The central role of density functional theory in the AI age

Science, Год журнала: 2023, Номер 381(6654), С. 170 - 175

Опубликована: Июль 13, 2023

Density functional theory (DFT) plays a pivotal role for the chemical and materials science due to its relatively high predictive power, applicability, versatility computational efficiency. We review recent progress in machine learning model developments which has relied heavily on density synthetic data generation design of architectures. The general relevance these is placed some broader context sciences. Resulting DFT based models with efficiency, accuracy, scalability, transferability (EAST), indicates probable ways routine use successful experimental planning software within self-driving laboratories.

Язык: Английский

---

Best-of-Both-Worlds Predictive Approach to Dissociative Chemisorption on Metals

The Journal of Physical Chemistry Letters, Год журнала: 2024, Номер 15(1), С. 307 - 315

Опубликована: Янв. 3, 2024

Predictive capability, accuracy, and affordability are essential features of a theory that is capable describing dissociative chemisorption on metal surface. This type reaction important for heterogeneous catalysis. Here we present an approach in which use diffusion Monte Carlo (DMC) to pin the minimum barrier height construct density functional reproduces this value. predictive allows construction potential energy surface at cost while retaining near DMC accuracy. Scrutinizing effects dissipation quantum tunneling, dynamics calculations suggest be chemical reproducing molecular beam sticking experiments showcase H2 + Al(110) system ∼1.4 kcal/mol.

Язык: Английский

---

Toward Accurate Quantum Mechanical Thermochemistry: (1) Extensible Implementation and Comparison of Bond Additivity Corrections and Isodesmic Reactions

The Journal of Physical Chemistry A, Год журнала: 2024, Номер 128(21), С. 4335 - 4352

Опубликована: Май 16, 2024

Obtaining accurate enthalpies of formation chemical species, Δ

Язык: Английский

---

Adapting hybrid density functionals with machine learning

Science Advances, Год журнала: 2025, Номер 11(5)

Опубликована: Янв. 31, 2025

Exact exchange contributions significantly affect electronic states, influencing covalent bond formation and breaking. Hybrid density functional approximations, which average exact admixtures empirically, have achieved success but fall short of high-level quantum chemistry accuracy due to delocalization errors. We propose adaptive hybrid functionals, generating optimal admixture ratios on the fly using data-efficient machine learning models with negligible overhead. The Perdew-Burke-Ernzerhof (aPBE0) improves energetics, electron densities, HOMO-LUMO gaps in QM9, QM7b, GMTKN55 benchmark datasets. A model uncertainty-based constraint reduces method smoothly PBE0 extrapolative regimes, ensuring general applicability limited training. By tuning fractions for different spin aPBE0 effectively addresses gap problem open-shell systems such as carbenes. also present a revised QM9 (revQM9) dataset more accurate properties, including stronger binding, larger bandgaps, localized dipole moments.

Язык: Английский

---

Rapid high-fidelity quantum simulations using multi-step nonlinear autoregression and graph embeddings

npj Computational Materials, Год журнала: 2025, Номер 11(1)

Опубликована: Март 2, 2025

Abstract The design and high-throughput screening of materials using machine-learning assisted quantum-mechanical simulations typically requires the existence a very large data set, often generated from at high level theory or fidelity. A single simulation fidelity can take on order days for complex molecule. Thus, although machine learning surrogate seem promising first glance, generation training defeat original purpose. For this reason, use to screen remains elusive many important applications. In paper we introduce new multi-fidelity approach based dual graph embedding extract features that are placed inside nonlinear multi-step autoregressive model. Experiments five benchmark problems, with 14 different quantities 27 levels theory, demonstrate generalizability accuracy approach. It few 10s 1000’s high-fidelity points, which is several orders magnitude lower than direct ML methods, be up two other methods. Furthermore, develop set 860 benzoquinone molecules atoms, containing energy, HOMO, LUMO dipole moment values four coupled cluster singles doubles.

Язык: Английский

---

Predicting Molecular Energies of Small Organic Molecules With Multi‐Fidelity Methods

Journal of Computational Chemistry, Год журнала: 2025, Номер 46(6)

Опубликована: Март 4, 2025

Multi-fidelity methods in machine learning (ML) have seen increasing usage for the prediction of quantum chemical properties. These methods, such as Δ$$ \Delta $$ -ML and Multifidelity Machine Learning (MFML), been shown to significantly reduce computational cost generating training data. This work implements analyzes several multi-fidelity including MFML electronic molecular energies at DLPNO-CCSD(T) level, that is, level coupled cluster theory single double excitations perturbative triples corrections. The models small organic molecules are evaluated not only on basis accuracy prediction, but also efficiency terms time-cost In addition, sampled from a public dataset, particular atmospherically relevant molecules, isomeric compounds, highly conjugated complex molecules.

Язык: Английский

---

Basis Set Incompleteness Errors in Fixed-Node Diffusion Monte Carlo Calculations on Noncovalent Interactions

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Апрель 30, 2025

Basis set incompleteness error (BSIE) is a common source of in quantum chemistry calculations, but it has not been comprehensively studied fixed-node Diffusion Monte Carlo (FN-DMC) calculations. FN-DMC, being projection method, often considered minimally affected by basis biases. Here, we show that this assumption always valid. While the relative introduced small total FN-DMC energy minor, can become significant binding (Eb) evaluations weakly interacting systems. We systematically investigated BSIEs FN-DMC-based Eb using A24 data set, well-known benchmark 24 noncovalently bound dimers. found are indeed when localized sets, such as cc-pVDZ and cc-pVTZ, employed. Our study shows aug-cc-pVTZ family strikes good balance between computational cost also augmenting sets with diffuse orbitals, counterpoise correction, or both, effectively mitigates BSIEs, allowing smaller aug-cc-pVDZ to be used.

Язык: Английский

---

Foundry-ML - Software and Services to Simplify Access to Machine Learning Datasets in Materials Science

The Journal of Open Source Software, Год журнала: 2024, Номер 9(93), С. 5467 - 5467

Опубликована: Янв. 23, 2024

Schmidt et al., (2024). Foundry-ML - Software and Services to Simplify Access Machine Learning Datasets in Materials Science. Journal of Open Source Software, 9(93), 5467, https://doi.org/10.21105/joss.05467

Язык: Английский

---

Accurate Quantum Monte Carlo Forces for Machine-Learned Force Fields: Ethanol as a Benchmark

Journal of Chemical Theory and Computation, Год журнала: 2024, Номер 20(14), С. 6020 - 6027

Опубликована: Июль 14, 2024

Quantum Monte Carlo (QMC) is a powerful method to calculate accurate energies and forces for molecular systems. In this work, we demonstrate how can obtain QMC the fluxional ethanol molecule at room temperature by using either multideterminant Jastrow-Slater wave functions in variational or just single determinant diffusion Carlo. The excellent performance of our protocols assessed against high-level coupled cluster calculations on diverse set representative configurations system. Finally, train machine-learning force fields compare them models trained reference data, showing that field based with faithfully reproduce power spectra dynamics simulations.

Язык: Английский

---

Investigation of Electronic and Molecular Features of Zn3S3/PEG4000 Nano-Composite Using the DFT Method

Journal of the Turkish Chemical Society Section A Chemistry, Год журнала: 2024, Номер 11(2), С. 565 - 574

Опубликована: Апрель 30, 2024

Molecular geometry structures were accurately optimized to low convergence energy thresholds for the Zn3S3 cluster before and after adding Polyethylene Glycol (PEG4000). Density functional theory DFT/ B3LYP calculations with 6-113G (d, p) basis set employed investigate structural electronic properties of Zn3S3/PEG4000 composite. The FTIR spectral lines analyzed where an agreement spectra titled molecules was evaluated between experimental theoretical findings active peaks O–H, C–H, C=O, C–O–C, Zn–S groups. vibrational modes frequencies systematically on distribution potential around range 0–4000 cm-1 observed 12 vibrations molecule, while 36 compound. Frontier high occupied, unoccupied molecular orbitals (HOMO&LUMO) calculated plotted obtain gap (E𝒈) resulting from difference those orbitals. promising indicator obtained at increasing E𝒈 (4.031 4.459) eV PEG4000, pointing out effect polymer ZnS surface as a capping agent. Additionally, features mentioned structures, such IP, EA, Ef, E𝒈, 𝐶𝑝, χ, η, Ѕ, ω, calculated. Finally, electrostatic (MEP) diagram Zn3S3/ PEG4000 charge densities isosurface contour diagrams estimated, showing nucleophilic electrophilic attack these compounds.

Язык: Английский

---