Generalized Bulgarian solitaire limit shapes

What Lean checks

Actual moves for arbitrary layer sets containing 1; all recurrent states; classification and maximality of stable references; one-column interlacing; the explicit uniform profile bound and convergence for every diverging scale. The separate fixed-scale counterexample remains checked.

Collection · Catalogue · Verification · Lean status

Status: both the fixed-scale counterexample and the separate corrected diverging-scale theorem have passed Lean source checks, standard-axiom audits, and independent kernel replay. See GeneralTheorem.lean and the exact checked scope. The note gives a counterexample to Conjecture 1 of Eriksson--Jonsson--Sjoestrand (2020) under its published definition, plus a proof of the conjectured conclusion for all layer-picking rules at any scaling a_n -> infinity. Unrefereed; originality and publication priority are not certified.

For n=101, the pick-layer set is L={1,...,51}. Under this single deterministic game rule, (76,25) is stable, while (101) <-> (51,50) is a recurrent two-cycle. For L_n={1,...,floor(n/2)+1} and constant horizontal scaling a_n=1, the reference stable limit has heights 3/4,1/4 on successive unit intervals, whereas the recurrent one-pile limit has height 1 on (0,1). The original Conjecture 1 allows this scaling; Theorem 4's separate assumptions require a_n -> infinity.

The second part proves an interlacing estimate for every layer-picking set. If rho is recurrent on n cards and zeta is the reference stable partition of size s<=n, then zeta[i+1]<=rho[i] and rho[i+1]<=zeta[i]. For x>=epsilon>0, a*epsilon>=2, one obtains

|D_a(rho,x)-D_a(zeta,x)| <= 4/(a*epsilon^2) + sqrt(2)/(epsilon*sqrt(n)).

Consequently the profiles agree uniformly away from zero whenever a_n -> infinity.

Run python3 verify.py to enumerate all admissible rules on decks 1 through 14 and verify the explicit counterexample through deck size 1000. The arguments in PROOF.md apply for arbitrary n.

Primary source: Eriksson, Jonsson, Sjoestrand (2020), Conjecture 1. For the known classification of fixed-bottom-B cycles, see Hopkins (2024), Theorem 5.