Stability relations for Hilbert space operators and a problem of Kaplansky

In his monograph on Infinite Abelian Groups, I. Kaplansky raised three ``test problems" concerning their structure and multiplicity. As noted by Azoff, these problems make sense for any category admitting a direct sum operation. Here, we are interested in the operator theoretic version of Kaplansky's second problem which asks: if $A$ and $B$ are operators on an infinite-dimensional, separable Hilbert space and $A \oplus A$ is equivalent to $B \oplus B$ in some (precise) sense, is $A$ equivalent to $B$? We examine this problem under a strengthening of the hypothesis, where a ``primitive" square root $J_2(A)$ of $A\oplus A$ is assumed to be equivalent to the corresponding square root $J_2(B)$ of $B \oplus B$. When ``equivalence" refers to similarity of operators and $A$ is a compact operator, we deduce from this stronger hypothesis that $A$ and $B$ are similar. We exhibit a counterexample (due to J. Bell) of this phenomenon in the setting of unital rings. Also, we exhibit an uncountable family $\{ U_\alpha\}_{\alpha \in \Omega}$ of unitary operators, no two of which are unitarily equivalent, such that each $U_\alpha$ is unitarily equivalent to $J_n(U_\alpha)$, a ``primitive" $n^{th}$ root of $U_\alpha \oplus U_\alpha \oplus \cdots \oplus U_\alpha$.