Empirical Indistinguishability in Quantum Theory

• Empirical indistinguishability is the idea that different physical states, models, or processes yield the same observable outcome statistics under all permitted measurements.
• It underpins fundamental limits in quantum ontology, statistical mechanics, and model selection, affecting quantum metrology and randomness testing.
• Measure concentration in high-dimensional systems and overlapping ontological models illustrate both theoretical constraints and operational challenges in differentiating states.

Empirical indistinguishability refers to the impossibility, in principle or in practice, of distinguishing between physical states, processes, models, or objects based solely on empirical data—that is, on all possible measurement outcomes permitted by the underlying theory and experimental constraints. In quantum theory, empirical indistinguishability is a pervasive phenomenon with foundational, operational, and methodological significance. It emerges at multiple levels, from the foundations of quantum ontology and statistical mechanics to practical questions in quantum information, statistical model selection, and control. Empirical indistinguishability imposes intrinsic limitations on knowledge: different underlying realities, distributions, or microstates may yield the same observable statistics and thus remain forever indistinguishable by any measurement.

1. Definitions and General Concepts

Empirical indistinguishability is defined in terms of observable outcome statistics. Two physical descriptions—whether states, probability distributions, or models—are empirically indistinguishable with respect to a prescribed class of measurements if they induce exactly the same distribution of outcomes for every measurement permitted by the theory and the experimental setup. This notion is often formalized as follows:

• Quantum states: Two density matrices $\rho,\sigma$ (or pure states $\psi, \phi$) are empirically indistinguishable relative to a measurement $M = \{E_z\}$ if $p(z|\rho) = p(z|\sigma)$ for all outcomes $z$, or, more generally, if the total-variation distance $\|\{p(z|\rho)\}_z - \{p(z|\sigma)\}_z\|_{TV}$ is less than a specified $\epsilon$ for all $M$.
• Ontological models: Given a set of ontic distributions $C$ and a map $\Phi: C \to$ (density matrices) $\psi, \phi$0, two distributions $\psi, \phi$1 with $\psi, \phi$2 yield exactly the same outcome statistics on all finite-outcome POVMs and are therefore empirically indistinguishable (Tumulka, 2022).
• Dynamical or statistical models: Two models are empirically indistinguishable if every input-output equation, transfer function, or observable trajectory produced by one model can be matched identically by some assignment of parameters in the other model, i.e., if their observable predictions coincide across all experiments (Bortner et al., 2023).
• Random number sources: Quantum and pseudo-random sources are empirically indistinguishable if no efficient test (randomness measure plus distinguisher) can tell them apart with anything more than negligible probability (Tsurumaru et al., 2023).

Empirical indistinguishability is either exact (for all possible experiments and resolutions) or $\psi, \phi$3-approximate (for outcomes indistinguishable up to statistical or practical error). In all cases, it circumscribes the empirical content of a physical description by the limitations of the measurement framework.

2. Empirical Indistinguishability in Quantum Ontology

In quantum mechanics, the empirical indistinguishability of certain underlying realities is an unavoidable consequence of the structure of quantum theory and its statistical predictions. In ontological models of quantum theory, where quantum states are associated with distributions over a space of hidden variables or “ontic states” $\psi, \phi$4, two distinct ontic distributions $\psi, \phi$5 can yield identical outcome statistics for all quantum measurements if they map to the same density matrix via the Born rule. This phenomenon is nontrivial:

• Tumulka’s theorem: In any empirically adequate ontological model (with convex set $\psi, \phi$6 of preparable distributions and the Born rule for all finite-outcome POVMs), the mapping $\psi, \phi$7 is never injective. There must exist pairs $\psi, \phi$8 with $\psi, \phi$9, i.e., empirically indistinguishable distributions (Tumulka, 2022).
• Implication: Even with unrestricted measurements, certain distinctions in the underlying “reality” are empirically inaccessible. There exist “facts in nature that cannot be discovered empirically”—one cannot, even in principle, determine which underlying distribution was present if all operational predictions match. This forms a quantum limitation to knowledge, distinct from classical theory.

Further, in $M = \{E_z\}$0-epistemic models, quantum indistinguishability of non-orthogonal states is often explained by overlapping ontic distributions. However, tight quantitative results show that in higher-dimensional Hilbert spaces, the classical overlaps cannot reproduce quantum indistinguishability, and the ability of epistemic overlap to explain empirical indistinguishability diminishes exponentially with system size (Barrett et al., 2013, Leifer, 2014).

3. Typicality and Observational Indistinguishability in High-Dimensional Quantum Systems

In high-dimensional Hilbert spaces, empirical indistinguishability is strengthened by measure concentration phenomena:

• Typicality theorems: For a large Hilbert subspace $M = \{E_z\}$1 (e.g., specified by a low-entropy macrostate via the Past Hypothesis), the overwhelming majority of pure states $M = \{E_z\}$2 (sampled with respect to Haar measure) yield, for any finite-outcome measurement $M = \{E_z\}$3, very nearly the same probability distributions for measurement outcomes as the maximally mixed state on $M = \{E_z\}$4. More precisely, for $M = \{E_z\}$5, the fraction of typical states for which $M = \{E_z\}$6 is exponentially small in the dimension $M = \{E_z\}$7 (Chen et al., 2024).
• Consequences:
  • Observational indistinguishability: No conceivable experiment can, even in principle, distinguish which typical pure state of $M = \{E_z\}$8 is realized. Any observation, unless astronomically unlikely, leaves the set of plausible initial conditions essentially unchanged.
  • Implications for cosmology and quantum statistical mechanics: The universal quantum state is empirically determined only up to this immense indistinguishable class.

This measure concentration differentiates quantum mechanics from classical statistical mechanics, where microstates typically yield sharply differing predictions. In quantum systems, empirical indistinguishability can be in-principle, not just in-practice, due to the geometry of Hilbert space (Chen et al., 2024).

4. Operational and Resource Perspectives in Quantum Information

Empirical indistinguishability underpins and quantifies specific resources and limitations in quantum information processing, quantum metrology, and experimental design.

• Indistinguishability of particles: The indistinguishability of identical bosons or fermions is quantified by the expectation value $M = \{E_z\}$9, where $p(z|\rho) = p(z|\sigma)$0 projects onto the fully symmetric $p(z|\rho) = p(z|\sigma)$1-particle subspace. $p(z|\rho) = p(z|\sigma)$2 measures the probability that the system behaves as fully indistinguishable bosons in an operational test (e.g., a perfect bunching experiment in linear optics). Tight lower bounds on $p(z|\rho) = p(z|\sigma)$3 can be derived from efficiently measurable two-particle projectors, enabling operational certification of how much "bosonic" indistinguishability is present (Englbrecht et al., 2023).
• In phase estimation: Quantum Fisher information (QFI)—dictating achievable precision in phase estimation—scales linearly with the degree of indistinguishability $p(z|\rho) = p(z|\sigma)$4 of input photons in an interferometer. Even minimal indistinguishability suffices to beat the standard quantum limit, indicating that empirical indistinguishability is a necessary and sufficient resource for quantum metrological advantage (Knoll et al., 2019).
• Entanglement from indistinguishability: Overlapping identical systems show operational entanglement detectable via spatially-localized measurements, which can be harnessed for quantum protocols such as teleportation. The ability to perform certain tasks thus arises from the indistinguishability not as a metaphysical principle, but as a strictly operational resource (Franco et al., 2017).

In all cases, the inability to distinguish certain under-the-hood microstates or particle configurations directly determines quantum operational capabilities or limitations.

5. Empirical Indistinguishability in Model Selection and System Identification

Empirical indistinguishability extends to classical and quantum modeling, especially in system identification, compartmental modeling, and statistical inference:

• Structural indistinguishability in compartmental models: Two distinct linear compartmental models are empirically indistinguishable if for any input and initial state, their outputs match for some parameters. The concept of permutation indistinguishability captures cases where models differ only by a relabeling of parameters, leading to exactly the same transfer functions. Sufficient graphical conditions (e.g., moving leaks, sliding detours in path-based models) guarantee indistinguishability based purely on model topology, enabling rapid certification without symbolic computation (Bortner et al., 2023).
• Implication: No experimental data—no matter how precise or extensive—can distinguish between permutation indistinguishable models unless extra assumptions or interventions are introduced. Empirical indistinguishability thus sets limits on parameter identification and model discrimination in applied sciences.

6. Philosophical, Metaphysical, and Contextual Nuances

Empirical indistinguishability also features in debates around identity, individuality, and the foundations of quantum theory.

• Empirical indistinguishability and the Gibbs paradox: The persistent entropy-of-mixing ambiguity in classical and quantum statistical mechanics reflects a pragmatic, context-dependent notion of indistinguishability. Whether two subsystems are treated as “the same” or “distinct” for entropy accounting depends on the set of allowed macroscopic manipulators and observables, not on ontological facts about the particles involved (Dieks, 2014).
• Contextuality and ontology: Quasi-set theory and related frameworks exploit a non-individual ontology of quantum particles to resolve paradoxes (e.g., Kochen-Specker contradictions) by taking indistinguishables as truly lacking identity. Contextual or probabilistic assignments then become non-problematic—each measurement may probe a different but indistinguishable property, nullifying demands for global, context-independent valuations (Barros et al., 2017).
• Symmetric observables and discernibility: Even within the strictures of permutation invariance (the Indistinguishability Postulate), quantum theory admits symmetric, relational observables (e.g., relative distance operators) that weakly discern quantum particles. This means empirical relations (but not absolute properties) can sometimes distinguish “identical” quantum systems in a physically legitimate and measurable way (Caulton, 2014).

7. Indistinguishability for Randomness and Computational Criteria

Empirical indistinguishability arises even in computational and cryptographic contexts:

• Quantum vs. pseudo-randomness: Assuming the existence of secure cryptographic pseudo-random number generators, no efficiently computable randomness measure and distinguisher can separate quantum random numbers from pseudo-random numbers produced by such a generator, provided the quantum process can be efficiently simulated classically. Empirical indistinguishability is thus cryptographically absolute except when computational assumptions or algorithmic constraints are abandoned (Tsurumaru et al., 2023).
• Experimental confirmation: In practice, quantum-generated random bit strings are found to be statistically indistinguishable from pseudo-random ones, as measured by all efficient algorithmic tests. This renders classical and quantum randomness operationally equivalent for any feasible application.

---

Empirical indistinguishability thus permeates modern physics, from limitations in state discrimination and knowledge in quantum foundations, through its status as a resource in information processing and metrology, to structural ambiguities in model identification and limitations in computational randomness tests. It reflects both foundational constraints of physical law and contextual, operational choices regarding what distinctions are considered empirically relevant.