Revisiting N$_2$ with Neural-Network-Supported CI

Yorick L. A. Schmerwitz,Louis Thirion,Gianluca Levi,Elvar Ö. Jónsson,Pavlo Bilous,Hannes Jónsson,Philipp Hansmann

2024-06-12

Abstract:We apply a recently proposed computational protocol for a neural-network-supported configuration interaction (NN CI) calculation to the paradigmatic N$_2$ molecule. By comparison of correlation energy, binding energy, and the full dissociation curve to experimental and full CI benchmarks, we demonstrate the applicability and robustness of our approach for the first time in the context of molecular systems, and offer thereby a new complementary tool in the family of machine-learning-based computation methods. The main advantage of the method lies in the efficiency of the neural-network-selected many-body basis set. Specifically, we approximate full CI results obtained on bases of $\approx 10^{10}$ Slater Determinants with only $\approx10^{5}$ determinants with good accuracy. The high efficiency of the NN CI approach underlines its potential for broader applications such as structural optimizations and even computation of spectroscopic observables in systems for which computational resources are a limiting factor.

Chemical Physics

What problem does this paper attempt to address?

This paper mainly discusses how to use the neural network-supported configuration interaction (NN CI) method to accurately determine the electronic ground state of molecular systems, especially for nitrogen molecule (N2). The traditional full configuration interaction (fCI) method is limited by the exponential growth of the Hilbert space, and accurate calculation of the binding energy of diatomic molecules like N2 is crucial for understanding bonding properties. In this paper, the authors applied a recently proposed computational protocol that uses convolutional neural networks to select many body basis sets and iteratively expand them to reduce the number of Slater determinants that need to be considered, thus efficiently approximating the fCI results. They compared the correlation energy, binding energy, and dissociation curves with experimental and fCI benchmarks, demonstrating the feasibility and robustness of this approach in molecular systems. The research results show that the NN CI method can approximate the fCI results obtained from about 10^10 Slater determinants using approximately 10^5 Slater determinants, demonstrating its high efficiency. This indicates the potential of the NN CI method in calculations of structural optimization and spectroscopic observables, especially in cases of limited computational resources. The paper also introduces the methods for calculating the one- and two-electron integrals using the Hartree-Fock approach and describes in detail the workflow of the neural network-supported CI algorithm, including a strategy based on active learning and the use of convolutional neural networks for the selection of Slater determinants. In conclusion, this paper proposes a new machine learning-assisted approach for effectively dealing with complex problems in quantum chemistry, particularly overcoming the "exponential wall" challenge faced by the full configuration interaction method.

Revisiting N$_2$ with Neural-Network-Supported CI

Neural-network-supported basis optimizer for the configuration interaction problem in quantum many-body clusters: Feasibility study and numerical proof

A neural network approach to running high-precision atomic computations

Deep-learning approach for the atomic configuration interaction problem on large basis sets

Selected non-orthogonal configuration interaction with compressed single and double excitations

Reducing the computational load – atomic multiconfiguration calculations based on configuration state function generators

Pair-correlated Configuration Interaction Method and Its Approximate Version for Solving the Electron Correlation Problem in Molecules

Massive-Parallel Implementation of the Resolution-of-Identity Coupled-Cluster Approaches in the Numeric Atom-Centered Orbital Framework for Molecular Systems

Improved Optimization for the Neural-network Quantum States and Tests on the Chromium Dimer

Go Green: Selected Configuration Interaction as a More Sustainable Alternative for High Accuracy

Comparison of many‐particle representations for selected‐CI I: A tree based approach

A New Angle on Benchmarking Noncovalent Interactions

Spin-Adapted Externally Contracted Multireference Configuration Interaction Method Based on Selected Reference Configurations

Ab initio Accuracy Neural Network Potential for Drug-like Molecules

Multi-dimensional neural network interatomic potentials for CO on NaCl(100)

Exhaustive state-specific dissociation study of the N2(Σg+1)+N(S4) system using QCT combined with a neural network method

Quantum Monte Carlo for Noncovalent Interactions: An Efficient Protocol Attaining Benchmark Accuracy

Seniority and Hierarchy Configuration Interaction for Radicals and Excited States

Ab Initio Molecular Dynamics Simulations of Atmospheric Molecular Clusters Boosted by Neural Networks

Accelerating Configuration Interaction Calculations for Nuclear Structure

Stochastic Resolution of Identity to CC2 for Large Systems: Ground State and Triplet Excitation Energy Calculations