Measurement No-Restriction in GPTs

• Measurement no-restriction is a principle in GPTs where every mathematically well-defined effect is physically realizable, linking the state space directly to measurement outcomes.
• Relaxing this hypothesis introduces intrinsic noise, enables self-dualization procedures, and alters the construction of joint state spaces in composite systems.
• Studies show that strict enforcement or relaxation of no-restriction influences Bell correlations, sequential measurement universality, and operational limits in quantum and post-quantum theories.

Measurement no-restriction is most commonly the no-restriction hypothesis of generalized probabilistic theories (GPTs): once a state space is fixed, every mathematically well-defined effect compatible with that state space is taken to be physically realizable. In this sense, measurements are determined by states through duality, with no further independent limitation on accessible effects. Subsequent work turned this hypothesis into a fault line in quantum foundations: some post-quantum correlation sets are incompatible with it, some operational restrictions act at the level of whole meters rather than individual effects, and several quantum and post-quantum results show that measurement behavior can be constrained by nonlocality, symmetry, or sequential structure even when the basic effect space is left otherwise unconstrained (Janotta et al., 2013).

1. Formal definition in GPTs

In the GPT framework, a system is specified by a convex state space $S$ inside a finite-dimensional real vector space $V$, a positive cone $V_+$, a unit effect $u$, an effect space, and a set of meters. Effects are affine maps $e:S\to[0,1]$, and a meter $A$ with finite outcome set $\Omega_A$ is a map $x\mapsto A_x$ with $A_x\in E(S)$ and $\sum_{x\in\Omega_A}A_x=u$. In equivalent cone language, the full mathematical effect space is

$V$0

while in another standard notation the hypothesis is written as $V$1, meaning that the physical effect space coincides with the dual of the state space (Janotta et al., 2013).

Operationally, the no-restriction hypothesis says that every linear functional that gives probabilities on all normalized states is an allowed measurement outcome. In the formulation used by Janotta and Lal, this is

$V$2

and the unrestricted case sets the physical effect set equal to $V$3. Quantum and classical theory satisfy this hypothesis in the usual formulations: all POVM effects $V$4 are allowed in quantum theory, and all probability-valued affine functionals are allowed in classical simplex theories (Janotta et al., 2013).

The hypothesis is strong because it removes any independent measurement postulate beyond state-space geometry. A physically restricted theory instead has

$V$5

so that some mathematically admissible effects are not physically realizable. This separates the specification of states from the specification of measurements, and turns “which effects exist?” into an independent structural question rather than a consequence of convex duality (Janotta et al., 2013).

2. GPTs without the no-restriction hypothesis

Janotta and Lal generalized the GPT framework precisely to allow $V$6, arguing that the no-restriction hypothesis is not physically motivated (Janotta et al., 2013). In that generalized setting, the effect set must still satisfy basic consistency conditions: inclusion of the unit effect, closure under complements $V$7, closure under coarse-graining, and stability under allowed transformations. What changes is that the effect cone need no longer be the full dual cone of the state cone.

One immediate use of restricted effects is to model intrinsic measurement noise. In noisy boxworld, the state space is kept fixed while the nontrivial extremal effects are shrunk toward $V$8,

$V$9

For $V_+$0, no nontrivial effect is certain on any state; only the unit effect remains probability one on all normalized states. This is a direct example of a theory in which measurement limitation is fundamental rather than a defect of experimental implementation (Janotta et al., 2013).

Relaxing no-restriction also enables a self-dualization procedure. The effect cone can be truncated so that, after a suitable change of representation, it coincides with the state cone. Janotta and Lal used this to construct self-dualized polygon theories and to show that the convex closure of Spekkens’ toy theory fits naturally into this framework. In that analysis, the unrestricted version of the Spekkens toy theory is boxworld, while the restricted version becomes local because the effect space is smaller than the full dual. The same work argued that self-dualized theories share many features of quantum theory, such as obeying Tsirelson’s bound for the maximally entangled state (Janotta et al., 2013).

The bipartite state space also changes. Without no-restriction, the usual maximal tensor product is too permissive because positivity on restricted local effects need not guarantee that conditional states remain in the original local state spaces. Janotta and Lal therefore introduced a generalized maximal tensor product,

$V_+$1

which characterizes exactly those joint states whose conditional states are valid on both sides (Janotta et al., 2013).

3. Correlation sets, Bell functionals, and almost quantum theory

A major use of measurement no-restriction is to test whether a proposed correlation set can arise from any GPT at all. In this direction, the paper on almost quantum correlations proved that no generalized probabilistic theory satisfying the no-restriction hypothesis can reproduce the almost quantum set $V_+$2 (Sainz et al., 2017). The result is not merely that $V_+$3 is post-quantum; it is that any GPT whose correlations are exactly, or even very close to, $V_+$4 must impose a nontrivial restriction on measurements.

The argument is formulated in terms of normalized Bell functionals (NBFs), linear functionals $V_+$5 on behaviors such that $V_+$6 for every behavior allowed by the theory. Under no-restriction, each NBF corresponds to a valid effect on the joint system, and complete families of NBFs define Bell measurements. Because such Bell measurements are then legitimate physical measurements, they must be composable: if $V_+$7 is a Bell measurement and $V_+$8 is another NBF, the composite

$V_+$9

must again be a valid NBF (Sainz et al., 2017).

The almost quantum set fails exactly this closure requirement. The paper constructs explicit almost quantum NBFs $u$0 such that the composed functional $u$1 is not an almost quantum NBF; numerically,

$u$2

Since a valid NBF cannot take negative values on allowed behaviors, this contradicts the no-restriction implication that Bell measurements should compose without leaving the effect space. A theory realizing $u$3 must therefore have $u$4 (Sainz et al., 2017).

This conclusion is complemented by work on joint states without no-restriction. Janotta and Lal later showed that the generalized maximal tensor product reduces to the standard maximal tensor product whenever at least one side obeys the no-restriction hypothesis, and that under certain linear-bijection restrictions on effect spaces, relaxing no-restriction does not allow stronger non-locality even if the generalized maximal tensor product admits new joint states (Janotta et al., 2014). Together, these results separate two issues: the geometry of allowed states and the independent operational content of the measurement set.

4. Operational restrictions on meters

The no-restriction hypothesis is effect-level: it asks whether every admissible effect exists. Heinosaari and coauthors reframed the problem at the level of whole meters, asking what restrictions on accessible measurements are operationally meaningful. They argued that all operational restrictions must be closed under simulation, where simulation means mixing and classical post-processing of meters (Filippov et al., 2019).

This leads to a broader classification than the state/effect duality alone captures. The paper distinguishes three classes of operational restrictions: restrictions on meters originating from restrictions on effects; restrictions on meters that do not restrict the set of effects in any way; and all other restrictions. The first class is fully characterized and linked to convex effect subalgebras. The second class shows that severe physical limitations can remain even when all effects are accessible; the example given is unambiguous discrimination of pure quantum states via effectively dichotomic meters. The third class contains further physically meaningful restrictions not reducible to either of the first two (Filippov et al., 2019).

This shifts the concept of measurement no-restriction from a statement about individual effects to a statement about the closure properties of the measurement architecture itself. A plausible implication is that effect-level no-restriction does not settle the accessibility of composite measurement procedures. In that sense, the hypothesis $u$5 is only one possible notion of unrestricted measurement, and not the most operationally complete one.

5. Sequential universality and symmetry-based limitations in quantum theory

Within standard quantum mechanics, there are both strong no-restriction results and strong restriction theorems. The clearest positive result is the universality of certain sequential measurements. For every observable $u$6, there exists an $u$7-channel $u$8 with the universal property: for every observable $u$9 jointly measurable with $e:S\to[0,1]$0, there is an observable $e:S\to[0,1]$1 such that

$e:S\to[0,1]$2

for all states $e:S\to[0,1]$3 and outcomes $e:S\to[0,1]$4. The construction uses a minimal Naimark dilation $e:S\to[0,1]$5 of $e:S\to[0,1]$6 and the channel

$e:S\to[0,1]$7

This means that, with the right first measurement scheme, sequential implementation imposes no additional limitation beyond joint measurability itself (Heinosaari et al., 2014).

The converse is equally important: not every $e:S\to[0,1]$8-channel is universal. The paper characterizes when a second observable $e:S\to[0,1]$9 can be realized after a fixed first measurement $A$0: this is possible iff the conjugate channel $A$1 is a $A$2-channel. In particular, Lüders measurements need not be universal for all jointly measurable partners, even in low-dimensional examples (Heinosaari et al., 2014). Measurement no-restriction in the sequential sense is therefore instrument-dependent, not automatic.

A contrasting restriction theorem is the generalized Wigner–Araki–Yanase analysis. Although every POVM can be realized as the measured observable of some normal measurement scheme, symmetry and compatibility constraints can forbid exact realizability within a given coupling. The incompatibility-based reformulation states that if the evolved pointer $A$3 commutes with a selfadjoint operator $A$4, then the measured sharp observable $A$5 must commute with

$A$6

In the additive conserved case $A$7, this recovers the standard WAY-type constraint $A$8 under the Yanase condition, but the formulation goes beyond additivity and even beyond conservation laws. The paper also derives POVM-level norm bounds such as

$A$9

showing that approximate measurability is limited by the incompatibility between the interaction and the constrained quantity (Tukiainen, 2016).

6. Measurement-disturbance constraints and broader operational usages

A different line of work uses “measurement no-restriction” more loosely, as the question whether measurement transformations can be chosen arbitrarily once state statistics are fixed. In a no-signaling box framework, Oppenheim–Wehner-type preparation uncertainty does not follow from no-signaling and Bell nonlocality alone, but a quantitative measurement uncertainty relation does. For a gentle measurement of a Bell-relevant observable $\Omega_A$0, with information gain $\Omega_A$1, average total disturbance $\Omega_A$2, Bell relevance $\Omega_A$3, and locality deficit $\Omega_A$4, the derived bound is

$\Omega_A$5

In the extremally nonlocal case $\Omega_A$6, any nonzero information gain about $\Omega_A$7 forces disturbance of other observables. From a measurement no-restriction viewpoint, this shows that not all conceivable measurement maps are compatible with no-signaling plus a given degree of nonlocality (Łodyga et al., 2017).

A still broader, outcome-based usage appears in model-independent metrology. If one assumes essentially no dynamical or structural constraints beyond a discrete outcome set $\Omega_A$8 and a probability distribution satisfying normalization and unbiasedness, there is a unique minimum error distribution supported on the two neighboring outcomes around the true parameter value. Its worst-case RMSE scales as $\Omega_A$9, which the paper interprets as a model-independent resource count $x\mapsto A_x$0. In that setting, the no-restriction stance is not about GPT duality but about allowing arbitrary probability models over a fixed outcome alphabet (Bharath et al., 2013).

A recent device-independent randomness result adds a network version of the theme. For bipartite and tripartite qubit scenarios, maximal global randomness can be certified without Bell inequalities, and the corresponding measurement incompatibility constraints are asymmetric: if one party’s measurements are sufficiently incompatible, the others’ can be arbitrarily close to compatible. A plausible implication is that “unrestricted” measurement freedom may be distributable across parties rather than required uniformly at every node (Zheng et al., 23 May 2025).

Taken together, these developments make measurement no-restriction a layered concept rather than a single axiom. In its strict GPT sense, it is the identification of physically allowed effects with the full dual set. In operational refinements, it concerns closure of meter sets under simulation. In quantum measurement theory, it appears as a question about whether sequential structure or symmetry imposes extra limitations beyond compatibility. And in post-quantum and metrological settings, it becomes a probe of which measurement transformations or probability models remain consistent with deeper operational principles.