Suppose that VV is a finite-dimensional vector space and that T∈L(V,W)T \in \mathcal L(V,W). Prove that there exists a subspace UU of VV such that U∩null T={0}U \cap \text{null } T = \{0\} and range T={Tu:u∈U}\text{range }T = \{Tu : u \in U\}.


Let n1,…,nkn_1,\dots,n_k be a basis for null T\text{null }T then extend this to a basis n1…,nk,u1,…,urn_1\dots,n_k,u_1,\dots,u_r of VV using 2.33. Let

U=span(u1,…,ur)U = \text{span}(u_1,\dots,u_r)

Clearly U∩null T={0}U \cap \text{null }T = \{0\}, and range T={Tu:u∈U}\text{range }T = \{Tu : u \in U\} since the nullspace components disappear by linearity

T(v)=T(a1n1+⋯+aknk+b1u1+⋯+brur)=T(b1u1+⋯+brur)=T(u)\begin{aligned} T(v) &= T(a_1n_1 + \dots + a_kn_k + b_1u_1+\dots+b_ru_r) \\ &= T(b_1u_1+\dots+b_ru_r) \\ &= T(u) \end{aligned}

I was tempted to try the set U′={v∈V:Tv≠0}∪{0}U' = \{v \in V : Tv \ne 0\} \cup \{0\} but this doesn't work as it isn't closed under addition. Specifically notice that n1+u1∈Un_1+u_1 \in U and n1−u1∈Un_1-u_1 \in U but their sum n1n_1 isn't. Having Tv≠0Tv \ne 0 isn't enough, we need vv to contain no nullspace component.