Alphabeta Math
ExampleConstruction: AI-adaptedVerification: AI-adaptedPipeline-generatedjudge pass (gpt-5.6-terra)audited 2026-09-13
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

A two-row hypercohomology spectral sequence

Example

Let K be bounded below with HqK=0 except for q=0,1, and supply the data of the second hypercohomology theorem for F. Put A=H0K, B=H1K. The only E2 rows are E2p,0=RpF(A) and E2p,1=RpF(B). With τp=d2p,1:RpF(B)→Rp+2F(A), the target RnF(K) fits into 0→coker⁡τn−2→RnF(K)→ker⁡τn−1→0, where negative-index terms are zero. The values of τ and these extensions need additional input for a general K.

Facts & Assumptions

Given: The functor, complex and supplied replacements, with DC or supplied comparisons for naturality.

[F1]

The second hypercohomology theorem gives this E2, finite decreasing filtration and differential (r,1−r) (Second hypercohomology spectral sequence).

[F2]

A computation record must retain unknown differentials and extensions explicitly (Spectral-sequence computation record).

Verification

1.1F1

For r=2 the only possible nonzero arrows are τp from row one to row zero. Hence E3p,0=coker⁡τp−2 and E3p,1=ker⁡τp. For r≥3, every outgoing arrow from either row has negative second coordinate and every incoming arrow starts above row one. These positions stay zero on successive pages, so E3=E∞.

2.1F1F2step 1.1

In total degree n, the only possible filtration quotients are at p=n and p=n−1. F1's zero/full endpoints identify the first as a subobject of the target and the second as its quotient, giving the displayed exact sequence. At n=0 it reduces to R0F(K)=F(A); at n=1 its subobject is R1F(A) and its quotient is ker⁡τ0. Below zero there are no surviving quotients, so the finite target filtration forces vanishing. This proves convergence and identifies the unresolved extension, rather than assuming a splitting.

3.1F1F2step 2.1∎

A fully numerical specialization takes abelian groups, F the identity, and K0=Z/2, K1=Z/3 with zero differential and supplied replacements. Exactness of identity means its positive derived objects vanish, since applying it preserves the exact resolution. Thus E20,0=Z/2, E20,1=Z/3, all other entries are zero, and every τp is zero. Each total degree has one nonzero quotient: the target is Z/2 in degree zero, Z/3 in degree one and zero otherwise. The upper edges are the identity under the augmentation identification. This specialization has no extension ambiguity and uses no choice beyond the supplied-data convention.

Depends on

Used by

Nothing in the library uses this result yet.

Dependency tree · two levels

20 results within two dependency steps of this one, each drawn at its shortest distance from it. An arrow runs from a result to what uses it, so the chart reads left to right and ends at this result, which carries a heavier outline. Every node is a link to that result. Click elsewhere on the chart to enlarge it.

Sources