12

I'm working on the following problem:

Let $A$ be the ring of $n \times n$ matrices over a field $\mathbb{F}$.

(a) Show that for any subspace $V$ of $\mathbb{F}^n$, the set $I_V$ of matrices whose kernel contains $V$ is a left ideal of $A$.

(b) Show that every left ideal of $A$ is principal.

I've done part $a)$, but would like to know if you can prove $b)$ directly from $a)$. It seems to me that given the left ideal $J$, it should be the case that if $V$ is the intersection of the kernels of matrices in $J$, then we should have $J = I_V$. I can show that $I_V$ is principal, and certainly $J$ is contained in $I_V$, but I can't show the other direction.

I think you can prove $b)$ by considering the subspace $W$ of $\mathbb{F}^n$ consisting of the rows of elements of $J$, which is of dimension $k \leq n$ say, and then showing that $J$ is generated by any matrix whose first $k$ rows are some basis for $W$ and whose final rows are all $0$. But it seems that we should be able to do the problem just using $a)$, and I'd like to know how to do it!

Tom Oldfield
  • 13,034
  • 1
  • 39
  • 77
  • Just out of curiosity, where did you get this problem from? The topic looks very interesting. – AnotherPerson Dec 23 '15 at 03:26
  • 1
    @SirJective It was an old prelim exam from here:

    https://www.ma.utexas.edu/academics/graduate/prelims/old_exams/

    It was from 2013_08 Algebra A, if you're interested.

    – Tom Oldfield Dec 23 '15 at 03:32
  • Muchas gracias amigo – AnotherPerson Dec 23 '15 at 03:38
  • 1
    I've never seen this angle on the problem. All of the ways I have in mind involve the semisimplicity or Von Neumann regularity of the ring, but this seems like a distinct path. – rschwieb Dec 23 '15 at 12:09
  • Closely related: http://math.stackexchange.com/questions/26726/what-are-the-left-and-right-ideals-of-matrix-ring-how-about-the-two-sided-ideal – user26857 Dec 23 '15 at 21:04

3 Answers3

12

The main result is the following:

Let $V$ be a finitely dimensional vector space. Then every left ideal in $\operatorname{End}(V)$ is of the form $\mathcal{I}(W) \colon =\{T \in \operatorname{End}(V) \mid T = 0 \textrm{ on } W\}$ ( $W$ is a subspace of $V$)

Indeed, let $I$ be a left ideal of $\operatorname{End}(V)$ and $$W = \mathcal{Z}(I) \colon= \{ w \in V \mid T(w) = 0 \textrm{ for all } T \in I\}.$$

Let us show that $I = \mathcal{I}(W)$, or, in other words $$I = \mathcal{I}(\mathcal{Z}(I)).$$ for every left ideal $I$. Note that by definition $$\mathcal{Z}(I) = \bigcap_{T \in I} \ker(T).$$ Since $V$ is a finite dimensional space there exist finitely many $T_1$, $\ldots $, $T_m \in I$ so that $$W=\mathcal{Z}(I) = \bigcap_{i=1}^m\ker(T_i).$$ Consider the operator $\tilde T= (T_1, \ldots, T_m)$ from $V$ to $V^m$, with kernel $\bigcap_{i=1}^m\ker(T_i) = W$.

Let now $S \in\operatorname{End}(V)$ that is $0$ on $W$. It follows (by a standard universality result) that there exists $L:\operatorname{Im}(\tilde T) \to V$ so that $$S = L \circ \tilde T.$$ Now $L$ can be extended to the full $V^m$. We know the form of linear maps from $V^m$ to $V$. They are given by $L = (L_1, \ldots , L_m)$ with $L_i \in\operatorname{End}(V)$. Therefore we have $$S = \sum_{i=1}^m L_i T_i,$$ so $S \in I$.

${\bf Added.}$ Let again $I$ be a left ideal, $W = \mathcal{Z}(I)$. We know from the above that $I = \mathcal{I}(W)$. From the proof above we see that any family $T_i$ with $\cap_{i=1}^m\ker(T_i) = W$ is a system of generators of $I$. So take $T$ so that $\ker T= W$. Then $T \in I$ and moreover, $T$ generates $I$. Therefore $I$ is a principal ideal.

Obs: Similarly (by duality say) one shows that every right ideal $J$ of $\operatorname{End}(V)$ is also of the form $$J = \mathcal{J}(W) = \{ T \in \operatorname{End}(V) \mid\operatorname{Im} T \subset W\}$$ Moreover, any family $T_i$ such that $\sum_{i=1}^m \operatorname{Im} T_i = \sum_{T \in J} \operatorname{Im} T$ generates $J$. Similarly, any right ideal is principal.


$\bf{Added:}$ The universality properties mentioned above:

  1. Let $\phi \colon M \to N$ a surjective morphisms ( of groups, modules, sets), $\phi' \colon M \to N'$ be another morphism such that $\operatorname{ker} \phi\subset \operatorname{ker}\phi'$. Then there exists a unique morphism $\chi \colon N \to N'$ such that $\phi' = \chi \circ \phi$.

  2. Let $\phi\colon V \to W$, $\phi\colon V \to W'$ linear maps such that $\operatorname{ker} \phi\subset \operatorname{ker}\phi'$. Then there exists a linear map $\chi \colon W \to W'$ ( not necessarily unique) such that $\phi' = \chi \circ \phi$.

    (2. )follows from 1. when we consider the surjective map from $V$ to $\operatorname{Im} \phi$, find a linear map from $\operatorname{Im} \phi$ to $W'$, and extend to a linear map from $W$ to $W'$ ( we use the fact that we are dealing with vector spaces.

    (1.) is basic in algebra and elsewhere, its proof is simple, should be checked.

orangeskid
  • 53,909
  • Hi, thanks for the answer. Could you clarify on why we can find a single $T$ in $I$ that has kernel $W$? – Tom Oldfield Dec 23 '15 at 20:01
  • 1
    @Tom Oldfiel: Yes, given any subspace $W$ there exists $T$ with $\ker T = W$. For take a complementary subspace $W'$ to $W$ and define $T(w + w') = w'$ for every $w + w' \in W\oplus W' = V$. – orangeskid Dec 23 '15 at 20:10
  • 1
    Sure, but how do we know that this $T$ lies in $I$? – Tom Oldfield Dec 23 '15 at 20:14
  • 1
    @Tom Oldfield: Remember, we already proved that if $T$ is zero on $W$ then $T$ is in $I$. Moreover, we also proved that if $\ker T = W$ and $S$ is zero on $W$ then $S = L T$ for some $L$ ( that is the case where $m=1$ above). Moreover, we know that every $S$ in $I$ is zero on $W$, by the definition of $W$ being $Z(I)$. – orangeskid Dec 23 '15 at 20:21
  • Oh right, so we're using the result that $I = I_W$, I thought we were doing it a priori. Thanks! – Tom Oldfield Dec 23 '15 at 20:39
  • Hi, I found this answer really helpful. There is one small thing that confused me for a while, and I thought perhaps you could clarify it in your answer. At the end you say "by duality, say" that the right ideals are of the form ${\text{Im} \subset W}$. That seems like the dual not of the result you proved, but of the (easier) result that all left ideals are of the form ${\text{Rowspace} \subset U}$. – Eric Auld Mar 01 '17 at 18:46
  • Sorry for being late. May I please ask what "standard universality result" you were referring to when claiming there exists $L: im(\tilde{T})\to V$ such that $S=L\circ \tilde{T}$? It seems to me, for direct sum, the universality property is that, for every $A_i: V\to V$, there exists a unique $f: V^m \to V$ such that $A_i=f\circ T_i$. That seems different to what you used. – user760 Jun 22 '23 at 08:05
  • But I can seem to make it work using basis. Suppose $B'\cup B$ is a basis of $V$ where $B'$ is a basis of subspace $W$. Then, $\dim(im(\tilde{T}))=|B|$, whereas $\dim(im(S))\leq |B|$. So there exists a $L: im(\tilde{T})\to im(S)$ s.t. $S=L\circ \tilde{T}$, since it's always possible to linearly map a space with larger dim to one with lower/equal dim surjectively. Correct? – user760 Jun 22 '23 at 08:47
  • 1
    @user760: Added some details about the universality properties mentioned. You are right, we are using some properties of vector spaces and linear maps, and those follows from existence of bases. So one way or another, bases are there. You can check that it also works if the field is not commutative ( like $\mathbb{H}$). Great! – orangeskid Jun 23 '23 at 20:09
  • For the right ideal $\mathcal{J}$, is the following correct? Suppose $T_1, \cdots, T_m\in \mathcal{J}$ and $\sum_1^m \text{Im}(T_i)=W$. Then, in $\mathcal{J}$, we can have $T_1', \cdots, T_m'$ defined as $T_j'=T_j\circ Q_j$ where $Q_j$'s are suitable projections. Then, $\text{Im}(T_i)\cap \text{Im}(T_j)=\emptyset$ for $i\neq j$, and $\oplus_1^m \text{Im}(T_i')=W$. Then, any $T$ having $W$ as its range would have the form $T=\sum_1^m T_j'\circ Q_j'$ with other suitable projections $Q_j'$'s, and so in the same right ideal $\mathcal{J}$. – user760 Jun 26 '23 at 12:40
  • 1
    @user760: I would try to prove the result for the ring of endomorphisms of a semi-simple module ( it behaves more or less like a vector space). The advantage is that we are forced to think in terms of modules rather than bases, and that can be also useful. I think your reasoning is good. You try to explain it to yourself and you will understand it better. – orangeskid Jun 27 '23 at 06:35
  • Thanks! I'll give that more general result a try! – user760 Jun 27 '23 at 07:10
2

Note that $I_V$ is principal left ideal of $M_n(\mathbb{F})$ whose generator can be taken to be any linear map $T \colon \mathbb{F}^n \rightarrow \mathbb{F}^n$ with $\ker(T) = V$. To see this, choose a basis $(w_1, \ldots, w_n)$ of $\mathbb{F}^n$ such that $(w_1, \ldots, w_k)$ is a basis of $\ker(T) = V$. Define $f_i = T(w_i)$ for $k + 1 \leq i \leq n$ and complete them to a basis $(f_1, \ldots, f_n)$ of $\mathbb{F}^n$.

Let $S \colon \mathbb{F}^n \rightarrow \mathbb{F}^n$ be a linear map with $V \subseteq \ker(S)$. We need to find a linear map $P \colon \mathbb{F}^n \rightarrow \mathbb{F}^n$ such that $PT = S$. Define $P$ by requiring that

$$ P(f_i) = \begin{cases} 0 & 1 \leq i \leq k, \\ S(w_i) & k + 1 \leq i \leq n. \end{cases}$$

Then

$$ P(T(w_i)) = \begin{cases} P(0) = 0 = S(w_i) & 1 \leq i \leq k, \\ P(f_i) = S(w_i) & k + 1 \leq i \leq n \end{cases} $$

which shows that $PT = S$.

Given any left-sided ideal $J$, let $V = \cap_{T \in J} \ker(T)$. Note that there must exist a linear map $T \in J$ with $\ker(T) = V$. Since $J$ is a left-sided ideal, we have

$$ (M_n(\mathbb{F}))T = I_V \subseteq J \subseteq I_V $$

which shows that $J = I_V$.

levap
  • 65,634
  • 5
  • 79
  • 122
  • Thanks a lot for the answer, I think the key thing that I was missing was the fact that any $T$ with the right kernel generates $I_V$. Could you be a little more specific though about how to find $T$ with $\ker(T) = V$? I know we can find finitely many $T_i$ such that the intersection of their kernels is $V$, and I can use this together with what you've done to prove the result, but I'm not sure how you found just one $T$. – Tom Oldfield Dec 23 '15 at 18:25
1

Hint:

  1. Given $A,B\in\operatorname{End}(V)$ where $V$ is a finite dimensional $\mathbb F$-v.s. Then $\exists T\in\operatorname{End}(V)\colon B=T\circ A\iff\ker A\subseteq\ker B$.
  2. Given finite number of subspaces $V_j\subseteq V,j=1,\dotsc,n$, we have $\sum_{j=1}^nI(V_j)=I(\bigcap_{j=1}^nV_j)$, where $I(W)=\{\,T\in\operatorname{End}(V)\,\vert\,W\subseteq\ker T\,\}$.
Yai0Phah
  • 9,733
  • Thanks for the answer! Do you have a good way of proving the left implication in part $1)$? Levap provides one below which is nice, but I'm not sure if you were thinking of something quicker. As for part $2)$, it's clear to me that the sum of the ideals is contained in the ideal of the intersection, but I'm not sure how to prove the other containment. – Tom Oldfield Dec 23 '15 at 18:31
  • @TomOldfield For the first one, you need to define $T(Ax)=Bx$ on the image $A(V)$. I give you a more chance to think throughly of the second point (the first point is needed here). – Yai0Phah Dec 23 '15 at 22:04
  • The idea for the first one sounds very much like what Levap did. Still not sure how this can be used for the second part, could you help me out please? – Tom Oldfield Dec 23 '15 at 22:45