I have a question about the coherent VQLS method. Since CA1, CA2,…CAL is arranged in series in the Hadamard test circuit. Does it mean the training circuit will become extremely high depth if L is large? Will this impact the potential of VQLS to be implemented on the real machines?
You are right, in a VQLS the assumption is that “L is not too large”. Basically one cannot apply a VQLS to a generic matrix A, but only to a matrix A that can be written as a linear combination of a small number of circuits.
Note: even in this case, this is still quite hard to implement on current quantum computers since each CA_j is a controlled-unitary. General controlled-circuits are very hard to implement experimentally.