Certify v0.10.0: The extremal graph for k(3,4) = r(I_3,L_4) = 21 is not unique — thirteen pairwise non-isomorphic, rigid, {I_3,TT_4}-free oriented graphs on 20 vertices, independently verified
Description
This is an addendum (Part J) to the determination of the Erdos-Rado oriented Ramsey number k(3,4) = r(I_3,L_4) = 21 (Part G, version DOI 10.5281/zenodo.21890619), answering its Question 8.1. Part G exhibits an explicit {I_3,TT_4}-free oriented graph W on 20 vertices (the lower bound for k(3,4) = 21) and asks whether W is the unique such graph up to isomorphism. The answer is NO. This release deposits thirteen {I_3,TT_4}-free oriented graphs on 20 vertices — W itself (w01_W.json, verified equal to Part G's Table 1) together with twelve more — that are pairwise non-isomorphic and every one of them RIGID (trivial automorphism group). So the 20-vertex extremal graph is not unique, and none of these thirteen is vertex-transitive; in particular none of them is a Cayley digraph. Vertex-transitive extremal graphs do exist (see the correction at the end). THE PUBLIC UNIT IS CERTIFICATES PLUS STANDARD-LIBRARY CHECKERS: each witness ships as a JSON arc list, and three stdlib-only checkers re-derive everything from the arc lists alone. verify_witnesses.py certifies, for each of the thirteen, validity as an oriented graph, {I_3,TT_4}-freeness by EXHAUSTIVE test over all C(20,3)=1140 triples and all C(20,4)=4845 quadruples, rigidity certified two independent ways (Weisfeiler-Leman colour refinement discrete at twenty singleton colours, cross-checked by an explicit automorphism count of 1), and pairwise non-isomorphism (a canonical form read off the discrete refinement, cross-checked by an invariant fingerprint of arc count, out- and in-degree pairs, and non-adjacency-graph degree sequence that already separates all thirteen). verify_qr7_lemma.py verifies the inputs to the QR_7 forcing lemma by exhaustive enumeration: there are exactly 240 labelled TT_4-free tournaments on 7 vertices and every one is isomorphic to the Paley tournament QR_7 = Cay(Z_7,{1,2,4}) (so it is the unique 7-vertex TT_4-free tournament; |Aut(QR_7)| = 21 and 240 = 7!/21), and there are none on 8 vertices (so every 8-vertex tournament contains a transitive TT_4). The forcing lemma follows: in any {I_3,TT_4}-free oriented graph a vertex has at most 7 non-neighbours, and a vertex with exactly 7 has its non-neighbourhood equal to QR_7 — the classical characterisation of A. Sanchez-Flores (Graphs Combin. 14 (1998), 181-200), re-established here by exhaustive enumeration and matching Part G, Prop. 3.3. blowup_bound.py checks that the largest {I_3,TT_4}-free tournament blow-up is QR_7[I_2] on 14 vertices (giving k(3,4) >= 15), which is six short of the extremal order 20, so no tournament blow-up is extremal. All three checkers import only the Python standard library, are not code-independent of the private search that produced the twelve additional witnesses, use no threads/subprocess/OS-specific calls, and return PASS on CPython 3.14.2 and 3.9.6. SCOPE, STATED PLAINLY: the result is a LOWER BOUND — at least thirteen pairwise non-isomorphic rigid extremal graphs; the exact number is not determined (the sampling that found the witnesses produced a new isomorphism class on every surviving completion, suggesting a large family, but no count is certified). Whether a QR_7 block is unavoidable — equivalently, whether the instance '{I_3,TT_4}-free on 20 vertices with every vertex having at most 6 non-neighbours' is unsatisfiable — is left OPEN. The search internals that produced the twelve additional witnesses are not part of this deposit; only the certificates and stdlib checkers are public. Computations carried out with Claude (Anthropic), directed by the author; the checkers are not code-independent of that search. The dated novelty and verification sweep ships as SWEEP-RECORD-K34ADD-2026-08-11.md and the note as k34add-paper/note.pdf. Correction, 17 November 2026: the earlier text of this description said that no 20-vertex extremal graph is vertex-transitive and that none is a Cayley digraph. That was wrong: the circulant Cay(Z_20,{2,4,5,7,11,14}) is an {I_3,TT_4}-free oriented graph on 20 vertices, as the reviewers at erdosproblemaday.com pointed out. The first arXiv version of the paper of Ihringer, Rajendraprasad and Weinert (arXiv:1707.09556v1, 2017) remarks that the largest {I_3,TT_4}-free Cayley graph has 20 vertices. The earlier text also concluded from the blow-up bound that the extremal family is substantially non-algebraic; that conclusion is withdrawn. The file SWEEP-RECORD-K34ADD-2026-08-11.md in this archive makes the same wrong statement about vertex-transitive graphs and cannot be changed here. Three other statements were changed on the same date to agree with the corrections published in versions 0.14.0 and 0.15.0 of Certify. The earlier text named a fingerprint (arc count, non-adjacency degree sequence, QR_7-block count) that does not separate all thirteen graphs; the fingerprint the checker computes does. The earlier text said twice that the checkers share no code with the search, and once that a fact was re-established from scratch; the checkers are not code-independent of the private search that produced the twelve additional witnesses.
Files
05oz/certify-v0.10.0.zip
Additional details
Related works
- Is supplement to
- Software: https://github.com/05oz/certify/tree/v0.10.0 (URL)
Software
- Repository URL
- https://github.com/05oz/certify