abc conjecture - Proof Program

This record contains a “corridor-certified” proof artifact for the abc conjecture, written as an explicit dependency DAG. Each node carries rigid Imports / Exports / Forbidden firewalls. Status note: the corridor is fully closed (no OPEN locks) under the external certified black boxes listed as CERT QST, CERT SUNIT, and CERT PADIC, and all thresholds are fixed once and for all by NODE 1 (Ledger) as explicit functions of ε. Target statement (tail sealing form) For each fixed ε∈(0,1), there exists an effectively determined constant C(ε) such that for every coprime triple a,b,c∈ℕ with a+b=c,rad(abc) < c^(1−ε) ⇒ c ≤ C(ε),where rad(n)=∏_{p|n} p. Consequently, for each ε>0 the set of ε-violations is finite. Core proof-line (Unicode, node-aligned, thesis-level) Normalization (deterministic): work with coprime positive integers a+b=c, and WLOG 1≤a≤b (swap if needed). Then a≤c/2 and x:=a/c∈(0,1/2]. NODE 1 (Ledger; fixed constants, no tuning): fix ε∈(0,1) and setθ(ε)=ε/10, V(ε)=⌈20/ε⌉+2, K(ε)=⌈200/ε²⌉,δ(ε)=θ(ε)/(V(ε)−1)>0, R(ε)=⌈2/δ(ε)⌉.Forbidden: any re-choice of these constants after Ledger; any triple-dependent tuning. Mass identity (bookkeeping): defineM(a,b,c)=log(abc)−log rad(abc)=∑_{p|abc} (v_p(abc)−1)·log p. NODE 2 (Asymmetry patch): for any ε-violation, either (A) a≤c^θ (asymmetric) or (B) a>c^θ (balanced-enough), and in case (B) log(abc)≥(1+θ)·log c. NODE 3 (Mass lower bound in balanced branch): if rad(abc)c^θ, thenM(a,b,c) ≥ (ε+θ)·log c.(If a≤c^θ, proceed with Nodes 4–6; Node 3 is not invoked in that branch.) NODE 4 (Support split): let S=S(abc)={p : p|abc}. Then either (S1) |S|≤K or (S2) |S|>K. NODE 5 (CERT SUNIT; bounded support closure): if |S|≤K, then c≤C_SUNIT(ε) effectively; hence no tail violations in (S1). NODE 6 (Heavy vs light split): in case |S|>K, either(H) ∃p with v_p(abc)≥V or (L) v_p(abc)c^θ, select distinct primes dividing a (or b) until the partial product exceeds c^δ. If it never exceeds c^δ, then rad(a)≤c^δ and hence a≤rad(a)^(V−1)≤c^θ, contradicting a>c^θ. Therefore the selection succeeds (or we are already in (H)). NODE 8 (Bridge; divisibility → proximity): for x=a/c∈ℚ{0,1}, if p^t|a then x is p-adically close to 0 by ≥min(t,V)·log p; if p^t|b then x is close to 1; if p^t|c then x is close to ∞. These contributions feed the truncated counting function used by CERT QST; “truncation” here is only the internal level V built into CERT QST (no external truncation theorem is imported). NODE 7 (CERT QST; light closure): in the light regime (L) with a>c^θ, Lemma 1 yields a deterministic lower bound log rad(a) ≥ δ·log c (or the same for b). Together with NODE 9 (prime selection) and NODE 8 (bridge), this yields a product-of-local-norms inequality for the linear forms L₁(X)=X and L₂(X)=1−X at finitely many places. By CERT QST, all such x∈ℚ lie in finitely many proper linear subspaces (in ℚ², equivalently ℚ³ in homogeneous coordinates (X,Y,Z)); pulling back to the fixed projective line X+Y=Z shows only finitely many rational x occur (degenerate relations X=0 or Y=0 are excluded by a,b>0 and gcd(a,b,c)=1). Hence c≤C_QST(ε) effectively in regime (L). NODE 10 (CERT PADIC; heavy closure, mandatory): in regime (H), CERT PADIC yields either (i) c≤C_PADIC(ε) or (ii) heavy reduces to bounded support, hence NODE 5 applies. NODE 11 (Glue; E1 execution): combining Nodes 2–10, defineC(ε)=max{C_SUNIT(ε), C_QST(ε), C_PADIC(ε)}.Then rad(abc)