Suppose WW is finite-dimensional and T∈L(V,W)T \in \mathcal L(V,W). Prove that TT is injective if and only if there exists S∈L(W,V)S \in \mathcal L(W,V) such that STST is the identity map on VV.


Suppose TT is injective and define S(Tv)=vS(Tv) = v. If Tv1=Tv2Tv_1=Tv_2 then (by injectivity) v1=v2v_1=v_2 meaning SS maps the same input to the same output thus it's a function (aka it's well defined).

It's easy to see SS is linear (it "inherits" linearity from TT)

Currently SS is only defined on the range of TT, we can extend it to be defined on the rest of WW using 2.34 (this is where we need WW to be finite-dimensional!): If W=range T⊕UW = \text{range }T \oplus U and SS is defined on range T\text{range }T we're free to define Su=0Su = 0 for all u∈Uu \in U. Now S∈L(W,V)S \in \mathcal L(W,V) as desired.

For the backwards direction suppose STST is the identity map, then Tv1=Tv2Tv_1=Tv_2 implies S(Tv1)=S(Tv2)=v1=v2S(Tv_1)=S(Tv_2) = v_1 = v_2 so TT is injective.

More generally for any function the existence of a left inverse is equivilant to injectivity. The interesting thing here is that the linear structure was preserved.


Old bad proof

Note that null T={0}\text{null }T = \{0\} is finite-dimensional, and range T⊆W\text{range }T \subseteq W is finite-dimensional, so 3b/16 implies VV is finite-dimensional and thus by 3.22

dim⁡V=range T\dim V = \text{range }T

Let w1,…,wrw_1,\dots,w_r be a basis for range T\text{range }T, since each wjw_j is in the range of TT we can let v1,…,vrv_1,\dots,v_r be a list in VV such that wj=Tvjw_j = Tv_j, then define Swj=vjSw_j = v_j for all 1≤j≤r1 \le j \le r. Use 2.33 to extend to a basis w1,…,wmw_1,\dots,w_m of WW and finish definining SS by setting Swj=0Sw_j = 0 for all r<j≤mr < j \le m.
The list v1,…,vrv_1,\dots,v_r is independent, because

a1v1+⋯+arvr=0  ⟹  T(a1v1+⋯+arvr)=0apply T on both sides  ⟹  a1w1+⋯+arwr=0linearity  ⟹  a1=⋯=ar=0w1,…,wr are independent\begin{aligned} a_1v_1 + \dots + a_rv_r = 0 &\implies T(a_1v_1 + \dots + a_rv_r) = 0 &&\text{apply $T$ on both sides} \\ &\implies a_1w_1 + \dots + a_rw_r = 0 &&\text{linearity} \\ &\implies a_1 = \dots = a_r = 0 &&\text{$w_1,\dots,w_r$ are independent} \end{aligned}

Since dim⁡range T=dim⁡V=r\dim \text{range }T = \dim V = r 2.39 implies the list v1,…,vrv_1,\dots,v_r forms a basis for VV. Now can finally show STST is the identity!

STv=ST(a1v1+⋯+arvr)since v1,…,vr is a basis=a1S(Tv1)+⋯+arS(Tvr)linearity=a1v1+⋯+arvr=vby definition of S\begin{aligned} STv &= ST(a_1v_1 + \dots + a_rv_r) &&\text{since $v_1,\dots,v_r$ is a basis} \\ &= a_1S(Tv_1) + \dots + a_rS(Tv_r) &&\text{linearity} \\ &= a_1v_1 + \dots + a_rv_r = v &&\text{by definition of $S$} \end{aligned}

This completes the forward direction.

Now suppose SS exists such that STST is the identity map on VV. Clearly null T⊆null ST={0}\text{null }T \subseteq \text{null }ST = \{0\} so TT is injective by 3.16. This completes the backwards direction.


This took me ~1h, way too long! I diden't realize that VV coulden't be infinite-dimensional since it would force null T\text{null }T to be infinite-dimensional, since range T⊆W\text{range }T \subseteq W is clearly finite-dimensional.

I'm proud of this solution though, my first solution was to define SS as the inverse of TT, then show it's well defined and linear. I like this solution much better since linearity is obvious (we just defined SS in terms of WW's basis).
edit: what the hell ^ this is garbage why was I ever proud of this, the only new thing here is linear structure being preserved. left inverses existing being the same as injectivity is true for all functions