Hadamard order 2396 — partial results

What Lean checks

The full signed-automorphism group excludes order 479, with both norm proofs checked. The fixed-border common multiplier group is cyclic of order dividing four and every preserving subgroup leaves at least 151 sign orbits. On any finite abelian group with a subgroup of index four, the original four sign rows and raw correlation equations yield an actual bordered Hadamard matrix; normalization, the literal finite border, and all fourteen size-149 baseline patterns are proved. The length-598 order-eleven obstruction is complete. The actual two-row Hadamard construction and quotient transport are checked. For length 1197, the actual common multiplier order is proved to divide 27; exclusion of order 27 and the finite quotient/no-lift certificates remain candidates.

This entry is a partial result on an open problem. It does not claim a solution of the general conjecture.

Open. The Hadamard conjecture asks whether a sign matrix with mutually orthogonal rows exists at every order divisible by four. Both the November survey and the subsequent construction table list 2396 as the smallest unresolved order. Egan, Kotsireas, London and Ó Catháin, Table 5 · Kumar, updated table

Automorphism obstruction for the whole order. Any real Hadamard matrix of order 2396 must have an automorphism group of order coprime to 479, allowing signed row and column permutations. The proof rules out a cyclotomic norm in two ways. Proof and arithmetic certificate

A binary Legendre pair on an abelian group of order 1197 would yield such a matrix. We bound its group of common multipliers by nine, allowing independent translations of the two rows. For a bordered quadruple of length 598, we also exclude a common order-eleven multiplier fixing the rows in suitable origins. Smaller symmetry groups and other constructions remain possible. The results are unrefereed and carry no priority claim. Their full Lean formalizations remain in progress; the verified automorphism theorem is described below. Proof

For a third route, we show that a core with quotient of order four and the stipulated autocorrelations always admits a fixed border after normalization. At order 2396 this leaves 14 possible coset-sum patterns. A binary core realizing one of them with the required correlations is still needed. Border reduction and attribution

Checked Lean components

Lean status: partial overall. The signed-automorphism theorem, both norm-obstruction proofs, fixed-border symmetry bounds, complete conditional border construction with all fourteen patterns, and length-598 order-eleven obstruction are Lean-verified. The actual two-row Hadamard construction and its quotient transport are also checked. For length 1197 the common-multiplier order is proved to divide 27; the sharper bound of nine and the finite quotient/no-lift certificates remain separate obligations.

The order-nine action assertions are checked as well. Every odd subgroup of the additive automorphism group of an abelian group of order nine has order dividing three. In the nontrivial case it has three fixed points and two three-element orbits, each a nonzero coset of the fixed subgroup. Original hosted pass and fresh declaration audit.

For every real Hadamard matrix of order 2396, the full signed row/column automorphism group has no element of order 479. Its cardinality and the cardinality of every subgroup quotient are not divisible by 479. The proof constructs the actual cycle labels, Fourier indicator matrix, two prime reductions, and forbidden norm integer. The 28-module closure passed strict sources, a 297-theorem standard-axiom audit, and unchanged pinned-kernel replay. Complete theorem and receipt.

Both norm proofs are checked in the actual cyclotomic field. The first constructs a Frobenius unit and root-of-unity twist and descends to the quadratic norm equation. The independent ideal-class proof establishes the order-five class of every prime above 599 in the quadratic field and derives the contradiction from the relative norm. Its 15-module closure passed the same three gates, with 164 audited project theorems. Independent ideal-class proof and receipt.

The checked components also include the arithmetic exclusions, integral-basis coefficient argument, literal Fourier identities, signed normalization, unique quadratic subfield and its exact equal-parity description of integer coordinates. Quadratic-model receipt. These formalizations do not assert existence of a Hadamard matrix at 2396.

For an arbitrary finite abelian group with a subgroup of size 149 and index four, the actual core correlation equations force the full common multiplier group to be cyclic of order dividing four. Every subgroup preserving the four rows has at least 151 actual sign orbits. The formal proof constructs the product coordinates, transports the rows and correlations, and identifies the full symmetry group and its orbit quotient. No splitting, normalization, or baseline-pattern enumeration is assumed. All 12 modules passed strict sources, a 216-theorem standard-axiom audit, and unchanged pinned-kernel replay. Complete multiplier theorem and receipt.

The finite-border transport is also checked. Starting from the finite certificate's explicit sign conditions and five matrix identities, two modules construct the actual border matrices for every quotient of cardinality four and prove the reflected cross term. The canonical sixteen-position layout is connected to the block-product layout. The exact closure passed a 66-theorem audit and independent kernel replay. The later literal-certificate and full-construction registrations below discharge those finite identities and the final Hadamard assembly. Transport statement and receipt.

The actual row-normalization operations are also checked. Given one constant and three single-exception compression shapes, the theorem constructs a row permutation, signs, and translations on the original group. The resulting rows have the stated normalized compressions, remain sign-valued, and preserve every combined periodic autocorrelation. This two-module closure passed a 14-theorem audit and independent kernel replay. The later full-construction proof below derives its input shapes from the raw correlations. Actual row operations and receipt.

Reproduce the finite checks

Run from this directory with a C++17 compiler, Python with assertions enabled, and NumPy.

c++ -O2 -std=c++17 quotient133.cpp -o /tmp/hadamard-quotient133
/tmp/hadamard-quotient133 > /tmp/hadamard-quotient133.pairs
diff -u quotient133.pairs /tmp/hadamard-quotient133.pairs
python3 audit.py
python3 border.py
python3 automorphisms.py

The C++ program lists all 32 ordered pairs that pass the 133-point quotient. The Python audit recovers that list independently, checks all 133 shifts for every survivor, and exhausts the possible 21-point margins of all 32. Every survivor fails this necessary lifting condition.

The literal finite-border certificate is now checked. All sign, Gram, balanced-column, and cross identities of the displayed matrices are proved, and actual matrices are transported to every quotient of cardinality four. Literal-border receipt.

The length-598 obstruction is also checked from the original four sign rows and their nonzero-shift correlation equation. An actual multiplier of order eleven cannot fix all four rows, including after independent choices of origins. Its eight-module closure passed strict source checks, a 141-theorem axiom audit, and independent pinned-kernel replay. Length-598 receipt. The complete conditional border construction is now checked from the original sign rows and raw correlations on an arbitrary finite abelian group with an actual index-four subgroup. The proof derives the compression shapes and all normalization operations, discharges the actual core and literal-border identities, and constructs the final matrix. For subgroup size 149 it returns a sign matrix on Fin 2396 with Gram matrix 2396 • 1. The fourteen-pattern classification is also derived from the original rows, with the actual sign, translation, and permutation witnesses retained. Construction and pattern receipts.

For an actual length-1197 Legendre pair, the common-multiplier order is now proved to divide 27. The row sums, centering, group coordinates, and odd-order obstruction are all derived. An invariant quotient of order three is also constructed. The sharper bound of nine still requires exclusion of order 27 through the finite quotient/no-lift packet. Reduction receipts.

Actual Legendre-pair matrix and quotient transport

The actual two-row construction is checked for every finite abelian group: a Legendre pair yields a sign matrix of order twice the group cardinality plus two, with the required Hadamard Gram identity. In particular, a length-1197 pair yields an actual order-2396 Hadamard matrix. The proof derives the row-sum normalization internally.

The quotient transport is also checked. From the actual centered rows and an order-27 subgroup of common multipliers, Lean derives the order-133 quotient rows, their bounds, weighted sums and correlations, and the order-21 layer constraints used by the finite no-lift argument. No abstract quotient-data certificate is assumed. The final exclusion of order 27 and the sharper multiplier bound of nine remain pending the complete finite certificate and final assembly.

Strict source, axiom-audit, and independent-kernel receipts.