3
$\begingroup$

Every matrix $A\in M_4(\mathbb{R})$ can be represented in the form of $$A=\sum_{i=1}^{n(A)} B_i\otimes C_i$$ for $B_i,C_i\in M_2(\mathbb{R})$.

What is the least uniform upper bound $M$ for such $n(A)$? In other words, what is the least integer $M$ such that every $A$ admit such a representation with $n(A)\leq M$?

Is this least upper bound equal to the corresponding least upper bound for all matrices $A$ which are a matrix representation of quaternions $a+bi+cj+dk$?

As another question about tensor product representation: What is a sufficient condition for a $4\times 4$ matrix $A$ to be represented in the form of $A=B\otimes C -C\otimes B$?

$\endgroup$
4
$\begingroup$

Because $M_4(\mathbb R) = M_2(\mathbb R) \otimes M_2(\mathbb R)$ as vector spaces (and as algebras, but we won't use this), we can replace $M_2(\mathbb R)$ by an arbitrary $4$-dimensional vector space $V$ and $M_4(\mathbb R)$ by $V \otimes V$. We can represent elements of $V\otimes V$ conveniently as $4\times 4$ matrices, where simple tensors are rank one matrices. The question is then equivalent to asking the minimum number of rank $1$ matrices it takes to write a $4 \times 4$ matrix as a sum of rank $1$ matrices. The answer is obviously $4$.

For quaternions acting by left multiplication by quaternions in the standard basis:

$$1= \begin{pmatrix} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \end{pmatrix} = \begin{pmatrix} 1 & 0 \\0 & 1 \end{pmatrix} \otimes\begin{pmatrix} 1 & 0 \\0 & 1 \end{pmatrix}$$

$$i= \begin{pmatrix} 0 & -1 & 0 & 0 \\ 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & -1 \\ 0 & 0 & 1 & 0 \end{pmatrix} = \begin{pmatrix} 0 & -1 \\1 & 0 \end{pmatrix} \otimes\begin{pmatrix} 1 & 0 \\0 & 1 \end{pmatrix}$$

$$j= \begin{pmatrix} 0 & 0 & -1 & 0 \\ 0 & 0 & 0 & 1 \\ 1 & 0 & 0 & 0 \\ 0 & -1 & 0 & 0 \end{pmatrix} = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix} \otimes\begin{pmatrix} 0 & -1 \\1 & 0 \end{pmatrix}$$

$$k= \begin{pmatrix} 0 & 0 & 0 & -1 \\ 0 & 0 & -1 & 0 \\ 0 & 1 & 0 & 0 \\ 1 & 0 & 0 & 0 \end{pmatrix} = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix} \otimes\begin{pmatrix} 0 & -1 \\1 & 0 \end{pmatrix}$$

Because only two terms appear on the right side, the matrices clearly have rank at most $2$ at tensor product, and some obtain rank exactly $2$, so the answer is two.

For the last question, the answer is isomorphic to the analogous answer for writing an element of $V \otimes V$ as $v_1 \otimes v_2 - v_2 \otimes v_1$ for $V$ a four-dimensional vector space. The condition is given by a skew-symmetry condition (i.e. 10 linear conditions) plus a Pfaffian condition (a quadratic condition). More precisely the general such matrix can be written as

$$ \begin{pmatrix} 0 & a & -a & 0 \\ b & c & d & e \\ -b & -d & -c & -e \\ 0 & f & -f & 0 \\ \end{pmatrix}$$

such that $af-be+cd=0$

$\endgroup$
  • 2
    $\begingroup$ "this is equivalent to asking the minimum number of rank 1 matrices it takes to write a 4×4 matrix as a sum of rank 1 matrices. The answer is obviously 4." @WillSavin I do not understand this sentence. $\endgroup$ – Paul Broussous May 15 at 9:01
  • 1
    $\begingroup$ @PaulBroussous We are asking what are the minimum $n$ such that all elements of $V \otimes W$ can be written as $\sum_{i=1}^{n} v_i \otimes w_i$ for $v_i \in V, w_i \in W$, where $V$ and $W$ are four-dimensional vector spaces. We can represent elements of $V \otimes W$ as $4 \times 4$ matrices, in which case the elements of the form $v_i \otimes w_i$ are the rank $1$ matrices. So we are asking for the minimum $n$ to write any $4 \times 4$ matrix as a sum of $n$ rank one matrix. The answer is $4$ - e.g. we can take the columns or the rows. $\endgroup$ – Will Sawin May 15 at 13:32
  • 1
    $\begingroup$ OK, I understand. Here you use the canonical isomorphism $V\otimes V^* \simeq {\rm End}(V)$, which has nothing to do with the natural isomorphism $M_4 ({\mathbb R})\simeq M_2 ({\mathbb R})\otimes M_2 ({\mathbb R})$. Indeed in that last isomorphim, $A\otimes B$, $A$, $B\not= 0$, is not necessarily a rank $1$ matrix ! $\endgroup$ – Paul Broussous May 15 at 13:57
  • 1
    $\begingroup$ @PaulBroussous Yes, the structure of $M_4(\mathbb R)$ is irrelevant for this problem, so we discard it, viewing $M_4(\mathbb R)$ only as $M_2(\mathbb R)\otimes M_2(\mathbb R)$, and then the structure of $M_2(\mathbb R)$ is irrelevant as well. I had already forgotten about this after the first line and so didn't realize that bringing in $4 \times 4$ matrices again would be potentially confusing. $\endgroup$ – Will Sawin May 15 at 15:39
  • 1
    $\begingroup$ @ZachTeitler It's skew symmetric when written in the standard matrix notation for elements of $V \otimes V$, $V$ a four-dimensional vector space, and the Pfaffian is the standard Pfaffian. But if we specialize to $V= M_2(\mathbb R)$ we have another way to write elements of the same vector space as matrices, using $M_2(\mathbb R) \otimes M_2(\mathbb R) = M_4(\mathbb R)$. I have written it in that way, so it no longer appears skew-symmetric. Concretely this just means we permute the entries. It might be fun to work out what this permutation does and why this formula is right - it was for me. $\endgroup$ – Will Sawin May 16 at 0:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.