Let $P$ be an invertible matrix with n rows and n columns. Let $L$ be the following vector space: The elements in $L$ are matrices $X$ with n rows and n columns, such that $tr(PX)=0$. Find $dimL$. Here's what I did, please tell me if this is correct: Let $(p_{i,j})$ be the elements of P. $P$ is invertible and thus $P \neq 0$. That means that there exists $i,j$ such that $p_{i,j} \neq 0$. Let $X$ be a matrices with n rows and columns. Name the elements of X, X=$(x_{i,j})$. Now, $tr(PX)=0$ is equivilent to:
$$\sum_{k=1}^{n}\sum_{m=1}^{n}(p_{k,m} \cdot x_{m,k})=0$$

When we open this sum, every element of $P$ and of $X$ appears once and only once.
Without loss of generality $p_{1,1} \neq 0$ (We mentioned that at least one element of $P$ is not zero).
Then the following equation is equivilent to:
$$x_{1,1} = */p_{1,1}$$ where $*$ is some sum of the $x_{i,j}$'s and $p_{i,j}$'s, but $x_{1,1}$ is not in that sum!
Thus, we can choose the elements of $X$ except for $x_{1,1}$, and then take $x_{1,1}=*/p_{1,1}$.
So $L$ is isomorphic to $R^{n^2-1}$ and thus $dimL=n^2-1$. Correct? I am doubtful because I almost didn't use the fact that $P$ is invertible, the saame proof would work if I wast told $P \neq 0$.

  • $\begingroup$ Indeed, your proof is not valid, for precisely the reason that you mention, but even were that not the case, your result would not follow: you've shown tha tthe dimension of $L$ is at most $n^2 - 1$. Why can there not be further dependencies between the $x_{i,j}$? $\endgroup$ – user3482749 Jan 12 at 17:11
  • $\begingroup$ @user3482749 But I mentioned that L is isomorphic to a vector space with dimension n^2-1, we can define $T: R^{n^2-1} \rightarrow L$: for every $v=(v_1,...,v_{n^2-1})$, start to fill the matrix X like that: skip x(1,1) and then go row by row and fill the elements of X to be the elements in v (x(1,2)=v1 x(1,3)=v2 and so on, go row by row) and than take x(1,1)=*/p(1,1).we get an element in L, this is obviously one to one, and is onto because if we have X=(x_(i,j)) in L, then it holds that x(1,1)=*/p(1,1). we take v=(x_(1,2), x(1,3), ..., x(n,n)) [go row by row and skip x(1,1)] Why is this wrong? $\endgroup$ – Omer Jan 12 at 17:23

You have some good ideas, but the computation is flawed.

Better see this as the composition of two linear maps: \begin{align} &f\colon M_n(\mathbb{R})\to M_n(\mathbb{R}) && f(X)=PX \\[4px] &g\colon M_n(\mathbb{R})\to \mathbb{R} && g(X)=\operatorname{Tr}(X) \end{align} where $M_n(\mathbb{R})$ is the space of $n\times n$ matrices.

What does the assumption that $P$ is invertible say about $f$? What's the dimension of the kernel of $g$? Use the rank-nullity theorem for this. Finally, what's the kernel of $g\circ f$?

An extension of the result is obvious. Using the rank-nullity theorem on $g\circ f$, we see that if it is surjective, then its kernel has dimension $n^2-1$. In order to show that this holds with only assuming $P\ne0$, you need to prove that $g\circ f$ is surjective, that is, there exists $X$ with $\operatorname{Tr}(PX)\ne0$ (not obvious, but true).

  • $\begingroup$ thanks for the comment, I'll try that later, but I want to know where am I wrong. please take a look at my comment to user3482749, I added a few more details to show that L is isomorphic to R^(n^2-1) (I defined in that comment an onto and one to one linear map), isn't it correct? $\endgroup$ – Omer Jan 12 at 17:26
  • $\begingroup$ @Omer The problem is the “handwaving” you do to conclude. The intuition is correct, but the argument is not sufficient: as you say, you need to use that $P$ is invertible, which you don't. $\endgroup$ – egreg Jan 12 at 17:30
  • $\begingroup$ P is invertible thus f is onto. f is onto and thus dimIm(gf)=dimIm(g)=dimR=1. thus, dimker(gf)=n^2-1 (by the rank nullity theorem). correct? $\endgroup$ – Omer Jan 12 at 17:36
  • $\begingroup$ @Omer Almost: $f$ is an isomorphism, so the dimension of $\ker(g\circ f)$ equals the dimension of $\ker g$, which is $n^2-1$ by rank-nullity. $\endgroup$ – egreg Jan 12 at 17:38
  • 1
    $\begingroup$ @omer You're right in that last comment... $\endgroup$ – DonAntonio Jan 12 at 17:42

The trace function is a linear functional from $\;M_n(F)=\;$ the vector space of all matrices of order $\;x\times n\;$ over a field $\;F\;$ to $\;F\;$, or simple a linear transformation between both vector spaces.

By the dimensions theorem, if a linear functional is not the zero functional, then it automatically is surjective and furthermore its kernel has dimension one less than the dimension of the domain of definition. That $\;P\;$ invertible helps us to deduce $\;f(x):=tr. (PX)\;$ is not the zero functional, thus $\;\dim\ker f= \dim M_n(F)-1=n^2-1\;$ and we're done.

The above is also true for any non-zero $\;P\;$ ... and thus the assumption that $\;P\;$ is invertible is way too much, in fact.


Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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