Upstream packet: sum-range-odd-eq-sq

Status: packet-ready · generated mechanically (ADR-020 / SPEC-020-A) · sponsor: Chris Barlow

The statement (as proved here)

import Mathlib

theorem sum_range_odd_eq_sq (n : ℕ) : ∑ i ∈ Finset.range n, (2 * i + 1) = n ^ 2 := by
  sorry

Kernel-verified on main: library/Unsorry/SumRangeOddEqSq.lean (theorem sum_range_odd_eq_sq), through Gate A (build --wfail, axiom audit against the standard whitelist, leanchecker kernel replay, regenerated ADR-011 binding obligation).

Proposed contribution

The git apply-able new-file diff is at sum-range-odd-eq-sq.patch. The target path Mathlib/Unsorry/SumRangeOddEqSq.lean is a placeholder — file placement and the final name are Zulip questions, not ours to decide. Content:

/-
Copyright (c) 2026 Chris Barlow. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Chris Barlow
-/
import Mathlib.Algebra.BigOperators.Group.Finset.Basic
import Mathlib.Tactic.Ring

theorem sum_range_odd_eq_sq (n : ℕ) : ∑ i ∈ Finset.range n, (2 * i + 1) = n ^ 2 := by
  induction n with
  | zero => simp
  | succ k ih =>
    rw [Finset.sum_range_succ, ih]
    ring

Dedup at mathlib HEAD

A name-grep is a pre-filter, not a proof of absence; the kernel build at HEAD (tools/upstream/verify_head.sh) is the strong evidence and its result belongs in the PR conversation.

Provenance dossier

Field	Value
source	classic identities
reference	Classical ‘gnomon’ identity (sum of consecutive odd numbers is a perfect square). Graham, Knuth & Patashnik, Concrete Mathematics, 2nd ed., §2.5; Rosen, Discrete Mathematics and Its Applications, 7th…
absence	machine-checked no-local-match (grep of pinned mathlib rev c5ea00351c28, 2026-06-10); related lemmas exist but are different identities
difficulty	1
decomposition sketch	Single induction on n. Base n=0 trivial. Step via Finset.sum_range_succ then ring/omega: n^2 + (2n+1) = (n+1)^2. No sub-lemmas. Optional Post⊆Pre edge: factor as 2*(∑ k) + n then apply existing Finset.sum_range_id.
title	For every natural n, the sum over i in 0..n-1 of (2i+1) equals n^2 (i.e. 1+3+5+…+(2n-1) = n^2).

Field

Value

source

classic identities

reference

Classical ‘gnomon’ identity (sum of consecutive odd numbers is a perfect square). Graham, Knuth & Patashnik, Concrete Mathematics, 2nd ed., §2.5; Rosen, Discrete Mathematics and Its Applications, 7th…

absence

machine-checked no-local-match (grep of pinned mathlib rev c5ea00351c28, 2026-06-10); related lemmas exist but are different identities

difficulty

decomposition sketch

Single induction on n. Base n=0 trivial. Step via Finset.sum_range_succ then ring/omega: n^2 + (2n+1) = (n+1)^2. No sub-lemmas. Optional Post⊆Pre edge: factor as 2*(∑ k) + n then apply existing Finset.sum_range_id.

title

For every natural n, the sum over i in 0..n-1 of (2i+1) equals n^2 (i.e. 1+3+5+…+(2n-1) = n^2).

Proof produced by an autonomous Claude agent swarm (model policy ADR-013/ADR-015: fable, progressive effort), merged with no human review through two CI gates (ADR-006 soundness, Gate B hygiene). Full machine history: the goal’s PR trail in this repository.

AI disclosure (paste-ready facts)

The Lean proof in this PR was produced by an autonomous LLM agent (Anthropic Claude, model fable) operating in the unsorry proof swarm (github.com/agenticsnz/unsorry), and was machine-verified there by kernel replay, an axiom audit against the standard whitelist (propext, Classical.choice, Quot.sound), and a CI-regenerated statement-binding obligation. I have read and understood the proof in full and can justify each step without AI assistance. Label: LLM-generated.

For the sponsor

Read the proof until you can justify every step without AI assistance — mathlib reviewers will expect exactly that.

Zulip first, in your own words: is the lemma wanted, where does it live, what should it be called? The PR-description narrative and every review reply likewise must be rewritten in your own words — mathlib policy forbids LLM-written conversation; only the lemma itself (disclosed) and the factual disclosure block above may be pasted.

Raise the draft PR with one command once you’ve done 1–2 — from the unsorry repo root:

python3 -m tools.upstream.raise_pr --goal sum-range-odd-eq-sq --fork <your-github-user> --understood

It clones mathlib master, applies the patch to a fresh branch, pushes to your fork, and opens a draft PR pre-filled with the factual disclosure and a placeholder where your narrative goes. (--understood is your attestation that you’ve read the proof; --dry-run shows the plan first.) The machine never marks it ready and never writes a review reply.

Write your narrative in the draft, apply the LLM-generated label, then you flip draft → ready. Expect the linter to want golfing (binder names, line length) — that editing is yours. See docs/upstreaming.md.

Record the outcome on the targets board (

in-discussion → pr-open →
merged | declined

). Declined is a valid, recorded result.