validation

We gather here the tests that users can perform to check that their installation of coral provides accurate results.

The Fourier transforms

The 2decomp&fft installation can be tested separately.

Reference benchmarks

Benchmark1 can be run as a quasi-automated test for linear solves. After running for about 20 minutes with a handful of cores, run python linearGrowthrate_numerical_vs_analytical.py. During the initial period of exponential growth, the growthrate of kinetic energy (blue line) should asymptote the analytic prediction (black dashed line).

Benchmark2 should be run to check that two mathematically equivalent formulations of the same problem (namely solving the incompressible Navier-Stokes equations using either a velocity-pressure formulation or a toroidal-poloidal decomposition) yield a common value for the transport coefficient of the flow. Running python compute_nusselt.py, one should obtain a Nusselt number fluctuating around 4.76 for both formulations.

Benchmark3 is a little more involved computationally. When computed over a long enough period of time (say 48 hours on a 48-cores compute node), the flow enters the so-called saturated dynamo regime described in the reference paper BCG20. Running python compute_Elsasser_Rm.py users may check that, in this regime, time series of kinetic and magnetic energy fluctuate around the values reported in BCG20: Rm = 131 ± 20 and Elsasser = 0.20 ± 0.04 for kinetic and magnetic energy, respectively (see their table 1).