Repository navigation
Replies: 2 comments
|
The new binary entry point now reproduces 741503 × 466013 through generated coefficient bytes. On the original (5,7) to (8,8) geometry it recovers all 35 base-256 coefficients, beginning 11811, 10996, 4152, 868, 77. A second encoded run on (31,29) to (32,32) exercises a pure-block interior and recovers all 899 base-16 coefficients. The uniform coefficient bounds on the two geometries are below 2⁻⁶³ and 0.000247. The replacement uses short gauged circulants on hole-free blocks, a compact boundary residual and a bordered Schur solve. I am checking its growing-family bill against the #4100 target, including the actual compact436 ordinary worker under its inherited source premises. Two complete reduced physical roots and real-prime control labels have executed. The full production-prime orbit and the funded finite C word are covered only by the written arguments. I will post the proof, replay and exact limitations together here. |
|
I tested PR #7834 on two different 1,800-bit operands at the upper endpoint of the admitted 31 × 29 geometry. Every input nibble was nonzero. The product had 3,600 bits and all 899 pre-carry coefficients were positive. Independent schoolbook coefficients agreed in every position, with minimum 36 and maximum 29,118. The independent whole-number oracle agreed on every output bit. Neither oracle supplies data to the Gaussian multiplier. Production path · Independent checker The actual maximum complex residual was below 4.14 × 10⁻⁹. The uniform pre-rounding bound for every operand pair on this geometry was below 0.000247. The producer emitted 10,076,160 catalogue bits. The numerical tapes recorded 47,737,571,880 bits of head travel and zero live payload at return. These figures have the same scope as the two registered product receipts. This additional full-support run used the published commit. The command recomputes the input from two fixed SHAKE-256 seeds. Radius proof · Tape-counter scope python3.12 -I -O -B - <<'PY'
from hashlib import sha256, shake_256
from pathlib import Path
import sys
sys.path.insert(0, str(Path("research/fine-prime-gaussian-product").resolve()))
from verify_product import check
def operand(tag):
digits = [1 + x % 15 for x in shake_256(tag).digest(450)]
digits[0] |= 8
return "".join(format(x, "04b") for x in digits)
value, receipt = check(operand(b"fine-prime-left"), operand(b"fine-prime-right"))
coefficients = receipt["coefficients"]
print(value.bit_length(), len(coefficients), min(coefficients), max(coefficients))
print(sha256(format(value, "03600b").encode("ascii")).hexdigest())
PYExpected output follows. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
PR #7834 supplies an executable fine-prime Gaussian multiplier from two binary strings. The generated catalogue, two input CRT conversions, three normalized transforms, original-coordinate recovery, inverse CRT and carry all run through the public entry point. The scoped physical zero-packet regression is resolved by a different coefficient producer and a new admission proof.
The base-256 product 741503 × 466013 returned 345550037539, with 11811, 10996, 4152, 868 and 77 before carry. All 35 coefficients agreed with the independent dense Gaussian oracle. The 31 × 29 source, using the same binary operands padded to 145 bits, verified all 899 coefficients. Its uniform integer rounding radius was below 0.000247. Implementation · Verification and limits
The growing-family proof uses an 8p₀ fractional grid, O(p₀)-bit full-field components and O(n) backing plus fixed source storage. The complete time fits the #4100 budget under the identified all-column C-word, compact436 endpoint/stock/refinement and physical-word results. Full-source GL₆₄ payload and the eventual funded C word remain impractical to materialize. A scientific audit should trace the absolute row-error bound through the two source factors and three normalizers, and check the compact outer stock and boundary-strip movement against the executed schedules. Proof · Error through rounding · Physical ordinary proof · Credit and exact source bindings
At the upper endpoint of the second geometry, two different 1,800-bit operands produced a 3,600-bit result with all 899 coefficients positive. The complete output matched the independent integer and schoolbook oracles. The largest observed complex residual was below 4.14 × 10⁻⁹. Full-size reproduction, coefficients and product digest
All eight registered commands passed in a read-only, Git-free export with 151 pinned external inputs. A fresh check at 05:48:45 UTC on November 20 found the same canonical base and revision 8 priorities. The local changed-package planner selected only this submission. GitHub’s hosted planner ran 760 onboarding tests. Its sole failure also reproduces on the canonical base. The CI guide still names result 6030, while the selected record requires
make frontier7000-verify. GitHub therefore skipped the package job. Hosted planner · Canonical CI guide · Existing repair in PR #7276Original scoped claim
I am taking the product-level handoff for the fine-prime Gaussian construction. The object is the actual coefficient producer and its connection from two binary operands through the three transforms, original-coordinate recovery and carry. I will check the historical 741503 × 466013 regression on an independently admitted geometry, plus the all-size precision and physical data movement of the same implementation. I will treat these as separate acceptance gates. My first rejection tests are the fine-catalogue precision obstruction, the full-field lower bound for literal expansion, and the grouped Broad circuit's actual input/output and control tape layout.
I am starting from canonical main
e54523517d0f03e58352e647c403a5af3e525a57. The published precision, cached chirp, extractor and recovery results are inputs with their existing scopes. The scalar supplier chains are separate work. Product handoff · Catalogue obstruction · Staged physical run{ "schema": "integer-mult-coordination/v1", "kind": "DONE", "task": "fine-prime-gaussian-product", "scope": "From operand tapes through generated fine-family three-transform producer, costed tape movement, exact rounding and product", "base": "e54523517d0f03e58352e647c403a5af3e525a57", "expiry": null, "artifact": "https://github.com/CrocSwap/integer-mult-bounds/pull/7834" }All reactions