Automatic State Interaction with Large Localized Active Spaces for Multimetallic Systems

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(11), P. 4654 - 4662

Published: May 24, 2024

The localized active space self-consistent field method factorizes a complete wave function into an antisymmetrized product of fragments. Correlation between fragments is then reintroduced through state interaction (LASSI), in which the Hamiltonian diagonalized model LAS states. However, optimal procedure for defining and LASSI unknown. We here present automated framework to explore systematically convergent sets spaces, we call LASSI[r, q]. This requires user select only r, number electron hops from one fragment another, q, basis functions per Hilbert space, converges CASCI limit q → ∞. Numerical tests this on trimetal oxo-centered complexes [Fe(III)Al(III)Fe(II)(μ3-O)(HCOO)6] [Fe(III)2Fe(II)(μ3-O)(HCOO)6] show efficient convergence with 4–10 orders magnitude fewer states than CASCI.

Language: Английский

---

Improving Bond Dissociations of Reactive Machine Learning Potentials through Physics-Constrained Data Augmentation

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 28, 2025

In the field of computational chemistry, predicting bond dissociation energies (BDEs) presents well-known challenges, particularly due to multireference character reactive systems. Many chemical reactions involve configurations where single-reference methods fall short, as electronic structure can significantly change during breaking. As generating training data for partially broken bonds is a challenging task, even state-of-the-art machine learning interatomic potentials (MLIPs) often fail predict reliable BDEs and smooth curves. By contrast, simple inexpensive physics-based models, such well-established Morse potential, do not suffer from any limitations. This work leverages potential improve MLIPs by augmenting set with along pathways. physics-constrained augmentation (PCDA) approach results in curves well near coupled-cluster level BDEs, all without requiring expensive quantum mechanical calculations. A case study methane combustion demonstrates how PCDA an existing MLIP, namely, ANI-1xnr. Not only are radicals molecules improved compared ANI-1xnr but PCDA-trained MLIP retains reliability when performing molecular dynamics simulations.

Language: Английский

---

Chemical Reaction Networks from Scratch with Reaction Prediction and Kinetics-Guided Exploration

Journal of Chemical Theory and Computation, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 30, 2025

Algorithmic reaction explorations based on transition state searches can now routinely predict relatively short sequences involving small molecules. However, applying these algorithms to deeper chemical network (CRN) exploration still requires the development of more efficient and accurate policies. Here, an algorithm, which we name yet another kinetic strategy (YAKS), is demonstrated that uses microkinetic simulations nascent achieve cost-effective, deep exploration. Key features algorithm are automatic incorporation bimolecular reactions between intermediates, compatibility with short-lived but kinetically important species, rate uncertainty into policy. In validation case studies glucose pyrolysis, rediscovers pathways previously discovered by heuristic policies elucidates new for experimentally obtained products. The resulting CRN first connect all major experimental pyrolysis products glucose. Additional presented investigate role rules, uncertainty, reactions. These show naïve exponential growth estimates vastly overestimate actual number relevant in physical networks. light this, further improvements prediction make it feasible CRNs might soon be predictable some contexts.

Language: Английский

---

Ten Problems in Polymer Reactivity Prediction

Macromolecules, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 17, 2025

Language: Английский

---

Automated Multireference Vertical Excitations for Transition-Metal Compounds

The Journal of Physical Chemistry A, Journal Year: 2025, Volume and Issue: unknown

Published: March 4, 2025

Excited states of transition metal complexes are generally strongly correlated due to the near-degeneracy d orbitals. Consequently, electronic structure calculations such species often necessitate multireference approaches. However, widespread use methods is hindered active space selection problem, which has historically required system-specific chemical knowledge and a trial-and-error approach. Here, we address this issue with an automated method combining approximate pair coefficient (APC) scheme for estimating orbital entropies discrete variational (DVS) approach evaluating quality. We apply DVS-APC calculation 67 vertical excitations in diatomics as well two larger complexes. show generated spaces yield NEVPT2 mean absolute errors 0.18 eV, line previous accuracies obtained organic systems, but than achieved hand-selected (0.14 eV). If instead using DVS identify best results from our trial wave functions, find improved performance (mean error 0.1 eV) over manually selected results. highlight deviation between hand possible measure bias introduced when selecting spaces. that multiconfiguration pair-density functional theory (MC-PDFT) tPBE tPBE0 functionals roughly 0.15 eV less accurate across class diatomic potentially accounting decreased DVS-APC, uses MC-PDFT energies select also showcase ability "down-sample" functions natural occupancies achieve smaller minimal retain accuracy starting Finally, proven be effective applied modeling excited complexes, suggesting may particular outstanding challenge

Language: Английский

---

Including Physics-Informed Atomization Constraints in Neural Networks for Reactive Chemistry

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: April 29, 2025

Machine learning interatomic potentials (MLIPs) have emerged as powerful tools for investigating atomistic systems with high accuracy and a relatively low computational cost. However, common unaddressed challenge many current neural network (NN) MLIP models is their limited ability to accurately predict the relative energies of containing isolated or nearly atoms, which appear in various reactive processes. To address this limitation, we present mathematical technique modifying any existing atom-centered NN architecture account atoms. The result produces consistent prediction atomization energy (AE) system using minimal constraints on model. Using technique, build model that call hierarchically interacting particle (HIP-NN)-AE, an AE-constrained version HIP-NN, well ANI-AE, accurate engine molecular (ANI). Our results demonstrate AE consistency models, drastically improves predictions models. We compare approach unconstrained from literature other scenarios, such bond dissociation energies, pathways, extensibility tests. These show improve performance some these tasks do not negatively affect tasks. constraint thus offers robust solution challenges posed by atoms

Language: Английский

---

Chemical Bond Overlap Descriptors From Multiconfiguration Wavefunctions

Journal of Computational Chemistry, Journal Year: 2024, Volume and Issue: 46(1)

Published: Nov. 28, 2024

The chemical bond is a fundamental concept in chemistry, and various models descriptors have evolved since the advent of quantum mechanics. This study extends overlap density its topological (OP/TOP) to multiconfigurational wavefunctions. We discuss comparative analysis OP/TOP using CASSCF DCD-CAS(2) wavefunctions for diverse range molecular systems, including X-O bonds X-OH (XH, Li, Na, H

Language: Английский

---

Distinguishing homolytic vs heterolytic bond dissociation of phenylsulfonium cations with localized active space methods

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 161(1)

Published: July 1, 2024

Modeling chemical reactions with quantum methods is challenging when the electronic structure varies significantly throughout reaction and excited states are involved. Multireference methods, such as complete active space self-consistent field (CASSCF), can handle these multiconfigurational situations. However, even if size of needed affordable, in many cases, does not change consistently from reactant to product, causing discontinuities potential energy surface. The localized SCF (LASSCF) a cheaper alternative CASSCF for strongly correlated systems weakly fragments. method used first time study reaction, namely bond dissociation mono-, di-, triphenylsulfonium cation. LASSCF calculations generate smooth scans more easily than corresponding, computationally expensive while predicting similar energies. Our suggest homolytic cleavage di- heterolytic pathway monophenylsulfonium.

Language: Английский

---