Accurate fundamental invariant-neural network representation of <i>ab initio</i> potential energy surfaces

Fu, Bina; Zhang, Dong H

doi:10.1093/nsr/nwad321

Cited by 6 publications

(4 citation statements)

References 98 publications

(76 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the large difference between ACE and PIPs is that the PIP approach is not atom-centered. A single basis is used to represent the PES (it should be noted that this comment also applies to the highly successful PIP-NN and PIP/FI-NN methods and to the automated PIP software “Robosurfer”). So, for the same size basis, the CPU effort for ACE scales roughly like N times the CPU time for PIPs.…”

Section: Discussionmentioning

confidence: 99%

No Headache for PIPs: A PIP Potential for Aspirin Runs Much Faster and with Similar Precision Than Other Machine-Learned Potentials

Houston,

Qu,

et al. 2024

J. Chem. Theory Comput.

View full text Add to dashboard Cite

Assessments of machine-learning (ML) potentials are an important aspect of the rapid development of this field. We recently reported an assessment of the linear-regression permutationally invariant polynomial (PIP) method for ethanol, using the widely used (revised) rMD17 data set. We demonstrated that the PIP approach outperformed numerous other methods, e.g., ANI, PhysNet, sGDML, and p-KRR, with respect to precision and notably with respect to speed [Houston et al., J. Chem. Phys. 2022, 156, 044120]. Here, we extend this assessment to the 21-atom aspirin molecule, using the rMD17 data set, with a focus on the speed of evaluation. Both energies and forces are used for training, and the precision of several PIPs is examined for both. Normal mode frequencies, the methyl torsional potential, and 1d vibrational energies for an OH stretch are presented. We show that the PIP approach achieves the level of precision obtained from other ML methods, e.g., atom-centered neural network methods, linear regression ACE, and kernel methods, as reported by Kovaćs et al. in J. Chem. Theory Comput. 2021, 17, 7696−7711. More significantly, we show that the PIP PESs run much faster than all other ML methods, whose timings were evaluated in that paper. We also show that the PIP PES extrapolates well enough to describe several internal motions of aspirin, including an OH stretch.

show abstract

Section: Discussionmentioning

confidence: 99%

No Headache for PIPs: A PIP Potential for Aspirin Runs Much Faster and with Similar Precision Than Other Machine-Learned Potentials

Houston,

Qu,

et al. 2024

J. Chem. Theory Comput.

View full text Add to dashboard Cite

show abstract

“…Based on those high-level data points, the full-dimensional PES of the NaCl + NaCl collision was fitted by the fundamental invariant-neural network (FI-NN) method. − A total of 7 FIs with a maximum degree of 3 were used as the input layer of the neural network. The structure of a standard feedforward neural network (NN) with two hidden layers can be expressed as I - J - K -1.…”

Section: Methodsmentioning

confidence: 99%

Supercollisions of NaCl + NaCl on an Accurate Full-Dimensional Potential Energy Surface

Peng,

Bai,

et al. 2024

J. Phys. Chem. A

Self Cite

View full text Add to dashboard Cite

An accurate, global, full-dimensional potential energy surface (PES) of NaCl + NaCl has been constructed by the fundamental invariant-neural network (FI-NN) fitting based on roughly 13,000 ab initio energies at the level of CCSD(T)-F12a/ aug-cc-pVTZ, with the small fitting error of 0.16 meV. Extensive quasiclassical trajectory (QCT) calculations were performed on this PES to investigate the energy transfer process of the NaCl + NaCl collision at four different collision energies. Various quantities were obtained, including the cross-sections, energy transfer probability, average energy transfer, and collision lifetime. The probabilities of energy transfer (P(ΔE)) for prompt trajectories, nonreactive trajectories, and reactive trajectories deviate from a simple exponential decay pattern. Instead, a noteworthy probability is observed in the high-energy transfer region, indicative of supercollisions. The formation of the (NaCl) 2 complex, coupled with a comparatively extended collision lifetime, promotes vibrational excitation in NaCl molecules. The reactive trajectories exhibit enhanced energy transfer, attributed to the longer lifetime of the NaCl dimer. This study not only provides an accurate and extensive understanding of the NaCl + NaCl collision dynamics but also reveals intriguing phenomena, such as supercollisions and enhanced energy transfer in reactive trajectories, shedding light on the complex intricacies of molecular interactions.

show abstract

“…One of the recent successful applications of machine learning in molecular sciences is the construction and use of neural network-based (NN-based) potential energy surfaces (PESs). The ability to conceive full-dimensional PESs for molecules, which goes back to the late 1990s, , has revolutionized the field of molecular simulations. − This is primarily due to their ease-of-use and versatility. Also, they have shown to reproduce the input data quite accurately, with representation errors well below chemical accuracy (1 kcal/mol).…”

mentioning

confidence: 99%

Numerical Accuracy Matters: Applications of Machine Learned Potential Energy Surfaces

Käser,

Meuwly

2024

J. Phys. Chem. Lett.

View full text Add to dashboard Cite

The role of numerical accuracy in training and evaluating neural network-based potential energy surfaces is examined for different experimental observables. For observables that require third-and fourth-order derivatives of the potential energy with respect to Cartesian coordinates single-precision arithmetics as is typically used in ML-based approaches is insufficient and leads to roughness of the underlying PES as is explicitly demonstrated. Increasing the numerical accuracy to double-precision gives a smooth PES with higher-order derivatives that are numerically stable and yield meaningful anharmonic frequencies and tunneling splitting as is demonstrated for H 2 CO and malonaldehyde. For molecular dynamics simulations, which only require first-order derivatives, single-precision arithmetics appears to be sufficient, though.

show abstract

Accurate fundamental invariant-neural network representation of ab initio potential energy surfaces

Cited by 6 publications

References 98 publications

No Headache for PIPs: A PIP Potential for Aspirin Runs Much Faster and with Similar Precision Than Other Machine-Learned Potentials

No Headache for PIPs: A PIP Potential for Aspirin Runs Much Faster and with Similar Precision Than Other Machine-Learned Potentials

Supercollisions of NaCl + NaCl on an Accurate Full-Dimensional Potential Energy Surface

Numerical Accuracy Matters: Applications of Machine Learned Potential Energy Surfaces

Contact Info

Product

Resources

About