Atomistic Simulations for Reactions and Vibrational Spectroscopy in the Era of Machine Learning─<i>Quo Vadis?</i>

Meuwly, Markus

doi:10.1021/acs.jpcb.2c00212

Cited by 10 publications

(7 citation statements)

References 192 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Here we focus on an application of the Δ-ML approach to AcAc, starting with a fragmented PIP basis that we used to obtained a new PES . We benefited from the dataset used for a neural-network-based PES for AcAc, but we decided to augment the dataset for that fit with additional MP2/aVTZ energies and gradients. The Δ-ML approach was then used to bring this surface up to LCCSD(T)-F12 accuracy.…”

Section: δ-Ml Pess For a Variety Of Molecular Systemsmentioning

confidence: 99%

Δ-Machine Learned Potential Energy Surfaces and Force Fields

Bowman

Qu²,

Conte

et al. 2022

J. Chem. Theory Comput.

View full text Add to dashboard Cite

There has been great progress in developing machinelearned potential energy surfaces (PESs) for molecules and clusters with more than 10 atoms. Unfortunately, this number of atoms generally limits the level of electronic structure theory to less than the "gold standard" CCSD(T) level. Indeed, for the well-known MD17 dataset for molecules with 9−20 atoms, all of the energies and forces were obtained with DFT calculations (PBE). This Perspective is focused on a Δmachine learning method that we recently proposed and applied to bring DFT-based PESs to close to CCSD(T) accuracy. This is demonstrated for hydronium, N-methylacetamide, acetyl acetone, and ethanol. For 15atom tropolone, it appears that special approaches (e.g., molecular tailoring, local CCSD(T)) are needed to obtain the CCSD(T) energies. A new aspect of this approach is the extension of Δ-machine learning to force fields. The approach is based on many-body corrections to polarizable force field potentials. This is examined in detail using the TTM2.1 water potential. The corrections make use of our recent CCSD(T) datasets for 2-b, 3-b, and 4-b interactions for water. These datasets were used to develop a new fully ab initio potential for water, termed q-AQUA.

show abstract

Section: δ-Ml Pess For a Variety Of Molecular Systemsmentioning

confidence: 99%

Δ-Machine Learned Potential Energy Surfaces and Force Fields

Bowman

Qu²,

Conte

et al. 2022

J. Chem. Theory Comput.

View full text Add to dashboard Cite

show abstract

“…3.3). Methods for deconvolution, 2D plots, and machine learning procedures should enhance the analysis of even more complex IR spectra extending the attainable information to the fingerprint range [76][77][78][79][80][81][82][83]. In summary, this simple and cost-effective setup has the potential to be applied in a wide variety of photo-induced reactions providing evidence at the molecular scale.…”

Section: Discussionmentioning

confidence: 99%

Direct detection of photo-induced reactions by IR: from Brook rearrangement to photo-catalysis

Glotz

Püschmann

Haas

et al. 2023

Photochem Photobiol Sci

View full text Add to dashboard Cite

In situ IR detection of photoreactions induced by the light of LEDs at appropriate wavelengths provides a simple, cost-effective, and versatile method to get insight into mechanistic details. In particular, conversions of functional groups can be selectively followed. Overlapping UV–Vis bands or fluorescence from the reactants and products and the incident light do not obstruct IR detection. Compared with in situ photo-NMR, our setup does not require tedious sample preparation (optical fibers) and offers a selective detection of reactions, even at positions where 1H-NMR lines overlap or 1H resonances are not clear-cut. We illustrate the applicability of our setup following the photo-Brook rearrangement of (adamant-1-yl-carbonyl)-tris(trimethylsilyl)silane, address photo-induced α-bond cleavage (1-hydroxycyclohexyl phenyl ketone), study photoreduction using tris(bipyridine)ruthenium(II), investigate photo-oxygenation of double bonds with molecular oxygen and the fluorescent 2,4,6-triphenylpyrylium photocatalyst, and address photo-polymerization. With the LED/FT-IR combination, reactions can be qualitatively followed in fluid solution, (highly) viscous environments, and in the solid state. Viscosity changes during the reaction (e.g., during a polymerization) do not obstruct the method. Graphical abstract

show abstract

“…In our recent study, a recursively embedded atom neural network (REANN) model 138,139 was proposed to address these issues. In the spirit of the MPNN, the orbital coefficients in Equation () or Equation () are now environment‐dependent and updated iteratively, 138,139

c_{j}^{t} = g_{j}^{t - 1} ({boldρ}_{j}^{t - 1} (c^{t - 1}, r_{j}^{t - 1})),

where

{boldρ}_{j}^{t - 1}

and

{boldc}^{t - 1}

are the EAD and orbital coefficients of the ( t − 1) iteration and

g_{j}^{t - 1}

is an atomic NN whose multiple outputs correspond to the orbital coefficients of the next ( t ) iteration. This process is repeated until the last iteration, where the atomic NN will output the atomic energy.…”

Section: Representabilitymentioning

confidence: 99%

Atomistic neural network representations for chemical dynamics simulations of molecular, condensed phase, and interfacial systems: Efficiency, representability, and generalization

Zhang

Lin

Jiang

2022

WIREs Comput Mol Sci

View full text Add to dashboard Cite

Machine learning techniques have been widely applied in many fields of chemistry, physics, biology, and materials science. One of the most fruitful applications is machine learning of the complicated multidimensional function of potential energy or related electronic properties from discrete quantum chemical data. In particular, substantial efforts have been dedicated to developing various atomistic neural network (AtNN) representations, which refer to a family of methods expressing the targeted physical quantity as a sum of atomic components represented by atomic NNs. This class of approaches not only fully preserves the physical symmetry of the system but also scales linearly with respect to the size of a system, enabling accurate and efficient chemical dynamics and spectroscopic simulations in complicated systems and even a number of variably sized systems across the phases. In this review, we discuss different strategies in developing highly efficient and representable AtNN potentials, and in generalizing these scalar AtNN models to learn vectorial and tensorial quantities with the correct rotational equivariance. We also review active learning algorithms to generate practical AtNN models and present selected examples of AtNN applications in gassurface systems to demonstrate their capabilities of accurately representing both molecular systems and condensed phase systems. We conclude this review by pointing out remaining challenges for the further development of more reliable, transferable, and scalable AtNN representations in more application scenarios.

show abstract

Atomistic Simulations for Reactions and Vibrational Spectroscopy in the Era of Machine Learning─Quo Vadis?

Cited by 10 publications

References 192 publications

Δ-Machine Learned Potential Energy Surfaces and Force Fields

Δ-Machine Learned Potential Energy Surfaces and Force Fields

Direct detection of photo-induced reactions by IR: from Brook rearrangement to photo-catalysis

Atomistic neural network representations for chemical dynamics simulations of molecular, condensed phase, and interfacial systems: Efficiency, representability, and generalization

Contact Info

Product

Resources

About