A Dimension Bound and Explicit Construction Prove the Deligne-Drinfeld Conjecture

Four objects can be bracketed in five ways. Reassociating from the first arrangement to the last gives two routes around a pentagon: one uses two steps, the other three. The compatibility rule requires the routes to agree. The manuscript asks which infinitesimal symmetries respect this and the other compatibility rules.

Its proposed answer is a free Lie algebra with one generator in each odd weight beginning at three: weights 3, 5, 7, and so on. “Free” means that these generators produce every solution, with no relations beyond the ordinary Lie algebra rules. The generators need not be canonical.

A Lie algebra combines expressions with a bracket that is bilinear, alternating, and satisfies the Jacobi identity. Bracketing adds weights: for example, the bracket of weight-3 and weight-5 generators has weight 8. Thus an odd-weight generating list does not mean that all even-weight components vanish. But bracketing the weight-3 generator with itself gives zero, not a new weight-6 element.

The manuscript formulates the problem using Lie polynomials in two letters, x and y, with rational coefficients; weight counts the total number of letters. A candidate must satisfy three tests: antisymmetry, a three-term identity, and a pentagon identity. Their common solution space is W.

The pentagon is evaluated in a four-strand braid Lie algebra, whose strand variables obey their own relations. The bracket on W is the Ihara bracket, not simply the ordinary bracket of polynomials. For a polynomial Θ, its associated derivation sends x to zero and y to [y, Θ], then extends by the derivation rule. These definitions specify the setting of the theorem.

At weight one, write a candidate as αx + βy. Substitution into the pentagon and cancellation of matching terms leaves −αt₁₂ − βt₃₄. The degree-one braid generators are independent, so α and β must both vanish. Thus W₁ = 0.

At weight two, every Lie polynomial in x and y is a multiple of [x,y]. Set z = −x−y, as required by the three-term test. Then [y,z] and [z,x] both equal [x,y], so the test becomes three times that bracket. Since the coefficients are rational and [x,y] is nonzero, the multiple must vanish: W₂ = 0.

This is a complete check of the first two weights, not a proof of the all-weight claim.

Let dₙ be the dimension in weight n predicted by the free Lie algebra on generators of weights 3, 5, 7, and so on. The manuscript’s main estimate is that dim Wₙ ≤ dₙ in every weight, not merely in a range checked by computation.

To obtain the bound, the proof reduces through characteristic two, where one plus one is zero. It does not reduce arbitrary rational coefficients. Instead, it intersects the rational solution space with a lattice whose denominators are odd. Saturation makes the lattice’s reduction inject into the ambient lattice after reduction and preserves dimension. The argument therefore counts reductions of rational solutions; it does not assume that the full solution space over the field with two elements is the same.

A decoding lemma provides one tool for this characteristic-two argument. In the associative algebra of words in x and y, define A = x², C = [x,y], and B = y. In characteristic two, C = xy + yx. Order equal-length words with x greater than y. The leading words of A, C, and B are xx, xy, and y. None is a prefix of another, so concatenations in the new letters decode uniquely. Distinct words in A, C, and B therefore have distinct leading words; in any finite relation, the largest leading word cannot cancel. This proves that the substitution is injective. It is an associative-algebra argument: x² is not an ordinary Lie polynomial in x and y.

The longer argument isolates terms with the lowest nonzero number of B’s and transfers the leading pentagon by comparing two ordering procedures. Deletion operators and a special identity show that setting A to zero loses no leading solution. The surviving projection is unchanged when B is replaced by B+s(C), for any polynomial s. The argument then restricts it to brackets of gₖ = ad_Cᵏ B, for k ≥ 1. Since C has weight two and B weight one, gₖ has weight 2k+1: precisely the predicted odd weights. The transfer, deletion, and image arguments are technical steps beyond the local decoding proof.

An upper bound alone does not show that the predicted generators exist. For the reverse inequality, the manuscript constructs a rational subalgebra V contained in W from compatible categorical derivations and regularized braid transport.

In every odd weight, the construction supplies a nonzero depth-one value—a term with exactly one y. Its nonvanishing ultimately follows from a convergent sum of positive terms. The proof does not assume arithmetic independence of these values. Instead, previously chosen generators supply independent Hall words, or nested brackets, whose leading projections match free Lie words.

At each new odd weight, the new depth-one value cannot be made from brackets of older generators, which have depth at least two. An integral induction chooses generators with the required reductions. Together with V contained in W and the upper bound, this yields dₙ = dim Vₙ = dim Wₙ. The generators have no additional Lie relations, and they exhaust the solution space.

The conclusion extends to completion by weight: an infinite sequence of homogeneous pieces is allowed, but this is formal completion, not numerical convergence. At any fixed weight, only finitely many pairs of pieces contribute to a bracket.