WorksPhysics and Astronomy
Quantum Reference Frames, Locality, and Microcausality in Quantum Field Theory
Abstract
As a consequence of the relativistic spectrum condition, quantum reference frames for translations on Minkowski space are never described by projection-valued measures, and in many cases fail a weaker spacetime localisation property known as the norm- condition. This has several implications for the local and observable structure of quantum field theories described relative to even the best available frame: for a large class of frame states, strict locality is replaced by almost locality, and microcausality by commutators that decay rapidly with increasing separation, corresponding to an effective smearing of the light cone by the localisation density of the frame state. Operations built from finitely many almost local Kraus operators, for example nonselective measurements of relativised local observables, satisfy an approximate no-signalling bound, with a residue attributable to the localisation uncertainty of the frame. Whether causality in its strongest operational form – non-disturbance under all physically realisable operations – survives relativisation is left open. An upshot of our finding is that quantum field theories formulated relative to a quantum reference frame cannot rest on the traditional axioms, and some general relational axiomatisation must replace at least some of them by taking into account physically realistic properties of the frame.
Introduction
If a physical system possesses a symmetry, every preparation procedure is accompanied by an orbit of others, related by the symmetry. The same applies to every measurement procedure. Specifying the nature of the procedure only specifies the orbit, not a particular instance within it. Without a way to break the symmetry and identify which preparation procedure on the orbit was used, the experimenter is restricted to those measurement procedures that are invariant with respect to the symmetry. In other words, physical observables are invariants.
One way to extend the range of available observables is to introduce a reference frame - a physical system that shares the same symmetry as the original, for then the invariant observables of the joint system become available. For example, while the position of a single particle in Euclidean space is not translationally invariant, the relative position of the particle with respect to a reference particle is invariant under common translations of both particles. If the reference particle is in a sufficiently well localised state, the relative position behaves, to good approximation, like the position of the original particle; the relative position is the translation invariant object that is measured in practice, with the absolute position being a good proxy in the limit of a perfectly localised reference.
Let us describe how this programme is implemented within quantum theory. Here we consider a quantum system , and a reference , itself a quantum system. For this reason, is called a quantum reference frame (QRF). We suppose that both and admit a symmetry described by a symmetry group , which is assumed to be locally compact. The mathematical description of and , and of , includes quantities that would be observable in the absence of the symmetry. For want of a better term, we temporarily describe these as ‘potential observables’. If is a quantum-mechanical system, one may always find an ideal QRF so that the algebra of potential observables of is isomorphic to an algebra of invariant relative observables of (e.g. [1, 2, 3, 4, 5, 6]), and this algebra can be recovered arbitrarily precisely via a physically well motivated conditional expectation, which gives the description of the system contingent on the specification of the state of the frame. This indicates that for quantum mechanical systems, it is to some degree a matter of taste whether to include the frame in the description, or not: in this sense, all potential observables become truly observable. Reference frames that are described by quantum mechanical systems are now well studied (e.g., [1, 4, 6, 7, 8, 9, 10]).
The situation changes when considering quantum field theory, where the axiomatic basis for the theory puts more stringent conditions on the frame; here the ideality property may be lost, and the above-mentioned isomorphism is typically no longer tenable. More precisely, given a locally compact group with a strongly continuous unitary representation in some Hilbert space , a (principal) quantum reference frame is defined to be a system of covariance based on , that is, the triple , where is a (possibly subnormalised) positive operator valued measure satisfying the covariance condition
From here, the relativisation map (e.g. [2, 3]) is given as
where the integral is understood in an appropriate manner ([3, 11, 12]).
Relativisation is understood as giving a description of relative to the frame ; is a normal completely positive subunital map whose image lies in the invariant algebra , and an injective -homomorphism if is projection-valued. Those operators in not invariant under are understood as not representing bona fide physical quantities, and acquire their physical meaning only relative to an appropriate frame. Given that one of the motivations for introducing quantum reference frames is based on the view that the frame should be described by the same theory as the system of interest, it is important to construct quantum reference frames appropriate to quantum field theory, ideally that are themselves QFTs.
In this paper, we study quantum reference frames in Minkowski space that are covariant under (4-)translation, hence with . It is known (due to a result of Toller, [13]; see also [14]) that no POVM satisfying the covariance requirement (I.1) plus the spectrum condition can be projection-valued, and therefore that multiplication is not preserved by in general, and some extra ‘fuzziness’ is present. Moreover, these POVMs typically do not satisfy a weaker ‘localisability’ condition, known as the norm-1 property, which allows the system algebra to be recovered up to arbitrarily good approximation under a natural conditional expectation. Our assumptions are weaker than those of Wightman/AQFT in Minkowski, so although our QRFs are not constructed from systems of fields, they would certainly apply in that setting.
Our main findings are as follows. Under the spectrum condition, no translation-covariant frame is projection-valued (Corollary II.2, following Toller [13]), and a quantitative bound rules out even the norm-1 property for a large class of frames (Theorem II.7). Relative to a frame whose localisation density is a Schwartz function, relativised local observables are almost local (Proposition III.4; see [15, 16]), their commutators at causal separation are (Lemma III.5), and the sharp light cone is replaced by its smearing with the autocorrelation of the localisation density (Remark III.7). Causality nonetheless survives in some form: relativisation preserves compatibility (Proposition IV.2), and operations with finitely many almost local Kraus operators, for instance the Lüders measurements of relativised local observables, disturb causally distant relativised observables only at order (Proposition IV.4, Corollary IV.7), locality of operations being, under Haag duality, equivalent to non-signalling (Remark IV.5). Finally, accurate recovery of local observables requires the frame to resolve small translations, and hence to carry correspondingly high energies (Proposition IV.8).
A complementary, axiomatic development has recently been given by Fedida and Glowacki [17], who formulate relational covariance and causality conditions for Poincaré-covariant frame observables. Their causality results are built on frame preparations with compactly supported localisation distributions; however, as follows from Proposition II.1 below, no such preparations exist once the frame itself satisfies the spectrum condition usually assumed in QFT [15, 18]. By contrast, in this work we enforce the spectrum condition at the expense of microcausality but, as we shall see, this does not necessarily entail acausal signalling.
Quantum Reference Frames for Translations in Minkowski space
Here we present Toller’s construction [13] of translation-covariant POVMs on (here, 1 + 3-dimensional) Minkowski space , adding some further particulars where they may be helpful. Throughout, the Minkowski product is , and Fourier transforms of vector-valued functions are normalised so that , unitary from to .
Let be a complex, separable Hilbert space carrying a strongly continuous unitary representation of translations
whose generators satisfy the spectrum condition - that the joint spectrum of the is contained in , the closed forward causal cone. We seek a POVM satisfying:
Note that the usual normalisation is too strong and is replaced by subnormalisation (II.3). If is interpreted as an effect associated with a certain event occurring in (or a measurement returning a value in ), then is the effect corresponding to the event occurring at all (or that the measurement succeeds in returning a result). As will be seen below (Theorem II.4), vanishes on the singular subspace of the energy-momentum spectral measure, which contains the vacuum and any single-particle states. Vectors in therefore do not define events in this sense; more generally, writing , one has , so the singular component of a state contributes only to the ‘non-occurrence event’.
Furthermore, in any QFT with a translation-invariant vacuum vector that is separating for the local algebras (as guaranteed by the Reeh–Schlieder theorem), no nonzero can belong to a local algebra, and under the cluster property not even to the quasilocal algebra [19]. The next proposition collects two related facts which together are the foundation of much of what follows. Both are due to Toller [13]; for part (ii) we write the proof out in detail.
Proposition II.1. Let be a POVM satisfying (II.2)–(II.3).
(i) For every nonempty open set , the positive operator is non-zero.
(ii) For every nonempty open set and every ,
Proof. (i) Suppose for some nonempty open . Choose such that contains an open ball of radius . Countably many translates of cover , and by (II.2),
Countable additivity of then gives , contradicting (II.3). (ii) Suppose for some nonempty open , and choose nonempty open sets with (two balls of half the radius of a ball contained in suffice). For we have , and therefore monotonicity of the POVM together with (II.2) gives
so for all . Now, fix any and define
The spectrum condition implies that is well defined as an -valued holomorphic function of in the lower tube , where denotes the interior of , with continuous boundary values on (by strong continuity of and uniform boundedness of on ). Hence is holomorphic in , continuous on the boundary, and by (II.5) vanishing on . Setting gives a holomorphic function on the upper tube whose boundary values agree with those of on . The edge-of-the-wedge theorem [18] then provides a holomorphic continuation across , giving a single holomorphic function on a connected open set containing . As this continuation vanishes on and the domain is connected, it vanishes identically. In particular, for every , and since was arbitrary, for every . Equivalently,
Applying the covering of part (i) to now yields .
Part (ii) of Proposition II.1 is enough to rule out projection-valued :
Corollary II.2. No POVM satisfying (II.2)–(II.3) is projection-valued.
Proof. Suppose, for a contradiction, that is a PVM. For any nonempty open and any , , so by Proposition II.1(ii), . Thus , equivalently . Monotonicity gives the reverse inclusion. As both are projections (since is assumed PVM), for every nonempty open . Now choose any bounded open ; then is open and nonempty, so Proposition II.1(i) gives . But , a contradiction.
We now turn to the structure of translation-covariant POVMs. The spectral theorem for the commuting self-adjoint operators provides a direct integral decomposition
where is supported in . The decomposition induces a corresponding orthogonal decomposition ; the vacuum and any single-particle sectors are contained in , on which vanishes. After rescaling the inner products on if necessary, the absolutely continuous part can be written as
where . In a theory with mass gap, with , one has .
Lemma II.3 (Toller, [13 Prop. 2]). For integers write , and let , equipped with the Fréchet topology generated by the seminorms ; is dense and translation-invariant in . For every , the complex measure is absolutely continuous with respect to Lebesgue measure on , and is represented by a smooth density :
The density is sesquilinear in , positive in the sense that , and satisfies the covariance relation
together with the bound
In particular, .
The next theorem gives a basis-free formulation of Toller’s representation of the density of translation-covariant POVMs, which can be cast as a Naimark dilation.
Theorem II.4. Let be a POVM satisfying (II.2)–(II.3). Then vanishes on , and there exist a Hilbert space and a measurable family of contractions
such that the map
defines a contraction , and the POVM is the (subnormalised, since is a contraction) compression of the canonical dilation on :
where denotes multiplication by the indicator function on . Equivalently,
for all and Borel , where is extended by zero on . Conversely, any such measurable family of contractions, not vanishing almost everywhere, defines a translation-covariant POVM (vanishing on ) via (II.10)–(II.11). Moreover, (the projection onto ) if and only if almost every is an isometry.
Proof. Translating Toller’s component formula [13 §3] into our notation, the density of any translation-covariant POVM satisfying (II.2)–(II.3) vanishes when either argument lies in , and on takes the form
for some Hilbert space and measurable field of contractions in the usual direct-integral sense, with a.e. on . Defining by (II.10), this reads . Plancherel’s theorem gives
so is a contraction, and integration of over gives (II.12), equivalently the Naimark-dilation formula (II.11). In particular ; hence iff for every , which by the Plancherel identity above is equivalent to a.e., i.e., to almost every being an isometry. The converse direction - that any measurable field of contractions, in the usual direct-integral sense, defines a translation-covariant POVM via (II.10)–(II.11) - is a direct computation.
If is chosen minimal, in the sense that the ranges of the together span a dense subspace, the pair is unique up to a unitary on .
Remark II.5. On , the equivalence class has a smooth pointwise representative given by
to be understood weakly, or as the value at of the Fourier transform on . In general does not extend to a bounded operator . The boundedness or otherwise of will play a structural role in the next subsection.
Localisability
Theorem II.4 allows us to address whether translation-covariant POVMs admit the norm-1 property.
Definition II.6. A POVM satisfies the norm-1 property if for every nonempty open .
This has two equivalent formulations (i) for every nonempty open there is a sequence of unit vectors for which (ii) for every nonempty open and any there is a unit vector for which . The norm-1 property is strictly weaker than being projection-valued, but characterises those observables for which arbitrarily sharp localisation in any open region is possible. By Corollary II.2, no translation-covariant POVM on is projection-valued, so it is natural to ask whether they may still satisfy the norm-1 condition.
Theorem II.7. Let be a translation-covariant POVM satisfying , and let be the contraction as given above, so that . Suppose that the map
initially defined on the domain , extends to a bounded operator . By translation covariance, this is equivalent to asserting that the pointwise representative of is given by
for all and , with bounded on . Then, for every Borel set of finite Lebesgue measure,
Consequently,
and does not have the norm-1 property.
Proof. For the asserted equivalence: given , choose with ; then uniformly, while in , so by Remark II.5 the continuous function represents . Now let with . By the hypothesis and unitarity of ,
for all . Hence
Taking the supremum over (and using that vanishes on ) gives (II.13). If had the norm-1 property, every nonempty open would satisfy ; choosing a nonempty bounded open with contradicts the estimate (and if the conclusion is immediate), so does not have the norm-1 property.
A simple, explicit sufficient condition for the hypothesis of Theorem II.7 is square-integrability of the Hilbert–Schmidt norm of in momentum space.
Corollary II.8. Suppose
Then the evaluation map extends to a bounded operator with . Consequently for every Borel of finite Lebesgue measure and does not have the norm-1 property.
Proof. Choose an orthonormal basis of and let be the section . By Fubini and the definition of the Hilbert–Schmidt norm, we have
For and ,
and Cauchy–Schwarz gives . Summing,
so the hypothesis of Theorem II.7 holds with .
Corollary II.9. If and for some with a.e. and , then is bounded in absolute value by , and
for every Borel with . Thus, does not have the norm-1 property.
Remark II.10. Theorem II.7 and Toller’s [13 Prop. 5] are complementary. Theorem II.7 shows that boundedness of , which is guaranteed, for instance, by condition (II.14), obstructs the norm-1 property being satisfied, with the quantitative bound (II.13). Toller gives a sufficient condition for the norm-1 property to hold, given by the asymptotic constancy of as , under which a translation-covariant POVM is in Toller’s terminology definite, which coincides with the norm-1 property. In particular, any definite translation-covariant POVM has unbounded , by Theorem II.7. These results thus correspond to opposite regimes of the asymptotic behaviour of —rapid decay excludes arbitrarily sharp localisation, and asymptotic constancy guarantees it. The intermediate regime, as well as the question of whether definite POVMs arise from physically natural QRFs in QFT, appears to be open.
Relative Observables
Let be a covariant POVM of the form described above, and let be a von Neumann algebra with a -weakly continuous action . For set
where the integral is understood weakly; the matrix elements determine uniquely. Existence, normality, complete positivity and subunitality of are established in [11] for an arbitrary von Neumann algebra and subnormalised (the generality needed in QFT, where local algebras are typically not of type I) provided extends to a normal map on , as it does here by Theorem II.4; explicitly, with denoting multiplication by .
Since and , the map is subunital in general. For , put , where . Then
Here
is an -density for arbitrary , and agrees with Toller’s smooth density when . For and , we write ; this is a subprobability density, and we refer to as a restricted relative observable.
Lemma III.1. If , then the essential support of is noncompact.
Proof. Suppose that has compact essential support. Then there is a compact set such that for almost every . Since
it follows that for almost every . Thus is a compactly supported element of . By Paley-Wiener, the Fourier transform of a compactly supported -function has an entire -valued extension to . On the other hand, by the Toller representation,
as an element of , with extended by zero outside . Hence vanishes on the nonempty open set . Analyticity forces to vanish identically, so . Consequently
which proves the result.
The decay of at infinity controls the localisation properties of the restricted relative observables. The next proposition gives a simple sufficient condition for rapid decay.
Proposition III.2. Let be given by for and by otherwise. If , then .
Proof. By Theorem II.4, as an -Fourier transform. Since the inverse Fourier transform is an automorphism of , one has . Moreover
For every multi-index ,
where is a multinomial coefficient. The Schwartz estimates for , together with Cauchy–Schwarz, imply the corresponding estimates for . For example,
where Schwartz seminorms are defined as usual, for , replacing by for .
Definition III.3. Let be a local net of von Neumann algebras with quasilocal algebra on Minkowski space . We assume throughout that the local algebras possess a common unit , (see [20 p. 849]). An element is almost local if, for some and future-pointing unit timelike vector , there is a family of local operators
where is the open double-cone with tips at , such that
The local approximant belongs to the local algebra of a double-cone about in the rest frame of . If is almost local then there is a family of local approximants obeying (III.10) and (III.11) for every and future-pointing unit timelike . Note that the double-cones have the property .
Proposition III.4. Let be bounded and local, with bounded. If , then
is almost local.
Proof. Let , fix any unit future-pointing timelike vector and suppose that is sufficiently large that , setting for all smaller values of . Choose with , for , and for , and set
understood in the weak sense. Since and is supported in , isotony gives . Moreover,
Since is Schwartz, the last integral is for every as (independently of the chosen ).
Lemma III.5. Let be almost local and fix any future-pointing timelike unit vector . Then, for every , there is such that, for all and all for which
i.e., belongs to the causal complement of , one has
In particular, as , uniformly in satisfying (III.15).
Proof. Fixing any and any future-pointing unit timelike vector , let be local approximating families for the . Now let and let with spacelike to . By covariance, , so locality gives
because every vector in is spacelike. Hence, writing
and using that are isometric, we have
from which, by (III.11) and , we arrive at the claim for ; for it holds trivially, since the left-hand side is at most .
Corollary III.6. Let be bounded local observables, with bounded, and suppose that . Put
Then the conclusion of Lemma III.5 holds for : for every there is such that
for all and satisfying (III.15).
Proof. This follows from Prop. III.4 and Lemma III.5.
At the formal level, we observe the following. If is a field, , and with , then
where
Thus a compactly supported test function is replaced by a smeared test function. If is Schwartz, then so is , and the resulting field is almost local in Haag’s sense [15]. This explains formally why exact microcausality is replaced by rapidly decaying commutators. To avoid domain questions for unbounded fields, we now pass to the Weyl algebra. For real-valued , let denote Weyl operators satisfying
where and are retarded/advanced Green operators, and suppose translations act by . Define
Then . The Weyl relations give
and hence
Note that depends only on . Even if and have causally disjoint supports, the integrand will be nonvanishing for suitable , so microcausality fails. Let
On the integrand vanishes. Therefore
This gives a bounded, quantitative version of the preceding formal picture: exact microcausality is replaced by a bound controlled by the probability that two independent frame translations fail to preserve causal disjointness.
Remark III.7. The bound (III.29) entails a statement about the operationally accessible causal structure. For let be the set indicator function corresponding to the situation that and fail to be causally disjoint, and let
denote the autocorrelation of the localisation density of the frame state . Then, for translated , , a change of variables in (III.29) gives
Relative to the frame, the sharp indicator of causal connectedness is replaced by its convolution with : the boundary of the light cone is smeared by the localisation uncertainty of the frame state. By Lemma III.1, , and hence , has non-compact support, so the smearing cannot be removed even in principle: no frame state restores a sharp causal boundary (although for there is rapid decay).
Commutativity, Compatibility and Disturbance
A simple way to appreciate the operational content of the microcausality condition in standard QFT rests on the Lüders theorem [21, 22]: A nonselective Lüders measurement of a discrete sharp observable , resulting in a state update , leaves the statistics of a second sharp observable undisturbed in every state if and only if . For observables with continuous spectrum, this holds for every discretisation [23]. For general POVMs commutativity is replaced by compatibility, which implies commutativity only in special cases.
Definition IV.1. Consider POVMs and on measurable spaces and respectively (which for the purposes of this paper, we may assume are given as ). and are called compatible if there is a joint POVM on such that for all and all , we have and .
This operational definition of compatibility requires the existence of joint probability distributions for and in all states, with the distributions for and obtained as marginals. If at least one of and is projection valued, the compatibility of and is equivalent to their commutativity (in the sense that for all and ); however, there are many examples of POVMs that are compatible but fail to commute [24, 25].
For POVMs the relationship between compatibility, (non-)disturbance and commutativity is nuanced. If some measurement of leaves undisturbed, then and are compatible (the sequential measurement, first and then , supplies a joint POVM with marginals and ), but the converse fails, and there are compatible pairs for which every measurement of one POVM disturbs the other [26]. Compatibility here is therefore not a statement of relativistic causality; the latter depends on the specifics of the available measurements even for compatible observables. The dependence of causal behaviour on how a measurement is implemented, rather than on the measured observable alone, is already present in ordinary QFT, where microcausality holds exactly. Let be sharp and local and consider the instrument that measures and, upon obtaining outcome , re-prepares the system in state . Here the instrument is then , and in the Heisenberg picture its nonselective channel is
This instrument measures with the same outcome statistics as the Lüders instrument, yet for essentially every , independently of the localisation of — this measurement signals, even though commutes with every causally disjoint observable. Indeed, we see that , so implies , which consists only of multiples of the unit under the extended locality property that holds for local nets of von Neumann algebras under standard conditions [27].
Causality is therefore not a property of the measured observable, but of the measurement operation, and must be formulated in terms of the localisation properties of the operations implementing the measurement. One way to proceed is to assume that the measurement is conducted by a probe QFT locally coupled to the system QFT with the coupled theory also described as a QFT [28], but for the moment we take a more general viewpoint. In Definition IV.3, we give a notion of locality for operations and show that, in the relativised theory, it yields an operational form of microcausality whose accuracy is set by the frame. First, however, we note that relativisation preserves compatibility.
Proposition IV.2. Let and be compatible POVMs and let be a normal unital completely positive map. Then and are compatible.
The proof follows immediately from noting that if is a joint POVM for and , then is a joint POVM for and [29]. The same conclusion holds for subnormalised POVMs and subunital normal completely positive maps, after making the appropriate changes.
Applying the completely positive map to a joint PVM makes the contrast between commutativity and compatibility explicit. For causally disjoint and the Weyl operators and commute, and there is a joint PVM whose marginals are the spectral measures and of and . By Proposition IV.2, is a joint POVM for the compatible pair and , whose first moments are the non-commuting operators and , as seen by the calculation
Thus relativisation (and restriction) may destroy commutativity, but nonetheless preserves compatibility.
We now turn to disturbance. Signalling concerns the nonselective channel defined by an instrument, and so we work directly at the level of channels. In addition to the localisation of observables, the localisation of operators is also part of the philosophy of local quantum physics; operations implemented by elements of a local algebra appear already in Haag and Kastler [20]. That the locality of an observable does not by itself ensure the causal behaviour of its measurement was stressed by Sorkin [30] (see also [31]), and the identification of operations that are both causal and physically realisable has since been pursued from several directions: measurement schemes in which the system is coupled locally to a probe [28, 32]; the direct analysis of state-update rules for scalar fields [33]; the construction of regularised, non-signalling measurements of field operators and their spacetime-extended generalisations [34, 35, 36]; and the general study of causal and realisable local operations [37].
Definition IV.3. By an operation we mean a normal unital completely positive map (a channel, in the Heisenberg picture). Every operation admits a Kraus decomposition
with both sums converging -weakly [38]. An operation is local in a bounded region if it admits such a decomposition with for every , and almost local if it admits one in which every is almost local. The Kraus decomposition of a given operation is not unique; locality and almost locality require one decomposition of the stated form; Remark IV.5 below pertains to every decomposition.
We caution immediately that ‘locality’ of an operation does not imply that it can be realised by local physical interactions within the given region. More comments on this appear below. (Note also that our terminology here differs from that of [37], in which an instrument is said to be local to a compact set if its measurement channel acts trivially on the algebra associated with the causal complement of ; see however Sec. IV of [37] for situations in which their definition implies that the instrument is local, in our sense, to any region containing .) The identity channel (a single Kraus operator ) is local in every region. The Lüders channel of a discrete POVM in — effects with , completed by — is local in : its Kraus operators are for , since the continuous functional calculus preserves . Thus the Lüders channel of every discrete local POVM is local in the sense of Definition IV.3. The measure-and-prepare channel of (IV.1) is, by contrast, local in no bounded region, as part (i) of the following proposition shows.
Proposition IV.4. Let be an operation with Kraus operators as in Definition IV.3.
(i) If is local in and is causally disjoint from , then for every .
(ii) Suppose is almost local with a finite Kraus decomposition , let be almost local, fix a future-pointing unit timelike vector , and for let denote the translated operation. Then, for every , there is such that, for all and all satisfying (III.15),
Proof. Since , with all sums converging -weakly,
For (i), take ; each , so every term vanishes by microcausality. For (ii), the decomposition is finite and , so, using that has Kraus operators ,
and Lemma III.5 bounds each term by , uniformly in as in the statement.
Remark IV.5. Part (i) admits a converse, showing that the notion of locality in Definition IV.3 is forced by causality. Suppose for every and every bounded causally disjoint from . For any Kraus decomposition of and any such ,
since and also lie in ; a vanishing sum of positive terms, whence for every [39, 40]. Thus each lies in the commutant of the von Neumann algebra generated by the local algebras of bounded regions causally disjoint from . Where Haag duality holds for (for instance, when is a double cone in a theory satisfying Haag duality), , so every Kraus operator of lies in : the operations that cannot signal out of are precisely those local in . This observation appears as Proposition 49 of [37]; the underlying fact—that a channel acting as the identity on a -algebra has Kraus operators in its commutant, is due to Lindblad [41], whose proof rests on the equality situation of the Schwarz inequality for completely positive maps [39]. The same identity, applied to the spectral projections of a sharp observable, yields the nontrivial part of the Lüders theorem.
We note that Proposition IV.4 and Remark IV.5 concern two regions: as Sorkin observed [30] (see [37 Sec. II.E] for a formulation in terms of channels), locality of the Kraus operators alone does not preclude superluminal signalling in protocols involving three or more regions, and the characterisation of the operations that are both physically realisable and fully causal is delicate [28, 33, 37]; these matters lie beyond the scope of this paper, but it is worth emphasising that the same issues also arise in classical field theories [42].
To bring measurements of relativised observables within the scope of part (ii), it remains to verify that the Kraus operators of their Lüders channels are again almost local.
Lemma IV.6. Let be an almost local effect. Then and are almost local.
Proof. Fix and a future-pointing unit timelike vector , and let be local approximants, which may be taken self-adjoint, with . Let ; then are effects with by the functional calculus, so . Since the square root is monotone and vanishes at , [43 Thm. X.1.1] (the proof there is dimension-independent, using only operator monotonicity of the square root), which is ; as , is almost local. Similarly, and , giving the second claim.
Corollary IV.7. *Suppose , let be effects with , and put and . Then the nonselective Lüders channel of the relativised observable ,
is an almost local operation, and its translate is the Lüders channel of the translated observable . Consequently, by Proposition IV.4(ii), a nonselective Lüders measurement of alters the expectation of any relativised local observable at causal distance in the sense of (III.15) by at most .
Proof. Since is completely positive and subunital and is unital and completely positive, each () is an effect, almost local by Proposition III.4, and hence so is , a sum of approximants for the (taken with a common and ) serving as approximants for . By Lemma IV.6, the Kraus operators () and are almost local. Hence is almost local, and since commutes with the functional calculus, is as claimed. Proposition IV.4(ii) applies.
Proposition IV.4 and Corollary IV.7 together provide what is left of of microcausality in the relativised theory: no operation with a finite almost local Kraus decomposition, including the Lüders measurements of every finite-outcome relativised local observable, can affect the expectation of a distant relativised observable by more than . We stress that (IV.4) should not be read as asserting the possibility of superluminal signalling. By Proposition II.1 and Lemma III.1, relativised observables admit no sharp localisation in bounded regions, so that the spacetime labels appearing in any relative no-signalling statement are themselves defined only to the accuracy set by the frame state.
Thus far we have considered how the local and causal properties of an observable fare under relativisation. We now turn to the question of the physical requirements of the frame for the relative and absolute descriptions to agree closely. Relative to a frame state , the actually realisable counterpart of a system effect is the restricted relative observable , and the quality of approximation is given by . In the quantum mechanical setting, bounds on were obtained in [44], where good approximation was shown to require a broad distribution of the symmetry generator in the frame state. For translations in Minkowski space the relevant property of the frame is its ability to resolve a displacement , quantified by the overlap of the frame state with its translate: close to 1 means that the frame cannot resolve . The following proposition shows that if is sensitive to the displacement , then accurate recovery of forces to be small. The bound also involves the unsharpness of , which vanishes exactly when is a projection.
Proposition IV.8. Let be a system effect, suppose the action is implemented by a strongly continuous unitary representation of , and put
For , let . If , then
Proof. Set and . Since and is completely positive and subunital, is an effect. The map is unital and completely positive, so the estimate given in [44 eq. (9)] gives
Moreover is invariant, so , while
The Cauchy–Schwarz inequality for completely positive maps [45 Lem. 1] gives
Dividing by and using gives (IV.10), up to the sign convention in the implementation of translations. □
Remark IV.9. If varies appreciably under a translation , then accurate recovery of by relativisation followed by restriction requires to be small and hence the frame state to have sufficiently small . In other words, and bearing in mind that , the frame must itself resolve the displacement . Since is related to the QRF energy-momentum distribution, resolving shorter displacement scales requires broader energy-momentum support.
To quantify this last point slightly more, we argue as follows. Recalling that , we note that
holds for any , where , which follows from the rather crude bound that for and otherwise. Supposing that and rearranging,
If is chosen small but nonzero, and , then . Thus, a significant proportion of the energy-momentum distribution of lies in the set . If, for example, is a timelike vector, then all momenta with have energy greater than in the rest-frame of . Alternatively, if is spacelike and then has energy greater than in any frame for which is purely spatial, i.e., orthogonal to the observer’s 4-velocity. To see this, let be the 4-velocity, then implies . In either case, resolving short spacetime displacements requires high energies in an associated frame of reference.
Concluding Remarks
The spectrum condition obstructs the existence of ideal reference frames for spacetime translations in QFT: translation-covariant POVMs are never projection-valued, and they fail the norm-1 property in many cases. Relative to such a frame, local observables become almost local, exact microcausality is replaced by rapid commutator decay controlled by the localisation density of the frame state. From an operational standpoint, in which spacetime structure is abstracted from relations among observables, the results above indicate that the causal structure relative to any realistic frame is intrinsically smeared (Remark III.7), and the sharp light cone appears as an idealised limit of frame localisation - a limit which, by Corollary II.2, no physical frame satisfies. We leave the systematic development of this perspective, and the associated question of the emergence of classical causal structure, to future work.
Several further questions remain open. The frames with unbounded , the only remaining candidates for the norm-1 property (Remark II.10), need a physical characterisation. Relativised two-point functions of Hadamard states may provide a natural class of examples. Whether the operational compatibility preserved under relativisation excludes obstructions to measurements of the kind identified by Sorkin [30] is open, and connects with the measurement framework of [28]. The relation between causal and realisable operations is delicate already in the absolute setting [37] and we leave the relational setting for the future. Finally, we note that the necessary unsharpness of time observables may play a role in the ubiquity of the phenomenon of type change [46, 47, 48, 49].
Note added. During the completion of this work, we became aware of the work of Agullo and Mele [50], who investigate QRFs also in a QFT setting with Gaussian-modulated POVMs as frame observables, which do not satisfy the norm-1/localisability property. We look forward to comparing our findings to theirs, particularly in connection with Corollaries II.8 and II.9. We also became aware of Freidel and Kirklin [51], in which the quantum reference frame for reparametrisations of a gravitational null ray is necessarily non-ideal; positivity of the null energy excludes a sharp frame observable. Though the context is different, this gives further indication that unsharp QRFs are likely to be typical in relativistic quantum settings quite generally.
Acknowledgments
We thank Daan Janssen, Kasia Reizner, Jan Glowacki and Samuel Fedida for useful discussions. We also thank Laurent Freidel for pointing out [51]. The work of CJF is supported in part by EPSRC Grant EP/Y000099/1 to the University of York. For the purpose of open access, the authors have applied a creative commons attribution (CC BY) licence to any author accepted manuscript version arising.
References
- [1]S. D. Bartlett, T. Rudolph, and R. W. Spekkens, Reference frames, superselection rules, and quantum information, Reviews of Modern Physics 79, 555 (2007).DOI
- [2]L. Loveridge, Quantum measurements in the presence of symmetry, Ph.D. thesis, University of York (2012).
- [3]L. Loveridge, T. Miyadera, and P. Busch, Symmetry, reference frames, and relational quantities in quantum mechanics, Foundations of Physics 48, 135 (2018).
- [4]A.-C. de la Hamette, T. D. Galley, P. A. Höhn, L. Loveridge, and M. P. Müller, Perspective-neutral approach to quantum frame covariance for general symmetry groups (2021), arXiv:2110.13824 [quant-ph].arxiv.org/abs/2110.13824
- [5]J. Glowacki, L. Loveridge, and J. Waldron, Quantum reference frames on finite homogeneous spaces, International Journal of Theoretical Physics 63, 137 (2024).
- [6]T. Carette, J. Glowacki, and L. Loveridge, Operational quantum reference frame transformations, Quantum 9, 1680 (2025).DOI
- [7]F. Giacomini, E. Castro-Ruiz, and Č. Brukner, Quantum mechanics and the covariance of physical laws in quantum reference frames, Nature Communications 10, 494 (2019).
- [8]A. Vanrietvelde, P. A. Höhn, F. Giacomini, and E. Castro-Ruiz, A change of perspective: switching quantum reference frames via a perspective-neutral framework, Quantum 4, 225 (2020).DOI
- [9]A.-C. de la Hamette and T. D. Galley, Quantum reference frames for general symmetry groups, Quantum 4, 367 (2020).
- [10]E. Castro-Ruiz and O. Oreshkov, Relative subsystems and quantum reference frame transformations, Communications Physics 8, 187 (2025).
- [11]L. Loveridge and J. Waldron, W*-algebraic quantum reference frames, In preparation (2026).
- [12]J. Glowacki and Y. Kuramochi, W*-algebraic integration theory, arXiv:2606.27366 (2026).
- [13]M. Toller, Localization of events in space-time, Physical Review A 59, 960 (1999).DOI
- [14]S. Mazzuccchi, On the observables describing a quantum reference frame, Journal of Mathematical Physics 42, 2477 (2001).DOI
- [15]R. Haag, Local Quantum Physics: Fields, Particles, Algebras, 2nd ed., Theoretical and Mathematical Physics (Springer, Berlin, Heidelberg, 1996).
- [16]W. Dybalski and C. Gérard, A criterion for asymptotic completeness in local relativistic QFT, Communications in Mathematical Physics 332, 1167 (2014).
- [17]S. Fedida and J. Glowacki, Foundations of relational quantum field theory I: Scalars (2025), arXiv:2507.21601 [quant-ph].arxiv.org/abs/2507.21601
- [18]R. F. Streater and A. S. Wightman, PCT, Spin and Statistics, and All That (Princeton University Press, 2000) originally published 1964 by W. A. Benjamin.DOI
- [19]R. Gianitrapani, Quantum coordinates of an event in local quantum physics, Journal of Mathematical Physics 39, 5180 (1998).
- [20]R. Haag and D. Kastler, An algebraic approach to quantum field theory, Journal of Mathematical Physics 5, 848 (1964).DOI
- [21]G. Lüders, Über die Zustandsänderung durch den Meßprozeß, Annalen der Physik 8, 322 (1951).DOI
- [22]P. Busch and P. Lahti, Lüders rule, in Compendium of Quantum Physics (Springer, 2009) pp. 356–358.
- [23]P. Busch, Unsharp localization and causality in relativistic quantum theory, Journal of Physics A: Mathematical and General 32, 6535 (1999).DOI
- [24]P. Busch, Unsharp reality and joint measurements for spin observables, Physical Review D 33, 2253 (1986).pubmed.ncbi.nlm.nih.gov/9956897
- [25]P. Busch, P. Lahti, J.-P. Pellonpää, and K. Ylinen, Quantum Measurement (Springer, Cham, 2016).DOI
- [26]T. Heinosaari and T. Miyadera, Qualitative noise-disturbance relation for quantum measurements, Phys. Rev. A 88, 042117 (2013).DOI
- [27]L. J. Landau, A note on extended locality, Comm. Math. Phys. 13, 246 (1969).DOI
- [28]C. J. Fewster and R. Verch, Quantum fields and local measurements, Communications in Mathematical Physics 378, 851 (2020).arxiv.org/abs/1810.06512
- [29]T. Heinosaari, J. Kiukas, D. Reitzner, and J. Schultz, Incompatibility breaking quantum channels, Journal of Physics A: Mathematical and Theoretical 48, 435301 (2015).DOI
- [30]R. D. Sorkin, Impossible measurements on quantum fields, in Directions in General Relativity: Proceedings of the 1993 International Symposium, Maryland, Vol. 2 (Cambridge University Press, 1993) pp. 293–305.arxiv.org/abs/gr-qc/9302018
- [31]L. Borsten, I. Jubb, and G. Kells, Impossible measurements revisited, Phys. Rev. D 104, 025012 (2021).DOI
- [32]H. Bostelmann, C. J. Fewster, and M. H. Ruep, Impossible measurements require impossible apparatus, Phys. Rev. D 103, 025017 (2021).DOI
- [33]I. Jubb, Causal state updates in real scalar quantum field theory, Physical Review D 105, 025003 (2022).DOI
- [34]R. Oeckl, Spectral decomposition of field operators and causal measurement in quantum field theory, J. Math. Phys. 66, 042302 (2025).DOI
- [35]R. Oeckl, Causal measurement in quantum field theory: spacetime, Phys. Rev. D 113, 085002 (2026).DOI
- [36]R. Oeckl and A. Zampeli, Towards local and compositional measurements in quantum field theory, Quantum 10, 2232 (2026), arXiv:2505.10968.DOI
- [37]J. Mandrysch, R. Simmons, and M. Navascués, Causality and realizability of local operations in quantum field theory (2026), arXiv:2607.12976.arxiv.org/abs/2607.12976
- [38]K. Kraus, States, Effects, and Operations: Fundamental Notions of Quantum Theory, Lecture Notes in Physics, Vol. 190 (Springer, Berlin, 1983).
- [39]M.-D. Choi, A Schwarz inequality for positive linear maps on C*-algebras, Illinois Journal of Mathematics 18, 565 (1974).DOI
- [40]A. Arias, A. Gheondea, and S. Gudder, Fixed points of quantum operations, Journal of Mathematical Physics 43, 5872 (2002).DOI
- [41]G. Lindblad, A general no-cloning theorem, Letters in Mathematical Physics 47, 189 (1999).
- [42]A. Much and R. Verch, Superluminal local operations in quantum field theory: A ping-pong ball test, Universe 9, 447 (2023).DOI
- [43]R. Bhatia, Matrix Analysis, 1st ed. (Springer New York, NY, 1997).
- [44]T. Miyadera, L. Loveridge, and P. Busch, Approximating relational observables by absolute quantities: a quantum accuracy–size trade-off, J. Phys. A: Math. Theor. 49, 185301 (2016).
- [45]B. Janssens, Unifying decoherence and the Heisenberg principle, Letters in Mathematical Physics 107, 1557 (2017).
- [46]V. Chandrasekaran, R. Longo, G. Penington, and E. Witten, An algebra of observables for de Sitter space, Journal of High Energy Physics 2023, 82 (2023).
- [47]C. J. Fewster, D. W. Janssen, L. D. Loveridge, K. Rejzner, and J. Waldron, Quantum reference frames, measurement schemes and the type of local algebras in quantum field theory, Communications in Mathematical Physics 406, 19 (2025).
- [48]J. De Vuyst, S. Eccles, P. A. Höhn, and J. Kirklin, Gravitational entropy is observer-dependent, Journal of High Energy Physics 2025, 146 (2025).DOI
- [49]J. De Vuyst, S. Eccles, P. A. Höhn, and J. Kirklin, Crossed products and quantum reference frames: on the observer-dependence of gravitational entropy, Journal of High Energy Physics 2025, 063 (2025).DOI
- [50]I. Agullo and F. M. Mele, A proposal for quantum reference frames in quantum field theory and their experimental exploration, In preparation (2026).arxiv.org/abs/2611.21740
- [51]L. Freidel and J. Kirklin, Gravitational null rays: Covariant quantization and the dressing time, J. High Energy. Phys 2026, 130 (2026).DOI