Lean normalization and density transfer

What Lean checks

All twelve example values and both critical-family reductions are checked. The unconditional four-speed gap theorem is proved: below 2/9, precisely the two scaled critical directions occur, both with value 1/5. This also proves the four-speed runner bound. The unconditional five-speed lower bound L(v) ≥ 1/6 is also proved by Renault’s argument. All 122,880 indexed plane cases and the five-speed coordinate cutoff are checked, and their gap input is now available. The bounded five-speed classification remains open.

These files prove normalization, coefficient-rounding transfer, and the arithmetic of the original lattice norm reduction. The original sharp lattice bound is not asserted by these files.

A later, independently checked pigeonhole argument supplies an alternative finite/infinite reduction. It gives the unconditional four-speed cutoff 5832 and, under the explicit four-speed gap, the five-speed nonfamily cutoff 279841. The finite critical-plane classification is fully checked in that route. See Elementary finite/infinite reduction. The two finite tuple classifications remain separate; the alternative does not require the lattice-volume theorems listed below.

Gcd normalization

LonelinessNormalization.lean defines the greatest common divisor of a list by an iterated Nat.gcd, and defines normalize ds by dividing every speed by that divisor.

value_scale proves, for every positive integer dd,

value(dds)=value(ds).\mathop{\mathrm{value}}(d\,ds)=\mathop{\mathrm{value}}(ds).

The upper bound reduces arbitrary real times modulo one. The lower bound uses the time t/dt/d for every t∈[0,1]t\in[0,1].

For a nonempty positive list, normalization preserves its actual value, length, positivity, and distinctness, and makes its greatest common divisor one. primitive_representative collects these conclusions in a single existence theorem.

Normalization also preserves containment of either critical quadruple. This removes the primitivity hypothesis from the previously proved classification:

Loneliness.Normalization.critical_family_below_threshold

For every positive speed list of length five containing a positive dilation of [1,2,3,4] or [1,3,4,7], a value below 1/51/5 therefore equals s/(5s+1)s/(5s+1) for some positive integer ss.

Density transfer by rounding coefficients

LonelinessSubspaceDensity.lean works directly with a real linear subspace E⊆ℝnE\subseteq\mathbb R^n. It defines

G(x)=mini∥xi∥,M(E)=supx∈EG(x).G(x)=\min_i\|x_i\|, \qquad M(E)=\sup_{x\in E}G(x).

The definitions are pointClearance and spaceValue. The theorem spaceValue_speed_line identifies M(ℝv)M(\mathbb Rv) with the existing actual real-time loneliness value of the natural speed list vv.

Let w1,…,wj∈ℤnw_1,\ldots,w_j\in\mathbb Z^n, and choose arbitrary approximating vectors vi∈Ev_i\in E. Suppose

|(wi)a−(vi)a|≤rifor every i,a.|(w_i)_a-(v_i)_a|\le r_i \quad\text{for every }i,a.

extension_density proves that every point

x∈E+spanℝ{w1,…,wj}x\in E+\mathop{\mathrm{span}}_{\mathbb R}\{w_1,\ldots,w_j\}

has a vector y∈Ey\in E and an integer vector z∈ℤnz\in\mathbb Z^n with

maxa|xa−za−ya|≤12∑i=1jri.\max_a|x_a-z_a-y_a|\le\frac12\sum_{i=1}^j r_i.

The construction is explicit. Express x=e+∑ibiwix=e+\sum_i b_iw_i, and put

z=∑iround(bi)wi,y=e+∑i(bi−round(bi))vi.z=\sum_i\mathop{\mathrm{round}}(b_i)w_i, \qquad y=e+\sum_i\bigl(b_i-\mathop{\mathrm{round}}(b_i)\bigr)v_i.

Then

x−z−y=∑i(bi−round(bi))(wi−vi).x-z-y=\sum_i\bigl(b_i-\mathop{\mathrm{round}}(b_i)\bigr)(w_i-v_i).

Each scalar coefficient has absolute value at most 1/21/2. The triangle inequality gives the asserted coordinate bound. This construction does not require the added vectors to be linearly independent.

point_transfer proves the required Lipschitz inequality modulo integer vectors. extension_value_bound and extension_value_bound_norm conclude

M(E+span{wi})≤M(E)+12∑i∥wi−vi∥∞.M\bigl(E+\mathop{\mathrm{span}}\{w_i\}\bigr) \le M(E)+\frac12\sum_i\|w_i-v_i\|_\infty.

Taking viv_i to be the orthogonal projection of wiw_i onto EE gives the coefficient-rounding transfer used by the lattice method. The existence of short projected lattice vectors is a separate mathematical step.

Exact arithmetic of the norm thresholds

LonelinessNormArithmetic.lean proves the numerical part with all lattice and subspace hypotheses explicit. It does not assert that those hypotheses have already been established for every speed tuple.

For ordered nonnegative numbers a≤b≤c≤da\le b\le c\le d, the inequalities

a+b>110,a+b+c>415,a+b+c+d>35a+b>\frac1{10},\qquad a+b+c>\frac4{15},\qquad a+b+c+d>\frac35

imply

b>120,c>445,d>320,bcd>11500.b>\frac1{20},\qquad c>\frac4{45},\qquad d>\frac3{20}, \qquad bcd>\frac1{1500}.

Given abcd≤1/Vabcd\le1/V, first_minimum_bound proves a<1500/Va<1500/V. five_speed_threshold combines this with the density inequalities and V≥33750V\ge33750 to prove that the containing two-dimensional space has value strictly below 2/92/9.

The analogous three-minimum calculation gives A<4374/5A<4374/5. squared_covolume_integer_bound proves that an integral squared covolume then satisfies A2≤765275A^2\le765275.

Dependencies of the original lattice route

The source for the external reduction is Yi, Gill and Bu, Four runners that are barely lonely, arXiv:2610.40694v1, especially Sections 3, 11, and 12. A readable PDF mirror contains the same paper.

Corollary 11.6(ii) uses the projected-lattice covolume identity, Vaaler's cube-section theorem, Minkowski's second theorem, coordinate coalescence, and Theorem 11.5. The latter uses the four-speed gap and an additional enumeration. The four-speed proof enumerates 800,816,099 quadruples. The five-coordinate g5enum2 enumeration visits 65,989,492,206 vectors, retains 30,485,136,454 normalized vectors, and has sixteen survivors. These searches are separate from this repository's final 317 five-speed candidates.

The pinned Mathlib contains lattice bases and covolumes in Mathlib.Algebra.Module.ZLattice.Basic and Mathlib.Algebra.Module.ZLattice.Covolume. It also contains Blichfeldt's theorem and Minkowski's first convex-body theorem in Mathlib.MeasureTheory.Group.GeometryOfNumbers. Searches of this pinned source did not find Vaaler's theorem, Minkowski's second theorem, successive minima, or a lonely-runner classification.

These particular files prove normalization, density transfer, and numerical threshold calculations. They leave the original lattice hypotheses explicit. The later pigeonhole route replaces those hypotheses with elementary finite approximation and the checked critical-plane certificate. Completion of the four-speed finite classification and the final five-speed finite classification is still needed for the full exhaustion theorem.