable species from the SET_GAS and SET_SMD sets for each Li+(S)n and (S)n cluster were further refined according to the procedure outlined in Section 2.4.
Within each SET_SMD set, the most stable structure was re-optimized according to the identical procedure but with the SMD solvation model turned on. More details on the geometry optimization procedure are provided in the ESI.†
The ,
and
terms in eqn (1) and (2) represent the continuum solvation free energies of molecular clusters and individual solvent molecules calculated within the continuum model. In this formalism, the solvation free energy is obtained as the difference between the molecular electronic energies in the solvent and gas phase. As many continuum solvation models were parameterized using the DFT geometries, energies, and electron densities, we have employed the PBE-D3 method for the calculation of the
terms. Since the DFT calculations are known to be less demanding for the basis set size, smaller minimally augmented ma-def2-tzvp85,86,88 basis sets were employed.
In our first protocol, denoted as SMD_OPT, the terms are calculated as the difference between the PBE-D3 SMD SP energy on the continuum solvent media optimized geometry and the PBE-D3 SP energy on respective gas-phase optimized geometry. In the second approach referred to as SMD_SP, we obtain the
terms as the difference between the PBE-D3 SP energies with and without the SMD continuum solvation model on the same gas-phase geometry. If the
and
contributions to eqn (1) and (2) computed via the SMD_OPT and SMD_SP schemes turn out to be similar, then the expensive geometry optimization and conformational search in the continuum can be avoided.
Our third and fourth protocols termed CPCM_OPT and CPCM_SP were obtained by replacing the SMD solvation model with the conductor-like PCM (CPCM)100 in the SMD_OPT and SMD_SP protocols, respectively, with all other details remaining unmodified. Unless otherwise noted, the non-electrostatic interactions were not included in the CPCM SP energies (see the ESI† for details).
![]() | ||
Fig. 1 Li+ solvation Gibbs free energies from pure continuum solvation models against their experimental “bulk” scale counterparts in kcal mol−1. |
For both SMD_SP and CPCM_SP strategies, the Pearson correlation coefficient (ρ) between predicted and experimental data was close to zero (see Table S1, ESI†), suggesting no correlation. Indeed, according to continuum solvation models, the lithium cation solvation free energies in MeCN and DMA are practically identical, while experimentally, there is a gap of 12 kcal mol−1. The poor predictive power of stand-alone dielectric solvation models is confirmed by previous reports. In particular, Bryantsev and co-workers59 highlighted the inability of the Jaguar Poisson–Boltzmann (PB) solvation model to demonstrate the difference between the MeCN and DMSO lithium solvation free energies captured in the experiment. Noteworthily, their PB solvation free energies turned out to be ca. 20 and 45 kcal mol−1 more negative compared to our CPCM_SP and SMD_SP calculated values, respectively.
To unravel the mismatch between solvation free energies predicted from various continuum models and their experimental counterparts, let us recall the basics of the dielectric model formalism. In practically all continuum models the solvation free energy is split into its electrostatic and non-electrostatic
components:
![]() | (3) |
In the case of a spherical ion, within the dielectric continuum model formalism the term is equal to its counterpart from the Born101 formula:
![]() | (4) |
Using the Born formula (4) and the CPCM ORCA 4.2 Li radius of 1.404 Å, we obtained the values identical to the CPCM_SP
quantities (see Table S1, ESI†). To unravel the difference between the SMD_SP and CPCM_SP strategies, we first subtracted the non-electrostatic contribution from the SMD_SP solvation energies to arrive at SMD_SP (EL)
parts (see Table S1, ESI†). The SMD_SP (EL)
values turned out to be equal to the Born formula
quantities obtained for the SMD Li radius of 1.82 Å. This means that the substantial difference between the SMD_SP and CPCM_SP predicted solvation free energies is originated from different atomic radii used for calculation of
. The
term calculated within the SMD formalism for the Li+ ion in the solvents considered in this work turned out to be 1–2 kcal mol−1 only. Furthermore, it has to be noted that the thus obtained SMD
term is not accurate as it contains only the cavitation-like contribution proportional to the molecular surface tension σ[M]. Other important non-electrostatic terms, apparently responsible for the difference between the MeCN and DMSO solvents, are not calculated as atomic surface tension parameters are not available for lithium in the original SMD article.36 Finally, using the Born formula and the lithium radius of 1.226 Å mentioned by Bryantsev and co-workers59 as Jaguar PB default, we arrived at solvation free energies of Li+ that are within 1–2 kcal mol−1 from their Jaguar PB values. As for the choice of the Li+ radius, we find the CPCM default reasonable as it is fairly close to the isoelectronic He atom van-der-Waals radius of 1.40 Å recommended by Bondi.102 Hence, the disagreement between the CPCM_SP and experimental solvation free energies is due to non-electrostatic terms such as cavitation, dispersion, repulsion, and specific Li ← S interactions. The SMD Li+ radius corresponds to the neutral lithium atom102 and is too large. In contrast, the Jaguar PB default Li+ radius is too small.
All these findings indicate that pure continuum solvation models are quite promising for the calculation of the solvation free energies of ions if respective radii are carefully chosen and non-electrostatic terms are properly taken care of. It has to be underlined that reasonable solvation free energies with these models can be achieved at the cost of gas-phase SP calculation without tedious manual work as in the “hybrid” approach discussed vide infra.
For the “monomer” cycle, the average span between the minimum and maximum Li+ solvation free energies obtained with the continuum model strategies tested in this work amounts to 4.5 kcal mol−1. Similarly, for the “cluster” cycle the average difference between the predicted minimum and maximum solvation free energies of the Li cation is 4.3 kcal mol−1. These variations in solvation free energies due to differences in continuum models are substantially smaller compared to the variations in the stand-alone SMD and CPCM predicted solvation Gibbs free energies (see Fig. 1). Hence, the introduction of the explicit solvent molecules in the cluster-continuum approach evens out the difference between the pure continuum models. The average difference between the SMD_OPT and SMD_SP predicted Li+ solvation free energies amounts to 0.8 and 1.0 kcal mol−1 for the “monomer” and “cluster” cycles, respectively. Similarly, the average deviation between the CPCM_OPT and CPCM_SP predicted values is equal to 0.7 and 1.0 kcal mol−1 for the “monomer” and “cluster” cycles. These observations suggest that computationally quite expensive conformational search and geometry optimization with continuum solvation effects included can, in general, be avoided. Finally, the average difference between the SMD_OPT and CPCM_OPT lithium cation solvation free energies amounts to 3.9 and 3.3 kcal mol−1 for the “monomer” and “cluster” cycles, respectively. Similarly, the average difference between the SMD_SP and CPCM_SP values is equal to 3.5 and 3.2 kcal mol−1 for the “monomer” and “cluster” cycles. Analysis suggests that the average difference of 3–4 kcal mol−1 between the SMD and CPCM predictions originates from (i) the use of different atomic radii in the calculation of the electrostatic part of solvation free energies; (ii) missing non-electrostatic interactions in the default CPCM ORCA 4.2 implementation (see Section 2.5).
The lowest Li+ solvation free energies obtained within the “monomer” cycle suggest considering the number of explicit solvent molecules n = 3 for DME and n = 4 for all other solvents in this work. According to the “cluster” cycle, slightly more negative solvation free energies can be obtained for larger numbers n for some combinations of solvents and continuum solvation model strategies. However, the found energy drop of 1–2 kcal mol−1 was not substantial, concluding that the results are starting to converge for the numbers of explicit solvent molecules already found for the “monomer” cycle. For these reasons, we used the “monomer” cycle numbers n for the “cluster” cycle as well. Moreover, these numbers n were in agreement with the “cluster” cycle study of Bryantsev and co-workers.59 For the sake of consistency, we first compare our predicted “monomer” cycle solvation free energies with their literature counterparts obtained for the same cycle. Afterward, we proceed to the “cluster” model results. All individual components of eqn (1) and (2) needed for the calculation of solvation free energies via both cycles are listed in Tables S4 and S5 (ESI†).
Using the “monomer” cycle, Carvalho and Pliego62 reported a Li+ bulk solvation free energy in water of −116.1 kcal mol−1 for Li+(H2O)4 coordination. Our predicted are in reasonable agreement with this value, and all fall in the range between −115.8 and −117.5 kcal mol−1 depending on the particular solvation model. The same authors predict the
in acetonitrile of −120.6 kcal mol−1. This number is more negative than our calculated values ranging from −110.5 to −115.2 kcal mol−1. This mismatch arises from our predicted
being 5.9 kcal mol−1 more negative due to considering CH3-related vibrational modes in Li+(MeCN)4 as free rotors by Carvalho and Pliego. For Li+ solvation free energy in DMSO, the authors report
of −123.6 kcal mol−1. It is reasonably close to our CPCM_OPT and CPCM_SP predictions of −122.7 and −121.2 kcal mol−1, and more positive compared to our SMD_OPT and SMD_SP estimates of −128.8 and −127.8 kcal mol−1. Pliego and Miguel65 reported a free energy of
in MeOH of −118.1 kcal mol−1. This value is reasonably close to our predictions ranging from −115.6 to −118.5 kcal mol−1, depending on the continuum solvation model strategy. Apart from small deviations in the
term, the mismatch in Li+ solvation free energies in methanol arises from our calculated
being 3.8 kcal mol−1 more positive compared to the corresponding term in ref. 65.
Bryantsev and co-workers exploited the “cluster” cycle to obtain lithium cation solvation free energies in a number of solvents.59 For the MeCN solvent the authors suggest a lithium cation solvation free energy of −121.7 kcal mol−1. Our “cluster” model results are systematically more negative and range from −128.3 to −132.9 kcal mol−1. Similarly, for the Li+ solvation process in DMA in the form of Li+(DMA)4 the authors derived a Gibbs free energy of −131.5 kcal mol−1. According to our calculations, in DMA for n = 4 varies from −134.0 to −140.1 kcal mol−1 depending on the particular continuum solvation model. For Li+ solvation in DME in the form of Li+(DME)3, Bryantsev and co-workers calculated a solvation free energy of −128.3 kcal mol−1. Our predictions for this number of explicitly coordinated DME molecules range from −133.3 to −138.3 kcal mol−1. Finally, for Li+ solvation in DMSO in the form of Li+(DMSO)4, the authors find a Gibbs free energy of −133.1 kcal mol−1. Again, for the number n = 4 of explicitly coordinated DMSO molecules, we predict Li+ solvation Gibbs free energies ranging from −138.5 to −143.7 kcal mol−1. Interestingly, the best agreement between our predicted solvation free energies and their counterparts from the work of Bryantsev and co-workers was achieved for the CPCM_SP results as this continuum model results in the most positive values.
Unfortunately, Bryantsev and co-workers did not provide the Cartesian coordinates as well as electronic and solvation free energies of all individual species studied in their work, making it difficult to identify the origin of disagreement with our predicted values via the same cycle. We assume, nevertheless, that the main source of discrepancies is due to a contribution from the term. While we utilized the default radii from the continuum solvation model, the authors of ref. 59 scaled the radii to reproduce experimental solvation free energies of some neutral molecules.
Irrespective of the continuum model, the “monomer” cycle lithium cation solvation free energies obtained in this work are found to be on average 14 kcal mol−1 more positive compared to those from the “cluster” cycle. The other research groups report similar deviations between the values derived from these cycles.60,61,66 The origin of the difference between the solvation Gibbs free energies obtained from the “monomer” and “cluster” cycles is to be discussed in the next section.
![]() | (5) |
Replacing the term in eqn (2) with (5) should result in eqn (1). However, according to our calculations, due to inaccuracies in approximations described in Section 2, the predicted solvation free energies of solvent clusters (S)n from all considered continuum models deviate substantially from their counterparts calculated viaeqn (5) (see Fig. 3 and Table S6 (ESI†)). This turns eqn (5) into an inequation leading to documented discrepancies between the solvation Gibbs free energies of the lithium cation predicted via the “monomer” and “cluster” cycles.
![]() | ||
Fig. 3 Difference between ![]() |
There are only three reasons for eqn (5) values not matching their counterparts from continuum models:
(1) The gas-phase Gibbs free energy of solvent cluster formation from individual solvent molecules, , is inaccurate.
(2) The solvation free energies of individual solvent molecules in eqn (5) predicted with continuum models are erroneous.
(3) The solvation free energies of solvent clusters obtained via continuum models and utilized in eqn (1) and (2), and left-hand side of eqn (5) are unreliable.
Experience suggests that reaction energies/enthalpies from the DLPNO-CCSD(T)/aug-cc-p(wC)VTZ//PBE-D3/ma-def2-svp protocol should be quite accurate.103,104 Due to a wealth of low-lying frequencies in the solvent clusters, the entropy changes calculated for gas-phase reaction in Scheme 3 could be less reliable, yet leading to total errors in of only a few kcal mol−1. For these reasons, the predicted in this work
and utilized in eqn (5) are most likely correct with potential inaccuracies of 2–3 kcal mol−1. Hence, explanation (1) above can be ruled out.
Further, we derived the experimental Gibbs free energies of self-solvation of all solvents from the available105–111 pressure of saturated vapors:
![]() | (6) |
The experimental are compared with their continuum solvation model equivalents in Fig. 4 and Table S7 (ESI†). The MUE deviations are 0.7, 0.6, and 1.1 kcal mol−1 for SMD_OPT, SMD_SP, and CPCM_OPT/CPCM_SP protocols, respectively. The MSEs are found to be low and all in the range from −0.3 to −0.6 kcal mol−1, and cannot be responsible for large systematic deviations reported in Fig. 3 and Table S6 (ESI†). Hence, explanation (2) can be eliminated as well.
![]() | ||
Fig. 4 Self-solvation Gibbs free energies from continuum solvation models against their experimental counterparts. |
For these reasons, we conclude that the free energies from the considered continuum solvation models are erroneous and entirely accountable for the mismatch between the lithium ion solvation free energies calculated via the “monomer” and “cluster” cycles. To illustrate the degree of incorrectness, we compare in Fig. 5 and Table S8 (ESI†) the continuum models
with their eqn (5) counterparts obtained via a combination of the DLPNO-CCSD(T)
and experimental
from Fig. 4 and Table S7 (ESI†). These findings are in line with the previous report of Bryantsev et al.61 on the solvation of the proton and Cu2+ cation in water.
![]() | ||
Fig. 5
Eqn (5)
![]() ![]() ![]() |
In our opinion, a reasonable way to define a set of recommended values is to compare our predicted values with their experimental counterparts reported by Marcus for bulk solvation8 and available for MeCN, DMA, DMF, DMSO, H2O, MeOH, and TMS solvents. The deviations between obtained via different strategies and their Marcus “bulk” scale counterparts in the form of ρ, MUE, and MSE are given in Fig. 7a–c and Table S10 (ESI†). According to Fig. 7a–c the lowest MUE and MSE and the highest ρ were obtained for the monomer cycle that inclines us to recommend these values. The “monomer” cycle performance is followed closely by that of the experimental-based monomer/cluster cycle. The worst performance was detected for the cluster cycle regardless of the number of explicitly coordinated solvent molecules considered. The large negative MSE values obtained for the “cluster” cycle can be reduced if using the CPA proton scale of Kelly et al.18,19 and correcting the Marcus8 experimental numbers accordingly. The somewhat lower ρ values obtained for the cluster cycle are harder to explain.
Footnote |
† Electronic supplementary information (ESI) available: Notes on computational details, Tables S1–S10, Cartesian coordinates (Å) and electronic energies (hartree) of all species studied in the present work. See DOI: 10.1039/d1cp01454d |
This journal is © the Owner Societies 2021 |