Strassen’s 2 × 2 matrix multiplication algorithm:

A conceptual perspective
Christian Ikenmeyer1 and Vladimir Lysikov2
1
Max Planck Institute for Software Systems, Saarland Informatics Campus
2
Department of Computer Science, Saarland University
cikenmey@mpi-sws.org vlysikov@cs.uni-saarland.de

February 14, 2019

arXiv:1708.08083v2 [cs.DS] 13 Feb 2019

Abstract
The main purpose of this paper is pedagogical.
Despite its importance, all proofs of the correctness of Strassen’s famous 1969 algorithm to multiply two
2 × 2 matrices with only seven multiplications involve some basis-dependent calculations such as explicitly
multiplying specific 2×2 matrices, expanding expressions to cancel terms with opposing signs, or expanding
tensors over the standard basis. This makes the proof nontrivial to memorize and many presentations of
the proof avoid showing all the details and leave a significant amount of verifications to the reader.
In this note we give a short, self-contained, basis-independent proof of the existence of Strassen’s
algorithm that avoids these types of calculations. We achieve this by focusing on symmetries and algebraic
properties.
Our proof can be seen as a coordinate-free version of the construction of Clausen from 1988, com-
bined with recent work on the geometry of Strassen’s algorithm by Chiantini, Ikenmeyer, Landsberg, and
Ottaviani from 2016.

1 Introduction
The discovery of Strassen’s matrix multiplication algorithm [28] was a breakthrough result in computational
linear algebra. The study of fast (subcubic) matrix multiplication algorithms initiated by this discovery has
become an important area of research (see [3] for a survey and [21] for the currently best upper bound on the
complexity of matrix multiplication). Fast matrix multiplication has countless applications as a subroutine
in algorithms for a wide variety of problems, see e.g. [7, §16] for numerous applications in computational
linear algebra. In practice, algorithms more sophisticated than Strassen’s are almost never implemented,
but Strassen’s algorithm is used for multiplication of large matrices (see [12, 25, 19] on practical fast matrix
multiplication).
The core of Strassen’s result is an algorithm for multiplying 2 × 2 matrices with only 7 multiplications
instead of 8. It is a bilinear algorithm, which means that it arises from a decomposition of the form
7
X
XY = uk (X)vk (Y )Wk , (⋆)
k=1

where uk and vk are cleverly chosen linear forms on the space of 2 × 2 matrices and Wk are seven explicit 2 × 2
matrices. Because of this structure it can be applied to block matrices, and its recursive application results in
an algorithm for the multiplication of two n × n matrices using O(nlog2 7 ) arithmetic operations (see [7, §15.2]
or [3] for details).
Because of the great importance of Strassen’s algorithm, our goal is to understand it on a deep level. In
Strassen’s original paper, the linear forms uk , vk , and the matrices Wk are given, but the verification of the
correctness of the algorithm is left to the reader. Unfortunately, such a description does not yield many further
immediate insights.

1
 Shortly after Strassen’s paper, Gastinel [14] published a proof of the existence of decomposition (⋆) using
simple algebraic transformations that is much easier to follow and verify. Many other papers provide alternative
descriptions of Strassen’s algorithm or proofs of its existence. Brent [4] and Paterson [26] present the algorithm
in a graphical form using 4 × 4 diagrams indicating which elements of the two matrices are used. A more
formal version of these diagrams are matrices of linear forms, which are used, for example, by Fiduccia [13]
(essentially the same proof appears in [29]), Brockett and Dobkin [5] and Lafon [20]. Makarov [22] gives a proof
that uses ideas of Karatsuba’s algorithm for the efficient multiplication of polynomials. Büchi and Clausen [6]
connect the existence of Strassen’s algorithm to the existence of special bases of the space of 2 × 2 matrices
in which the multiplication table has a specific structure (their results are more general and apply not only to
matrix multiplication). Alexeyev [1] describes several algorithms for matrix multiplication as embeddings of
the matrix algebra into a 7-dimensional nonassociative algebra with a special properties.
Verification of these proofs usually requires simple, but lengthy computations: expansion of explicit decom-
positions in some basis, multiplication of several matrices or following chains of algebraic transformations in
which careful attention to details is required. To obtain a more conceptual proof of the existence of Strassen’s
algorithm, we do not focus on the explicit algorithm, but on the algebraic properties of the 2 × 2 matrices,
their transformations and symmetries of Strassen’s algorithm. It is well-known that the decomposition (⋆) is
not unique. Given one decomposition, we can obtain another one by applying the identity

XY = A−1 (AXB −1 )(BY C −1 ) C

 

and using the original decomposition for the product in the square brackets. Alternatively, we can talk about
2 × 2 matrices as linear maps between 2-dimensional vector spaces. Any choice of bases in these vector spaces
gives a new bilinear algorithm. De Groote [18] proved that the algorithm with seven multiplications is unique
up to these transformations (this result is also announced without a proof in [23], see also [24]). Thus, Strassen’s
algorithm is unique in this sense and there should be a coordinate-free description of this algorithm which does
not use explicit matrices. One such description is given in [10] and the proof of its correctness uses the fact that
matrix multiplication is the unique (up to scale) bilinear map invariant under the transformations described
above. This is a nontrivial fact which requires representation theory to prove. Moreover, the verification of
the correctness in [10] is left to the reader.
Symmetries of Strassen’s algorithm are also useful for its understanding. Clausen [11] gives a description
of Strassen’s algorithm in terms of special bases, as in [6], and notices that the elements of these bases form
orbits under the action of the symmetric group S3 on the space of 2 × 2 matrices defined via conjugation
with specific matrices, i. e., Strassen’s algorithm is invariant under this action. Clausen’s construction is also
describled in [7, Ch.1]. Grochow and Moore [16, 17] generalize Clausen’s construction to n × n matrices using
other finite group orbits. Another symmetry is only apparent in the trilinear representation of the algorithm:
the decompositions (⋆) are in one-to-one correspondence with decompositions of the trilinear form tr(XY Z)
of the form
X7
tr(XY Z) = uk (X)vk (Y )wk (Z)
k=1

where uk , vk and wk are linear forms. The decomposition corresponding to Strassen’s algorithm is then invari-
ant under the cyclic permutation of matrices X, Y, Z. This symmetry is exploited in the proof of Chatelin [9],
which uses properties of polynomials invariant under this symmetry. He also notices the importance of a matrix
which is related to the S3 symmetry discussed above. The symmetries of Strassen’s algorithm are explored in
detail in [8, 10]. Several earlier publications note their importance [15, 27]. The paper [2] explores symmetries
of algorithms for 3 × 3 matrix multiplication.
In this paper we provide a proof of Strassen’s result which is
• coordinate-free — we do not use explicit matrices, which allows us to focus on the algebraic properties
required to prove the correctness of the algorithm. We avoid all tedious explicit calculations, in particular
any expansions of expressions and any verification of explicit sign cancellations. Our proof can be seen
as a coordinate-free version of Clausen’s construction.
• elementary — our proof uses only simple facts from basic linear algebra and does not require knowledge
of representation theory. This is also why we do not use tensor language. Proofs from [10] and [17] are
based on more complicated mathematics and may offer other insights.
Formally, the result that we prove is the following.

2
Theorem 1 (Strassen [28]). Fix any field F. There exist fourteen linear forms u1 , . . . , u7 , v1 , . . . , v7 : F2×2 → F
and seven matrices W1 , . . . , W7 ∈ F2×2 such that for all pairs of 2 × 2 matrices X and Y the product satisfies
7
X
XY = uk (X)vk (Y )Wk . (⋆)
k=1

Acknowledgements. The authors thank Alin Bostan, Joshua Grochow and anonymous referees for com-
ments and pointers to the literature.

2 Preliminaries from linear algebra

The trace tr(A) of a square matrix A is the sum of its diagonal entries. If tr(A) is zero, then the matrix
A is called traceless. Taking the trace of a product of (rectangular) matrices is invariant under cyclic shifts:
tr(A1 A2 · · · An ) = tr(A2 · · · An A1 ). As a consequence, the trace of a matrix is invariant under conjugations:
tr(B −1 AB) = tr(ABB −1 ) = tr(A). Another implication is that if u is a column vector and v T is a row vector,
then v T u = tr(v T u) = tr(uv T ).
The characteristic polynomial of a 2 × 2 matrix A is λ2 − tr(A)λ + det(A). The Cayley—Hamilton theorem
says that substituting A for λ yields the zero matrix.

3 Rotational symmetry
In this section we collect some standard facts about rotation matrices. We think of the 2 × 2 matrix D
as a rotation of the plane by 120◦, but to make our approach work over every field we use a more algebraic
definition for D.
Let D have determinant 1 and trace −1, that is, D has characteristic polynomial λ2 + λ + 1. We assume
that D is not a multiple
 of theidentity id (this is implicitly satisfied if the characteristicisnot3).
 For
 example,

0 −1 1 0 −1
we could choose D = , the matrix that cyclically permutes the three vectors , , .
1 −1 0 1 −1

Claim 2. The matrix D has the following properties: D3 = id and id +D + D−1 = 0 and tr(D−1 ) = −1.
Proof. The characteristic polynomial of D is λ2 + λ + 1. By the Cayley—Hamilton theorem D2 + D + id = 0.
Multiplying by D we obtain D + D2 + D3 = 0 = id +D + D2 and hence D3 = id. Therefore D−1 = D2 and
thus id +D + D−1 = 0. This implies tr(D−1 ) = − tr(id) − tr(D) = −1.
For every column vector u define u⊥ as the row vector satisfying conditions u⊥ u = 0 and u⊥ Du = 1. If u
is not an eigenvector of D, then u and Du are linearly independent, so u⊥ is uniquely defined. If, on the other
hand, u is an eigenvector of D, the two conditions are inconsistent and u⊥ is undefined.
We
  fix a vector u that is not an eigenvector
 of D and define u⊥ as above. In our example we could choose
1 0 −1
u= , which is not an eigenvector of .
0 1 −1
A first simple observation relates u⊥ and (Du)⊥ :
Claim 3. u⊥ D−1 = (Du)⊥ .
Proof. We need to verify the two defining properties for (Du)⊥ . We have (u⊥ D−1 )(Du) = u⊥ u = 0 and
(u⊥ D−1 )D(Du) = u⊥ Du = 1 as required.
The following observation complements the fact that u⊥ Du = 1.
Claim 4. u⊥ D−1 u = −1.
Proof. Using Claim 2 we have id +D + D−1 = 0 and thus

u⊥ u + u⊥ Du + u⊥ D−1 u = 0.

Since u⊥ u = 0 and u⊥ Du = 1, the claim follows.

3
4 Seven multiplications suffice
In this section we apply our structural insights from Section 3 to prove Theorem 1. We set M := uu⊥ .
Clearly tr(M ) = u⊥ u = 0 and we obtain the following identities that can be used to simplify products of M ,
D, and D−1 :
Claim 5. M 2 = 0 and M DM = M and M D−1 M = −M .
Proof.

M2 = (uu⊥ )(uu⊥ ) = u(u⊥ u)u⊥ = 0.

M DM = (uu⊥ )D(uu⊥ ) = u(u⊥ Du)u⊥ = uu⊥ = M.
M D−1 M = (uu⊥ )D−1 (uu⊥ ) = u(u⊥ D−1 u)u⊥ = −uu⊥ = −M,

where in the last line we used Claim 4.

By Claim 2, conjugation with D is a map of order 3 on the vector space of all 2 × 2 matrices, i.e., for any
matrix A there is a triple of conjugates A 7→ D−1 AD 7→ DAD−1 7→ A. Moreover, if A is traceless, then so are
its conjugates.
Claim 6. The matrices M , D−1 M D, and DM D−1 form a basis of the vector space of traceless matrices.
Proof. Since M is traceless, its conjugates are also traceless. Hence it is enough to prove that M , D−1 M D
and DM D−1 are linearly independent.
Since u is not an eigenvector of D, the vectors u and Du are linearly independent and thus form a basis
of the space of column vectors. The row vectors u⊥ and u⊥ D−1 = (Du)⊥ (Claim 3) are orthogonal to u and
Du, respectively. Therefore they form a basis of the space of row vectors. Thus, the four matrices

u · u⊥ = M, u · u⊥ D−1 = M D−1 , Du · u⊥ = DM, Du · u⊥ D−1 = DM D−1

form a basis of the space of 2 × 2 matrices. The matrices M and DM D−1 are contained in this basis. Adding
up all four matrices, we get (id +D)M (id +D−1 ), which can be simplified to (−D−1 )M (−D) = D−1 M D using
Claim 2. Therefore the matrices M , DM D−1 , D−1 M D are linearly independent.
Since D and D−1 have trace −1 6= 0 (Claim 2), adding D or D−1 to the basis in Claim 6 yields two bases
for the full space of 2 × 2 matrices: {D, M, D−1 M D, DM D−1 } and {D−1 , M, D−1 M D, DM D−1 }.
Using the properties D2 = D−1 , D−2 = D and M 2 = 0 from Claim 2 and Claim 5, we can write down the
multiplication table with respect to these two bases. We further simplify it using the identities M DM = M
and M D−1 M = −M from Claim 5.
D−1 M D−1 M D DM D−1
D id DM MD D−1 M D−1

M M D−1 0 M D−1 M D M DM D−1

= −M D = M D−1

D−1 M D D−1 M D−1 M DM 0 D−1 M D−1 M D−1

= D−1 M = −D−1 M D−1

DM D−1 DM D DM D−1 M DM DM D
0
= −DM = DM D

Proof of Theorem 1. Notice that in the body of the table only (scalar multiples of) 7 matrices are used, and
the entries are aligned in such a way that two occurrences of the same matrix are either in the same row or
in the same column. At this point we are done proving Theorem 1, because the existence of such a pattern
gives a simple way to construct a matrix multiplication algorithm as follows. To multiply matrices A and B,
represent them in the bases {D, M, D−1 M D, DM D−1 } and {D−1 , M, D−1 M D, DM D−1 }, respectively:

X = x1 D + x2 M + x3 D−1 M D + x4 DM D−1
(4.1)
Y = y1 D−1 + y2 M + y3 D−1 M D + y4 DM D−1

4
 Note that the xi are linear forms in the entries of X and the yj are linear forms in the entries of Y . We
expand the product XY and group together summands according to the table:

XY = x1 × y1 × id
+ x2 × (y1 + y4 ) × M D−1
+ x3 × (y1 + y2 ) × D−1 M
+ x4 × (y1 + y3 ) × DM D
+ (x1 − x4 ) × y2 × DM
+ (x1 − x2 ) × y3 × MD
+ (x1 − x3 ) × y4 × D−1 M D−1
↑ ↑ ↑
uk (X) vk (Y ) Wk
This finishes the proof.
Remark. Taking the trace in (4.1) and using the fact that M and its conjugates are traceless, we see that
tr(X) = x1 tr(D) = −x1 , and tr(Y ) = −y1 . Thus the first of the 7 summands is tr(X) tr(Y ) id.

References
[1] Valery B. Alekseyev. Maximal extensions with simple multiplication for the algebra of matrices of the
second order. Discrete Math. Appl., 7(1):89–102, 1996. doi:10.1515/dma.1997.7.1.89.
[2] Grey Ballard, Christian Ikenmeyer, Joseph M. Landsberg, and Nick Ryder. The geometry of rank
decompositions of matrix multiplication II: 3 × 3 matrices. Preprint arXiv:1801.00843, arXiv, 2018.
arXiv:1801.00843.
[3] Markus Bläser. Fast matrix multiplication. Theory Comput. Grad. Surv., 5:1–60, 2013.
doi:10.4086/toc.gs.2013.005.
[4] Richard P. Brent. Algorithms for matrix multiplication. Tech. Report STAN-CS-70-157, Stanford Uni-
versity, Department of Computer Science, 1970. doi:10.21236/ad0705509.

[5] Roger W. Brockett and David Dobkin. On the optimal evaluation of a set of bilinear forms. In Proc. 5th
ACM STOC, pages 88–95, 1973. doi:10.1145/800125.804039.
[6] Werner Büchi and Michael Clausen. On a class of primary algebras of minimal rank. Linear Algebra
Appl., 69:249–268, 1985. doi:10.1016/0024-3795(85)90080-1.
[7] Peter Bürgisser, Michael Clausen, and M. Amin Shokrollahi. Algebraic Complexity The-
ory, volume 315 of Grundlehren der mathematischen Wissenschaften. Springer, Berlin, 1997.
doi:10.1007/978-3-662-03338-8.
[8] Vladimir P. Burichenko. On symmetries of the Strassen algorithm. Preprint arXiv:1408.6273, arXiv, 2014.
arXiv:1408.6273.
[9] Philippe Chatelin. On transformations of algorithms to multiply 2 × 2 matrices. Inf. Process. Lett.,
22(1):1–5, 1986. doi:10.1016/0020-0190(86)90033-5.

[10] Luca Chiantini, Christian Ikenmeyer, Joseph M. Landsberg, and Giorgio Ottaviani. The geometry of rank
decompositions of matrix multiplication I: 2 × 2 matrices. Exp. Math., Advance online publication, 2017.
doi:10.1080/10586458.2017.1403981.
[11] Michael Clausen. Beiträge zum Entwurf schneller Spektraltransformationen. Habilitationsschrift, Univer-
sität Karlsruhe, 1988.
[12] Jean-Guillaume Dumas and Victor Y. Pan. Fast matrix multiplication and symbolic computation. Preprint
arXiv:1612.05766, arXiv, 2016. arXiv:1612.05766.
[13] Charles M. Fiduccia. On obtaining upper bounds on the complexity of matrix multiplication. In Com-
plexity of Computer Computations, pages 31–40, 1972. doi:10.1007/978-1-4684-2001-2_4.

5
[14] Noël Gastinel. Sur le calcul des produits de matrices. Numer. Math., 17(3):222–229, 1971.
doi:10.1007/BF01436378.
[15] Ann Q. Gates and Vladik Kreinovich. Strassen’s algorithm made (somewhat) more natural: A pedagogical
remark. Bull. EATCS, 73:142–145, 2001. URL: https://digitalcommons.utep.edu/cs_techrep/502/.
[16] Joshua A. Grochow and Cristopher Moore. Matrix multiplication algorithms from group orbits. Preprint
arXiv:1612.01527, arXiv, 2016. arXiv:1612.01527.
[17] Joshua A. Grochow and Cristopher Moore. Designing Strassen’s algorithm. Preprint arXiv:1708.09398,
arXiv, 2017. arXiv:1708.09398.
[18] Hans F. de Groote. On varieties of optimal algorithms for the computation of bilinear mappings
II: Optimal algorithms for 2 × 2-matrix multiplication. Theor. Comp. Sci., 7(2):127–184, 1978.
doi:10.1016/0304-3975(78)90045-2.
[19] Jianyu Huang, Leslie Rice, Devin A. Matthews, and Robert A. van de Geijn. Generating fami-
lies of practical fast matrix multiplication algorithms. In Proc. IPDPS 2017, pages 656–667, 2017.
doi:10.1109/IPDPS.2017.56.
[20] Jean-Claude Lafon. Optimum computation of p bilinear forms. Linear Algebra Appl., 10(3):225–240,
1975. doi:10.1016/0024-3795(75)90071-3.
[21] François Le Gall. Powers of tensors and fast matrix multiplication. In Proc. ISSAC 2014, pages 296–303,
2014. doi:10.1145/2608628.2608664.
[22] Oleg M. Makarov. The connection between two multiplication algorithms. USSR Comput. Math. Math.
Phys., 15(1):218–223, 1975. doi:10.1016/0041-5553(75)90149-4.
[23] Victor Y. Pan. О схемах вычисления произведений матриц и обратной матрицы [On algorithms for
matrix multiplication and inversion]. Усп. мат. наук, 27(5(167)):249–250, 1972. Translation available
in [24]. URL: http://mi.mathnet.ru/umn5125.
[24] Victor Y. Pan. Better late than never: Filling a void in the history of fast matrix multiplication and
tensor decompositions. Preprint arXiv:1411.1972, arXiv, 2014. arXiv:1411.1972.

[25] Victor Y. Pan. Fast matrix multiplication and its algebraic neighbourhood. Sb. Math., 208(11):1661–1704,
2017. doi:10.1070/SM8833.
[26] Mike Paterson. Complexity of product and closure algorithms for ma-
trices. In Proc. ICM 1974, volume 2, pages 483–489, 1974. URL:
https://www.mathunion.org/fileadmin/ICM/Proceedings/ICM1974.2/ICM1974.2.ocr.pdf#page=491
[cited 2018-02-03].
[27] Mike Paterson. Strassen symmetries. Presentation at Leslie Valiant’s 60th
birthday celebration, 30.05.2009, Bethesda, Maryland, USA, 2009. URL:
https://www.cis.upenn.edu/~mkearns/valiant/paterson.ppt [cited 2018-02-03].
[28] Volker Strassen. Gaussian elimination is not optimal. Numer. Math., 13(4):354–356, 1969.
doi:10.1007/BF02165411.
[29] Gideon Yuval. A simple proof of Strassen’s result. Inf. Process. Lett., 7(6):285–286, 1978.
doi:10.1016/0020-0190(78)90018-2.