Reject numerical ties in field-line return selection#153
Open
krystophny wants to merge 28 commits into
Open
Conversation
isw_qflux_NA=1 with in_file_pert='none' assembles an identically zero non-axisymmetric source. That zero vector is the Arnoldi seed, so its normalization divides by zero (0/0 -> NaN) and zhseqr then fails to converge at every step, spinning without qflux output. Reject the configuration in check() with a clear message instead.
Add an optional right-hand side to the non-axisymmetric solve: the drive of a single-helicity perturbation given by a radial corrugation field (delta B^s/B_0^phi)_mn and an electrostatic potential harmonic Phi_mn. The vpar corrugation piece enters the A1/A2 force channels with velocity weights a3m and x^2 (sigma-odd), the ExB piece with weights x^-1 and a1m (sigma-even); the two new velocity moments are computed in compute_sources. Input via new &ntv_input variables (isw_hel_drive, m_theta_hel, hel_brad_re/im, hel_phim_re/im), default off. Band profiles, moments and dimensional prefactors follow a CAS-checked derivation (see issue #118). Off by default: golden-record QL summary.h5 is reproduced to 1e-15 against a main build.
Review findings on the drive commit: fold the x^(3k-4) weight into a single power so the x=0 quadrature node stays finite; skip the moment quadrature on the precomputed-matrix-element path and reject isw_hel_drive at run time when the moments are unavailable; reject the unvalidated relativistic branch, reconstruction mode and unset species data; deallocate asource_hel in collop_unload.
The zero-source guard added upstream rejects isw_qflux_NA=1 with in_file_pert='none'. With the misalignment drive, that configuration is valid: the drive supplies the non-axisymmetric source. Relax the guard so it fires only when the drive is off or all its amplitudes are zero.
The opposite-helicity drive (m,n) -> (m,-n) enters add_helical_source as the complex conjugate of the band profile (phase convention exp(+i(m theta + n phi))). The assembly stencil is a real-coefficient linear map, so it must conjugate the assembled source pointwise; a sign or phase error in the stencil breaks this. The existing checks use a profile whose real and imaginary parts vary independently, so they do not exercise conjugate symmetry. Add a check that conjg(profile) yields conjg(source) to 1e-14, plus a vacuity guard that the source carries an imaginary part. Verification: - ctest helical_source_test: Passed (0.07 s) - ctest -LE external-data: 18/18 passed - red-before: injecting a complex coefficient (1,0.1) into one stencil term makes the new check error stop with 'conjugate profile did not conjugate the source pointwise' (exit 1)
Decompose the phi-integrated current channel qflux(2,k) = sum_col flux_vector(2,col)*source_vector(col,k) into its per-field-line-point contributions in the mode-1 local (axi) solver, which HELCORE uses. The new pointwise_current_mod reuses the already-verified flux_vector current row: its counter-passing block enters with the explicit minus sign (parallel current is odd under sigma -> -sigma) and the step_factor phi-quadrature weight is baked in, so the decomposition introduces no new sign, weight, or normalization. flux_vector's columns partition into disjoint per-istep blocks, so the block dot products sum back to qflux(2,k) exactly. The solver writes current_profile.dat (phi_mfl, per-drive contributions, and current densities) guarded by isw_hel_drive, so runs without the helical drive are bitwise unchanged. A hard self-check stops the run if the field-line sum does not reproduce qflux(2,k). Verification: - 17/17 TEST ctests pass; new pointwise_current_test added. - Red-before: breaking the block partition (nblk without the npassing+1) gives "ERROR STOP FAIL: per-point contributions do not sum to the qflux current channel"; restored -> Passed. - Local HELCORE simple_l1 run, all three drive cases (nminus_phi, nminus_phi_im, nplus_phi): gamma_current/D31/D32 reproduce the accepted aCluster response (-7.338706/-20.944508, -2.711060/-3.138948, -2.369674/-1.825414) to ~1e-11; each current_profile.dat reconstructs qflux(2,k) to ~1e-15 relative over 961 finite points.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The change leaves field integration, return-distance evaluation, propagator construction, ABI, and memory layout unchanged. It only changes the comparison used to select the terminating return.
Verification
Failing before, with fixed geometry and increasing steps per period:
The 3840-step result selected a numerical near-tie that was 3.5E-9 smaller than the period-1 distance.
Passing after:
The formatter reports the existing legacy whole-file style warning for
COMMON/mag_interface.f90; no broad reformat is included.