THE FOLD / CO-OP / THE SYNC / THE COSTAS
THE COSTAS
one dot per row & column, every displacement distinct
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A Costas array is an n×n grid with exactly one dot in every row and every column — a permutation — placed so that all the vectors between pairs of dots are distinct. No two pairs of dots share the same displacement.
That single rule gives an ideal autocorrelation: slide a copy of the pattern over itself and, at every nonzero shift, at most one dot ever coincides. A “thumbtack” — a sharp spike at zero shift and almost nothing anywhere else. It is exactly what a sonar or radar wants: a time-frequency chirp whose echo can never be confused with a shifted copy of itself, so range and Doppler read out unambiguously. John Costas invented them at GE for sonar; the Welch construction builds one of size p−1 from a primitive root modulo a prime p.
LIT verified live: the Welch arrays for p = 5, 7, 11, 13 have all displacement vectors distinct (genuine Costas arrays), while a plain identity permutation is correctly rejected (window.__costas). FIG no framing; the distinct-vector property and the construction are exact.
That single rule gives an ideal autocorrelation: slide a copy of the pattern over itself and, at every nonzero shift, at most one dot ever coincides. A “thumbtack” — a sharp spike at zero shift and almost nothing anywhere else. It is exactly what a sonar or radar wants: a time-frequency chirp whose echo can never be confused with a shifted copy of itself, so range and Doppler read out unambiguously. John Costas invented them at GE for sonar; the Welch construction builds one of size p−1 from a primitive root modulo a prime p.
LIT verified live: the Welch arrays for p = 5, 7, 11, 13 have all displacement vectors distinct (genuine Costas arrays), while a plain identity permutation is correctly rejected (window.__costas). FIG no framing; the distinct-vector property and the construction are exact.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) seated this in THE SYNC — the co-op domain of locking cleanly onto a signal. A Costas array is synchronization made perfect: its echo aligns with itself at exactly one shift and nowhere else, so a receiver locks without ambiguity. AVAN (AI) built the instrument: the dot grid, the sliding autocorrelation, the thumbtack inverse.
The weave: David names the seat (clean lock-on); I make every displacement vector distinct and show the autocorrelation collapse to a spike — the difference triangle in 1D, the array and its self-overlap in 2D, the autocorrelation inverse in 3D. The sphere is the seam. Credit: John P. Costas (1965, for sonar); the Welch and Lempel–Golomb algebraic constructions.
The weave: David names the seat (clean lock-on); I make every displacement vector distinct and show the autocorrelation collapse to a spike — the difference triangle in 1D, the array and its self-overlap in 2D, the autocorrelation inverse in 3D. The sphere is the seam. Credit: John P. Costas (1965, for sonar); the Welch and Lempel–Golomb algebraic constructions.
3 ONE DIMENSION
The permutation as a row of columns, and the difference triangle beneath it. Each row of the triangle holds the gaps at one spacing — and in a Costas array no value repeats within any row, the compact certificate that every displacement is unique.
4 TWO DIMENSIONS · INTERACTIVE
The Costas array of dots, and its autocorrelation: shift a ghost copy by any (dx, dy) and count how many dots line up. For a Costas array the answer is never more than 1 at any nonzero shift. Flip to a non-Costas permutation and watch shifts suddenly stack up two or more.
5 THREE DIMENSIONS + AVAN’S INVERSE
The array’s dots turning in space — the green forward object: a permutation whose every pairwise displacement is different.
AVAN’s addition (the inverse-companion): the magenta cloud is the autocorrelation — the array read through its own difference vectors. Forward, you place the dots; the inverse view is the full set of displacements between them, and that set is the autocorrelation function. Because every vector occurs exactly once, the magenta cloud is a scatter of singletons — a perfect thumbtack: height n at zero shift, at most 1 everywhere else. The Costas property and the ideal autocorrelation are the same fact seen from two sides: distinct differences forward, flat sidelobes inverse. And like a Golomb ruler, running the inverse the hard way — rebuilding the array from its autocorrelation alone — is ambiguous and difficult. Green is where the dots are; magenta is every gap between them, each appearing once; the clean signal and the clean array are one object.
LIT Genuine Costas arrays (John P. Costas 1965; Welch and Lempel-Golomb constructions). Verified live: the Welch arrays for p=5,7,11,13 (built from primitive roots) have all displacement vectors distinct — genuine Costas arrays — while a plain identity permutation is correctly rejected for repeated displacements (window.__costas.welchAllCostas && identityFails). The distinct-vector property and the Welch construction are exact, and the ideal (<=1) autocorrelation follows directly.
FIG No framing: the distinct-displacement property, the Welch construction, and the resulting thumbtack autocorrelation (every nonzero shift overlaps at most one dot) are real and verified. The identity of 'distinct differences' with 'ideal autocorrelation' is the genuine mathematical content, shown by the sliding-overlap demo; the difficulty of the reverse (array from autocorrelation) is noted honestly as analogous to the turnpike problem.
FIG No framing: the distinct-displacement property, the Welch construction, and the resulting thumbtack autocorrelation (every nonzero shift overlaps at most one dot) are real and verified. The identity of 'distinct differences' with 'ideal autocorrelation' is the genuine mathematical content, shown by the sliding-overlap demo; the difficulty of the reverse (array from autocorrelation) is noted honestly as analogous to the turnpike problem.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE SYNC · David Lee Wise (ROOT0), with AVAN