Inter-branch message transfer on superconducting quantum processors: a multi-architecture benchmark

TL;DR

The paper tackles quantifying inter-branch message transfer in a Wigner's-friend circuit as a practical benchmark for near-term superconducting processors. It presents a reproducible Qiskit implementation of Violaris' unitary message-transfer primitive, scaling to $n=32$ across four IBM backends and evaluating three message families with multiple diagnostics, all without error mitigation. Sparse messages reveal depth-controlled, device-noise-limited performance with $p_{\mathrm{all}}$ spanning roughly $0.07$ to $0.68$, while half and dense messages suffer from routing overhead and significant transpiler-seed variability; amplitude and divergence tests further show non-amplification of branch weight and performance degradation with branch-conditioned complexity. The work delivers a hardware-centered, openly available benchmark suite to evaluate near-term superconducting devices and guide mitigation strategies for inter-branch information transfer.

Abstract

We treat inter-branch message transfer in a Wigner's-friend circuit as a practical benchmark for near-term superconducting quantum processors. Implementing Violaris' unitary message-transfer primitive, we compare performance across IBM Eagle, Nighthawk, and Heron (r2/r3) processors for message sizes up to $n=32$, without error mitigation. We study three message families -- sparse (one-hot), half-weight, and dense -- and measure conditional string success $p_{\mathrm{all}}=\Pr(P=μ\mid R=0)$, memory erasure after uncomputation, and correlation diagnostics (branch contrast and bitwise mutual information). The sparse family compiles to essentially constant two-qubit depth, yielding a depth-controlled probe of device noise: at $n=32$ we observe $p_{\mathrm{all}}$ spanning $\approx0.07$ to $\approx0.68$ across backends. In contrast, half and dense messages incur rapidly growing routing overhead, and transpiler-seed variability becomes a practical limitation near the coherence frontier. We further report an amplitude sweep (no-amplification test) and a divergence ``cousins'' sweep that quantifies degradation with branch-conditioned complexity. All data and figure-generation scripts are released.

Figures (9)

• Figure 1: Pilot benchmark for string-level transfer success at $n=1$ for the protocol and controls.
• Figure 2: Pilot benchmark for memory erasure at $n=1$ for the protocol and controls.
• Figure 3: Scaling of string transfer success for sparse messages. The compiled two-qubit depth is essentially constant, so performance differences largely reflect device-level noise.
• Figure 4: Scaling of string transfer success for half-weight messages.
• Figure 5: Scaling of string transfer success for dense messages.