xtensor
and Eigen
Kokkos
implementation will be providedThe algebra is not complete in samurai.
For example, it is not possible to write
A comprehensive rewrite of the system is underway to address these limitations.
With this new implementation, we can write:
The goal is to evenly divide the workload amongst processors.
Mesh parts (and associated data) must be exchanged between neighbours.
Loïc Strafella’s work
What was achieved?
What remains?
without: \(\epsilon = 1e-2\), \(\underline{\ell} = 2\), \(\bar{\ell} = 12\)
with: \(\epsilon = 1e-2\), \(\underline{\ell} = 2\), \(\bar{\ell} = 12\)
What was achieved?
What remains?
2 new recruits: Sébastien Dubois and Alexandre Hoffmann
samurai is available on conda-forge
and conan
.
In the NumPEx context, samurai must also be installed with Spack and Guix. In the next few months, we will provide a way to install it through these package managers.
It’s coming!
… but some issues remain:
An IMEX strategy for stiff PDEs:
Belousov-Zhabotinsky (very stiff source - 3 eq) \[ \left\{ \begin{aligned} \partial_t a - D_a \, \Delta a &= \frac{1}{\mu} ( -qa - ab + fc) \\ \partial_t b - D_b \, \Delta b &= \frac{1}{\varepsilon} ( qa - ab + b\,(1-b)) \\ \partial_t c - D_c \, \Delta c &= b - c \end{aligned} \right. \]
\(\epsilon = 1e-3\), \(\underline{\ell} = 2\), \(\bar{\ell} = 10\)
Requested
Nice-to-have
What you can use now?
What is planned?
Journées OpenNumDef - - 13 and 14 February 2025