Against probability: A quantum state is more than a list of probability distributions

This paper demonstrates a no-go theorem proving that a topologically robust probability representation of quantum states, while necessary for meaningful physical statements, cannot simultaneously preserve essential structural features such as the subsystem composition.

The Gist

The Big Idea: The Map is Not the Territory

Imagine you have a complex, 3D sculpture (a quantum state). To describe it to someone who can't see it, you decide to take a series of 2D photographs (probability distributions) from every possible angle.

The authors of this paper argue that while these photos can uniquely identify the sculpture, relying on them to describe the sculpture has a fatal flaw: The photos can lie about how close two sculptures are.

In the world of quantum physics, scientists often try to describe a system not by its "true" state (the density operator), but by a list of probabilities for what happens when you measure it. The paper claims that if you try to do this in a way that is mathematically "robust" (meaning small errors in the photos don't lead to huge errors in your understanding), you lose the ability to describe how parts of the system fit together.

The Problem: The "Blurry Photo" Trap

To understand why this matters, imagine you are trying to generate a truly random string of numbers (like a secret code) using a quantum machine.

1. The Goal: You want the output to be perfectly random. In the "real" quantum world, you can prove a sequence is random if it is impossible to distinguish from a perfect random source, even if an enemy has access to all the machine's internal secrets.
2. The Trap: The authors show a scenario where a quantum system looks almost perfectly random if you only look at the "photos" (the probability distributions of local measurements). The photos say, "Hey, this looks random!"
3. The Reality: But if you look at the actual quantum state, it is not random at all. It is highly entangled and structured.

The Analogy:
Imagine two people standing very far apart.

• The Real Distance (Trace Distance): If you measure the actual distance between them, they are 100 miles apart.
• The Photo Distance (Probability Metric): You take a photo of them from a specific angle. In the photo, they look like they are standing right next to each other.

If you only trust the photo, you think they are close. But in reality, they are far apart. The paper calls this non-robustness. It means that a "small" difference in the probability list (the photo) can actually hide a "massive" difference in the real physical state.

The Dilemma: You Can't Have It All

The authors prove a "No-Go" theorem. You cannot have a probability-based description of a quantum system that has all three of these desirable features at the same time:

1. Robustness: Small changes in the description shouldn't mean the system is totally different. (The photo should match reality).
2. Subsystem Structure: You must be able to describe parts of the system separately (e.g., Alice's part and Bob's part) without losing the connection between them.
3. Efficiency: The description should be compact and manageable (not requiring an infinite amount of data).

The Trade-off Table:

• Standard Quantum Mechanics (Density Operators): Has all three. It's robust, handles parts well, and is efficient.
• Probability Representations (Local Measurements): You can have the parts and efficiency, but you lose robustness. The photos lie about closeness.
• Probability Representations (All Measurements): You can have robustness and parts, but you lose efficiency. The list of probabilities becomes so huge it's useless.

Why This Matters (According to the Paper)

The authors point out that many popular ideas in physics rely on these probability lists, and this discovery breaks them:

• QBism (Quantum Bayesianism): This theory treats quantum states as just a list of an agent's beliefs (probabilities). The paper says this view fails for complex systems (like a vibrating string or a particle in space) because the "belief list" isn't robust enough to describe reality accurately.
• Reconstructing Quantum Theory: Scientists try to rebuild quantum physics from simple rules about probabilities. The paper says you can't do this successfully for large systems unless you add extra, artificial rules to fix the "closeness" problem.
• Quantum Cryptography: If you try to prove a secret code is secure using only probability lists (without assuming quantum mechanics is true), you might think it's secure because the "photos" look random. But the paper warns that the code might actually be broken because the "photos" are misleading.
• Quantum Field Theory: Physicists often describe the universe using "correlation functions" (a type of probability list). The paper suggests these descriptions might fail to capture the true nature of complex, non-local connections in the universe.

The Bottom Line

The paper concludes that while probability lists are a very popular and convenient way to talk about quantum physics, they are fundamentally flawed as a complete replacement for the standard "density operator" description.

The Metaphor:
Trying to describe a quantum system only with probability lists is like trying to navigate a city using only a 2D map that has been stretched and distorted. It might show you the names of the streets (the structure), and it might be easy to carry in your pocket (efficient), but if you try to judge the distance between two buildings based on that map, you will get lost. To navigate safely, you need the 3D reality (the density operator), not just the distorted map.

Technical Summary

Technical Summary: "Against probability: A quantum state is more than a list of probability distributions"

Problem Statement
Quantum states are frequently represented not by density operators ($\rho$) but by lists of outcome probabilities derived from a tomographically complete set of measurements ($M$). Such representations are ubiquitous in physics, appearing in quantum field theory (via correlation functions) and quantum foundations (within generalized probabilistic frameworks). While such a probability representation $P_M(\rho)$ is injective (uniquely specifying the state), the authors question whether it is faithful in a physical sense. Specifically, they investigate whether approximate statements derived from probability representations (e.g., "the output is approximately random") remain valid when "pulled back" to the level of density operators. The core problem is that probability representations may lack topological robustness: a sequence of states that appears to converge to an ideal behavior in the probability representation space may not converge in the physical state space (defined by trace distance).

Methodology
The authors employ a topological and information-theoretic approach to analyze the relationship between the metric induced by a probability representation and the standard trace distance metric on the space of density operators.

1. Definition of Robustness: They define a representation $P_M$ as topologically robust if convergence in the induced metric $d_M$ implies convergence in the trace distance $\delta$. Formally, for any sequence of states $(\rho^{(n)})$ and a set of states $\Sigma$:
   $$\lim_{n\to\infty} d_M(\rho^{(n)}, \Sigma) = 0 \implies \lim_{n\to\infty} \delta(\rho^{(n)}, \Sigma) = 0$$
   where $d_M(\rho, \sigma) = \frac{1}{2} \sup_{M \in \mathcal{M}} \|P_M(\rho) - P_M(\sigma)\|_1$.

2. Counter-Example Construction: They construct a specific example involving a randomness generation protocol using $n$ unstable atoms. They define a sequence of entangled states $\rho^{(n)}_{RE}$ that are far from the set of perfectly random states $\Sigma_{rand}$ in terms of trace distance ($\delta \geq 1/4$). However, when restricted to local measurements ($M_\otimes$), the probability distributions of these states converge to $\Sigma_{rand}$ as $n \to \infty$. This demonstrates that the representation based on local measurements is not robust.

3. Topological Characterization: The authors analyze the vector space spanned by density operators, $\text{span}(D(H))$. They prove that while the topologies induced by $d_M$ and $\delta$ are identical on the set of density matrices $D(H)$ (except in pathological "fragile" cases), robustness requires the equivalence of these topologies on the entire vector space $\text{span}(D(H))$. This is equivalent to the completeness of $\text{span}(D(H))$ with respect to the norm $\|\cdot\|_M$.

4. Quantifying Structure via Entropy: To generalize the result beyond specific examples, they introduce an information-theoretic measure of "structure." They define the asymptotic entropy of a representation based on the minimum size of a compressed list of data required to $\epsilon$-decode the probabilities for any measurement. They consider representations that are "efficient" (where the number of effects grows polynomially with dimension) or those based on local operations.

Key Contributions and Results

• The No-Go Theorem (Theorem 6): The central result is a no-go theorem stating that if a probability representation $P_M$ possesses non-negligible structure (specifically, if its asymptotic entropy is bounded by a sequence significantly smaller than the trivial upper bound of $2^{2^n}$), then the representation cannot be topologically robust.
• Incompatibility of Structure and Robustness: The paper establishes a fundamental trade-off:
  • Robustness requires the representation to be "structure-less" (essentially requiring an exponentially large number of measurements to distinguish states robustly).
  • Structure (such as subsystem decomposition, locality, or efficiency) inevitably leads to non-robustness.
• Specific Corollaries:
  • Corollary 7: The representation based on local measurements ($P_{M_\otimes}$) is not robust. This implies that security definitions in post-quantum cryptography based solely on non-signaling boxes (which correspond to $P_{M_\otimes}$) may not guarantee security under standard quantum definitions.
  • Corollary 8: Any efficient representation (where the number of measurement effects grows polynomially with the Hilbert space dimension, such as minimal tomographically complete sets like SIC-POVMs) is not robust. This applies to systems of unbounded dimension.
• Generalization to GPTs (Theorem 9): The authors extend their findings to Generalized Probabilistic Theories (GPTs), showing that there exist GPTs where the topologies induced by local measurements and the generalized trace distance differ, even when the set of measurements is not fragile.

Significance and Claims
The paper argues that probability representations, while mathematically convenient and foundational to many frameworks (including QBism and reconstruction programs), suffer from a fundamental flaw: they fail to preserve the notion of "closeness" between states when the representation includes physically relevant structure.

• Implications for Quantum Foundations: The results pose a challenge to programs that attempt to reconstruct quantum theory from probabilistic postulates using efficient measurement sets (e.g., SIC-POVMs) or to interpret quantum states purely as catalogs of belief (QBism) for unbounded systems. Without robustness, approximate probabilistic statements cannot be reliably mapped back to physical states.
• Implications for Quantum Information: In quantum cryptography, the non-robustness of $P_{M_\otimes}$ suggests that "post-quantum" schemes relying on non-signaling boxes might be less secure than their quantum counterparts, as the probabilistic definition of security does not guarantee security in the trace-distance sense.
• Implications for Quantum Field Theory (QFT): Since QFT states are often characterized by correlation functions (a form of probability representation), the authors suggest these representations are not robust, particularly for non-local observables (e.g., Hawking radiation correlations).

Conclusion
The authors conclude that the density operator formalism possesses favorable properties—robustness, subsystem structure, and efficiency—that probability representations cannot simultaneously achieve. While probability representations offer theory-independence, the paper argues that for any representation with non-negligible structure (which is necessary for practical physics), topological robustness is lost. This necessitates a search for alternative approaches that unify the generality of probability representations with the robustness of density operators.