When one quantum prover can replace many

A network’s computers may each know only part of the information needed to check a claim about the whole system. An outside supplier could help by sending evidence to each computer. But if that supplier cannot be trusted, how should the network check its evidence—and does having several separate suppliers offer an advantage?

In the preprint “Reducing Provers in Distributed Quantum Verification,” Alex Della Schiava, Atsuya Hasegawa, François Le Gall and Harumichi Nishimura examine this question mathematically. Posted to arXiv and submitted to SIROCCO 2027, the work presents theoretical protocols and proofs, not experiments on quantum hardware.

Why combining quantum evidence is difficult

The suppliers, called provers, send quantum states as evidence. The network’s computers, or nodes, then perform local operations and exchange quantum messages with their neighbors. The claim is accepted only if every node accepts.

Two probabilities matter. Completeness measures how reliably suitable evidence gets a true claim accepted. Soundness bounds the chance that any allowed evidence gets a false claim accepted. Perfect completeness means that an honest proof of a true claim succeeds with certainty.

The authors impose two important restrictions: different provers’ proofs must be unentangled, and each prover’s proof must also be unentangled across network nodes. Entanglement is a quantum connection that prevents a joint state from being described as a mixture of separate states for its parts. These restrictions are assumptions of the model, not capabilities demonstrated in a laboratory.

Classical proofs from several suppliers can simply be bundled together. Quantum proofs are trickier: one supplier could entangle the bundled components, creating cheating strategies unavailable to separate, unentangled suppliers.

From many suppliers to two

The first construction asks two provers each to provide a complete set of the original proofs. This doubles the total evidence received by each node, compared with the original total across all provers.

Each node performs quantum comparisons called SWAP tests on corresponding registers from the two suppliers. Together, these comparisons form a product test that constrains cheating through entanglement within either supplier’s block. The nodes then run the original verification using one block.

A node whose comparison fails still participates, but ultimately rejects. Because acceptance requires every node’s approval, the nodes need not exchange test results or coordinate a shared random choice between testing and verification. The construction adds no communication rounds.

It does weaken the protection against false acceptance. If the original soundness bound is s, the new bound is (6 + s)/7, while completeness is preserved. The theorem requires original completeness to exceed that new bound; it is not an unrestricted conversion of every protocol’s error guarantees.

Using the network to reach one

For a single prover, the authors exploit the assumed absence of entanglement between nodes. They divide nodes into two colors and arrange for each node to receive one proof block directly and another through a nearby supplier node of the opposite color.

The routing structure uses pairs and triples of nodes. Every supplier is at most two network steps from its recipient and serves at most two recipients. Nodes check the supplied routing labels locally.

Grouping registers by the colors of the nodes to which the prover initially supplies them—including supplier nodes holding registers for other recipients—produces two unentangled blocks, allowing the product-test construction to work with just one prover.

For a network of N nodes and an original maximum local proof size of p qubits, the resulting proof uses at most 3p plus an additional term growing logarithmically with N. Routing adds four synchronous communication rounds. These are round-count guarantees, not claims that message volume, hardware demands or elapsed execution time stay unchanged.

Making false acceptance unlikely

The authors also show how repetition reduces error when completeness is perfect. Repetitions run in parallel, and each node accepts only if every run succeeds. Proof size grows with the number of repetitions, but the four-round overhead stays fixed.

Proving this works is nontrivial: a dishonest prover may entangle evidence across repetitions. The authors bound the verification’s acceptance behavior using a mathematical object that preserves the needed error bound under repetition.

For any fixed soundness target below one, this establishes that one prover can match multiple provers with a constant-factor proof-size increase, an additive logarithmic term and four extra rounds. Extensions without perfect completeness require stronger conditions on initial errors and network size.

The scope remains narrow. Nodes have unrestricted local computation and ideal quantum operations, networks have no isolated nodes, and the main result depends on initially node-separable proofs. Allowing entanglement across nodes remains an open direction. The authors also disclose AI assistance in developing parts of the arguments and manuscript. The contribution is a resource trade-off within a formal model—not a demonstrated quantum-network technology or a resolution of the broader single-versus-multiple-prover question.