A global girth obstruction for Garg–Mineyev taiko product structures

The construction problem

Over the two-element field F₂, the target is a pair of nonzero group-ring elements whose product is zero, or a nontrivial unit: an invertible element that is not a group element. The assembly seeks such examples in group rings of torsion-free CAT(0) groups, a class defined by nonpositively curved geometry. Torsion-free means that no group element other than the identity has finite order.

In a group ring over the two-element field, a finite sum is described by its support. Multiplying two such sums amounts to multiplying every element in one support by every element in the other. Terms cancel when the same product occurs an even number of times. A product board records these encounters: its rows form a set A, its columns a set B, and its positions form A × B.

Mineyev developed a geometric assembly method, and Garg–Mineyev formulated its finite support data as product structures. Paired board positions encode the intended cancellations. In the even case all positions are paired; in the odd case one position remains. Coherent orientation, a no-fold condition, and three possible combinations of graph-girth requirements make this a sufficient route to the intended group-ring constructions. [2]

Each paired cell {(a, b), (a′, b′)} joins distinct rows and columns. It determines an edge aa′ in the row graph and an edge bb′ in the column graph. Repeated edges with the same endpoints are counted once, and the cell pairings determine their colors. A coherent orientation points the two edges of each pair in corresponding directions. No-fold means that a vertex has at most one incoming edge and one outgoing edge of any color.

A graph’s girth is the length of its shortest cycle, with girth taken as infinite when there is no cycle. Here girth ignores the edge directions. The triple-girth condition asks the two horizontal graphs and a middle link to satisfy at least one of these three regimes:

The assembly framework is the starting point. The paper’s contribution is the incidence analysis that proves its requirements incompatible at every support size in the theorem.

Choose the incidence description

Fix a coherent orientation of the colored horizontal graphs. Splitting each color into its outgoing and incoming versions produces signed colors. Each signed color s determines a row set A_s and a column set B_s, hence a combinatorial rectangle A_s × B_s.

Under the standing orientation and no-fold hypotheses, middle-link girth at least six is equivalent to a simple incidence restriction: two distinct signed colors share at most one point across A and B together. Their rectangles partition the board, apart from the possible single defect. They may share a row or a column vertex, but no covered board position belongs to two rectangles. The incoming and outgoing rectangles of each color have matching side sizes. Structure Theorem 3.3.

This description makes the competing demands visible. Large rectangles cover efficiently, but their incidences have to fit together without repeated intersections. Girth bounds also restrict the horizontal edge supply.

What coarse counts miss

Write (x_c, y_c) for the matched side sizes of an unsigned color c. Let m and n be the board dimensions, and let δ be zero in the even case and one in the odd case. Coverage gives an exact identity:

Coverage alone is permissive. On an 8 × 8 board, propose two unsigned colors, each with side sizes (4, 4). This is a proposed numerical profile, not an actual product structure.

The pressure inequality couples a color’s size on one side to the incidence capacity on the other:

Here e_A is the number of edges in the horizontal graph on A. The proposed profile gives e_A = 8, so either color requires 4(8 − 4) ≤ 16 − 4. Even the looser upper budget E₈ = 9 would give 16 ≤ 14. The omitted coupling changes the answer. The profile also fails the dual Fisher constraint; pressure is one direct way to expose why these coarse checks are insufficient. Remark 7.4.

Twelve required meetings. Six available pairs.

One residual case distils the obstruction into a short count. In the branch where all three link graphs have girth at least six, the general estimates reduce a six-row board to a 6 × 6 board. The horizontal graphs must both be six-cycles. Coverage and the side-size bounds then force exactly two unsigned colors with profile (3, 3), hence four signed colors.

Every vertex in A ∪ B has degree two in the middle link. Across twelve vertices, this demands twelve meetings of pairs of signed colors. Near-disjointness allows each pair to meet at most once. Four signed colors offer only six pairs. The requirements would force 12 ≤ 6.

This is a small case reached by the proof, not a substitute for the argument at arbitrary support sizes. Its value is that the reader can see exactly where the required incidence capacity runs out. Lemma 6.4.

Why the obstruction is global

To go beyond small examples, the proof combines constraints that control different features of the same geometry. Coverage fixes how much of the board must be filled. Pressure couples the two sides. Fisher bounds limit repeated pairs. High-girth graph bounds restrict the available horizontal edges.

These inequalities explain the repeated search failures across an unbounded range of support sizes.

The sharp frontier

Consequently, when the middle link has girth six, the minimum of the two horizontal girths is at most four. The examples show that this bound is sharp. They are allowed combinatorial structures on the boundary, not counterexamples produced by the excluded assembly route.

The computer-assisted part belongs to this sharper frontier. Its certificate checks the finite reduction and the remaining integer color profiles, including twenty-four surviving corner types. The main no-triple-girth theorem does not depend on that computation. Keeping these roles separate shows exactly which conclusion each part establishes. See Theorem 8.1 and Section 8.1.

Carry the answer back

The incidence inequalities are necessary conditions: every product structure satisfying the hypotheses must pass them. Proving them incompatible therefore rules out the original input. A profile that passed the inequalities would not, by itself, construct a product structure. The direction of the bridge determines what the calculation proves.

This is the practice I want to develop further: take a difficult group-theoretic question, find finite structures in which its decisive relationships can be worked with, and return a rigorous conclusion. Incidence geometry, graph algorithms, exact combinatorics, and matrix methods offer different ways to do that. An effective matrix formulation should yield a new bound, construction, or cheaper certificate.

There is a close numerical precedent. Netzer and Thom searched for a positive semidefinite Gram matrix, then turned their numerical output into an exact sum-of-squares identity in the group algebra of SL(3, Z). The resulting identity establishes a spectral gap and a new quantitative proof of property (T). A theorem connecting the certificate to the group, together with rigorous error correction, makes the numerical search decisive. [3]

Kaluba and Nowak make the additional obligation explicit. Their spectral-gap algorithm must choose a finite search subspace and turn its numerical output into a certified bound. Enlarging that subspace can improve the ideal bound while increasing both the optimization cost and the penalty attached to residual error. In one reported SL(2,19) calculation, expanding from word radius five to seven worsened the certified bound. Revised optimization constraints improve accuracy, while validated interval arithmetic makes the final check cheaper. Representation choice, numerical error, and computational cost already interact within a group-theoretic problem. [4] Related methods subsequently proved property (T) for Aut(F₅) and then for Aut(F_n) for every n ≥ 6, where F_n is the free group on n generators. [5]

Martin Nitsche subsequently resolved the remaining Aut(F₄) case through a better formulation of the certificate search. He interpreted the dual semidefinite program geometrically in terms of harmonic cocycles and used the resulting constraints and symmetries to simplify the optimization. That simplification made a larger search space computationally feasible and yielded a proof of property (T). [6]

The transferable move is to test a coarse description beyond the obvious checks, identify the relationship it omits, and make that relationship explicit. In the Taiko argument, this strengthens the constraints enough to prove an obstruction. Sparse numerical research adds a further obligation: the missing information must be detected and retained at a cost the improved computation can repay. Retaining every interaction or correction direction can erase the computational saving that made the representation useful. The research task is to discover which additions are necessary, which can remain omitted, and how to justify that choice.

The essay’s four-cycle example makes the distinction concrete. The changed matrix retains its size, sparsity pattern, diagonal, and spectrum, yet the old coarse space misses an important error direction. Including that direction in the coarse space lets the exact coarse correction remove that error. The larger research problem is to discover such diagnoses economically on unfamiliar inputs.

My essay AI research and the remaining human advantage makes the strategic claim bluntly: the old refuge in abstraction is over. I would concentrate research effort on discovering better diagnostic and repair rules for sparse numerical representations, where spectral graph algorithms, matrix analysis, and numerical analysis meet. Success means meeting the required accuracy and reliability at a lower total cost, including diagnosis, setup, solves, checking, and recovery.

Once a representation is made explicit, it becomes a tool that other researchers and agents can use. The Taiko paper illustrates the discovery of such a tool. The proposed specialist premium must therefore be earned through further discoveries on unfamiliar problems, after earlier methods, counterexamples, and code have been shared with every competitor. The essay’s testing programme asks whether specialist participation continues to produce better diagnostics, representations, or repairs under the same accuracy, reliability, and total-cost requirements.

For an entry into the method, a subgroup becomes a machine shows a group question becoming a finite graph procedure. The Refacte approach explains the common discipline: choose the working object, derive its operations, and prove the bridge.