Efficient state vector simulation of mid-circuit measurements

Mid-circuit measurement with feedforward presents interesting challenges for state vector simulation, since the non-unitarity of measurement interrupts the straightforward process of applying a sequence of gates to a state vector. This forces the simulator implementation to augment the simulation beyond simple matrix-vector multiplication. Existing approaches include rerunning the simulation for each shot, using deferred measurement to convert all mid-circuit measurements to terminal measurements, and copying the state vector at each conditional statement encountered in the simulation. Each of these approaches has benefits and drawbacks in terms of various factors, such as memory and runtime.

In this talk, we present an approach to mid-circuit measurement simulation that combines the benefits from different methods. It requires at most one full state vector's worth of memory at any time, executes fewer simulation repetitions than the number of shots when number of measurements is sufficiently small, and is easily parallelizable to multiple compute nodes. Furthermore, the implementation parses OpenQASM directly, allowing developers to make full use of powerful language features, such as subroutines, custom gates, and loops.