Circuit Cutting

Circuit cutting addresses a fundamental near-term constraint: current quantum processors have limited qubit counts and connectivity, making it impossible to run many circuits of practical interest directly. The core idea is to identify wires or gates in a circuit where dependencies between subsystems can be broken, replace those connections with local operations and classical communication, and combine the results of multiple smaller executions to reconstruct the output of the original large circuit. Two main variants exist: wire cutting, which severs a qubit wire mid-circuit and reconstructs the channel using sampled mid-circuit states, and gate cutting, which decomposes a two-qubit gate (such as CNOT) into a weighted sum of local operations on each subsystem separately.

The classical overhead of circuit cutting is fundamental rather than an implementation artifact. Cutting k wires or gates requires running O(4^k) subcircuit executions and combining them with a classical post-processing step that applies signed weights (a quasiprobability decomposition). For a single wire cut, the overhead is a factor of 16 in the number of shots required to achieve the same statistical precision as an uncut circuit. This overhead grows exponentially with k, which limits the practical number of cuts to roughly 2 to 4 before the classical sampling cost becomes prohibitive. The technique is therefore best suited to circuits that are almost separable, with a small number of cross-partition entangling operations.

IBM’s Qiskit provides a circuit-cutting toolbox (part of the Qiskit Addon suite) that automates the decomposition, subcircuit execution, and reconstruction steps. The toolbox accepts a target circuit and a cut specification, compiles the subcircuits, distributes them across one or more smaller quantum processing units, and applies the quasiprobability recombination classically. Practical use cases include running a 100-qubit circuit across two 50-qubit processors with a small number of inter-processor cuts, and noise characterization workflows where isolated subcircuits are easier to validate. The technique is also used in distributed quantum computing research as a software analog of quantum interconnects.

The central trade-off in circuit cutting is shots versus qubits: cutting allows larger logical circuits to run on smaller physical hardware, but requires exponentially more shots to compensate for the classical post-processing overhead. For noise benchmarking or proof-of-concept studies at small scale this is often acceptable, but for computations intended to demonstrate quantum advantage the overhead typically erases any quantum speedup. Research directions include finding tighter overhead bounds for structured circuits, leveraging entanglement forging for specific problem symmetries, and combining cutting with error mitigation to control both shot overhead and hardware noise simultaneously.