353 lines
22 KiB
Markdown
353 lines
22 KiB
Markdown
# Shape–field weak response investigation — 2026-09-18
|
||
|
||
## Prespecified scope
|
||
|
||
Follow up the large, nearly cancelling shape/field Newton direction for n=1.5,
|
||
J=0 without a central-density constraint. This study distinguishes within-element
|
||
physical field transport from surface variation and tests the specific hypothesis
|
||
that the Q3 displacement/Q4 reference geometry mismatch contributes to weak
|
||
response. Cancellation by itself is not evidence of a gauge or bad element orders.
|
||
|
||
The baseline branch is `research/shape-field-weak-response-20260918`, based on
|
||
`db854c0f7cdf5137e377273499f0ac5b6053b752`. The isolated order variant is
|
||
`research/shape-field-displacement-q4-20260918`. Preserve original checkout,
|
||
immutable executables, source revisions, mesh/input hashes and serial run records.
|
||
|
||
## Observation and controls
|
||
|
||
The diagnostic prepares a frozen state with fixed normalization, independently
|
||
verifies its full-J linear defect, saves every value-block Jacobian contribution,
|
||
and checks restoration of accepted buffers and operator action. Three small raw
|
||
Taylor trials test the same original operator. Physical observations use independent
|
||
grid functions and the prepared displacement; no accepted state is modified.
|
||
|
||
At common quadrature orders 10 and 14, record physical scalar pullback changes,
|
||
transport `grad(f).v`, and Eulerian changes `delta f - grad(f).v`. Density and
|
||
potential gradients are broken/local: facet jumps and RT Piola variations are not
|
||
included, so these observations cannot certify a pure relabelling nullspace.
|
||
Surface radial parameters are unweighted nodal samples. An independent affine
|
||
identity projection checks ambient displacement representation, separately from
|
||
the actual r-squared radial extension. Small-span SVDs describe only the sampled
|
||
block directions, never the full Jacobian rank.
|
||
|
||
Baseline frozen cases: the recorded n=1.5 seed and late directions at eta=0.03,
|
||
and a fresh cold solve at the six-update coupled endpoint. Then a fresh six-update
|
||
baseline with gravity-Schur inverse, gravity-curvature globalization, restart 160,
|
||
maximum 200 Krylov iterations, eta=0.03, and a fresh final direction. Match this
|
||
live run with the Q4 variant from the same analytic seed, including a Q4 seed
|
||
probe. Cases, builds and expensive audits run serially. A failed linear gate or
|
||
preparation remains a failure; do not silently relax controls.
|
||
|
||
## Order intervention (conditional on baseline observation)
|
||
|
||
Only change `Displacement::vectorOrder` from `3 + uniform increment` to
|
||
`4 + uniform increment` on the variant branch; keep uniform increment zero.
|
||
The loaded reference geometry remains Q4, density/potential L2 Q2, gravity RT
|
||
index 2, and enthalpy H1 Q3. The shared displacement collection means the scalar
|
||
surface trace also becomes Q4. This introduces shape modes and changes derived
|
||
quadrature orders and cost; it is not a pure interpolation-error intervention.
|
||
Document exact source line, date, rationale, DOFs, and compiled quadrature changes.
|
||
Use a separate build directory and new analytic seeds: existing coefficient
|
||
snapshots cannot be transferred unchanged between orders.
|
||
|
||
## Decision criteria
|
||
|
||
Compare verified linear accuracy, accepted step lengths, within-variant growth
|
||
of weak-direction indicators, common physical norms, virials, positivity/EOS
|
||
metrics and elapsed cost. Across orders, coefficient norms and discrete residuals
|
||
are not physical error measures. Q10/Q14 agreement diagnoses integration
|
||
sensitivity, not convergence certification. Virial <=1e-5 remains required but
|
||
cannot replace internal pointwise checks. Six updates are a bounded mechanistic
|
||
comparison, not evidence of high-n/high-rotation convergence.
|
||
|
||
## Results
|
||
|
||
The baseline branch retains Q3 displacement. The separate Q4 branch is an
|
||
isolated numerical experiment, not a recommendation to change the default.
|
||
|
||
### Reproducible baseline
|
||
|
||
Evidence root: `../../diagnostics/shape_field_weak_response_2026-09-18` relative
|
||
to this checkout. Frozen probes `baseline_probes_v1` use source `d79db94`; the
|
||
six-update control `q3_live_v1` uses `2c4a5e0`. Both include immutable binaries,
|
||
mesh/input hashes, commands, clean-source provenance, and independent elapsed
|
||
and CPU clocks. All full-J linear gates passed. Independent audits
|
||
`baseline_probes_audit_v2.json` and `q3_live_audit_v1.json` pass; accepted buffers,
|
||
factors and restored actions agree exactly. Q3 curved/tensor/geometry regression:
|
||
5 cases, 21,368 assertions passed.
|
||
|
||
The first checker version omitted the new cache-preservation invariant from its
|
||
schema. Version 2 requires it. A small finite synthetic matrix reproduced
|
||
NumPy/Accelerate matmul warnings; direct finite-checked contractions avoid that
|
||
backend path, without changing tolerances. Raw FEM artifacts remain unchanged.
|
||
The artifact checker passes 38 tiny algebra/corruption fixtures. The angular
|
||
analyzer independently checks its degree-8 recurrence, orthogonality, cubic
|
||
symmetries, rank, and file binding on analytic fixtures.
|
||
|
||
| Observation | Seed | Old late state (12 updates) | Fresh control (6 updates) | Coupled endpoint (6 updates) |
|
||
|---|---:|---:|---:|---:|
|
||
| Normalized residual | 2.73659e-4 | 1.39670e-5 | 1.49720e-5 | 1.24583e-5 |
|
||
| Sampled block-span minimum singular response | 5.80574e-2 | 1.43543e-4 | 8.16558e-4 | 6.94894e-4 |
|
||
| Sampled surface radius span / nodal mean | 3.42403e-5 | 1.24122e-2 | 1.09849e-2 | 1.11125e-2 |
|
||
|
||
The largest sampled singular response remains approximately 22.1–22.6. The SVD
|
||
uses the six nonzero disjoint value-block pieces of each p, normalized by their
|
||
coefficient norms. The direction-dependent subspace changes between observations;
|
||
the comparison combines changes in the operator and the probe subspace. It is
|
||
not a full-J spectrum or rank test. In particular,
|
||
comparing the 12-step late state with six-step endpoints does not demonstrate a
|
||
causal benefit of the coupled correction: the matched six-step control has a
|
||
slightly stronger minimum response than the coupled endpoint.
|
||
|
||
Late q14 Eulerian/pullback change norm ratios are density 0.05769, enthalpy
|
||
0.18870, potential 0.01288, versus seed 0.3243, 0.3080, 0.1993. This is substantial
|
||
transport compensation but not physically invisible motion. Broken DG gradients
|
||
omit jumps. The maximum q10/q14 relative difference in the three original cases'
|
||
Eulerian squared integrals is 1.69e-5, an integration sensitivity observation.
|
||
|
||
The radius observations expose a validation gap: the old late state has nodal
|
||
radii 0.997674–1.010093, while its q18 P(h)-based virial error is 4.06909e-6.
|
||
The fresh six-step control has q18 virial error 4.22590e-6 and reproduces the old
|
||
six-step residual trajectory. Its sampled hydrostatic-gradient RMS increases
|
||
from 0.00152810 to 0.0106811 despite the lower residual and virial error.
|
||
Neither is converged or physically certified;
|
||
negative density samples remain. Virial success can coexist with appreciable
|
||
nonrotating surface asphericity at these unfinished states.
|
||
|
||
`q3_surface_modes_v1.json` contains unweighted fits, not surface-integral power.
|
||
The 12-step late direction is 99.9995107% cubic-invariant in sampled squared
|
||
norm, but degree <=8 spherical harmonics leave 0.893679 of its nodal norm.
|
||
Degree <=2 leaves 0.981646. Thus most of the nodal direction is unresolved by
|
||
degree <=8 fits and is nearly cubic-invariant. This is consistent with
|
||
mesh/discretization coupling, without identifying its source: cubic symmetry
|
||
also exists in the bulk and gravity discretizations. These are not surface-area
|
||
spectra or evidence that the surface discretization alone causes the problem.
|
||
|
||
The affine control on the actual stellar mesh gives Q3 relative value error
|
||
9.01277e-5 and gradient error 1.23472e-3. This warrants the isolated Q4 comparison,
|
||
but adding surface modes could also worsen weak response.
|
||
|
||
### Structural interpretation and a rejected shortcut
|
||
|
||
At J=0, bulk Bernoulli rows are unweighted moments. Their direct shape derivative
|
||
at fixed reference-field coefficients is `integral(q * B * div(v))`; the EOS
|
||
shape column similarly contains its local closure imbalance. Thus pointwise
|
||
near-equilibrium shape information increasingly relies on gravity and mechanical
|
||
coupling. Small projected residuals alone do not prove small pointwise imbalance.
|
||
Current shape rows are the prescribed lift's transpose applied to volume force.
|
||
For regular surface h~s, density~s^n and pressure~s^(n+1), weakening near-surface
|
||
coupling as n increases is plausible. DG jumps and the independent RT field
|
||
prevent concluding exact discrete redundancy from the continuum identities.
|
||
|
||
Relevant assembly: `libmeanfield/impl/operators/prepared_hydrostatic_equilibrium.cpp`
|
||
(lines 864–911, 1354–1397), `operators/kernels/barotropic_closure_kernels.cpp`
|
||
(line 655), and `operators/prepared_stellar_equilibrium.cpp` (lines 806–811).
|
||
|
||
A naive replacement of mechanical shape rows by surface Bernoulli moments is
|
||
rank deficient at the desired pointwise B=0, J=0 root with the current trace
|
||
spaces. The actual closed stellar surface has 96 quads, 192 edges, 98 vertices.
|
||
Continuous surface Qp has 96*p*p+2 degrees of freedom; broken Q(p-1) potential
|
||
traces supply only 96*p*p. Q3 surface therefore has 866 rows versus 864 DG Q2
|
||
stellar trace coefficients, after eliminating enforced h=0 trace variations.
|
||
C adds no independent trace dimension; the direct shape measure term vanishes
|
||
with B. Surface-only Q4 widens this hypothetical deficit to 1538-864=674.
|
||
This concerns that proposed replacement, not a proven deficiency of the current
|
||
mechanical Jacobian. Any boundary formulation requires a compatible trace/test
|
||
space and explicit rank/stability analysis.
|
||
|
||
An invertible linear/tangent coordinate or preconditioner change preserves the
|
||
exact mapped physical Newton correction at a fixed state. It can improve
|
||
coordinate-dependent singular values and inexact linear solves. A nonlinear
|
||
coordinate map or retraction can also change finite trial curvature; none of
|
||
these observations alone removes weak physical response. Shape representation,
|
||
step control, and operator formulation remain distinct hypotheses.
|
||
|
||
### Q4 intervention
|
||
|
||
Applied 2026-09-18 at approximately 17:45 UTC on
|
||
`research/shape-field-displacement-q4-20260918`, branched from baseline `211be17`.
|
||
Changed `libmeanfield/interface/field/field_registry.cppm`,
|
||
`Displacement::vectorOrder`, from `3 + uniformPolynomialOrderIncrement` to
|
||
`4 + uniformPolynomialOrderIncrement`; configured uniform increment stays zero.
|
||
The dated source comment records the rationale. Eleven expectations and their
|
||
comments in six test files change to enforce the new displacement degree and
|
||
derived quadrature, while assertions for all other field orders stay fixed.
|
||
|
||
The purpose is to remove measured ambient affine representation error and test
|
||
its effect on the weak direction. The shared scalar surface increases from 866
|
||
to 1538 parameters, so this also adds shape modes. Derived pressure-force order
|
||
increases by one, gravity-force order by two, and geometry-inspection request
|
||
from 18 to 24, including the geometry-admissibility sampling; common physical
|
||
diagnostic quadratures remain 10/14 (weak-direction
|
||
observations) and 14/18 (state metrics). The surface metric, normalization and
|
||
preconditioner also change with their FE space. Equal relative linear tolerance
|
||
does not imply equal physical error or an identical line-search norm. The default
|
||
r-squared radial lift stays
|
||
unchanged and is not made globally affine by this ambient enrichment.
|
||
|
||
Use the separate `cmake-build-release-displacement-q4` directory with the same
|
||
LLVM 22.1.8, Release/IPO, MPI and library configuration. Rebuild all affected
|
||
modules and registered-space tests together. Q4 cases start from the analytic
|
||
seed; no coefficient snapshot is transferred between orders. The fixed
|
||
200-iteration linear budget and eta=0.03 remain unchanged. A 3600-second process
|
||
limit bounds each Q4 numerical case; no silent control relaxation is permitted.
|
||
|
||
### Completed Q3/Q4 comparison
|
||
|
||
Q4 source is `4be1a23c3ec2650af6a2a49528a8f5248cb001c3`. Its only production
|
||
change is the compile-time displacement degree at
|
||
`libmeanfield/interface/field/field_registry.cppm:163`; all instrumentation and
|
||
analysis remain in `experiments/`. No physics-facing API changed. Build and
|
||
dependency configuration values match the Q3 build. The Q4 regression campaign
|
||
passed 22 cases and 380,856 assertions: registry/quadrature contracts, field
|
||
spaces, curved geometry, surface/radial/domain Jacobians, Lane–Emden projection,
|
||
and prepared pressure/gravity force Jacobians. Cases, builds and full artifact
|
||
audits were serialized.
|
||
|
||
`q4_v1` contains the completed seed and live campaigns. The independent audit
|
||
`q4_audit_v1.json` passes 101,282 and 101,310 checks respectively, including raw
|
||
vector algebra, runner hashes, physical-integral identities, and exact buffer,
|
||
normalization, residual and Jacobian-action restoration. An artifact audit pass
|
||
does not turn a failed solve into a successful solve. The endpoint direction is
|
||
explicitly classified `finite_direction_failed_true_gate`.
|
||
|
||
The analytic seed has identical physical volume diagnostics in both variants.
|
||
Q4 reduces the ambient affine relative value error from 9.01277e-5 to
|
||
2.11629e-16 and gradient error from 1.23472e-3 to 3.24701e-15. Nevertheless,
|
||
the seed Newton direction's stellar displacement L2 grows from 0.00121967 to
|
||
0.0385749, a factor of 31.63. The seed linear solve grows from 26 to 67 iterations;
|
||
both meet the same independently checked eta=0.03 criterion. All three raw small
|
||
Q4 seed trials prepare and exhibit a quadratic residual remainder. This is not
|
||
evidence of a missing first-order derivative in the tested direction.
|
||
|
||
The fixed six-iteration-budget comparison gives:
|
||
|
||
| Measurement | Q3 displacement / surface | Q4 displacement / surface |
|
||
|---|---:|---:|
|
||
| Surface parameters | 866 | 1538 |
|
||
| Total root unknowns | 178073 | 178745 |
|
||
| Accepted updates | 6 | 3 |
|
||
| Termination | Nonlinear iteration limit | Fourth linear solve exhausted 200 iterations |
|
||
| First three accepted-step time | 269.31 s | 505.65 s |
|
||
| Residual after three updates / own initial residual | 0.056755 | 0.906188 |
|
||
| Final residual / own initial residual | 0.054710 | 0.906188 |
|
||
| All Newton-iteration time, including failed attempt | 547.34 s | 638.42 s |
|
||
| Outer Krylov iterations, including failed attempt | 354 | 492 |
|
||
| Endpoint probe time, excluded above | 93.07 s | 222.45 s |
|
||
| q18 P(h) potential-energy virial error | 4.22590e-6 | 8.89666e-5 |
|
||
| q18 P(h) force virial error | 2.71367e-6 | 8.27467e-6 |
|
||
| q18 mass error | 3.41685e-6 | 1.96178e-5 |
|
||
| q18 scaled EOS density mismatch RMS | 1.27948e-4 | 1.61156e-4 |
|
||
| q18 hydrostatic-gradient RMS | 0.0106811 | 0.325413 |
|
||
| q18 gravity versus broken potential-gradient RMS | 0.0121494 | 0.326881 |
|
||
| Negative density mass fraction | 2.89683e-5 | 4.19384e-5 |
|
||
|
||
These are differing stopped endpoints, not equal-accuracy or converged solutions.
|
||
The three-update comparison isolates the accepted-count difference in the table,
|
||
but all the previously listed order/space/quadrature/scaling confounders remain.
|
||
Iteration time excludes initialization, final physical measurements and the
|
||
diagnostic probe; total live-process times were 705.76 s and 944.98 s.
|
||
The stdlib helper `compare_shape_order_runs.py` binds the small comparison inputs
|
||
to runner hashes and reproduces `q3_q4_comparison_v1.json`.
|
||
|
||
Q3 accepts steps 1, 1/16, 1/16, 1/16, 1/32, 1/32. Q4 accepts approximately
|
||
0.1770, 0.04425, 0.04425. Its fourth linear solve and independently repeated cold
|
||
endpoint solve both stop after 200 iterations with relative full-J defect
|
||
0.0868564, above 0.03. The finite failed direction remains a directional
|
||
observation only. It is not used as an equally accurate Newton comparison.
|
||
|
||
Q4 also changes the observed physical character of the seed direction: the q14
|
||
Eulerian/pullback norm ratios for density, enthalpy and potential become
|
||
13.0825, 13.4289 and 13.2393. Thus the added deformation is not largely cancelled
|
||
by scalar transport, unlike the late Q3 direction. These are broken scalar
|
||
observations, not a complete distributional or RT transport analysis. The
|
||
largest q10/q14 relative change in Q4 endpoint Eulerian squared integrals is
|
||
4.35e-5; this is an integration sensitivity check, not discretization convergence.
|
||
|
||
`q4_surface_modes_v1.json` passes its independent angular analysis. At the stopped
|
||
Q4 endpoint, sampled radii span 0.927235–1.049839, or 12.2559% of their nodal
|
||
mean. The Q3 endpoint's observed span is 1.09849%. These are different nodal
|
||
sample sets, not common-point or full-surface extrema. In particular, zero
|
||
initial radius variation at Q4 nodes does not make the curved reference surface
|
||
an exact sphere: the same physical mesh sampled at Q3 nodes has a small nonzero
|
||
span. The Q4 seed direction's degree <=8 nodal fit leaves 0.982810 of its norm;
|
||
its sampled squared norm is 99.9999989% cubic-invariant. The corresponding failed
|
||
endpoint direction leaves 0.988475 of its norm. This remains fine angular
|
||
structure, without identifying which discretization component causes it.
|
||
|
||
The saved figure `surface_order_comparison_v2.png` uses a common color scale and
|
||
labels the different sample and accepted-update counts. Version 1 is retained
|
||
as an unused layout draft; version 2 fixes the caption/colorbar overlap.
|
||
|
||
### Conclusions and the next isolated hypotheses
|
||
|
||
**Do not promote the Q4-only change.** It repairs ambient affine representation
|
||
but worsens useful nonlinear progress, linear cost, interior gradient checks,
|
||
and potential-energy virial balance in this case. This rules out order matching
|
||
alone as a sufficient repair under the tested configuration. It does not show
|
||
that the Q3 representation error contributes nothing, nor that a compatible
|
||
enrichment of the other fields could not help.
|
||
|
||
The leading hypothesis is an unfavorable combination of surface freedom and
|
||
the resolving power of the coupled field equations. The observed directions
|
||
have strong fine angular structure, and adding surface freedom produced much
|
||
larger physical motion without useful residual reduction. Normalization and
|
||
preconditioning remain possible contributors; full-J residual accuracy does
|
||
not bound physical correction error for a weak operator. More Krylov work might
|
||
recover a linear solution, but the first three gate-passing Q4 directions already
|
||
gave poor accepted progress. No exact nullspace or discrete stability theorem
|
||
has been established for the present mechanical operator.
|
||
|
||
The next direct experiment should separate independent surface parameters from
|
||
ambient displacement interpolation. A fixed map `q=q0+B*a` with small smooth
|
||
bases of increasing degree can test this hypothesis around the same frozen
|
||
state. Inject physical variations and restrict the existing dual shape force
|
||
with `B^T`; the reduced blocks are the corresponding Galerkin restrictions.
|
||
Pull back the surface metric and preconditioning surrogate consistently, then
|
||
apply state/residual normalization. Do not transpose an already normalized value
|
||
basis and assume it is the physical dual restriction. A surface-mass-orthonormal
|
||
basis is a controlled choice. The deformation contract already separates
|
||
parameter count from boundary-displacement size; the default problem constructor,
|
||
PhysicalRiesz surface metric, and material-surface preconditioners still contain
|
||
nodal assumptions that must be changed together.
|
||
|
||
That reduced solve changes the approximation/test space. Its excluded original
|
||
shape forces and full original residual must remain visible. A frozen `q0`
|
||
retains its existing surface distortion; success there would justify a later
|
||
analytic-seed sequence with increasing surface resolution, not certify a smooth
|
||
physical root. Suggested first frozen bases use spherical degrees 2/4/6/8
|
||
(9/25/49/81 parameters). High-rotation work must demonstrate adequate shape
|
||
resolution rather than enforce a low-mode answer. This experiment has not been
|
||
implemented here.
|
||
|
||
Existing `experiments/README.md`, `gravity_completed_rigid_motion.cpp`, and the
|
||
surface-mode/reachability/frequency probes provide relevant mode construction
|
||
and gravity-completion infrastructure to reuse. The proposed addition is the
|
||
fully coupled restricted Newton solve with a consistent dual restriction and
|
||
excluded-force checks. Historical CSVs in the original checkout were not rerun
|
||
or incorporated into this controlled order comparison.
|
||
|
||
A separate extension test addresses the virial dilation mode. Q4 ambient fields
|
||
can represent the reference position X, but the production logical-radius-squared
|
||
extension E0 does not generate X from its surface trace b. At a retained logical
|
||
radius 1/8 vertex, lifting b gives only one eighth of the required affine value;
|
||
this remains true with Q4 interpolation. A fixed rank-one correction
|
||
`E(z)=E0(z)+(X-E0(b))*ell(z)`, with `ell(b)=1`, preserves the supplied trace and
|
||
gives `E(b)=X`. Its exact transpose is
|
||
`E0^T*f + ell^T*((X-E0(b))^T*f)`; its Jacobian is constant. With an interior
|
||
state displacement `u=E(z)`, direction `b+z` then generates current physical
|
||
dilation `X+u`. Reference-coordinate access, MPI weights, and adjoint virtual
|
||
work require explicit tests. Keep the fixed-infinity vacuum policy unchanged.
|
||
|
||
This would place the discrete dilation test in the mechanical force space; it
|
||
does not automatically enforce the potential-energy virial using the separate
|
||
mixed potential. The Q4 endpoint demonstrates why both independent virials are
|
||
needed. The rank-one lift also does not remove the observed fine angular modes,
|
||
so it is a separate structural control, not a presumed cure. It remains
|
||
unimplemented and must not be bundled with surface reduction in a causal test.
|
||
|
||
All reported states remain nonconverged and include negative density samples;
|
||
P(rho)-based virials are undefined there. The required 1e-5 virial milestone and
|
||
internal pointwise validation have not been achieved by this order intervention.
|
||
No claim about higher-n or rapidly rotating convergence follows from this study.
|