Data-Driven Acceleration of the Coupled-Cluster Singles and Doubles Iterative Solver.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Machine Learning Applications

Background:

• Solving coupled-cluster (CC) equations is computationally expensive and scales poorly with system size.
• Current methods require iterative determination of amplitudes to minimize system energy.
• Accurate quantum chemical calculations are crucial for understanding molecular behavior.

Purpose of the Study:

• To develop a novel machine learning approach for predicting coupled-cluster singles and doubles (CCSD) amplitudes.
• To accelerate the computation of the coupled-cluster wave function.
• To explore accurate energy calculations by potentially bypassing traditional CC equation solving.

Main Methods:

• Utilized machine learning models trained on electronic structure properties from the MP2 level.
• Extracted features including orbital energies, one-electron Hamiltonian, and Coulomb/exchange terms.
• Developed a data-driven CCSD (DDCCSD) method that still solves the actual CC equations.

Main Results:

• Demonstrated the potential for significant speedups in solving CCSD equations.
• Showcased that accurate energetics can be achieved by bypassing iterative CC equation solving.
• Preliminary data indicate the effectiveness of encoding physical principles in ML models.

Conclusions:

• The data-driven CCSD (DDCCSD) approach offers a promising avenue for accelerating quantum chemistry calculations.
• Machine learning, when informed by physical principles, can effectively predict complex quantum chemical properties.
• This methodology has the potential to make high-level electronic structure calculations more accessible.