Theory days on Computational tools in physics : performances vs. reproducibility . – (Meeting / LPT), 27-29/11/2024

Organizers : M. Belkacem, P. M. Dinh, and E. Suraud

Aim and objectives

Following the tradition of the previous successful meetings that have started in 2009, the Theory Days workshop aims at stimulating constructive discussions, collaborations, and knowledge exchange among a selected group of theoreticians. This year, the 16th edition will be dedicated to Computational tools in physics : performances vs. reproducibility.

The development of computational codes for structure and dynamics of electronic systems is experiencing new directions of research in relation to the growing availability of artificial intelligence tools. Machine learning is now applied to numerous computational issues in physics and chemistry, such as for example structure optimization in clusters and molecules. The aim of this year meeting is to address this question in the specific context of dynamical problems, on the basis of existing computational softwares. To restrict the field, we shall focus on fully dynamical approaches allowing us to address dynamical scenarios beyond the domain of linear response. The meeting will thus try to summarize today’s situation in the field, by considering dedicated questions raised by dynamical codes.

The problem of reproducibility is becoming a major issue in computational science. The issue is becoming especially important because of the ever increasing complexity of the codes and as computations become bigger and bigger up to a point at which it may become too expansive to make systematic checks. It thus raises the key question of how large differences can be acceptable between results obtained from a given code using various “reasonable’’ parameters and/or various machines. There is a hardware aspect and a physical aspect. We shall try in the meeting to address both aspects on typical examples.

We plan to gather people from various groups working in this domain, mostly in western Europe. The workshop is intended to be a working forum, thus rather informal. The format will consist in a series of long detailed talks, typically 1 hour (40 mn talk + 10 mn for questions) focusing in particular on ongoing problems and open issues. We thus urge participants to prepare their talks in this spirit. Ultimately a most efficient organization would consist in having a limited number of talks per representative group in order to avoid doublings.

see more details : here