Title: Numerical reproducibility and optimization of a Quantum Monte Carlo Kernel Library (QMCkl)
Date: 2020-10-09 10:05
Slug: job_e210f4b289455a0938f31ffd569b9c53
Category: job
Authors: Pablo de Oliveira Castro
Email: pablo.oliveira@uvsq.fr
Job_Type: Post-doctorat
Tags: postdoc
Template: job_offer
Job_Location: Saint-Quentin-en-Yvelines
Job_Duration: 12 mois
Job_Website: https://trex-coe.eu/
Job_Employer: Université de Versailles St Quentin
Expiration_Date: 2021-01-30
Attachment: job_e210f4b289455a0938f31ffd569b9c53_attachment.pdf
Numerical reproducibility and optimization of a Quantum Monte Carlo Kernel Library (QMCkl)
Keywords: High performance computing, quantum chemistry, numerical analysis, reproducibility, floating point, variable precision
Location: Saint Quentin en Yvelines, France.
Type of contract : Post-doc / Research Engineer
Duration: 12 months (extendable)
Starting: November 2020 - January 2021
Salary: 3000€ / month (salaire net)
Candidate profile
Post-doc / Research Engineer with a strong background in high performance computing. Experience in the following fields is a plus but not required: numerical accuracy, compiler development, quantum chemistry. Please send a motivation letter and your CV, including names and contact information of two references.
Context and objectives
TREX (<https://trex-coe.eu/>) is a European Center of Excellence (CoE) federating European scientists, HPC centers, and SMEs. It aims at developing, promoting, and maintaining open-source high-performance software solutions in the field of quantum chemistry, that are ready to take advantage of upcoming exascale architectures.
Quantum Monte Carlo (QMC) methods account for the fully correlated nature of the quantum electronic problem and offer an extremely accurate and systematically improvable solution of the Schrödinger equation. QMC calculations are extremely demanding from the point of view of computational resources and calculations are generally carried
out thanks to the massively parallel character of the underlying algorithms.
The CoE will design QMCkl, a high-performance library for QMC calculations for molecules and solids, which will then be integrated in the codes of the CoE. One goal is to ensure that the library is numerically robust across different architectures and accelerators. Moreover, the library should optimize its floating point representation to take advantage of emerging hardware capabilities. To measure numerical reproducibility and find optimization opportunities, we will leverage Verificarlo (<https://github.com/verificarlo/verificarlo>), a compiler for numerical debugging and optimization built upon LLVM. Verificarlo will be extended to address QMCkl library and ported on the TREX project architectures.
Verificarlo is an active project with many contributors, in particular University of Versailles and Intel are actively contributing to the project.
Location
The University of Versailles Saint-Quentin-en-Yvelines (UVSQ, <http://www.uvsq.fr>) is a major contributor to the creation of the University of Paris-Saclay, especially in the field of Computer Science. UVSQ is strongly involved in the field of High Performance Computing through various research laboratories and industrial projects. The HPC team, now part of Exascale Computing Research (ECR) laboratory and Li-PaRAD (<http://www.liparad.uvsq.fr/>) laboratory is specialized since the nineties in the field of HPC architectures, performance evaluation and optimization tools, intensive numerical computation and advanced communication architectures. ECR is a joint collaboration between CEA, Intel and UVSQ.
Contact
Email: pablo.oliveira@uvsq.fr
Pablo de Oliveira Castro
Laboratoire d'Informatique Parallélisme Réseaux et Algorithmes Distribués (Li-PaRAD)
Saint Quentin en Yvelines, France