Proposition: Let A and B be nxn matrices. Also, let
AB=I
then
BA=I
Proof: Every matrix is row equivalent to a single row-reduced echelon matrix. That is, for every matrix, there is a finite sequence of elementary row operations such that if we apply this sequence of row operations on this matrix, we get a row-reduced echelon matrix. If we denote R to be this matrix,
er(er-1(………e1(R)…..))=A
for some elementary row operations, er,er-1,…….,e1. Every elementary row operation on any matrix corresponds to multiplication of an elementary matrix with that matrix. This elementary matrix is obtained by doing the row operation on Imxm where m is equal to row number of the matrix that we apply the elementary row operation. Denote Ei to be correspondent elementary matrix of ei. Then,
ErEr-1…….E1R=A
(ErEr-1…….E1R)B=I
multiply both sides Er‘, inverse of Er
Er‘(ErEr-1…….E1R)B=Er‘I
keep multiplying,
Er-1‘Er‘(ErEr-1…….E1R)B=Er-1‘Er‘I
E1‘……..Er-1‘Er‘(ErEr-1…….E1R)B=E1‘……….Er-1‘Er‘I
RB=E1‘……….Er-1‘Er‘I
Can R contain a zero row? No, because if it were, then RB would contain a zero row and any matrix that is row equivalent to I can’t contain a zero row. Why? Because if there were such matrix, we would obtain a contradiction. Let’s assume that RB contains some zero rows and find (RB)’ (equal to RB without zero rows) matrix’s row-reduced echelon matrix, and then add zero rows at the end of this row-reduced echelon matrix. Then, this final matrix, say M, is row reduced echelon matrix of RB, and clearly it is not equal to I. This is a contradiction, since we have two distinct row reduced echelon matrix of RB; one is I and the other is M. This is not possible.
So, R must not have zero row, then it must be I (easy to verify). Then,
RB=IB=B=E1‘……….Er-1‘Er‘I
Is BA=I?
BA=(E1‘……….Er-1‘Er‘I)(ErEr-1…….E1R)
=(E1‘……….Er-1‘Er‘I)(ErEr-1…….E1I)=I
Yes, it is