13.07.2015 Views

Semigroup of generalized inverses of matrices

Semigroup of generalized inverses of matrices

Semigroup of generalized inverses of matrices

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Semigroup</strong> <strong>of</strong> <strong>generalized</strong> <strong>inverses</strong> <strong>of</strong> <strong>matrices</strong> 151( )Q be two nonsingular <strong>matrices</strong> such that A = Q −1 Ir 0P . Then it is sufficient( )0 00 0to take A ′ = Q −1 P with w ∈ M0 w(m−r)×(n−r) (K) andrank(w) = min (m, n) − r.b) Let rank(A) = r, rank(B) = s. According to 1, there exist <strong>matrices</strong> A ′ andB ′ ∈ M m×n (K) <strong>of</strong> ranks min (m, n) − r and min (m, n) − s such that rank (A + A ′ ) =rank(A) + rank(A ′ ) = min (m, n) and rank (B + B ′ ) = rank(B) + rank(B ′ ) =min (m, n). It follows that (A + A ′ ) {1} ⊂ A {1} ∩ A ′{1} , (B + B ′ ) {1} ⊂ B {1} ∩ B ′{1} .Since A + A ′ , B + B ′ have the same rank, they are equivalent. So, there existsan isomorphism ϕ from A {1} on ϕ ( A {1}) (such that ϕ (A + A ′ ) {1}) = (B + B ′ ) {1} .Hence we have(ϕ (A + A ′ ) {1}) (⊂ ϕ A {1} ∩ A ′{1}) (= ϕ A {1}) ∩ ϕ(A ′{1})(and ϕ (A + A ′ ) {1}) ⊂ B {1} ∩ B ′{1} . Finally, we have ϕ ( A {1}) ∩ B {1} ≠ ❢ . □The natural question arises whether it is possible to conclude that for any <strong>matrices</strong>A, B ∈ M m×n (K) we have A {1} ∩ B {1} ≠ ❢ . The answer is negative. In fact if wetake B = αA for a scalar α which is different <strong>of</strong> 1 and 0, then if X ∈ A {1} ∩ B {1} ,we have αA = αAXαA = α 2 A. Thus α ∈ {0, 1} which contradicts our assumption.Consequently we have A {1} ∩ B {1} = ❢ .3.2.3 Partial order in M {1}m×n (K)Definition 3.1. ([5]) A minus partial order, denoted by ≺ − , is defined as follows forA, B ∈ M m×n (K), then A ≺ − B if rank(B) = rank(A) + rank (B − A).Theorem 3.6. 1) The inclusion is a partial order in M m×n(K) {a} induced by the minusorder in the reverse order.b) Let m 0 = min (m, n). For any matrix A ∈ M m×n (K) <strong>of</strong> rank r there exists asequence <strong>of</strong> <strong>matrices</strong> A = A r ≺ − A r+1 ≺ − .... ≺ − A m0 in M m×n (K) such thatrank (A r ) = rank (A) = r, rankA r+i = r + i for i = 1, ...m 0 − r. Thus, there existsa sequence A {1}m 0⊂ · · · ⊂ A {1}r = A {1} and A {1}m 0is the last term.Pro<strong>of</strong>. a) For A, B ∈ M m×n (K), if A ≺ − B, then rank(B) = rank(A)+rank (B − A).By Lemma 3.2 it follows that B {1} ⊂ A {1} . Hence we have the partial order ⊂ inM m×n(K). {1} b) Let {v 1 , v 2 , ...v r } be a basis <strong>of</strong> R(A) and {v r+1 , v r+2 , ...v m } be suchtogether with the basis <strong>of</strong> R(A) form a basis for K m . Let {e 1 , e 2 , ...e n } be a basis forK n such that A r e j = Ae j = v j for j = 1, ...r and A r e j = 0 for j = r + 1, ...mand for i = 1, ...m 0 − r, A r+i e j = v j for j = 1, ...r + i and A r+i e j = 0 forj = r + i + 1, ...m. ( In these ) bases the <strong>matrices</strong> ( A = ) A r and A r+i are <strong>of</strong> the formA = A r = Q −1 Ir 0P , A0 0r+i = Q −1 Ir+i 0P for i = 1, ...m0 00 − r. Thenwe have rank (A r ) = rank(A) = r , andrank (A r+i ) = rank (A r ) + i = r + i = rank(A r ) + rank (A r+i − A r ) ,

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!