Analytical gradients for tensor hyper-contracted MP2 and SOS-MP2 on graphical processing units.

Area of Science:

• Computational chemistry
• Quantum chemistry
• Theoretical chemistry

Background:

• Accurate calculation of molecular properties is crucial for understanding chemical reactions and material properties.
• Traditional methods like Møller-Plesset perturbation theory (MP2) face significant computational cost for large systems.
• Tensor hyper-contraction (THC) offers a promising approach to reduce this computational burden.

Purpose of the Study:

• To derive and implement analytic energy gradients for MP2 and SOS-MP2 using the THC approximation.
• To reduce the computational scaling of gradient calculations for these methods.
• To demonstrate the accuracy and efficiency of the THC-MP2 implementation for molecular simulations.

Main Methods:

• Derivation of analytic energy gradients for THC-MP2 and THC-SOS-MP2.
• Implementation utilizing graphics processing units (GPUs) and sparse tensor techniques.
• Application to geometry optimization and ab initio molecular dynamics (AIMD) simulations.

Main Results:

• Reduced formal scaling of MP2 and SOS-MP2 gradient calculations to quartic and cubic, respectively.
• Efficient implementation leveraging GPUs and sparse tensor methods.
• Accurate energy conservation in micro-canonical AIMD simulations, validating the nuclear gradients.

Conclusions:

• The developed THC-MP2 gradient implementation provides accurate results for geometry optimizations and AIMD.
• This method significantly improves the efficiency of MP2 and SOS-MP2 calculations for larger systems.
• The approach is suitable for large-scale quantum chemistry simulations and molecular dynamics.