IQC Student Seminar Featuring Matthew Duschenes

Quantum computing devices require exceptional control of their experimental parameters to prepare quantum states and simulate other quantum systems, in particular while subject to noise. Of interest here are notions of trainability, how difficult is it to classically optimize parameterized, realistic quantum systems to represent target states or operators of interest, and expressivity, how much of a desired set of these targets is our parameterized ansatze even capable of representing? We observe that overparameterization phenomena, where systems are adequately parameterized, are resilient in noisy settings at short times and optimization can converge exponentially with circuit depth. However fidelities decay to zero past a critical depth due to accumulation of either quantum or classical noise. To help explain these noise-induced phenomena, we introduce the notion of expressivity of non-unitary, trace preserving operations, and highlight differences in average behaviours of unitary versus non-unitary ensembles. We rigorously prove that highly-expressive noisy quantum circuits will suffer from barren plateaus, thus generalizing reasons behind noise-induced phenomena. Our results demonstrate that appropriately parameterized ansatze can mitigate entropic effects from their environment, and care must be taken when selecting ansatze of channels.