Researchers Simulate 70-Qubit Experiment in 37.3 Minutes

Hidetaka Manabe of the Singapore University of Technology and Design and colleagues have, for the first time, classically simulated IBM’s 70-qubit doped Clifford random circuit sampling experiment using a deterministic tensor network approach. Previously, simulating such a complex quantum circuit required substantial computational resources, but the team completed all 2051 amplitude batches in 37.3 minutes using 32 nodes. The largest intermediate tensor required only 235 complex64 entries, a times smaller than IBM’s initial estimation.

A more efficient classical simulation technique for complex quantum computations has been achieved, successfully modelling IBM’s 70-qubit experiment. This method required considerably fewer computational resources than previously thought, completing the simulation in under an hour using 32 processing nodes. The advance provides a key verification tool for quantum computer outputs and enables the design of future quantum experiments by offering a means to quantitatively assess their complexity.

The team achieved an advancement in simulating complex quantum computations by successfully modelling IBM’s 70-qubit experiment with a new classical approach. They employed a tensor network technique, which can be visualised as a flow chart representing the quantum calculation as a network of interconnected nodes, allowing for a more efficient organisation of the data and calculations. This work provides a valuable set of tools for verifying the outputs of actual quantum computers and designing future experiments.

Efficient tensor network simulation validates 70-qubit quantum circuit fidelity estimates

The largest intermediate tensor required only 235 complex64 entries, a reduction of 256 times compared to IBM’s initial estimation, and was distributed across eight GPUs per node. This decrease in tensor size enabled the complete simulation of IBM’s 70-qubit experiment in 37.3 minutes using 32 nodes equipped with NVIDIA H100 GPUs. Previously, such a simulation would have been computationally prohibitive due to the exponential growth of required memory. The deterministic temporal-boundary tensor network approach efficiently handles open-boundary circuits, circumventing limitations of methods requiring closed systems and allowing for streamlined calculations.

A log-XEB estimate of 0.35034, with a 95 percent interval of 2.29763 to 0.40305, shows numerical compatibility with IBM’s independently reported fidelity lower bound for their 70-qubit experiment. Furthermore, the approach accurately models open-boundary one-dimensional circuits, unlike some techniques limited to closed systems, and allows for analysis of how factors like entangling-gate bond dimension and circuit depth affect computational cost. However, scalability to circuits with substantially more complex connectivity or non-uniform gate arrangements remains a continuing challenge for classical simulation.

Validating quantum computations through large-scale circuit emulation

Classical simulation of a 70-qubit quantum circuit offers a powerful new method for validating results from actual quantum hardware, and for designing more effective experiments. The technique relies on a specific type of quantum circuit, one with limited connectivity and a particular structure of entangling gates, and scaling it to circuits with more complex arrangements presents a significant hurdle. Despite these limitations regarding circuit complexity, this demonstration of classical simulation remains striking, providing an independent verification method for quantum computations.

Validating quantum results is vital, as errors can occur in building and operating these delicate systems, and this technique offers a pathway to improve experimental design and assess the reliability of quantum processors. The researchers NVIDIA demonstrated a classical simulation of a complex 70-qubit quantum circuit using 32 nodes with eight GPUs each, completing the task in just over thirty-seven minutes.

Employing a ‘temporal-boundary’ tensor network, a method of organising complex calculations as interconnected nodes, allowed for efficient handling of the quantum circuit’s structure, specifically its open-ended design. This technique significantly reduced the computational burden; the largest data structure required to be contained 2 35 entries, a substantial decrease compared to previous estimates.

The researchers successfully classically simulated a 70-qubit quantum circuit, verifying its output probabilities and achieving a log-XEB estimate of 0.35034. This simulation provides an independent check on results obtained from quantum hardware and helps to refine experimental design. Their tensor network approach required a tensor containing 235 complex entries, substantially less data than previously anticipated for such a calculation. The method is currently limited to circuits with specific connectivity, but offers a valuable tool for validating quantum computations as they become more complex.