A Lean 4 / Mathlib proof of the fundamental theorem of projective geometry, in its
lattice-theoretic form, with no sorry and no axioms beyond Lean's standard three
(propext, Classical.choice, Quot.sound). It also includes a proof that the
completeness hypothesis cannot be dropped.
ftpg_proof_limit : ftpg_statement_limit. Let L be a complete lattice that is
modular, complemented, and compactly generated. Suppose L is irreducible (every line
through two distinct atoms contains a third atom) and has height at least 4 (a chain
⊥ < a < b < c < d). Then there is a division ring D and a D-vector space V
with
L ≃o Submodule D V
ftpg_proof_finite : ftpg_statement_finite. The same conclusion for a bounded
lattice that is modular, complemented, and atomistic, irreducible, of height at least
4, and of finite Krull dimension. This is derived from the limit form
(ftpg_finite_of_limit): finite height implies completeness and compact generation.
not_ftpg_statement. Without completeness or finite height, the statement is
false. The lattice of subspaces of (ℕ ⊕ ℕ) →₀ ℚ that are finite-dimensional or of
finite codimension is complemented, modular, atomistic, irreducible, and of height at
least 4, but it is not complete (hollow_no_lub). Every Submodule D V is complete,
so no such isomorphism exists.
The hypotheses of the limit form are tight, one counterexample family each: non-Desarguesian planes are excluded by the height condition, continuous geometries by compact generation, and the lattice above by completeness.
The division ring is built from the lattice itself, by the von Staudt–von Neumann
coordinatization. A coordinate frame (coordFrame_exists) is found from irreducibility
and height alone. Addition and multiplication are defined by incidence on a line, and
each field law is proven from Desargues' theorem. Planar Desargues is itself proven from
the modular law by lifting the configuration out of the plane, which is what height ≥ 4
provides (desargues_planar; Bridges/FTPG/Instance.lean assembles DivisionRing). Then a
Veblen–Young coordinatization (pointSystem_exists) assigns atoms to projective points
over the opposite ring: first the plane, then 3-space, then finite-dimensional windows,
then a directed limit over all of L.
Bridges/FTPG/README.md is the original working log, file by file.
lake exe cache get
lake build
Lean v4.31.0, Mathlib v4.31.0. Audit.lean checks the axiom dependencies of the
three headline theorems at build time, and every file under Bridges/FTPG/ carries its
own #guard_msgs axiom receipts.
This was developed inside lightward/foam, a
broader project whose vocabulary shows up in the docstrings and comments. Every file
under Bridges/ is byte-identical to lightward/foam at commit
91b62c0f
(2026-07-08), which is why the module paths are Bridges.FTPG.* and the namespace is
Foam.Bridges. To verify, from a checkout of foam next to this repo:
git -C foam checkout 91b62c0f
(cd ftpg && find Bridges -type f) | while read f; do cmp "foam/bridges/$f" "ftpg/$f"; done
No output means every file matches.
The proof was completed at
cf0e8135 and the classical
axiom it replaced was retired at
91b62c0f. Work on it began in
April 2026. The full commit messages there are the most detailed account of the climb.
Two notes on what's in the archive:
Bridges/FTPG/Hollow.leanstill declares the retiredaxiom ftpg(the statement without completeness, which is false). It is used only byftpg_refuted : False, as an exhibit; the main theorems do not depend on it, asAudit.leanchecks.Charge.leanfrom the same directory is left out. It depends on foam's core and isn't part of the proof.- New here: this README,
Audit.lean, the rootBridges.leanimport list, and the build files.lake-manifest.jsonis foam'sbridgesmanifest with thefoamdependency removed, so the Mathlib pin is the one the proof was checked against.
Proven by Isaac Bowen, with Claude (Opus 4.8 and Fable 5). Archived by Isaac Bowen, with Claude (Opus 5.5).
Public domain (UNLICENSE).