January 1698

The quantum no-broadcasting theorem states that it is impossible to produce perfect copies of an arbitrary quantum state, even if the copies are allowed to be correlated. Here we show that, although quantum broadcasting cannot be achieved by any physical process, it can be achieved by a virtual process, described by a Hermitian-preserving trace-preserving map. This virtual process is canonical: it is the only map that broadcasts all quantum states, is covariant under unitary evolution, is invariant under permutations of the copies, and reduces to the classical broadcasting map when subjected to decoherence. We show that the optimal physical approximation to the canonical broadcasting map is the optimal universal quantum cloning, and we also show that virtual broadcasting can be achieved by a virtual measure-and-prepare protocol, where a virtual measurement is performed, and, depending on the outcomes, two copies of a virtual quantum state are generated. Finally, we use canonical virtual broadcasting to prove a uniqueness result for quantum states over time.

Deep neural networks are a powerful tool for predicting properties of quantum states from limited measurement data. Here we develop a network model that can simultaneously predict multiple quantum properties, including not only expectation values of quantum observables, but also general nonlinear functions of the quantum state, like entanglement entropies and many-body topological invariants. Remarkably, we find that a model trained on a given set of properties can also discover new properties outside that set. Multi-purpose training also enables the model to infer global properties of many-body quantum systems from local measurements, to classify symmetry protected topological phases of matter, and to discover unknown boundaries between different phases.