Current Research

Quantum-chemical Simulation of Large Molecules

OUTLINE
We are developing new quantum chemical methods and a new supercomputer to predict the properties of molecules and materials. We are working on a quantum theory based on reduced density matrices, a new hybrid QM/MM method, and a special-purpose computer for molecular simulations.
TOPICS
(1) Development of a new quantum theory based on reduced density matrices (RDMs)
The electronic wave function, the solution of the Schrodinger equation, completely determines the behavior of electrons in matter. However, due to electron repulsion the exact wave function is beyond our reach. Since only two-body forces exist in this world, RDMs, which are much simpler mathematical objects than wave functions, contain the same information as wave functions. We have developed a quantum theory in which RDMs serve as the basic variables. This density matrix theory is much simpler than the density functional theory widely used in first-principle simulations. Using the quantum field theory we proposed a systematic method to determine the correlation energy functional, the remaining unknown piece. It will become a new standard to accurately and efficiently predict the properties of matter.
(2) Ab initio simulation of biomolecules by quantum mechanical/molecular mechanical (QM/MM) method
The rapid progress of structural biology continues to increase the importance of quantum chemical simulations of biomolecules. An important issue is the elucidation of the enzyme reaction mechanism. Due to the huge size and the high accuracy required, QM/MM is the only practical choice. In this method only the important portion is treated by quantum mechanics, while the rest is treated with classical mechanics. A difficult problem is the junction between the two subsystems. Several crude empirical models have been proposed. To eliminate the empirical elements we determine the QM/MM model theoretically by successive elimination of the variables in the MM part by renormalization. This new method enables fast yet accurate simulation of large molecules.
(3) Quantum chemistry on the single-instruction multiple-data stream (SIMD) parallel computer
Recent CPUs need fast internal memory (cache) to compensate for the slow speed of external memory. Thus most of the transistors are used to implement it. Colleagues Ebisuzaki (RIKEN) and Makino (University of Tokyo) have proposed improved architecture for scientific calculation by accumulating a thousand simple CPUs on a chip instead of a large cache. The performance of this SIMD computer will reach tera flops, which is one or two order of magnitude faster than the present ones. To perform molecular simulations on this SIMD computer, we devised a completely different algorithm because the architecture is greatly different from other parallel computers. Within two years we can perform quantum and classical molecular simulations on this special-purpose computer.
FUTURE WORK
By combining these studies everyone will be able to perform large-scale molecular simulation much faster and more accurately. This will contribute to both pure and applied sciences, including nanotechnology, material design, structural biology, and drug design.