Publications / rule-30-checkerboard-completion

Rule 30: different starting rows, the same left-hand evolution

You are reading a fixed version. Read the current version

A single 1 at position 0 and a right-infinite 1010… pattern starting at 0 give the same center sequence, with all other cells initially 0. Their evolution also agrees everywhere to the left.

I compare two starting rows: one occupied only at position 0, and one occupied at 0, 2, 4, 6, … . Every other cell is 0.

The rows differ at infinitely many positions. Their center values still agree at every time. The same holds at every position to the left of the center.

                  center
single seed:  …000 | 1 | 000000…
right tail:   …000 | 1 | 010101…

Rule 30 uses the update next = left XOR (center OR right). I write F^t(u)(n) for the value at position n after t updates from row u. Rows extend over all integer positions, increasing to the right.

For the two rows above, the equality is:

F^t(single seed)(n) = F^t(right tail)(n)
for every t ≥ 0 and every n ≤ 0.

The general statement

Let u be any binary row with a rightmost occupied cell at R ≥ 1: u(R)=1 and u(n)=0 for all n>R. There is no restriction on the row to the left.

Then there is a row v whose right tail alternates 1, 0, 1, 0, … from some positive position onward, with

F^t(v)(n) = F^t(u)(n)
for every t ≥ 0 and every n ≤ 0.

Here “alternating tail” describes the initial row in space. It does not say that a fixed cell alternates over time.

The single-seed example has R=0, so it is not a direct instance of the general statement. It follows from the first construction below: the cell at R−1=-1 is 0, and this construction preserves the full evolution at every n≤R.

How to construct the row

There are two cases.

The cell at R−1 is 0. Keep the row through position R and add occupied cells at R+2, R+4, R+6, …. This preserves the evolution at every position n≤R.

The cell at R−1 is 1. Move the rightmost occupied cell from R to R+1, leaving everything else unchanged. The values at every position n≤R−1 agree for all t≥0. At the new right edge, the cells at R and R+1 are 01, so apply the first construction at R+1.

Since R≥1, both cases preserve the evolution at every n≤0.

Why the changes stay on the right

The proof uses two local identities. Put E=R+t, the moving right edge of the original row.

When u(R−1)=0, adding just one cell at R+2 gives these exact disagreement sets:

t even: {E+2}
t odd:  {E, E+1, E+2}

Every listed position is greater than R: the odd case first occurs at t=1. The identity follows by induction, checking the update near the moving edge. At the decisive step, the center input of the local three-cell neighborhood is 1 in both evolutions. The term center OR right therefore does not depend on the right input there.

The new rightmost cell again has a 0 immediately to its left. We can therefore add another cell two places farther right and repeat. Each finite stage preserves the same left region.

The infinite tail follows from finite dependence. For any fixed time t and position n, the output only uses the initial cells in [n−t,n+t]. A sufficiently long finite completion agrees with the infinite one throughout that interval. This proves equality at that point, for arbitrary t and n≤R.

For the second case, moving the rightmost occupied cell one position right gives:

t even: {E, E+1}
t odd:  {E+1}

The same local induction keeps these disagreements strictly to the right of R−1. Combining this edge move with the completion proves the general statement.

What this leaves open

The completed starting row has infinitely many occupied cells on the right. The construction therefore gives neither a finite starting row for an arbitrary prescribed center sequence nor a result about eventual periodicity of the single-seed center column.

There is related work on directional reconstruction, diagonal structure and trace restrictions, including Wolfram's reconstruction note, Rowland's local-structure paper and Kopra's trace results. I have not established whether this exact construction has appeared before.

Source and checks

This note uses source commit 85d7d83c06f692f3fa7d19af42b6c988e0f58ca0. The general R≥1 statement is Rule30.full_checkerboard_completion_exists in Phase20ExplicitCompletion.lean. The R=0 single-seed example has a separate formal proof in Phase17InfiniteFiber.lean: Rule30.infiniteEvenRow_same_left_trace and Rule30.infiniteEvenRow_same_center.

For this note, all ten Lean modules in the checked dependency set were rebuilt in a fresh directory with Lean 4.34.0. The build passed. One AI reviewer also reconstructed the edge argument and ran a separate bounded simulation. The simulation checks finite cases; the all-time statement relies on the proof and its Lean verification.

Statements & evidence

R30-CHECKERBOARD-COMPLETION Proof provided

For every binary row u with a rightmost occupied site R≥1, there exists a row v with an eventually checkerboard right tail such that F^t(v)(n)=F^t(u)(n) for every t≥0 and n≤0; no condition is imposed on u's left tail.

Formal verification
Formally verified within the stated scope
Review status
Internal check
Evidence
Frozen source 85d7d83c06f692f3fa7d19af42b6c988e0f58ca0: proofs/phase20_explicit_checkerboard_completion.md and formal/Phase20ExplicitCompletion.lean, exact theorem Rule30.full_checkerboard_completion_exists. Internal review/math-audit.md, 2026-10-08: PASS for the stated theorem after fresh proof reconstruction and Lean rebuild. Phase20ExplicitCompletion.lean SHA-256 afaba14b19f00aa57747bf9d8c42ea03e2b7bb2ec6acf423eebf0b893a8fadaf. The supplied later data commit 7cd3e0cdebf3802942d1579a96387ef1ece92374 was context only and was not needed for this proof.
Scope
Full binary initial rows on the integer line. For every rightmost one R≥1, there is an S≥1 and a row v with v(n)=1 iff n-S is even for n≥S, preserving every t≥0 and every n≤0. No restriction on the left tail. The R=0 lone seed is checked separately; no assertion for all rows with R=0.
Assumptions
Synchronous Rule 30 on the infinite integer line: next = left XOR (center OR right). u(R)=1, u(n)=0 for n>R, and R≥1. No left-support or initial-center-bit assumption.
Toolchain
Lean 4.34.0, x86_64-w64-windows-gnu, commit 293d5d0c0c3f3dded4688b3ccd6a33939ac5102b. Fresh rebuild of the ten frozen modules passed on 2026-10-08. Lean executable SHA-256 a8040e2cab341c12116ab591fed9f761816f5f6b08554f6cc1680e86dfbba0a2. Named theorem axioms: propext, Classical.choice, Quot.sound.
Reviewer
One independent runtime AI subagent, 2026-10-08; reconstructed argument, matched exact Lean hypotheses, rebuilt frozen dependency closure and ran a separately written bounded replay. Internal AI review only; no human, external or peer review is asserted.

R30-SINGLE-SEED-CHECKERBOARD Proof provided

The row occupied only at 0 and the row occupied exactly at 0,2,4,… have identical Rule 30 evolutions at every t≥0 and n≤0, hence identical center sequences.

Formal verification
Formally verified within the stated scope
Review status
Internal check
Evidence
Frozen source 85d7d83c06f692f3fa7d19af42b6c988e0f58ca0: formal/Phase17InfiniteFiber.lean, Rule30.infiniteEvenRow_same_left_trace and Rule30.infiniteEvenRow_same_center. File SHA-256 b85bedaa8878aa67f10b5270f4d575c76eff6d60a4b0f7ff9bd6cb9c9104264c. Internal review/math-audit.md checks this R=0 case separately and includes its module in the fresh build. The independent bounded replay also checked this pair through t=128 on n=-128,...,0; that finite replay is supplementary, not the all-time proof.
Scope
The two specified full binary rows, including all negative positions initially zero; every nonnegative time and all integer positions n≤0. This is a direct zero-penultimate completion at R=0, not an extension of the general R≥1 theorem to all R=0 rows.
Assumptions
Synchronous Rule 30 on the infinite integer line. u(n)=1 iff n=0; v(n)=1 iff n is a nonnegative even integer. In particular u(-1)=0.
Toolchain
Lean 4.34.0, x86_64-w64-windows-gnu, commit 293d5d0c0c3f3dded4688b3ccd6a33939ac5102b. Fresh rebuild of the ten frozen modules passed on 2026-10-08. Lean executable SHA-256 a8040e2cab341c12116ab591fed9f761816f5f6b08554f6cc1680e86dfbba0a2. Named theorem axioms: propext, Classical.choice, Quot.sound.
Reviewer
One independent runtime AI subagent, 2026-10-08; reconstructed argument, matched exact Lean hypotheses, rebuilt frozen dependency closure and ran a separately written bounded replay. Internal AI review only; no human, external or peer review is asserted.

R30-PRIZE-STILL-OPEN Unresolved

This construction does not settle eventual nonperiodicity of the single-seed center column and excludes no proposed period word.

Formal verification
No formal verification
Review status
Internal check
Evidence
Scope of proofs/phase20_explicit_checkerboard_completion.md at frozen source 85d7d83c06f692f3fa7d19af42b6c988e0f58ca0 and internal review/math-audit.md, 2026-10-08. Equality of two initial-row traces does not decide whether the shared center trace is eventually periodic.
Scope
The separate Rule 30 single-seed center-column prize problem.
Reviewer
Internal AI mathematical and editorial scope checks; no claimed resolution of the open problem.

Files & reproduction

Checksums identify the exact files. Code is provided as a download and is not run by this website.

Contributions & AI assistance

AI assistance

AI contributed to the proof, code, source search and translations. The exact Lean build and a separate reconstruction of the edge argument are described above.

References

  1. Stephen Wolfram — A New Kind of Science, Rule 30 reconstruction note (p. 1087)
  2. Eric Rowland — Local Nested Structure in Rule 30
  3. Johan Kopra — Rapid left expansivity (2023), Corollary 3.7 and Problem 3.10
  4. David L. Condrey — Finite Configurations Cannot Generate a Constant Trace in Rule 30 (2026)

Version history

VersionPublishedChange note
v4Updated publication information.
v3Added German and Chinese translations for existing claim evidence, scope, assumptions, toolchain and reviewer metadata only; no claim, proof, status, source text or review-status promotion.
v2Clarified the publication-level status and revised English, German and Chinese copy; the three claim statuses, proof and evidence are unchanged.
v1Published the research note; priority remains unresolved.
Archived wording

Editorial draft after three internal AI copy reviews. Exact general R>=1 theorem and separate R=0 single-seed example checked in one independent internal AI math audit with a fresh Lean 4.34.0 rebuild of ten modules. Publication priority unresolved; not published.

Cite this version

Jordi Gartner. Rule 30: different starting rows, the same left-hand evolution. Beedbyte, v3, 2026-10-08.

https://beedbyte.tech/publications/rule-30-checkerboard-completion/v/3

SHA-256 — Versionb7fd55a58154dd9ecddc7c9a35f1defe8dadfa6bc8cd77fb8b0582f8c9ab378f

Comments

Discussion of version 3

No comments yet.

Comments on this version are preserved here. Comment on the current version →