Decomposition of matrices into product of idempotents and separativity of regular rings

(1) We have r=er=euu^{-1}r. Hence our hypothesis shows that r \in \Pi E. We also have u^{-1}ru \in \Pi E and hence ru=eru=euu^{-1}ru\in \Pi E. Finally ur=u(ru)u^{-1}\Pi E.

(2) Let u^{-1}=\Sigma_{j\ne i} e_{jj} + (1-\Sigma_{j\ne i}\alpha_j)e_{ii}\in Gl_n(R). We then have have u^{-1}A=A_i=(1-e_{ii})A and we check that the result follows by applying (1).

(3) If A is a product of idempotent matrices then there exists E=E^2\in M_n(R) such that A=EA. Since R is projective free there exists an invertible matrix P\in M_n(R) such that PEP^{-1} is a diagonal matrix with zero and one on the diagonal. We thus have PAP^{-1}= PEP^{-1}PAP^{-1} and hence PAP^{-1} has a zero row.

Assume that A is a quasi elementary matrix. We note that a quasi elementary matrix always has a zero row and a zero column. Such a matrix is always similar to a matrix having last row zero. In other words, we may assume that our quasi elementary matrix A is of the form A=(I_{n}-e_{nn})Q where Q is an elementary matrix of the form Q=I_{n}+be_{ij}, where 1\le i \ne j\le n. If j=n, then A=(I_{n}-e_{nn})(I_{n}+be_{in})=I_{n}+be_{in}-e_{nn}. This matrix is easily seen to be idempotent. If j\ne n, we have

We can easily check that all the three factors are idempotent matrices.

First suppose that the matrix A has its i^{th} row zero and is a product of idempotent matrices.

Let Q=Q_{k,l}=Id +ae_{kl} with a\in R be an elementary matrix. If i\notin \{k,l\} then comparing the rows of the matrices on both sides we get that Q_{k,l}A=(id -e_{ii})Q_{k,l}A. If i=k, we have Q_{i,l}A= (Id.-e_{ii} + ae_{il})A (remark that Id.-e_{ii} + ae_{il} is an idempotent). Finally if i=l we have Q_{k,i}A=A . In the three cases we conclude that if A is a product of idempotents then the same is true for Q_{k,l}A. Conversely, noting that Q^{-1} is also an elementary matrix, we know, by Proposition 3, that the quasi elementary matrix (1-e_{ii})Q^{-1} is a product of idempotent matrices. Since the i^{th} row of A is a zero row, we have A=(Id-e_{ii})A=(1-e_{ii})Q^{-1}QA. This shows that if QA is a product of idempotent matrices, the same is true for A.

We leave the proof of the case when Q is a permutation matrix to the reader.

(1) B is a product of idempotent matrices, say B=E_1\cdots E_l then

is a product of idempotent matrices.

(2) If B is invertible we can write

(3) If the column C is right linear combination of the columns of B, say C=BQ, for some column Q then we have

Consider

Let us write B+A=E_1\cdots E_l where, for any 1\le i \le n, E_i^2=E_i we then have

This gives the proof.

It is enough to consider the case when r=1. Indeed if r>1 and we assume the result true for r=1, then the matrix A'=\begin{pmatrix} B & 0 \\ 0 & 0 \end{pmatrix}\in M_{n+1}(R) is a product of idempotent matrices. Hence A=\begin{pmatrix} A' & 0 \\ 0 & 0 \end{pmatrix} is a product of idempotent matrices.

(1) Let \tilde{B}=(1-e_{ii})B be the matrix obtained by replacing the i^{th} row of B by a zero row. We can then write \tilde{B}=B - e_iB_i, where B_i is the i^{th} row of B and e_i is the column with all entries zero except the i^{th} one which is 1. Proposition 7 then applies and gives the case when r=1.

(2) It is clear that A is in fact a product of quasi elementray matrices and hence Proposition 3 shows that A is a product of idempotent matrices.

(3) Since A is clearly a quasi permutation matrix, it is a product of idempotent matrices.

(4) We consider only the case of upper triangular matrices. The proof proceeds by induction on n\ge 1. If n=1 we have the following decomposition

Assuming the result for upper triangular matrices of size n-1, we consider the case of a matrix A=\begin{pmatrix} B & 0 \\ 0 & 0 \end{pmatrix} where B\in M_n(R) is upper triangular. We consider, as in (1) above, the matrix \tilde{B}=B(I_n-e_{nn}) obtained by replacing the last column of B by zeros. Since \tilde{B} is upper triangular, the induction hypothesis shows that the matrix \tilde{B} is a product of idempotent and hence (1) above shows that A is also a product of idempotent matrices.

If the ring R is supposed to be von Neumann regular and separative then, the Morita invariance of these properties implies that the same is true for any matrix ring over R. Let B be any matrix in M_n(R). The main result of [M3] shows that the matrix

is a product of idempotent matrices if and only if the following relations between the annihilators are satisfied:

where S=M_{n+r}(R). Let us prove the first equality. We let

with X_1\in M_{n}(R) and X_4\in M_r(R) and other matrices are of the appropriate size. We notice that if X_1=0 and X_3=0 then X\in lann(A). We then get that lann(A) contains the matrices with the first n columns all zero.

Since the right ideal of S generated by these matrices is S=M_{n+r}(R), we get that lann(A)S=S. Similarly we also have Srann(A)=S. Since A=\begin{pmatrix} B & 0 \\ 0 & 0 \end{pmatrix}, we can show that the ideal generated by I_{n+r}-A is the ring S itself. This concludes the proof.

If n=1 we can consider the decomposition obtained in the proof of Proposition 8 (4).

So suppose that n>1, and that R is a GE ring. Since B is invertible, the matrix B is a product of elementary matrices and invertible diagonal matrices. The conclusion follows easily from Proposition 8.

We first remark that, as in the begining of the proof of Proposition 8, it is enough to consider the case when l=2n. So we assume that A\in M_{2n}(R). Proposition 4 shows that that the given matrix is a product of idempotents if and only if the matrix

is a product of idempotents. But for this matrix we have the obvious decompositon