SANTANA Project: Exploring algorithms for accelerating Newton-Krylov RANS 
convergence in CODA solver

Within the Algo-COOP team, the Algo group conducts research in the fundamentals 
of high performance simulation. This includes a wide range of topics in applied 
mathematics, such as scalable algorithms in numerical linear algebra, iterative 
and direct algorithms for large linear systems, novel methods for solving 
partial differential equations, data assimilation, optimisation, uncertainty 
quantification and scientific machine learning.

This research project will investigate iterative solvers for use in the software 
suite CODA developed by Onera, Airbus and DLR. CODA is a CFD solver for aircraft 
design and it features innovative algorithms as well as advanced software 
technology concepts dedicated to HPC. The Spliss library, used in CODA to handle 
linear algebra problems, provides an excellent framework for parallel computing. 
However, the high number of iterations currently observed for the block-Jacobi 
preconditioned Newton-Krylov solver clearly indicates that further
improvements are necessary. In this project, we will carefully analyze and 
identify bottlenecks using the mathematical/physical structure of the system and 
its HPC implementation.

The linear solvers used for time integration interact with many of the 
ingredients of the overall algorithm, and a better understanding of these 
interdependencies is needed to consider an overall improvement in CODA's
performance. We will systematically analyze the impact of the way linear systems 
are implemented (e.g. differentiation, integration of forces and source terms, 
linearization) the impact of line-search and CFL relaxation procedures on the 
convergence properties of linear systems, particularly for turbulence models. 
This will enable us to identify algorithms that can be expected to converge 
rapidly. In addition to manual fine-tuning of algorithms and solver interfaces, 
we will explore the possibilities of exploiting randomized linear algebra 
techniques. This will enable us to exploit synergies with other research 
projects currently underway at CERFACS, where this axis is being explored in a 
related but different context. The post-doc researcher in this position is 
expected to collaborate closely with another researcher working on the same 
project.

Skills required: 
•You have defended your thesis less than 3 years from the date of this job 
offer.
• PhD in applied mathematics, fluid dynamics, computer science, iterative 
solvers, numerical linear algebra or
high performance computing.
• Knowledge of pre-conditioners applied to large-scale problem
• Proficiency in programming languages C++ and Python.
• Familiarity with HPC environments and performance optimization.


To apply, please send your CV and covering letter to mpkumar@cerfacs.fr, 
applications are open until 03/03/2025.

https://cerfacs.fr/en/offer/santana-project-exploring-algorithms-for-
accelerating-newton-krylov-ransconvergence-in-coda-solver/