The Wigner's Friend Thought Experiment

Put a competent observer inside the very system you are describing quantum-mechanically, and the measurement problem stops being an abstraction and becomes a clash between two people. Your friend $F$ sits in a sealed lab and measures a spin prepared in $\alpha\lvert\uparrow\rangle+\beta\lvert\downarrow\rangle$ ; from inside, $F$ sees a definite result and writes it down. You, Wigner $W$ , stand outside with no information leaking out, so the only description consistent with "the lab is a closed quantum system" is a coherent superposition of " $F$ recorded up" and " $F$ recorded down." Two descriptions of one event; one contains a fact the other insists does not yet exist. This is not a disagreement about numbers. Trace the friend out of $W$ 's pure state and you recover exactly the mixed state encoding $F$ 's probabilities, so no measurement on the spin alone can ever tell the two stories apart. The disagreement is about what is real, and about where the Heisenberg cut between "quantum system" and "observer who collapses things" sits. Every ingredient — entanglement, the reduced density matrix, the partial trace — you already own; Wigner's friend aims them at the observer. What keeps the puzzle alive is reversibility. If $W$ could run the sealed lab backwards with the inverse of the friend's measurement unitary, the record would un-write and an interference experiment would prove no collapse ever happened. Decoherence makes that reversal hopeless in any real macroscopic lab — but "hopeless in practice" is not "forbidden in principle," and that gap is where modern no-go theorems (Frauchiger–Renner and the local-friendliness inequalities) turn the parable into hard mathematics that rules interpretations in or out.

Key takeaways

A friend $F$ measures a spin inside a sealed lab and sees a definite outcome; the outside observer $W$ describes the whole lab as a unitary system and assigns the entangled superposition $\alpha\lvert\uparrow\rangle\lvert F_\uparrow\rangle+\beta\lvert\downarrow\rangle\lvert F_\downarrow\rangle$ — no collapse.
The two descriptions are observationally identical on the spin alone: tracing out the friend gives the friend's classical mixture $\lvert\alpha\rvert^2\lvert\uparrow\rangle\langle\uparrow\rvert+\lvert\beta\rvert^2\lvert\downarrow\rangle\langle\downarrow\rvert$ , so the clash is ontological, not statistical.
$W$ could tell them apart only with a coherent interference (or reversal) measurement on the entire lab — allowed by unitary QM, ruined by decoherence in any real lab.
The scenario sharpens the measurement problem: the tension owned by "The Quantum Measurement Problem" and "Von Neumann Measurement and the Projection Postulate," now made interpersonal.
Extended versions (Frauchiger–Renner 2018; local-friendliness, Bong et al. 2020) prove universal quantum theory, single observer-independent outcomes, and locality/free-choice cannot all hold — a Bell-like no-go for observers.
Interpretations declare themselves: Copenhagen restricts the cut, many-worlds denies single outcomes, objective-collapse modifies the dynamics, relational/QBist make outcomes observer-relative.
Key idea: Wigner's pure entangled state and the friend's collapsed state predict identical spin statistics because the reduced density matrix is the same — yet they disagree about whether a single fact exists, and settling that disagreement is exactly choosing how to solve the measurement problem.

The core insight

Wigner's pure entangled description and the friend's collapsed description of the same measurement are observationally indistinguishable on the system alone — tracing out the friend reproduces the friend's classical statistics — yet they disagree about what is factual; resolving that disagreement is equivalent to choosing how the measurement problem is solved.

Where this lives today

Wigner's friend has moved from the seminar room to the optics table. In 2020 Bong and collaborators built an extended-Wigner-friend test with single photons as "friends": a photon's path is a measurement record that a superobserver can either read out or coherently reverse and interfere. The measured correlations violate a local-friendliness inequality, so — for these simple systems — observer-independent facts, locality, and free choice cannot all hold, connecting the parable directly to the loophole-free Bell-test program. The parable also drives falsifiable physics. Because the conflict hinges on whether macroscopic superpositions survive, objective-collapse models (GRW, CSL) make distinct predictions, and matter-wave interferometry with ever-larger molecules is narrowing the collapse-parameter window. Frauchiger–Renner, by treating observers as qubits in a reversible circuit, is a touchstone for work on consistent multi-agent reasoning and quantum-information foundations.

The physical picture

Picture a soundproof, light-tight, thermally isolated room. Inside, your friend measures a quantum coin and records a definite answer, up or down. From outside, knowing only that the room is isolated and obeys the Schrödinger equation, you must describe it as the superposition "friend-recorded-up" plus "friend-recorded-down" — nothing has leaked out to select a branch. Both stories are internally consistent; they disagree on one thing only: whether a single fact already exists. The lever that pries them apart is reversibility. If you could coherently run the room backwards — apply the inverse of the friend's measurement unitary — you would un-write the record, restore the original superposition, and a follow-up interference measurement would show a sharp result no collapsed state could give. The friend's "definite fact" is only as robust as the irreversibility of the record: the paradox lives in the idealized reversible limit, and decoherence hides it in the real world.

Figure

Nested observers and the Heisenberg cut

The payoff is a single, sharp question you can put on a plot: is the sealed lab in a pure entangled state (Wigner is right, no collapse) or a mixed diagonal state (the friend is right, a fact exists)? The two give different answers to one global observable, but identical answers to everything measurable on the spin alone — the whole tension in one frame.

Figure

The one experiment that could decide — and why it is hopeless

The core equations

Every equation here is standard measurement theory; pointing it at an observer rather than a dial makes "when does collapse happen" inescapable. We use $\hbar=1$ , a spin- $\tfrac12$ system, and a friend whose memory has orthonormal record states. The spin starts in a coherent superposition — the object the friend is about to measure, an ordinary pure state.

\lvert\psi\rangle_S=\alpha\lvert\uparrow\rangle+\beta\lvert\downarrow\rangle,\qquad |\alpha|^2+|\beta|^2=1 \tag{1}

where: $\lvert\psi\rangle_S$ is the pre-measurement spin state; $\alpha,\beta\in\mathbb{C}$ are the up/down amplitudes; $\lvert\uparrow\rangle,\lvert\downarrow\rangle$ are eigenstates of the measured observable (say $S_z$ ). Read physically: the spin has no definite $S_z$ value until measured, and $|\alpha|^2,|\beta|^2$ are the Born weights the friend will see. A measurement, described microscopically, is a unitary interaction that correlates the friend's memory with the spin — this is the "Process 2" premeasurement owned by "Von Neumann Measurement and the Projection Postulate." It is defined by how it acts on the eigenbasis.

\hat U_{\text{meas}}\big(\lvert\uparrow\rangle\lvert F_0\rangle\big)=\lvert\uparrow\rangle\lvert F_\uparrow\rangle,\qquad \hat U_{\text{meas}}\big(\lvert\downarrow\rangle\lvert F_0\rangle\big)=\lvert\downarrow\rangle\lvert F_\downarrow\rangle \tag{2}

where: $\lvert F_0\rangle$ is the friend's "ready" state; $\lvert F_\uparrow\rangle,\lvert F_\downarrow\rangle$ are orthonormal records, $\langle F_\uparrow|F_\downarrow\rangle=0$ . Read physically: a good measurement copies "which outcome" into a distinguishable memory — the minimal model of a faithful detector. Because $\hat U_{\text{meas}}$ is linear, it cannot turn the superposition (1) into one branch; it distributes over the sum, entangling spin and friend into the state Wigner assigns.

\lvert\Psi\rangle_{SF}=\alpha\,\lvert\uparrow\rangle\lvert F_\uparrow\rangle+\beta\,\lvert\downarrow\rangle\lvert F_\downarrow\rangle \tag{3}

where: $\lvert\Psi\rangle_{SF}$ is Wigner's global state — a pure entangled superposition. Read physically: from the outside, "the friend measured" means "the friend got entangled," full stop. There is no fact about which outcome in this description — exactly the point owned by "The Quantum Measurement Problem," which states that unitary evolution "never produces a single definite outcome." Any observer who can touch only the spin — including Wigner, if he refuses to open the lab — sees the reduced state, the partial trace over the friend. This is where the two descriptions secretly agree.

\rho_S=\mathrm{Tr}_F\,\lvert\Psi\rangle_{SF}\langle\Psi\rvert=|\alpha|^2\lvert\uparrow\rangle\langle\uparrow\rvert+|\beta|^2\lvert\downarrow\rangle\langle\downarrow\rvert \tag{4}

where: $\mathrm{Tr}_F$ is the partial trace over the friend's Hilbert space; $\rho_S$ is the state available to any spin-only probe. Read physically: this is identical to the friend's collapsed mixture. "The Reduced Density Matrix and Partial Trace" owns the general fact that $\rho_A=\mathrm{Tr}_B\rho_{AB}$ is "mixed exactly when the joint pure state is entangled" — here that mixedness is precisely the friend's classical uncertainty in disguise. To distinguish pure from collapsed, one must measure a genuinely global observable — the projector onto Wigner's coherent state, the one experiment that separates the descriptions.

\hat O_{\text{interf}}=\lvert\Psi\rangle_{SF}\langle\Psi\rvert,\qquad \langle\hat O_{\text{interf}}\rangle_{\text{pure}}=1,\qquad \langle\hat O_{\text{interf}}\rangle_{\text{mix}}=|\alpha|^4+|\beta|^4 \tag{6}

where: $\hat O_{\text{interf}}$ projects onto $\lvert\Psi\rangle_{SF}$ ; "mix" is the diagonal collapsed lab state $|\alpha|^2\lvert\uparrow F_\uparrow\rangle\langle\uparrow F_\uparrow\rvert+|\beta|^2\lvert\downarrow F_\downarrow\rangle\langle\downarrow F_\downarrow\rvert$ . Read physically: for $\alpha=\beta=1/\sqrt2$ the pure case gives $1$ and the mixture gives $\tfrac12$ — a factor-of-two gap that exists only if Wigner can coherently address the whole macroscopic lab. Finally, the spin–friend entanglement in Wigner's description is quantified by the von Neumann entropy of the reduced state — the functional owned by "Entanglement Entropy and Measures of Entanglement."

S(\rho_S)=-|\alpha|^2\log_2|\alpha|^2-|\beta|^2\log_2|\beta|^2 \tag{7}

where: $S(\rho_S)$ is in bits. Read physically: $S=1$ bit at $\alpha=\beta=1/\sqrt2$ (maximal entanglement, "which-outcome" fully in the friend) and $S=0$ for a product state (a definite spin).

Derivation AWhy Wigner cannot detect the friend's collapse from the spin alone

We prove the sharp version of the paradox's subtlety: every measurement statistic restricted to the spin is identical whether one uses Wigner's pure entangled state or the friend's collapsed mixture. The strategy is to take an arbitrary system observable $\hat A$ , compute its expectation in Wigner's description by expanding the entangled state, and watch the friend-orthogonality kill exactly the terms that could carry a signature of coherence. Watch that "the spin alone" means we tensor $\hat A$ with the identity on the friend, which is what makes the friend factor collapse into an inner product.

Step 1. Wigner measures only the spin, so the relevant operator is $\hat A\otimes\mathbb{1}_F$ and its expectation in the global pure state is $\langle\hat A\otimes\mathbb{1}_F\rangle_\Psi=\langle\Psi|(\hat A\otimes\mathbb{1}_F)|\Psi\rangle$ . Why: probing only the spin means doing nothing to the friend — the identity on $F$ encodes "no interference experiment on the lab."

Step 2. Expand with $\lvert\Psi\rangle_{SF}=\alpha\lvert\uparrow\rangle\lvert F_\uparrow\rangle+\beta\lvert\downarrow\rangle\lvert F_\downarrow\rangle$ : the diagonal terms carry $\langle F_k|F_k\rangle=1$ , while the cross terms — the only ones that could reveal coherence — carry $\alpha^*\beta\,\langle\uparrow|\hat A|\downarrow\rangle\langle F_\uparrow|F_\downarrow\rangle+\text{c.c.}$ Why: the friend overlap $\langle F_\uparrow|F_\downarrow\rangle$ sits precisely on the coherence-carrying cross terms.

Step 3. Friend records are orthonormal, $\langle F_\uparrow|F_\downarrow\rangle=0$ , so the cross terms vanish and

\langle\hat A\rangle_\Psi=|\alpha|^2\langle\uparrow|\hat A|\uparrow\rangle+|\beta|^2\langle\downarrow|\hat A|\downarrow\rangle.

Why: a faithful record is exactly one that is perfectly distinguishable — and that same distinguishability is what erases the interference from the spin's viewpoint.

Step 4. The friend, holding a definite-but-unknown outcome, predicts $\langle\hat A\rangle_F=|\alpha|^2\langle\uparrow|\hat A|\uparrow\rangle+|\beta|^2\langle\downarrow|\hat A|\downarrow\rangle$ — the same number. Since $\hat A$ was arbitrary, no spin-only experiment discriminates the descriptions; the disagreement cannot be statistical. Verified: reproduces the partial-trace identity $\langle\hat A\rangle_\Psi=\mathrm{Tr}_S(\hat A\rho_S)$ ; the verifier confirms $\rho_S$ is diagonal with the Born weights for three amplitude choices, off-diagonals below $10^{-12}$ .

Go deeper · Derivation B · 1 minReversal and revival — when a collapse can be undone

Derivation BReversal and revival — when a collapse can be undone

We show that if the friend's measurement is a genuine unitary $\hat U_{\text{meas}}$ , then Wigner — controlling the entire isolated lab — can apply $\hat U_{\text{meas}}^\dagger$ and recover the original superposition, proving the friend's "collapse" was never an irreversible physical fact. The route is direct: act with the inverse on the entangled state (3), use linearity, and invert the defining relations (2). The subtlety is that this requires reversing the friend's memory too — exactly what decoherence forbids in a real lab.

Step 1. Start from Wigner's post-measurement state $\lvert\Psi\rangle_{SF}=\alpha\lvert\uparrow\rangle\lvert F_\uparrow\rangle+\beta\lvert\downarrow\rangle\lvert F_\downarrow\rangle$ . Why: this is the only description available before the lab is opened — the friend's outcome is not a separate fact in it.

Step 2. Apply the inverse interaction, using linearity: $\hat U_{\text{meas}}^\dagger\lvert\Psi\rangle_{SF}=\alpha\,\hat U_{\text{meas}}^\dagger\lvert\uparrow\rangle\lvert F_\uparrow\rangle+\beta\,\hat U_{\text{meas}}^\dagger\lvert\downarrow\rangle\lvert F_\downarrow\rangle$ . Why: $\hat U_{\text{meas}}^\dagger$ exists because a closed-system measurement interaction is unitary; inverting (2) sends $\lvert\uparrow\rangle\lvert F_\uparrow\rangle\to\lvert\uparrow\rangle\lvert F_0\rangle$ and likewise for down.

Step 3. Collect the terms:

\hat U_{\text{meas}}^\dagger\lvert\Psi\rangle_{SF}=(\alpha\lvert\uparrow\rangle+\beta\lvert\downarrow\rangle)\otimes\lvert F_0\rangle=\lvert\psi\rangle_S\otimes\lvert F_0\rangle.

The friend's record is erased and the original spin superposition is restored, so a subsequent $S_x$ measurement gives a sharp $+1$ that a truly collapsed state could never produce. Why: if the friend's experience had corresponded to an objective, irreducible collapse, this exact revival would be impossible — the recovered coherence is the smoking gun that unitary evolution never removed a branch.

Step 4. The lesson is a forced trilemma: deny universal unitarity (objective collapse), deny single outcomes (many-worlds), or make outcomes relational. In any realistic lab, coupling to $\sim10^{23}$ environmental degrees of freedom makes $\hat U_{\text{meas}}^\dagger$ unattainable, hiding but not resolving the conflict. Verified: modeling $\hat U_{\text{meas}}$ as a CNOT (control = spin, target = friend memory), the verifier confirms $\hat U^\dagger\hat U=\mathbb 1$ , that the premeasurement produces $\alpha\lvert\uparrow F_\uparrow\rangle+\beta\lvert\downarrow F_\downarrow\rangle$ , and that $\hat U^\dagger$ restores the product state with the reduced spin purity back to $1$ .

Go deeper · Derivation C · 1 minThe reversibility clash in one number — pure versus collapsed

Derivation CThe reversibility clash in one number — pure versus collapsed

This is the structural result the article owns: the same joint state is simultaneously "pure" (Wigner) and "effectively collapsed" (the friend), and a single global observable exposes the difference. We compute the joint purity, the reduced-spin purity, and the interference-projector expectation in both labs — watching the bookkeeping, since purity of the joint lab and purity of the spin answer different questions.

Step 1. For Wigner's pure state, $\rho_{SF}=\lvert\Psi\rangle\langle\Psi\rvert$ satisfies $\mathrm{Tr}(\rho_{SF}^2)=1$ — the lab is in a definite (pure) global state even though no outcome is definite. Why: purity $1$ means a coherent interference measurement can, in principle, return a sharp answer.

Step 2. The reduced spin (Eq. 4) has purity $\mathrm{Tr}(\rho_S^2)=|\alpha|^4+|\beta|^4<1$ (equal to $\tfrac12$ at $\alpha=\beta=1/\sqrt2$ ), so the spin looks mixed — indistinguishable from a genuinely collapsed spin. Why: pure whole, mixed part is the mathematical face of entanglement — here the friend's "which-outcome" record does the entangling.

Step 3. Measure $\hat O_{\text{interf}}=\lvert\Psi\rangle\langle\Psi\rvert$ : on the pure state $\langle\hat O_{\text{interf}}\rangle=1$ , while on the collapsed diagonal mixture $\rho^{\text{mix}}_{SF}$ ,

\mathrm{Tr}\!\big(\hat O_{\text{interf}}\rho^{\text{mix}}_{SF}\big)=|\alpha|^2|\langle\Psi|\uparrow F_\uparrow\rangle|^2+|\beta|^2|\langle\Psi|\downarrow F_\downarrow\rangle|^2=|\alpha|^4+|\beta|^4.

Why: $\langle\Psi|\uparrow F_\uparrow\rangle=\alpha^*$ , $\langle\Psi|\downarrow F_\downarrow\rangle=\beta^*$ , so the mixture's overlap with the coherent state is $<1$ whenever both amplitudes are nonzero.

Step 4. The two answers, $1$ versus $|\alpha|^4+|\beta|^4$ , differ — a factor-of-two gap at the symmetric point that is real and in-principle measurable, yet operationally invisible because reading $\hat O_{\text{interf}}$ demands coherent control of every macroscopic degree of freedom. Verified: for $\alpha=\beta=1/\sqrt2$ and $\alpha^2=0.3$ , the verifier confirms joint purity $=1$ , reduced purity $=|\alpha|^4+|\beta|^4$ , $\langle\hat O\rangle_{\text{pure}}=1$ and $\langle\hat O\rangle_{\text{mix}}=|\alpha|^4+|\beta|^4$ , and that the two are unequal.

Limiting cases

The bite of the paradox is set by how isolated and reversible the friend's lab is — from a maximal in-principle conflict down to complete operational suppression.

Case	Result	Meaning
Perfectly isolated, reversible lab	$\rho_{\text{lab}}=\lvert\Psi\rangle\langle\Psi\rvert$	Wigner's pure state; $\langle\hat O_{\text{interf}}\rangle=1$ . In principle $W$ can interfere or reverse the whole lab and confirm no collapse. Paradox maximal.
Decohered lab (realistic)	$\rho_{\text{lab}}\to\|\alpha\|^2\lvert\uparrow F_\uparrow\rangle\langle\cdots\rvert+\|\beta\|^2\lvert\downarrow F_\downarrow\rangle\langle\cdots\rvert$	Environment einselects a pointer basis; $\langle\hat O_{\text{interf}}\rangle\to\|\alpha\|^4+\|\beta\|^4$ . Wigner's effective state matches the friend's mixture — suppressed operationally, intact conceptually ("Quantum Decoherence and the Density Matrix").
System-only view (any isolation)	$\rho_S=\|\alpha\|^2\lvert\uparrow\rangle\langle\uparrow\rvert+\|\beta\|^2\lvert\downarrow\rangle\langle\downarrow\rvert$	Identical for both observers. No spin-only measurement reveals the disagreement — why it is subtle, not statistical.
Product outcome $\alpha=1,\beta=0$	$S(\rho_S)=0$ , $\langle\hat O_{\text{interf}}\rangle_{\text{pure}}=\langle\hat O_{\text{interf}}\rangle_{\text{mix}}=1$	Spin already definite; no entanglement, no clash — a genuine fact everyone agrees on.
Extended (multi-friend, spacelike labs)	Local-Friendliness bound $\|\langle\text{CHSH}\rangle\|\le 2$ , quantum up to $2\sqrt2$	Friends as systems in separated labs; quantum theory predicts violation, a testable no-go — the register of "Bell's Theorem and the CHSH Inequality."

Worked examples

Example 1 — The measurable gap, and how hard it is to see. Take the friend measuring $S_z$ on the symmetric spin $\lvert\psi\rangle_S=\tfrac{1}{\sqrt2}(\lvert\uparrow\rangle+\lvert\downarrow\rangle)$ . Compute (a) the reduced-spin purity, (b) the interference-projector expectations for pure vs collapsed lab, and (c) the entanglement entropy.

Step 1. Amplitudes $\alpha=\beta=1/\sqrt2$ , so $|\alpha|^2=|\beta|^2=\tfrac12$ . The reduced state (Eq. 4) is $\rho_S=\tfrac12\lvert\uparrow\rangle\langle\uparrow\rvert+\tfrac12\lvert\downarrow\rangle\langle\downarrow\rvert=\tfrac12\mathbb 1$ .

Step 2. Reduced-spin purity: $\mathrm{Tr}(\rho_S^2)=|\alpha|^4+|\beta|^4=\left(\tfrac12\right)^2+\left(\tfrac12\right)^2=\tfrac14+\tfrac14=\mathbf{0.5}$ . The spin looks maximally mixed — exactly like a genuinely collapsed coin.

Step 3. Interference projector (Eq. 6): on the pure lab $\langle\hat O_{\text{interf}}\rangle=\mathbf{1}$ ; on the collapsed lab $\langle\hat O_{\text{interf}}\rangle=|\alpha|^4+|\beta|^4=\mathbf{0.5}$ . The gap is $1-0.5=\mathbf{0.5}$ — a factor of two, unambiguous if you can read $\hat O_{\text{interf}}$ .

Step 4. Entanglement entropy (Eq. 7): $S(\rho_S)=-\tfrac12\log_2\tfrac12-\tfrac12\log_2\tfrac12=\mathbf{1\ \text{bit}}$ — the "which-outcome" information is fully stored in the friend.

Step 5. The catch: reading $\hat O_{\text{interf}}$ requires coherently controlling every degree of freedom of the friend and lab ( $\sim10^{23}$ of them, with an unimaginably short decoherence time). So the $\mathbf{0.5}$ gap is real yet unmeasurable — the paradox in one line.

Example 2 — A biased friend and the entropy of ignorance. Now the friend measures an unequal superposition with $|\alpha|^2=0.3$ , $|\beta|^2=0.7$ . Find the reduced purity, the interference gap, and the entanglement entropy.

Step 1. Reduced state $\rho_S=0.3\,\lvert\uparrow\rangle\langle\uparrow\rvert+0.7\,\lvert\downarrow\rangle\langle\downarrow\rvert$ ; purity $\mathrm{Tr}(\rho_S^2)=0.3^2+0.7^2=0.09+0.49=\mathbf{0.58}$ .

Step 2. Interference expectations: pure $\langle\hat O_{\text{interf}}\rangle=\mathbf{1}$ ; collapsed $\langle\hat O_{\text{interf}}\rangle=|\alpha|^4+|\beta|^4=0.58$ . Gap $=1-0.58=\mathbf{0.42}$ — smaller than the symmetric case because a more biased state is "closer" to a definite outcome.

Step 3. Entanglement entropy $S(\rho_S)=-0.3\log_2 0.3-0.7\log_2 0.7=0.521+0.360=\mathbf{0.881\ \text{bits}}$ .

Step 4. Reading: less entanglement (0.881 vs 1 bit) means the friend stores less which-outcome information and the interference gap shrinks — the biased limit smoothly approaches a genuine product state as $|\alpha|^2\to0$ or $1$ .

Origins · 1 mineugene wigner and how this was discovered

eugene wigner

1902–1995 · Hungarian-American physicist, 1963 Nobel laureate · Posed the friend paradox to argue that consciousness might play a role in collapse

Eugene Wigner brought the measurement problem to a head by putting a conscious observer inside the quantum description. In his 1961 essay "Remarks on the Mind-Body Question," he argued that a friend measuring inside a sealed lab forces a choice: either the friend's mind collapses the wavefunction (his own tentative, now unpopular, reading), or quantum mechanics applies to observers too and one must accept a superposition of "friend saw up" and "friend saw down." Wigner's deeper legacy here is structural — his 1939 classification of the unitary representations of the Poincaré group is what tells us particles carry mass and spin, underlying the transformation theory recalled by "Lorentz Covariance of the Dirac Equation." The friend paradox stayed a curiosity for decades until David Deutsch (1985) proposed reversing the friend as a test, and Frauchiger and Renner (2018) turned it into a formal no-go theorem.

History

1961

— Wigner poses the friend thought experiment in "Remarks on the Mind-Body Question"

1985

— Deutsch proposes reversing the friend's measurement as an in-principle test of collapse

2018

— Frauchiger and Renner prove universal quantum theory, single outcomes, and consistency are jointly inconsistent

2018

— Brukner proves a no-go theorem for observer-independent facts

2020

— Bong et al. violate a local-friendliness inequality with photonic "friends"

Going further

References. Wigner, "Remarks on the Mind-Body Question," in The Scientist Speculates (1961) · Deutsch, Int. J. Theor. Phys. 24, 1 (1985) · Frauchiger & Renner, Nat. Commun. 9, 3711 (2018) · Brukner, Entropy 20, 350 (2018), "A No-Go Theorem for Observer-Independent Facts" · Bong et al., Nat. Phys. 16, 1199 (2020), "A strong no-go theorem on the Wigner's friend paradox" · Nielsen & Chuang, Quantum Computation and Quantum Information, Ch. 2 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. I (measurement postulates).

Provenance. The unitary premeasurement, entangled global state, and partial-trace reduction are standard measurement theory (Nielsen & Chuang Ch. 2; Cohen-Tannoudji measurement postulates), recomputable directly. The joint-purity / reduced-purity / interference-projector values and the entanglement entropies are machine-checked in the accompanying verifier for $\alpha=\beta=1/\sqrt2$ and $|\alpha|^2=0.3$ . The Frauchiger–Renner coin weight $1/3$ and joint outcome $1/12$ are the published values (Nat. Commun. 9, 3711); the local-friendliness inequality and its experimental violation are from Bong et al. (Nat. Phys. 16, 1199), the observer-independence no-go from Brukner (Entropy 20, 350). This article frames the paradox and states the no-go theorems; it does not claim to resolve Wigner's friend.

Builds on: The Density Operator and Mixed States · The Reduced Density Matrix and Partial Trace · Quantum Entanglement and Non-Separability · The Schmidt Decomposition · The Quantum Measurement Problem · Von Neumann Measurement and the Projection Postulate. Leads to: Bell's Theorem and the CHSH Inequality · The EPR Paradox and Local Realism · Hidden Variable Theories · Experimental Tests and Violation of Bell Inequalities.

Is the friend's experience "wrong" in Wigner's description?

No — and that is the point. In Wigner's pure entangled state, "friend saw up" and "friend saw down" both appear as branches, and each branch contains a perfectly ordinary friend with a definite, correct memory. Nothing in the friend's experience is falsified; what is denied is only that there is a single, observer-independent fact about which outcome occurred before the lab is opened. Whether that denial is acceptable is exactly what separates the interpretations.

Doesn't decoherence just solve this by collapsing the superposition?

Decoherence explains why you never see the paradox: coupling to the environment einselects a pointer basis and drives the reduced lab state to the diagonal mixture, so Wigner's effective description matches the friend's. But — as "Quantum Decoherence and the Density Matrix" states — the global system-plus-environment state is still one pure superposition. Decoherence delocalizes the coherence into an unmeasurable environment; it never removes a branch. It converts "interference could reveal no collapse" into "interference is hopeless," suppressing the paradox operationally without answering which outcome is real.

What exactly is the Heisenberg cut, and why does Wigner's friend move it?

The Heisenberg cut is the boundary between the part of the world treated as a quantum system (unitary) and the part treated as a classical measuring device (Born rule). "The Quantum Measurement Problem" stresses that this cut "has no physical location the formalism can point to." Wigner's friend dramatizes it: put the cut just outside the spin and the friend is a classical detector who collapses it; put it outside the whole lab and the friend is a quantum system in superposition. Both give the same spin statistics, so the physics does not fix where the cut goes — you decide, and that decision is an interpretation.

Could a real experiment ever run the friend backwards?

For a literal human friend, no — the environmental degrees of freedom entangled with a macroscopic brain and lab put $\hat U_{\text{meas}}^\dagger$ astronomically beyond reach, with a decoherence time far shorter than any conceivable control timescale. What is feasible is a microscopic "friend": a single photon, atom, or few-qubit register whose record is small enough to coherently reverse or interfere. Those are the systems in the Bong et al. (2020) local-friendliness experiment, which is why the no-go theorem could be tested at all.

How is Frauchiger–Renner different from ordinary Wigner's friend?

Plain Wigner's friend is a tension — two descriptions disagree about facts but never produce a numerical contradiction. Frauchiger–Renner nests two friend-plus-superobserver setups and has the agents reason about each other via the Born rule. Under three assumptions — quantum theory universally valid (Q), agents' conclusions mutually consistent (C), single outcome per measurement (S) — the chained reasoning concludes a joint outcome is impossible, yet quantum theory predicts it with probability $1/12$ : an outright logical contradiction, so at least one of Q, C, S must go, and which one you drop is which interpretation you hold.

Does the local-friendliness theorem assume hidden variables, like Bell?

No, and that is its strength. Bell's theorem starts from local hidden variables. The local-friendliness no-go (Brukner 2018; Bong et al. 2020) instead assumes only "absoluteness of observed events" (every measurement yields a single outcome that is a fact for everyone), locality, and freedom of choice — no realism about pre-existing values. These are strictly weaker than Bell's assumptions, so the inequality is harder to satisfy and its violation rules out a larger class of worldviews — a stronger no-go than Bell's for the question of observer-independent facts.