On normal forms for the similarity classes of matrices and pairs of matrices

Klaus Bongartz · 2025

Read On normal forms for the similarity classes of matrices and pairs of matrices by Klaus Bongartz (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-10926.txt.

Introduction

Klaus Bongartz answers two questions left open since 1998 in Arnold's problem book, in this 2025 paper on matrix normal forms. Over any field, the similarity classes of square matrices admit a tidy system of representatives, built from finitely many flat affine pieces. Pairs of matrices refuse the same tidiness: for size above one, over an algebraically closed field of characteristic zero, no such finite flat system exists. The split answer draws a clean line between one matrix and two.

Catalog record

CREATORKlaus Bongartz
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-10926

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On normal forms for the similarity classes of matrices and pairs of matrices Klaus Bongartz Abstract We answer two questions posed 1998 in the book ‘Arnold’s problems’. First, over any field k k there is a representative system for the similarity classes of n × n n\times n -matrices which is a finite disjoint union of affine subspaces. And second, for n ≥ 2 n\geq 2 an analogous statement fails for pairs of n × n n\times n - matrices over any algebraically closed field of characteristic 0. The positive answer rests on a slight modification of the rational normal form whereas the negative answer requires more work. On page 127 of [ 1 ] Kontsevich asks whether there is in the space of complex matrices of a fixed size a representative system for the similarity classes which is the union of affine subspaces. In the comments he writes on page 613 that this has an easy positive answer for one matrix which he does not remember but that the analogous question for pairs of matrices is also interesting and open. In this note we describe first an easy solution for one matrix given already 1989 in [ 2 ] , then we give the negative answer for pairs of 2\times 2 -matrices using a result of [ 3 ] from 1996 and finally we extend the answer to all n\geq 3 using appropriate vector bundles as in [ 4 ] . We fix some notations and we recall basic facts about the rational normal form. Let k be an arbitrary field. The set of all n\times n -matrices with coefficients in k is denoted by k^{n\times n} and the canonical base vectors in k^{n} by e_{i} . Definition 1 . A rational normal form R\in k^{n\times n} is a block-diagonal matrix \left[\begin{array}[]{cccc}B(P_{1})&0&...&0\\ 0&B(P_{2})&...&0\\ ...&...&...&...\\ 0&0&...&B(P_{r})\end{array}\right] with companion-matrices B(P_{1}),B(P_{2}),\ldots,B(P_{r}) on the main diagonal, such that all polynomials P_{i} are normed of degree \geq 1 with coefficients in k and such that P_{i+1} divides P_{i} for each i with 1\leq i\leq r-1 . We write R=R(P_{1},P_{2},\ldots,P_{r}) for such a matrix. There is the following important classical result ( see [ 5 ] for a short proof ). Theorem 1 . Each matrix A\in k^{n\times n} is similar over k to exactly one rational normal form R(A)=R(P_{1},P_{2},\ldots,P_{r}) . Here R(A) as well as an invertible matrix T with T^{-1}AT=R are obtained from the coefficients of A by finitely many rational operations the number of which grows polynomially with n . Definition 2 . Let p=(p_{1},p_{2},\ldots,p_{r}) be a partition of n , i.e. a sequence of natural numbers p_{1}\geq p_{2}\geq\ldots\geq p_{r}\geq 1 adding up to n . We denote by R(p) the set of rational normal forms as above with blocks of sizes p_{1},p_{2},\ldots,p_{r} . Thus the disjoint union of the various R(p) is a representative system for the similarity classes in k^{n\times n} . Each R(p) is a closed subset of k^{n\times n} but not an affine subspace as soon as two different p_{i} ’s occur. To remedy this we replace the companion-matrices by ‘generalized companion matrices’ as defined below. Definition 3 . Let q=(q_{1},q_{2},\ldots,q_{s}) be an s-tuple of natural numbers q_{j}\geq 1 . Given normed polynomials Q_{j} of degree q_{j} the generalized companion-matrix C=C(Q_{1},Q_{2},\ldots,Q_{s}) is the quadratic matrix of size m=\sum q_{i} with the companion-matrices B(Q_{j}) as diagonal blocks ordered in the natural way. Outside of these blocks all entries are 0 except that c_{i+1,i}=1 holds for all i<m . By C(q) we denote the set of all these generalized companion matrices. As an illustration we write down all matrices in C(q) for q=(2,1,2) . Here a,b,c,d,e are arbitrary scalars. \left[\begin{array}[]{cccccccccccc}0&e&0&0&0\\ 1&d&0&0&0\\ 0&1&c&0&0\\ 0&0&1&0&b\\ 0&0&0&1&a\\ \end{array}\right] . The following observation is crucial: Lemma 1 . The matrix C=C(Q_{1},Q_{2},\dots,Q_{s}) from above is similar to the companion-matrix of P=Q_{1}\cdot Q_{2}\cdots Q_{s} . Furthermore C(q) contains exactly one nilpotent matrix N and the differences A-N with A\in C(q) form a linear subspace of k^{m\times m} of dimension m . Proof: We make k^{m} to a k[X] -module M via X\cdot v=Cv . It is enough to prove that e_{1} generates M and that P annihilates e_{1} . Namely then the map 1\mapsto e_{1} induces a surjection k[X]/(P)\rightarrow M between spaces of the same dimension and so an isomorphism. Thus B(P) is similar to C . We proceed by induction on s . The case s=1 is obvious. To go from s-1 to s suppose Q_{1}=a_{0}+a_{1}X+\ldots a_{q_{1}-1}X^{q_{1}-1}+X^{q_{1}} . We have Xe_{1}=e_{2},Xe_{2}=e_{3},\ldots,Xe_{q_{1}-1}=e_{q_{1}} and Xe_{q_{1}}=e_{q_{1}+1}-(a_{0}e_{1}+a_{1}e_{2}+\ldots a_{q_{1}-1}e_{q_{1}}) . Thus we get Q_{1}e_{1}=e_{q_{1}+1} . The action of X on the submodule U of M spanned by the e_{j} with j\geq q_{1}+1 is given by a generalized companion-matrix with s-1 blocks, whence by induction U is generated by e_{q_{1}+1} which is annihilated by Q_{2}\cdot Q_{3}\cdots Q_{s} . So e_{1} generates M and it is annihilated by P . Now B(P) is nilpotent iff P=X^{m} . The rest is obvious. Our aim is to prove: Theorem 2 . Let p=(p_{1},p_{2},\ldots,p_{r}) be a partition of n . Then there is an affine subspace A(p) of dimension p_{1} in k^{n\times n} such that the map A\mapsto R(A) induces a bijection between A(p) and R(p) . Proof: First we take a closer look at the partition. Assume that there are s different p_{i} ’s occurring in p . In case s=1 we set A(p)=R(p) which is already an affine subspace. For s\geq 2 we number the ‘jump indices’ i<r where p_{i}\neq p_{i+1} in ascending order as j_{1}<j_{2}<\ldots<j_{s-1} and we add j(s)=r,j_{0}=0 . Thus we obtain for 1\leq i\leq s-1 non-zero natural numbers q_{i}=p_{j_{i}}-p_{j_{i}+1} to which we add q_{s}=p_{r} . For any s -tuple (Q_{1},Q_{2},\ldots,Q_{s}) of normed polynomials with degQ_{j}=q_{j} we consider a block-diagonal matrix A(Q_{1},Q_{2},\ldots Q_{s}) of the shape \left[\begin{array}[]{cccc}D_{1}&0&...&0\\ 0&D_{2}&...&0\\ ...&...&...&...\\ 0&0&...&D_{r}\end{array}\right]. Here we set D_{i}=C(Q_{k},Q_{k+1},\ldots,Q_{s}) for j_{k-1}<i\leq j_{k} . This matrix depends only on the tuple of polynomials and on the partition p . The set of all these matrices is the wanted affine subspace. Namely each block D_{i} of A(Q_{1},Q_{2},\ldots,Q_{s}) is just a ‘right lower’ sub-block of D_{1} the size of which depends only on p and the possible D_{1} ’s form an affine subspace. As an illustration we write down all matrices in D(p) for the partition (5,3,2,2) of 12 . Here a,b,c,d,e are arbitrary scalars. \left[\begin{array}[]{cccccccccccc}0&e&0&0&0&0&0&0&0&0&0&0\\ 1&d&0&0&0&0&0&0&0&0&0&0\\ 0&1&c&0&0&0&0&0&0&0&0&0\\ 0&0&1&0&b&0&0&0&0&0&0&0\\ 0&0&0&1&a&0&0&0&0&0&0&0\\ 0&0&0&0&0&c&0&0&0&0&0&0\\ 0&0&0&0&0&1&0&b&0&0&0&0\\ 0&0&0&0&0&0&1&a&0&0&0&0\\ 0&0&0&0&0&0&0&0&0&b&0&0\\ 0&0&0&0&0&0&0&0&1&a&0&0\\ 0&0&0&0&0&0&0&0&0&0&0&b\\ 0&0&0&0&0&0&0&0&0&0&1&a\\ \end{array}\right] To prove the theorem we verify that A\mapsto R(A) is the wanted bijection. The product P_{i}=Q_{k}Q_{k+1}\cdots Q_{s} is normed of degree p_{i} and D_{i} is similar to B(P_{i}) by lemma 1. Obviously P_{i} divides P_{i-1} for all i>1 and so the rational normal form of the constructed matrix is R(P_{1},P_{2},\ldots,P_{r}) and it belongs to R(p) . Reversely, given a rational normal form R=R(P_{1},P_{2},\ldots,P_{r}) one gets the s-1 quotients Q_{i}=P_{j_{i}}/P_{j_{i}+1} having degree q_{i} and one adds Q_{s}=P_{r} . Then A(Q_{1},Q_{2},\ldots Q_{s}) belongs to A(p) and its rational normal form is R . In [ 6 ] Daugulis has given another construction of the affine subspaces starting from the Jordan normal forms so that his proof is more complicated. I thank him for bringing Kontsevichs problem to my attention. In the following all varieties are quasi-projective and defined over an algebraically closed field

Read the complete text