We know that the trace of a matrix is a linear map for all square matrices and that $\operatorname{tr}(AB)=\operatorname{tr}(BA)$ when the multiplication makes sense.

On the Wikipedia page for trace, under properties, it says that these properties characterize the trace completely in the following sense: If $f$ is a linear function on the space of square matrices satisfying $f(xy)=f(yx)$, then $f$ and $\operatorname{tr}$ are proportional. A note on the bottom of the page gives the justification, but I do not understand the logic of it. Thanks

  • $\begingroup$ The word linear means, here, $f(A+B) = f(A) + f(B),$ and, for a constant $c,$ also $f(cA) = c f(A).$ $\endgroup$ – Will Jagy Feb 2 '12 at 5:47
  • $\begingroup$ In the note of the bottom of the wiki page, $e_{ij}$ means the matrix with $(i,j)$-entry being equal to $1$, and all the other entries are zero. $\endgroup$ – Paul Feb 2 '12 at 5:50
  • $\begingroup$ @Paul, thanks for your response, but I am having trouble understanding the proof in general, not just what the symbols stand for. $\endgroup$ – Edison Feb 2 '12 at 5:59
  • $\begingroup$ As explained in Wikipedia, the key point is that the space $ \mathfrak{sl}_n(k)$ of matrices with zero trace is one-dimensional, and $M_n(k) =: \mathfrak{gl}_n(k) = \mathfrak{sl}_n(k) \oplus k$ satisfies $[ \mathfrak{gl}_n(k), \mathfrak{gl}_n(k)] = \mathfrak{sl}_n(k)$. $\endgroup$ – Watson Apr 27 '18 at 15:24

The proof in that footnote depends on the following facts that were left out.

  1. If $i\neq j$, then $[e_{ii},e_{ij}]=e_{ii}e_{ij}-e_{ij}e_{ii}=e_{ij}-0=e_{ij}$. So we can write $e_{ij}$ as a commutator $e_{ij}=xy-yx$ and thus $f(e_{ij})=f(xy)-f(yx)=0.$
  2. If $i\neq j$, then similarly $[e_{ij},e_{ji}]=e_{ii}-e_{jj}$, and we get that $$ f(e_{jj})=f(e_{jj})+0=f(e_{jj})+f([e_{ij},e_{ji}]=f(e_{jj}+(e_{ii}-e_{jj}))=f(e_{ii}). $$

Let $f$ be such a map and $E_{ij}$ be the matrix with a one in $(i,j)$-entry and zero elsewhere. We have

$$E_{ij}E_{kl}=\left\{\begin{array}{cc} 0, &\mbox{ if } \; j \neq k \\E_{il}, & \mbox{ if } \; j=k\end{array}\right.$$

Therefore, if $i\not=j, \; E_{ij}=E_{i1}E_{1j},$ then by hypothesis,




for each $1 \leq i \leq n.$

Now, let $C=\sum_{1\leq i,j \leq n}a_{i,j}E_{ij}$ be a vector in the above basis, then

$$f(C)=\sum_{1\leq i,j \leq n}a_{ij}f(E_{ij})=\sum_{i=1}^n a_{ii}f(E_{11})=f(E_{11})\sum_{i=1}^n a_{ii}=f(E_{11})tr(C).$$

  • $\begingroup$ I was searching all over the internet for an easy and understandable answer to this question. Thank you so much ! $\endgroup$ – Sauhard Sharma Apr 25 at 5:53

Yeah, we are given a function on square matrices of a fixed size, call it $f,$ with three properties, square matrices $A,B$ and constant $c.$ So: $$f(A + B ) = f(A) + f(B), $$ $$ f(AB) = f(BA), $$ $$ f(cA) = c f(A).$$

As Paul pointed out, the notation $e_{ij}$ means the matrix with a 1 at position $ij$ and 0 everywhere else.

There is some value for $f(e_{11}).$ E do not know what that is.

First,for some $i \neq 1,$ define $$ S_i = e_{i1} + e_{1i} $$ The main thing is that $$ S_i e_{11} S_i = e_{ii} $$ and $$ S_i^2 = I. $$ So $$ f(e_{ii}) = f(S_i (e_{11} S_i)) = f( (e_{11} S_i) S_i) = f( e_{11} S_i^2) = f(e_{11}). $$

Next, with $i \neq j,$ we use $$ e_{ii} e_{ij} = e_{ij} $$ while $$ e_{ij} e_{ii} = 0, $$ the matrix of all 0's.

Begin with any $B,$ $$f(0) = f(0B) = 0 f(B) = 0.$$

Now, for any $i \neq j,$ $$ f(e_{ij}) = f(e_{ii} e_{ij}) = f(e_{ij} e_{ii}) = f(0) = 0. $$

Finally, if the entries of $A$ are $A_{ij},$ we have $$ A = \sum_{i,j = 1}^n A_{ij} e_{ij}, $$ so $$ f(A) = f(\sum_{i,j = 1}^n A_{ij} e_{ij}) = \sum_{i,j = 1}^n A_{ij} f(e_{ij}) = \sum_{i=1}^n A_{ii} f(e_{ii}) = \sum_{i=1}^n A_{ii} f(e_{11}) = f(e_{11}) \sum_{i=1}^n A_{ii} = f(e_{11}) \mbox{trace} A $$


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.