Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
59 changes: 59 additions & 0 deletions .github/workflows/side24-periodization-firstvar.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,59 @@
name: SIDE24 periodization first-variation checks

'on':
pull_request:
paths:
- 'frontiers/side24_periodization_20260929/**'
- '.github/workflows/side24-periodization-firstvar.yml'
workflow_dispatch:

permissions:
contents: read

jobs:
exact-and-probe:
runs-on: ubuntu-latest
timeout-minutes: 10
steps:
- uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262
with:
persist-credentials: false
- uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065
with:
python-version: '3.11.16'
- name: Verify sources, replay both modes, reject mutants
env:
PYTHONDONTWRITEBYTECODE: '1'
run: |
set -euo pipefail
python -B -S - <<'PY'
import hashlib, json, pathlib, subprocess, sys
root = pathlib.Path('frontiers/side24_periodization_20260929').resolve()
manifest = json.loads((root / 'SOURCE_FILES.json').read_text())['files']
present = sorted(p.name for p in root.iterdir())
expected_names = sorted([e['path'] for e in manifest] + ['SOURCE_FILES.json'])
if present != expected_names:
raise ValueError('packet tree differs from manifest: ' + repr(present))
for entry in manifest:
path = root / entry['path']
if pathlib.PurePosixPath(entry['path']).name != entry['path'] or path.is_symlink():
raise ValueError('flat regular source required: ' + entry['path'])
data = path.read_bytes()
if len(data) != entry['bytes'] or hashlib.sha256(data).hexdigest() != entry['sha256']:
raise ValueError('source identity mismatch: ' + entry['path'])
script = str(root / 'first_variation_check.py')
expected = (root / 'RESULTS.json').read_bytes()
mutants = ['M1', 'M2', 'M3']
for flags in (['-B', '-S'], ['-B', '-O', '-S']):
out = subprocess.run([sys.executable, *flags, script],
check=True, capture_output=True, timeout=300).stdout
if out != expected:
raise RuntimeError('output differs from RESULTS.json under ' + ' '.join(flags))
for m in mutants:
rc = subprocess.run([sys.executable, *flags, script, '--mutant', m],
capture_output=True, timeout=300).returncode
if rc != 1:
raise RuntimeError('mutant ' + m + ' not rejected')
print(json.dumps({'scientific_effect': 'NONE', 'scientific_acceptance': False}))
PY
git diff --exit-code
181 changes: 181 additions & 0 deletions frontiers/side24_periodization_20260929/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,181 @@
# SIDE24 periodization first-variation: the exact nearest-image correction

**Object:** CL-SIDE24-PFV-20260929-v1. **Author:** Anthropic Claude (session `session_01NMeKEismAyeqgdB4sy2NJU`).
**Disposition:** author-side candidate; **nonauthor review open**. No register, prize, `lemma_closed` flag or scientific
status is changed by this note. Scientific effect: NONE until a nonauthor lane reviews it.

## Scope and exact parent

This note evaluates the **leading finite-size correction** to the SIDE24 lifetime-density coefficient defined by the
reconciled parent chain: `imports/lifetime_parent_20260925/UNIFORM_MATRIX_CAP_AND_LIFETIME.md` eq. (15.2)
(blob `dfed3b8d…`, SHA256 `9350ad6e…`), read with the reading rule of
`reviews/d1_chain_reconciliation_20260928/RECONCILIATION.md`, and its `d=3` reference specialization in
`coefficients/side24_v1/PROOF.md` (blob `44b66f04…`). It does **not** re-derive, re-accept, or modify that parent
or that coefficient; it derives one additional scalar, the first image-shell variation, at the parent's own
existential scope. It asserts no numerical enclosure of the coefficient itself.

Let `Φ(J)` be the `O(3)`-invariant coefficient functional on the full even covariance-jet tensors through order
six, defined by the parent's §15 angular integral. The moments `(a, m4, χ)` parametrize its restriction to the
isotropic jet family used in §4; the side-`L` periodized jet `J_L` need not be isotropic. In the reference
normalization, `c_{3,L} = Φ(J_L)`. Write `J_∞` for the isotropic nonperiodic Euclidean jet
(`a=1, m4=3, χ=15`, `Ω = aχ − m4² = 6`), which is fixed by rotations. Poisson summation gives
`J_L = J_∞ + J^{(1)}_L + R_L`, where `J^{(1)}_L` is the term linear in `q = e^{−L²/2}` from the six nearest
images `±L e_i` after normalization. The remaining normalized jet `R_L` includes normalization/product terms
as well as higher image shells.

## Theorem (candidate)

With `P_3(L) = −L²(10L⁴ − 147L² + 315)/105` (equivalently `−(2/21)L⁶ + (7/5)L⁴ − 3L²`),

DΦ(J_∞)[J^{(1)}_L] / Φ(J_∞) = P_3(L) · e^{−L²/2},

and hence the relative side-`L` correction is

c_{3,L}/c_{3,∞} − 1 = P_3(L) e^{−L²/2} + ε_L,

where `ε_L` consists of the nonlinear functional and remaining normalized-jet terms described in §5; no new
sharp bound on `ε_L` is asserted here.

At `L = 24`: `P_3(24) = −620813376/35`, so the leading relative correction is
`−(620813376/35) e^{−288} ≈ −1.4862 × 10^{−118}`, and it sits strictly inside the coarse periodization band
`|c_{3,24}/c_{3,∞} − 1| < 10^{−106}` proved (outward-rational) in `coefficients/side24_v1/PROOF.md` §§2–4. The
present note sharpens the *leading term* of that already-reviewed band; it does not weaken or replace the band.

## 1. Rotational projection is exact because Φ is O(3)-invariant

The full coefficient functional is `O(3)`-invariant because rotating a covariance jet rotates the angular
integration variable in the parent's §15 integral. Since `J_∞` is fixed by rotations, its derivative satisfies
`DΦ(J_∞)[U·H] = DΦ(J_∞)[H]` for every rotation `U` and jet perturbation `H`. By linearity,

DΦ(J_∞)[H] = DΦ(J_∞)[∫_{O(3)} U·H dU],

where Haar measure is normalized to mass one. Consequently only the Haar projection of `H` onto the isotropic
three-moment subspace contributes to this first variation; no such reduction is asserted for `Φ` at a general
anisotropic jet. The relevant projection onto `(a, m4, χ)` uses the directional-derivative averages

⟨∂²_v⟩ = (1/3) Σ_i d_ii,
⟨∂⁴_v⟩ = (1/5) Σ_i d_iiii + (2/5) Σ_{i<j} d_iijj,
⟨∂⁶_v⟩ = (1/7) Σ_i d_i⁶ + (3/7) Σ_{i≠j} d_i⁴d_j² + (6/7) d_112233,

whose coefficients are the exact even moments of a uniform unit vector on `S²`
(`E v₁² = 1/3, E v₁⁴ = 1/5, E v₁²v₂² = 1/15, E v₁⁶ = 1/7, E v₁⁴v₂² = 1/35, E v₁²v₂²v₃² = 1/105`). The checker
derives these two ways: directly, and via the Gaussian factorization `g = R v`, `R² ~ χ²₃ ⟂ v`, which gives
`E v₁^{2a}v₂^{2b}v₃^{2c} = (2a−1)!!(2b−1)!!(2c−1)!! / (2n+1)!!`. Spectral moments are `a = −⟨∂²_v⟩`,
`m4 = +⟨∂⁴_v⟩`, `χ = −⟨∂⁶_v⟩`.

## 2. First-order image perturbation of the 1-D moments

The side-`L` kernel factorizes over axes. For `j = 1, 2, 3` and `L ≥ 1`, each 1-D normalized even moment is
`m_{2j}(L) = He_{2j}(0) + 2q(He_{2j}(L) − He_{2j}(0)) + O_j((1 + L^{2j})q²)`.
Indeed, `(d/dz)^k e^{−z²/2} = (−1)^k He_k(z) e^{−z²/2}` (probabilists' Hermite), so the raw numerator is
`h_0 + 2h_L q + O_j((1 + L^{2j})q⁴)`, where `h_0 = He_{2j}(0)`, `h_L = He_{2j}(L)`, and
`Z_L = 1 + 2q + O(q⁴)`. The raw tail bound follows from Hermite polynomial growth and the convergence of
`Σ_{n≥2} n^{2j} exp(−(n² − 4)/2)`. Normalization creates a quadratic term, as the exact identity

(h_0 + 2h_L q)/(1 + 2q) − h_0 − 2(h_L − h_0)q = −4(h_L − h_0)q²/(1 + 2q)

shows. Thus the next raw 1-D images start at `q⁴`, but the normalized remainder starts at `q²` scale; the
polynomial factor in `L` is retained in the uniform bound. With `He_2(0), He_4(0), He_6(0) = −1, 3, −15`, the per-`q` variations remain
`δm_{2j} = 2(He_{2j}(L) − He_{2j}(0))`.

## 3. Assembling the projected moment variations

Only the matching 1-D factor is perturbed at first order (six images `±L e_i` perturb the axis-`i` factor). Feeding
the per-axis / per-pair perturbations through the §1 projection gives, per unit `q`,

δa = −2L²,
δm4 = (6/5)L⁴ − 12L²,
δχ = −(6/7)L⁶ + 18L⁴ − 90L².

(These are the exact closed forms the checker verifies as `H5_*`.)

## 4. The chain rule through Φ: the three exponents derived, not read off

(v1.1, answering the xAI Slice A HOLD: every exponent below is an identity of finite Gaussian conditioning,
verifiable in any lane without the parent's prose.)

**The isotropic family.** Let `J = (a, m4, χ)` be the even jet of a covariance that is isotropic through order six,
in the parent's sign convention `a = −⟨∂²_v K⟩(0)`, `m4 = +⟨∂⁴_v K⟩(0)`, `χ = −⟨∂⁶_v K⟩(0)`. Isotropy fixes the
derivative tensors up to these scalars: `∂_i∂_j K(0) = −a δ_ij`, `∂_i∂_j∂_k∂_l K(0) = (m4/3)(δ_ij δ_kl + δ_ik δ_jl
+ δ_il δ_jk)` (so `d_iiii = m4`, `d_iijj = m4/3`), and the axial sixth derivative `∂_x⁶ K(0) = −χ`. Using
`Cov(∂^α f, ∂^β f) = (−1)^{|β|} ∂^{α+β} K(0)`, the §15 objects at a contact point with axis `u = e_x` and
transverse coordinates `y ∈ ℝ^m`, `m = d − 1`, are:

1. **Gradient.** `Cov(G) = a I_d`, so `p_G(0) = (2π)^{−d/2} a^{−d/2}`.
2. **The `V_u` block.** `V_u = (f_xx, f_{xy_1}, …, f_{xy_m})` has `Var f_xx = m4`, `Var f_{xy_j} = d_{xxy_jy_j} = m4/3`,
and all cross-covariances vanish (each would be a fourth derivative with an odd number of one index). Hence
`det Cov(V_u) = m4 (m4/3)^m` and `p_{V_u}(0) = (2π)^{−d/2} 3^{m/2} m4^{−d/2}`.
3. **The axial third derivative.** `Var f_xxx = χ`, `Cov(f_xxx, f_x) = −m4`, `Var f_x = a`, and `f_xxx` is
uncorrelated with the transverse gradient components. So `τ_u² = Var(f_xxx | G = 0) = χ − m4²/a = Ω/a`,
`Ω := aχ − m4²`, and `τ_u^{4/3} = Ω^{2/3} a^{−2/3}`. (Reference values: `15 − 9 = 6`.)
4. **The transverse Hessian given `V_u = 0`.** Unconditionally `Cov(A_ii, A_jj) = (m4/3)(1 + 2δ_ij)`,
`Var A_ij = m4/3` for `i ≠ j`, off-diagonal entries uncorrelated with the diagonal ones and with `V_u`;
`Cov(A_ii, f_xx) = d_{xxy_iy_i} = m4/3`; `Cov(A_ij, f_{xy_k}) = 0` (odd in `x`). Conditioning on `f_xx = 0`
(the only correlated coordinate of `V_u`) subtracts `(m4/3)²/m4 = m4/9` from every diagonal–diagonal
covariance:

Cov(A_ii, A_jj | V_u = 0) = (m4/3)(2/3 + 2δ_ij), Var(A_ij | V_u = 0) = m4/3 (i ≠ j).

At `m4 = 3` this is exactly `(2/3)δ_ijδ_kl + δ_ikδ_jl + δ_ilδ_jk`, i.e. `A = Q + √(2/3) Z I_m` of
`coefficients/side24_v1` §1 (and `Var A = 8/3` for `m = 1`). In general the conditional law is the fixed
`m4 = 3` law scaled by `√(m4/3)`; since `h(A) = det(A)² 1{A < 0}` is homogeneous of degree `2m` and the cone
is scale-invariant, `D_u = (m4/3)^m D_u^{(3)}`, i.e. **`D_u ∝ m4^{d−1}`**.

**Exponents in every `d`.** Multiplying 1–4, the isotropic-family dependence of the (15.2) integrand — hence of
`Φ`, since the angular integral of a constant is a constant — is

Φ ∝ a^{−d/2 − 2/3} · m4^{d/2 − 1} · Ω^{2/3}.

For `d = 3`: `Φ ∝ Ω^{2/3} m4^{1/2} a^{−13/6}`. (For `d = 2` the `m4` exponent is `0`, consistent with
`coefficients/side24_v1` eq. (1) where `D_1 = 4/3` carries `m4¹` against `p_V ∝ m4^{−1}`.) The checker verifies
items 1–4 and the exponent sums exactly (group `H6`), including the conditional-covariance scaling at several
rational `m4` values. Hence at the planar point `(1,3,15)`, `Ω = 6`,

d log Φ = (2/3)(dΩ/6) + (1/2)(dm4/3) − (13/6) da = dΩ/9 + dm4/6 − (13/6) da,
dΩ = χ da + a dχ − 2 m4 dm4 = 15 da + dχ − 6 dm4.

Substituting §3 and dividing by `q` yields exactly `P_3(L)`. The amplitude-normalization check (`c ∝ A^{−2/3}`
under `f → Af`, matching `ν_{Af}(ℓ) = A^{−1}ν_f(ℓ/A)`) is the parent's own consistency identity — with the general
exponents above it reads `−d/2·2 + 4/3 + 2(d−1) = −2/3` and is reproduced in the D1-E full-depth review.

**What §4 consumes from the parent.** Only the *shape* of (15.2): that the leading coefficient is the angular
integral of `p_G(0) p_{V_u}(0) τ_u^{4/3} D_u` times a jet-independent constant. The values of the four factors
under any isotropic jet are computed here from the covariance alone.

## 5. Remainder (what a reviewer must still bound analytically)

With `R_L = J_L − J_∞ − J^{(1)}_L`, the remainder `ε_L` has two parts: (i) the nonlinear functional Taylor
remainder at `J_∞` for the full perturbation `J_L − J_∞`; (ii) `DΦ(J_∞)[R_L]/Φ(J_∞)`. The remaining normalized
jet `R_L` includes normalization terms and products between axis factors as well as deep image shells
`|n| ≥ √2 L`. In particular, §2's normalization already creates `q²` terms even though the next raw 1-D
images start at `q⁴`.

No separate `D²Φ` or remaining normalized-jet majorant is proved here. The existing
`coefficients/side24_v1/PROOF.md` certifies the *total* relative correction at `L = 24` below `10^{−106}`;
the triangle inequality therefore gives only the inherited coarse bound
`|ε_24| < 10^{−106} + |P_3(24)| e^{−288}`. The present first-variation term `≈ 10^{−118}` is far inside the
existing band. A sharper remainder enclosure or a sign certificate for the complete correction requires a
separate analytic bound and is outside this packet's leading-first-variation claim.

## 6. What this note is and is not

- **Is:** an exact, independently-derived closed form for the *leading* nearest-image correction to the reference
coefficient, with a stdlib fail-closed checker (`first_variation_check.py`, `-O` byte-identical, three semantic
mutants rejected). It answers, in the parent's own conventions, the question "what is the side-24 correction?"
that the superseded 2026-08 line raised.
- **Is not:** a numerical enclosure of `c_{3,24}` itself (that is side24_v1's reviewed arithmetic scope); a second-
order remainder certificate; any change to the parent's existential scope; or nonauthor-accepted. It uses the
parent §15 exponent structure as an input — a reviewer who rejects that structure must revisit §4 here too.

## Review requested

Priority nonauthor (non-Claude) check: (i) the §1 projection weights and the O(3)-invariance argument that makes
`DΦ` a function of `(a,m4,χ)` only; (ii) the §4 exponent read-off from the parent §15 factors; (iii) the §3
closed forms; (iv) whether the §5 remainder deserves a first-class `D²Φ` bound in these conventions or inherits
side24_v1's band as stated. xAI/Grok or OpenAI/ChatGPT are the natural lanes (Claude authored this).

## Reproduce

python -B -S first_variation_check.py # prints RESULTS.json byte for byte
python -B -O -S first_variation_check.py # identical
python -B -S first_variation_check.py --mutant M # exit 1 for M in {M1, M2, M3}
52 changes: 52 additions & 0 deletions frontiers/side24_periodization_20260929/RESULTS.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,52 @@
{
"object": "CL-SIDE24-PFV-20260929-v1-CHECKS",
"passed": true,
"results": {
"H1_hermite": {
"He2": "L^2-1",
"He4": "L^4-6L^2+3",
"He6": "L^6-15L^4+45L^2-15",
"at_zero": [
1,
-1,
3,
-15
]
},
"H2_sphere_moments": {
"v1^2": "1/3",
"v1^2v2^2": "1/15",
"v1^2v2^2v3^2": "1/105",
"v1^4": "1/5",
"v1^4v2^2": "1/35",
"v1^6": "1/7"
},
"H3_projection_weights": {
"d4": [
"1/5",
"2/5"
],
"d6": [
"1/7",
"3/7",
"6/7"
]
},
"H5_moment_variations_per_q": {
"da": "-2L^2",
"dchi": "-(6/7)L^6+18L^4-90L^2",
"dm4": "(6/5)L^4-12L^2"
},
"H6_chain": "exponents derived: a^(-d/2-2/3) m4^(d/2-1) Omega^(2/3); d=3 -> (-13/6, 1/2, 2/3); dlogc = dOmega/9 + dm4/6 - (13/6) da at (1,3,15)",
"H7_P3": {
"P3(24)": "-620813376/35",
"P3(L)": "-L^2(10L^4-147L^2+315)/105"
},
"H8_bounds": {
"coarse_band": "1e-106 (side24_v1 (4)); linear term inside",
"e^-288": "<1e-125 (exact rational)",
"|P3(24)|e^-288": "<1e-117 (exact); ~1.4862e-118 (float)"
}
},
"scope": "finite exact controls for the nearest-image first-variation candidate; not the Gaussian/analytic proof"
}
50 changes: 50 additions & 0 deletions frontiers/side24_periodization_20260929/SOURCE_FILES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,50 @@
{
"author": "Anthropic Claude",
"consumes": [
{
"path": "imports/lifetime_parent_20260925/UNIFORM_MATRIX_CAP_AND_LIFETIME.md",
"role": "parent (15.2) exponent structure",
"sha256": "9350ad6eaba6626b93c3dedeef9e2ff816e5cdf1c8318e85fb27499141c84bc7"
},
{
"blob": "44b66f04f89fcd87383b3603fa69f1feb64cdddd",
"path": "coefficients/side24_v1/PROOF.md",
"role": "reference specialization + coarse band"
}
],
"delivered_under": "owner instruction in claude.ai session 2026-09-29; OWNER_STOP left in place; no self-merge; scientific effect NONE",
"files": [
{
"bytes": 12391,
"path": "PROOF.md",
"sha256": "f540b9b9907ba3e7fa21a9aff03c2af9c7c26c8af63589a79fe72f9f0a62ec50"
},
{
"bytes": 13368,
"path": "first_variation_check.py",
"sha256": "57d0cac78975ee6c1a23321d8aa44cff92681fb2a42aff8fe820f727c975d6f3"
},
{
"bytes": 1307,
"path": "RESULTS.json",
"sha256": "b9a28da8aa4332fcf67fc6a1a61d0c8c5caead4de6f15d653a8aafa0426c268c"
}
],
"mutants": [
"M1",
"M2",
"M3"
],
"object": "CL-SIDE24-PFV-20260929-v1.1",
"result": {
"P3(24)": "-620813376/35",
"P3(L)": "-L^2(10L^4-147L^2+315)/105",
"coarse_band": "< 1e-106 (side24_v1, reviewed)",
"leading_relative_correction_L24": "-(620813376/35) e^-288 ~ -1.4862e-118"
},
"reviewer_session": "https://claude.ai/code/session_01NMeKEismAyeqgdB4sy2NJU",
"scientific_acceptance": false,
"scientific_effect": "NONE",
"scope": "leading nearest-image first-variation of the reference SIDE24 coefficient; author-side; nonauthor review open",
"v1_1_change": "\u00a74 rewritten: the three exponents are derived from finite Gaussian conditioning of the isotropic jet (general d), not read off; checker H6 extended (conditional-covariance scaling, tau^2 Schur complement, general-d exponent sums). \u00a71-3, \u00a75 and the closed forms unchanged."
}
Loading
Loading