Wigner's-Friend Scenarios

Wigner’s-friend scenarios examine the joint use of quantum descriptions, recorded outcomes, and reasoning across agents. Their research value comes from explicit incompatibility theorems and operational protocols, not from claims that laboratory interference has disproved ordinary facts. Reviewed 11 October 2026, the evidence horizon includes Bong and colleagues’ 2020 local-friendliness theorem and Baumann and Brukner’s 2024 memory analysis. The former’s publicly linked code supplies a reproducible baseline.

Helpful background. Wigner’s friend introduces the setup; the Frauchiger–Renner argument explains a different reasoning chain. Their assumptions must not be combined silently.

The operational content of a friend

A reversible premeasurement can correlate a system with a memory:

$$\frac{\lvert0\rangle+\lvert1\rangle}{\sqrt2}\lvert R_*\rangle \longmapsto \frac{\lvert0,R_0\rangle+\lvert1,R_1\rangle}{\sqrt2}.$$

A superobserver who controls the full laboratory can, in principle, measure interference between these branches. Whether an actual memory qualifies as an observed outcome for a given theorem is a separate modeling premise. A single photonic degree of freedom can implement the quantum circuit without settling the philosophy of macroscopic observers.

The key experimental tradeoff is that a stable external record leaks information about the branch, while interference requires suitable coherence and control. The exact tradeoff depends on the memory, environment, and operations being allowed.

How much external record can interference tolerate?

Represent the controlled laboratory branches by orthogonal states $\lvert L_0\rangle,\lvert L_1\rangle$, and an inaccessible external record by normalized pure states $\lvert e_0\rangle,\lvert e_1\rangle$. Consider $\lvert\Psi\rangle= \frac{\lvert L_0\rangle\lvert e_0\rangle+ \lvert L_1\rangle\lvert e_1\rangle}{\sqrt2}.$ With $\gamma=\langle e_1|e_0\rangle$, an ideal branch-interference measurement within the laboratory has maximum visibility $V=|\gamma|$. Meanwhile, the trace distance between the two external record states is $D_E=\sqrt{1-|\gamma|^2}, \qquad P_{\mathrm{guess}}(E)=\frac{1+D_E}{2}.$ Thus $V^2+D_E^2=1$ for these equal-prior pure conditional records, a special case of the which-way distinguishability relation studied by Englert. If the external record supports 90% optimal branch-guessing success, this model permits at most $V=0.6$. A perfectly distinguishable retained record gives zero laboratory-only visibility.

This is a resource constraint, not a local-friendliness inequality. It does not decide whether the modelled premeasurement counts as an observed event. It tells an experimental proposal what is needed to combine a persistent copy of that event with an interference operation. Mixed records, unequal priors, and restricted readout require the corresponding generalized distinguishability calculation.

Measuring or erasing the external register conditionally can change the subensemble interference. Ignoring the conditioning record restores the original average statistics. A postselected interference pattern must therefore not be reported as an unconditional reversal while retaining the very which-branch information the model traces out.

Established theorem, debated interpretation

Bong and colleagues derive inequalities from a specified set of assumptions including locality-related conditions, freedom of settings, and absoluteness of observed events, then discuss quantum violations and an experiment. The incompatibility is conditional and precise. It is not a proof that every possible meaning of an observer-independent world is inconsistent.

An important research discipline is to separate a theorem’s mathematical premises from the physical interpretation assigned to the modeled friend. Another is to identify which records survive until the final measurement and which must be erased or coherently reversed. If two claimed observer reports cannot coexist as jointly accessible data, treating them as entries in a single observed table is an additional assumption.

Baumann and Brukner analyze an additional register that would record a friend’s awareness of a change in her memory, finding a conflict with no-signalling under their extended-scenario assumptions. This is a claim about a particular memory construction and probability assignment. It reinforces the need to model extra records explicitly rather than adding a remembered history verbally after the circuit has been specified.

Schlosshauer’s decoherence analysis supplies the background for modeling record leakage. It does not independently settle which no-go premise should be abandoned.

An entry project: reconstruct one local-friendliness bound

Baseline. Use the numerical construction associated with Bong and colleagues to reproduce one inequality and its quantum value. Document the operational variables and permitted probability marginals before running the optimization.

Resources. Linear programming or small semidefinite programs, qubit circuits, and a laptop. The paper’s publisher page links the LocalFriendliness code; pin the exact revision used. Start with a single chosen inequality and list all probability marginals on which its bound depends.

Deliverable. An independently checked inequality table, a circuit calculation, and a noise sweep in which memory dephasing and measurement visibility are varied separately. Show which terms in the witness require coherent control of the friend’s memory. Independently reproduce the $V=1$ and $V=0$ endpoints above before calculating the full witness. The deliverable should distinguish an inequality violation, a record-readout error, and a conditional interference measurement.

Failure criterion. If the reproduction uses a different causal or marginal-consistency assumption, it is a different theorem. If the implemented circuit retains a branch record that the mathematical protocol erases, recalculate the probabilities including that record.

Research extension. Replace the ideal memory by a limited-access register and derive the strongest witness allowed by a specified control budget. Alternatively, compare two assumption sets by constructing an explicit model that satisfies one but not the other. Both approaches yield checkable results without relying on ambiguous language about consciousness. The new result must improve on the finite-memory and record assumptions already treated in the literature; adding dephasing to an otherwise unchanged circuit is a calibration exercise until that comparison is made.

How to judge a proposed direction

Favor projects with a finite operational scenario and a falsifiable inequality or simulation. A purely verbal contradiction often disappears once conditioning, time ordering, and the persistence of records are made explicit. A good philosophical project can still be rigorous: it should identify a premise, show its role in a proof, and state the cost of rejecting it.

Reassessment is warranted by an improved theorem with demonstrably weaker assumptions or an experiment implementing a materially stronger record-and-control regime. A larger quantum memory is relevant only if its actual operations match the intended operational contract.

References

• V. Baumann and Č. Brukner, “Wigner’s friend’s memory and the no-signaling principle,” Quantum 8, 1481 (2024). Memory-register analysis.
• K.-W. Bong, A. Utreras-Alarcón, F. Ghafari, Y.-C. Liang, N. Tischler, E. G. Cavalcanti, G. J. Pryde, and H. M. Wiseman, “A strong no-go theorem on the Wigner’s friend paradox,” Nature Physics 16, 1199–1205 (2020). Article and code link.
• B.-G. Englert, “Fringe Visibility and Which-Way Information: An Inequality,” Physical Review Letters 77, 2154–2157 (1996). Distinguishability bound.
• M. Schlosshauer, “Decoherence, the measurement problem, and interpretations of quantum mechanics,” Reviews of Modern Physics 76, 1267–1305 (2004). arXiv:quant-ph/0312059.

Entry points

Begin with the inequality’s probability variables and supplemental derivation. Draw a circuit marking every surviving classical record; this frequently clarifies what a proposed test can and cannot establish.