Isomorphism by restricting surjective linear mapping between finite dimensional spaces, and quotient space by null space

1.3k Views Asked by At

In Axler's book Linear Algebra Done Right, Section 3D, Exercise 8:

Suppose $V$ is finite-dimensional and $T:V\rightarrow W$ is a surjective linear map of $V$ onto $W$. Prove that there is a subspace $U$ of $V$ such that $\left.T\right|_U$ is an isomorphism of $U$ onto $W$. (Here $\left.T\right|_U$ means the function $T$ restricted to $U$. In other words, $\left.T\right|_U$ is the function whose domain is $U$, with $\left.T\right|_U$ defined by $\left.T\right|_U (u)=Tu$ for every $u\in U$.)

An immediate solution would be considering a basis $w_1, \dots, w_m$of $W$. Then there exists $v_1, \dots, v_m \in V$, linearly independent, and $Tv_i=w_i$. Let $U =$ span($v_1, \dots, v_m$), then we get an isomorphism $\left.T\right|_U : U\rightarrow W$.

My question is that do we have $U = V/(\operatorname{null}T)$, where $V/(\operatorname{null}T)$ is the quotient space of $V$ by null$T$? And do we have $\left.T\right|_U = \tilde{T}$, where $\tilde{T}: V/(\operatorname{null}\to W)$ is defined by $\tilde{T}(v+\operatorname{null}T)=Tv$?

1

There are 1 best solutions below

2
On BEST ANSWER

Given finite-dimensional $k$-vector spaces $V$ and $W$ with a surjective linear transformation $T : V \to W,$ by the First Isomorphism Theorem, there exists a unique linear transformation $S : V / \ker T \to W$ such that $T = S \circ P,$ where $P : V \to V / \ker (T)$ is the canonical surjection defined by $P(v) = v + \ker T.$ Consequently, your intuition is essentially correct; however, it is not technically true that $U = V / \ker T.$

We will need to do things in a slightly different order than your proof, but we will ultimately establish that $V = U \oplus \ker T$ so that $U$ can be identified (by the First Isomorphism Theorem) with the quotient space $V / \ker T.$ Recall that any linearly independent set of fewer than $\dim V = n$ vectors can be extended to a basis for $V.$ Consequently, we may extend a basis $\{v_{m + 1}, \dots, v_n \}$ of $\ker T$ to a basis $\{v_1, \dots, v_m, v_{m + 1}, \dots, v_n \}$ of $V,$ where we denote $\dim W = m.$ We may define (as you have) the subspace $U = \operatorname{span}_k \{v_1, \dots, v_m \}.$ We claim that $T(U) = W$ so that $T|_U : U \to W$ is surjective and therefore an isomorphism. Given any vector $w$ in $W,$ there exists a vector $v$ in $V$ such that $T(v) = w$ by hypothesis that $T$ is surjective. Using our basis, we may write $v = a_1 v_1 + \cdots + a_m v_m + a_{m + 1} v_{m + 1} + \cdots + a_n v_n,$ from which it follows that $$\begin{align} w = T(v) &= T(a_1 v_1 + \cdots + a_m v_m + a_{m + 1} v_{m + 1} + \cdots + a_n v_n) \\ \\ &= T(a_1 v_1 + \cdots + a_m v_m) + a_{m + 1} T(v_{m + 1}) + \cdots + a_n T(v_n) \\ \\ &= T(a_1 v_1 + \cdots + a_m v_m) \end{align}$$ by hypothesis that $v_{m + 1}, \dots, v_n$ are basis elements for $\ker T.$ We have therefore that every vector $w$ of $W$ can be written as $T(u)$ for some vector $u$ in $U$ so that $T(U) = W,$ as desired.

Last, in order to prove that $V = U \oplus \ker T,$ we must show two things: (1.) $V = U + \ker T$ and (2.) $U \cap \ker T = \{0\}.$ Considering that $\{v_1, \dots, v_m, v_{m + 1}, \dots, v_n \}$ is a basis for $V$ such that $\{v_1, \dots, v_m \}$ is a basis for $U$ and $\{v_{m + 1}, \dots, v_n \}$ is a basis for $\ker T,$ the first condition holds. Given a vector $v \in U \cap \ker T,$ we have that $v = a_1 v_1 + \cdots + a_m v_m$ and $v = a_{m + 1} v_{m + 1} + \cdots + a_n v_n,$ from which it follows that $$a_1 v_1 + \cdots + a_m v_m - a_{m + 1} v_{m + 1} - \cdots - a_n v_n = 0.$$ But this implies that $a_1 = \cdots = a_m = a_{m + 1} = \cdots = a_n = 0$ so that $v = 0,$ as desired.


Perhaps this means nothing to you at the moment, but this is a consequence of a more general notion: we say that a short exact sequence of $R$-modules $$0 \to L \xrightarrow \varphi M \xrightarrow \psi N \to 0$$ splits if there exists an isomorphism $\gamma : M \to L \oplus N$ such that $\gamma \circ \varphi : L \to L \oplus N$ is the inclusion map and $\psi \circ \gamma^{-1} : L \oplus N \to N$ is the canonical surjection into the second coordinate. Given that $R = k$ is a field, we refer to an $R$-module (or a $k$-module, as it were) as a $k$-vector space. Every $k$-vector space is free as an $R$-module, and every free $R$-module is projective. Every short exact sequence of $R$-modules that ends with a projective $R$-module splits. By the First Isomorphism Theorem, we obtain the short exact sequence $$0 \to \ker T \xrightarrow i V \xrightarrow T W \to 0,$$ where $i : \ker T \to V$ is the inclusion map. By identifying $W$ and $V / \ker T,$ we obtain the short exact sequence $$0 \to \ker T \xrightarrow i V \xrightarrow P \frac V {\ker T} \to 0.$$ Ultimately, this is all to say that this behavior you intuitively detected is predictable and always works.