Current research in the Martínez Group aims to make molecular modeling both predictive and routine. New approaches to interactive molecular simulation are being developed, in which users interact with a virtual-reality based molecular modeling kit that fully understands quantum mechanics. New techniques to discover heretofore unknown chemical reactions are being developed and tested, exploiting the many efficient methods that the Martínez group has introduced for solving quantum mechanical problems quickly, using a combination of physical/chemical insights and commodity videogaming hardware.
TeraChem is a powerful general-purpose quantum chemistry package developed in our group. By exploiting graphical processing units and designing the code from the ground up to expose parallel concurrency, we are able to accelerate Hartree-Fock and DFT calculations by several orders of magnitude over CPU based programs. This allows ab initio quantum mechanical calculations to be routinely performed on extremely large systems.
Reactions involving multiple electronic states are important in studing light-harvesting molecule, photochemistry, UV damage, isomerization, etc. To simulate these processes we use ab initio methods to calculate the potential energies and couplings. In this approach the motions of the electrons and nuclei are described quantum mechanically
The electron repulsion integral (ERI) tensor is a demanding source of computational complexity in many ab initio methods. In the THC approximation, we reduce this fourth-order tensor to a product of five second-order tensors. This new approximation allows ab initio methods to be evaluated with reduced scaling. We have shown that this reduces the scaling to O(N4) for MP2, MP3, CC2, EOM-CC2, and CCSD. Other efforts to increase computational efficiency include the use of the Hubbard type correction to minimal basis sets.
Due to the computational efficiency of GPUs, it is now possible to perform quantum computations on protein systems. The computations compare favorably to both molecular mechanics simulations as well as experiment, especially for relatively disordered systems.
Chemical understanding is driven by the experimental discovery of new compounds and reactivity, and is supported by theory and computation that provide detailed physical insight. The ab initio nanoreactor is a highly accelerated first-principles molecular dynamics simulation of chemical reactions that discovers new molecules and mechanisms without preordained reaction coordinates or elementary steps. Using the nanoreactor, we can find new pathways which highlight the emergence of theoretical and computational chemistry as a tool for discovery.
A virtual molecular modeling kit is developed based on GPU-enabled interactive ab initio molecular dynamics (MD). The code uses the TeraChem and VMD programs with a modified IMD interface. Interactive models have a long history in chemical research. The venerable ball and stick model, invented more than a century ago, still plays an important role in shaping our understanding of chemistry. In that same spirit, interactive ab initio calculations represent a new synthesis of intuitive human interfaces with accurate numerical methods. AI-IMD can already be applied to systems containing up to a few dozen atoms at the Hartree-Fock level of theory. As computers and algorithms continue to improve, the scope of on-the-fly calculations will continue to widen.