Towards the efficient simulation of near-term quantum devices using tensor network states, we introduce an improved real-space parallelizable matrix-product state (MPS) compression method. This method enables efficient compression of all virtual bonds in constant time, irrespective of the system size, with controlled accuracy, while it maintains the stability of the wavefunction norm without necessitating sequential renormalization procedures. In addition, we introduce a parallel regauging technique to partially restore the deviated canonical form, thereby improving the accuracy of the simulation in subsequent steps. We further apply this method to simulate unitary quantum dynamics and introduce a parallel time-evolving block-decimation (pTEBD) algorithm. We employ the pTEBD algorithm for extensive simulations of typical one- and two-dimensional quantum circuits, involving over 1000 qubits. The obtained numerical results unequivocally demonstrate that the pTEBD algorithm achieves the same level of simulation precision as the current state-of-the-art MPS algorithm but in polynomially shorter time, exhibiting nearly perfect weak scaling performance on a modern supercomputer.
Comment: 19 pages, 11 figures