Let $\textbf A$ denote the space of symmetric $(n\times n)$ matrices over the field $\mathbb K$, and $\textbf B$ the space of skew-symmetric $(n\times n)$ matrices over the field $\mathbb K$. Then $\dim (\textbf A)=n(n+1)/2$ and $\dim (\textbf B)=n(n-1)/2$.

Short question: is there any short explanation (maybe with combinatorics) why this statement is true?

EDIT: $\dim$ refers to linear spaces.

  • 1
    $\begingroup$ Do you mean symmetric (not normal) in the title? And do you mean the $\dim$ of linear spaces of such matrices, not the $\dim$ of the matrices, right? $\endgroup$
    – enzotib
    Aug 23 '12 at 9:59
  • $\begingroup$ I did edit it - thanks for the reminder! $\endgroup$ Aug 23 '12 at 10:02
  • $\begingroup$ You did not edit it correctly; $\mathbf A$ still refers to just one matrix, not a subspace. I will edit it for you. $\endgroup$ Aug 23 '12 at 12:38
  • $\begingroup$ And you should say that $\mathbb K$ is not of characteristic $2$, or otherwise symmetric and anti-symmetric matrices are the same thing and your equations cannot both be true. $\endgroup$ Aug 23 '12 at 12:40

All square matrices of a given size $n$ constitute a linear space of dimension $n^2$, because to every matrix element corresponds a member of the canonical base, i.e. the set of matrices having a single $1$ and all other elements $0$.

The skew-symmetric matrices have arbitrary elements on one side with respect to the diagonal, and those elements determine the other triangle of the matrix. So they are in number of $(n^2-n)/2=n(n-1)/2$, ($-n$ to remove the diagonal).

For the symmetric matrices the reasoning is the same, but we have to add back the elements on the diagonal: $(n^2-n)/2+n=(n^2+n)/2=n(n+1)/2$.


Here is my two cents:

\begin{eqnarray} M_{n \times n}(\mathbb{R}) & \text{has form} & \begin{pmatrix} *&*&*&*&\cdots \\ *&*&*&*& \\ *&*&*&*& \\ *&*&*&*& \\ \vdots&&&&\ddots \end{pmatrix} \hspace{.5cm} \text{with $n^2$ elements}\\ \\ \\ Skew_{n \times n}(\mathbb{R}) & \text{has form} & \begin{pmatrix} 0&*'&*'&*'&\cdots \\ *&0&*'&*'& \\ *&*&0&*'& \\ *&*&*&0& \\ \vdots&&&&\ddots \end{pmatrix} \end{eqnarray} For this bottom formation the (*)s and (*')s are just operationally inverted forms of the same number, so the array here only takes $\frac{(n^2 - n)}{2}$ elements as an argument to describe it. This appears to be an array geometry question really... I suppose if what I'm saying is true, then I conclude that because $\dim(Skew_{n \times n}(\mathbb{R}) + Sym_{n \times n}(\mathbb{R})) = \dim(M_{n \times n}(\mathbb{R}))$ and $\dim(Skew_{n \times n}(\mathbb{R}))=\frac{n^2-n}{2}$ then we have that \begin{eqnarray} \frac{n^2-n}{2}+\dim(Sym_{n \times n}(\mathbb{R})))=n^2 \end{eqnarray} or \begin{eqnarray} \dim(Sym_{n \times n}(\mathbb{R})))=\frac{n^2+n}{2}. \end{eqnarray}

  • $\begingroup$ $Skew_{n\times n}(\mathbb{R})=(Sym_{n\times n}(\mathbb{R}))^{\perp}?$ $\endgroup$
    – eraldcoil
    Aug 18 at 5:50

This is a bit late, but I figured I would chime in since the other answers don't really give the combinatorial intuition asked for in the original question.

In order to specify a skew-symmetric matrix, we need to choose a number $A_{ij}$ for each set $\{i,j\}$ of distinct indices. How many such sets are there? Combinatorics tells us that there are $\left( \begin{array}{@{}c@{}} n \\ 2 \end{array} \right) = \frac{n(n-1)}{2}$ such sets.

Similarly, in order to specify a symmetric matrix, we need to choose a number $A_{ij}$ for each set $\{i,j\}$ of (not necessarily distinct) indices. We can encode such sets by adding a special symbol $n+1$ that indicates a repeated index. Thus, we need to choose a number $A_{ij}$ for each set $\{i,j\}$ of distinct symbols, where now a symbol is either an index ($1 , \ldots , n$), or our special symbol $n+1$ that is used for a repeated index. Combinatorics tells us that there are $\left( \begin{array}{@{}c@{}} n+1 \\ 2 \end{array} \right) = \frac{n(n+1)}{2}$ such sets.


The dimension of symmetric matrices is $\frac{n(n+1)}2$ because they have one basis as the matrices $\{M_{ij}\}_{n \ge i \ge j \ge 1}$, having $1$ at the $(i,j)$ and $(j,i)$ positions and $0$ elsewhere. For skew symmetric matrices, the corresponding basis is $\{M_{ij}\}_{n \ge i > j \ge 1}$ with $1$ at the $(i,j)$ position, $-1$ at the $(j,i)$ position, and $0$ elsewhere.

Note that the diagonal elements of skew symmetric matrices are $0$, hence their dimension is $n$ less than the dimension of normal symmetric matrices.

  • $\begingroup$ But this is no explanation why the symmetric matrices have the specified $\dim$. $\endgroup$ Aug 23 '12 at 10:03
  • $\begingroup$ Do you mean $n^2$? $\endgroup$
    – enzotib
    Aug 23 '12 at 10:04
  • $\begingroup$ Yeah, I didn't notice earlier that he asked for proofs of the dimensions too, and have edited. $\endgroup$ Aug 23 '12 at 10:07
  • $\begingroup$ @RijulSaini: Thanks, but enzotib's answer seems to be easier to understand! $\endgroup$ Aug 23 '12 at 10:13

There have been many very good answers, so this is going to be another perspective of counting. We tackle the case of skew-symmetric first, which yields the condition that for any matrix $A$ with real entries, $a_{ij} = -a_{ij}$, so this means one side of the matrix is completely determined by the other as shown.

$$\begin{matrix} \begin{pmatrix} 0 &*' \\ *& 0 \\ \end{pmatrix} & * \end{matrix}$$

$$\begin{matrix} \begin{pmatrix} 0 &*' & *' \\ *& 0 & *' \\ * & * & 0 \\ \end{pmatrix} & \begin{matrix} *& \\ * & * \end{matrix} \end{matrix}$$

$$\begin{matrix} \begin{pmatrix} 0 &*' & *' &*'\\ *& 0 & *' &*'\\ * & * & 0 &*'\\ * & * &* & 0 \end{pmatrix} & \begin{matrix} * & & \\ *&* & \\ * & * &* \end{matrix} \end{matrix}$$

$$\begin{matrix} \begin{pmatrix} 0&*'&*'&*'&\cdots \\ *&0&*'&*'& \\ *&*&0&*'& \\ *&*&*&0& \\ \vdots&&&&\ddots \end{pmatrix} & \begin{matrix} *& \\ *&*& \\ *&*&*& \\ \vdots&&&&\ddots \end{matrix} \end{matrix}$$

Notice that the triangle keeps getting larger and larger and at each stage and is incremented by $+1$. Since we have $n-1$ entries to consider we add up the number of entries, that is $$1+ 2 + \dots + (n-1).$$

So by Gauss, we have $$\frac{(n-1)(n-1+1)}{2} = \frac{(n-1)n}{2}$$ degree of freedom and that is the dimension we seek.

For the symmetric case we need to add back the $n$ objects in diagonal zeroed out, so $$\frac{(n-1)n}{2} + n = \frac{n^2 + n}{2}$$


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.