Exploiting the Structure in Tensor Decompositions for Matrix Multiplication

Isaac Wood; Jakob Moosbauer; Manuel Kauers

arxiv: 2602.11041 · v3 · pith:UJAGZ47Enew · submitted 2026-02-11 · 💻 cs.SC

Exploiting the Structure in Tensor Decompositions for Matrix Multiplication

Manuel Kauers , Jakob Moosbauer , Isaac Wood This is my paper

Pith reviewed 2026-05-22 11:30 UTC · model grok-4.3

classification 💻 cs.SC

keywords tensor decompositionsmatrix multiplicationfast algorithmsalgebraic complexityexponent reductionstructural features6x6 matrices

0 comments

The pith

A new algorithm exploits special structural features in tensor decompositions to achieve a lower exponent for matrix multiplication than the rank bound indicates.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a method that takes advantage of particular structural properties in certain tensor decompositions used for matrix multiplication. These properties can be identified and used algorithmically to reduce the asymptotic number of multiplications below what the decomposition rank alone would require. For 6 by 6 matrices the approach improves the exponent from the recent value of 2.8075 down to 2.8019. The improvement is obtained while preserving a reasonable leading coefficient, so the resulting algorithm remains practical in its constant factors.

Core claim

By identifying and algorithmically exploiting special structural features present in chosen tensor decompositions for matrix multiplication, it is possible to obtain algorithms whose exponent is strictly smaller than the one implied by the rank of the decomposition. For the concrete case of 6 by 6 matrix multiplication this produces an exponent of 2.8019 rather than 2.8075.

What carries the argument

Special structural features inside tensor decompositions that can be algorithmically exploited to derive a lower exponent than the rank bound.

If this is right

For 6 by 6 matrix multiplication the exponent can be lowered to 2.8019 while retaining a usable leading coefficient.
Existing tensor decompositions can yield better algorithms once their structural features are used, without the need to search for lower-rank decompositions.
The same exploitation technique may apply to decompositions for other matrix sizes whenever comparable structures are present.
Improvements arise from algorithmic use of structure rather than from changes to the underlying decomposition itself.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Future searches for fast matrix multiplication could prioritize finding decompositions that possess exploitable structure rather than minimal rank alone.
The approach may extend to other problems that rely on tensor decompositions in algebraic complexity, such as polynomial multiplication or bilinear forms.
Combining this structural exploitation with existing optimization heuristics could produce further incremental gains for larger matrix sizes.

Load-bearing premise

The selected tensor decompositions contain identifiable structural features that can be exploited to produce a strictly lower exponent than the rank without invalidating the decomposition or inflating the leading coefficient beyond a reasonable size.

What would settle it

A direct count of arithmetic operations in the resulting 6 by 6 algorithm that shows it requires at least as many multiplications as the original rank bound, or a proof that any exploitation of the claimed structures either breaks correctness or forces an impractically large leading coefficient.

Figures

Figures reproduced from arXiv: 2602.11041 by Isaac Wood, Jakob Moosbauer, Manuel Kauers.

read the original abstract

We present a new algorithm for fast matrix multiplication using tensor decompositions which have special features. Thanks to these features we obtain exponents lower than what the rank of the tensor decomposition suggests. In particular for $6\times 6$ matrix multiplication we reduce the exponent of the recent algorithm by Moosbauer and Poole from $2.8075$ to $2.8019$, while retaining a reasonable leading coefficient.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper reports a modest improvement on the 6x6 matrix multiplication exponent by exploiting structural features in tensor decompositions to beat the rank bound.

read the letter

The main thing here is a small but concrete drop in the exponent for 6x6 matrix multiplication. They improve on the recent Moosbauer and Poole result by using special features in the tensor decomposition to reach 2.8019 instead of 2.8075 while keeping the leading coefficient reasonable. This is what stands out: a way to get below the usual rank bound without invalidating the underlying map. The work is a direct extension of the prior algorithm rather than a new framework, but it does deliver a fresh numerical value that was not in the literature before. They earn credit for stating the gain clearly and for paying attention to the leading term, which matters if anyone wants to use these exponents in practice. The approach is incremental and focused on one fixed size, which is fine for this corner of algebraic complexity. The soft spot is the thin description of the actual exploitation step. The abstract claims the lower exponent but does not lay out the counting argument or verification, so the result depends on whether the full text shows a clean, reproducible way to turn the structural features into fewer operations. If that part is solid and the decomposition remains correct, the central claim holds. A minor additional question is how far the trick travels beyond this particular decomposition. This paper is for researchers who track small-case matrix multiplication exponents and tensor methods. A reader following the recent literature on 6x6 or similar sizes would find the new number useful. It deserves peer review because it adds a checkable data point on a standard problem with enough context to evaluate. I would send it to referees.

Referee Report

0 major / 2 minor

Summary. The manuscript presents a new algorithm for fast matrix multiplication that exploits special structural features in chosen tensor decompositions. These features are used to obtain matrix-multiplication exponents strictly lower than the value implied by the decomposition rank. The concrete claim is an improvement for 6×6 matrix multiplication, reducing the exponent from 2.8075 (Moosbauer–Poole) to 2.8019 while retaining a reasonable leading coefficient.

Significance. If the structural exploitation is shown to preserve correctness of the underlying bilinear map and to deliver the stated asymptotic improvement without an implicit increase in effective rank or leading term, the result would constitute a modest but concrete advance in the search for faster matrix-multiplication algorithms. The approach of going beyond the rank bound by algorithmic exploitation of tensor structure is potentially reusable and therefore of interest to the community.

minor comments (2)

The abstract states a numerical improvement but does not indicate the verification method, error analysis, or how the new exponent is derived from the structural features. Adding a brief outline of the verification procedure (e.g., explicit rank computation or machine-checked certificate) would strengthen the presentation.
It would be helpful to include a short comparison table showing the previous and new exponents together with the associated leading coefficients for the 6×6 case and, if available, for a few other small sizes.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment of our manuscript, the accurate summary of our contribution, and the recommendation for minor revision. We appreciate the recognition that exploiting structural features beyond the rank bound may be of broader interest.

read point-by-point responses

Referee: If the structural exploitation is shown to preserve correctness of the underlying bilinear map and to deliver the stated asymptotic improvement without an implicit increase in effective rank or leading term, the result would constitute a modest but concrete advance in the search for faster matrix-multiplication algorithms. The approach of going beyond the rank bound by algorithmic exploitation of tensor structure is potentially reusable and therefore of interest to the community.

Authors: We have explicitly constructed the bilinear map in the manuscript and verified that the structural features (detailed in Section 3) yield a valid algorithm whose multiplication count produces the claimed exponent of 2.8019 for 6×6 matrices. The derivation follows the standard relation between the number of multiplications and the resulting exponent; no effective rank inflation occurs because the structure is used only to schedule the existing multiplications more efficiently. We will add a short clarifying paragraph in the revision to make the correctness argument more self-contained. revision: partial

Circularity Check

0 steps flagged

Minor self-citation of prior algorithm; derivation remains self-contained

full rationale

The paper claims an improvement in matrix multiplication exponent by exploiting special structural features in chosen tensor decompositions, yielding a concrete reduction from 2.8075 to 2.8019 for the 6×6 case while retaining a reasonable leading coefficient. The abstract references the 'recent algorithm by Moosbauer and Poole,' and since Moosbauer is a co-author this is a self-citation. However, the central claim rests on a new algorithmic exploitation of structure rather than re-using or fitting the prior result as an input; no equations, parameter fitting, self-definitional steps, or load-bearing uniqueness theorems appear in the abstract or description. The derivation is therefore self-contained against external benchmarks with only a non-load-bearing self-citation, warranting a low score.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract mentions no explicit free parameters, axioms, or invented entities; the central claim rests on the existence and exploitability of unspecified structural features in tensor decompositions.

pith-pipeline@v0.9.0 · 5584 in / 1187 out tokens · 34365 ms · 2026-05-22T11:30:29.155707+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We present a new algorithm for fast matrix multiplication using tensor decompositions which have special features... ⟨6,6,6⟩ ≤ 137 ⊙ ⟨1⟩ ⊕ 8 ⊙ ⟨1,1,2⟩ ... exponent 2.8016
IndisputableMonolith/Foundation/DimensionForcing.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Theorem 8 ... lim ω_k = ω_0 where n^ω0 = 3 Σ s_i (n_i m_i p_i)^ω0/3

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages · 1 internal anchor

[1]

Josh Alman, Ran Duan, Virginia Vassilevska Williams, Yinzhan Xu, Zixuan Xu, and Renfei Zhou. 2024. More Asymmetry Yields Faster Matrix Multiplication. arXiv:2404.16349 [cs.DS] https://arxiv.org/abs/2404.16349

work page arXiv 2024
[2]

Gal Beniamini and Oded Schwartz. 2019. Faster Matrix Multiplication via Sparse Decomposition. InThe 31st ACM Symposium on Parallelism in Algorithms and Ar- chitectures(Phoenix, AZ, USA)(SPAA ’19). Association for Computing Machinery, The results in this paper are flawed. Please do not use them. We are working on a revised version. Exploiting the Structure...

work page doi:10.1145/3323165.3323188 2019
[3]

Dario Bini, Milvio Capovani, Francesco Romani, and Grazia Lotti. 1979. O(n2.7799) complexity for n×n approximate matrix multiplication.Inform. Process. Lett.8 (06 1979). https://doi.org/10.1016/0020-0190(79)90113-3

work page doi:10.1016/0020-0190(79)90113-3 1979
[4]

H. Cohn, R. Kleinberg, B. Szegedy, and C. Umans. 2005. Group-theoretic algo- rithms for matrix multiplication. In46th Annual IEEE Symposium on Foundations of Computer Science (FOCS’05). 379–388. https://doi.org/10.1109/SFCS.2005.39

work page doi:10.1109/sfcs.2005.39 2005
[5]

Cohn and C

H. Cohn and C. Umans. 2003. A group-theoretic approach to fast matrix multi- plication. In44th Annual IEEE Symposium on Foundations of Computer Science,

work page 2003
[6]

https://doi.org/10.1109/SFCS.2003.1238217

Proceedings.438–449. https://doi.org/10.1109/SFCS.2003.1238217

work page doi:10.1109/sfcs.2003.1238217 2003
[7]

Don Coppersmith and Shmuel Winograd. 1990. Matrix multiplication via arith- metic progressions.Journal of Symbolic Computation9, 3 (1990), 251–280. https://doi.org/10.1016/S0747-7171(08)80013-2 Computational algebraic com- plexity editorial

work page doi:10.1016/s0747-7171(08)80013-2 1990
[8]

Hans F de Groote. 1978. On varieties of optimal algorithms for the computation of bilinear mappings I. the isotropy group of a bilinear mapping.Theoretical Computer Science7, 1 (1978), 1–24. https://doi.org/10.1016/0304-3975(78)90038-5

work page doi:10.1016/0304-3975(78)90038-5 1978
[9]

The (generaliz ed) Post correspondence problem with lists consisting of two words is decidable

Hans F. de Groote. 1978. On varieties of optimal algorithms for the computa- tion of bilinear mappings II. optimal algorithms for 2×2-matrix multiplication. Theoretical Computer Science7, 2 (1978), 127–148. https://doi.org/10.1016/0304- 3975(78)90045-2

work page doi:10.1016/0304- 1978
[10]

Jean-Guillaume Dumas, Clément Pernet, and Alexandre Sedoglavic. 2025. A non-commutative algorithm for multiplying 4x4 matrices using 48 non-complex multiplications. arXiv:2506.13242 [cs.SC] https://arxiv.org/abs/2506.13242

work page arXiv 2025
[11]

Alhussein Fawzi, Matej Balog, Aja Huang, Thomas Hubert, Bernardino Romera- Paredes, Mohammadamin Barekatain, Alexander Novikov, Francisco J. R. Ruiz, Julian Schrittwieser, Grzegorz Swirszcz, David Silver, Demis Hassabis, and Pushmeet Kohli. 2022. Discovering faster matrix multiplication algorithms with reinforcement learning.Nature610, 7930 (Oct 2022), 47...

work page doi:10.1038/s41586-022-05172-4 2022
[12]

2003.Modern Computer Algebra (2 ed.)

Joachim Von Zur Gathen and Jurgen Gerhard. 2003.Modern Computer Algebra (2 ed.). Cambridge University Press, USA

work page 2003
[13]

IE Kaporin. 2024. Semi-analytical solution of Brent equations.Doklady Mathe- matics518, 1 (2024), 29–34. https://journals.eco-vector.com/2686-9543/article/ view/647987

work page 2024
[14]

Elaye Karstadt and Oded Schwartz. 2017. Matrix Multiplication, a Little Faster. In Proceedings of the 29th ACM Symposium on Parallelism in Algorithms and Architec- tures(Washington, DC, USA)(SPAA ’17). Association for Computing Machinery, New York, NY, USA, 101–110. https://doi.org/10.1145/3087556.3087579

work page doi:10.1145/3087556.3087579 2017
[15]

Manuel Kauers and Jakob Moosbauer. 2022. The FBHHRBNRSSSHK-Algorithm for Multiplication inZ5×5 2 is still not the end of the story. arXiv:2210.04045 [cs.SC] https://arxiv.org/abs/2210.04045

work page arXiv 2022
[16]

Manuel Kauers and Jakob Moosbauer. 2023. Flip Graphs for Matrix Multiplication. InProceedings of the 2023 International Symposium on Symbolic and Algebraic Computation(Tromsø, Norway)(ISSAC ’23). Association for Computing Machin- ery, New York, NY, USA, 381–388. https://doi.org/10.1145/3597066.3597120

work page doi:10.1145/3597066.3597120 2023
[17]

Manuel Kauers and Jakob Moosbauer. 2023. Some New Non-Commutative Matrix Multiplication Algorithms of Size (𝑛, 𝑚, 6). arXiv:2306.00882 [cs.SC] https://arxiv.org/abs/2306.00882

work page arXiv 2023
[18]

Manuel Kauers and Isaac Wood. 2025. Consequences of the Moosbauer-Poole Algorithms. arXiv:2505.05896 [cs.SC] https://arxiv.org/abs/2505.05896

work page arXiv 2025
[19]

Manuel Kauers and Isaac Wood. 2025. Exploring the Meta Flip Graph for Matrix Multiplication. arXiv:2510.19787 [cs.SC] https://arxiv.org/abs/2510.19787

work page arXiv 2025
[20]

Axel Kemper. 2025. From 𝐹2 to Z Solutions of Brent Equations. https://github. com/a1880/matrix-multiplication Preprint. Available at https://github.com/a1880/ matrix-multiplication

work page 2025
[21]

Laderman

Julian D. Laderman. 1976. A Noncommutative Algorithm for Multiplying3 × 3 Matrices Using 23 Multiplications.Bull. Amer. Math. Soc.82, 1 (1976), 126–128. https://doi.org/10.1090/S0002-9904-1976-13988-2

work page doi:10.1090/s0002-9904-1976-13988-2 1976
[22]

Jakob Moosbauer and Michael Poole. 2025. Flip Graphs with Symmetry and New Matrix Multiplication Schemes. InProceedings of the 2025 International Symposium on Symbolic and Algebraic Computation (ISSAC ’25). Association for Computing Machinery, New York, NY, USA, 233–239. https://doi.org/10.1145/ 3747199.3747566

work page arXiv 2025
[23]

2025.The Number of the Beast: Reducing Additions in Fast Matrix Multiplication Algorithms for Dimensions up to

Erik Mårtensson and Paul Stankovski Wagner. 2025.The Number of the Beast: Reducing Additions in Fast Matrix Multiplication Algorithms for Dimensions up to

work page 2025
[24]

https://doi.org/10.1137/1.9781611978759.4

47–60. https://doi.org/10.1137/1.9781611978759.4

work page doi:10.1137/1.9781611978759.4
[26]

Alexander Novikov, Ngân V ˜u, Marvin Eisenberger, Emilien Dupont, Po-Sen Huang, Adam Zsolt Wagner, Sergey Shirobokov, Borislav Kozlovskii, Francisco J. R. Ruiz, Abbas Mehrabian, M. Pawan Kumar, Abigail See, Swarat Chaudhuri, George Holland, Alex Davies, Sebastian Nowozin, Pushmeet Kohli, and Matej Balog. 2025. AlphaEvolve: A coding agent for scientific an...

work page internal anchor Pith review Pith/arXiv arXiv 2025
[27]

1984.How to multiply matrices faster

Victor Pan. 1984.How to multiply matrices faster. Springer-Verlag, Berlin, Hei- delberg

work page 1984
[28]

A. I. Perminov. 2025. Fast Matrix Multiplication via Ternary Meta Flip Graphs. arXiv:2511.20317 [cs.SC] https://arxiv.org/abs/2511.20317

work page arXiv 2025
[29]

Robert L. Probert. 1976. On the Additive Complexity of Matrix Multiplication. SIAM J. Comput.5, 2 (1976), 187–203. https://doi.org/10.1137/0205016

work page doi:10.1137/0205016 1976
[30]

Schönhage

A. Schönhage. 1981. Partial and Total Matrix Multiplication.SIAM J. Comput.10, 3 (Aug. 1981), 434–455. https://doi.org/10.1137/0210032

work page doi:10.1137/0210032 1981
[31]

Oded Schwartz, Sivan Toledo, Noa Vaknin, and Gal Wiernik. 2024. Alternative Basis Matrix Multiplication is Fast and Stable. In2024 IEEE International Parallel and Distributed Processing Symposium (IPDPS). 38–51. https://doi.org/10.1109/ IPDPS57955.2024.00013

work page arXiv 2024
[32]

Oded Schwartz and Eyal Zwecher. 2025. Towards Faster Feasible Matrix Multi- plication by Trilinear Aggregation. arXiv:2508.01748 [cs.DS] https://arxiv.org/ abs/2508.01748

work page arXiv 2025
[33]

2025.Collection of fast matrix multiplication algorithms

Alexandre Sedoglavic. 2025.Collection of fast matrix multiplication algorithms. Université de Lille. https://fmm.univ-lille.fr/ Accessed: 2026-02-03

work page 2025
[34]

Alexey Smirnov. 2013. The bilinear complexity and practical algorithms for matrix multiplication.Computational Mathematics and Mathematical Physics53 (12 2013). https://doi.org/10.1134/S0965542513120129

work page doi:10.1134/s0965542513120129 2013
[35]

STRASSEN

V. STRASSEN. 1969. Gaussian Elimination is not Optimal.Numer. Math.13 (1969), 354–356. http://eudml.org/doc/131927

work page 1969
[36]

Strassen

V. Strassen. 1986. The asymptotic spectrum of tensors and the exponent of matrix multiplication. In27th Annual Symposium on Foundations of Computer Science (sfcs 1986). 49–54. https://doi.org/10.1109/SFCS.1986.52

work page doi:10.1109/sfcs.1986.52 1986

[1] [1]

Josh Alman, Ran Duan, Virginia Vassilevska Williams, Yinzhan Xu, Zixuan Xu, and Renfei Zhou. 2024. More Asymmetry Yields Faster Matrix Multiplication. arXiv:2404.16349 [cs.DS] https://arxiv.org/abs/2404.16349

work page arXiv 2024

[2] [2]

Gal Beniamini and Oded Schwartz. 2019. Faster Matrix Multiplication via Sparse Decomposition. InThe 31st ACM Symposium on Parallelism in Algorithms and Ar- chitectures(Phoenix, AZ, USA)(SPAA ’19). Association for Computing Machinery, The results in this paper are flawed. Please do not use them. We are working on a revised version. Exploiting the Structure...

work page doi:10.1145/3323165.3323188 2019

[3] [3]

Dario Bini, Milvio Capovani, Francesco Romani, and Grazia Lotti. 1979. O(n2.7799) complexity for n×n approximate matrix multiplication.Inform. Process. Lett.8 (06 1979). https://doi.org/10.1016/0020-0190(79)90113-3

work page doi:10.1016/0020-0190(79)90113-3 1979

[4] [4]

H. Cohn, R. Kleinberg, B. Szegedy, and C. Umans. 2005. Group-theoretic algo- rithms for matrix multiplication. In46th Annual IEEE Symposium on Foundations of Computer Science (FOCS’05). 379–388. https://doi.org/10.1109/SFCS.2005.39

work page doi:10.1109/sfcs.2005.39 2005

[5] [5]

Cohn and C

H. Cohn and C. Umans. 2003. A group-theoretic approach to fast matrix multi- plication. In44th Annual IEEE Symposium on Foundations of Computer Science,

work page 2003

[6] [6]

https://doi.org/10.1109/SFCS.2003.1238217

Proceedings.438–449. https://doi.org/10.1109/SFCS.2003.1238217

work page doi:10.1109/sfcs.2003.1238217 2003

[7] [7]

Don Coppersmith and Shmuel Winograd. 1990. Matrix multiplication via arith- metic progressions.Journal of Symbolic Computation9, 3 (1990), 251–280. https://doi.org/10.1016/S0747-7171(08)80013-2 Computational algebraic com- plexity editorial

work page doi:10.1016/s0747-7171(08)80013-2 1990

[8] [8]

Hans F de Groote. 1978. On varieties of optimal algorithms for the computation of bilinear mappings I. the isotropy group of a bilinear mapping.Theoretical Computer Science7, 1 (1978), 1–24. https://doi.org/10.1016/0304-3975(78)90038-5

work page doi:10.1016/0304-3975(78)90038-5 1978

[9] [9]

The (generaliz ed) Post correspondence problem with lists consisting of two words is decidable

Hans F. de Groote. 1978. On varieties of optimal algorithms for the computa- tion of bilinear mappings II. optimal algorithms for 2×2-matrix multiplication. Theoretical Computer Science7, 2 (1978), 127–148. https://doi.org/10.1016/0304- 3975(78)90045-2

work page doi:10.1016/0304- 1978

[10] [10]

Jean-Guillaume Dumas, Clément Pernet, and Alexandre Sedoglavic. 2025. A non-commutative algorithm for multiplying 4x4 matrices using 48 non-complex multiplications. arXiv:2506.13242 [cs.SC] https://arxiv.org/abs/2506.13242

work page arXiv 2025

[11] [11]

Alhussein Fawzi, Matej Balog, Aja Huang, Thomas Hubert, Bernardino Romera- Paredes, Mohammadamin Barekatain, Alexander Novikov, Francisco J. R. Ruiz, Julian Schrittwieser, Grzegorz Swirszcz, David Silver, Demis Hassabis, and Pushmeet Kohli. 2022. Discovering faster matrix multiplication algorithms with reinforcement learning.Nature610, 7930 (Oct 2022), 47...

work page doi:10.1038/s41586-022-05172-4 2022

[12] [12]

2003.Modern Computer Algebra (2 ed.)

Joachim Von Zur Gathen and Jurgen Gerhard. 2003.Modern Computer Algebra (2 ed.). Cambridge University Press, USA

work page 2003

[13] [13]

IE Kaporin. 2024. Semi-analytical solution of Brent equations.Doklady Mathe- matics518, 1 (2024), 29–34. https://journals.eco-vector.com/2686-9543/article/ view/647987

work page 2024

[14] [14]

Elaye Karstadt and Oded Schwartz. 2017. Matrix Multiplication, a Little Faster. In Proceedings of the 29th ACM Symposium on Parallelism in Algorithms and Architec- tures(Washington, DC, USA)(SPAA ’17). Association for Computing Machinery, New York, NY, USA, 101–110. https://doi.org/10.1145/3087556.3087579

work page doi:10.1145/3087556.3087579 2017

[15] [15]

Manuel Kauers and Jakob Moosbauer. 2022. The FBHHRBNRSSSHK-Algorithm for Multiplication inZ5×5 2 is still not the end of the story. arXiv:2210.04045 [cs.SC] https://arxiv.org/abs/2210.04045

work page arXiv 2022

[16] [16]

Manuel Kauers and Jakob Moosbauer. 2023. Flip Graphs for Matrix Multiplication. InProceedings of the 2023 International Symposium on Symbolic and Algebraic Computation(Tromsø, Norway)(ISSAC ’23). Association for Computing Machin- ery, New York, NY, USA, 381–388. https://doi.org/10.1145/3597066.3597120

work page doi:10.1145/3597066.3597120 2023

[17] [17]

Manuel Kauers and Jakob Moosbauer. 2023. Some New Non-Commutative Matrix Multiplication Algorithms of Size (𝑛, 𝑚, 6). arXiv:2306.00882 [cs.SC] https://arxiv.org/abs/2306.00882

work page arXiv 2023

[18] [18]

Manuel Kauers and Isaac Wood. 2025. Consequences of the Moosbauer-Poole Algorithms. arXiv:2505.05896 [cs.SC] https://arxiv.org/abs/2505.05896

work page arXiv 2025

[19] [19]

Manuel Kauers and Isaac Wood. 2025. Exploring the Meta Flip Graph for Matrix Multiplication. arXiv:2510.19787 [cs.SC] https://arxiv.org/abs/2510.19787

work page arXiv 2025

[20] [20]

Axel Kemper. 2025. From 𝐹2 to Z Solutions of Brent Equations. https://github. com/a1880/matrix-multiplication Preprint. Available at https://github.com/a1880/ matrix-multiplication

work page 2025

[21] [21]

Laderman

Julian D. Laderman. 1976. A Noncommutative Algorithm for Multiplying3 × 3 Matrices Using 23 Multiplications.Bull. Amer. Math. Soc.82, 1 (1976), 126–128. https://doi.org/10.1090/S0002-9904-1976-13988-2

work page doi:10.1090/s0002-9904-1976-13988-2 1976

[22] [22]

Jakob Moosbauer and Michael Poole. 2025. Flip Graphs with Symmetry and New Matrix Multiplication Schemes. InProceedings of the 2025 International Symposium on Symbolic and Algebraic Computation (ISSAC ’25). Association for Computing Machinery, New York, NY, USA, 233–239. https://doi.org/10.1145/ 3747199.3747566

work page arXiv 2025

[23] [23]

2025.The Number of the Beast: Reducing Additions in Fast Matrix Multiplication Algorithms for Dimensions up to

Erik Mårtensson and Paul Stankovski Wagner. 2025.The Number of the Beast: Reducing Additions in Fast Matrix Multiplication Algorithms for Dimensions up to

work page 2025

[24] [24]

https://doi.org/10.1137/1.9781611978759.4

47–60. https://doi.org/10.1137/1.9781611978759.4

work page doi:10.1137/1.9781611978759.4

[25] [26]

Alexander Novikov, Ngân V ˜u, Marvin Eisenberger, Emilien Dupont, Po-Sen Huang, Adam Zsolt Wagner, Sergey Shirobokov, Borislav Kozlovskii, Francisco J. R. Ruiz, Abbas Mehrabian, M. Pawan Kumar, Abigail See, Swarat Chaudhuri, George Holland, Alex Davies, Sebastian Nowozin, Pushmeet Kohli, and Matej Balog. 2025. AlphaEvolve: A coding agent for scientific an...

work page internal anchor Pith review Pith/arXiv arXiv 2025

[26] [27]

1984.How to multiply matrices faster

Victor Pan. 1984.How to multiply matrices faster. Springer-Verlag, Berlin, Hei- delberg

work page 1984

[27] [28]

A. I. Perminov. 2025. Fast Matrix Multiplication via Ternary Meta Flip Graphs. arXiv:2511.20317 [cs.SC] https://arxiv.org/abs/2511.20317

work page arXiv 2025

[28] [29]

Robert L. Probert. 1976. On the Additive Complexity of Matrix Multiplication. SIAM J. Comput.5, 2 (1976), 187–203. https://doi.org/10.1137/0205016

work page doi:10.1137/0205016 1976

[29] [30]

Schönhage

A. Schönhage. 1981. Partial and Total Matrix Multiplication.SIAM J. Comput.10, 3 (Aug. 1981), 434–455. https://doi.org/10.1137/0210032

work page doi:10.1137/0210032 1981

[30] [31]

Oded Schwartz, Sivan Toledo, Noa Vaknin, and Gal Wiernik. 2024. Alternative Basis Matrix Multiplication is Fast and Stable. In2024 IEEE International Parallel and Distributed Processing Symposium (IPDPS). 38–51. https://doi.org/10.1109/ IPDPS57955.2024.00013

work page arXiv 2024

[31] [32]

Oded Schwartz and Eyal Zwecher. 2025. Towards Faster Feasible Matrix Multi- plication by Trilinear Aggregation. arXiv:2508.01748 [cs.DS] https://arxiv.org/ abs/2508.01748

work page arXiv 2025

[32] [33]

2025.Collection of fast matrix multiplication algorithms

Alexandre Sedoglavic. 2025.Collection of fast matrix multiplication algorithms. Université de Lille. https://fmm.univ-lille.fr/ Accessed: 2026-02-03

work page 2025

[33] [34]

Alexey Smirnov. 2013. The bilinear complexity and practical algorithms for matrix multiplication.Computational Mathematics and Mathematical Physics53 (12 2013). https://doi.org/10.1134/S0965542513120129

work page doi:10.1134/s0965542513120129 2013

[34] [35]

STRASSEN

V. STRASSEN. 1969. Gaussian Elimination is not Optimal.Numer. Math.13 (1969), 354–356. http://eudml.org/doc/131927

work page 1969

[35] [36]

Strassen

V. Strassen. 1986. The asymptotic spectrum of tensors and the exponent of matrix multiplication. In27th Annual Symposium on Foundations of Computer Science (sfcs 1986). 49–54. https://doi.org/10.1109/SFCS.1986.52

work page doi:10.1109/sfcs.1986.52 1986