Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $A:E\rightarrow F$ be a Linear Transformation between finite dimensional vector spaces, with $\mathrm{Rank}(A)=r$ and $\dim E=n$, $\dim F=m$. Prove that there are basis in $E$ and $F$ such that the matrix of $A$ has $a_{11}=\cdots=a_{rr}=1$ and $a_{ij}=0$ everywhere else, as entries.

I thought in the change of basis $ap=qa'$ where $a'$ would be the matrix we want but as I got no information about $a$, $p$, $q$ then this is not a way out definetely. Then as the rank is the maximum number of independent columns and rows I thought I could just erase the ones that are linear dependent but this doesn't guarantee me that the transformation would be the same transformation without the deleted linear dependent columns and rows.

A hint would be apreciated, Thanks in advance.

share|cite|improve this question
If you want to kill a fly with a sledgehammer, you could apply the Singular Value Decomposition. – wildildildlife Jun 25 '11 at 16:35
up vote 3 down vote accepted

The columns of the matrix $A$ are the components of the images of the basis vectors of $E$, i.e. $A(u_i)= A\cdot u_i$ for some basis $<u_1,...,u_n>=E$. Now the null space $\ker A=\{v\in E|\, A(v)=0\in F\}$ is a vector subspace of $E$ and the image space $\text{im} A=\{w\in F\, |\,\exists v\in E, w=A(v)\}$ a subspace of $F$ with both satisfying the Rank-Nullity theorem $\text{rank}(A)+\text{nullity}(A)=n$, that is $$ \dim(\text{im}\,A)+\dim(\ker A) = \dim E $$

This means $E$ is decomposed into the space of vectors which map to $0$ in $F$ through $A$, that is $\ker A$, and its complementary subspace $E-\ker A=:\widetilde{\text{im}} A < E$, the space of vectors which map to all the nontrivial vectors into the image of $A$, i.e. $E=\widetilde{\text{im}} A\oplus \ker A$. In fact this is just another way of stating the first isomorphism theorem for vector spaces (where $<$ means vector subspace) $$ E>\widetilde{\text{im}} A\cong \frac{E}{\ker A}\cong \text{im}\,A<F. $$ With this decomposition of $E$, you choose a basis of $E$ formed by the $r$ basis vectors $\{u_1,...,u_r\}$ of $\widetilde{\text{im}} A$ and the $n-r$ basis vectors $\{u_{r+1},...,u_n\}$ of $\ker A$. In this basis the matrix of $A$ whose columns are the components of the vectors $A(u_i)$, has zeros in the last $n-r$ columns since they are images of vectors in the null space. Finally the first $r$ columns are $A(u_1)...A(u_r)$, but since $A$ has rank $r$ these must be linearly independent and thus a basis of $\text{im}\, A$. Choose this set of $\{A(u_1),...,A(u_r)\}$ as new basis of the image, so that you can complete it to a basis of $F$ adding some $\{v_1,..., v_{m-r}\}\subset F$. Therefore you have arrived to basis of $E$ and $F$ where the decompositions in $F$ of the linear transformation of the $E$ basis are $$ \begin{aligned} & A(u_i)=\sum_{j=1}^r \delta_{ij}A(u_j)+\sum_{j=1}^{m-r} 0\cdot v_j\;\;\text{for } u_i\in\widetilde{\text{im}} A \\ & A(u_k)=\sum_{j=1}^r 0\cdot A(u_j)+\sum_{j=1}^{m-r} 0\cdot v_j\;\;\text{for } u_k\in\ker A \end{aligned} $$ gives you the matrix of $A$ as $$ A=\begin{pmatrix} 1 & 0 & \cdots & 0 & 0 & \cdots & 0 \\ 0 & 1 & \cdots & 0 & 0 & \cdots & 0 \\ \vdots & \vdots & \ddots & 0 & \vdots & \ddots & 0 \\ 0 & 0 & \cdots & 1 & 0 & \cdots & 0 \\ 0 & 0 & \cdots & 0 & 0 & \cdots & 0 \\ \vdots & \vdots & \ddots & 0 & 0 & \ddots & 0 \\ 0 & 0 & \cdots & 0 & 0 & \cdots & 0 \\ \end{pmatrix}= \begin{pmatrix} \textbf{I}_{r\times r} & | & \mathbf{0}_{r\times (m-r)} \\ ----- & | & ----- \\ \mathbf{0}_{(n-r)\times r} & | & \mathbf{0}_{(n-r)\times (m-r)} \\ \end{pmatrix}. $$

share|cite|improve this answer
Dim(Im(A))+Dim(Ker(A))=Dim(E)$\not\Rightarrow$E=Im(A)$\oplus$Ker(A) in general. See A(x,y)=(x-y,x-y) and take w=(1,1) then w$\in Ker(A)\cap Im(A)$ – Ivan3.14 Jun 25 '11 at 20:42
@missing314: No, $E/\ker A\cong\text{im} A\cong\widetilde{\text{im}}A:=E-\ker A$. I never stated $\text{im}A\oplus\ker A=E$, that could only be the case for endomorphism $A:E\rightarrow E$, but your target is other $F$. What I mean by $\widetilde{\text{im}}A$ is the complementary subspace of $\ker A$ within $E$ whereas for nonendomorphism you have $\text{im}A$ subspace of $F$, so the equation you write does not make sense. The rank-nullity th. is a result from the first isomorph. theorem since the dim of a quotient is $\dim (E/\ker A)=n-\dim\ker A=n-r$ which is precisely $\text{rank}A$. – Javier Álvarez Jun 25 '11 at 21:39
@missing314: the example you mention has $\ker A=<(1,1)>=\text{im} A$, but what I say is that $\widetilde{\text{im}}A=\mathbb{R}^2-<(1,1)>=<(-1,1)>$, i.e. the complementary of $\ker A$ satisfies $\widetilde{\text{im}}A\cong\text{im}A\cong E/\ker A\Rightarrow E=\widetilde{\text{im}}A\oplus\ker A$ which is still true for your example by the isomorphism $\widetilde{\text{im}}A\ni (-x,x)\mapsto (x,x)\in\ker A$. This is because of the special case when $\text{im}A\subseteq\ker A$ for endomorphisms, but for the GENERIC CASE $\text{im}A\cap\ker A=\{0\}$ and thus $\text{im}A=\widetilde{\text{im}}A$. – Javier Álvarez Jun 25 '11 at 22:10

Let $\left\{ u_1, \dots , u_n\right\}$ be a basis of $E$. Then, $\left\{ Au_1, \dots , Au_n \right\}$ is a system of generators of the subspace $\mathrm{im} A \subset F$. Whenever you have a system of generators of a vector (sub)space, you can delete some of them in order to obtain a basis. Since $r= \mathrm{rank} A = \mathrm{dim}\ \mathrm{im}A$, you can take $r$ of those $Au_i$ to form a basis of $\mathrm{im}A$. Reordering the original basis $u_1, \dots , u_n$ if necessary, we can assume that these are the first ones. So $Au_1, \dots , Au_r$ are a basis of $\mathrm{im}A$.

Now, you have $r\leq m$ linearly independent vectors $Au_1, \dots , Au_r$ in $F$. You can always complete a set of linearly independent vectors in order to order to obtain a basis of your vector space. So, choose $m-r$ vectors $v_{r+1}, \dots , v_m \in F$ such that $Au_1, \dots , Au_r, v_{r+1}, \dots , v_m$ is a basis of $F$.

And you're done: $\left\{ u_1, \dots , u_n\right\} \subset E$ and $\left\{ Au_1, \dots , Au_r, v_{r+1}, \dots v_m\right\} \subset F$ are bases you were looking for.

EDIT. I'm afraid my answer is wrong. If you perform the steps in it, the matrix you'll obtain looks like

$$ \begin{pmatrix} 1 & 0 & \dots & 0 & a^1_{r+1} & \dots & a^1_n \\ 0 & 1 & \dots & 0 & a^2_{r+1} & \dots & a^2_n \\ \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \dots & 1 & a^r_{r+1} & \dots & a^r_n \\ 0 & 0 & \dots & 0 & 0 & \dots & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \dots & 0 & 0 & \dots & 0 \end{pmatrix} $$

But you have no control on the remaining $a^i_j$.

share|cite|improve this answer
The remaining $\{v_{r+1},...,v_m\}\subset F$ do not matter as long as you choose a basis of $E$ formed by the $u_1,...,u_r$ and choose the remaining $\dim\ker A=n-r$ vectors belonging to $\ker A$. In this way their image under $A$ is null in any basis of $F$ (in particular in the one constructed by completing $Au_1,...,Au_r$ with $v_{r+1},...,v_m$) and therefore your last $n-r$ columns of your matrix are all zeros whereas the first $r$ columns are as you obtained. – Javier Álvarez Jun 26 '11 at 18:17
Yes, but I should re-write all my answer. And you already did it. :-) – a.r. Jun 28 '11 at 23:19

This is a mixed of Javier and Agustí answers, I can take any basis of E, say $\{v_{1},\cdots v_{n}\}$ then look at the images $\{Av_{1},\cdots Av_{n}\}$ and find a basis of the image here with this basis $\{Av_{1},\cdots Av_{r}\}$ propertly ordered is obvious that the members of $\{v_{1},\cdots v_{r}\}$ don't belong to $Ker(A)$ then I can take a basis of the kernel as the dimension of it is DimE-Dim(Im(A))=n-r, let that basis be $\{v'_{r+1}\cdots v'_{n}\}$ then is easy to see that the basis asked for E is $\{v_{1},\cdots v_{r},v'_{r+1}\cdots v'_{n}\}$ and for F the basis would be $\{Av_{1},\cdots Av_{r},w_{r+1}\cdots w_{m}\}$ as I can always complete the basis of the image to a basis of the whole space F.

Thanks both (Javier and Agustí)

share|cite|improve this answer
looking well at my asnwer seems like there is a gap in the part where a I choose a basis of the kernel and joint it with the $\{v_{1}\cdots v_{r}\}$ as this doesn't guarantee me that the whole $\{v_{1}\cdots v_{r},v'_{r+1},\cdots v'_{n}\}$ would be linear indenpent. So to correct this I think a basis of the kernel should be chosen first then complete to a basis of the whole E – Ivan3.14 Jun 25 '11 at 22:35
indeed, find a basis of the image by expressing $A\cdot v$ as a linear combination of independent vectors of $F$. Then complete it to a basis of $F$. Now, solve the system $A\cdot u=0$ giving you the basis of $\ker A$; complete it to a basis of $E$. Then in those basis $A$ has the required matrix form. – Javier Álvarez Jun 26 '11 at 1:35

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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