Why finite quantum circuits cannot reproduce the quantum switch

When order itself is a quantum resource

What if a quantum device could send a system through two unknown operations in an order that is not fixed in advance? The quantum switch makes this question precise. It uses a control qubit to coherently select between two sequences: one control state applies channel A and then B, while the other applies B and then A. Crucially, the device preserves coherence between these alternatives, so the output can retain interference between the two orders.

That is different from an ordinary circuit that chooses an order using classical information. Such a circuit can still use quantum memory and processing, and it may adapt later choices to earlier measurement outcomes. But each run has a definite sequence of calls; its order is not held in a coherent superposition. Could repeated queries to the same unknown channels make up for that difference?

In “Simulating the quantum switch with quantum circuits is impossible,” Caisheng Cheng proves that the answer is no under a specific, demanding standard: exact, deterministic, universal simulation. For every fixed target dimension of at least two, no quantum circuit with classical control of causal order can reproduce the complete two-channel switch for all possible input channels using any finite number of calls to each. The claim already holds for qubit channels. The proposed simulator must work without knowing which channels it will receive, and must match the whole output channel, including its action on inputs entangled with another system.

Why repeated calls do not help

The proof turns repeated access into a contradiction. Repeated calls might seem to give a circuit additional freedom: its output can depend on the unknown channels many times over. Cheng asks what happens if every call except one to A and one to B is replaced by a known identity operation. Averaging over which pair of calls is retained produces a one-call-per-channel circuit related to the target switch. The relation follows by probing the assumed simulator with mixtures of an unknown channel and the identity, then comparing the terms that depend on both unknown channels.

For qubit inputs, the argument then separates the part of each circuit history that carries the switched action from other contributions. The control’s two diagonal sectors distinguish the two orders; comparing them isolates the switched component. But physical circuit order imposes a normalization constraint: after tracing out the final output, the last queried channel’s output must satisfy the causal trace-preservation condition. Cheng constructs constraints that cancel the residual contributions while retaining the switched component. The simulation identity requires this component to have weight one divided by the product of the two call counts, which is positive for every finite pair of counts. The causal constraints require the same quantity to vanish. No finite counts can satisfy both.

The result also covers larger target systems: the paper embeds qubit channels into the larger query space while preserving the number of calls, extending the obstruction to every fixed dimension at least two. It then treats the complete r-switch, which coherently controls every permutation of r input channels. By choosing two control branches that differ only by exchanging a pair of channels, and setting the others to identity operations, one can recover the two-channel switch. A finite deterministic simulator for the larger operation would therefore imply a forbidden finite simulator for that pair.

What the impossibility does—and does not—say

The boundary matters. This is an impossibility result for exact deterministic simulation of arbitrary channels, not a claim that every approximate or restricted-input simulation is impossible. The paper contrasts its setting with known possibilities for unitary inputs and for protocols that use postselection, where unsuccessful runs are discarded. Nor does the theorem say that a switch cannot be implemented with a physical device that directly provides coherent control of order: that device is the operation being simulated. This is a 2026 arXiv preprint, and its author describes this first version as a presentation of the main results, with a fuller manuscript in development. The contribution is a sharp limit on what repeated, causally ordered calls alone can reproduce.