The Prefix Conjecture for almost Golomb sequences

From papers

Let ara_r be the order-rr almost Golomb sequence, defined by the sliding-window greedy recurrence in the source, and let GG denote Golomb's sequence. Prefix Conjecture. For all r3r\ge 3,

ar(r)=G(r1).a_r(r)=G(r-1).

This has been verified computationally for all r200r\le 200. It asserts that the common initial prefix of the order-rr sequences, evaluated at n=rn=r, agrees with the corresponding value of Golomb's sequence; proving this boundary identity remains open.

Progress summary

Open

The conjecture has been checked through r=200r=200, but no public proof or counterexample has been found.

The conjecture asserts that the order-rr almost Golomb sequence agrees with Golomb’s sequence at the boundary value n=rn=r, for every r3r\ge 3. It is presented as an open conjecture in the 2026 paper Almost Golomb Sequences.

Known results

  • The identity has been verified computationally for every r200r\le 200.
  • The first rr terms of ara_r coincide with the common full-memory greedy prefix.
  • A proved run-length identity, Nr(m)=ar(m+1)ar(m+1r)N_r(m)=a_r(m+1)-a_r(m+1-r), reduces a broader maximal-multiplicity conjecture to this conjecture plus a separate Domination Lemma.

April 2026 paper

The paper formalized the Prefix Conjecture as Conjecture 9.3 and explicitly stated that a complete proof of the surrounding maximal-multiplicity conjecture remains open. The scan found no later proof, counterexample, verification, or model-based solution announcement.

Current status (as of August 2026): The identity is computationally established for r200r\le 200, but the assertion for all r3r\ge 3 remains open, with no publicly recorded proof or counterexample.

Sources
Sources & referencesView supporting material

Primary source

Benoit Cloitre, “Almost Golomb Sequences”, arXiv:2604.02404 (2026).

Solutions 1

Proof

A stronger exact maximal-prefix theorem holds. Let GG be Golomb's sequence, and write

Cm=j=1mG(j),C0=0.C_m=\sum_{j=1}^mG(j),\qquad C_0=0.

For every r3r\ge3, the order-rr almost Golomb sequence satisfies

ar(n)=G(n1)(3nCr).\boxed{ a_r(n)=G(n-1) \qquad(3\le n\le C_r). }

Moreover, the next term is the first discrepancy:

ar(Cr+1)=r+1r=G(Cr).a_r(C_r+1)=r+1 \ne r=G(C_r).

In particular, rCrr\le C_r, so the original conjecture

ar(r)=G(r1)a_r(r)=G(r-1)

follows immediately.

To prove the stronger assertion, write a=ara=a_r, extend a(j)=0a(j)=0 for j0j\le0, and put

Sm=i=0r1a(mi).S_m=\sum_{i=0}^{r-1}a(m-i).

The established run characterization gives, for every m3m\ge3,

a(n)=mSmn<Sm+1.(1)a(n)=m \quad\Longleftrightarrow\quad S_m\le n<S_{m+1}. \tag{1}

The initial values are a(1)=1a(1)=1, a(2)=a(3)=2a(2)=a(3)=2; thus S3=5S_3=5, and a(4)=2a(4)=2.

First, we prove by strong induction that

a(n)=G(n1)(3nCr1+1).(2)a(n)=G(n-1) \qquad(3\le n\le C_{r-1}+1). \tag{2}

The cases n=3,4n=3,4 hold by the initial values. Suppose n5n\ge5, and put

g=G(n1).g=G(n-1).

The Golomb run boundaries are

Cg1+1n1Cg.(3)C_{g-1}+1\le n-1\le C_g. \tag{3}

Since nCr1+1n\le C_{r-1}+1, we have gr1g\le r-1. Also g3g\ge3 and g+1n1g+1\le n-1. Consequently, both windows Sg,Sg+1S_g,S_{g+1} are untruncated, and all entries in them are known inductively. Hence, for m=g,g+1m=g,g+1,

Sm=j=1ma(j)=1+2+j=3mG(j1)=2+Cm1.S_m = \sum_{j=1}^m a(j) = 1+2+\sum_{j=3}^mG(j-1) = 2+C_{m-1}.

By (3),

Sg=Cg1+2n<Cg+2=Sg+1.S_g=C_{g-1}+2 \le n < C_g+2=S_{g+1}.

Equation (1) now gives a(n)=g=G(n1)a(n)=g=G(n-1), proving (2).

Because Cr1rC_{r-1}\ge r, equation (2) also gives

a(r+1)=G(r).a(r+1)=G(r).

Therefore

Sr=2+Cr1,S_r=2+C_{r-1},

and the next sliding window satisfies

Sr+1=Sr+a(r+1)a(1)=Cr+1.S_{r+1} = S_r+a(r+1)-a(1) = C_r+1.

Thus the value rr occupies precisely the positions

Cr1+2,,CrC_{r-1}+2,\ldots,C_r

in ara_r. The shifted Golomb sequence G(n1)G(n-1) occupies the same positions with value rr, and additionally has value rr at n=Cr+1n=C_r+1.

Combining this with (2) proves equality throughout

3nCr.3\le n\le C_r.

At n=Cr+1=Sr+1n=C_r+1=S_{r+1}, equation (1) gives

ar(Cr+1)=r+1,G(Cr)=r.a_r(C_r+1)=r+1, \qquad G(C_r)=r.

Hence the agreement interval is exact, proving the conjecture and its sharp maximal-prefix strengthening.

0 endorsements
Shivam Patel ·