Toward DMC Accuracy Across Chemical Space with Scalable Δ-QML

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Machine Learning

Background:

• Quantum diffusion Monte Carlo (DMC) accurately predicts molecular energetics and properties by solving the Schrödinger equation.
• DMC's O(N^3) scaling limits its application to larger systems, hindering its role as a reference method.
• Traditional methods like CCSD(T) are computationally expensive for large systems.

Purpose of the Study:

• To assess the accuracy of DMC for smaller molecules as a foundation for its use in larger systems.
• To explore the potential of combining quantum Monte Carlo (QMC) with quantum machine learning (QML) to alleviate computational burdens.
• To develop high-quality computational chemistry descriptions across diverse chemical spaces.

Main Methods:

• Utilizing quantum diffusion Monte Carlo (DMC) for accurate energy calculations.
• Employing quantum machine learning (QML) surrogate models to reduce computational cost.
• Implementing three key approximations: fixed-node approximation, accurate references for bond dissociation energies, and scalable amons-set-based QML (AQML) models.

Main Results:

• Converged DMC results were obtained for over 1000 small organic molecules (up to five heavy atoms) and 50 medium-sized organic molecules (nine heavy atoms).
• AQML models were validated using these DMC results.
• Numerical evidence for Δ-AQML models indicates that modest QMC training datasets of amons are sufficient for predicting total energies with near chemical accuracy.

Conclusions:

• Coupling QMC with QML significantly alleviates computational demands, making QMC a viable reference method.
• The developed AQML models demonstrate the potential for accurate predictions across chemical space with reduced computational cost.
• This approach paves the way for high-quality computational chemistry descriptions of larger and more complex systems.