arXiv:2610.33294
Twirled and swirled worlds: a compositional approach
Sina Soltani, Thomas D. Galley, Marco Erba, John H. Selby
Abstract
Twirled and swirled worlds describe generalised probabilistic theories (GPTs) restricted by symmetry-imposed superselection rules. We extend these constructions to a fully compositional setting for finite-dimensional GPTs and compact groups, treating states, effects, and multipartite transformations uniformly as processes. We distinguish system actions from process-space actions. We define actions on a theory as a whole by requiring the corresponding process-space actions to respect sequential and parallel composition. Averaging physical theory actions defines twirling, while averaging nonphysical theory actions defines swirling. We show that invariant subtheories might not satisfy process tomography, and establish conditions under which they do, involving invariant dynamically faithful states or internal classical reference frames. For locally tomographic parent theories, we prove that every theory symmetry admits a representation by linear system actions and that the invariant subtheory satisfies process tomography. An invariant multipartite process space fails local tomography if and only if some bipartition has sides sharing a non-trivial irreducible representation. We characterise when ancillary systems are necessary, which representations they must carry, and when entanglement is required for ancilla-free process tomography. We develop a framework for unifying twirled subtheories and extend incoherent simulations to compact groups. Finally, we recover bilocal classical theory as a twirled subtheory of classical theory and real quantum theory as a swirled subtheory of complex quantum theory, deriving their tomographic properties from their representation structure.
AI Summary
The paper asks how symmetry-restricted generalised probabilistic theories (GPTs) can be defined consistently for arbitrary multipartite processes, and which forms of tomography survive the restriction. It treats finite-dimensional GPTs and compact groups, distinguishing actions on systems, process spaces, and entire theories. Theory actions must preserve sequential and parallel composition; Haar averaging of physical actions is called twirling, and averaging of nonphysical actions swirling. The resulting invariant processes form compositionally closed subtheories. Invariant states and effects always distinguish one another, but this argument does not automatically establish process tomography. Sufficient conditions include invariant dynamically faithful states—or faithful vectors in the span of invariant states—and internal classical reference frames with suitable control. For locally tomographic parent GPTs, every theory action admits a linear system-level representation, and every invariant subtheory satisfies process tomography. Representation theory characterises failure of local tomography: it occurs precisely when some bipartition has sides sharing a nontrivial irreducible representation. Related criteria identify ancillary representation requirements and when entanglement is necessary for ancilla-free process tomography. The framework also unifies different system-induced twirled worlds, extends incoherent simulations to compact groups, and recovers bilocal classical theory and real quantum theory as symmetry restrictions. The proofs use compositional identities, Haar averaging, and invariant-sector decompositions. Beyond locally tomographic parents, necessary and sufficient conditions for process tomography remain open; no explicit counterexample exhibiting its failure is supplied. Continuous classical reference frames require infinite-dimensional extensions whose detailed construction is deferred.
Central claim may not hold: The abstract presents possible failure of process tomography as an established result, whereas the body establishes sufficient conditions for tomography and leaves whether failure actually occurs as an open question.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.