The Signal
Liu, Bai and Scarani Unify Quantum Smoothing Theories
Mingxuan Liu, Ge Bai and Valerio Scarani present a framework that treats competing quantum state-smoothing approaches as different choices of prior correlations. Their paper, published in npj Quantum Information, uses extended Bayesian retr
Quantum.world Team · Written with AI · fact-checked ·
Two quantum state-smoothing approaches that have seemed incompatible can be expressed as special cases of one retrodictive framework. In a paper published on November 1 in npj Quantum Information, Mingxuan Liu, Ge Bai and Valerio Scarani connect the competing methods to different assumptions about prior correlations. Their approach uses extended Bayesian retrodiction to estimate an open quantum system’s state from measurements made both before and after the time of interest. It also gives researchers a way to describe other choices of prior, beyond the established approaches.
Future measurements reshape the estimate
Filtering estimates a system’s state using measurements recorded up to a chosen time. Smoothing adds later measurements to that record, using information gathered afterward to improve an estimate of the earlier state. The method does not alter the system’s past; it changes what an observer can infer about it.
That inference is less straightforward in quantum mechanics than in classical statistics. Quantum measurements can be incompatible, and retrodiction—the process of reasoning backward from later information—depends on what correlations the observer includes in the prior. The authors make those correlations explicit by introducing auxiliary systems into the prior belief, representing correlations formed during preparation and subsequent evolution.
Prior correlations explain the split
The paper’s central move is to treat prior correlations not as a technical detail, but as part of the definition of the estimate. Under this account, earlier quantum smoothing proposals correspond to different ways of extending the prior. The framework therefore identifies a common structure without claiming that every choice produces the same smoothed state.
The authors write: “Our results establish quantum state smoothing as a fundamentally retrodictive process just like classical smoothing, with proper quantum features clearly identified.”
“Our results establish quantum state smoothing as a fundamentally retrodictive process just like classical smoothing, with proper quantum features clearly identified.”
Mingxuan Liu, Ge Bai and Valerio Scarani, “Unifying quantum smoothing theories with extended retrodiction”
Earlier formulations by Guevara and Wiseman, and by Petz and Fuchs, appear as particular cases in the broader account. By identifying what differs between them—the prior correlations—the authors give researchers a shared way to compare methods that previously began from distinct assumptions.
Average entropy compares the choices
The paper also characterizes different prior choices using the average entropy of the resulting smoothed states. Entropy provides an information-theoretic measure for comparing those estimates across measurement outcomes. The authors derive upper and lower bounds for the average entropy, linking the range to specific smoothing constructions.
That comparison does not supply a single ranking that holds for every later measurement. In the preprint version, the authors state that no future-measurement-independent measure can universally order all the candidate smoothed states. Which estimate is most useful therefore depends in part on the measurement record and the correlations an observer has chosen to represent.
A shared theory, not a new measurement device
The result is a theoretical account of state estimation in monitored open systems, not a report of a new apparatus or experimental demonstration. Its immediate contribution is conceptual: it places established smoothing theories inside one formalism and shows how to extend that formalism to other priors.
For experiments, the choice still has to be grounded in the system being studied. Researchers must specify which correlations formed during preparation and evolution belong in the prior; the framework makes those assumptions visible, but does not choose them for a particular setup.