The Signal
CV-DIPE Uses Homodyne Data to Compare Quantum States
A paper describes a protocol for estimating the inner product of continuous-variable quantum states using local homodyne measurements at shared phase settings. Its sufficient bounds show improved accuracy scaling for Gaussian states, while
Quantum.world Team · Written with AI · fact-checked ·
“Our results enable independent CV platforms to benchmark their state preparations against one another, even when neither state admits a tractable classical description.”
Jun Tang, Congcong Zheng, Xutao Yu and Zaichen Zhang, the paper’s authors
Local measurements, shared phase settings
A paper posted to arXiv on October 29 sets out a way for two separate quantum platforms to estimate how similar their continuous-variable states are without bringing the states together or performing a joint measurement. The protocol, called continuous-variable distributed inner-product estimation, or CV-DIPE, relies on local homodyne measurements at shared phase settings, followed by classical post-processing.
The quantity being estimated is Tr(ρσ), the inner product of the two states’ density operators. The authors target cases where neither platform has a tractable classical description of its state, making direct comparison difficult.
Each platform measures copies of its own state using the same sequence of randomly selected phase settings. The measurement records are then combined classically. A common phase reference and consistent quadrature normalization are assumptions of the scheme: the same setting must point along the same phase-space directions on both devices.
Gaussian states have improved sufficient bounds
CV systems have an infinite-dimensional state space, so the protocol truncates the phase-space integral used to calculate the inner product. That creates two sources of error: the omitted tail outside the chosen radius, and statistical noise from the finite measurement sample. The truncation radius controls the balance between them.
Under a known upper bound E on mean energy per mode, the authors derive a sufficient number of phase settings that, for fixed mode count M and energy bound, scales as O(ε−(2M+2) log(2/δ)). Here ε is the target additive error and 1−δ the confidence level. For Gaussian states under the same energy constraint, their bound instead scales as O(ε−2[log(2M/ε)]2M log(2/δ)). Stronger Gaussian tail control accounts for that difference; the improvement does not remove the mode-count and energy dependence.
Those expressions bound phase settings, not the total number of copies independently of the number of measurements per setting. With n shots per state at each setting, the paper counts 2nNθ copies across both platforms. The authors also derive a Bernstein bound that incorporates n and can reduce the sufficient setting count; their simpler Hoeffding-based bounds do not reflect that finite-shot dependence.
Simulations show the cost of adding modes
The paper tests the estimator numerically on Gaussian and non-Gaussian examples, including coherent, squeezed, Fock-sector and even-cat states. In one set of simulations, the mean-squared error increases with mode number across all four families. At four modes, the reported error is about 3.2×10−3 for coherent states, around 10−2 for Fock states, and on the order of 10−1 for squeezed and even-cat states.
Those results describe simulations, not measurements on operating quantum hardware. They also show why the paper’s efficiency claim needs qualification: its formal guarantees depend strongly on state assumptions, and the authors say their bounds can be conservative. In their tests, empirically chosen truncation radii sometimes produced a better bias–variance balance than the sufficient radii from the theoretical bounds.
Noise and drift remain outside the analysis
CV-DIPE offers a route to compare independently prepared multimode states without reconstructing either state in full or requiring a joint measurement across platforms. The trade-off is that both systems must coordinate their phase settings, and the copy requirements can grow steeply with mode count for general energy-constrained states.
The authors identify analysis of experimental noise and drift, along with reliable data-driven selection of the truncation radius, as future work. Until those effects are incorporated, the result is a theoretical protocol supported by numerical simulations—not an experimental demonstration of lower comparison costs.