The Signal
Three Researchers Derive a Unique Quantum Bayes Rule
A paper posted to arXiv on October 29 proves that four consistency requirements uniquely select the Petz transpose map for quantum retrodiction. The result applies to finite-dimensional channels and leaves a family of rotated maps when one
Quantum.world Team · Written with AI · fact-checked ·
A paper posted to arXiv on October 29 argues that quantum retrodiction has a uniquely determined rule—if it meets four specified consistency conditions. Jacopo Surace, Chenxu Li and Francesco Buscemi show that those conditions select the Petz transpose map, a quantum counterpart to Bayes’ rule for reasoning backward from an observed output to a prior state.
The result addresses a basic difference between classical and quantum inference. A classical prior and a stochastic process define a joint probability distribution that can be conditioned in either direction. In quantum theory, a state and a channel do not provide an equally unambiguous joint-state description, leaving open what should count as a principled reverse rule.
“Our result thus explains why the Petz transpose map arises in seemingly different areas of quantum theory: it is the unique quantum analogue of Bayes’ rule compatible with these consistency requirements.”
Jacopo Surace, Chenxu Li and Francesco Buscemi
Four conditions constrain the reverse channel
The authors start with a prior state π and a quantum channel Φ that sends it to an output state σ. Retrodiction must map σ back to π, but it need not undo Φ on every possible input state. The reverse channel depends on the chosen prior as well as on the forward process.
They require that reversing twice restores the original channel; that reversal respects sequential composition, inverting the order of successive processes; and that it respects parallel composition, leaving an independent process independent. The fourth condition requires ordinary unitary inversion when the forward process is unitary. The theorem covers channels between finite-dimensional full matrix algebras, with systems restricted to the supports of the reference states when those states are not full-rank.
Under those conditions, the reverse map is the Petz assignment: it uses the adjoint of the forward channel, weighted on either side by the square roots of the prior, and by inverse square roots of the predicted output. The authors write the map as R(X) = √π Φ†(σ⁻¹ᐟ² X σ⁻¹ᐟ²) √π. The choice of prior remains part of the inference problem; the theorem says the four conditions leave no further freedom in the rule.
Drop involutivity, and rotated maps remain
The paper identifies the role of the double-reversal condition. Without involutivity, composition and unitary inversion still constrain the answer, but allow a family of doubly rotated Petz maps. Those rotations cannot be chosen independently: they must remain consistent under unitary changes of basis and when independent systems are combined.
If the authors also require continuity in the prior, the allowed family narrows to rotated Petz assignments indexed by one common real parameter. Involutivity then selects the unrotated map. The paper therefore distinguishes the conditions that enforce operational consistency from the added symmetry that removes the remaining alternatives.
Measurements turn retrodiction into state inference
For a measurement described by effects Eₓ and outcome probability pₓ = Tr(πEₓ), the rule assigns the outcome a conditional input state √π Eₓ √π / pₓ, for outcomes with nonzero probability. Averaging those states using the outcome probabilities recovers the prior. When the prior and measurement effects are diagonal in the same basis, this update reduces to ordinary Bayesian conditioning.
The conditional state describes an inference about the system before measurement, not necessarily the physical state left behind by a particular measurement instrument. Different instruments can implement the same measurement statistics while producing different post-measurement states. The theorem’s scope is also bounded: the authors identify extensions to general infinite-dimensional C*-algebras and von Neumann algebras as open questions.