Add compute implementation for surface geometry terms #884
Conversation
| benchmark_name | dt(%) | dt(s) | t_new(s) | t_old(s) |
| -------------------------------------- | ---------------------- | ---------------------- | ---------------------- | ---------------------- |
test_build_transform_fft_lowres | -7.83 +/- 2.88 | -1.06e-03 +/- 3.91e-04 | 1.25e-02 +/- 8.5e-05 | 1.36e-02 +/- 3.8e-04 |
test_build_transform_fft_midres | -4.24 +/- 1.53 | -4.07e-03 +/- 1.46e-03 | 9.19e-02 +/- 1.2e-03 | 9.59e-02 +/- 8.6e-04 |
test_build_transform_fft_highres | -1.30 +/- 0.96 | -6.11e-03 +/- 4.52e-03 | 4.62e-01 +/- 3.0e-03 | 4.69e-01 +/- 3.4e-03 |
test_equilibrium_init_lowres | +1.83 +/- 2.92 | +1.10e-02 +/- 1.75e-02 | 6.12e-01 +/- 1.5e-02 | 6.01e-01 +/- 9.0e-03 |
test_equilibrium_init_medres | -0.08 +/- 2.21 | -9.91e-04 +/- 2.78e-02 | 1.26e+00 +/- 2.0e-02 | 1.26e+00 +/- 2.0e-02 |
test_equilibrium_init_highres | +0.16 +/- 0.96 | +6.47e-03 +/- 4.02e-02 | 4.17e+00 +/- 3.2e-02 | 4.17e+00 +/- 2.4e-02 |
test_objective_compile_dshape_current | +3.49 +/- 11.20 | +1.53e-01 +/- 4.90e-01 | 4.52e+00 +/- 3.2e-01 | 4.37e+00 +/- 3.7e-01 |
test_objective_compile_atf | +0.45 +/- 10.83 | +4.26e-02 +/- 1.02e+00 | 9.49e+00 +/- 8.4e-01 | 9.45e+00 +/- 5.8e-01 |
test_objective_compute_dshape_current | +2.87 +/- 2.67 | +6.18e-05 +/- 5.73e-05 | 2.21e-03 +/- 4.8e-05 | 2.15e-03 +/- 3.2e-05 |
test_objective_compute_atf | -0.05 +/- 2.66 | -3.98e-06 +/- 2.04e-04 | 7.66e-03 +/- 6.7e-05 | 7.66e-03 +/- 1.9e-04 |
test_objective_jac_dshape_current | -2.59 +/- 14.48 | -1.20e-03 +/- 6.70e-03 | 4.50e-02 +/- 4.3e-03 | 4.62e-02 +/- 5.1e-03 |
test_objective_jac_atf | +1.13 +/- 4.09 | +2.54e-02 +/- 9.19e-02 | 2.28e+00 +/- 8.0e-02 | 2.25e+00 +/- 4.6e-02 |
test_perturb_1 | -0.09 +/- 16.02 | -8.05e-03 +/- 1.40e+00 | 8.75e+00 +/- 9.6e-01 | 8.76e+00 +/- 1.0e+00 |
test_perturb_2 | -0.08 +/- 3.92 | -1.23e-02 +/- 5.85e-01 | 1.49e+01 +/- 2.9e-01 | 1.49e+01 +/- 5.1e-01 | |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #884 +/- ##
==========================================
- Coverage 94.91% 94.91% -0.01%
==========================================
Files 80 80
Lines 19947 19966 +19
==========================================
+ Hits 18933 18951 +18
- Misses 1014 1015 +1
|
f0uriest
requested review from
ddudt,
dpanici,
rahulgaur104,
kianorr,
unalmis and
YigitElma
| ) | ||
| def _a_major_over_a_minor(params, transforms, profiles, data, **kwargs): | ||
| A = transforms["grid"].compress(data["A(z)"], surface_label="zeta") | ||
| P = transforms["grid"].compress(data["perimeter(z)"], surface_label="zeta") | ||
| # derived from Ramanujan approximation for the perimeter of an ellipse |
There was a problem hiding this comment.
Does this always work? What is the shape is like a triangle?
There was a problem hiding this comment.
it's always an approximation for any non-elliptical cross section, since the definitions of major and minor axes and elongation aren't uniquely defined in those cases. In practice it still gives reasonable results in all the cases we've tried it. For a triangular cross section you'd basically be getting the elongation of an equivalent elipse which would likely be close to 1, which is what we'd expect.
| x = eq.compute(key)[key].max() # max needed for elongation broadcasting | ||
| y = eq.surface.compute(key)[key].max() | ||
| if key == "a_major/a_minor": | ||
| rtol, atol = 1e-2, 0 # need looser tol here bc of different grids |
There was a problem hiding this comment.
If I increase the resolution of both the grids, can I reduce rtol and still pass the test?
There was a problem hiding this comment.
in theory yes. The main difference is that the eq.compute uses a quadrature grid to compute the perimeter, which doesn't include nodes at rho=1, so we extrapolate what the true perimeter is based on an approximation from rho~0.9x. As you increase the resolution, the outermost node in the quadrature grid will approach rho=1 but you'd need to go to really high resolution.
FourierRZToroidalSurfaceElongationandAspectratio objectives to target surfaces as well as equilibriaResolves #795