Suppose VV and WW are finite-dimensional and T∈L(V,W)T \in \mathcal L(V,W). Show that with respect to each choice of bases of VV and WW, the matrix of TT has at least dim⁡range T\dim \text{range }T nonzero entries.


Let v1,…,vnv_1,\dots,v_n and w1,…,wmw_1,\dots,w_m be the bases for VV and WW that will be used for the matrix (we don't control these).

Let AA be the matrix of TT, we can clearly find at least dim⁡range T\dim \text{range }T vectors in v1,…,vnv_1,\dots,v_n with Tvk≠0Tv_k \ne 0. This implies there is a nonzero element in each row because

0≠Tvk=∑j=1mAj,kwj0 \ne Tv_k = \sum_{j=1}^m A_{j,k}w_j

Implies at least one Aj,k≠0A_{j,k} \ne 0 (independence of WW basis). doing rr times implies the number of nonzero entries is at least dim⁡range T\dim \text{range }T.


This is the best bound we can get using only the dimension of the range, since the identity matrix has the number of nonzero entries equal to the dimension of the range.

todo (easy): Remove "clearly" and make rigorous, but make sure it's still understandable!