Surjectivity of `δ` on a tree nerve.
ProvedSheafCohomologyRobustness.NerveBetti.delta_surjective_of_treeaether-catalogmachinelearning
Surjectivity of δ on a tree nerve. If the nerve is connected with
|E| = |V| − 1, then every overlap discrepancy is a coboundary: local
certificates always glue.
theorem SheafCohomologyRobustness.NerveBetti.delta_surjective_of_tree[Nonempty ι] (hconn : IsConnectedNerve (edgeAdj G))
(htree : Fintype.card Edge + 1 = Fintype.card ι) :
Function.Surjective (delta G) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source MachineLearning/SheafCohomologyRobustness/NerveBetti.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from MachineLearning/SheafCohomologyRobustness/NerveBetti.lean
import Mathlib
import Definitions.Def_MachineLearning_SheafCohomologyRobustness_GraphNervePoincare
import Definitions.Def_MachineLearning_SheafCohomologyRobustness_NerveBetti
/-
Copyright (c) 2025 Harmonic. All rights reserved.
Released under Apache 2.0 license.
# The Betti Number of a Nerve: `dim H¹ = |E| − |V| + 1`
The previous files computed the first cohomology of three specific nerves: the
path (`H¹ = 0`), the loop (`dim H¹ = 1`), and the doubly periodic torus
(`dim H¹ = 2`, with plaquette relations). This file proves the general law they
are instances of.
For a **finite oriented nerve graph** `G` — a finite set of cover regions `ι`
(vertices) and a finite set of overlaps `Edge` with endpoints `src`, `tgt` —
the Čech `0`-cochains are `ι → ℝ` and the `1`-cochains are `Edge → ℝ`, with
coboundary `(δ f) e = f (tgt e) − f (src e)`. The main results:
* `ker_delta_eq_constants` — on a connected nerve the kernel of `δ` is exactly
the line of constant certificates (`H⁰ ≅ ℝ`);
* `finrank_range_delta` — hence the space of gluable discrepancies has dimension
`|V| − 1`;
* `finrank_nerveH1_add`, `finrank_nerveH1` — therefore
`dim H¹(nerve) = |E| − |V| + 1`,
the **first Betti number** (cycle rank) of the nerve graph: the number of
independent adversarial obstruction classes equals the number of independent
loops of the cover.
* `nerveH1_eq_zero_iff_card` — in particular `H¹ = 0` exactly when the nerve has
`|E| = |V| − 1`, i.e. is a spanning tree: **certified gluing is possible for
every local datum iff the nerve is acyclic**.
-- !-- Lab Notes -- !--
* Hypothesis (Hypothesizer, grand challenge): "the number of independent
adversarial obstructions of a cover is a topological invariant of its nerve,
equal to the first Betti number `|E| − |V| + 1`."
* Experiment (Experimenter): a spanning-tree decomposition was attempted first
and proved painful in Lean; the rank–nullity route is far cleaner — the only
geometric input needed is `H⁰ = ℝ`, i.e. that a connected nerve has no
nonconstant flat certificate, which is proved by transporting along walks.
* Analysis (Analyst): this explains the earlier computations uniformly. Loop:
`|E| = |V|`, Betti `1`, matching `finrank_cyclicH1`. Path: `|E| = |V| − 1`,
Betti `0`, matching `H1_path_vanishes`. Torus: the `2`-dimensional nerve has
plaquettes, so its `H¹` is the *flat* subspace modulo coboundaries and drops
from `|V| + 1` to `2`; the discrepancy is exactly the rank of the plaquette
relations, which is the next conjecture in `FUTURE_DIRECTIONS.md`.
* Critique (Critic): the theorem needs `Nonempty ι` (an empty cover has no
base region and `H⁰ = 0`), and connectivity is genuinely used — for `k`
components the correct statement is `|E| − |V| + k`, stated as the corollary
hypothesis rather than silently assumed.
* Synthesis (PI): adversarial obstruction counting is Betti-number counting.
-/
open Finset
open SheafCohomologyRobustness
open NerveBetti
open GraphNerve
variable {ι Edge : Type*} (G : NerveGraph ι Edge)
variable [Fintype ι] [Fintype Edge]Formal statement
theorem SheafCohomologyRobustness.NerveBetti.delta_surjective_of_tree[Nonempty ι] (hconn : IsConnectedNerve (edgeAdj G))
(htree : Fintype.card Edge + 1 = Fintype.card ι) :
Function.Surjective (delta G) := by sorrySource