Contextuality

Contextuality is the obstruction to explaining operational statistics by an ontological model whose representations respect a specified context-independence principle. In the traditional sharp-measurement setting, the same event must keep its value across compatible measurement contexts. In the generalized setting, operationally equivalent procedures must have identical ontological representations. These are related but nonidentical concepts:

1. traditional Kochen–Specker contextuality, formulated for sharp measurements with predetermined outcomes; and
2. generalized operational contextuality, formulated through operational equivalences among preparations, measurement events, and transformations.

The distinction matters. A Kochen–Specker contradiction is a contextuality proof, but not every modern contextuality proof is a Kochen–Specker coloring argument. Contextuality is also not merely noncommutativity, measurement disturbance, or Bell nonlocality.

Required background. Projectors supplies sharp yes-no events and complete orthogonal alternatives.

Helpful background. Projection-Valued Measures supplies the measure-theoretic meaning of a sharp measurement; Gleason’s Theorem constrains context-independent additive probability assignments on projectors.

Contexts for sharp measurements

For a finite-dimensional system, a sharp measurement context can be represented by a projective measurement

$$\mathcal C = \{P_1,\ldots,P_n\}, \qquad P_iP_j=\delta_{ij}P_i, \qquad \sum_{i=1}^nP_i=I.$$

The projectors are mutually exclusive outcomes, and the last equation says that the outcomes are exhaustive. More generally, a context is a maximal or otherwise specified set of compatible observables whose joint measurement is under discussion. The exact choice of maximal versus nonmaximal contexts must be stated in any theorem or experiment.

A projector can occur in more than one context. For example,

$$\mathcal C=\{P,Q,R\}, \qquad \mathcal C'=\{P,S,T\}$$

may be two qutrit bases that share the event $P$. The operational question associated with $P$ is the same yes-no question in both bases, even though the alternative outcomes differ.

Traditional measurement noncontextuality requires the value attributed to that shared event to be the same in both contexts. It is not the claim that all observables commute. It is a consistency requirement on the representation of an event wherever that event appears.

The Kochen–Specker value assignment

The traditional framework assumes that every sharp event has a predetermined response. An ontic state $\lambda$ therefore induces a map

$$v_\lambda:\mathcal P(\mathcal H)\longrightarrow\{0,1\},$$

where $v_\lambda(P)=1$ means that the outcome represented by $P$ would occur. For every complete projective context,

$$\sum_{P\in\mathcal C}v_\lambda(P)=1.$$

Exactly one outcome is assigned value one. Crucially, $v_\lambda(P)$ has no context label: if $P$ belongs to several PVMs, it keeps the same value in all of them.

The equivalent observable formulation uses a value $v_\lambda(A)$ in the spectrum of each sharp observable and imposes functional consistency,

$$v_\lambda\!\left(f(A)\right) = f\!\left(v_\lambda(A)\right).$$

For commuting observables, the same idea preserves joint algebraic relations. Spectral projectors translate between the projector and observable formulations. These constraints are what make a collection of local within-context assignments into one proposed global assignment.

The Kochen–Specker Theorem proves that no such global assignment exists for all projectors when $\dim\mathcal H\geq3$. Each context can be colored separately; the contradiction appears when overlapping contexts must agree.

Operational statistics and ontological models

Generalized contextuality begins one level closer to laboratory data. Let $\mathsf P$ denote a preparation, $M$ a measurement, and $k$ an outcome. An operational theory supplies probabilities

$$p(k\mid M,\mathsf P).$$

An ontological model introduces an ontic state $\lambda$, a preparation distribution $\mu(\lambda\mid\mathsf P)$, and a measurement response $\xi(k\mid M,\lambda)$ such that

$$p(k\mid M,\mathsf P) = \int_\Lambda \xi(k\mid M,\lambda)\, \mu(\lambda\mid\mathsf P)\,d\lambda.$$

The response function is a conditional probability:

$$0\leq\xi(k\mid M,\lambda)\leq1, \qquad \sum_k\xi(k\mid M,\lambda)=1.$$

It need not be deterministic. The ontological-model question is whether these distributions and response functions can reproduce the observed operational probabilities while respecting specified equivalences among laboratory procedures.

Generalized operational noncontextuality

Two preparation procedures $\mathsf P$ and $\mathsf P'$ are operationally equivalent when no available measurement distinguishes them:

$$p(k\mid M,\mathsf P) = p(k\mid M,\mathsf P')$$

for every allowed $M$ and $k$. Preparation noncontextuality then requires

$$\mu(\lambda\mid\mathsf P) = \mu(\lambda\mid\mathsf P').$$

Likewise, two measurement events $[k\mid M]$ and $[k'\mid M']$ are operationally equivalent when they have the same probability for every allowed preparation. Measurement noncontextuality requires

$$\xi(k\mid M,\lambda) = \xi(k'\mid M',\lambda)$$

for every $\lambda$. Transformation noncontextuality applies the analogous condition to operationally equivalent transformations and their ontic transition kernels.

The principle can be summarized as:

Operationally indistinguishable procedures receive identical ontological representations.

This is a conditional methodological principle, not a theorem of probability theory. A contextuality experiment first specifies an operational fragment, the equivalences within that fragment, and the data to be reproduced. Only then does it ask whether a noncontextual ontological model exists.

Traditional and generalized notions compared

Question	Kochen–Specker framework	Generalized operational framework
Primary objects	Sharp projectors or observables	Preparations, measurement events, and transformations
Equivalence	The same projector or functional observable relation	Empirical indistinguishability within an operational fragment
Response functions	Outcome deterministic for sharp measurements	May be stochastic
Measurement scope	Projective measurements	Can include POVMs and unsharp measurements
Preparation and transformation assumptions	Usually not part of the theorem	Stated separately through operational equivalences
Single-qubit scope	No contradiction from projective basis coloring alone	Contextuality proofs can exist with broader procedures and equivalences

The generalized framework contains the traditional case only after the additional sharp-measurement and determinism assumptions are justified. It is therefore unsafe to use “noncontextual” without saying which equivalences and which ontological representations are meant.

Where outcome determinism enters

Outcome determinism is an explicit premise of the traditional Kochen–Specker value-assignment problem. It is not built into generalized measurement noncontextuality. There, $\xi(k\mid M,\lambda)$ may lie strictly between zero and one.

For certain ideal sharp measurements, outcome determinism can be derived from perfect predictability together with additional assumptions, typically including preparation noncontextuality and access to suitable operationally equivalent mixtures. That derivation is scenario dependent. Perfect predictability by itself does not license replacing every response function by zero or one.

This assumption ledger prevents two common mistakes:

• using a deterministic Kochen–Specker contradiction against a model that was never required to be deterministic; and
• claiming a generalized no-go result while silently assuming the sharp, ideal measurement structure that the generalized framework was designed to avoid.

State-independent and state-dependent witnesses

The original Kochen–Specker theorem is state independent: no quantum state enters the impossibility of coloring the relevant projectors. Finite Kochen–Specker sets and parity constructions exhibit the obstruction using a finite compatibility hypergraph. The Peres–Mermin square developed on the Kochen–Specker theorem page is a compact example in dimension four.

Other contextuality tests are state dependent. They select a state and a finite set of measurements whose correlations violate an inequality obeyed by the chosen class of noncontextual models. The spin-1 construction of Klyachko, Can, Binicioğlu, and Shumovsky is a standard example.

A logical contradiction and a statistical witness answer different questions:

• a coloring proof excludes an exact deterministic assignment for an ideal measurement structure;
• a noncontextuality inequality supplies a noise-robust data constraint, but only relative to its operational equivalences, compatibility assumptions, and treatment of experimental imperfections.

Finite precision cannot be handled by pretending approximate projectors are exactly identical. A credible experiment reports how operational equivalences and compatibility are established or bounded, then propagates those errors into the noncontextual bound.

The single-qubit boundary

A rank-one qubit projector can be written

$$P_{\boldsymbol n} = \frac12 \left(I+\boldsymbol n\cdot\boldsymbol\sigma\right),$$

and its orthogonal complement is $P_{-\boldsymbol n}$. Every rank-one projective context consists only of this antipodal pair. One can select exactly one point from each pair on the Bloch sphere and assign it value one, assigning zero to its antipode. Thus the traditional projector-coloring contradiction does not arise in dimension two.

This exception is narrow. It does not show that every operational fragment of qubit quantum theory has a generalized-noncontextual model. Preparations, unsharp measurements, POVMs, or transformations introduce operational equivalences not captured by antipodal projective pairs. Spekkens’s generalized framework provides qubit no-go results precisely in these broader scenarios.

Two qubits also form a four-dimensional system, so state-independent Kochen–Specker arguments such as the Peres–Mermin square apply even though each subsystem separately has dimension two.

What contextuality is not

Not merely noncommutativity. Noncommuting observables cannot belong to one sharp joint context, but a contextuality proof requires a network of compatible contexts and a failed global representation. One incompatible pair is not a contextuality proof.

Not merely disturbance. Sequentially measuring incompatible observables can disturb a system. Kochen–Specker contextuality instead compares a shared event across compatible contexts. Experiments must still control disturbance because it can imitate a violation if compatibility is assumed too casually.

Not Bell nonlocality. Bell scenarios add spacelike separated parties and a local-causality factorization. Kochen–Specker contextuality requires neither spatial separation nor locality. Bell nonlocality and contextuality share convex and ontological-model techniques, but their premise ledgers are different; see Bell’s Theorem.

Not the exclusion of all hidden variables. Contextual ontological models remain logically possible. The theorem excludes the stated noncontextual class, not every completion or interpretation of quantum mechanics.

Interpretive force and limits

Contextuality shows that quantum predictions cannot be represented as if every sharp property carried one pre-existing, context-independent value satisfying all quantum functional relations. Generalized contextuality extends the lesson: even stochastic ontological representations can fail when operationally equivalent procedures are required to have identical representations.

Neither conclusion selects a unique interpretation. A model may retain underlying variables by making outcome responses depend on the full measurement arrangement, by rejecting a relevant operational equivalence at the ontic level, or by changing another premise. The scientific content lies in identifying exactly which classical representation fails, not in attaching the word “contextual” to every form of measurement dependence.

Common pitfalls

Leaving the context undefined. State whether contexts are PVMs, compatible sets, or specified laboratory implementations. Different choices define different model classes.

Treating a projector label as an operational equivalence. Two procedures represented by the same ideal operator are operationally equivalent only relative to the operational theory or tested fragment being used.

Smuggling in determinism. Traditional Kochen–Specker arguments assume zero-one values. Generalized measurement noncontextuality alone permits stochastic responses.

Forgetting the qubit exception. Projective Kochen–Specker coloring does not yield a single-qubit contradiction. Broader qubit contextuality results use additional procedures or equivalences.

Equating a violation with apparatus independence. A robust conclusion depends on how imperfect compatibility and approximate equivalences enter the noncontextual bound.

Exercises

1. A shared projector

Let

$$\mathcal C=\{P,Q,R\}, \qquad \mathcal C'=\{P,S,T\}$$

be two qutrit projective contexts. If $v(P)=1$, determine the values of the other four projectors in a Kochen–Specker assignment. What additional requirement connects the two contexts?

Solution

Exactly one projector in each complete context has value one. Therefore

$$v(Q)=v(R)=v(S)=v(T)=0.$$

The connection is measurement noncontextuality: the shared projector has one value $v(P)$, not separate values $v(P\mid\mathcal C)$ and $v(P\mid\mathcal C')$.

2. Why a qubit coloring exists

Show that choosing one projector from every antipodal pair $\{P_{\boldsymbol n},P_{-\boldsymbol n}\}$ defines a valid deterministic assignment for all rank-one qubit projective measurements.

Solution

Every nontrivial rank-one qubit PVM is exactly

$$\{P_{\boldsymbol n},P_{-\boldsymbol n}\}.$$

Assign value one to the selected member of each antipodal pair and zero to the other. Then

$$v(P_{\boldsymbol n})+v(P_{-\boldsymbol n})=1$$

for every qubit context. No third mutually orthogonal ray exists to generate the overlapping-basis constraints used in dimension three.

3. Operational equivalence

Suppose two measurement events have equal probabilities for three tested preparations but the operational fragment contains a fourth preparation that distinguishes them. May measurement noncontextuality equate their response functions?

Solution

No. Operational equivalence requires equality for every preparation in the declared fragment. The fourth preparation breaks the equivalence, so measurement noncontextuality imposes no equality between the two response functions. Testing only the first three preparations would define a smaller fragment and a different claim.

4. Determinism audit

An ontological model assigns $\xi(1\mid M,\lambda)=1/3$ for some $\lambda$. Does this fact alone make the model measurement contextual?

Solution

No. Generalized measurement noncontextuality compares response functions for operationally equivalent events; it does not require those functions to be zero or one. The model is measurement contextual only if two equivalent events receive different responses for some $\lambda$. It does, however, fall outside the deterministic value-assignment class used by a traditional Kochen–Specker proof.

References

• A. A. Klyachko, M. A. Can, S. Binicioğlu, and A. S. Shumovsky, “Simple Test for Hidden Variables in Spin-1 Systems,” Physical Review Letters 101, 020403, 2008, doi:10.1103/PhysRevLett.101.020403.
• S. Kochen and E. P. Specker, “The Problem of Hidden Variables in Quantum Mechanics,” Indiana University Mathematics Journal 17, 59–87, 1968, doi:10.1512/iumj.1968.17.17004.
• M. D. Mazurek, M. F. Pusey, R. Kunjwal, K. J. Resch, and R. W. Spekkens, “An Experimental Test of Noncontextuality without Unphysical Idealizations,” Nature Communications 7, 11780, 2016, doi:10.1038/ncomms11780.
• R. W. Spekkens, “Contextuality for Preparations, Transformations, and Unsharp Measurements,” Physical Review A 71, 052108, 2005, doi:10.1103/PhysRevA.71.052108.