Positive decompositions of selfadjoint operators (Q989965)

Let \(H\) be a separable complex Hilbert space and \(L(H)\) be the Banach algebra of all bounded linear operators on \(H\). For \(a\in L(H)\), let \(R(a)\) and \(N(a)\) be the range and the kernel of \(a\), respectively. If \(M\) and \(N\) are closed subspaces of \(H\), then the minimal angle between \(M\) and \(N\) is that number on the interval \([0,\pi/2]\) the cosine of which is \(c_0(M,N)=\sup\{ |\langle x,y\rangle|: x\in M\), \(y\in N\), \(\| x\| \leq 1\), \(\| y\| \leq 1\}\). If \(a\in L(H)\) is a selfadjoint operator and \(c_1, c_2\in L(H)\) are positive operators such that \(a=c_1-c_2\) and \(c_0\bigl(\overline{R(c_1)},\overline{R(c_1)}\bigr)<1\), then \(a=c_1-c_2\) is said to be a positive decomposition of \(a\). Note that each selfadjoint operator \(a\) has a unique positive decomposition \(a=a_1-a_2\) satisfying the additional condition \(R(a_1)\perp R(a_2)\). This decomposition is called the positive orthogonal decomposition of \(a\). However, there can be many other non-orthogonal positive decompositions of \(a\). One of the main results of the paper is a characterization of positive decompositions. Let \(a=c_1-c_2\) be a positive decomposition of a selfadjoint operator \(a\in L(H)\). If \(p\in L(H)\) is the oblique projection with range \(\overline{R(c_1)}\) and kernel \(\overline{R(c_2)}\oplus N(a)\), then the reflection \(w=2p^*-1\) is \(a\)-positive, i.e., \(\langle aw x,x\rangle \geq 0\) for all \(x\in H\). On the other hand, if \(w\) is an \(a\)-positive reflection, then \(q=\frac{1}{2}(w+1)\) is an oblique projection and \(a=c_1-c_2\), where \(c_1=aq\), \(c_2=a(q-1)\), is a positive decomposition of \(a\). Operators \(a, b\in L(H)\) are congruent if there exists an invertible operator \(g\in L(H)\) such that \(b=gag^*\). If \(a=a_1-a_2\) is the positive orthogonal decomposition of a selfadjoint operator \(a\), then a selfadjoint operator \(b\) is congruent to \(a\) if and only if there exists a positive decomposition \(b=b_1-b_2\) of \(b\) such that, for \(i=1,2\), \(R(b_i)=u_i\bigl(R(a_i)\bigr)\), where \(u_i\in L(H)\) are unitary operators, and \(\dim N(a)=\dim N(b)\). Let \({\mathcal I}_{a}=\{ g \in L(H):~g~\text{invertible~and}~gag^*=a\}\) be the isotropy group of \(a\in L(H)\). If \(a\) is a selfadjoint operator with closed range and the positive orthogonal decomposition \(a=a_1-a_2\), then \(\{ ga_1 g^{*}-ga_2 g^*:~g\in {\mathcal I}_{a}\}\) is the set of all positive decompositions of \(a\).