Physics-Informed Neural Networks and Beyond: Enforcing Physical Constraints in Quantum Dissipative Dynamics

Digital Discovery, Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 1, 2024

This study shows that NNs do not conserve trace in quantum dissipative dynamics and proposes an uncertainty-aware approach enforces perfect conservation by design.

Language: Английский

---

Surprising new dynamics phenomena in Diels–Alder reaction of C60 uncovered with AI

Published: May 22, 2024

Our recently developed physics-informed active learning allowed us to perform extensive AI-accelerated quasi-classical molecular dynamics investigation of the time-resolved mechanism two Diels–Alder reactions. This revealed that despite high similarity between static transition state geometries in reactions with ethene and fullerene C60 as dienophiles, around transitions are remarkably different. In a substantial fraction (10%) reactive trajectories, larger non-covalently attracts 1,3-dimethyl-butadiene long before barrier so diene undergoes series complex motions including roaming, somersaults, twisting, twisting somersaults until it aligns itself pass over barrier. These complicated processes could be easily missed typically performed quantum chemical simulations shorter fewer trajectories. After passing barrier, bonds take longer form than case reaction: consequence markedly different topology PES region reactants products. effects not captured by intrinsic reaction coordinate (IRC) calculations do reveal any difference widths.

Language: Английский

---

Charting electronic-state manifolds across molecules with multi-state learning and gap-driven dynamics via efficient and robust active learning

Published: Aug. 6, 2024

We present a robust protocol for affordable learning of the electronic-state manifold to accelerate photophysical and photochemical molecular simulations. The solves several pertinent issues precluding widespread use machine (ML) in excited-state introduce novel physics-informed multi-state ML model that can learn an arbitrary number excited states across molecules with accuracy better or similar ground-state energies established potentials. also gap-driven dynamics meticulous accelerated sampling small-gap regions: which proves crucial stable surface-hopping dynamics. Put together, enable efficient active furnishing models Our active-learning includes based on uncertainty quantification, ensuring quality each adiabatic surface, low error energy gaps, precise calculation hopping probability. thresholds quantification are automatically chosen statistical physical considerations. will be made available next release open-source MLatom as described at https://github.com/dralgroup/al-namd

Language: Английский

---

Surprising new dynamics phenomena in Diels–Alder reaction of C60 uncovered with AI

Published: May 27, 2024

Our recently developed physics-informed active learning allowed us to perform extensive AI-accelerated quasi-classical molecular dynamics investigation of the time-resolved mechanism two Diels–Alder reactions. This revealed that despite high similarity between static transition state geometries in reactions with ethene and fullerene C60 as dienophiles, around transitions are remarkably different. In a substantial fraction (10%) reactive trajectories, larger non-covalently attracts 2,3-dimethyl-1,3-butadiene long before barrier so diene undergoes series complex motions including roaming, somersaults, twisting, twisting somersaults until it aligns itself pass over barrier. These complicated processes could be easily missed typically performed quantum chemical simulations shorter fewer trajectories. After passing barrier, bonds take longer form than case reaction: consequence markedly different topology PES region reactants products. effects not captured by intrinsic reaction coordinate (IRC) calculations do reveal any difference widths.

Language: Английский

---

Surprising new dynamics phenomena in Diels–Alder reaction of C60 uncovered with AI

Published: June 20, 2024

Our recently developed physics-informed active learning allowed us to perform extensive AI-accelerated quasi-classical molecular dynamics investigation of the time-resolved mechanism Diels–Alder reaction fullerene C60 with 2,3-dimethyl-1,3-butadiene. In a substantial fraction (10%) reactive trajectories, larger non-covalently attracts 2,3-dimethyl-1,3-butadiene long before barrier so that diene undergoes series complex motions including roaming, somersaults, twisting, and twisting somersaults around until it aligns itself pass over barrier. These complicated processes could be easily missed in typically performed quantum chemical simulations shorter fewer trajectories. After passing barrier, bonds take longer form compared simplest prototypical ethene 1,3-butadiene despite high similarities transition states widths evaluated intrinsic coordinate (IRC) calculations.

Language: Английский

---

Surprising Dynamics Phenomena in the Diels–Alder Reaction of C₆₀ Uncovered with AI

The Journal of Organic Chemistry, Journal Year: 2024, Volume and Issue: 89(20), P. 15041 - 15047

Published: Oct. 3, 2024

We performed an extensive artificial intelligence-accelerated quasi-classical molecular dynamics investigation of the time-resolved mechanism Diels-Alder reaction fullerene C

Language: Английский

---