arXiv:2609.20729
Blind Quantum Computation with a Small Quantum Server
Daniel Lovsted, Filipa C. R. Peres, Joshua Nevin, Selman Ipek, Anne Broadbent
Abstract
Blind quantum computation (BQC) allows low-resource clients to securely delegate computations to a quantum server, but server resource costs scale with the computation size, posing a bottleneck for implementations. By leveraging Pauli-based computation (PBC), we achieve BQC with a server whose size depends only on the non-Clifford gate count. Our protocol inherits fault tolerance and qubit virtualization from PBC and reveals that classical-client BQC is possible even in the absence of classical simulability. Finally, we present an entanglement-based dual protocol that performs a resource state computation with a dramatically reduced execution cost.
AI Summary
The paper asks how much server-side quantum memory is needed for information-theoretically secure delegation with classical inputs and outputs. It introduces Blind Pauli-based Computation (Blind PBC), which hides an input $\mathbf{x}$ while sampling the output of a public Clifford+$T$ circuit $C$. For $n$ circuit qubits and $t$ $T$ gates, the server uses $t$ qubits. The client prepares $t$ single-qubit states from four possibilities and sends them in one quantum-communication round, followed by adaptive classical interaction. The comparison protocol, quantum computing on encrypted data, requires $n+t$ server qubits with one quantum transmission, or $n+1$ with $t$ just-in-time transmissions. Blind PBC encrypts the initial resource states and modifies the classical selection of Pauli measurements. Correctness follows from propagating input Pauli operators through the circuit and tracking resource-state rotations and final output corrections. A simulator-based argument establishes perfect input privacy against a deviating server. The protocol inherits Pauli-measurement fault-tolerant implementation and qubit virtualization; virtualizing $k$ qubits incurs costs exponential in $k$, including client and communication costs. For a circuit subclass including certain IQP circuits, the protocol needs only a classical client and final classical output correction; classical sampling hardness is conjectural. An entanglement-based, time-reordered construction additionally prepares a circuit-specific, input-independent $t$-qubit resource state, executable using adaptive single-qubit $X/Y$ measurements. The guarantees concern input hiding, not circuit hiding or verification. The server-memory advantage is regime-dependent, and the resource-state construction reduces execution resources by moving quantum work into preparation.
Error found: The stated preparation interchanges the phase-mask and Z-mask bits relative to Algorithm 1, invalidating its matching correctness argument and making the auxiliary protocol's claimed perfect security false.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.