Solvation Entropy Made Simple Alejandro J. Garza∗ The Dow Chemical Company, 1776 Building Midland, Michigan 48674, United States E-mail: [email protected] Abstract The entropies of molecules in solution are routinely calculated using gas phase formulae. It is assumed that, because implicit solvation models are fitted to reproduce free energies, this is sufficient for modeling reactions in solution. However, this procedure exaggerates entropic effects in processes that change molecularity. Here, computationally efficient (i.e., having similar cost as gas phase entropy calculations) approximations for determining solvation entropy are proposed to address this issue. The S!, S, and Sα models are nonempirical and rely only on physical arguments and elementary properties of the medium (e.g., density and relative permittivity). For all three methods, average errors as compared to experiment are within chemical accuracy for 110 solvation entropies, 11 activation entropies in solution, and 32 vaporization enthalpies. The models also make predictions regarding microscopic and bulk properties of liquids which prove to be accurate. These results imply that ∆Hsol and ∆Ssol can be described separately and with less reliance on parametrization by a combination of the methods presented here with existing, reparametrized implicit solvation models. Introduction isfactory as it inevitably underestimates ∆Sreac effects and can result in unphysical negative It is standard practice in computational chem- free energy barriers (a recent publication10 dis- istry to calculate the free energy of molecules cusses this and related ad hoc methods to com- in solution utilizing gas phase entropies.1,2 Gas pute entropies in solution from entropies in the phase entropies are well described by analyti- gas phase). Another workaround is to scale gas cal formulas of statistical mechanics, but there phase entropies by a rule-of-thumb factor of ≈ is no similarly efficient way of estimating en- 0.65.1,11{14 While scaling gas phase entropies tropies in solution rigorously. However, the may be reasonable for small molecules, it is not use of gas phase entropies to model processes justified for larger molecules for which the vi- that change molecularity (e.g., adsorption and brational and cavity, rather than translational binding) in solution often exaggerates entropy and rotational, entropy terms dominate (there changes during the reaction, ∆S .1{8 In con- reac is no reason for which vibrations should change densed media, the translational and rotational arXiv:1901.11128v1 [physics.chem-ph] 30 Jan 2019 drastically upon solvation; numerical results motions of a molecule are hindered, reducing support this view15). Implicit solvation mod- the entropy as compared to the gas phase. Re- els are often fitted to reproduce experimental arrangement of the solvent to form a cavity for free energies under standard conditions,2,16{20 the solute further lowers the entropy of the sys- so one could expect them to improve ∆G . tem. The typical error in ∆S for bimolec- reac reac However, unless a model designed to account ular reactions in solution is so large that some for temperature effects is used and appropriate authors have suggested to completely neglect derivatives are taken (see, e.g., refs. 21,22), the translational and rotational entropy5,6 (or just fact that the T ∆S term in ∆G relies on the former9). This approach is, however, unsat- reac reac gas phase entropies will result in an incorrect 1 temperature dependence of the reaction. Fur- Theory thermore, in actual applications, implicit solva- The total entropy of an atom or molecule in tion models often do not improve the too-high solution is decomposed into contributions from binding free energies caused by the use of gas vibrations, translations, rotations, and the sol- phase entropies.1{9 Molecular dynamics and al- vent cavity: chemical free energy methods can in principle determine free energies in solution without re- S = Sv + St + Sr + Sc: (1) lying on gas phase formulas.23{25 Nonetheless, such calculations require significant additional The vibrational entropy can be computed from work from the user and are computationally de- the harmonic oscillator approximation in the manding, which makes them unsuitable for rou- same way as in the gas phase. It is well known, tine or high-throughput calculations. They also however, that this approximation yields un- suffer from inherent reproducibility issues due physically large contributions to the entropy to the sensitivity of Newtonian dynamics to ini- from low-frequency modes. To avoid this is- tial conditions and the lingering effects of such sue, Sv is calculated with the method proposed conditions if sampling is insufficient.24,25 by Grimme,26 which can be seen as a quasi hin- The purpose of this work is to formulate an dered rotor that interpolates between the har- efficient approximation for calculating molecu- monic oscillator and free rotor entropy formu- lar entropy in solution based only on physical las. One can, however, compute Sv with any ap- and geometric arguments. Three models, S!, proximation deemed appropriate as Sv does not S, and Sα, are derived that rely only on such influence solvation for the methods presented arguments|no empirically fitted parameters| here. and elementary solvent properties (e.g., mass density). The methods differ only in how the Translational Entropy cavitation entropy is calculated: S!, S, and Sα utilize, respectively, the Pitzer acentric fac- The contributions from St in terms of the trans- 27,28 tor (!), the relative permittivity (r), and r lational partition function are as well as the isobaric thermal expansion coef- @ ln (q ) ficient (α). This establishes a connection be- S = k ln (q ) + k + kT t ; (2) t t @T tween ! (a microscopic property), r, and α V (macroscopic properties). The cost of evaluat- where q is approximated by the familiar ex- ing entropy with these models is comparable to t pression obtained from the eigenenergies of a the cost of calculating gas phase entropies with particle of mass m confined in a volume V : the ideal gas/rigid rotor/harmonic approxima- tion. Their accuracy is tested by constructing 2πmkT 3=2 a database of 110 experimental solvation en- qt = V: (3) h2 tropies and 11 activation entropies in solution; the resulting average errors are in the range of All quantities in eq. 3 are unambiguously de- 2{3 cal/mol-K, which is comparable to the ac- fined except for V . For an ideal gas V = kT=P , curacy of gas phase entropies and within what is but in condensed media V will depend on prop- considered chemical accuracy (≤ 1 kcal/mol at erties of the medium such as, e.g., its density 300 K). Additionally, the models make testable and particle volume. Here, we define V in terms predictions regarding molecular and bulk prop- of the volume of the solute cavity, vc, as well as erties of liquids that prove to be in agreement the average number of accessible cavities Nc, with experiment. V = Ncvc; (4) so that we can evaluate V based on a physical interpretation of Nc and vc. In our model, vc is 2 the volume of a sphere with a radius equal to Typically, Nc ≈ 1, but the hopping terms can the sum of the spherical equivalent radii of the be important in cases of small solutes in bulky solute and the volume of free space per solvent or low density solvents. Note that previous particle. This definition is equivalent to works have also utilized a cavity volume to 30 determine St. However, the definition of vc 3 1=3 1=3 and consideration of the possibility of hopping vc = VM + Vfree (5) distinguish the present approach from previous with ones. S Mw Although we have written in eq. 1 separate Vfree = − VS; (6) NAρ terms for St, Sr, and Sc, these are actually in- tertwined. As we see next, the definition of St where VS/M is the volume of a solvent/solute molecule, ρ the mass density of the medium, is important to determine Sr, and Sr is in turn used to derive an approximation for Sc. The NA Avogadro's number, and Mw the molec- ular weight (throughout this document, sub- way to think about St to more easily under- scripts/superscripts \M" and \S" denote solute stand Sr as conceptualized here is simply as the entropy of a point particle in a box, as opposed and solvent, respectively). Here, VM and VS are determined from van der Waals radii,29 though to the entropy of an object that has rotations. vc is relatively insensitive to how molecular vol- umes are defined (vide infra). Rotational Entropy To determine Nc, we define the probability We define Sr in terms of the contributions from per solvent particle of \hopping" to an adjacent the rigid rotor approximation and the transla- cavity, x, based on the solvent, solute, and free tional entropy lost by virtue of acquiring a gyra- volumes as tion radius while being confined to V = Ncvc. 2=3 2=3 Assuming that rotation is fast, the radius rc maxfVfree − VM ; 0g x = 2=3 2=3 : (7) of the cavity in which the centroid of a linear Vfree + VS or spherically symmetric rigid rotor can move That is, the solute can only hop if the cross freely is effectively reduced by its radius of gy- ration rg, sectional area of VM is smaller than Vfree (given that all cross sectional areas are defined Natoms identically in terms of volume, regardless of 2 1 X 2 rg = (rk − rmean) : (10) Natoms shape). Furthermore, assuming effective spher- k=1 ical shapes for each volume and introducing 1=3 For nonsymmetric rotors, the reduction by of rc = [3vc=(4π)] as the radius of the cavity, there will be rc by rg is also assumed as an averaged radius is necessary to preserve rotational invariance.