The entire problem statement is:

Let $V$ be a finite dimensional vector space and $T:V\to V$ be the projection of $W$ along $W'$, where $W$ and $W'$ are subspaces of $V$. Find an ordered basis $\beta$ of $V$ such that $[T]_\beta$ is a diagonal matrix.

The book I took this problem from is Linear Algebra by Friedberg, Insel, & Spence.

My attempt of the proof is as follows:

We have that $V=W\oplus W'$, so let $\beta_1=\{v_1,\dots,v_k\}$ and $\beta_2=\{v_{k+1},\dots,v_n\}$ be bases for $W$ and $W'$, respectively. Then we have that $\beta_1\cap\beta_2=\varnothing$ and $\beta=\beta_1\cup\beta_2$ is a basis for $V$. Now to determine the matrix representation of $T$ in the ordered basis $\beta$ consider $T(v_j)$ for $1\leq j\leq k$, \begin{align} T(v_j)&=\sum_{i=1}^ka_{ij}v_i\nonumber\\ &=1\cdot v_j\nonumber \end{align} The last equality occurs since $\beta_1$ is a basis for $W$ and hence linearly independent. Considering $[T]_\beta$ as a block matrix, it follows from above that the sub-matrix composed of the first $k$ rows and $k$ columns is the identity matrix of size $k$, that is, $I_k$. Moreover, since $T$ is the projection of $W$ along $W'$, the entries of $[T]_\beta$ can be entirely specified using only $\beta_1$. Hence, the entries outside of the $k\times k$ sub-matrix are zero. Thus we have that $$ [T]_\beta=\left( \begin{array}{c|c} I_k & 0 \\ \hline 0& 0\\ \end{array}\right) $$ where the off diagonal zero matrices have size $(n-k)\times k$ and $k\times(n-k)$. Also, the diagonal zero matrix has size $(n-k)\times (n-k).$

My main concern is when I stated that $V=W\oplus W'$, which then allowed me to conclude that $\beta=\beta_1\cup\beta_2$ was a basis for $V$. The reason I'm assuming this is because the definition provided for the projection $T$ on $W$ along $W'$, it is assumed that $V$ is the direct sum of $W$ and $W'$.

Thanks for any help or feedback!


1 Answer 1


Well, you are correct that $V=W\oplus W'$, but a much simpler approach is to use the fact that for any projection we have $T^2=T$ so that $T(T-I)=0$. So in general we then have that the minimal polynomial of $T$ is $m_T(x)=x(x-1)$ (the special "trivial" cases being $T= 0$ so that $W=0$ and $T = I$ so that $W'=0$). Since this polynomial consist only of linear factors, $T$ is diagonalizable with eigenvalues $1$ and $0$.

Now we have $T(w)=w$ if and only if $w \in W$, so that the eigenspace associate with $1$ is $W$. Similarly we have $T(w')=0$ If and only if $w' \in W'$ so that the eigenspace associated with $0$ is $W'$. So any basis $\beta=\beta_W \cup \beta_{W'}$ where $\beta_W$ and $\beta_{W'}$ are bases for $W$ and $W'$ respectively is such that $[T]_\beta$ is diagonal.

  • $\begingroup$ Thanks for the response. So far in the book, they have not discussed eigenvalues or eigenspaces. So far I only have the definition of a linear function, some associated theorems with regards to the null space and range. The section the problem is taken from introduces the idea of an ordered basis. $\endgroup$
    – S.D.
    Apr 27, 2014 at 8:57
  • $\begingroup$ ah ok I see... your approach looks sound, I would just explicitly show that $T(v_j)=0$ when $j>k$. And then you must make sure you know why $V=W\oplus W'$, maybe in answering the question you may assume it, but for yourself it would be good if you can work it out. You must prove that $V=W+W'$ and $W \cap W' = \{0\}$ OR that dim$(W)+$ dim$(W')=$ dim$(V)$. This last equation you can prove using the properties of the projection. $\endgroup$ Apr 27, 2014 at 9:31
  • $\begingroup$ @ShantDanielian Are you using Friedberg as a textbook for a course? If so, is it postgrad? Or are you just self-studying? $\endgroup$ Apr 27, 2014 at 16:37
  • $\begingroup$ Yes, I'm using Friedberg's textbook and yes I'm in a course using this book as well. Now, since $T$ is defined as the projection on $W$ along $W'$, and how I defined each basis, doesn't the fact $T(v_j)=0$ for $j>k$ just drop out? Moreover, I went through a problem which stated that if the intersection of two bases of a subspace are empty and their union make a basis for the parent vector space, then $V=W\oplus W'$. Moreover, from the definition of projection $T$, it assumes the fact V is written as a direct sum. Isn't that the case here? $\endgroup$
    – S.D.
    Apr 27, 2014 at 20:55
  • $\begingroup$ @ShantDanielian ok a few subtle problems here...I will refer to your original question: `Moreover, since $T$ is the projection of $W$ along $W′$, the entries of $[T]_\beta$ can be entirely specified using only $\beta_1$.' This statement is quite simply false. A zero entry isn't nothing. You have shown that the first $k$ columns of $[T]_\beta$ are the columns of $I_k$, but you have to show what the last $n-k$ columns are too. Regarding the moreover part: where did you prove that $\beta_1 \cap \beta_2 = \emptyset$? And sure it implies that $V$ is written as a direct sum, but do you know why? $\endgroup$ Apr 28, 2014 at 5:43

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.