I am studying some [edit: bounded] operators that have the property that the partial isometry in the polar decomposition $T = U|T|$ is not only unitary but also self-adjoint.
Is this an already established class of operators? Is there an exact condition when $U$ in the polar decomposition is self-adjoint? (I mean, obviously it gives that the Duggal transform is equal to $T^*$...) How often does this happen? Can the partial isometry in the PD ever be self-adjoint but not unitary?
The operator I am working on is not normal, or even hyponormal.
Thanks,
Derek
The class of operators you're describing is self-adjoint operators in Krein spaces. Your $U$ is nice enough to allow us to introduce some additional structure on our vector space where $T$ behaves nicely.
To sum-up your setting (and put it into a slightly different notation to have a better reference to the existing literature).
We have a Hilbert space ($\mathcal K, (\cdot,\cdot)$) and a self-adjoint operator $A (=|T|)$ as well as an operator $J (=U)$ which fulfills $J^*=J^{-1}=J$. Although it is not a self-adjoint operator, we are interested in some properties of $T=JA$, for instance its spectrum.
We can define an indefinite inner product on $\mathcal K$ by defining $[x,y]=(Jx,y)$. This is not positive definite, but fulfills all other properties of a scalar product. We call $(\mathcal K,[\cdot,\cdot])$ a Krein space.
We have $(Ax,y)=(JJAx,y)=[JAx,y]=[Tx,y],$ hence as $A$ is symmetric w.r.t. $(\cdot,\cdot)$, so is $T$ w.r.t. $[\cdot,\cdot]$. With some work (less if $T$ is bounded), we can show that the $[\cdot,\cdot]$-adjoint of $T$ is equal to $T$, hence $T$ is self-adjoint in the Krein space ($\mathcal K, [\cdot,\cdot]$).
For some literature on Krein spaces, there's this introduction.
Notable books are
T.A. Azizov, I.S. Iokhvidov: Linear Operators in spaces with an indefinite metric (1989)
J. Bognár: Indefinite Inner Product Spaces (1974)
There are actually not too many results on the general class of self-adjoint operators in Krein spaces, but some sub-classes of those, such as definitizable operators are particularily nice and some advancement on those has been made in the recent years. It's worth noting that if $J$ (your $U$) has spectrum which lies only on the positive real line, except for a finite number of negative eigenvalues with finite-dimensional eigenspaces, you are in the nice situation of a Pontrjagin space.