Skip to content

Commit f68f13a

Browse files
apourshclaude
andcommitted
Bound the exact mode-fiber rounding radius: finite reduction and 1 <= R <= 6
Answers #6312's first open question in part. For the PR4933 mode matrix, L = M^T Z^112 is the Hamming construction of sqrt2*E8; R <= 1 iff every lattice point of the open zonotope is a 0/1 subset sum of rows; an explicit tau = 6 basis certificate with Beck-Fiala rounding gives R <= 6. 1,864 sampled interior near-face points are all subset sums (evidence only). No kappa or frontier change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1 parent 2485805 commit f68f13a

10 files changed

Lines changed: 669 additions & 0 deletions

File tree

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,15 @@
1+
{
2+
"schema": "mode-fiber-radius-basis-v1",
3+
"basis_rows": [
4+
7,
5+
35,
6+
57,
7+
73,
8+
87,
9+
90,
10+
92,
11+
99
12+
],
13+
"tau": 6,
14+
"scope": "Rows of M forming a basis of L = M^T Z^112; every other row has integer coordinates in this basis with l1 norm at most tau. Found by search/basis_search.py; verify.py recomputes everything exactly."
15+
}
Lines changed: 132 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,132 @@
1+
# Mode-fiber rounding radius: a finite reduction and 1 ≤ R ≤ 6
2+
3+
Prepared for apoursh with substantial Anthropic Claude assistance. Apache-2.0.
4+
5+
## Setting
6+
7+
M is the 112×8 mode matrix of Dillon Amburgey's
8+
[PR4933](https://github.com/CrocSwap/integer-mult-bounds/pull/4933), read from
9+
`research/coded-native-caller/NORMALIZERS.json` (`mode_columns`), with rows
10+
m_0, …, m_111. Nick Petalas's #6312 (`research/mode-preserving-rounding`, open
11+
at head `46593234bfe658b63fb829f0d22b7eb216e392a2`) studies rounding x ∈ ℝ^112 to
12+
z ∈ ℤ^112 with Mᵀz = Mᵀx, and asks in its first open question for the least
13+
uniform radius
14+
15+
R = sup { ρ(x) : the fiber F(x) = {z ∈ ℤ^112 : Mᵀz = Mᵀx} is nonempty },
16+
ρ(x) = min { ‖z − x‖∞ : z ∈ F(x) }.
17+
18+
The minimum exists because F(x) is discrete. #6312 proves R ≥ 1, using
19+
x = (e_0+e_7+e_20+e_111)/2. This note gives a finite characterisation of R = 1
20+
and a first explicit upper bound.
21+
22+
Write L = Mᵀℤ^112 ⊂ ℤ^8, V = ker Mᵀ ⊂ ℝ^112, Z = Mᵀ[0,1]^112 (a zonotope) and
23+
S = {Mᵀz : z ∈ {0,1}^112} (subset sums of rows).
24+
25+
## 1. Structure of M (exact, verify.py)
26+
27+
* Every row has four nonzero entries, all ±1. The rows are distinct, and MᵀM = 56·I.
28+
* The 14 row supports are exactly the 14 weight-four words of the extended
29+
Hamming code H8 = [8,4,4]. Each support carries 8 rows, so 8 of its 16 sign
30+
patterns occur. The row set is **not** closed under negation.
31+
* L = {x ∈ ℤ^8 : x mod 2 ∈ H8}, the Hamming construction of √2·E8, with
32+
[ℤ^8 : L] = 16. verify.py proves this by comparing Hermite normal forms
33+
of the rows and of H8 ∪ 2ℤ^8.
34+
* For every i, the other 111 rows have rank 8.
35+
36+
## 2. Theorem A (finite reduction)
37+
38+
**R ≤ 1 if and only if L ∩ int Z ⊆ S.** If some t ∈ L ∩ int Z is not a 0/1
39+
subset sum of the rows, then every y ∈ (0,1)^112 with Mᵀy = t has ρ(y) > 1,
40+
so R > 1. With #6312's lower bound:
41+
42+
R = 1 ⇔ every lattice point of L in the open zonotope int Z is a subset sum of rows.
43+
44+
*Proof.*
45+
46+
(a) *Invariance and continuity.* F(x+n) = F(x)+n for n ∈ ℤ^112, so ρ is
47+
ℤ^112-periodic. For v ∈ V, F(x+v) = F(x), so |ρ(x+v) − ρ(x)| ≤ ‖v‖∞.
48+
49+
(b) *Generic points suffice.* Fix x with F(x) ≠ ∅. The coordinate functional
50+
v ↦ v_i is not identically zero on V. Otherwise e_i ∈ V^⊥ = col span(M), say
51+
e_i = Ma. Then m_k·a = 0 for all k ≠ i, and since those rows have rank 8
52+
(§1), a = 0, a contradiction. So {v ∈ V : x_i + v_i ∈ ℤ} is a countable union
53+
of proper affine hyperplanes of V for each i. The union over i is a null set,
54+
so there are v ∈ V with arbitrarily small ‖v‖∞ such that x+v has no integral
55+
coordinate. By (a), R is the supremum of ρ over points with no integral coordinate.
56+
57+
(c) *Reduction to the open cube.* For such x, put y = x − ⌊x⌋ ∈ (0,1)^112.
58+
Then ρ(y) = ρ(x), and t := Mᵀy ∈ L, because Mᵀx ∈ L (the fiber is nonempty)
59+
and Mᵀ⌊x⌋ ∈ L. For y_i ∈ (0,1), an integer z_i satisfies |z_i − y_i| ≤ 1 if
60+
and only if z_i ∈ {0,1}. Hence ρ(y) ≤ 1 ⇔ t ∈ S, and otherwise ρ(y) > 1.
61+
62+
(d) *Which t occur.* Mᵀ is a surjective linear map ℝ^112 → ℝ^8. The image of the
63+
open cube is the interior of the image of the closed cube, so
64+
{Mᵀy : y ∈ (0,1)^112} = int Z.
65+
66+
Combining (b)–(d): R ≤ 1 ⇔ L ∩ int Z ⊆ S. If t ∈ (L ∩ int Z) \ S, every y in
67+
the open cube over t has ρ(y) > 1. ∎
68+
69+
Consequences. Deciding R = 1 is a **finite** problem, since Z is bounded. A
70+
counterexample has a finite exact certificate: a lattice point t, a rational
71+
y ∈ (0,1)^112 with Mᵀy = t, and infeasibility of the 0/1 system Mᵀz = t.
72+
Conversely, R = 1 needs every one of the finitely many interior lattice points
73+
to be a subset sum. Boundary lattice points of Z impose no condition.
74+
75+
## 3. Theorem B (explicit upper bound): R ≤ 6
76+
77+
Let B be the eight rows `[7, 35, 57, 73, 87, 90, 92, 99]` from CERTIFICATE.json.
78+
verify.py checks |det M_B| = 16 = covol(L). Since the rows lie in L, B is a basis
79+
of L. Each other row m_i (i ∈ N, |N| = 104) has integer coordinates w_i ∈ ℤ^8
80+
in this basis, and max_i ‖w_i‖₁ = τ = 6. Let W be the 8×104 matrix with columns w_i.
81+
82+
Take y ∈ (0,1)^112 with Mᵀy = t ∈ L, as in Theorem A(b,c).
83+
84+
*Step 1 (Beck–Fiala rounding of y_N against W).* Keep x ∈ [0,1]^N, starting
85+
from x = y_N, and let F be its fractional coordinates. Call row b of W
86+
*active* if Σ_{i∈F}|W_{bi}| > τ. Since Σ_b Σ_{i∈F}|W_{bi}| ≤ τ|F|, there are
87+
fewer than |F| active rows. So some nonzero d supported on F satisfies
88+
W_active d = 0. Move x along d until one more coordinate is integral, and
89+
repeat until F = ∅. Each active row value is preserved. Once a row becomes
90+
inactive, only coordinates of F change afterwards, each by less than 1. Its
91+
total later change is therefore less than Σ_{i∈F}|W_{bi}| ≤ τ, or zero. The
92+
result z_N ∈ {0,1}^N has ‖W(y_N − z_N)‖∞ < τ.
93+
94+
*Step 2 (absorb into the basis).* Set z_B = y_B + W(y_N − z_N). Then
95+
96+
Σ_{b∈B} z_b m_b = Σ_B y_b m_b + Σ_N (y_i − z_i) m_i = t − Σ_N z_i m_i ∈ L,
97+
98+
and B is a basis of L, so z_B ∈ ℤ^8. Hence z ∈ F(y), with |z_i − y_i| < 1 on N
99+
and |z_b − y_b| < τ on B.
100+
101+
So ρ(y) < 6 for every generic y, and R ≤ 6 by Theorem A(a,b). ∎
102+
103+
Local search over row bases (`search/basis_search.py`, three seeds × 200 restarts)
104+
found no basis with τ < 6. This method therefore stops at 6; a better bound
105+
needs a different argument (see the open questions).
106+
107+
## 4. Evidence for R = 1 (not a proof)
108+
109+
`search/evidence.py` (scipy/HiGHS) samples lattice points t ∈ L near the boundary
110+
of Z, where Theorem A's condition is hardest. It does this in two ways:
111+
112+
- near a vertex (face rank 0): the vertex minus a small inward lattice vector;
113+
- near faces spanned by 1–7 random rows: rows off the face are pinned by the side
114+
of the supporting functional, rows on the face get fractional weights with
115+
denominator 2, 3, 4 or 6, and a small inward lattice vector is subtracted.
116+
117+
It keeps t with a positive LP interior margin and solves the 0/1 system. All
118+
1,864 interior points found (by face rank 0–7: 401, 350, 402, 372, 175, 60, 72, 32)
119+
are subset sums. verify.py re-checks each witness exactly, t ∈ L and Σ_{i∈z} m_i = t.
120+
Interiority comes only from a floating-point LP and is **not** certified.
121+
This sample supports, but does not prove, the conjecture R = 1.
122+
123+
## Scope
124+
125+
R is a property of the stated exact-mode contract for this M, inherited from
126+
#6312. Nothing here shows that an actual caller must preserve these modes, or
127+
supplies a physical rounder, error ledger, tape/precision bill, supplier
128+
admission, new κ or frontier change. The five framework assumptions keep
129+
their status. Theorems A and B are written proofs whose finite inputs are
130+
checked exactly; no proof-assistant or independent human review is claimed.
131+
Beck–Fiala's iterative rounding (1981) and the Hamming construction of E8 are
132+
classical and credited as such.
Lines changed: 72 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,72 @@
1+
# Mode-fiber rounding radius: R = 1 is a finite question, and R ≤ 6
2+
3+
A scoped partial answer to the first open question of Nick Petalas's #6312
4+
(`research/mode-preserving-rounding`, open at head `46593234`). For the PR4933
5+
mode matrix M (112×8), let R be the least ℓ∞ radius such that every x with a
6+
nonempty integer fiber {z ∈ ℤ^112 : Mᵀz = Mᵀx} has a point of that fiber within R.
7+
#6312 proves R ≥ 1. This package proves:
8+
9+
- **Structure.** L = Mᵀℤ^112 = {x ∈ ℤ^8 : x mod 2 ∈ H8}, the Hamming construction
10+
of √2·E8, with index 16. The 14 row supports are the weight-four words of
11+
[8,4,4], with 8 rows each. MᵀM = 56·I. The row set is not closed under negation.
12+
- **Theorem A (finite reduction).** R ≤ 1 if and only if every lattice point of L
13+
in the *open* zonotope Mᵀ(0,1)^112 is a 0/1 subset sum of rows. One interior
14+
lattice point that is not a subset sum forces R > 1. So **R = 1 is a finite
15+
decision problem**, and either answer has a finite exact certificate.
16+
- **Theorem B.** **R ≤ 6.** This is the first explicit upper bound. It uses an 8-row
17+
basis of L in which every other row has coordinates of ℓ₁ norm at most 6,
18+
plus Beck–Fiala iterative rounding.
19+
- **Evidence, not proof.** All 1,864 sampled interior lattice points near vertices
20+
and faces (face rank 0–7) are subset sums. This supports the conjecture R = 1.
21+
22+
[PROOF.md](PROOF.md) has the arguments and exact scope. The basis is in
23+
[CERTIFICATE.json](CERTIFICATE.json), the witnesses in [SEARCH.json](SEARCH.json),
24+
and the pins in [SOURCES.json](SOURCES.json). Complete κ and the selected frontier
25+
are unchanged. This is a property of the exact-mode contract. It does not claim
26+
that any caller needs the contract, and supplies no physical rounder or error/tape bill.
27+
28+
## Reproduce
29+
30+
From the repository root; the standard library suffices:
31+
32+
```sh
33+
python3 -I -W error -B research/mode-fiber-radius-bounds/verify.py
34+
python3 -I -O -W error -B research/mode-fiber-radius-bounds/verify.py
35+
python3.12 -B scripts/research_package.py --package research/mode-fiber-radius-bounds
36+
```
37+
38+
Each prints `PASS_MODE_FIBER_RADIUS_BOUNDS` in about a second. It checks the
39+
structure facts, the HNF lattice identity, the rank condition Theorem A needs,
40+
the #6312 witness relation, the τ = 6 basis (|det| = 16, integral coordinates,
41+
maximum ℓ₁ norm exactly 6) and every subset-sum witness exactly. It also checks
42+
seven rejection controls: a flipped matrix entry, a repeated basis row, a claimed
43+
τ = 5, a non-basis, a wrong witness mask, an off-lattice target and a broken
44+
witness relation. Interiority of the sampled targets came from a floating-point
45+
LP and is not certified.
46+
47+
The non-certifying search scripts in `search/` need numpy and scipy:
48+
`evidence.py SEED TRIALS` regenerates one evidence run (seeds 11–14, 320 trials
49+
each), and `basis_search.py SEED RESTARTS` repeats the basis search.
50+
51+
## Open next questions
52+
53+
1. **Decide R = 1.** Is every lattice point of L in the open zonotope a subset
54+
sum? This would settle #6312's question exactly. First step: reduce the
55+
near-boundary cases with the signed-permutation symmetry of the rows, and try
56+
a face recursion, where deep points are handled by discrepancy rounding and
57+
points near a face reduce to the rows in that flat. Blocker: about 10¹²
58+
lattice points rule out brute force, and the recursion is unproved.
59+
2. **Lower the bound below 6.** Local search finds no basis with τ < 6, so this
60+
method is exhausted. First step: move y to a vertex of its fiber polytope, with
61+
at most 8 independent fractional rows F. Then R < 2 follows if every lattice
62+
point of the zonotope of F is a {−1,0,1} combination of the other rows.
63+
Check this over non-primitive F up to symmetry. Blocker: about C(112,8) sets F
64+
before reduction.
65+
66+
## Credit
67+
68+
Dillon Amburgey (PR4933) supplies M. Nick Petalas (#6312) supplies the contract,
69+
the radius-one witness and the question. Its predecessor credits (JoyboyBrian,
70+
Dugongue, Ali Romman) carry over. Beck–Fiala rounding and the Hamming
71+
construction of E8 are classical. Prepared by apoursh with substantial Anthropic
72+
Claude (Claude Code) assistance. New work is Apache-2.0.

‎research/mode-fiber-radius-bounds/SEARCH.json‎

Lines changed: 1 addition & 0 deletions
Large diffs are not rendered by default.
Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,24 @@
1+
{
2+
"schema_version": 1,
3+
"base_commit": "2485805a6d47f36ea83ddde6327d453ac82d44dc",
4+
"files": [
5+
{
6+
"path": "certificates/selected-result.json",
7+
"usage": "provenance",
8+
"git_blob": "d59b5035e7b4cbaa9d688a1d32dd0d6eb79c466a",
9+
"sha256": "9b8d0eea95b3bf0dfcb7938a732a6439b87a2685dec2a5ab9e525b3b34debbfa"
10+
},
11+
{
12+
"path": "docs/research/agent-task-catalog.json",
13+
"usage": "provenance",
14+
"git_blob": "3404fbdd5e7529a0674eef26114b5071e4586cec",
15+
"sha256": "ead6cdc9d0af6d58e8a49744d8ec8c7c15e13d2ba93c276c434bcf9b51b93a81"
16+
},
17+
{
18+
"path": "research/coded-native-caller/NORMALIZERS.json",
19+
"usage": "input",
20+
"git_blob": "bca92c222c395d3d75c047fbb678f3cd0947c60f",
21+
"sha256": "ae8aad06aa5d1811d725895b3ab5a4f4daa9dcdd88294aa0d5f723e78bb1e0da"
22+
}
23+
]
24+
}
Lines changed: 89 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,89 @@
1+
{
2+
"schema_version": 1,
3+
"task_id": "uniform-transfer-audit",
4+
"priority_revision": 8,
5+
"base_commit": "2485805a6d47f36ea83ddde6327d453ac82d44dc",
6+
"original_card": {
7+
"id": "uniform-transfer-audit",
8+
"question": "Can one of the selected record's five inherited assumptions be proved at its actual caller?",
9+
"starting_points": [
10+
"docs/research/current-priorities.md",
11+
"docs/research/research-family-status.md",
12+
"certificates/selected-result.json",
13+
"docs/research/pr-intake-evidence/pr5599-completion/PROOF-REVIEW.md"
14+
],
15+
"methods": [
16+
"Construct and independently check one source-bound instance of the stated first experiment.",
17+
"Seek an exact counterexample or target-specific complete-budget obstruction to that same proposed change."
18+
],
19+
"acceptance_gate": "Identify the proposition and what would change downstream; do not equate a finite replay with a universal proof.",
20+
"first_step": {
21+
"command": [
22+
"python3",
23+
"-B",
24+
"scripts/research_selected_baseline.py"
25+
],
26+
"expected": "PASS_RESEARCH_BASELINE_BINDINGS_ONLY",
27+
"inspect": "Read revision 8, the selected review and live exact-source handoffs. Then follow this card's bounded first_experiment.",
28+
"scope": "Source-bound orientation only, not a supplier replay or theorem proof."
29+
},
30+
"first_experiment": "Choose one assumption ID (native_all_columns, owned_erasure, compiler, precision_tape, analytic_recovery), one quantified implication and exact dependency closure. Check current handoffs, including #5625 row reentry and #5707 root-prefix lowering, without treating their partial scope as the whole theorem. Prove or falsify that implication with a caller-bound argument; stop at the chosen obligation. Finite tests or root digits do not discharge an all-size statement.",
31+
"bottleneck_link": "Judge the actual current supplier contract and complete signed target/paid bill; frozen-profile ceilings require an applicability proof.",
32+
"allowed_variations": [
33+
"One exact source head, helper/cut region or quantified proposition and changed hypothesis; check live handoffs first. Different seeds are not different scopes."
34+
],
35+
"scope_change_triggers": [
36+
"Changing source head, helper/cut domain, operation alphabet or caller boundary without recording the new scope.",
37+
"Treating an archived proposal or local numerical pass as complete supplier admission."
38+
],
39+
"overlap_terms": [
40+
"native_all_columns",
41+
"owned_erasure",
42+
"compiler",
43+
"precision_tape",
44+
"analytic_recovery"
45+
],
46+
"existing_checks": [
47+
{
48+
"scope": "PR #4100: proposed unconditional construction and finite checks, without independent acceptance.",
49+
"evidence": "research/unconditional-batched-crt/README.md",
50+
"remaining": "Independent proof audit of one caller-bound all-size implication; all other premises remain explicit.",
51+
"status": "author-checked",
52+
"source_commit": "337d04939571a637aeab54a0f58f54fde811663a",
53+
"selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
54+
},
55+
{
56+
"scope": "Selected PR5599: complete helper-subspace construction and independent caller replay; five framework assumptions remain.",
57+
"evidence": "docs/research/pr-intake-evidence/pr5599-completion/PROOF-REVIEW.md",
58+
"remaining": "The next experiment must identify uncovered source-specific work. Successor archive intake does not transfer this admission.",
59+
"status": "reviewed",
60+
"source_commit": "09dd0965711ca7eeb169057a420abab1c9052137",
61+
"selected_record_sha256": "2a0220e866417774606e0f991e41b9c1d26b1f43b97857238cf478dbc7d330b5"
62+
}
63+
]
64+
},
65+
"scope_status": "changed",
66+
"actual_scope": "Open question 1 of #6312 (research/mode-preserving-rounding, head 46593234bfe658b63fb829f0d22b7eb216e392a2): the least uniform l_inf radius R for exact mode-preserving integer rounding on all nonempty fibers of the PR4933 mode map M.",
67+
"scope_change_reason": "The session drew recursive-interfaces cards needing a full construction context. #6312 had just changed this card to a precision side result and left its first open question with only a lower bound. That self-contained question is answered here in part.",
68+
"bottleneck_link": "Side result on the precision contract of the coded inverse. It bounds what an exact mode-preserving rounder can achieve on this M; it does not show that any caller needs one.",
69+
"budget_minutes": 90,
70+
"stop_rule": "Stop at a rigorous characterisation plus an explicit upper bound, or at one exact counterexample t in L cap int Z outside S. Do not claim R = 1 from search evidence.",
71+
"outcome": "partial",
72+
"evidence": "PROOF.md Theorems A and B, with verify.py exact checks: L is the Hamming E8 lattice; the genericity rank condition; the tau = 6 basis certificate; 1,864 exact subset-sum witnesses for sampled interior points.",
73+
"author": "apoursh",
74+
"ai_assistance": "Anthropic Claude (Claude Code) designed the argument, wrote the code and ran the searches under apoursh's direction.",
75+
"open_questions": [
76+
{
77+
"question": "Is every lattice point of L in the open zonotope int Z a 0/1 subset sum of rows (equivalently, R = 1)?",
78+
"why_it_matters": "It settles #6312's open question exactly; the search sample supports yes.",
79+
"first_step": "Use the signed-permutation automorphism group of the 112 rows to reduce the finitely many near-boundary cases. Try a face recursion: points deep in Z are handled by discrepancy rounding, and points near a face reduce to the rows lying in that flat.",
80+
"blocked_by": "Brute force is infeasible (roughly 1e12 lattice points), and the face recursion is not proved."
81+
},
82+
{
83+
"question": "Can the rigorous bound R <= 6 be lowered, for example to R < 2?",
84+
"why_it_matters": "The current bound comes from a generic Beck-Fiala argument, and local search over row bases finds no basis with tau < 6.",
85+
"first_step": "Move y to a vertex y* of its fiber polytope, with at most 8 independent fractional rows F. R < 2 follows if every r in L cap Z(F) is a {-1,0,1} combination of rows outside F; enumerate the non-primitive independent F up to symmetry.",
86+
"blocked_by": "About C(112,8) candidate sets F before symmetry reduction."
87+
}
88+
]
89+
}
Lines changed: 36 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,36 @@
1+
{
2+
"schema_version": 1,
3+
"status": "ready",
4+
"python": "3.12",
5+
"requirements": [],
6+
"inputs": [
7+
"research/coded-native-caller/NORMALIZERS.json"
8+
],
9+
"commands": [
10+
{
11+
"argv": [
12+
"{python}",
13+
"-I",
14+
"-W",
15+
"error",
16+
"-B",
17+
"{package}/verify.py"
18+
],
19+
"timeout_seconds": 60,
20+
"checks": "Mode-matrix structure, L = {x : x mod 2 in H8} by HNF, genericity ranks, PR6312 witness relation, tau = 6 basis certificate, 1,864 exact subset-sum witnesses and seven rejection controls."
21+
},
22+
{
23+
"argv": [
24+
"{python}",
25+
"-I",
26+
"-O",
27+
"-W",
28+
"error",
29+
"-B",
30+
"{package}/verify.py"
31+
],
32+
"timeout_seconds": 60,
33+
"checks": "The same explicit checks under optimized Python."
34+
}
35+
]
36+
}

0 commit comments

Comments
 (0)