IDEMPOTENT GROWTH IN BORDER-RANK APPROXIMATIONS OF MONOMIAL ALGEBRAS
We study optimal divergence rates in approximations of structure tensors of finite-dimensional commutative monomial algebras over ℂ by as many rank-one terms as their border rank. The divergence exponent measures the smallest possible growth of the rank-one summands as the approximation error tends to zero. We relate this growth to that of primitive idempotents in smoothings by distinct points. For the algebras R_{e,k} = ℂ[x₁,…,x_e]/(x₁,…,x_e)^k we prove that the exponent is k−1 for e,k ≥ 2, in contrast with (k−1)/k in the curvilinear case. The key estimate shows that the points associated with every n-term ε-approximation, where n = dim_ℂ R_{e,k}, lie within distance O(ε) of the origin; its proof uses the Koszul relations among the generators of (x₁,…,x_e)^k. For the box algebras ℂ[x₁,…,x_e]/(x₁^{a₁},…,x_e^{a_e}) we obtain the exponent Σᵢ(aᵢ−1)/aᵢ. For monomial ideals J ⊂ ℂ[x,y] of finite colength we derive a lower bound λ(J) from their syzygies. This bound is sharp for the infinite family J = (x^{2c}, x^c y^d, y^{d+1}), c,d ≥ 2, where θ(J) = 2 − (d+1)/(2cd), but no weighted rescaling of a fixed point configuration attains the optimum; an explicit construction with two scales does. Another infinite family, (x^a, x^c y, y²) with a > 2c ≥ 4, satisfies θ(J) = 2 − 1/c > λ(J), showing that the separation of the points imposes costs beyond those detected by the global sizes of their coordinates. Supplementary material: idempotent_growth_scripts.zip (43 files), Python 3 scripts (NumPy and SciPy) with their complete outputs, and a README mapping each script to the statement it checks. The scripts compute the lower bound λ(J) exactly in rational arithmetic, count the ideals for which the union-of-boxes construction determines θ, and check numerically the explicit configurations of the upper bounds. None of the computations is used in the proofs.
Authors
- JOSÉ TORREGROSA JIMÉNEZ
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24418561
- Primary Topic
- Tensor decomposition and applications
- Type
- preprint