Suppose VV is finite-dimensional and E\mathcal E is a subspace of L(V)\mathcal L(V) such that ST∈EST \in \mathcal E and TS∈ETS \in \mathcal E for all S∈L(V)S \in \mathcal L(V) and all T∈ET \in \mathcal E. Prove that E={0}\mathcal E = \{0\} or E=L(V)\mathcal E = \mathcal L(V).


Suppose E≠{0}\mathcal E \ne \{0\}, I'm going to show this implies E=L(V)\mathcal E = \mathcal L(V).

Let TT be a nonzero map in E\mathcal E and let v∈Vv \in V be such that Tv≠0Tv \ne 0.
Let v1=vv_1 = v and extend to a basis v1,…,vnv_1,\dots,v_n of VV, let AA be the matrix of TT with respect to this basis.

Basically, we extract the nonzero component of AA by multiplying it by elementary matrices (row operations) to get

[1000]and[0001]\begin{bmatrix} 1 & 0 \\ 0 & 0 \end{bmatrix} \quad\text{and}\quad \begin{bmatrix} 0 & 0 \\ 0 & 1 \\ \end{bmatrix}

(In general more then 2d but latexifying matrices is annoying)

Then we add them to get that the identity is in E\mathcal E (subspace of E\mathcal E is closed under linear combinations), which allows us to show E=L(V)\mathcal E = \mathcal L(V) because for any S∈L(V)S \in \mathcal L(V), IS=S∈EIS = S \in \mathcal E.

todo: Provide details (I'm lazy right now)