Calculated state-of-the art results for solvation and ionization energies of thousands of organic molecules relevant to battery design
Jan Weinreich, Konstantin Karandashev, Daniel Jose Arismendi Arrieta, Kersti Hermansson, O. Anatole von Lilienfeld · 2024
Read Calculated state-of-the art results for solvation and ionization energies of thousands of organic molecules relevant to battery design by Jan Weinreich, Konstantin Karandashev, Daniel Jose Arismendi Arrieta, Kersti Hermansson, O. Anatole von Lilienfeld (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2411-00994.txt.
Description
Abstract We present high-quality reference data for two fundamentally important groups of molecular properties related to a compound’s utility as a lithium battery electrolyte. The first one is energy changes associated with charge excitations of molecules, namely ionization potential and electron affinity.
Catalog record
| CREATOR | Jan Weinreich, Konstantin Karandashev, Daniel Jose Arismendi Arrieta, Kersti Hermansson, O. Anatole von Lilienfeld |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2411-00994
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Calculated state-of-the art results for solvation and ionization energies of thousands of organic molecules relevant to battery design Jan Weinreich, Konstantin Karandashev, Daniel Jose Arismendi Arrieta, Kersti Hermansson, O. Anatole von Lilienfeld Abstract We present high-quality reference data for two fundamentally important groups of molecular properties related to a compound’s utility as a lithium battery electrolyte. The first one is energy changes associated with charge excitations of molecules, namely ionization potential and electron affinity. They were estimated for 7000 randomly chosen molecules with up to 9 non-hydrogen atoms C, N, O, and F (QM9 dataset) using DH-HF, DF-HF-CABS, PNO-LMP2-F12, and PNO-LCCSD(T)-F12 methods as implemented in Molpro software with aug-cc-pVTZ basis set; additionally, we provide the corresponding atomization energies at these levels of theory, as well as CPU time and disk space used during the calculations. The second one is solvation energies for 39 different solvents, which we estimate for 18361 molecules connected to battery design (Electrolyte Genome Project dataset), 309463 randomly chosen molecules with up to 17 non-hydrogen atoms C, N, O, S, and halogens (GDB17 dataset), as well as 88418 amons of ZINC database of commercially available compounds and 37772 amons of GDB17. For these calculations we used the COnductor-like Screening MOdel for Real Solvents (COSMO-RS) method; we additionally provide estimates of gas-phase atomization energies, as well as information about conformers considered during the COSMO-RS calculations, namely coordinates, energies, and dipole moments. 1 Introduction Identifying candidate molecules for lithium battery electrolyte components is an important part of creating batteries suited for different modes of operation. The discovery of molecules and materials with improved functionality is expected to greatly accelerate with machine learning (ML) algorithms in coming years [ 35 , 23 , 63 ] . However, achieving adequate accuracy of new ML models critically depends on the availability of high-quality reference data, both for validation of the new models and of the underlying ML algorithms, and for training new models to guide exploration of chemical space. This motivated us to collect high-quality data for two fundamentally important molecular properties. The first one is energy changes associated with charge excitations of molecules, namely ionization potential and electron affinity. The second one is solvation energies which are related to how usable a compound is as an additive to lithium battery electrolytes. Ionization potential (IP) and electron affinity (EA) are related to how readily a molecule enters reduction and oxidation reactions. Although the utility of these quantities in battery design has been put into question, [ 9 ] they are often used in preliminary screening for battery electrolyte components, [ 31 , 11 , 47 , 34 ] where these quantities are evaluated with relatively computationally cheap calculations. Thus, our goal here is to provide accurate reference data to allow the testing of the fidelity of such calculations. When choosing our reference calculation methods, we have avoided dependence on any kind of empirical fits, disqualifying density functional theory methods and limiting our considerations to post-Hartree Fock approaches. We also aimed to balance feasible computational time for medium-sized molecules that could be considered for battery electrolytes (as too large molecules tend to create issues with solvent viscosity and ion conductivity [ 10 ] ) with accuracy, drawing our attention to PNO-LMP2-F12 [ 68 , 38 , 32 ] and PNO-LCCSD(T)-F12 [ 54 , 37 , 40 , 41 ] methods as implemented in Molpro. [ 69 , 70 , 71 ] The methods improve upon standard coupled-cluster method [ 62 , 6 ] with single and double excitations and a perturbative treatment of triple excitation [ 48 , 7 , 57 ] [CCSD(T)] and second-order Møller-Plesset perturbation theory [ 42 , 12 ] (MP2) in several ways [ 39 ] . For each calculation we also present the corresponding computational time to allow using the data for developing novel multilevel machine learning schemes. [ 73 , 20 ] More calculation details are given in Section 2 . The resulting dataset is named “QM9-IPEA”. Solvation energies are of similar importance for understanding interactions between the solute and the solvent, though in the context of battery research they are primarily useful for approximating trends in solubility in a given battery electrolyte, [ 11 ] with the latter quantity being a fundamentally important factor of how usable a compound is as an electrolyte additive. Our method of choice for estimating these quantities is the COnductor-like Screening MOdel for Real Solvents [ 29 , 25 , 28 , 26 ] (COSMO-RS), which is based on quantum chemical calculations and trades a much higher cost for smaller number of additional parameters [ 27 ] and robustness. The accuracy of COSMO-RS is among the highest for describing equilibrium fluid thermodynamics: for instance, a mean unsigned error of 0.5\text{\,}\mathrm{k} was achieved for experimental solvation energies [ 74 ] making it popular among chemical engineering applications. [ 18 ] In the context of battery material research, COSMO-RS also demonstrated reasonable mean absolute errors (as low as 0.14 V) when predicting experimental values of redox potentials. [ 61 ] The balance of accuracy and computational cost provided by COSMO-RS allowed us to construct a dataset with a significant chemical variety of both solvated molecules and the solvents they are solvated in, while accurately reflecting physical trends over such a wide chemical space. We cover a broad spectrum of organic chemistry with our database, with a particular focus on electrolyte applications. The foundation of our database is the Electrolyte Genome Project (EGP) dataset [ 47 ] , which we have expanded by randomly selecting structures from the GDB17 [ 52 ] database, as well as with fragment structures of AGZ7 [ 21 ] designed to cover the chemical space of ZINC and GDB17. Details are given in Section 3 . Our resulting dataset contains data on 418185 molecules and is named “SolQuest”. 2 QM9-IPEA dataset details 2.1 Molecules included The calculations were performed for a subset of the QM9 [ 52 , 49 ] dataset, which is often used for benchmarking in machine learning algorithm studies. We randomly chose 7000 molecules from the dataset and performed calculations for their states with charges 0, 1, and -1 using the geometries present in the dataset. We used aug-cc-pVTZ [ 13 , 24 , 72 ] basis set. When running Molpro we set the following thresholds: 10^{-8} a.u. for energy convergence, 10^{-7} for orthonormality check, 5\cdot 10^{-10} for smallest allowed eigenvalue of the overlap matrix. Note that calculations for 4000 of the QM9 molecules considered in this work previously appeared in Ref. 20 along with 4000 calculations for EGP. [ 47 ] We had generated the latter with exactly the same methodology, except we additionally used def2-TZVPP [ 64 ] basis set for Li, Be, and Ca atoms and aug-cc-pVTZ-PP [ 15 , 44 , 45 ] basis set for Zn and Br atoms. 2.2 IP and EA calculations As mentioned in the Introduction, the local correlation treatment in PNO-LMP2-F23 and PNO-LCCSD(T)-F12 methods used here improve on the CCSD(T) method. Firstly, the scaling of the methods’ cost with system size is improved by using their localized versions [LMP2 [ 46 ] and LCCSD(T) [ 19 , 53 ] ] that only consider excitations between localized orbitals