[Triumf-seminars] TRIUMF Theory Seminar today at 13:00

Wed Apr 13 05:00:00 PDT 2022

Date/Time: Wed 2022-04-13 at 13:00

Location:  Remote              

Speaker:   Mehdi Drissi (TRIUMF)

Title:     Improving variational Monte-Carlo methods with deep neural networks

Abstract: Variational Monte-Carlo (VMC) is a well-founded method aiming at solving the quantum many-body problem via the minimization of the expectation value of an Hamiltonian for a given class of trial wave-functions. In the last couple of years, the traditional ansatzes used in VMC have been revisited with the idea of using artificial neural networks (ANNs) to parametrize the trial wave-functions [1]. This has lead to tremendous progress, in particular in quantum chemistry [2,3,4] where VMC calculations can now reach systems with up to 30 electrons with an accuracy on par or better than state-of-the-art coupled-cluster calculations with single, double and perturbative triple excitations (CCSD(T)).

In this talk, I will discuss our recent attempts to further improve this technology in order to increase the reach of VMC+ANNs for general quantum many-body systems. After briefly reviewing the general idea of VMC, I will discuss how ANNs can be used to parametrize the wave-function of many fermions. In particular, I will focus on a 1D spinless toy-model we are using to test different computing strategies. Specific emphasis will be put on the on-going development of a promising new optimizer, specifically tailored to VMC+ANNs calculations.

[1] Carleo and Troyer, Science, 355, p. 602 (2017)
[2] Pfau, Spencer, Matthews and Foulkes, Physical Review Research, 2(3), 033429 (2020).
[3] Choo, Mezzacapo and Carleo, Nature Communication 11, 2368 (2020)
[4] Hermann, Schätzle and Noé, Nature Chemistry 12, 891–897 (2020)

https://ubc.zoom.us/j/62535266430?pwd=THVrUzhMdjAyVkNnM05OUWRZYUppZz09
Meeting ID: 625 3526 6430
Passcode: 307647

______________________________

General TRIUMF seminar information available at http://www.triumf.ca/home/upcoming-events/about-seminars-lectures