Given a bounded self-adjoint operator $T$ on a Hilbert space $H$, the Jordan decomposition asserts that there exists a unique pair of bounded positive operators $(T_+,T_-)$ such that $$ T = T_+ - T_-, \quad \text{and}\quad T_+T_-=0. $$ For that reason $T_+$ and $T_-$ are often called the positive and negative part of $T$, respectlvely. In particular $T\leq T_+$.

Question. Is $T_+$ the smallest positive operator bigger than $T$? In other words, if $S$ is a bounded positive operator on $H$, such that $T\leq S$, does it follow that $T_+\leq S$.


EDIT: The following is true in the case that $T_+$ is invertible.

We have that $T_+-T_-\leq S$. Note that since $T_-T_+=T_+T_-=0$, we have that $T_-p(T_+)=p(T_+)T_-=0$ for any polynomial $p(z)$ with constant term $0$. Now since $T_+$ is invertible, $0\not\in\sigma(T_+)$, so the spectrum of $T_+$ is a closed set contained in an interval $[\varepsilon, M]$, where $0<\varepsilon<M<\infty$. The functions $f_n(t)=t^{1/n}$ defined on $\sigma(T_+)$ are continuous and each one can be uniformly approximated by polynomials with constant term $0$, so we have that $f_n(T_+)T_-=T_-f_n(T_+)=0$ for all $n$. But note that since $f_n\to1$ uniformly on $\sigma(T_+)$ we have that $f_n(T_+)\to\text{id}_H$ in norm. This actually proves that $T_-=0$, and the result follows trivially.

In the case that $T_+$ is not invertible, we cannot say anything: As OP commented, consider for example $T=\pmatrix{1&0\\0&-1}$, then $T_+=\pmatrix{1&0\\0&0}$. Consider the matrix $S=\pmatrix{9&-6\\-6&4}$, then we have that $$S=\pmatrix{3&-2\\0&0}^*\cdot\pmatrix{3&-2\\0&0}\geq0$$ and we have that $$S-T=\pmatrix{8&-6\\-6&5}\geq0$$ since the eigenvalues of $S-T$ are $\frac{13\pm 3\sqrt{17}}{2}$. On the other hand, we have that

$$S-T_+=\pmatrix{8&-6\\-6&4}$$ and the eigenvalues of $S-T_+$ are $6\pm 2\sqrt{10}$, one of which is negative.

| cite | improve this answer | |
  • $\begingroup$ I think your statement that $T_+^{1/n}\to\text{Id}_H$, as $n\to\infty$, does not hold. For example, if $T=\pmatrix {1 & 0 \cr 0 & -1}$, then $$T_+=\pmatrix {1 & 0 \cr 0 & 0} = T_+^{1/n}.$$ $\endgroup$ – Black Oct 17 at 18:19
  • $\begingroup$ @Black you are right, it was a stupid mistake of mine. I have edited my post. $\endgroup$ – JustDroppedIn Oct 17 at 22:50
  • $\begingroup$ Two points: when $T_+$ is invertible then $T_-=0$, so $T$ is already positive and my conjecture becomes tautological. Secondly, your $S$ is not positive so it doesn't give a counter example. $\endgroup$ – Black Oct 18 at 0:28
  • $\begingroup$ your first point is exactly what I am saying. About your second point, I didn't notice you wanted $S$ to be positive. I am editing again. $\endgroup$ – JustDroppedIn Oct 18 at 1:31
  • $\begingroup$ @Black I think we are done here $\endgroup$ – JustDroppedIn Oct 18 at 1:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.