Sheaf-Theoretic Preparation Contextuality

Imagine you are a detective trying to solve a mystery, but instead of looking for a single culprit, you are trying to figure out if a set of local clues can be explained by one single, consistent "master story" that happened behind the scenes.

This paper introduces a new way of looking at a famous quantum mystery called contextuality. Usually, scientists look at contextuality through the lens of measurements (asking questions and getting answers). This paper flips the script and looks at it through the lens of preparations (setting up the experiment).

Here is the breakdown using simple analogies:

1. The Two Sides of the Coin: Measuring vs. Preparing

In the quantum world, there are two main ways to interact with a system:

Measurement (The Old Way): You set up a machine, ask it a question, and get an answer. The "context" is which other questions you asked at the same time.
- The Analogy: Imagine you are looking at a painting through a small window. If you look through the top-left window, you see a blue sky. If you look through the bottom-right, you see a green tree. "Measurement contextuality" asks: Is there one single, complete painting behind the wall that explains all these views? If the views contradict each other (e.g., the sky is blue in one window but red in another overlapping window), there is no single painting. The views are "contextual."
Preparation (The New Way): You set up a machine to create a specific state (like preparing a specific card from a deck). The "context" is which other machines you could have used to prepare it.
- The Analogy: Imagine you are a chef. You have different stations (sources) to prepare ingredients. Station A can make a "Red Sauce" or a "Blue Sauce." Station B can make a "Spicy Sauce" or a "Sweet Sauce."
- The paper asks: If I tell you I used Station A to make Red Sauce, and Station B to make Spicy Sauce, can we imagine a single, master recipe book (a global response) that explains how all possible combinations of sauces were made, even the ones we didn't actually mix?

2. The Core Problem: Filling in the Blanks

The paper's main insight is about how we "fill in the blanks" when we don't have full information.

In Measurement (The Old Way): If you know the global picture, you can easily figure out the local picture by just "zooming out" or ignoring details. It's like taking a high-res photo and cropping it. There is only one right way to crop it.
In Preparation (The New Way): If you know the local picture (the specific sauce made at Station A), figuring out the global picture (the master recipe) is much harder. There isn't just one way to guess what happened at the other stations. You have to make a stochastic guess (a probability guess).
- The Metaphor: Imagine you find a half-eaten cookie on a table. You know it came from a specific jar (local context). But to guess what the whole jar looked like (global context), you have to imagine what the other cookies were. You could guess they were all chocolate, or all oatmeal, or a mix. There are many ways to "complete" the story.

3. The Rules of the Game

The authors realized that because there are many ways to guess the global story, we need strict rules to make the game fair. They proposed two rules for how we are allowed to "fill in the blanks":

Input Independence: Your guess about the missing ingredients shouldn't depend on what you already know about the ingredients you do have. If I tell you "I used Red Sauce," your guess about the Spicy Sauce shouldn't change just because I told you that. The sources are independent.
Compositionality: If you guess the global story in two steps (first guess the middle, then guess the end), it should be the same as guessing the whole thing in one step. The order of your guessing shouldn't matter.

When you follow these two rules, the paper proves something surprising: The only way to guess the global story is to treat every source as a separate, independent coin flip. You can't have a complex, interwoven global story; it has to be a simple product of individual parts.

4. The Big Reveal: The PBR Example

The authors tested this new framework using a famous quantum setup called the PBR scenario (named after Pusey, Barrett, and Rudolph).

The Setup: Imagine two chefs (Alice and Bob) each have two ways to prepare a dish. They combine their dishes and serve them to a judge.
The Result: The paper shows that even if you follow the strict rules of "Input Independence" and "Compositionality," you cannot construct a single, consistent "Master Recipe Book" that explains all the dishes Alice and Bob served.
The Conclusion: No matter how you try to fill in the blanks to create a global story, the local clues (the actual dishes served) contradict the global story. The "Master Recipe" simply does not exist.

Summary

This paper introduces a new mathematical tool (using "sheaf theory," which is just a fancy way of organizing local and global data) to prove that in the quantum world, how you prepare a system matters just as much as how you measure it.

They showed that if you try to explain quantum preparation statistics as if they came from a single, hidden, classical reality (a global recipe), you hit a wall. The local statistics cannot be "stochastically extended" to a global whole without breaking the rules of independence. This proves that the quantum world is "contextual" not just when we look at it, but even when we set it up.

1. Problem Statement

The paper addresses a gap in the mathematical formulation of quantum contextuality. While contextuality is well-understood in the context of measurements (where local statistics cannot be extended to a global joint distribution), the formulation for preparations has been less rigorous in the sheaf-theoretic framework.

Existing approaches to preparation contextuality (e.g., Spekkens' operational framework) rely on ontological models constrained by operational equivalences. However, they do not frame preparation contextuality as a structural obstruction to extending local data to a global object, which is the hallmark of the sheaf-theoretic approach to measurement contextuality.

The core challenge identified by the authors is the fundamental asymmetry between measurement and preparation:

Measurement: Local data is obtained from global data via canonical marginalization (restriction maps). The restriction is unique and fixed by physics.
Preparation: Local data arises from global data via stochastic averaging over unobserved degrees of freedom. There is no canonical "completion" map; there are many inequivalent ways to extend a partial preparation specification to a global one.

The paper asks: Can we formulate preparation contextuality as an obstruction to the existence of a global response matrix, given that the extension mechanism itself is non-unique and stochastic?

2. Methodology

The authors develop a preparation-dual framework within the sheaf-theoretic formalism, utilizing explicit linear algebra and matrix theory.

A. Mathematical Framework

Setup: They define a set of sources $Y$ (preparation devices) and a set of preparation instances $I$ for each source. A "source context" $\Gamma$ is a subset of sources that can be jointly implemented.
Empirical Model: An empirical model is a family of conditional outcome distributions $E_{m|\Gamma}$ for each context $\Gamma$ , describing the statistics of a fixed measurement $m$ given specific preparation choices.
The Extension Problem: The goal is to determine if there exists a global response matrix $D_{m|Y}$ (mapping global preparation instances to outcomes) and a family of stochastic extension matrices $S_{Y|\Gamma}$ (mapping local instances to global instances) such that:
$E_{m|\Gamma} = D_{m|Y} S_{Y|\Gamma}$
This equation represents lifting local statistics to a global model.

B. Defining "Admissible" Extensions

Since extension matrices are not canonical, the authors impose two minimal structural constraints to define an admissible extension family:

Input Independence: The distribution of instances outside a context $\Gamma$ must be independent of the specific instances chosen within $\Gamma$ . This prevents artificial correlations from being built into the extension rule.
Compositionality: Extending in stages (e.g., $\Gamma \to \Gamma' \to Y$ ) must yield the same result as a single-step extension. This ensures the refinement procedure is coherent and path-independent.

C. Key Theoretical Derivation

Using the constraints of input independence and compositionality, the authors prove Theorem 1 (Product-form Extension Theorem).

Result: Any admissible extension matrix $S_{Y|\Gamma}$ must factorize into a product of single-site distributions.
$S_{Y|\Gamma}(\sigma_Y | \sigma_\Gamma) = \mathbb{1}[(\sigma_Y)|_\Gamma = \sigma_\Gamma] \prod_{p \in Y \setminus \Gamma} \mu_p(\sigma_p)$
Implication: This rigid product form drastically reduces the space of possible extensions, turning the problem of finding a global model into a feasibility problem for stochastic matrices with specific structural constraints.

3. Key Contributions

Formalization of Preparation Contextuality: The paper provides the first sheaf-theoretic formulation of preparation contextuality defined strictly as an obstruction to stochastic extension, mirroring the measurement case but accounting for the non-uniqueness of extension maps.
Product-Form Theorem: The derivation that admissible extensions must factorize across sources is a critical structural result. It shows that "noncontextual" preparation models require a specific type of independence between sources.
Definition of Preparation Compatibility: The authors introduce a condition (Eq. 25) ensuring that the induced statistics on overlapping source contexts are consistent regardless of which larger context is used to derive them. This distinguishes between trivial incompatibility (where no extension exists) and genuine contextuality (where a compatible extension exists, but no global response matrix can reproduce the data).
Possibilistic Reduction: They establish that if a model is possibilistically contextual (based on support/zero-patterns), it is also probabilistically contextual. This allows for combinatorial proofs using Boolean matrices.

4. Results

The authors demonstrate their framework using a Pusey-Barrett-Rudolph (PBR)-type scenario:

Setup: Two parties (Alice and Bob) each have two sources ( $Z$ and $X$ bases) with two preparation instances each. The contexts are pairs of sources (e.g., $\{a, b\}$ ).
Quantum Realization: The preparation instances correspond to pure states $|0\rangle, |1\rangle, |+\rangle, |-\rangle$ . The measurement is a specific 4-outcome POVM.
Analysis:
1. Compatibility: They show that the empirical model is preparation compatible. There exists an admissible extension family (specifically, uniform single-site distributions) that satisfies the consistency conditions on overlaps.
2. Contextuality: They prove that no global response matrix $D_{m|Y}$ exists that can reproduce the empirical statistics for any admissible extension family.
Proof Mechanism: The proof relies on a parity-based combinatorial argument.
- The empirical data contains specific "forbidden" outcomes (zeros) for every local preparation.
- Due to the product-form of the extension, these zeros propagate to the global level.
- For global instances with even parity, the propagated constraints forbid all four possible measurement outcomes simultaneously, which is impossible for a valid probability distribution.
- For odd parity, the constraints leave insufficient support to satisfy the empirical data.
- Conclusion: The empirical model is preparation contextual.

5. Significance

Theoretical Unification: The work successfully extends the powerful sheaf-theoretic machinery (previously limited to measurements) to the preparation side, creating a symmetric duality between restriction (measurement) and stochastic extension (preparation).
Clarification of Noncontextuality: It clarifies that preparation noncontextuality is not just about the existence of an ontological model, but specifically about the existence of a global response matrix compatible with coherent, independent extensions of local data.
Operational Implications: By framing the problem as a linear-algebraic feasibility problem (stochastic matrix extension), the authors open the door to algorithmic checks for contextuality and potential cohomological characterizations (analogous to measurement contextuality).
Foundational Insight: The results reinforce the idea that quantum states cannot be viewed as mere "preparations" of underlying classical variables that are independent of the context, even when the preparation devices themselves are treated as classical control interfaces. The PBR-type example shows that the "reality" of the quantum state (in the sense of preparation independence) is incompatible with a noncontextual global description.

In summary, the paper establishes a rigorous, matrix-based framework for preparation contextuality, proving that quantum mechanics exhibits a fundamental obstruction to extending local preparation statistics to a global noncontextual model, even when the extension rules are constrained only by minimal structural independence.