Related Experiment Video
Updated: Jun 27, 2026

Targeting Neuronal Fiber Tracts for Deep Brain Stimulation Therapy Using Interactive, Patient-Specific Models
Published on: August 12, 2018
Using neural networks, optimized coordinates, and high-dimensional model representations to obtain a vinyl bromide
Sergei Manzhos1, Tucker Carrington
1Département de chimie, Université de Montréal, Case postale 6128, succursale Centre-ville Montréal, (Québec) H3C 3J7 Canada. sergei.manzhos@gmail.com
High-dimensional model representations (HDMRs) combined with neural networks (NNs) efficiently generate accurate molecular potentials. Optimized redundant coordinates simplify complex potentials into a sum-of-products form, crucial for quantum dynamics calculations.
Area of Science:
- Computational Chemistry
- Quantum Mechanics
- Machine Learning
Background:
- Accurate potential energy surfaces are essential for simulating molecular dynamics.
- High-dimensional model representations (HDMRs) offer a way to approximate these surfaces.
- Traditional HDMR methods face challenges with the combinatorial explosion of terms in high dimensions.
Purpose of the Study:
- To develop an efficient method for constructing high-dimensional potential energy surfaces.
- To leverage neural networks (NNs) within the HDMR framework.
- To enable the representation of complex potentials in a form suitable for quantum dynamics.
Main Methods:
- Utilizing neural networks (NNs) to fit high-dimensional model representations (HDMRs).
- Employing optimized redundant coordinates to reduce the number of terms in the HDMR.
- Using exponential neurons to achieve a sum-of-products form for the potential.
- Fitting 12- and 15-dimensional potential surfaces for vinyl bromide.
Main Results:
- Demonstrated successful fitting of 12- and 15-dimensional potentials using NN-based HDMRs.
- Showed that redundant coordinates effectively bypass the combinatorial complexity of traditional methods.
- Achieved a sum-of-products potential representation comparable in quality to full-dimensional fits.
- Reduced computational cost by building the full surface from lower-dimensional (e.g., 6D) NN fits.
Conclusions:
- NN-based HDMRs provide an effective approach for generating accurate potential energy surfaces.
- The sum-of-products form obtained is highly beneficial for quantum dynamics simulations.
- Optimized redundant coordinates are key to managing the complexity of high-dimensional problems.
Related Concept Videos
Newman Projections
The organic molecules rotate across the single bonds leading to numerous temporary three-dimensional structures of varying energy known as conformers.
Fischer Projections
Predicting Molecular Geometry
Hybridization of Atomic Orbitals I
Molecular Models
Valence Bond Theory
