Classical simulability of Clifford+T circuits with Clifford-augmented matrix product states

Abstract: Generic quantum circuits typically require exponential resources for classical simulation, yet understanding the limits of classical simulability remains a fundamental question. In this work, we investigate the classical simulability of $N$-qubit Clifford circuits doped with $t$ number of $T$-gates by converting the circuits into Clifford-augmented matrix product states (CAMPS). We develop a simple disentangling algorithm to reduce the entanglement of the MPS component in CAMPS using control-Pauli gates, which replaces the standard algorithm relying on heuristic optimization when $t\lesssim N$, ensuring that the entanglement of the MPS component of CAMPS does not increase for $N$ specific $T$-gates. Using a simplified model, we explore in what cases these $N$ $T$-gates happen sufficiently early in the circuit to make classical simulatability of $t$-doped circuits out to $t=N$ possible. We give evidence that in one-dimension where the $T$-gates are uniformly distributed over the qubits and in higher spatial dimensions where the $T$-gates are deep enough we generically expect polynomial or quasi-polynomial simulations when $t \leq N$. We further explore the representability of CAMPS in the regime of $t>N$, uncovering a non-trivial dependence of the MPS entanglement on the distribution of $T$-gates. While it is polynomially efficient to evaluate the expectation of Pauli observable or the quantum magic in CAMPS, we propose algorithms for sampling, probability and amplitude estimation of bitstrings, and evaluation of entanglement R\'enyi entropy from CAMPS, which, though still having exponential complexity, improve efficiency over the standard MPS simulations. This work establishes a versatile framework based on CAMPS for understanding classical simulatability of $t$-doped circuits and exploring the interplay between quantum entanglement and quantum magic on quantum systems.