WENO5 running, on boundaries reduces to first order

[dere4]

for now, need to add verification and optimize, waste of resources rn

[dere4]

https://www.sciencedirect.com/science/article/pii/S002199910096443X https://www.tandfonline.com/doi/abs/10.1080/1061856031000104851

used these as a reference. First one recommends LLE, Roe or RF flux as building blocks. Its implemented now that either HLLE, rusanov or upwind can be used as building blocks. Need to yet do OVS. Also probably worth implementing the third order scheme, uses smaller stencil. Also talks about how Roe averaging or Lax-Friedrichs splitting recommended for hyperbolic conservation laws, thats what our HLLE and rusanov building blocks take care of.

[dere4]

takes 140s for me now for 1ms, can be cut quite a lot I think

[archermarx]

OVS is currently failing for neutrals and ions

[archermarx]

takes 140s for me now for 1ms, can be cut quite a lot I think

how many cells?

[dere4]

140 seconds 100 cells

[dere4]

Tests are running now. Added WENO as argument to HyperbolicScheme and implemented your suggestions. order of accuracy for WENO 1 now cause still didnt care about boundaries. next step:

do OVS with weno and periodic boundaries so that high order scheme applied everywhere and see if it actually gets to order of accuracy 5. once successful and confirmed that WENO is implemented correctly, implement higher order on boundaries as well.

[archermarx]

How's the runtime? And does the number of allocations increase with increasing simulation time?

[archermarx]

Fixed minor merge conflicts

[dere4]

yep number of allocations increases with simtime. twice the simtime, twice the allocations so quite bad. will investigate

[dere4]

Runtime and allocation discrepancy, WENO vs no WENO:

julia> sol = run_sim(1e-3; ncells=100, nsave=10000, case = 1, dt = 1.3e-8, flux = HallThruster.rusanov, WENO = true) 18.502903 seconds (358.37 M allocations: 8.834 GiB, 11.05% gc time) Hall thruster solution with 10000 saved frames Retcode: Success End time: 0.001 seconds

julia> sol = run_sim(1e-3; ncells=100, nsave=10000, case = 1, dt = 1.3e-8, flux = HallThruster.rusanov, WENO = false) 3.661543 seconds (212.10 k allocations: 165.046 MiB) Hall thruster solution with 10000 saved frames Retcode: Success End time: 0.001 seconds

[dere4]

Also WENO does not seem to be compatible with HLLE, rusanov with the splitting might be required I think

[archermarx]

ok, yeah we're still allocating somewhere. also tests are broken but it's my fault this time. will fix on my end soon

[archermarx]

tried to fix tests, let's see if it worked

[dere4]

fixed allocations. almost same runtime for WENO on and off. also fixed WENO neutrals.

[dere4]

julia> sol = run_sim(0.5e-3; ncells=100, nsave=10000, case = 1, dt = 1.3e-8, flux = HallThruster.rusanov, WENO = true) 2.282685 seconds (212.11 k allocations: 165.047 MiB, 1.24% gc time) Hall thruster solution with 10000 saved frames Retcode: Success End time: 0.0005 seconds

julia> sol = run_sim(0.5e-3; ncells=100, nsave=10000, case = 1, dt = 1.3e-8, flux = HallThruster.rusanov, WENO = false) 1.881572 seconds (212.71 k allocations: 165.084 MiB, 3.53% gc time, 0.19% compilation time) Hall thruster solution with 10000 saved frames Retcode: Success End time: 0.0005 seconds

[dere4]

tests passing on my end

[archermarx]

i broke ionization tests when i added krypton support, fixed that

[archermarx]

Fixed tests maybe for real this time