Suppose VV and WW are finite-dimensional and TL(V,W)T \in \mathcal L(V,W). Prove that there exists a basis of VV and a basis of WW such that with respect to these bases, all entries of M(T)\mathcal M(T) are 00 except that the entries in row jj, column jj equal 11 for 1jdimrange T1 \le j \le \dim \text{range }T.

(This generalizes 3c/2)


Let v1,,vnv_1,\dots,v_n be a basis of VV such that Tv1,,TvrTv_1,\dots,Tv_r is a basis for range T\text{range }T and vr+1,,vnv_{r+1},\dots,v_n is a basis for null T\text{null }T. The existance of such a basis is shown in the proof of 3.22 though it isn't in the theorem statement. (todo: ugly! construct it using book results)
Let wj=Tvjw_j = Tv_j for 1jr1 \le j \le r then extend to a basis w1,,wmw_1,\dots,w_m of WW.

Clearly M(T)\mathcal M(T) will have ones in the diagonal up to r=dimrange Tr = \dim \text{range }T since writing TvkTv_k in terms of the matrix, where 1kr1 \le k \le r gives

Tvk=j=1mAj,kwjTv_k = \sum_{j=1}^m A_{j,k} w_j

Now since wk=Tvkw_k = Tv_k and the ww's are independent we must have

Aj,k={1if j=k0otherwiseA_{j,k} = \begin{cases} 1 &\text{if $j = k$} \\ 0 &\text{otherwise} \end{cases}

To finish we need to show that the (r+1),,m(r+1),\dots,m columns are all zero. This is because Tvk=0Tv_k = 0 for k>rk > r and

Tvk=0=j=1mAj,kwjTv_k = 0 = \sum_{j=1}^m A_{j,k} w_j

Implies (by independence of w1,,wmw_1,\dots,w_m) that Aj,k=0A_{j,k} = 0 for all jj.


This was intuitively obvious to me after doing 3c/2, but took a while to make rigorous. I need to improve my "rigorous articulation"