Northwestern Logo

Shiozaki Group :: Research Interests

Predictive Electronic Structure Theory for Complex Systems

At the intersection of chemistry, physics, and computer science, we seek to solve challenging problems of chemical and materials interests by developing quantum theories and computer codes.

Below, you will find some examples we are currently interested in. Most of them will be investigated partly through collaboration with Northwestern's great experimental and theory colleagues.

» Model Hamiltonians for Exciton Dynamics from Many-Body Wave Functions

ASD We have developed a method, called active space decomposition (ASD), which uses molecular geometries to compress active-space wave functions. This new method not only realizes active space computation of unprecedented size, but also allows us to extract few-state model Hamiltonians of exciton/electron dynamics in the quasi-diabatic representation. We have also shown a natural link between ASD and low-entanglement wave functions by demonstrating a multi-state analogue of ASD using a density matrix renormalization group algorithm. The method has been applied to, e.g., singlet fission processes in organic crystals.

» Heavy-Element Chemistry with Fully Relativistic Electronic Structure Theory

Dirac Simulation of properties of molecules containing heavy elements is a great challenge since spin–orbit and spin–spin couplings are too large to be treated as perturbations. Therefore, we must use electronic structure theories that are based on the four-component Dirac equation.

We are developing an efficient parallel program for fully relativistic 4-component electronic structure theory to simulate properties of molecules with 100 atoms and a few heavy elements. For instance, we are in collaboration with Danna's group to discover new magnetic molecules and materials that would replace lanthanide magnets, whose supply is limited for many reasons.

» Electronic structure of extended systems

We are interested in electronic structures of adsorbed molecules on surfaces. Usually they are treated by DFT, often with plane waves. However, especially for strongly correlated systems, there is a need for rigorous computation on the basis of many-body wave functions. We are currently working on extending real-space renormalization group formalism (such as contractor renormalization group) to ab initio Hamiltonians.

» Massively Parallel Codes

Besides our imagination, computer codes are among the most important tools; and it is important to have in-house programs that are useful for problems you are interested in. We are primarily working on a in-house program written by the group members, which features a very competitive multireference electron correlation implementation. Technically speaking, our code is object-oriented and very flexible to future additions of new codes.

Top500

Our codes are tailored specifically for massively parallel computing environments (more specifically, a large number of CPUs and a huge distributed memory space). See the development of the supercomputer performance above (taken from top500.org). In addition to our own InfiniBand'ed 1,000 core machine (to be purchased), we have possibility of using Northwestern's supercomputer 'Quest,' national facilities, and those overseas.

» Automated Implementation Technique

Conventionally, many efforts in the electronic structure theory community have been devoted to computer code implementations of awfully complicated equations that arise from relatively simple Ansätze. In the past decade, the schemes based on automated symbolic algebra have been reported by Kállay, by Hirata, and by others to circumvent this problem. We further advance this strategy so that it can handle spin-free multireference theories. As a consequence, the burden of programming is minimized in our group.

SMITH

...No worries, you can still code a lot (with much improved scientific productivity).