Compact wave function methods for chemical systems

点击翻译按钮获取中文摘要

With this award, the Chemical Theory, Models and Computational Methods program in the Chemistry Division is supporting Dr. Alan Aspuru-Guzik of Harvard University to develop an efficient and accurate method to study the electronic structure of molecules and solids that are difficult to study by existing approaches. The Computational and Data-Enabled Science and Engineering programs in Chemistry and Physics, as well the CISE/ACI CIF21 Venture Fund for Software Reuse will contribute to the award. The exact quantum mechanical description of the electronic structure of atoms, molecules and solids is computationally intractable. Accurate computational descriptions that can be carried out with current computers for systems of practically relevant sizes remains a challenge. If available, accurate descriptions promise a transformative acceleration in the cycle of materials design. The proposed work advances a new idea for the compact representation of the wave functions of atoms, molecules and materials by borrowing ideas of the field of signal processing. The Aspuru-Guzik group is working to optimize the computer code, and then to release it both as open source and incorporated in popular quantum chemistry computer packages. The ideas and software developed in this project will be incorporated into a massive open online course (MOOC), entitled " The Quantum World", that Aspuru-Guzik is developing under the HarvardX platform. A technique borrowed from the field of signal processing, matching pursuit, is combined with a non-orthogonal orbital expansion. Preliminary results show that the method is competitive with the state-of-the-art methods but still requires considerable software development and theoretical extension. The main objective of the current project is to advance the proposed method, namely the non-orthogonal multicomponent adaptive greedy iterative compression (NOMAGIC). NOMAGIC already has shown excellent wave function compression of the order of tens of determinants for obtaining chemically accurate solutions. Aspuru-Guzik and coworkers are developing the NOMAGIC method as a black box method to be incorporated with several electronic structure packages. The researchers are developing methods for excited state simulation as well as to extend the ansatz considered by the approach to tensor networks.