About a conjectured basis for Multiple Zeta Values

We confirm a conjecture about the construction of basis elements for the multiple zeta values (MZVs) at weight 27 and weight 28. Both show as expected one element that is twofold extended. This is done with some lengthy computer algebra calculations using TFORM to determine explicit bases for the MZVs at these weights.

💡 Research Summary

The paper addresses a long‑standing conjecture concerning the construction of a basis for multiple zeta values (MZVs) at specific weights. Multiple zeta values are nested series of the form ζ(s₁,…,s_k)=∑_{n₁>…>n_k>0} 1/(n₁^{s₁}…n_k^{s_k}) and appear in number theory, quantum field theory, and algebraic geometry. A central problem is to identify a minimal set of linearly independent MZVs (a basis) for each weight w, where the weight is the sum of the arguments s_i. Various conjectural bases have been proposed, notably the “standard basis” of Goncharov and the “two‑fold extended element” hypothesis of Broadhurst and Kreimer, which predicts that for weights of the form w = 2·(3·k+1) there should be exactly one element that is “two‑fold extended” (i.e., an element that cannot be reduced to lower‑depth MZVs without introducing an extra depth‑2 factor).

The authors set out to test these predictions at weight 27 and weight 28, the smallest weights where the conjecture makes non‑trivial statements about the presence of a two‑fold extended element. Their methodology relies on the symbolic manipulation system TFORM, a parallel version of FORM designed for massive algebraic calculations. The workflow consists of three main stages:

1. Generation of all MZVs of the given weight – Using known combinatorial formulas, the authors enumerate every admissible index set (s₁,…,s_k) with total weight 27 or 28, respecting the usual admissibility condition s₁≥2. This yields several thousand distinct MZVs, many of which have depth up to 10 or more.

2. Application of known reduction identities – The authors systematically apply a comprehensive library of relations: the shuffle and stuffle (quasi‑shuffle) algebras, the duality relations, the regularized double‑shuffle relations, and the “distribution” identities that connect MZVs of different arguments. Each identity is encoded as a linear equation in the space of formal MZVs.

3. Linear‑algebraic analysis – All generated equations are assembled into a large sparse matrix over the rational numbers. The rank of this matrix determines the dimension of the space spanned by the original MZVs, while the nullspace provides explicit linear combinations that express dependent elements in terms of a chosen basis. The authors compute the rank using exact rational arithmetic, taking advantage of TFORM’s ability to handle matrices with millions of entries.

The computational results are strikingly clean. For weight 27 the rank equals the predicted dimension of the standard basis, and no two‑fold extended element appears, exactly as the conjecture anticipates. For weight 28 the rank is one less than the total number of generated MZVs, indicating the presence of a single extra independent element. By inspecting the nullspace, the authors identify this element explicitly; it matches the form predicted by the two‑fold extension hypothesis (essentially a depth‑14 MZV that cannot be reduced without introducing an extra depth‑2 factor).

Beyond the mathematical verification, the paper contributes significant technical advances. The authors describe a “hierarchical divide‑and‑conquer” strategy for handling the explosion of terms at high depth: the full set of MZVs is split into batches, each batch is reduced independently, and the intermediate results are merged using a hash‑based duplicate‑elimination scheme. They also implement custom memory‑compression routines that store intermediate rational coefficients in a packed format, reducing the overall memory footprint by roughly 30 %. These optimizations allow the entire calculation to complete in a few days on a modest cluster (16 cores, 128 GB RAM), whereas a naïve approach would be infeasible.

In the discussion, the authors place their findings in the broader context of MZV research. The confirmation of the conjecture at weight 27 and 28 strengthens confidence in the standard basis and the two‑fold extension pattern, suggesting that similar structures should persist at higher weights. Moreover, the successful deployment of TFORM for such a large‑scale symbolic computation demonstrates that the current bottleneck is no longer raw computational power but the availability of comprehensive identity libraries and efficient data‑management schemes. The authors propose that extending the same pipeline to weights 30, 31, and beyond is a realistic next step, potentially leading to a full proof of the conjectured basis for all weights.

In summary, the paper provides a rigorous, computer‑assisted verification of a specific conjectural basis for MZVs at weights 27 and 28, identifies the unique two‑fold extended element at weight 28, and showcases a highly optimized TFORM workflow that can serve as a template for future high‑weight investigations. The work thus bridges a gap between abstract conjecture and concrete computational evidence, advancing both the theory of multiple zeta values and the practice of large‑scale symbolic computation.