Significant advances in quantum chemical simulations of large systems

Yesterday at the Condensed Phase Dynamics meeting in Telluride there we nice talks by Eran Rabani and Todd Martinez highlighting recent advances they have independently made in simulating large systems. Rabani is largely concerned with solid systems such as silicon nanoparticles and Martinez with hundreds of reacting molecules.

Rabani's work with Roi Baer and Daniel Neuhauser is a highly creative and original stochastic approach to doing electronic structure calculations.  Instead of explicitly determining wave-functions [or Kohn-Sham orbitals in DFT] they use stochastic wave functions [a random phase is assigned to each point on a spatial grid] and compute one-particle expectation values as averages over the ensemble of stochastic wave-functions.

For Kohn-Sham Density Functional Theory they obtain sub-linear scaling of the computational cost with system size, as reported in this PRL. Some of the original problems with the approach [slow converge for systems involving weakly coupled sub-systems, e.g. a pair of buckyballs] are solved with an embedded fragment approach, as described in this recent preprint.

Over the past decade Martinez' group has pioneered the use of Graphics Processing Units (GPUs) [think computer games] for quantum chemistry calculations. The figure below from this recent paper highlights the computational speedup compared to conventional codes and architectures.

One can now do a DFT-based calculation of the energy of about one thousand water molecules in less than one minute. The bottle neck is the cubic scaling of diagonalisation routines.

Todd described recent work "discovering chemical reactions with the nano reactor".

Most quantum chemistry calculations require "knowing the answer" before you begin. i.e, you know the reactants and the products and you make certain guesses as to transition states between them. However, given a set of reactants is it possible to have an automated computational procedure that allows discovery of new products and the mechanisms for their production?
The answer is yes, and this represents a significant advance.

Todd presented dynamical simulations of a set of 100-300 atoms for 1 nanosecond at temperatures of about 1000-1500 K.

This high temperature is used to reduce the effects of non-covalent interactions and allow many reactions to proceed within a nanosecond.

Indeed, one does discover new reactants that one would not just predict based on a knowledge of "freshman" chemistry. Furthermore, one can find the reaction paths.

They have also simulated the famous and controversial Urey-Miller experiment aimed at producing amino acids [the building blocks of proteins] from simple initial reactants [water, methane, ammonia, hydrogen].
The simulation did find glycine is formed, and via three different reaction pathways.