Suspected error in proof about partitioned matrices in Matrix analysis by Roger Horn

82 Views Asked by At

I'm quite confused when reading Horn and Johnson's Matrix analysis (second edition) p.20 , 0.7.6 Rank in a partitioned matrix and rank-principal matrices .

Denote $M_{m,n}(F)$ as the set of all $m\times n $ matrices over field $F$ (arbitrary) , for square matrices use abbreviation . Partition $A \in M_n(F)$ as $$ A = \left(\begin{array}{rr} A_{11} & A_{12} \\ A_{21} & A_{22} \end{array}\right) ,\quad A_{11} \in M_r(F) , A_{22} \in M_{n-r}(F) $$ The book attempts to prove (actually it's $\iff$ but I don't think it's relevant here)

$rk(A)=rk[A_{11}\;A_{12} ] = rk\begin{bmatrix}A_{11}\\ A_{21}\end{bmatrix} \implies A_{11}$ is nonsingular .

as follow

Suppose $A_{11}$ is singular , then $rk(A_{11})=k<r $ . There are nonsingular $S,T\in M_r(F)$ such that $$ SA_{11}T = \left(\begin{array}{rr} I_{k} & 0 \\ 0 & 0_{r-k} \end{array}\right) $$ Therefore, $$ \hat{A} = \left(\begin{array}{rr} S & 0 \\ 0 & I_{n-r} \end{array}\right) A \left(\begin{array}{rr} T & 0 \\ 0 & I_{n-r} \end{array}\right) = \left(\begin{array}{rr} SA_{11}T & SA_{12} \\ A_{21}T & A_{22} \end{array}\right) $$ has rank $r$ , as do its block row and column . Because the $r^{th}$ row of the first block column of $\hat{A} $ is zero , there must be some column in $SA_{12}$ whose $r^{th}$ entry is not zero, which means that $\hat{A} $ has at least $r + 1$ independent columns. This contradicts $rank \hat{A} = rank A = r$ ,so $A_{11}$ must be nonsingular. $\square$

Question : Why $rank \hat{A} = rank A = r ? $ I only know $rank \hat{A} = rank A$ since multiplication with nonsingular matrices won't change rank . I don't recall making any assumption on $A_{12}\in M_{r,n-r}(F),A_{21}\in M_{n-r,r}(F)$ and $A_{22}$ , so let all of them be $0$ , then $rk(\hat{A}) = k < r $ .

Nonetheless , if assume $rank \hat{A} = rank A = r $ , then the rest of the proof follows . I couldn't find such assumption . Am I missing something ?