Preparation Distinguishability in Operational Theories

Updated 5 July 2026

Preparation distinguishability is the study of differentiating various preparation procedures using operational tools, characterizing distinctions in GPTs, quantum, and process theories.
It employs methodologies such as minimum-error discrimination in convex-operational frameworks and programmable gate criteria in process theories to establish discrimination limits.
The concept is further refined via ontological and contextual analyses, linking to resource constraints in multipartite locality, interference phenomena, and stochastic dynamics.

Preparation distinguishability denotes a family of operational questions about whether distinct preparations, preparation procedures, or preparation-induced ensembles can be told apart by the admissible primitives of a theory. In the cited literature, the term spans minimum-error discrimination of ensembles in generalized probabilistic theories (GPTs), probability-free programmability notions in process theories, impossibility theorems for distinguishing different preparations of the same pure state, ontological and contextual refinements of operational distinguishability, and specialized formulations for multipartite locality, interference, boson sampling, and stochastic trajectories (Bae, 2012, Chiribella, 2014, Pati, 2019).

1. Conceptual scope and formal variants

The literature does not use a single universal definition. Instead, it introduces several task-dependent notions, each tied to a specific operational setting.

Setting	Object being distinguished	Criterion
Convex-operational GPTs	Ensemble $\{q_x,w_x\}$	Maximum success / minimum error
General process theories	States $\{\rho_x\}$	Programmability for every gate-family
Standard quantum theory	Two preparations of the same pure state	Physical impossibility of non-disturbing discrimination
Ontological models	Preparations $P_x$ and epistemic states $\mu_x(\lambda)$	Comparison of $s^{OM}$ and $s^\Lambda$
Two-copy comparison games	Preparation procedures under a SWAP-type test	Operational score $D_{\rm op}$
Quantum information processing	Density operators, channels, strategies	Fidelity and trace distance

Representative definitions include the GPT optimization problem for minimum-error discrimination, the process-theoretic criterion that a family of states is distinguishable iff it can serve as a program for any family of gates, the quantum no-go result forbidding discrimination of two preparation procedures that yield the same pure state, the ontological comparison between operational distinguishability $s^{OM}$ and ontic distinctness $s^\Lambda$ , the SWAP-type two-copy score $D_{\rm op}$ , and fidelity- and trace-distance-based distinguishability measures with direct operational meanings (Bae, 2012, Chiribella, 2014, Pati, 2019, Chaturvedi et al., 2019, Ghoreishi, 20 Apr 2026, Rethinasamy et al., 2021).

This suggests that “preparation distinguishability” is best treated as a structured family of tasks rather than a single invariant. What is held fixed varies across frameworks: the ensemble, the gate family, the preparation history, the ontic representation, or the measurement architecture.

2. Convex-operational discrimination in generalized probabilistic theories

In the convex-operational GPT framework, one is given an ensemble of normalized states $\{\rho_x\}$ 0, where $\{\rho_x\}$ 1, $\{\rho_x\}$ 2, and $\{\rho_x\}$ 3. A measurement is a collection of effects $\{\rho_x\}$ 4 with $\{\rho_x\}$ 5 for all states and $\{\rho_x\}$ 6. The primal minimum-error problem is to maximize

$\{\rho_x\}$ 7

over effects satisfying $\{\rho_x\}$ 8 and $\{\rho_x\}$ 9. The dual problem minimizes $P_x$ 0 subject to $P_x$ 1 for each $P_x$ 2, and strong duality follows from Slater’s condition in finite-dimensional GPTs (Bae, 2012).

The Karush–Kuhn–Tucker conditions introduce a symmetry operator $P_x$ 3, nonnegative scalars $P_x$ 4, and normalized complementary states $P_x$ 5 such that

$P_x$ 6

These conditions encode both stationarity and complementary slackness. Operationally, the same $P_x$ 7 admits $P_x$ 8 decompositions into a known weighted state plus a complementary term, while the optimal effect $P_x$ 9 is orthogonal to its complementary component.

A central conclusion is that minimum-error distinguishability is generally a global property of the whole ensemble, not a function of pairwise overlaps alone. Geometrically, if one forms the polytope with vertices $\mu_x(\lambda)$ 0, then the complementary polytope with vertices $\mu_x(\lambda)$ 1 is congruent and anti-parallel, satisfying

$\mu_x(\lambda)$ 2

The success probability is then determined by the overall size relation between the two polytopes rather than by any single pairwise relation.

The same paper places discrimination in a bipartite steering scenario. If Alice can steer Bob’s marginal into decompositions

$\mu_x(\lambda)$ 3

then no-signaling implies

$\mu_x(\lambda)$ 4

The dual/KKT solution yields the same relation, so the bound is tight. The stated conclusion is that distinguishability is fixed by the compatibility of ensemble steering on states and the no-signaling principle on outcome probabilities, independently of the specific geometry of the GPT state space.

3. Probability-free distinguishability, copiability, and disturbance

A different formulation replaces probability of correct guessing by a structural notion in general process theories. For a family of states $\mu_x(\lambda)$ 5 of a system $\mu_x(\lambda)$ 6, distinguishability means that for every choice of data-bearing system $\mu_x(\lambda)$ 7, target system $\mu_x(\lambda)$ 8, and family of gates $\mu_x(\lambda)$ 9, there exists a single programmable gate

$s^{OM}$ 0

such that inserting $s^{OM}$ 1 on the $s^{OM}$ 2 port programs $s^{OM}$ 3 to perform $s^{OM}$ 4 on $s^{OM}$ 5 (Chiribella, 2014).

For finite families, distinguishability is equivalent to copiability. Concretely, there exists a copying gate

$s^{OM}$ 6

with

$s^{OM}$ 7

for all $s^{OM}$ 8 iff the family is distinguishable in the programmable-gate sense. The forward implication is immediate by choosing the target gates to be state-preparation maps; the reverse implication is obtained through faithful side information and asymptotic programming arguments.

The same framework yields a theory-independent no-information-without-disturbance statement for pure states. If $s^{OM}$ 9 are pure and not distinguishable, then no gate

$s^\Lambda$ 0

can both preserve the reduced state on $s^\Lambda$ 1 and extract side information into $s^\Lambda$ 2 that depends on whether $s^\Lambda$ 3 or $s^\Lambda$ 4 was input. In the proof, no disturbance plus purity forces

$s^\Lambda$ 5

and nontrivial side information $s^\Lambda$ 6 would imply distinguishability.

In finite-dimensional quantum theory, the abstract criterion specializes to the standard zero-error notion: a set $s^\Lambda$ 7 is distinguishable in this sense iff it can be discriminated with zero error, equivalently iff the states have mutually orthogonal supports (Chiribella, 2014). The process-theoretic formulation therefore reconstructs cloning and no-disturbance structure without appealing to probabilities.

4. Distinguishing preparation procedures for the same pure state

In standard quantum theory, a pure state may be prepared in infinitely many ways, but two preparation procedures that yield the same pure state cannot be physically distinguished without contradiction. Pati models a hypothetical preparation-distinguishing machine $s^\Lambda$ 8 by a unitary $s^\Lambda$ 9 acting on system $D_{\rm op}$ 0 and ancilla $D_{\rm op}$ 1, initially in $D_{\rm op}$ 2, such that

$D_{\rm op}$ 3

$D_{\rm op}$ 4

with

$D_{\rm op}$ 5

The machine states $D_{\rm op}$ 6 and $D_{\rm op}$ 7 are supposed to encode which preparation was used while leaving the system in $D_{\rm op}$ 8 (Pati, 2019).

Unitarity preserves inner products, leading to the key equation

$D_{\rm op}$ 9

Perfect distinguishability of the machine states would require $s^{OM}$ 0, hence $s^{OM}$ 1. But one may choose non-orthogonal fiducials such as $s^{OM}$ 2 and $s^{OM}$ 3, so the machine cannot exist in general. The paper’s conclusion is stronger: the only consistent possibility is $s^{OM}$ 4, in which case no information about the preparation procedure is ever written into the ancilla.

The no-go theorem is reinforced by a signaling argument. If Bob could distinguish preparation procedures for the same pure state, then in an EPR scenario Alice could choose different local unitaries on her half of singlets, inducing different preparation decompositions on Bob’s side. A perfect preparation-distinguishing machine would make Bob’s local density operators distinguishable, enabling superluminal signaling (Pati, 2019). The result is therefore tied directly to no-signaling rather than merely to no-cloning or no-measurement-without-disturbance.

The same reasoning extends to pure bipartite states, and the mixed-state case follows by purification. If two preparations of the same mixed state were distinguishable, then the corresponding purifications would also become distinguishable, contradicting the pure-state no-go theorem.

Preparation distinguishability also appears in settings where the standard Helstrom task is not the primary object. One example is the distinction between computable and truly random mixing. Bendersky and collaborators consider a computable mixture

$s^{OM}$ 5

where the sequence of pure states is generated by a fixed classical program. Even though $s^{OM}$ 6 in operator norm, they show that a maximally mixed source and a computable mixture can be distinguished in finite time and with arbitrarily high success probability by combining an informationally complete, unbiased POVM with Martin–Löf randomness tests. The measurement uses $s^{OM}$ 7 POVM elements, and the protocol halts almost surely (Grande et al., 2017). A proof-of-concept experiment further showed that two random-looking preparation sequences become distinguishable when true randomness is replaced with pseudorandomness.

A second refinement compares operational distinguishability with ontological distinctness. In the framework of Chaturvedi and Saha, for $s^{OM}$ 8 preparations $s^{OM}$ 9 chosen uniformly, the operational distinguishability is

$s^\Lambda$ 0

while ontic access gives

$s^\Lambda$ 1

Bounded ontological distinctness for preparations, $s^\Lambda$ 2, demands $s^\Lambda$ 3. For three qubit states with Bloch vectors at $s^\Lambda$ 4 in the equatorial plane, the optimal three-outcome POVM yields $s^\Lambda$ 5, while the minimal ontological distinctness forced by the analysis satisfies $s^\Lambda$ 6, giving excess ontological distinctness $s^\Lambda$ 7 (Chaturvedi et al., 2019). The paper also derives the inequality

$s^\Lambda$ 8

which is violated by the qubit triplet.

A third refinement is the two-copy comparison game of operational discriminability. Given preparations satisfying

$s^\Lambda$ 9

and a binary SWAP-type test with

$D_{\rm op}$ 0

the score is

$D_{\rm op}$ 1

In the qubit realization, $D_{\rm op}$ 2 equals a fidelity-based discriminability, but under a preparation-noncontextual ontological model with a SWAP-like rule and a sharp single-copy test it obeys

$D_{\rm op}$ 3

The natural qubit model saturates this bound, so the direct game alone does not witness contextuality in that regime. In a Bell-coupled scenario, however, the same comparison measurements on Bob’s conditional preparations produce the CHSH bound

$D_{\rm op}$ 4

establishing a quantitative link between discriminability and Bell-contextual correlations (Ghoreishi, 20 Apr 2026).

These variants show that preparation distinguishability can diagnose non-randomness, constrain ontological models, and act as an operational resource in contextuality-oriented scenarios.

6. Infinite families, multipartite locality, interference, and dynamics

For countably infinite pure-state families, Kawakubo and Koike distinguish three levels of unambiguous discrimination: simple distinguishability, uniform distinguishability, and perfect distinguishability. They prove the exact correspondences

distinguishable $D_{\rm op}$ 5 minimal,
uniformly distinguishable $D_{\rm op}$ 6 Riesz–Fischer,
perfectly distinguishable $D_{\rm op}$ 7 orthonormal (Kawakubo et al., 2016).

Applied to the von Neumann lattice of coherent states with fundamental-cell area $D_{\rm op}$ 8, the criterion becomes sharp: if $D_{\rm op}$ 9, the entire lattice is not unambiguously distinguishable, while if $\{\rho_x\}$ 00, it is uniformly distinguishable. At the threshold $\{\rho_x\}$ 01, deleting two points restores minimality but not the Riesz–Fischer property.

In multipartite quantum systems, distinguishability under locality constraints can be embedding-invariant. Shu shows that for perfect or unambiguous LOCC $\{\rho_x\}$ 02, PPT, and SEP discrimination, if a set of states is indistinguishable in $\{\rho_x\}$ 03, then it remains indistinguishable after embedding into $\{\rho_x\}$ 04 (Shu, 2020). The result states that such distinguishabilities are properties of the states themselves and are independent of the local system dimension. Related work on $\{\rho_x\}$ 05-partitions classifies tripartite product sets into classes according to which bipartitions permit perfect LOCC discrimination, and identifies cases in higher-party systems where a set can be fully bi-distinguishable yet locally indistinguishable across some finer $\{\rho_x\}$ 06-partition (Halder et al., 2019).

In interference theory, distinguishability is tied to internal structure and entanglement. For squeezed-vacuum sources on linear interferometers, Shchesnovich defines

$\{\rho_x\}$ 07

the probability that $\{\rho_x\}$ 08 photons interfere as indistinguishable. The four-photon case is

$\{\rho_x\}$ 09

where $\{\rho_x\}$ 10 is the purity of the squeezed states, and for fixed purity $\{\rho_x\}$ 11 decreases exponentially fast in $\{\rho_x\}$ 12 (Shchesnovich, 2021). In multipath interference with a path detector, distinguishability and coherence satisfy

$\{\rho_x\}$ 13

while

$\{\rho_x\}$ 14

so entanglement quantitatively links path predictability to distinguishability (Qureshi, 2020). Lahiri and collaborators further show that in a two-path single-photon interferometer partial polarization arises only when both an erasable distinguishability $\{\rho_x\}$ 15 and an inerasable distinguishability $\{\rho_x\}$ 16 are present, with

$\{\rho_x\}$ 17

(Lahiri et al., 2015).

In boson sampling and time-resolved photodetection, distinguishability due to mixed single-photon states is not removable by detector design. For Gaussian photons with timing jitter, the relevant symmetric-subspace weight is

$\{\rho_x\}$ 18

and the total-variation distance from the ideal indistinguishable distribution is bounded by

$\{\rho_x\}$ 19

The comparison between slow detectors and fast time-resolved detectors yields the same dependence on single-photon purity, leading to the conclusion that distinguishability due to mixed states is an intrinsic property of photons, whatever the photodetection scheme (Shchesnovich et al., 2020).

Finally, outside quantum-state preparation proper, the same conceptual pattern appears for stochastic processes. For two continuous-time Markov processes $\{\rho_x\}$ 20 and $\{\rho_x\}$ 21, the trajectory-ensemble distinguishability is the path-space KL divergence

$\{\rho_x\}$ 22

which decomposes into initial, edge-wise, and temporal contributions. Each edge contribution has the nonnegative “current–force” form

$\{\rho_x\}$ 23

where the force is a KL divergence between exponential waiting-time densities (Pagare et al., 2024). This is not a preparation-history problem in the narrow quantum sense, but it shows that distinguishability can also be resolved into local dynamical contributions rather than only global histogram comparisons.

Across these extensions, a consistent theme emerges: distinguishability is constrained not just by state overlap, but by the permitted operational architecture—steering, no-signaling, cloning structure, ontological access, locality class, detector resolution, or trajectory-level observation.