Closed kevinhkhuynh closed 2 years ago
u = [1.02, 1.03, 1.04, 1.06, 1.07, 1.08] (changing rdim to be less than e-6 error) u_test = [1.05] tf = 10.0 (1000 time steps) -rdim 40 rel. error = 0.0117 Relative l2 error of global ROM solution 0.0002377676966473762
u = [1.0, 1.05, 1.15, 1.2] (changing rdim to be less than e-6 error) u_test = [1.1] tf = 10.0 (1000 time steps) -rdim 80 rel. error = 0.147 Relative l2 error of global ROM solution 0.0003884699514716798
u = [1.0, 1.025, 1.05, 1.075, 1.125, 1.15, 1.175, 1.2] (changing rdim to be less than e-6 error) u_test = [1.1] tf = 10.0 (1000 time steps) -rdim 80 rel. error = 0.0799
u = [1.0, 1.2, 1.4, 1.6, 1.8, 2.0] (changing rdim to be less than e-6 error) u_test = [1.5] tf = 10.0 (1009 time steps) -rdim 40 rel. error = inf tf = 10.0 (1000 time steps) -rdim 80 rel. error = 510.846 Relative l2 error of global ROM solution 0.0002546485578641565
u = [1.02, 1.03, 1.04, 1.06, 1.07, 1.08] (changing rdim to be less than e-6 error) u_test = [1.05] tf = 10.0 (1000 time steps) -rdim 40 rel. error = 0.0117
u = [1.0, 1.025, 1.05, 1.075, 1.125, 1.15, 1.175, 1.2] (changing rdim to be less than e-6 error) u_test = [1.1] tf = 1.0 (100 time steps) -rdim 80 rel. error = 0.02409
u = [1.0, 1.05, 1.1, 1.15, 1.2, 1.30, 1.35, 1.4, 1.45, 1.5] (changing rdim to be less than e-6 error) u_test = [1.25] tf = 0.5 (50 time steps) -rdim 30 rel. error = 0.0459
It seems like matrix interpolation is having trouble getting below 1% error as the parameter distance increases.
When this PR goes through, I can update the spack repo to fix the libROM package there (already made the changes, but it depends on this PR to work correctly since some files are moved around).
Files of interest are :
Interpolater.h Interpolater.C VectorInterpolater.h VectorInterpolater.C MatrixInterpolater.h MatrixInterpolater.C examples/prom/dg_advection_local_rom_matrix_interp.cpp
If anyone wants to read up on matrix interpolation, it is section 5 of this paper: https://arxiv.org/abs/1506.07849
All Laghos tests pass.
@siuwuncheung when you get a chance, would you review this PR?