Transfer learned potential energy surfaces: accurate anharmonic vibrational dynamics and dissociation energies for the formic acid monomer and dimer

Abstract

The vibrational dynamics of formic acid monomer (FAM) and dimer (FAD) is investigated from machine-learned potential energy surfaces at the MP2 (PESMP2) and transfer-learned (PESTL) to the CCSD(T) levels of...

Quantitative molecular simulations

“… 197 Comparing computed IR spectra from MD simulations on a morphed PES 178 with those measured experimentally yielded an estimated barrier for DPT of 7.2 kcal mol −1 197 which compares with 7.3 kcal mol −1 from a subsequent analysis of microwave spectra 198 and 8.2 kcal mol −1 from fitting CCSD(T)-F12a energies calculated with the cc-pVTZ and aug-cc-pVTZ basis sets for H and C/O atoms (CCSD(T)-F12a/haTZ) to permutationally invariant polynomials. 199 A recent full dimensional PES which was transfer learned from the MP2 level of theory to CCSD(T) energies reported 200 a barrier for DPT of 7.92 kcal mol −1 and a dissociation energy for FAD in the gas phase from diffusion Monte Carlo simulations of D 0 = −14.23 ± 0.08 kcal mol −1 in excellent agreement with an experimentally determined value of −14.22 ± 0.12 kcal mol −1 . 201 These examples illustrate the benefit of accurate, reactive and full-dimensional PESs for quantitative simulations of gas phase processes.…”

Quantitative Molecular Simulations

“…ML-based techniques provide an opportunity to include such couplings from rigorous electronic structure data and first examples demonstrate their accuracy for chemical reactions and spectroscopy. 173,250 Dynamical simulations and the chemical understanding of surface processes are a crucial factor for innovations in heterogeneous catalysis. The difficulties and large effort in performing accurate, high-level quantum electronic computations together with incorporating non-adiabatic effects during molecule-surface interactions has often hindered quantitative agreement between simulations and experimental observations.…”

Computational Vibrational Spectroscopy

“…36 Although such effects can be "encoded" in an empirical force field, capturing such effects from a globally valid, machine-learned energy function is more readily possible as has recently been done for formic acid dimer. 37,38 As an example for the performance of state-of-the art ML-based methods for vibrational spectroscopy, formic acid monomer and dimer (FAM and FAD) in the gas phase is considered. 39 Using PhysNet 40 a reference machine-learned PES was determined at the MP2/aug-cc-pVTZ level of theory for FAM and FAD.…”