# Matrix tensor construction research toolkit

Edition: 8 October 2026. Initial decision: **Defer performance product**. The product interpretation and pricing are hypotheses; independent proof and buyer validation remain open.

## Research finding

The square matrix result bounds the complex arithmetic exponent by nine fourths, with related rectangular and every-field bounds. Scope notes explicitly distinguish arithmetic complexity from bit complexity and practical crossover sizes.

## Problem and buyer

An algebraic-complexity research team needs to manipulate tensor constructions, verify parameter inequalities and explore their tradeoffs. An ML systems team instead needs fast, accurate hardware kernels at finite sizes.

## What the finding could enable

A symbolic tensor-construction toolkit could improve research reproducibility and parameter search. A production matrix library would require additional finite constructions, numerical stability and hardware evidence.

## Technical and commercial limits

Complex asymptotic arithmetic does not directly establish a faster GPU kernel, stable mixed-precision behavior or cheap extraction. Finite-field applicability varies across the companion bounds.

## Minimal architecture

Construction parser -> exact parameter arithmetic -> tensor or restriction records -> certificate export -> small finite benchmarks. Avoid claiming a hardware advantage before comparing complete end-to-end kernels.

## Existing alternatives and differentiation

Research scripts and existing linear algebra libraries serve different users. The realistic initial buyer is a research group funding tooling, with a limited market size.

## Monetization hypothesis

Hypothesis: grant-funded tooling or specialist contract work. A lucrative infrastructure company requires a demonstrated production crossover or broader reuse outside tensor research.

## Validation experiment

Validate a small construction symbolically, measure extraction overhead and coefficient growth, and compare stable finite kernels with established matrix multiplication at realistic dimensions.

## Conditions to reject or defer

Defer if only asymptotic existence is accessible or if finite constructions exceed feasible memory and precision budgets.

## Next concrete action

Read the extraction and uniformization details and list which parts can be materialized at small dimension.

## Source evidence

Repository sources are pinned to revision `fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb`. Selected scope notes and manuscript statements have been reviewed to the extent described above; these links do not represent successful kernel checks.

### Family 107

Subject: Matrix multiplication with exponent at most 9/4.

- [An Upper Bound of 9/4 for the Matrix Multiplication Exponent](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/Matrix-Multiplication-Nine-Fourths-October-2-2026/paper.pdf)
- [Complex Matrix Multiplication Below 2.258 and Rectangular Bounds](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/Complex-Matrix-Multiplication-Below-2.258-and-Rectangular-Bounds-September-24-2026/Complex-Matrix-Multiplication-Below-2.258-and-Rectangular-Bounds-September-24-2026.pdf)
- [Staggered extraction for exact matrix multiplication over every field](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/Staggered-extraction-for-exact-matrix-multiplication-over-every-field-September-24-2026/Staggered-extraction-for-exact-matrix-multiplication-over-every-field-September-24-2026.pdf)
- [Formal scope notes](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/lean/docs/107.md)