arXiv:1912.00033v3 [quant-ph] 13 Apr 2021 ∗ † I.Cvrattm bevbe 7 time Covariant III. V eainlqatmdnmc nDrcand Dirac in dynamics quantum Relational IV. [email protected];sae rtauthorshi first shared [email protected]; authorship. first shared [email protected]; I hs pc tutr n eainlDirac relational and structure space Phase II. .Itouto 2 Introduction I. eue uniain11 quantization reduced observables .Dnmc :Rltoa ia bevbe 12 observables Dirac Relational I: Dynamics A. .Cascltm bevbe 7 observables time Classical A. .Casclrltoa yais4 dynamics relational Classical A. .Rdcdpaesaeqatzto 14 quantization space phase Reduced B. .Qatmtm bevbe 8 observables time Quantum B. .Dcmoiino h hs pc noa into space phase the of Decomposition B. .Casclpaesaerdcin14 reduction space phase Classical 1. .Dsrt pcrmcok 10 9 quantum non-degenerate of Examples clocks 3. spectrum Discrete 2. clocks spectrum Continuous 1. lc n ytmo neet6 interest of system and clock 2 4 knw nttt fSineadTcnlg rdaeUnive Graduate Technology and Science of Institute Okinawa lcs11 clocks eateto hsc n srnm,DrmuhClee Ha College, Dartmouth Astronomy, and Physics of Department xed oa rirr ubro conditionings. of number systems arbitrary ancilla an or to clocks extends Un ideal con approximations, probabilities. how invoke transition showing not correct by the propagators give wrong permit observables yields finally formalism trinity The PW the formalism. normal PW reported the in previously gauge-invarentanglement a a resolves obse furthermore of relational trinity description of t evaluation gauge-fixed violate gauge-invariant a formalism manifestly of PW the analog sho of quantum We probabilities effect. conditional nonlocality the temporal relat dependent clock transforming a frames, strate reference temporal observables changing relational exten these to q gauge-invariantly of The of properties Wald. algebraic analog lish and quantum Unruh the by as and reported procedure clocks clocks quantum non-ideal realistic encompass quantization of which a develop POVMs we covariant process, using the In trinity. the lence bandvasmer euto.Cntttn he faces three Constituting an reduction. symmetry formalism, clock-n via Schr¨odinger picture obtained the (PW) be in Wootters’ equivalence observables and unknown relational Page 1) previously a dynamics: quantum establish we maps, tion 1 h rbe ftm nqatmgaiyclsfrarelational a for calls gravity quantum in time of problem The eateto hsc n srnm,Uiest olg Lo College University Astronomy, and Physics of Department 3 eateto hsc,SitAsl olg,Mnhse,N Manchester, College, Anselm Saint Physics, of Department hlp .H¨ohn, A. Philipp CONTENTS 5 nttt o unu pisadQatmIfrain(IQOQ Information Quantum and Optics Quantum for Institute h rnt fRltoa unu Dynamics Quantum Relational of Trinity The utinAaeyo cecs -00Ven,Austria Vienna, A-1090 Sciences, of Academy Austrian ,2, 1, ∗ lxne .H Smith, H. R. Alexander Dtd pi 4 2021) 14, April (Dated: p. 4 I.Cagn eprlrfrnefae 25 frames reference temporal Changing VII. I ietnln h aeWotr omls 22 formalism Page-Wootters the Disentangling VI. eetn eet‘lc-eta’approach ‘clock-neutral’ recent a extend We . .Tetiiyo eainlqatmdnmc 16 dynamics quantum relational of trinity The V. vbe ntepyia ne rdc.The product. inner physical the in rvables iepeiu rpsl,orrslto does resolution our proposals, previous like ,4, 3, zto miut n lrfisterl of role the clarifies and ambiguity ization smnfsl ag-nain,adeasily and gauge-invariant, manifestly is , oa bevbe n tts n demon- and states, and observables ional rcdr o eainlDrcobservables Dirac relational for procedure .Saetasomtos25 transformations State A. .Ritrrtn h rnt 22 trinity the Reinterpreting A. Page-Wootters The II: Dynamics A. ftesm yais ecl hsequiva- this call we dynamics, same the of .Euvlneo yaisI n I 22 III and II Dynamics of Equivalence C. .Smlfigcmuaos25 Simplifying C. .Osral rnfrain 27 transformations B. .Casclaao ftetiilzto 24 trivialization the of analog Classical B. .Dnmc I:Rltoa esnegpicture Heisenberg Relational III: Dynamics B. urlpcueo ia uniain 2) quantization, Dirac of picture eutral ecntan,i norc.Te r a are They incorrect. is constraint, he htKcarsciiim leigthat alleging Kuchaˇr’s criticism, that w igguefie uniis eestab- We quantities. gauge-fixed ding † )terltoa esnegpicture Heisenberg relational the 3) d st,On,Oiaa9409,Japan 0495, 904 Okinawa Onna, rsity, st eov uhˇ’ rtcs that Kuchaˇr’s criticism resolve to us s .Rdcdqatzto 15 quantization Reduced 2. .Osral rnfrain nthe in transformations Observable 1. omls 16 space phase reduced with 19 Relation 2. 17 and reduction symmetry Quantum 1. deparametrization quantum through II and I Dynamics of Equivalence 2. Page-Wootters the Introducing 1. formalism we he prahst relational to approaches three tween n aiiinP .Lock E. P. Maximilian and atqatt n qiaett the to equivalent and quantity iant oe,NwHmsie075 USA 03755, Hampshire New nover, atmrdcinmp eelthis reveal maps reduction uantum eov h o-oooiiyissue non-monotonicity the resolve eainlSh¨dne itr 27 Schr¨odinger picture relational uniain21 17 19 quantization I Dynamics with equivalence formalism dn odn ntdKingdom United London, ndon, iinlpoaiiiso relational of probabilities ditional ouin sn unu reduc- quantum Using solution. wHmsie012 USA 03102, Hampshire ew I), 5 2

2. Observable transformations in the is described. Recognizing that any employed reference relational 29 frame is itself a physical system, it too must be subject C. Temporal localization is frame dependent 29 to dynamics and interact with the degrees of freedom it D. Connection with past work on quantum wishes to describe. In particular, the famous ‘rods and temporalframechanges 31 clocks’ that formed Einstein’s conception of a reference frame must be quantized. This insight has long been VIII. Implications of the trinity 32 recognized in the community [1–34], by A. Quantum analog of gauge-invariant those interested in foundational issues aimed at removing extensionofgauge-fixedquantities 32 the background structure inherent in standard quantum B. Conditional inner product as quantum theory [35–55], and more recently applied in the context gauge-fixedphysicalinnerproduct 33 of science [37, 56–61]. C. Resolving Kuchaˇr’s three criticisms 33 Background independence leads to a dynamical conun- D. There is no normalization ambiguity in the drum in the context of canonical quantum gravity: the Page-Woottersformalism 36 Hamiltonian of a generally covariant theory, such as gen- eral relativity, is constrained to vanish in the absence of IX. Conclusion 36 boundaries [1, 3, 62]. As a consequence, in the quantum theory it appears as if one obtains a ‘frozen formalism’ Acknowledgments 38 and physical states (of the spatial geometry and matter) do not evolve in time. This is known as the problem of References 38 time in quantum gravity [10, 11, 63]. However, upon closer inspection, it is clear that the quantum theory is A. Comment on the validity of the absence of not ‘timeless’ as often stated. The problem of time is interactions 40 rather a manifestation of background independence and B. Freedom of choice in classical and quantum means that physical states do not evolve relative to an ex- time observables 41 ternal background time. Instead, one must extract a time evolution in a relational manner, i.e. pick some quantized C. Proofs of lemmas and theorems of Secs. IV degrees of freedom to serve as an internal time — a tem- and V 41 poral quantum reference frame — relative to which the remaining quantum degrees of freedom evolve [1–34, 44– D. Derivation referenced in Sec. VIIC 52 47, 49–52]. In this regard, given the a priori many pos- sible choices of internal time, we shall extend arguments E. Mathematical details 52 that it is more appropriate to consider the ensuing quan- 1. Canonical transformation separating gauge tum theory as being ‘clock-neutral’ [25, 26] rather than and gauge-invariant degrees of freedom 52 ‘timeless’; it is a description of physics prior to having 2. Correct propagator from gauge-invariant chosen a temporal reference frame relative to which the conditional probability 53 other degrees of freedom evolve. We will refer to such temporal reference frames loosely as ‘clocks’. We emphasize that, depending on the con- I. INTRODUCTION crete model at hand, they may represent clocks in an op- erational laboratory situation or describe global degrees Background independence is the lesson of general rel- of freedom, such as the dynamical ‘size’ of the Universe ativity: a physical theory should not depend on external in a cosmological setting, which can serve as a cosmic structures. In pre-relativistic physics, space and time time standard. are external entities with respect to which the dynamics We focus on three of the main approaches to solv- of matter unfolds. In contrast, general relativity unites ing the problem of time through a relational notion of space and time into a single object, spacetime, which . The first approach (Dynamics I), is is dynamical and interacts with matter as described by formulated in terms of gauge invariant relational Dirac Einstein’s field equations. observables that correspond to the simultaneous reading However, standard quantization techniques often rely of a clock and observable of interest [1–30] within the a on background structures, such as imposing the canon- priori ‘clock-neutral’ picture of Dirac quantization (‘first ical commutation relations on constant-time hypersur- quantize, then constrain’). A second approach (Dynam- faces. These techniques cannot be applied unaltered in a ics II), put forward by Page and Wootters [44, 45] and quantum theory of gravity where the aim is to quantize further developed in [46, 47, 49–51, 64–71], describes re- spacetime itself, rather than to quantize matter in space- lational quantum dynamics in terms of quantum correla- time. New tools that allow for a background independent tions between a clock and system and yields a relational quantization scheme are thus required [1–3]. Schr¨odinger picture. Finally, a third approach known as Often the external structures in a theory appear as quantum symmetry reduction (Dynamics III), draws its reference frames with respect to which matter and motion inspiration from, and in some cases is equivalent to, re- 3

Clock-neutral sidesteps pathologies of some classical time func- picture tions. Furthermore, by appealing to the more gen- in which dynamics is eral notion of an observable as characterized by a defined by relational Dirac observables POVM, this allows for a resolution of the appar- ent non-monotonicity of realistic quantum clocks, as used by Unruh and Wald [77] to argue against the viability of a relational approach to the problem of time. Indeed, our POVM-based time observable Relational will be monotonic for bounded Hamiltonians and Relational Schrodinger¨ picture Heisenberg picture admits a consistent probability interpretation. Page-Wootters resulting from quantum formalism, time from • symmetry reduction Employing such clock POVMs, we construct a sys- quantum correlations tematic quantization procedure for relational Dirac observables. This amounts to a G-twirl, i.e. an av- FIG. 1. The trinity of relational quantum dynamics posits eraging over the group generated by the constraint, that the dynamics described by relational Dirac observables of the (kinematical) observable of interest and a in the clock-neutral picture of Dirac quantization, the rela- projection onto a chosen reading of the quantum tional Schr¨odinger picture of the Page-Wootters formalism, clock. This extends the use of G-twirling tech- and the relational Heisenberg picture obtained upon a quan- niques, often used in the literature on spatial quan- tum symmetry reduction of the clock-neutral theory are three tum reference frames without constraints, e.g. see manifestations of the same relational quantum theory. [37, 61, 78], to the context of Hamiltonian con- straints and temporal quantum reference frames. We prove various algebraic properties of the thus duced phase space quantization, which singles out a time constructed relational quantum observables. observable at the classical level that is then used to con- struct a quantum theory [2, 10, 11, 25, 26, 72] (see [31– • The quantum reductions which map the clock- 34] for a discussion in loop quantum gravity/cosmology). neutral Dirac quantized theory into either the re- This yields a relational Heisenberg picture. Each of these lational Schr¨odinger picture of the Page-Wootters approaches is a manifestation of the relational paradigm formalism or the relational Heisenberg picture of in physics: localization in both space and time is only the symmetry reduced theory, reveal our procedure meaningful in relation to other physical systems, and not of quantizing relational observables as the quan- relative to absolute or external structures. tum analog of so-called gauge-invariant extensions Due to their different motivations and dissimilar ways of gauge-fixed quantities [14–17, 79]. in which dynamics arises in each, these three proposals • for a relational quantum dynamics were long thought to We place the Page-Wootters formalism on a more be distinct [10, 11]. A main contribution of this article rigorous foundation and bring it into conversation is to show that under certain conditions, amounting to with the modern techniques of quantum gravity. the requirement that a physical clock be “well-behaved” The trinity implies that the dynamics arising in the and does not couple to the evolving degrees of freedom, Page-Wootters formalism should be regarded as the these three proposals are actually a manifestation of the quantum analog of the dynamics defined on a clas- same relational quantum theory, as summarized in Fig. 1. sical reduced phase space resulting from choosing a We thus refer to these relational quantum theories as specific gauge related to the choice of clock. the trinity of relational quantum dynamics, or simply the • We fully resolve Kuchaˇr’s criticism that the con- trinity. This equivalence enables us to prove a number of ditional probabilities of the Page-Wootters formal- further results in the context of relational dynamics and ism violate the constraints [10]. We show that they resolve past issues reported in the literature. coincide with expectation values of relational ob- We list here the further contributions of this article: servables in the clock-neutral picture and thus can be viewed as quantum analogs of gauge-fixed ex- • Using the most general notion of a quantum observ- pressions of gauge-invariant quantities. This also able defined as a positive -valued measure clarifies that the alleged normalization ambiguity (POVM) [73–75], we construct a novel quantiza- reported in [49] does not arise. tion of classical (kinematical) time observables as- sociated with a clock using only the quantization of • We generalize the clock-neutral approach to chang- the clock Hamiltonian HˆC . This extends the dis- ing temporal quantum reference frames developed cussion of clock POVMs in the context of relational in [25, 26] to the case where clocks are described us- dynamics in [50, 51, 76]. This procedure does not ing POVMs. This extends the perspective-neutral rely on a self-adjoint time operator that is canoni- approach to quantum reference frames [25, 26, 39, cally conjugate to HˆC, which in general situations 40, 80], which identifies the gauge-invariant quan- of physical interest does not exist. This elegantly tum theory obtained through Dirac quantization as 4

a description of the physics prior to having chosen plays in the Page-Wootters formalism in Sec. VI. Next, a quantum frame from which to describe the re- we construct temporal frame change maps between clock maining degrees of freedom. Here, the quantum perspectives and illustrate a novel time nonlocality effect reduction maps of the trinity are associated with a in Sec. VII. In Sec. VIII we discuss the quantum analog choice of clock, and assume the role of ‘quantum co- of the gauge-invariant extension of gauge fixed quanti- ordinate’ maps to the ‘perspective’ of that clock. In ties, resolve Kuchaˇr’s criticisms (pointing out differences analogy to coordinate changes on a manifold, one with past attempts at resolutions), and explain why there can then change clock perspective by concatenating is no normalization ambiguity in the Page-Wootters for- reduction maps associated to different clocks. This malism. We conclude in Sec. IX. procedure always passes through the clock-neutral Classical phase space functions and their quantum op- picture, and transforms both states and observables erator equivalent will be distinguished with hats, and between different clock perspectives. throughout we work in units such that ~ = 1. • Using this temporal frame change method, we demonstrate a clock dependent temporal nonlocal- ity effect. When a clock is in a superposition read- ing different times, the dynamics of a system of II. PHASE SPACE STRUCTURE AND interest with respect to that clock will be in a su- RELATIONAL DIRAC OBSERVABLES perposition of time evolutions. This complements a similar effect reported in [65], and is the tempo- A. Classical relational dynamics ral analog of the quantum frame dependent spatial correlations observed in [38, 39]. Using this clock The diffeomorphism-invariance of general relativity change method, we also find a new ‘self-reference’ leads to a so-called Hamiltonian constraint, i.e. a Hamil- phenomenon of quantum clocks. tonian that is constrained to vanish (in the absence of • The trinity allows us to completely resolve Kuchaˇr’s boundaries) [1, 3, 62]. The Hamiltonian of general rela- criticism that the Page-Wootters formalism yields tivity thereby not only generates the dynamics, but also the wrong propagators [10]. We introduce a new temporal diffeomorphisms, which are gauge transforma- two-time conditional probability using relational tions. However, a gauge-invariant form of dynamics can observables at the level of the a priori clock-neutral be encoded in so-called relational observables [1, 6–9, 14– picture. Upon quantum reduction, this always 18]. We review here the concept of relational observables yields the correct transition probabilities in the re- for finite-dimensional models subject to a Hamiltonian lational Schr¨odinger picture of the Page-Wotters constraint. formalism as expected from standard quantum me- Consider a system on an N-dimensional configuration a a chanics. In contrast to previous proposals [21, 49], space described by the action S = =R ds L(q , q˙ ), our resolution does not rely on approximations, whereq ˙a denotes differentiation withM respect to s and ideal clocks or auxiliary ancilla systems, and au- a =0, 1,...,N. Suppose the action isR reparametrization- tomatically extends to an arbitrary number of con- invariant (i.e. invariant under one-dimensional diffeo- ditionings. morphisms), meaning the Lagrangian transforms as a a a a a • scalar density L(q , q˙ ) L(q , dq /ds˜) ds/ds˜ under a We clarify the role entanglement plays in giving rise reparametrization s 7→s˜(s). It follows that the Leg- to relational dynamics in the Page-Wootters for- endre transformation7→ will then produce a Hamiltonian malism by emphasizing that this entanglement is H = N(s) CH , where N(s) is an arbitrary (lapse) func- kinematical and demonstrating that the same dy- tion and CH is a so-called Hamiltonian constraint namics can arise in the absence of this kinematical entanglement. N a a a We begin in Sec. II by reviewing the classical theory of CH = pa q˙ L(q , q˙ ) 0, − ≈ a=1 Hamiltonian constrained systems and relational observ- X ables, and subsequently specializing to a direct sum of a phase spaces describing a clock and a system whose which has to vanish due to the reparametrization invari- dynamics the clock will track. In Sec. III the quantiza- ance of L(qa, q˙a). This condition defines a (2N 1)- − tion of kinematical time observables as so-called covariant dimensional submanifold kin, referred to as the con- POVMs is described. In Sec. IV A, we introduce Dynam- straint hypersurface, in theC⊂P 2N-dimensional kinematical ics I defined in terms of quantum relational Dirac observ- phase space kin, which is parametrized by the canoni- P a a a ables and also discuss reduced phase space quantization, cal coordinates q and pa satisfying q ,pb = δb . The which, while not comprising an element of the trinity, image of the Legendre transformation{ is thus} a lower- will be of conceptual importance. In Sec. V the equiv- dimensional subset of kin. In this context, denotes alence of the relational dynamics comprising the trin- a weak equality meaningP that the equality only≈ holds on ity is established. We then clarify the role entanglement [79, 81]. Such a setting is schematically depicted in C 5

Fig. 2.1 Setting henceforth N(s) = 1, the Hamiltonian H co- P incides with the constraint function CH and generates kin dynamical equations on the kinematical phase space df := f, C , ds { H } where f : R is an arbitrary phase space function. Pkin → This defines a dynamical flow on the phase space kin, s P αCH : R kin, with flow parameter s that transforms any function→Pf as

∞ sn f αs f := f, C , (1) 7→ CH · n! { H }n n=0 X Gauge orbits where f, CH n+1 := f, CH n, CH is the iterated { } {{ } } 2 FIG. 2. Depicted is the unconstrained phase space P (rect- Poisson bracket with the convention f, CH 0 := f. The kin dynamical orbits, corresponding to solutions{ } to the equa- angular prism), the constraint surface C (green surface), gauge tions of motion, must lie on the constraint surface . Be- orbits/dynamical trajectories in C generated by CH (black C curves on C), the gauge-fixing surface T = τ (red plane), and ing the only constraint, C is first class and its action red H the reduced phase space P (thick black line, see Sec. IV B). on corresponds to (active) temporal diffeomorphisms S The relational Dirac observable Ff,T (τ) is a gauge-invariant on theC manifold = R underlying the action S, which M function on C corresponding to the question “what is the value are equivalent to reparametrizations (passive diffeomor- of the function f when the clock T reads τ?” Hence, it cor- phisms) s s˜(s). Since the action S is invariant under responds to the value of the function f on the intersection of reparametrizations,7→ the evolution with respect to the flow the gauge-fixing surface T = τ with C. Letting the parameter parameter s is not physical; it is a gauge transformation τ run unfolds the relational dynamics and thus corresponds on . This mimics the situation in general relativity. In- to ‘scanning’ C with the family of gauge-fixing surfaces T = τ. deed,C general relativistic cosmological models satisfy all the structure introduced here [82]. Physical observables are represented by functions F instead of the original non-dynamical parameter s. Such on the constraint surface that are invariant under the a choice of T is therefore often called an internal time C flow generated by the constraint CH and known as Dirac or clock function in the gravity literature. This suggests s observables. This requirement amounts to the condition that we construct Ff,T (τ) by solving αCH T = τ for s, using the expansion in Eq. (1) and denoting· the solution F, CH 0. (2) as s (τ), and then evaluating the flow of f at s = s (τ), { }≈ T T which yields Using so-called relational Dirac observables (aka evolv- ing constants of motion) [1, 3, 6–9, 14–18], it is possible s Ff,T (τ) := αCH f to establish a gauge-invariant dynamics. Relational Dirac · s=sT (τ) observables encode how one observable evolves relative to n ∞ (τ T ) CH another along the flow generated by CH . That is, they − f, . (3) ≈ n! T, C are Dirac observables Ff,T (τ) (in this case also known n=0  { H }n as complete observables) corresponding to the value a X phase space function f (a partial observable) takes on The expansion in the second equality was first derived when the phase space function T (another partial observ-C (as a special case of a general framework) in [14–17]. As able) takes the value τ. Hence, the partial observable T shown in these works, it is a simple exercise to demon- assumes the role of a dynamical reference degree of free- strate that the functions Ff,T (τ) satisfy Eq. (2) and are dom, which we can choose to parametrize the flow αCH thus Dirac observables. Notice that Ff,T (τ) is only de- fined where T, C = 0, i.e. where T defines a good { H } 6 parametrization of the flow αCH . For later purposes, we note that this construction of 1 The familiar Hamiltonian mechanics of a system without con- Ff,T (τ) constitutes a so-called gauge-invariant extension straints can be recovered from the special case CH = p0 + of a gauge-fixed quantity [14–17, 79]. Since CH generates HS(qi,pi), where HS (qi,pi) is the Hamiltonian for a system de- scribed by the coordinates qi,pi with i = 1,...,N − 1 [79]. not only the dynamics, but also gauge transformations, 2 More precisely, this is a pull-back. Let x denote a point in every dynamical trajectory in is also a gauge orbit. In n s s ∞ sn d f(x) C P . Then α · f(x) := f(α (x)) = n = kin CH CH n=0 n! ds any region of where T, CH = 0, T defines a good ∞ sn P clock and theC gauge-fixing{ condition} 6 T = τ singles out n=0 n! {f, CH }n(x). For notational simplicity, we henceforth dropP reference to the points x ∈ Pkin, which are specified by the a point on each gauge orbit in this region (which later a N coordinates (q ,pa)a=1. will be all of ). F (τ) is a gauge-invariant quantity C f,T 6 defined in this entire region of and it encodes a gauge- parametrize its orbits.3 The clock function T will be used C fixed quantity, namely the value of f at the point on the as one coordinate on C. gauge orbit fixed by the condition that T = τ. This Based on this partition,P we shall henceforth further re- construction is schematically represented in Fig. 2. strict to classical theories described by an autonomous (i.e. independent of flow parameter s) Hamiltonian con- Notice that τ is now an evolution parameter and so straint of the form Ff,T (τ) in Eq. (3) really is a one-parameter family of Dirac observables. Letting τ run over its set of permis- CH = HC + HS 0, (4) ≈ sible values then describes the relational evolution of f where H is a function on , which we refer to as the relative to the clock T . We stress that this construction C PC holds for an arbitrary phase space function f. clock Hamiltonian, and HS is a function on S, which we refer to as the system Hamiltonian. That is,P we assume While relational Dirac observables can in principle be that the clock and system do not interact. quantized once their classical form is known, the quan- This is an assumption usually made in the literature tum analog of this systematic construction procedure, to on the Page-Wootters formalism [44, 45], which is why gauge-invariantly extend gauge-fixed quantities, has thus we shall likewise adopt this assumption in order to prove far not been established in the literature. The reason equivalence with other approaches (see [50] in which this is that Dirac quantization immediately yields a gauge- assumption is relaxed in the context of the Page-Wootters invariant (cf. Sec. IV A), so that a gauge- formalism). We emphasize that Eq. (4) is, of course, an fixing as above is not feasible in the quantum theory and idealization. If the constraint modelled a laboratory sit- one has to proceed differently. One of our results below uation, one might interpret this as a reasonable situation and in [83, 84] is to develop precisely the quantum analog in which the clock and system are so far apart that their of the gauge-invariant extension of gauge-fixed quantities interaction may be neglected. However, in general rela- procedure for a class of models. tivity, Eq. (4) is a strong restriction. Being a field theory, finite dimensional general relativistic systems correspond to models with symmetry, such as homogenous cosmolog- ical models or certain black hole spacetimes. In this case, the phase space variables correspond to global and there- fore not localized degrees of freedom, such as the scale factor or certain anisotropy parameters. In this case, one cannot conceive of an absence of interactions between ‘clock’ and ‘system’ as corresponding to them being far B. Decomposition of the phase space into a clock and system of interest removed from one another. In fact, generic general rel- ativistic systems do not satisfy the idealization Eq. (4) [3, 10–12, 19, 29, 63]. Nonetheless, important examples of relativistic systems satisfying Eq. (4) exist, such as ho- As just described, Hamiltonian constraints force us to mogenous vacuum cosmologies [85] or homogeneous cos- consider dynamical degrees of freedom as time variables. mologies with a massless scalar field [26, 86–88], which While the above considerations hold true for general sys- are often studied in . tems with a single Hamiltonian constraint on finite di- In Appendix A, we argue in more detail why the ab- mensional phase spaces, we shall henceforth work under sence of interactions between clock and system as in further restrictions, which will considerably simplify the Eq. (4) are, in fact, untenable in generic models, featur- subsequent analysis. The reason is that these restrictions ing a non-integrable dynamics. This is also to highlight will permit us to go beyond the formal level in the quan- that the resolution of the ‘clock ambiguity problem’ (re- tum theory and to exhibit the links between three a priori lated to the ‘multiple choice problem’ in quantum gravity distinct approaches to quantum relational dynamics. [10, 11]) proposed in [69] does not apply to generic mod- els. Instead, a change method, such as For the remainder of this article, we consider theo- ries, which permit us to globally partition the degrees of freedom into a clock C and a system S. More pre- cisely, we shall assume for simplicity that the kinematical 3 The assumption dim PC = 2 is not in conflict with the clock phase space can be globally decomposed into a product system possibly being a composite system of many degrees of kin C S, where C and S denote the clock and freedom. In that case, the clock function T may be a collective Psystem≃P phase×P space, respectively.P P While a general phase degree of freedom that is chosen as a time standard, relative space may not globally decompose in this form (e.g. if to which all other degrees of freedom (including the remaining ones in the clock system) evolve. That is, with a choice of time it is compact), locally this can always be achieved. We standard one effectively decomposes the clock system phase space shall also assume that dim C = 2, while dim S can be into the time standard part, PC , and its other degrees of freedom, arbitrary but finite. The reasonP is that a singleP Hamil- which here we simply think of as being contained in the system tonian constraint requires only a single clock function to phase space PS . 7 the one introduced in [25–29] and further developed in (including the Page-Wootters formalism) that is based on Sec. VII and in [65, 83], will become indispensable for using dynamical time observables.4 addressing the ‘clock ambiguity problem’. As we will now show, it is possible to sidestep the is- sue raised by Unruh and Wald by relaxing the require- ment that observables in quantum theory have to be III. COVARIANT TIME OBSERVABLES self-adjoint operators. Instead we will adopt the notion of a generalized observable defined by a POVM, which In the spirit of Misner, Thorne, and Wheeler [89], who is standard in quantum information [91] and quantum remarked “Time is defined so that motion looks simple!”, metrology [75]. In particular, this will permit us to de- we will suppose that the partial time observable T is fine monotonic (covariant) time observables with a well- covariant (simple) with respect to the group generated defined probability interpretation even for bounded clock by the Hamiltonian HC . This will amount to T essen- Hamiltonians. However, the set of possible clock read- tially being canonically conjugate to HC and thus be- ings over which the probability distribution is defined ing monotonic along the orbits generated by the latter. need not be perfectly distinguishable. Nonetheless, this Such time observables are first described in the classical is common to many quantum measurements and not a theory as clock functions and then in the quantum the- fundamental obstruction. ory as positive operator-valued measures (POVMs). In We consider this a resolution of the issue raised by all cases, they capture what we intuitively have in mind Unruh and Wald: by appealing to a more general notion when thinking of a clock, and will be employed in the fol- of an observable characterized by a POVM, the relational lowing sections when discussing the trinity of relational approach to the problem of time is viable also in the quantum dynamics. Henceforth, we will simply refer to presence of realistic Hamiltonians (see also the follow-up T as a time observable. However, we emphasize that T work [83, 84]). is a partial observable, not a complete Dirac observable, since by construction T is not gauge invariant. This section will resolve an apparent monotonicity is- A. Classical time observables sue of relational time observables reported in [77]. As is well-known, and originally observed by Pauli [90], there An (autonomous) Hamiltonian system on a two- ˆ dimensional phase space is completely integrable. cannot exist a self-adjoint time operator T that is canon- PC ically conjugate to a bounded, self-adjoint Hamiltonian Assuming that the phase space flow generated by the 5 HˆC . This observation was refined somewhat by Unruh clock Hamiltonian HC is complete, it follows from Liou- and Wald in [77] who showed that for a bounded Hamilto- ville’s integrability theorem (e.g. see [92]) that we can ˜ nian HˆC there cannot exist a self-adjoint time operator always find some clock function T on C , such that ˆ T,H˜ = u(H ) is a constant of motionP for some func- T which satisfies the following monotonicity (“Heracli- { C } C tian”) property in Schr¨odinger : tion u. Accordingly, the clock T˜ changes at a constant rate along the dynamical trajectories (or remains static (i) There exists an infinite sequence of states for u(HC ) = 0). In this case, we can always choose an- T0 , T1 , T2 ,... with T0 < T1 < T2 < . . . such ˜ | i | i | i other clock function T := T/u(HC), which is canoni- that Tn is an eigenstate of the projection operator | i cally conjugate to the clock Hamiltonian T,HC =1on onto the spectral interval centered around Tn. . This is what we mean classically by simplicity{ } of the PC (ii) For each n there exists m>n such that the tran- clock, i.e. its covariance with respect to HC . Since u may vanish for some trajectories, such a choice T may not be sition amplitude f (t)= T exp( itHˆ ) T to mn m C n globally valid on (e.g. see [25, 26, 83]), although usu- go from T to the larger Th is| non-vanishing− | i for C n m ally one can findP a T with such properties on the (owing some t > 0, so that the clock has a nonvanishing to its integrability) dense subset of where dH = 0.6 probability to run forward. PC C 6

(iii) For each n and all t> 0, fmn(t) = 0 for all m

The choice of T is clearly not unique since T + h(HC ) B. Quantum time observables enjoys the same properties for an arbitrary differentiable function h. In the quantum theory, by a ‘simple’ time observ- Using such a ‘simple’ T and Eq. (4), the power series able we mean a POVM that is covariant with re- expansion of relational Dirac observables in Eq. (3) sim- spect to the group generated by the clock Hamiltonian ˆ plifies for a phase space function fS on S: HC [73, 74, 94]. We describe here such covariant POVMs P and the relation between their properties and the spec- trum of HˆC . Covariant clock POVMs were introduced ∞ (τ T )n F (τ) − f ,H . (5) into relational dynamics in [50, 51, 76], and also recently fS ,T ≈ n! { S S}n n=0 considered in [95]. Here we expound their properties. X Since HˆC is assumed to be a self-adjoint operator, by Stone’s theorem [96] it generates a one-dimensional group For our discussion it will be relevant whether the clock G whose unitary representation on the clock Hilbert ˆ has non-degenerate or degenerate energy levels. Classi- iHC t space C is UC (t) := e− for all t G R, where cally, this means that constant energy surfaces are con- G denotesH the set of values necessary to∈ parametrize⊆ G. nected in the former case and comprised of disconnected The group G can either be compact or noncompact. In pieces in the latter case, such that each connected piece the former case, this implies that for some group element, contains a single dynamical orbit. Liouville’s integrabil- parametrized by tmax G, ity theorem, together with our assumption that the flow ∈ iϕ of HC is complete, further implies that the connected UC (tmax)= e IC , ϕ [0, 2π). (6) components of the constant energy surfaces of the clock ∈ 1 Hamiltonian HC on C are diffeomorphic to either S or The phase ϕ takes into account that the R [92]. Consequently,P the clock function T , being conju- of a system is a ray in Hilbert space. As such, Eq. (6) gate to HC, will be periodic in the former case and run is the condition that UC (t) yields a projective unitary monotonically over an infinite range in the latter case. representation of G, i.e., a representation up to phase. While for periodic clocks T will only take values in a fi- Let (G) denote the Borel σ-algebra of G, so that (G, (GB)) is a measurable space, and let ( ) de- nite interval [0,tmax), one still has to keep track of the B LB HC clock’s ‘winding numbers’ in order to monitor the evolu- note the set of bounded operators on C . A POVM E : (G) ( ) is defined throughH the following tion of S’s degrees of freedom, which may not be periodic T B → LB HC resulting in Eq. (5) being multivalued [84].7 three measure properties (e.g. see [74]) Simple examples of non-degenerate clock Hamiltoni- 1. Positivity: E (X) 0 for all X (G); T ≥ ∈B ans with orbits diffeomorphic to R are HC = cp, 2 2. Normalization: E (G)= I ; with a dimensionful constant c, and HC = p /2m + T C a2 q a1 e , with positive dimensionful constants ai and 3. σ-additivity: ET ( iXi) = ET (Xi) for any se- q R. In the former case, a covariant time ob- ∪ i ∈ quence Xi of disjoint sets in (G). servable is given by T = q/c, and in the latter, by PB 2m p 1 2m A POVM E is said to be covariant with respect to G − T T = p2 a √H coth p2 HC (for p = 0). Non- − 2 C 6 if the self-adjoint effect operators ET (X) satisfy the co- compactq clocks of this kindq will be considered in sec- variance condition tions IV and V. By contrast, an obvious example of a clock with a non-degenerate Hamiltonian with orbits ET (X + t)= UC (t)ET (X)U † (t), (7) 1 C diffeomorphic to S is the harmonic oscillator, HC = 2 2 2 p /2m + mω q /2. In this case, the periodic clock for all X (G) and t G. If a POVM ET is covariant ∈B ∈ function is simply the phase observable T := φ(q,p) = with respect to G, the group generated by HˆC , then we p 1/ω arctan( − ), which satisfies T,H = 1 (so-called will refer to E as a time observable of the clock C. mωq { C} T action-angle variables [93]). Such periodic clocks will We restrict our attention to time observables described be discussed in the following subsection and explored in by effect densities proportional to one-dimensional ‘pro- greater depth in a follow-up article [84]. An example of jection operators’ onto what we will refer to as (possibly a degenerate clock Hamiltonian, H = p2/2m, with or- unnormalizable) clock states t , C | i bits diffeomorphic to R, is studied in the context of the E (dt)= µ dt t t , (8) trinity in a companion article [83]. T | ih | where µ R is a constant. We will explain shortly how the clock∈ states are constructed using the eigenstates of HˆC . The constant µ is fixed by the normalization condi- tion 7 Clocks in everyday life are also periodic, but through calendar days we keep track of the clocks’ ‘winding numbers’ to monitor ET (G)= ET (dt)= IC , (9) a monotonic passage of time. ZG 9 and dt denoting the G invariant Haar measure on G. The is motivation for the above assumption is that effect den- sities not described by one-dimensional ‘projectors’ have Hˆ = dε ε ε ε , (13) C | ih | less resolution [74, 97]. Furthermore, the effect operator Zσc for any X (G) is now given by ET (X)= X ET (dt). From Eq.∈B (9) it follows that the clock states form a where ε denotes an eigenstate of the clock Hamiltonian R with eigenvalue| i ε. The covariance condition in Eq. (10) resolution of the identity and thus a basis for C . How- ever, the clock states need not be orthogonal, andH if they implies that the clock states take the form are not, then this basis is overcomplete. The covariance ig(ε) iεt condition in Eq. (7) then implies that the clock states t = dε e e− ε , (14) | i | i transform under the action of G as Zσc where g(ε) is an arbitrary real function encoding a free- t′ = U (t′ t) t . (10) | i C − | i dom in the choice of clock states. This freedom is the th quantum incarnation of the classical freedom in defin- The n moment operator of the time observable ET is ing a clock that is canonically conjugate to HC (see Ap- pendix B). The overlap of two clock states is given by Tˆ(n) := µ dt tn t t . (11) G | ih | Z iε(t t′) t t′ = dε e − = χ(t t′), (15) ˆ ˆ(1) h | i − We define a time operator T := T as the first moment Zσc operator of the time observable E T where we have defined the function ˆ T = µ dt t t t . (12) 2πδ(x) σc =R, G | ih | Z iεminx 1 χ(x) :=e πδ(x)+ iP x σc =(εmin, ), (16) iε x iεmaxx ∞ This time operator Tˆ is symmetric but not necessarily i e min e σ =(ε ,ε ),  −x  c min max self-adjoint [74], a property we shall revisit shortly. We emphasize that the quantization of a classical clock func- and P denotes the Cauchy principal value. From Eq. (15) tion T should not be associated with the time operator it follows that the clock states have infinite norm and thus 8 Tˆ. Instead, the quantum analog of T is the covariant are not elements of the clock Hilbert space, unless σc is time observable ET , which is a POVM, and therefore bounded above and below. Further, only for σc = R are fully characterized by all of its moment operators Tˆ(n). the clock states orthogonal. In this case, the POVM cor- Nonetheless, considering the time operator Tˆ allows us responds to a projective measurement, and Eq. (12) is to compare the covariant time observable with previous then simply the spectral decomposition of the time op- work. erator. Such clocks are often considered in the literature The possible non-self-adjointness notwithstanding, the and represent an idealization in which the clock states moment operators Tˆ(n) are viable quantum observables are in principle perfectly distinguishable. We henceforth with a consistent probability interpretation; however, refer to such clocks as ideal. That the spectrum of the measurement outcomes t X may not be perfectly dis- clock Hamiltonian is unbounded below in this case is the tinguishable because the∈ clock states need not be or- content of Pauli’s famous remark on the (apparent) im- thogonal. This resolves the issue raised by Unruh and possibility of a physically meaningful time operator [90]. Wald [77]: thanks to the covariance property in Eqs. (7) On the other hand, when σc (R the clock states are not and (10), we have a viable monotonic time observable orthogonal. which we will now describe in more detail. The group G generated by a clock Hamiltonian with In general, the spectrum of the clock Hamiltonian continuous spectrum is noncompact and G = R. This is σC := Spec(HˆC )= σc σp is the union of its continuous because if G were compact, then from Eqs. (6) and (13), spectrum σ and point∪ (discrete) spectrum σ . For sim- it would follow that eiεtmax = eiϕ for all ε σ . However, c p ∈ c plicity, we will only consider non-degenerate clock Hamil- this condition cannot be satisfied since σc contains irra- tonians with spectra that are either entirely continuous tional numbers. This is, of course, the quantum analog σC = σc or entirely discrete σC = σp in the following two of the classical discussion above: a classical Hamiltonian subsections. In [83], we describe the analogous proper- ties for an example of a degenerate, continuous spectrum clock Hamiltonian. 8 More precisely [98], one considers a rigged Hilbert space defined ′ by the triplet Φ ⊂ HC ⊂ Φ , where Φ is a proper subset dense in ′ 1. Continuous spectrum clocks HC and Φ is the dual of Φ, defined through the inner product on HC . In this case, Φ is the Schwarz space of smooth rapidly decreasing functions on R and Φ′ is the space of tempered dis- For non-degenerate continuous spectrum clocks, σC = tributions on R. The clock states are tempered distributions, |ti ∈ Φ′. σc, the spectral decomposition of the clock Hamiltonian 10

HC generating a noncompact flow on a two-dimensional It follows that the clock states are orthogonal if e.g. σp = phase space (usually) leads to a quantum Hamiltonian Z [97]. HˆC with continuous spectrum. Having established that As noted above, if the group G generated by the G = R, and given that the energy eigenstates form a res- clock Hamiltonian is noncompact, then G = R. Insert- olution of the identity, IC = dε ε ε , and Eqs. (9) ing Eq. (19) into the normalization condition, Eq. (9), σc | ih | and (13), it follows that the normalization constant ap- one finds that the result diverges in this case. We R pearing in Eq. (8) is fixed to be µ = 1 . therefore cannot construct a covariant time observable 2π ˆ Using Eqs. (12) and (15), one can verify that the clock in the manner described above when Spec(HC ) = σp states in Eq. (14) are eigenstates of the first moment and G is noncompact. For G to be compact, so that ˆ ˆ G = [0,tmax) R, it follows from Eq. (6) that T , i.e. T t = t t , only in the special case of the ideal ⊂ | i | i ˆ clock, where Spec(HC )= R. In this special case, we also iεj tmax iϕ e = e , εj σp. (20) have that Tˆ is self-adjoint and that Tˆn is equal to the ∀ ∈ nth moment operator of the clock POVM Tˆ(n) given in For Eq. (20) to be satisfied it must be the case that Eq. (11). Differentiating U (s) TUˆ † (s) = Tˆ s I (which fol- nj Z s.t. εj tmax =2πnj + ϕ, εj σp, C C C ∃ ∈ ∀ ∈ lows from Eq. (10), the invariance of the− Haar measure, ˆ and G = R) with respect to s and setting s = 0, one finds which implies that the spectrum of HC reads that the time operator and clock Hamiltonian (formally) 2πnj + ϕ satisfy the canonical commutation relation εj = , εj σp. tmax ∀ ∈

T,ˆ HˆC = iIC . (17) Hence, for G to be compact, the spectrum of HˆC must h i also be rational (see [97] for a related discussion).9 This While this holds for any continuous (non-degenerate) is again the quantum analog of the classical discussion ˆ HC , we note that the time operator and clock Hamil- above: a classical Hamiltonian generating a flow homeo- tonian form a Heisenberg pair (which requires both to morphic to S1 in a two-dimensional phase space (usually) be self-adjoint [99]) only in the case of the ideal clock, in leads to a quantum Hamiltonian with discrete, rational accordance with Pauli’s remark noted above. This point spectrum. Note that the global phase ϕ is only unique has been discussed in another context in [100]. up to multiples of 2π. Finally, using Eq. (14), one also finds that Once more, the normalization condition in Eq. (8) fixes 1 the constant µ = tmax− and Eq. (19) allows for the time 1 i g(ε) i ε t ε = dt e− e t , (18) operator Tˆ to be expressed as | i 2π R | i Z i[g(εj ) g(εk )] which generalizes the Fourier transform to a canonical tmax e − Tˆ = IC + i εj εk . (21) pair with a not necessarily self-adjoint Tˆ in Eq. (17). 2 εj εk | ih | εj ,εk σp − Xj=∈k 6 2. Discrete spectrum clocks The action of the time operator on a clock state is

ig(εj ) tmax e iεk t The spectral decomposition of a clock Hamiltonian Tˆ t = t + i e− εj | i 2 | i εj εk | i with non-degenerate discrete spectrum, σC = σp, is εj ,εk σp − jX=∈k 6 Hˆ = ε ε ε , C j | j ih j | εj σp X∈ where ε denotes an eigenstate of the clock Hamiltonian 9 Generic Hamiltonians featuring an irrational spectrum, how- | j i with eigenvalue εj . The covariance condition in Eq. (10) ever, usually correspond either to complex many body systems implies that the clock states take the form or to classically non-integrable systems. As such, they typi- cally do not arise in the quantization of two-dimensional phase spaces, like that of the clock. Nevertheless, it is interesting ig(εj ) iεj t t = e e− ε . (19) | i | ji to note what would happen for Hamiltonians with irrational εj σp spectrum. The evolution of states on H could be written as ∈ C X −i εk t |ψC (t)i = k ck e |εki and since the ratios of eigenvalues where again g(εj) is an arbitrary real function encoding εk are notP rational numbers, it is impossible to satisfy Eq. (20) a freedom in the choice of clock states. The overlap of for any finite t 6= 0. Hence, the clock has infinite range and the state will never exactly return to its initial state |ψC (0)i. How- two clock states is ever, in aperiodic intervals, the state may get arbitrarily close to ′ |ψC (0)i in the sense that their difference gets arbitrarily close to iεj (t t ) t t′ = e − . h | i the zero-vector. This is the content of the quantum recurrence εj σp theorems [101–103]. X∈ 11

from which it is seen the clock states are not eigenstates self-adjoint in this case, being isomorphic to the posi- of the time operator. Note that the time observable is tion operator on the real line, and HˆC is unbounded be- a POVM with measurement outcomes t G and the low [90]. As a second example, we consider a Hamilto- ˆ ∈ time operator T is defined as its first moment. Thus nian whose spectrum is bounded below, namely HˆC = 2 a2 qˆ one should not expect the clock states to necessarily be pˆ /2m + a1 e on C L2(R), with a1,a2 > 0 eigenstates of Tˆ; see also [74] for a related discussion. and the boundary conditionH ≃ that energy eigenstates van- Using Eq. (21), the of the time operator ish for q where the potential diverges. Defining 10 → ∞ and clock Hamiltonian can be evaluated, ν(ε) :=2 √2mε , the energy are then given a2 by ψ (q) = K ( 2√2ma1 ea2q/2), where K (z) are the i[g(εj ) g(εk )] ε iν(ε) a ν T,ˆ Hˆ = iI i e − ε ε 2 C C − | j ih k| modified Bessel functions of the second kind, from which εj ,εk σp ˆ h i X∈ φt(q) and then T can be constructed as described above. = i I t t . Since σ = R+ is not equal to R in this case, the clock C − | maxih max| c wavefunctions φt(q) t are not orthogonal. As a third ex- ˆ ˆ  { } ˆ qˆ Thus T and HC form a Heisenberg pair on the subspace ample, consider the Hamiltonian HC = c , with the posi- tion operator acting on L (0,a). This Hamiltonian HC ≃ 2 := ψ t ψ =0 , therefore has a doubly-bounded spectrum σc = (0,a/c). D {| i ∈ HC h max| i } ⊂ HC We have energy eigenfunctions ψε(q) = δ(q cε), and − i q t which is dense in the clock Hilbert space C when its (again, non-orthogonal) clock states φt(q) = e− c , and H 2 dimensionality is infinite [99, 104]. Despite this domain hence T (q, q′)= ic δ′(q q′). This example was consid- − restriction, the eigenstates of HˆC can be expressed via a ered in [99], though with restrictions on the domain of Fourier transform of the clock states what we have called T (q, q′). On the other hand, an obvious example of a rational, 1 ig(εj ) iεj t non-degenerate clock spectrum is the Harmonic oscilla- εj = dt e− e t . | i tmax | i tor. In this case, the quantization of the phase observ- ZG able mentioned above serves as the (self-adjoint) clock operator Tˆ = φˆ [74, 105]. The clock states given in 3. Examples of non-degenerate quantum clocks Eq. (19) then fail to be orthogonal. For completeness we have included here a discussion of discrete spectrum clock To illustrate the quantum time observables discussed Hamiltonians and discuss such clocks in detail in the con- above, we now consider some examples. For clocks text of relational quantum dynamics in [84], henceforth governed by a non-degenerate Hamiltonian with a con- considering only noncompact clocks. tinuous spectrum, we construct the time operator Tˆ via a wavefunction representation. Denoting the set of energy eigenfunctions with respect to observable IV. RELATIONAL QUANTUM DYNAMICS IN qˆ by ψε(q) ε, one can then use Eq. (14) to find DIRAC AND REDUCED QUANTIZATION the wavefunctions{ } of the clock states via the Fourier transform φ (q) := dε e iεtψ (q), where for sim- t σc − ε Prior to describing the trinity in Sec. V, we first in- plicity we have chosen g(ε) = 0. The time opera- troduce the formulation of relational quantum dynamics R tor is then given by Tˆ = dq dq′ T (q, q′) q q′ , with in the language of relational observables in Dirac quan- | ih | T (q, q ) := 1 dt t φ (q)φ (q ). tization (‘first quantize, then constrain’). This formula- ′ 2π R t t R′ We now give three examples of non-degenerate, tion will produce the clock-neutral element of the trinity. continuous-spectrumR clocks. First, in analogy to the The word relational is used because the formulation de- classical examples discussed in Sec. IIIA, consider the fines the quantum dynamics of the system S with respect clock governed by HˆC = c pˆ on C L2(R), with to the dynamical clock C, which is described in terms [ˆq, pˆ]= i. Such a clock Hamiltonian hasH a≃ non-degenerate of a covariant time observable (POVM) as discussed in Sec. III. For simplicity, we henceforth restrict our consid- spectrum Spec(HˆC ) = R (i.e. an ideal clock). In this εq 1 i c eration to clocks which possess a non-degenerate, contin- case, we have ψε(q) = √ e , so the clock states 2π uous spectrum Hamiltonian HˆC , and discuss the trinity φt(q) = √2πδ(t q/c) are orthogonal, as anticipated, for degenerate clock Hamiltonians in a companion arti- q− qˆ and T (q, q′) = δ(q q′), i.e. Tˆ = . Clearly Tˆ is cle [83], and postpone the discussion of the trinity for c − c discrete spectrum Hamiltonians to [84]. We also introduce an alternative formulation of rela- tional quantum dynamics obtained through phase space 10 This result can also be derived by differentiating reduction and subsequent quantization (‘first constrain, ˆ † ˆ s UC (s) T UC (s) = T − sIC + 0 dt |tiht|, which follows from the then quantize’), although this will not a priori be an invariance of the Haar measure,R adjusting integration labels and element of the trinity. The other two formulations of re- † limits, and noting that UC (tmax) |tiht| UC (tmax)= |tiht|. lational quantum dynamics which complete the trinity 12

in Sec. V are obtained through the quantum analog of where IC and IS denote the identity operators acting on phase space reduction. The relation among these latter C and S, respectively. three formulations will be studied in Sec. V. H AssumingH this equation has a non-trivial solution, by assumption zero will lie in the continuous spectrum of 14 CˆH since HˆC has a continuous spectrum. Accordingly, A. Dynamics I: Relational Dirac observables solutions to Eq. (23) will be improper eigenstates of CˆH and so not be normalizable in kin. That is, ψphys / 11 H | i ∈ Dirac’s constraint quantization algorithm begins by kin. Using group averaging [3, 107, 109, 110], we can H quantizing the kinematical phase space kin C S, project an arbitrary kinematical state onto a physical by promoting suitable phase space coordinatesP ≃P to×P op- state, erators on what is known as the kinematical Hilbert ˆ space . The direct sum structure of the classical ψphys = δ(CH ) ψkin Hkin | i | i phase space suggests a preferred partitioning of the kine- 1 isCˆH = ds e ψkin matical Hilbert space kin C S, where C and 2π R | i are the Hilbert spacesH describing≃ H ⊗ the H clock andH system Z HS degrees of freedom, which here are simply quantizations = ψkin( E, E) E C E S , (24) E σSC − |− i | i of and , respectively. We assume that this quanti- Z ∈ PC PS P zation leads to a self-adjoint and non-degenerate clock where Hamiltonian HˆC acting on C with continuous spec- H σ := Spec(Hˆ ) Spec( Hˆ ). (25) trum. The clock variable is then quantized via the co- SC S ∩ − C variant clock POVM ET , defined through the clock states In order to normalize physical states, we define a new ˆ ˆ in Eq. (14), yielding a canonical pair [T, HC ]= i thanks inner product on the space of solutions to Eq. (23), using to Eq. (17). Recall that Tˆ need not necessarily be self- the group averaging projector and the kinematical inner adjoint. Similarly, we assume that a suitable Poisson product kin on kin, subalgebra S of phase space observables on S are pro- h· |·i H A Q P 12 ˆ moted to a quantum representation S on S, from ψphys φphys phys := ψkin δ(CH ) φkin kin (26) which the full set of self-adjoint systemA observablesH on h | i h | | i S, assumed to include the quantum Hamiltonian HˆS, = ψkin∗ ( E, E) φkin( E, E). H E σSC − − can be constructed (usually involving a choice of factor Z ∈ P ordering). For our purposes, it will not be necessary to Here, φkin is any representative of the equivalence class Q | i specify the properties of S any further. of states in kin, which project under Eq. (24) onto the A 13 H Under our assumptions, an arbitrary kinematical same physical state φphys , and similarly for ψkin . This state can expanded as defines an inner product| i on the space of solutionsh | to Eq. (23). Modulo subtleties irrelevant for the present ψ = dε ψ (ε, E) ε E , (22) discussion, the space of solutions can then be Cauchy | kini kin | iC | iS Zσc ZE completed to a Hilbert space of physical states phys P H where the sum-integral notation here and below accounts [3, 107, 109, 110]. We stress that phys kin. We can think of the physical HilbertH space6⊂ H as the for the discrete or continuous nature of the system Hamil- Hphys tonian’s spectrum. ‘quantum constraint surface’. Note, however, that phys- ical states are gauge invariant since U (s) ψ = The constraint in Eq. (4) is implemented by demand- CS | physi i s CˆH isHˆC isHˆS ing that physical states of the quantum theory are an- ψphys , where UCS(s) := e− = e− e− . | i ⊗ nihilated by the associated constraint operator, assumed In other words, physical states do not change under the ˆ to be self-adjoint on , resulting in a Wheeler-DeWitt evolution generated by CH . This is in contrast with the Hkin type equation classical case, where CH generates a non-trivial flow on . In the context of quantum gravity, this leads to whatC is CˆH ψphys = HˆC IS + IC HˆS ψphys =0, (23) known as the problem of time or the ‘frozen formalism’ | i ⊗ ⊗ | i [10, 11, 63]. As such, physical states are often considered  as ‘timeless’. However, we argue, in line with [25, 26], that it is more appropriate to regard physical states as 11 The precise technical formulation of the algorithm has evolved ‘clock-neutral’; they correspond to a global description of over time [2, 3, 79, 81, 106, 107]. Here, we implement the algo- physics, prior to choosing a temporal reference system. rithm using group averaging techniques [3, 107]. 12 Q In Dirac quantization, one usually attempts to solve AS is in general a small subset of the linear operators L(HS) due to the Groenewold-van-Hove theorem which implies that one can- the problem of time relationally by promoting a choice of not map the full Poisson algebra of classical phase space functions homomorphically into a quantum commutator algebra [108]. 13 If the spectrum of HˆS were degenerate, we would have to intro- duce additional labels, but this would not change the 14 Usually, this means that the flow generated by the classical con- subsequent discussion. straint CH is non-compact in Pkin. 13

relational Dirac observables to operators acting on phys. time observable’ is therefore not recognized as a Dirac This involves a choice of clock, of which there areHa pri- observable.16 Furthermore, while completing this work ori many among the kinematical operators on kin. The we noticed that a similar expression to the fourth line in physical Hilbert space encodes simultaneouslyH a multi- Eq. (27) was recently carefully constructed as a quantiza- tude of these different choices and their associated re- tion of relational Dirac observables in [30]. The starting lational quantum dynamics because the choice of clock point of [30] is different: it begins with integral tech- is made after constructing phys [25, 26] (“tempus post niques for relational observables [13, 19], rather than the quantum” [11]). Accordingly,H we consider Dirac quanti- power-series expansions [14–17] used here, and it also zation as producing an a priori clock-neutral picture. does not employ covariant clock POVMs. We will fur- Here we choose as a temporal reference system the ther discuss the relation with our work in Sec. VIII A. clock C associated with the Hilbert space , Hamil- The following theorem shows that Fˆ (τ) is (for- HC fS ,T tonian HˆC , and covariant time observable ET . Using mally) a family of Dirac observables and thus gauge in- th the n moment operator of ET given in Eq. (11), about variant. t = τ, we define the quantization of the (formal) power ˆ 15 Theorem 1. FfS ,T (τ) is a (strong) Dirac observable, series in Eq. (5) of relational Dirac observables as ˆ that is, FfS ,T (τ) commutes algebraically with the con- ∞ n straint operator of Cˆ ˆ 1 i n ˆ ˆ H FfS ,T (τ) := dt t t (t τ) fS, HS 2π R | ih |⊗ n! − n n=0 Cˆ , Fˆ (τ) =0. (28) Z X h i H fS ,T 1 ˆ ˆ i(τ t)HS ˆ i(τ t)HS h i = dt t t e − fS e− − Proof. The proof is in Appendix C. 2π R | ih |⊗ Z 1 itHˆS itHˆS While the operator families in Eq. (27) are thus strong = dt t + τ t + τ e− fˆS e 2π R | ih |⊗ quantum Dirac observables, we will only be interested Z in their weak action, i.e. their action on . To sim- 1 ˆ Hphys = dt UCS(t) τ τ fS UCS† (t) plify notation, we introduce the notion of a quantum weak 2π R | ih |⊗ Z   equality between operators in analogy to the classical =: τ τ fˆ , (27) case, indicating their equality on the ‘quantum constraint G | ih |⊗ S   surface’ phys: th H where [fˆS, HˆS]n := [[fˆS, HˆS]n 1, HˆS] is the n -order − Oˆ1 Oˆ2 (29) nested commutator, with the convention [fˆ , Hˆ ] := fˆ , ≈ S S 0 S ˆ ˆ ˆ O1 ψphys = O2 ψphys , ψphys phys . and where fS is the quantization of the classical function ⇔ | i | i ∀ | i ∈ H fS. The second equality is obtained from the Baker- Furthermore, Let ΠσSC be the projector from S to its Campbell-Hausdorff formula, the third equality follows subspace spanned by all system energy eigenstatesH E | iS from changing integration variables t t + τ and noting with E σSC , with σSC given in Eq. (25), i.e. those that the Haar measure dt is invariant→ under the action permitted∈ upon solving the constraint. As such, we will of G, and the last equality makes use of the definition of denote this system Hilbert subspace phys := Π ( ) HS σSC HS UCS(t). The fourth line makes clear that this construc- and refer to it as the physical system Hilbert space. For tion can be viewed as a group averaging of the kinemati- later purpose, let us denote by ˆ cal operator τ τ fS. Such a group averaging is known ˆphys ˆ | ih |⊗ fS := ΠσSC fS ΠσSC , as a G-twirl operation of τ τ fˆS over the noncom- G | ih |⊗ phys pact one-parameter unitary group generated by CˆH (see ˆ the projection of an arbitrary fS ( S) to S . [37, 61, 78] for a discussion of G-twirl operations in the ∈ L H L H We are now in a position to see that the quantum re- context of spatial quantum reference frames).   lational Dirac observables in Eq. (27) form weak equiva- An expression similar to the one in the second line lence classes, as shown by the following result. of Eq. (27) was also recently proposed in the context of covariant clock POVMs as a “relative time observable” Lemma 1. The quantum relational Dirac observables ˆ ˆ in [95]. However, the interpretation in [95] is very dif- FfS ,T (τ) and F phys (τ) are weakly equal, i.e. coincide fS ,T ferent: a constraint is not considered and the ‘relative on phys. Hence, the relational Dirac observables associ- atedH to system observables form equivalence classes where ˆ ˆ ˆ FfS ,T (τ) and FgS ,T (τ) are equivalent if ΠσSC fS ΠσSC = Π gˆ Π . 15 As usual, the Groenewold-van-Hove-theorem [108] implies that σSC S σSC only a strict subset of the Poisson-algebra of Dirac observables on C will be homomorphically mapped to a commutator algebra of quantum Dirac observables under this quantization prescrip- 16 Instead, the authors of [95] propose to use it to describe how a tion. We assume that a suitable choice of such a subalgebra has clock evolves relative to some other reference system. The invari- been made. This is combined with the choice of AS above, its Q ant ‘relative time observable’ is then evaluated in non-invariant quantum representation AS and may involve a choice of factor states (kinematical states in the language of constraint quanti- ordering in the quantization fS 7→ fˆS. zation), which we consider undesirable. 14

Proof. The proof is given in Appendix C. As an aside, we note that only in the special case of an ideal clock (Spec(HˆC ) = R), in which case Tˆ t = t t ˆ | i | i When ΠσSC is non-trivial, the set of relational Dirac and T is self-adjoint, can we simplify Eq. (27) to observables Fˆf ,T (τ) associated to system observables fˆS S ˆ ˆ ˆ ˆ ˆ i(τ T ) HS ˆ i(τ T ) HS evolving relative to ET is therefore “not as big” on the FfS ,T (τ)= e − ⊗ IC fS e− − ⊗ . physical Hilbert space as the set of system observables ⊗ ˆ ˆphys Relational Dirac observables in this form have previously fS on S itself. The operators fS thus label the weak equivalenceH classes of relational Dirac observables with appeared in the context of homogeneous quantum cos- respect to ET . This will become crucial when showing mology, e.g. see [85]. equivalence with the other approaches to relational quan- The relational quantum dynamics on phys then H tum dynamics below. In particular, fˆphys will turn out to amount to letting the parameter τ run, which corre- S ˆ be the system operators of the Page-Wootters formalism. sponds to the values that the time observable T can take. In [14] it was shown that classically the relational Dirac In particular, one can evaluate the relational Dirac ob- observables in Eq. (3) define weakly an algebra homo- servables on physical states using the physical inner prod- uct, ψphys Fˆf ,T (τ) ψphys phys, as defined in Eq. (26). morphism f Ff,T (τ) with respect to addition, multi- h | S | i plication, and7→ the Poisson bracket. The following the- This provides a sense of evolution, despite physical states ˆ orem proves that the appropriate analog is also true in not evolving under the action of CH . the quantum theory: the equivalence classes of relational

Dirac observables inherit their algebraic properties on the 18 physical Hilbert space directly from the algebraic prop- B. Reduced phase space quantization ˆphys phys erties of the operators fS acting on S . H We separate this discussion into two parts, the first Theorem 2. The map deals with classical phase space reduction and the second with the quantization of the reduced phase space (see phys also [72] for general comments on this topic in the context FT (τ): S ( phys) L H → L H of relational Dirac observables).  ˆphys ˆ fS Ffˆphys,T (τ) 7→ S is weakly an algebra homomorphism with respect to ad- 1. Classical phase space reduction dition, multiplication and the commutator. That is, the following holds: The clock-neutral constraint surface is not a phase space, but rather a presymplectic manifoldC of ˆ Ff phys+gphys hphys,T (τ) dim kin 1. However, the description of the dynam- S S · S P − ˆ ˆ ˆ ics relative to our choice of temporal reference system T Ff phys,T (τ)+ Fgphys,T (τ) Fhphys,T (τ), ≈ S S · S will lead to a phase space description. This is achieved

through a gauge fixing procedure. Since FfS ,T (τ) is con- and stant along the gauge orbits while nevertheless fully en- coding the dynamics through the parameter τ, we are ˆ ˆ ˆ Ff phys,T (τ), Fgphys ,T (τ) F[f phys,gphys],T (τ), S S ≈ S S free to gauge fix to remove the now-redundant clock de- h i grees of freedom without losing information. (We do where is the quantum weak equality of Eq. (29). not want to evolve the clock relative to itself [25, 26].) ≈ Since T, CH = 1, one can choose for simplicity T = 0. Proof. The proof is given in Appendix C. For unbounded{ } clocks with G = R, to which we have restricted, this singles out exactly one point on each ˆ ˜ Usually, FfS ,T (τ) is required to be self-adjoint on gauge orbit for which T = T/h(HC) is well-defined (cf. .17 However, at this formal level, we shall not ad- Sec. IIIA). In line with the integrability property of the Hphys dress this issue here, but only comment on it later in clock, we shall assume this to be the case on a dense Sec. V. By utilizing covariant POVMs, our procedure per- subset of orbits, so that T = 0 constitutes a good gauge mits us to extend the construction of quantum relational Dirac observables to covariant time observables ET not necessarily described by a self-adjoint time operator. We shall discuss this further in Sec. VIII. 18 The following subsection is not strictly necessary for understand- ing the trinity in Sec. V and may be skipped on a first reading. We include it here for completeness as this method is an often em- ployed formulation of relational dynamics. We will later discuss the relation between reduced phase space quantization and the 17 Alternatively, in line with the sprit of this paper, one could ex- trinity. It will also become useful for understanding the quantum plore the generalization of observables on Hphys defined in terms analog of ‘gauge-invariant extensions of gauge-fixed quantities’ in of POVMs rather that self-adjoint operators. Sec. VIIIA. 15

fixing condition for almost all orbits. In the special case 2. Reduced quantization that HC = cp, setting T = q/c to zero is in fact valid for all orbits. We proceed with the quantization of the gauge fixed re- The reduced phase space is the space of gauge orbits, red red duced phase space S on a suitable Hilbert space S . i.e., the quotient space / , where identifies points on Given that red mayP not be isomorphic to , redHmay C ∼ ∼ PS PS HS a given orbit generated by CH . With our (possibly only differ from the system Hilbert space S used in Dirac almost) globally valid gauge fixing condition at hand, red H quantization. On S we choose a suitable Poisson sub- T =0, where T =0 is the gauge fixing surface in kin P ˜ C ∩ S S P algebra of functions S and promote it to a quantum defined by T = 0 (see Fig. 2), is equivalent to / (or Q A representation ˜ on red, from which the self-adjoint a dense subset thereof). The Dirac bracket [79,C 81],∼ in- S S system observables,A includingH the reduced system Hamil- ducing the Poisson structure on this gauge fixed reduced tonian Hˆ red, are constructed. Arbitrary states of the sys- phase space from that on , reads in this case S kin 19 ˆ red P tem can then be expanded in the eigenbasis of HS F, G := F, G F, C T, G + F,T C , G , { }D { }−{ H }{ } { }{ H } red red for all F, G on . All Dirac brackets involving the now ψS = ψS (E) E S , (32) C | i E σred | i redundant clock variable T and the constraint CH vanish, Z ∈ S P while fS,gS D fS,gS for fS,gS functions on S. { } ≡ { } P red ˆ red We can thus simply drop the redundant and fixed clock where σS = Spec(HS ). Assuming as usual that variables (T = 0,H = H ) and are left with a gauge C S f(E) E E′ = f(E′) for an arbitrary complex − red h | i fixed reduced phase space [79, 81], henceforth denoted by E σS red Z ∈ S T =0. functionP f, their inner product reads P To≃C∩S emphasize that the functions corresponding to sys- red red red tem degrees of freedom now live on the phase space S , ψ φ = ψ∗ (E) φ (E). (33) red P S S S S we equip them with the label , although as functions h | i E σred Z ∈ S of the phase space coordinates they will be the same. P Note that red need not necessarily be isomorphic to We emphasize that ˜Q may differ from Q used in PS AS AS S (see [25, 111] for simple examples). Indeed, the S Dirac quantization (e.g. see [25, 111, 112]), leading to P red degrees of freedom may be further restricted on S : possibly different spectral properties of self-adjoint ob- redP red due to having solved the constraints, image red (H )= ˆ S S servables. In general, the reduced Hamiltonian HS may P ˆ image S (HS ) image C ( HC), where imageX (f) de- not have the same spectrum as HS does on phys, that P ∩ P − notes the image of function f on domain X. Notice also ˆ red ˆ ˆ H is, Spec(HS ) Spec(HS) Spec( HC)= σSC may not that here we are making use of the non-degeneracy condi- ≡ ∩ − red hold. Firstly, our gauge-fixed phase space S , which tion on the clock Hamiltonian HC . Since the HC = const we are quantizing, may only be a dense subsetP of the full surfaces are connected by assumption, the procedure reduced phase space / , as discussed above. The latter yields a single reduced phase space. This will no longer may thus actually requireC ∼ a parametrization in terms of be the case when the Hamiltonian is degenerate (e.g. red different coordinates than those used on S . This is rel- [25, 26, 83, 111]). evant as the procedure of Cauchy completionP leading to This reduced phase space red is interpreted as the dy- red PS S should render the quantization of / equivalent to namics described relative to the temporal reference sys- theH quantization of a dense subset thereof.C ∼ Secondly, the tem T [25, 26]. Indeed, under the gauge fixing condition value set of Hred on red may only be a strict subset of S PS T = 0, the relational Dirac observables in Eq. (5) reduce that of HS on kin due to having solved the constraints. to Thus, while locallyP the structures of red and may PS PS ∞ τ n agree, it is well known that global phase space properties F red(τ)= f redHred , (30) fS n! { S S }n severely influence which observables can be promoted to n=0 X self-adjoint operators at all and, if they can, what their where we made use of fS,gS D fS,gS , as noted domain is, thereby directly affecting their spectral prop- above. It is clear that they{ satisfy} the≡ standard { } equations erties [113]. of motion of the system S This entails repercussions for the relation between Dirac and reduced quantization, which, for the systems dF red fS red red red red considered here, we can not always expect to be exactly = F ,H D F ,H , (31) dτ { fS S } ≡{ fS S } equivalent. Our work thereby adds to the previous lit- but now interpreted relative to the dynamical clock T . erature on the relation of the two quantization methods In particular, given that the evolution parameter τ runs (e.g. [34, 106, 111, 112, 114–116]). There are, however, over all the possible values of T , we have τ G = R. In the context of relational dynamics, this∈ reduction procedure is often called a classical ‘deparametrization’ 19 ˆ red with respect to the clock choice T . We construct the Again, if HS had a degenerate spectrum, we would have to quantum analog in Sec. VB. introduce additional labels. 16

models for which we will be able to establish an exact in [84]). This is accomplished by formulating Dynamics ˆ red equivalence. A sufficient condition is Spec(HS )= σSC , II and III in terms of invertible quantum reduction maps which clearly holds for arbitrary HˆS in the simple case from the physical Hilbert space phys, defined by the 2 H where HˆC = c pˆ on C = L (R). The equivalence will constraint in Eq. (23), to reduced Hilbert spaces associ- H 2 a2 qˆ ated with the relational Schr¨odinger picture of Dynamics also hold when HˆC =p ˆ /2+ a1 e and HˆS is (minus) an arbitrary positive Hamiltonian. II and the relational Heisenberg picture of Dynamics III. red The relation between these three relational dynamics is On S we can now define the quantization of the reducedH evolving observables in Eq. (30) as summarized in Fig. 3. While this immediately establishes the equivalence be- ∞ ( iτ)n tween quantum relational Dirac observables, the Page- Fˆred(τ)= − [fˆredHˆ red] fS n! S S n Wootters formalism, and the relational Heisenberg pic- n=0 ture obtained through quantum reduction, it will not al- Xˆ red ˆ red i HS τ ˆred i HS τ ˆred ways be the case that the latter coincides with the rela- = e fS e− fS (τ), (34) ≡ tional Heisenberg picture of reduced phase space quanti- where in the last line we have made use of the Baker- zation described in Sec. IVB. ˆred Campbell-Hausdorff formula. For FfS (τ) to be self- Moreover, the quantum reduction maps referenced red above are isometries that can be used to map observables adjoint on S , the classical function fS must be pro- moted to aH self-adjoint operator, which may require a in both the relational Schr¨odinger and Heisenberg pic- choice of factor ordering. tures to quantum relational Dirac observables on phys H It is clear that the reduced evolving observables in of the form given Eq. (27), and vice versa. As a by- Eq. (34) satisfy the quantum analog of Eq. (31), namely product, this provides a new construction procedure for the Heisenberg equations of motion with respect to τ quantum relational Dirac observables from observables on the reduced Hilbert spaces associated with Dynam- dFˆred ics II and III. fS = i [Hˆ red, Fˆred]= i [Hˆ red, fˆred]. dτ S fS S S To help keep track of the numerous Hilbert spaces in- volved in establishing the trinity, we summarize them in In terms of expectation values, relational evolution takes Table I. the form ψred Fˆred(τ) ψred . Recall again that τ G = h S | fS | S i ∈ R. Hilbert space Description

Altogether, the states in the reduced quantum theory HC Clock C Hilbert space do not evolve in τ, while observables do. Hence the result HS System S Hilbert space of reduced phase space quantization20 yields a relational H ≃ HC ⊗ HS Kinematical Hilbert space Heisenberg picture. Another relational Heisenberg pic- kin Hphys ≃ δ(CˆH )(Hkin) Physical Hilbert space ture will be obtained through quantum symmetry reduc- phys tion in Sec. VB1, which will be shown to be equivalent HS =ΠσSC (HS ) ⊆ HS Physical system Hilbert space to the one above under certain conditions. red System Hilbert space obtained HS ⊆ HS by reduced quantization

V. THE TRINITY OF RELATIONAL TABLE I. The various Hilbert spaces used in the following phys QUANTUM DYNAMICS discussion are summarized here. While Hphys ≃ HS are always isometric, they are only isometric to the Hilbert space red HS of reduced phase space quantization when Eq. (57) is Having introduced Dynamics I in Sec. IV A, defined satisfied. in terms of relational Dirac observables, we now de- scribe two additional a priori distinct formulations of re- lational quantum dynamics: the Page-Wootters formal- ism (Dynamics II) and the relational Heisenberg picture A. Dynamics II: The Page-Wootters formalism obtained from a quantum symmetry reduction procedure, which constitutes a quantum deparametrization (Dynam- ics III). We establish the equivalence between these three The proposal of Page and Wootters [44, 45, 117, 118] relational dynamics under the condition that the clock also begins by quantizing the constraint in Eq. (23), and ˆ from a physical state ψphys seeks to recover a relational Hamiltonian HC has a continuous non-degenerate spec- | i 21 trum (this is generalized to a doubly degenerate spec- quantum dynamics between the clock and system. This trum in [83] and to periodic i.e. discrete-spectrum clocks is accomplished by phrasing any statement we would nor- mally make about the time dependence of a system as a

20 The reduced quantum theory obtained through a clock gauge fix- ing is often also called a quantum theory that is ‘deparametrized’ 21 In most of the literature on the Page-Wootters formalism the with respect to a clock choice. physical state |ψphysi is denoted as | ψii. 17

question conditional on the clock: What is the proba- where we have used Eq. (10) to write the first equality bility of an observable fˆS associated with the system S and Eq. (23) to moving from the second to third equality. giving a particular outcome f, if the a clock measurement Let us suppose that the physical state is normalized of the time observable E yields the time τ? such that ψS(τ) ψS (τ) = 1 for all τ G. This can be T h | i ∈ We first introduce the Page-Wootters formalism and related to the normalization of physical states if we de- subsequently show its equivalence to the relational dy- fine the following inner product, first introduced in [50], namics in terms of quantum relational Dirac observables on the space of solutions to the quantum constraint in defined in Sec. IV A. Eq. (23), ψ ψ := ψ τ τ I ψ h phys| physiPW h phys| | ih |⊗ S | physikin 1. Introducing the Page-Wootters formalism = ψS (τ) ψS (τ) =1, (38) h | i  for all τ G = R. Notice that this defines a priori a dif- The clock states τ are again taken to be the covariant ferent normalization∈ on the space of solutions to Eq. (23) ones defined in Eq.| (i14) (recall that we have restricted than the physical inner product in Eq. (26) obtained via to noncompact clocks for the remainder). Let eT (τ) := group averaging. As such, the two inner products could τ τ be the ‘effect operator’ corresponding to the clock a priori lead to two different Cauchy completions of the | ih | reading τ. Similarly, suppose that the effect operator space of solutions to Eq. (23). However, in Sec. VA2 we ˆ efS (f) is associated with the system observable fS taking shall show that the physical inner product and the Page- the value f. It is standard in the literature on the Page Wootters inner product in Eq. (38) are, in fact, equiva- and Wootters approach to then compute the probability lent, and thereby do not give rise to two different physical of f given that the clock reads the time τ by postulating Hilbert spaces. the in the following form:22 The definition of the Page-Wootters inner product in Eq. (38) allows us to express the probability in Eq. (35) Prob(f and τ) Prob(f when τ)= (35) purely in terms of the conditional state Prob(τ) Prob(f when τ)= ψS (τ) efS (f) ψS(τ) . (39) ψphys eT (τ) efS (f) ψphys h | | i = h | ⊗ | ikin . Given that the conditional state ψS(t) satisfies the ψphys eT (τ) IS ψphys h | ⊗ | ikin Schr¨odinger equation (37), this agrees| withi the standard We write here ‘postulate’ as it has so far not been clar- time-dependent probability for the outcome f of the sys- ified in the literature whether these expectation values tem observable fˆS. In particular, the expectation value are actually gauge-invariant. In Sec. VA2, we shall of fˆS evolves as usual fˆS (τ) = ψS (τ) fˆS ψS(τ) . Ac- show that these expectation values can be equivalently cordingly, it is justifiedh to henceforthi h call| the| conditionali written in terms of the quantum relational Dirac ob- state formulation of Page and Wootters the relational servables and the physical inner product on of Hphys Schr¨odinger picture. Sec. IV A. Since these structures are manifestly gauge- We mention an often neglected subtlety: since the con- invariant, this shows that indeed the conditional proba- ditional states in Eq. (36) come from conditioning the bility above is gauge-invariant. physical states in Eq. (24), it is clear that the space From a physical perspective, the conditional probabili- spanned by them is simply the physical system Hilbert ties in Eq. (35) are usually justified as follows. To recover space phys, which may be a proper subspace of the sys- the Schr¨odinger equation, let us define the conditional HS tem Hilbert space S used in kinematical quantization. state of the system given that the clock reads τ as For consistency, weH will thus restrict the permissible set of system observables in the conditional state formula- ψS (τ) := τ IS ψphys . (36) | i h |⊗ | i tion to any observable fˆS, acting on S , which leaves As shown in Refs. [44, 45], these conditional states satisfy phys H its subspace S invariant. This will become relevant the Schr¨odinger equation in the clock time τ: when showingH equivalence with quantum relational ob-

′ servables below. Note that in the often considered special d d iHˆC (τ τ ) i ψS(τ) = i τ ′ e − IS ψphys ˆ phys dτ | i dτ h | ⊗ | i case HC = c pˆ, S = S so that no such restriction applies. H H = τ Hˆ I ψ − h | C ⊗ S | physi = τ Cˆ I Hˆ ψ − h | H − C ⊗ S | physi ˆ 2. Equivalence of Dynamics I and II = HS ψS(τ) ,  (37) | i The central ingredient in the Page-Wootters formal- isim is the definition of the conditional state in Eq. (36), 22 We highlight here the labels ‘kin’ and ‘phys’ to clarify the rela- which defines what we will call the Page-Wootters reduc- tion to the structures in Dirac quantization. This is not usually phys tion map S : phys S , defined as done in the literature on the Page and Wootters approach where R H → H subtleties of Dirac quantization are often ignored. S(τ) := τ I . (40) R h |⊗ S 18

The label S on the reduction map stands for ‘Schr¨odinger Dynamics I: picture’ to distinguish it from the Heisenberg picture re- Clock-neutral picture duction map of the following subsection. We write this Specified by relational label in bold face in order to also distinguish it from the Dirac observables italic S which stands for ‘system’. This map has a left phys R R − inverse phys from solutions ψS(t = τ) of the H H S ) 1 H → H | 23i S S Schr¨odinger equation Eq. (37) at the fixed time t = τ I I = = e ⊗ T − | −  ⊗ T i τ ε τ 1 1  | ∗ 1 τ  S− (τ) := dt t US(t τ) )( e ig ( R ) = H  R 2π | i⊗ − τ τ ˆ ( C ε Z ( | ( ∗ δ ⊗ = δ(Cˆ ) ( τ I ). (41) S ) H S | | i⊗ R ε I ∗ C ) =  ) τ C ( T Indeed, 1 ⊗ T − S I R S 1 dt  S− (τ) S(τ) ψphys = t τ US(t τ) ψphys R R | i R 2π | ih |⊗ − | i Dynamics II: Dynamics III: Z US (τ) dt Relational Relational = t t ψphys Schr¨odingerpicture Heisenberg picture R 2π | ih | i Z Page-Wootters U † τ Quantum symmetry = ψ , S ( ) | physi formalism reduction using that the clock states form a resolution of the iden- tity, Eq. (9). In particular, FIG. 3. The trinity of relational quantum dynamics. This figure depicts the reduction maps from the physical Hilbert phys 1 space Hphys to the physical system Hilbert space HS and S− (τ) S(τ)= δ(Cˆ ) ( τ τ I )= I . (42) R R H | ih |⊗ S phys their inverses. These maps are used to transform states and observables between the clock-neutral picture given by Dirac Conversely, one finds the identity acting on conditional quantization, the relational Schr¨odinger picture derived from states (defined by clock C) in the form the Page-Wootters formalism, and the relational Heisenberg picture. It is these maps that are used to prove the equiva- 1 S(τ) S− (τ)= τ δ(CˆH ) τ lence between these three relational quantum dynamics com- R R h | | i prising the trinity. 1 = dt χ∗(t τ) US(t τ) 2π R − − Z = Πσ , Theorem 3. Let fˆ ( ). The quantum relational SC S ∈ L HS Dirac observable Fˆ (τ) acting on , Eq. (27), re- fS ,T Hphys where ΠσSC is the projector onto the physical system duces under S(τ) to the corresponding projected observ- Hilbert space and the last line follows from Eq. (C3). R phys able in the relational Schr¨odinger picture on S , The Page-Wootters reduction map and its inverse H can be used to construct an encoding operation ˆ 1 ˆ S (τ) FfS ,T (τ) S− (τ) = ΠσSC fS ΠσSC . τ phys R R S : ( ), where ( ) denotes the set E L HS → L Hphys L H ˆphys phys of linear operators from to itself. This operation en- Conversely, let fS S . The encoding oper- physH ∈ L H codes observables on S into Dirac observables acting ation in Eq. (43) of system observables coincides on the on the physical HilbertH space and is defined as Hphys physical Hilbert space phys with the quantum relational Dirac observables in Eq.H (27), i.e. τ ˆphys 1 ˆphys S fS := S− (τ) fS S(τ) E R R τ ˆphys ˆ   S fS Ff phys,T (τ), (44) = δ(Cˆ ) τ τ fˆphys . (43) E ≈ S H | ih |⊗ S     where is the quantum weak equality of Eq. (29). Indeed, as the following theorem shows, this encoding re- ≈ produces precisely the quantum relational Dirac observ- Proof. The proof is in Appendix C. ables from Dirac quantization in sec. IV A. In particular, when ΠσSC = IS, we have a many-to-one relation 6 ˆ ˆphys ΠσSC fS ΠσSC = fS .

23 ˆ ˆ The input to the inverse map has to be a state |ψS(t = τ)i, not Lemma 1 asserts that FfS ,T (τ) coincides with F phys (τ) fS ,T a family of states |ψS(t)i. on , which when combined with Theorem 3 estab- Hphys 19 lishes a (formal) equivalence between the full sets of re- B. Dynamics III: Relational Heisenberg picture lational Dirac observables Eq. (27) on phys and system through quantum deparametrization phys H observables on S . This constructionH of Dirac observables in terms of the First, we showcase the quantum symmetry reduction τ ˆphys encoding map elucidates that S(fS ) corresponds to procedure taking us from the clock-neutral Dirac quanti- ˆphys E zation to a relational Heisenberg picture relative to clock the system observable fS “when the clock observ- able yields the value τ”. This becomes especially clear observable ET . Thereafter, we explore the relation with through the next theorem. It shows that the expecta- reduced quantization. As shown in [25, 26] (see also tion values of quantum relational Dirac observables in [39, 40] for spatial quantum reference frames), this pro- the physical inner product, Eq. (26), coincide with the cedure consists of two steps: expectation values of the encoded system observables in 1. Constraint trivialization: A transformation of the the Page-Wootters inner product, Eq. (38), and with the constraint such that it only acts on the chosen ref- expectation values of the system observables in the rela- erence system (here a clock), fixing its degrees of tional Schr¨odinger picture, Eq. (39). freedom. ˆ ˆphys Theorem 4. Let fS ( S) and fS = 2. Conditioning on classical gauge fixing conditions: ˆ ∈ L H phys ΠσSC fS ΠσSC be its associated operator on . Then A ‘projection’ which removes the now redundant HS reference frame degrees of freedom.24 ˆ φphys FfS ,T (τ) ψphys h | | iphys This quantum symmetry reduction procedure consti- ˆphys = φS(τ) f ψS(τ) tutes a quantum deparametrization. h | S | i τ ˆphys = φphys S fS ψphys , h |E | iPW   1. Quantum symmetry reduction and equivalence with where ψS (τ) = S(τ) ψphys . Dynamics I | i R | i Proof. The proof is in Appendix C. We define the trivialization map on and Hkin Hphys An important corollary immediate follows. relative to the covariant time observable in terms of its nth moment operators: ˆ Corollary 1. Setting fS = ΠσSC in Theorem 4 shows ∞ n n the equivalence of the physical inner product in Eq. (26) i (n) T := Tˆ HˆS + ε and the Page-Wootters inner product in Eq. (38) on T n! ⊗ ∗ n=0   phys, and therefore that the Page-Wootters reduction X 1 ˆ H phys it(HS +ε∗) map S(τ) defines an isometry . That is, = dt t t e , (45) phys S 2π R | ih |⊗ R H → H Z φphys ψphys phys = φphys ψphys PW ˆ h | i h | i and require that ε Spec(HC ). The reason for the ∗ ∈ = φS (τ) ψS (τ) , latter requirement will become clear shortly. Let us also h | i define for all conditional and physical states related by 1 ˆ ψS(τ) = S(τ) ψphys . 1 it(HS +ε∗) T− := dt t t e− , (46) | i R | i T 2π R | ih |⊗ Hence, the two inner products for physical states (for- Z mally) define the same physical Hilbert space . Fur- which will turn out to be the inverse of the trivialization Hphys thermore, since the Page-Wootters reduction map S(τ) on phys, as established in the following Lemma. We R H 1 is invertible, this section proves the formal equivalence note that the trivialization map and T− need not be (n) T of the relational quantum dynamics on phys as encoded unitary on kin since Tˆ need not be self-adjoint (cf. H phys H in quantum relational Dirac observables and on Sec. IIIB1). However, this will not be a problem as HS as encoded in the relational Schr¨odinger picture of the we are only interested in its action on phys, where the H Page-Wootters formalism. In particular, the above re- following holds: sults show that the Page-Wootters formalism is mani- festly gauge invariant (and therefore physically further justified), which to the best of our knowledge was not explicitly established before. 24 While it is a true projection on the kinematical Hilbert space, it is As a final remark, note that Theorem 4 shows for- not a projection when applied to the physical Hilbert space, as it ˆphys only removes redundant information, namely degrees of freedom mally that if the system observable fS is self-adjoint already fixed through the constraint. No physical information is on phys, then so should be Fˆ (τ) on , given the lost. Hence, we put projection into quotation marks as it can be HS fS ,T Hphys invertibility of S(τ). inverted for physical (but not for kinematical) states. R 20

Lemma 2. The trivialization map given in Eq. (45) triv- a function of the clock reading τ, in contrast to the Page- ializes the constraint to the clock degrees of freedom Wootters reduction in Eq. (40); on physical states the a priori τ-dependence on the right hand side of Eq. (49) ˆ 1 ˆ T CH T− = HC ε IS , (47) drops out. We can interpret this as a state in a rela- T T − ∗ ⊗ tional Heisenberg picture, provided we make the further   1 identification for any ε R. Furthermore, for ε Spec(HˆC ), − ∗ ∈ ∗ ∈ TT is the left inverse of T on physical states, i g( E) T ψS(E) e− − ψkin( E, E). (51) 1 ≡ − T− T Iphys, T ◦ T ≈ Notice that since ψS(E) is square-integrable/summable, phys and the trivialization transforms physical states into we have H( phys) S S, and so again find the product states with a fixed and redundant clock factor physicalR systemH Hilbert≃ H space⊆ as H the image of the quan- tum symmetry reduction. The label H will henceforth i g(ε∗) T ψphys = e ε C (48) signify ‘Heisenberg picture’. T | i | ∗i i g( E) The inverse of the reduction map in Eq. (49) is [25, 26] e− − ψkin( E, E) E S . ⊗ E σSC − | i Z ∈ 1 1 i g(ε∗) P H− = T− e ε C IS . (52) Proof. The proof is given in Appendix C. R T | ∗i ⊗   Equation (47) holds regardless of the value of ε , while Indeed, we have the following: ∗ Eq. (48) is only true for ε Spec(HˆC ). Indeed, CˆH ∗ ∈ Lemma 3. On physical states, the quantum symmetry and HˆC ε will only have the same spectrum if T is − ∗ T reduction map is equal to unitary on kin, which is only true if Spec(HˆC ) = R, H (n) in which case the clock states are orthogonal and Tˆ H τ U † (τ) (53) R ≈ h |⊗ S is self-adjoint (cf. Sec. IIIB1). For example, if HˆC is while its inverse can be written as bounded and HˆS is unbounded, then CˆH and HˆC ε − ∗ will have distinct spectra. However, this is of no concern 1 H− = δ(Cˆ ) ( τ U (τ)) . (54) to us since we are not interested in the full spectrum of R H | i⊗ S ˆ CH on kin, but only in its zero-eigenspace, namely the Moreover, the two maps are the appropriate inverses of H space phys of physical states. Here, we will need T to one another: be invertibleH and to preserve the zero-eigenvalue, whichT ˆ 1 is the case when ε Spec(HC ). H− H = Iphys, ∗ ∈ R ◦ R We emphasize that T is not a transformation 1 T H H− = ΠσSC . on phys, but instead a transformation of it, since R ◦ R clearlyH Eq. (48) no longer satisfies the original constraint Proof. The proof is in Appendix C. Eq. (23), but instead the transformed constraint Eq. (47). Note that the trivialization map disentangles the clock This permits us to define a new encoding map of and system, which were originally entangled in the phys- phys evolving observables on S into Dirac observables, ical state given in Eq. (24). We will discuss this point in phys H H : ( ) ( ). We may choose any Heisen- more depth in Sec. VI. E L HS → L Hphys The redundant clock factor in Eq. (48) carries no more berg picture observable information about the original state ψphys and can be ˆ ˆ | i ˆphys iτHS ˆphys iτHS removed by a ‘projection’ onto the classical gauge fix- fS (τ)= e fS e− (55) ing condition T = τ, cf. Sec. IVB1. Accordingly, we phys phys phys define the complete quantum symmetry reduction map on , which is why we may set fˆ (0) = fˆ , and HS S S phys H( phys) to the relational Heisenberg picture define the encoding as (generalizingH → R theH procedure of [25, 26] to include also non- phys 1 phys orthogonal clock states) as H fˆ (τ) := H− fˆ (τ) H. (56) E S R S R i ε∗ τ   H := e− ( τ I ) . (49) R h |⊗ S TT Note, that we therefore do not equip the Heisenberg red It follows from Eqs. (14) and (48) that picture observables with the label , in contrast to Sec. IVB on reduced quantization; the relation between phys red i g( E) S and S remains to be investigated. The following H ψphys = e− − ψkin( E, E) E S , theoremH confirmsH that the encoded observables coincide R | i E σSC − | i Z ∈ with the quantum relational Dirac observables on . P (50) Hphys ˆ which is independent of the parameter τ. For this reason Theorem 5. Let fS ( S ). The quantum relational ∈ L H we do not write this quantum deparametrization map as Dirac observables Fˆ (τ) on , Eq. (27), reduce fS ,T Hphys 21 under H to the corresponding projected evolving ob- Theorem 6 formally implies that the same applies to R phys ˆ servables of the relational Heisenberg picture on , FfS ,T (τ) on phys. S H Eq. (55), i.e. H This generalizes the quantum symmetry reduction pro- cedure introduced in Refs. [25, 26], to which we refer the ˆ 1 ˆ H FfS ,T (τ) H− = ΠσSC fS(τ) ΠσSC . reader for an explicit exposition in two concrete models. R R We note that it seems fruitful to explore the connection ˆphys phys with a recent algebraic approach to establishing a quan- Conversely, let fS (τ) S be any evolving ob- ∈ L H tum version of symplectic reduction [119], which may be servable, Eq. (55). In analogy to Eq. (44), related to the procedure exhibited here. phys H ˆ ˆ phys fS (τ) Ff ,T (τ). E ≈ S   2. Relation with reduced phase space quantization Proof. The proof is given in Appendix C.

It is evident that Lastly, we comment on the relation with reduced phase ˆ ˆ ˆ iτHS ˆ iτHS space quantization of Sec. IVB. ΠσSC fS(τ) ΠσSC = e ΠσSC fS ΠσSC e−

phys Corollary 3. The relational Heisenberg picture on is an element of . Owing to Lemma 1, this the- phys L HS , obtained through the quantum symmetry reduc- HS orem thereby establishes  an equivalence between the full tion H, is only equivalent to the relational Heisenberg ˆ R sets of relational Dirac observables FfS ,T (τ) on phys picture of reduced phase space quantization described in ˆphys H and of the evolving system observables fS (τ) of the Sec. IV B if relational Heisenberg picture on phys (see also the dis- HS Spec(Hˆ red) = Spec(Hˆ ) Spec( Hˆ )= σ . (57) cussion below Theorem 3). S S ∩ − C SC The next result shows that the expectation values of the quantum relational Dirac observables, Eq. (27), in Specifically, in this case, the physical inner product, Eq. (26), coincide with the phys (i) red := H ( ), expectation values of the corresponding evolving observ- HS ≃ HS R Hphys phys ables of the relational Heisenberg picture on . red phys 1 S (ii) Hˆ Hˆ := H Hˆ H− , and H S ≡ S R S R Theorem 6. ˆ ˆphys Let fS ( S ) and fS (τ) = (iii) The set of quantum symmetry re- ˆ ˆ ∈ L H iτHS ˆ iτHS e ΠσSC fS ΠσSC e− be its associated evolving duced evolving observables in Eq. (55), phys phys 1 Heisenberg operator on . Then ˆ H ˆ phys S fS (τ)= Ff ,T (τ) H− , coincides with H R S R ˆred φ Fˆ (τ) ψ = φ fˆphys(τ) ψ , the set of evolving observables fS (τ), Eq. (34), h phys| fS ,T | physiphys h S | S | S i resulting from reduced phase space quantization. phys In particular, under the appropriate identi- where ψS = H ψphys S . red | i R | i ∈ H fications, ψ ψS = H ψphys and | S i ≡ | i R | i Proof. The proof is given in Appendix C. fˆphys(τ) fˆred(τ), we have S ≡ S Again, an important corollary immediately follows. φred fˆred(τ) ψred φ fˆphys(τ) ψ h S | S | S i ≡ h S | S | S i Corollary 2. Setting fˆS = IS in Theorem 6 shows that ˆ = φphys Ff phys,T (τ) ψphys phys . the quantum symmetry reduction map H preserves the h | S | i inner product R Proof. The proof is given in Appendix C. φphys ψphys phys = φS ψS , h | i h | i Hence, if Eq. (57) is satisfied, then the relational dy- where and denote the inner products on namics of the quantization of the reduced phase space h·|·iphys h·|·i and phys, respectively, and physical and reduced and Dirac quantization are equivalent. The simplest Hphys HS states are related by ψ = H ψ . Hence, H example is the special case of the ideal clock where S phys ˆ 2 ˆ (formally) defines an isometry.| i R | i R HC = c pˆ on L (R) and HS arbitrary. Another exam- 2 a2 qˆ 2 ple is HˆC =p ˆ /2+ a1 e , ai > 0, on L (R) (but energy Given that the quantum symmetry reduction proce- eigenstates only required to vanish as q + ) and HˆS dure is invertible, we have thereby established a formal equal to (minus) the harmonic oscillator→ or free∞ particle equivalence between the dynamics encoded in the quan- Hamiltonian. tum relational Dirac observables on the clock-neutral If Eq. (57) is not satisfied, then reduced and Dirac physical Hilbert space phys and the relational Heisen- quantization will not be exactly equivalent (see also [72, phys H berg picture on S . Specifically, if the evolving re- 111, 112, 114–116]). In this case it can still happen that H phys phys red 1 duced observables fˆ (τ) are self-adjoint on , then one can embed into [111], here through H− . S HS HS Hphys R 22

C. Equivalence of Dynamics II and III of freedom, while still using a Page-Wootters reduction scheme. This observation relies on a reinterpretation of In the previous two subsections, we have demonstrated the trinity which we have just established and in partic- the formal equivalence of Dirac quantization with the ular Lemma 2. Page-Wootters formalism, as well as with the relational Heisenberg picture obtained through a quantum symme- try reduction procedure. Therefore, the Page-Wootters A. Reinterpreting the trinity formalism, which we had already identified as the rela- tional Schr¨odinger picture, is equivalent with this rela- tional Heisenberg picture. It is thus obvious that the Recall that the quantum symmetry reduction map in Page-Wootters formalism and the relational Heisenberg Eq. (49) is a two-step process which we may write us- picture of the quantum reduction must be related by the ing the Page-Wootters reduction map in Eq. (40) as ′ iε∗τ unitary evolution US(τ). Indeed, Eqs. (40) and (41), as H = ′S(τ ′) T , where ′S(τ ′) := e− S(τ ′). Re- well as Eqs. (53) and (54), directly imply callR alsoR fromT Sec. VB1Rthat the trivializationR map yields a transformation of the physical Hilbert space, S(τ) U (τ) H, R ≈ S · R which we may interpret as a new physical Hilbert space 1 1 ′ := T ( phys). This permits us to reinterpret the S− (τ)= H− US† (τ). phys R R · trinityH diagramT H of Fig. 3 in terms of two Page-Wootters This completes the proof of the formal equivalence of reductions on two different physical Hilbert space repre- the three elements of the trinity of relational quantum sentations (not depicting inverse maps) as follows: dynamics depicted in Fig. 3 for clock Hamiltonians with non-degenerate and continuous spectrum. T T ′ Hphys Hphys

VI. DISENTANGLING THE PAGE-WOOTTERS H ′ ′ S(τ) R S(τ ) FORMALISM R R

† phys US (τ) phys In the context of the Page-Wootters formalism, it is HS HS sometimes stressed that the emergence of time from the ‘timeless’ quantum theory defined by the Hamiltonian (Recall that the image of H does not depend on τ ′.) constraint Eq. (23) originates in the entanglement be- The left and right Page-WoottersR reductions produce, of tween the clock and system (e.g. [44, 45, 49, 69, 120]). course, the relational Schr¨odinger and Heisenberg pic- This is suggested by the shape of physical states in tures, respectively, which both live on the same physical Eq. (24) or by expanding the physical state in the clock system Hilbert space phys. state basis HS Equation (52) implies that the inverse map from the 1 relational Heisenberg picture on phys to the trivialized ψphys = dτ τ ψS(τ) . HS | i 2π R | i | i physical Hilbert space is τ -independent and given Z phys′ ′ by H Wootters emphasizes this point [45]

1 ig(ε∗) One motivation for considering such a “con- S′ − = e ε IS densation” of history [i.e. physical state] is R | ∗i⊗ ˆ the desire for economy as regards the num- = δ(HC ε ) t =0 IS , − ∗ | i ⊗ ber of basic elements of the theory: quantum   correlations are an integral part of quantum where we have made use of Eq. (18). This is a product theory already; so one is not adding a new version of Eq. (41), relative to the trivialized constraint element to the theory. And yet an old ele- Eq. (47). ment, time, is being eliminated, becoming a We have seen in Lemma 2 that the trivialization map secondary and even approximate concept. acts as a disentangling map on the physical Hilbert TT Enticing though this may be, we shall now explain why space; states in ′ are product states between clock Hphys one has to be careful with this picture of the emergence of and system relative to the tensor factorization of kin. H time evolution. In short, this entanglement within phys- Using the reduction maps, it is now straightforward to ical states Eq. (24) is not gauge-invariant, but defined show that all relational observables on ′ , i.e. the Hphys with respect to a tensor factorization of the kinemati- trivialization of the relational Dirac observables from cal Hilbert space which is not inherited by the physical phys, are also product observables. To this end, we H Hilbert space. As we shall demonstrate, one can also first define a new encoding of the evolving observables obtain the same relational dynamics without any (kine- of the relational Heisenberg picture on phys. Denot- HS matical) entanglement between clock and system degrees ing ψ′ := ψ in Eq. (48), we find weakly on | physi TT | physi 23

phys′ especially when zero lies in the continuous spectrum of H the constraint(s) as in this article.25 It is correct that phys 1 phys physical states Eq. (24) are entangled with respect to S′ fˆ (τ) ψ′ = S′ − fˆ (τ) S′ (τ ′) ψ′ E S | physi R S R | physi the kinematical structure in the sense of ′   ˆ iε∗τ not being separable. However, given that is not a = δ(HC ε) t =0 τ ′ e− Hphys − | ih | subspace of kin (thanks to Eq. (26) physical states can  phys  H fˆ (τ) ψ′ be thought of as distributions on ), physical states ⊗ S | physi Hkin ˆphys do not give rise to all the probabilistic consequences of = ′ 0 0 fS (τ) ψphys′ entanglement on , in particular in terms of correla- G | ih |⊗ | i Hkin  phys  tions, because they are not normalizable with respect to = I fˆ (τ) ψ′ , C ⊗ S | physi the kinematical inner product. This notion of entangle- ment is in any case kinematical, and not gauge-invariant. where ′ denotes the G-twirl with respect to the group G As we shall now argue, it cannot be probed using gauge- generated by the trivialized constraint (HˆC ε ) IS. Notice that − ∗ ⊗ invariant Dirac observables. ˆ ˆ A physical notion of entanglement must be defined in Ff′S ,T (τ) := ′ 0 0 fS(τ) G | ih |⊗ terms of structures on phys. Let us now argue that the   H are the adaptations of the relational Dirac observables kinematical tensor product decomposition between clock and system, used to construct phys, in fact does not in Eq. (27) to the new representation on phys′ with re- H spect to the trivialized constraint. ExploitingH the trinity survive on the latter. This is a consequence of the re- of Sec. V, it is also clear that these coincide with the dundancy on the physical Hilbert space. As a result of the constraint defining the physical Hilbert space not all trivialized relational Dirac observables from phys: H of the physical degrees of freedom are independent be- 1 Fˆ′ (τ) ψ′ = Fˆ (τ) − ψ′ cause some get fixed, while others will be algebraically fS ,T | physi TT fS ,T TT | physi phys related. This is especially evident from the trivialized = I fˆ (τ) ψ′ . (58) C S phys physical Hilbert space ′ and the shape of its states ⊗ | i Hphys Since the trivialized constraint only acts on the clock Eq. (48); their clock factor is entirely redundant. But it factor, this result is to be expected. is also apparent from an algebraic perspective: a gauge- The entire relational dynamics relative to the covari- invariant tensor factorization of phys must manifest it- self in terms of commuting subalgebrasH of Dirac observ- ant time observable ET is therefore encoded in product states, Eq. (48), and product observables, Eq. (58), on ables. Are there subalgebras of Dirac observables that depend only on clock and system degrees of freedom, re- phys′ with respect to the kinematical tensor product. H The fact that one can always change a tensor factor- spectively, which commute and can thereby establish that ization on a Hilbert space through an entangling unitary the physical Hilbert space factors into a clock and system may lead one at first to think that this observation is un- decomposition? The only independent clock Dirac ob- ˆ ˆ surprising. Let us explain why the situation is, in fact, servable is its Hamiltonian HC , but due to Eq. (23), HC ˆ more subtle. While we may also interpret the trivializa- is the same observable as HS on phys, up to an over- all negative sign. Owing to the redundancyH on , tion T as a passive transformation which changes the Hphys partitioningT of the theory into clock and system, it leads there do not exist independent commuting subalgebras to crucial differences compared to standard unitary re- of Dirac observables corresponding purely to clock and partitionings of a Hilbert space: system degrees of freedom, respectively. In this sense, phys does not inherit the kinematical tensor decompo- (a) The trivialization map is generally not a unitary H TT on kin with respect to which the tensor factoriza- tionH is defined. (It is unitary if the clock states in Eq. (14) are orthogonal.) In fact, it may not even be invertible on . By contrast, Lemma 2 proves Hkin that is invertible between and ′ , TT Hphys Hphys which is why Eq. (58) only holds weakly. 25 A extreme example exhibiting the difference between kinematical (b) The clock factor for all observables and states and gauge-invariant entanglement is 3D vacuum quantum grav- ity. Kinematically, the theory has local degrees of freedom and on phys′ is completely fixed through the con- straintH and contains no more information about the accordingly there may be all kinds of entanglement on its kine- matical Hilbert space. However, upon imposing the constraints, physics; it is redundant. All non-trivial physical in- the theory becomes topological and thus devoid of local gauge- formation is encoded in the system factor. invariant degrees of freedom. The physical Hilbert space turns out to be one-dimensional for 3D vacuum quantum gravity (with This highlights that one has to be careful with the pic- genus-one spatial hypersurfaces) [121]: it has a unique physical ture that dynamics emerges from entanglement. Indeed, state which is also not part of the kinematical Hilbert space. the notion of entanglement in gauge theories is subtle, Kinematical entanglement has become physically irrelevant. 24

26 sition between clock and system. We can now formally Dirac quantize kin using the In conclusion, entanglement does play a role in the new canonically conjugate pairs. The followingP discus- emergence of time evolution, but only a kinematical sion is formal because the canonical transformation TT notion of it and even this is not strictly necessary. may not always be globally valid, so that the new canoni- i j Upon Page-Wootters reduction, kinematically entangled cal coordinates (T, PT ; QS(τ), PS (τ)) may not be defined physical states yield the relational Schr¨odinger picture. everywhere on kin. For example, we have already seen However, one obtains the unitarily equivalent relational in Sec. IIIA thatP T may be ill-defined on subsets of Pkin Heisenberg picture also through Page-Wootters reduc- and, depending on HC and HS, the new clock momen- tion, but in this case of kinematically unentangled states tum PT may not actually take values in the full real line. from phys′ . To strengthen this last point, we argue now In that case, we can not simply promote the pair (T, PT ) that thisH trivialized physical Hilbert space can sometimes to a pair of canonically conjugate self-adjoint operators be regarded as the result of a Dirac quantization of the on a new clock Hilbert space ′ . Instead, one could HC same classical system Eq. (4), but with respect to a dif- employ affine quantization [25, 113], promoting PT to ferent set of phase space coordinates. a self-adjoint operator on ′ and defining the quan- HC tum analog of T on ′ , as in Sec. IIIB, in terms of HC a covariant clock POVM, this time with respect to PˆT . B. Classical analog of the trivialization More generally, it may be necessary to resort to geomet- ric quantization techniques [113, 124]. Leaving such global challenges aside, formally the kine- For this section only, let us assume that the sys- matical Hilbert space = ′ ′ is spanned by tem phase space is parametrized by canonical pairs kin′ C S S the states H H ⊗ H (qi ,pi )N and theP clock phase space is parametrized S S i=1 PC by a canonical pair (t,pt), for simplicity all taking values j j j ψ′ = dP dP ψ′ (P , P ) P P . in the full reals. The classical analog of the trivialization | kini T S kin T { S} | T i | Si Z j T is a canonical transformation TT on kin = C S, Y Twhich splits the new canonical coordinatesP intoP ⊕P pure The constraint we need to now impose is Pˆ and thus gauge degrees of freedom on the one hand, and pure Dirac T already trivialized. Hence, physical states defining a new observables on the other: physical Hilbert space ′′ are Hphys i i N i j (t,pt; qS,pS)i=1 T, PT := CH ; QS(τ), PS (τ) , 7→ ψ′′ := δ(PˆT ) IS ψ′   | physi ⊗ | kini where   j j j = P =0 dP ψ′ (0, P ) P , | T i⊗ S kin { S} | Si i j j Q (τ) := F i (τ) , P (τ) := F i (τ), Z S qS ,T S pS ,T Y in analogy to the trivialized physical states ψ TT | physi and FfS ,T (τ) is given in Eq. (5); for systems with con- of Eq. (48). Similarly, it is clear that a complete set straints linear in the momenta see also [116, 122, 123]. of Dirac observables in this decomposition is simply the T The transformation T is shown to be canonical in Ap- kinematical operators pendix E1 and is sometimes called an abelianization of constraints when there are several [14, 79]. ˆi ˆj IC QS(τ) and IC PS(τ), We note that we can also interpret this as a passive ⊗ ⊗ transformation which changes the decomposition of the in analogy to the trivialized relational Dirac observables kinematical phase space from = into = in Eq. (58); all other Dirac observables will be functions Pkin PC ×PS Pkin ′ ′ , where, e.g., ′ is now parametrized by the of these Dirac observables. The physical Hilbert space of PC ×PS PC canonical pair (T, CH ) and thereby depends on the old this Dirac quantization is trivialized by construction. degrees of freedom. What is the relation between this new physical PS Hilbert space ′′ and the trivialized Hilbert space Hphys ′ := ( )? When H is classically unbounded Hphys TT Hphys C in both directions, and thus Spec(HˆC ) = Spec(PˆT )= R, 26 Something similar happens when considering two qubits, H ≃ the two coincide, phys′ phys′′ . In this case, the canon- C2 ⊗C2, and restricting to the three-dimensional subspace of the H ≃ H ical transformation TT is globally defined on kin and the symmetric sector, Hsym ⊂ H. On this subspace the observables i j P relative to one qubit can be considered as dependent on those relational Dirac observables QS(τ), PS (τ) take values in of the other. Likewise, this subspace does not inherit the tensor all of the reals, even on the constraint surface . In par- ˆ ˆ ˆi Cˆj product structure of H of which it is a subspace in the sense ticular, one can quantize (T, PT ) and (QS(τ), PS (τ)) as that it cannot be written as a non-trivial tensor product (after canonically conjugate self-adjoint operators on kin′ and all, it is three-dimensional). The difference is that in the qubit H this extends to ′′ for the latter pairs. Hence, their case there is no gauge symmetry. Hence, H is already ‘physical’ Hphys spectrum on ′′ is the full reals. Likewise, in this case and thus so too is the entanglement with respect to its tensor Hphys product structure. we have σ = Spec(Hˆ ) on , i.e. the system energy SC S Hphys 25 does not get restricted on the physical Hilbert space and canonically conjugate, if fˆS andg ˆS are canonically con- phys we have S = S . Hence, we can identify a complete jugate. set of trivializedH DiracH observables in Eq. (58) with This is the quantum analog of how, classically, the Poisson-algebra of relational Dirac observables on the i j IC qˆS(τ) and IC pˆS(τ), constraint surface is determined using the Dirac bracket ⊗ ⊗ on the gauge fixingC surfaces [14–17]. More generally, re- i j ˆphys ˆ ˆphys ˆ where theq ˆS, pˆS are the system observables defining the calling that fS (τ) = exp(iHSτ) fS exp( iHSτ), it ˆ ˆ − relational Dirac observables Fqi ,T (τ), F j (τ). Their is clear that Eqs. (58) and (59) are a manifestation of S pS ,T spectrum will likewise be the full real line, given that the (weak) quantum algebra homomorphism established phys in Theorem 2. S = S. Accordingly, we have phys′ phys′′ and weH can identifyH the two quantum theoriesH on≃ Hthem. We note that in this special case the trivialization is actu- ally a unitary operator on and we have ′ = VII. CHANGING TEMPORAL REFERENCE Hkin Hkin T ( kin). FRAMES T WhileH there may be other cases in which this equiv- alence holds, it is unlikely that the two quantum theo- We now explain how a change of temporal reference ries on ′ and ′′ coincide in general, even if one Hphys Hphys frame is performed in both the Page-Wootters formalism could cope with the global challenges alluded to above. and the relational Heisenberg picture obtained through In fact, their relation will generally be of a similar kind as quantum symmetry reduction, and, owing to the trinity, that between Dirac and reduced quantization discussed in changes between these pictures. Recall that a temporal Sec. VB1. The Groenewold-van-Hove theorem [108, 124] reference frame (system) is a clock C associated with a implies that two quantizations of the same system with Hilbert space C , a Hamiltonian HˆC , and a time observ- respect to different sets of canonically conjugate coordi- H able ET associated with a POVM that is covariant with nates cannot in general be unitarily equivalent. In our respect to the group generated by HˆC and defined by the case, this means that and ′ will not in general Hkin Hkin set of clock states t , t G . A change of temporal be unitarily equivalent and this is consistent with the reference frame therefore{| i ∀ means∈ } changing the clock with fact that T is not in general unitary on kin. This will respect to which the dynamics of a system is specified. render theT question of whether the spectraH of Dirac ob- We examine in sequence how states and observables servables coincide in the two theories a complicated one. transform under a change of temporal reference frame. In the context of quantum gravity, this point has been To construct the temporal frame change (TFC) map, raised before [10, 11] (see also [85] where an equivalence we will make use of the reduction maps and their in- between Dirac quantization of homogeneous cosmological verses, given for the relational Schr¨odinger picture (Page- models with respect to two different canonical coordinate Wootters formalism) in Eqs. (40) and (41) and for the sets could be established). relational Heisenberg picture in Eqs. (53) and (54). We Regardless of whether the trivialized Hilbert space and then use the TFC map to briefly examine the relativity the Dirac quantization of the classically trivialized the- of temporal locality. In what follows we thereby gen- ory coincide, the trivialization map can in general be T eralize (and recover) the recent temporal frame change viewed as the quantum analog of theT classical canonical operations developed in Ref. [25, 26] for the relational transformation T . T Heisenberg picture and reduced quantization, and later in Ref. [65] for the Page-Wootters formalism. In partic- ular, we will show that they are equivalent. C. Simplifying commutators

As an aside, the above observations are useful for the computation of commutators of relational Dirac observ- A. State transformations ables on phys. Observe that H Consider two clocks (temporal reference frames), A ˆ ˆ 1 and B, and a system S whose dynamics we are inter- T FfS ,T (τ), FgS ,T (τ) T− ψphys′ (59) T T | i ested in describing with respect to either clock. Suppose h ˆ i ˆ = Ff′S ,T (τ), Fg′S ,T (τ) ψphys′ the physical states of the theory satisfy the constraint | i equation h phys physi = I fˆ (τ), gˆ (τ) ψ′ . C ⊗ S S | physi h i Cˆ ψ = Hˆ + Hˆ + Hˆ ψ =0, (60) ˆ ˆphys phys H | physi A B S | physi For example, suppose OS := [fS (τ), gˆS (τ)] is a con- phys   stant of motion on S . Then it immediately follows where for simplicity we have suppressed tensor products ˆ ˆ H ˆ that [FfS ,T (τ), FgS ,T (τ)] = IC OS on phys. This of identity operators (e.g. HˆA = HˆA IB IS). In the ˆ ⊗ˆ H ⊗ ⊗ demonstrates that FfS ,T (τ) and FgS ,T (τ) are (weakly) relational Schr¨odinger and Heisenberg pictures, the state 26

of clock B and system S with respect to clock A is27 Temporal frame change using the Page-Wootters formalism

ψBS A(τA) := S(τA) ψphys , | | i R | i R − ψBS A := H,A ψphys , S 1 ) (τ (τB | | i R | i A ) RS while the state of A and S with respect to B is U † U U † U BS(τA) BS(τA) |ψphys AS(τB) AS(τB) ψAS B(τB ) := S(τB) ψphys , | | i R | i −1 RH ψAS B := H,B ψphys . H,A ,B | | i R | i R ΛA→B −1 For clarity in the frame change procedure below, we at- H = RH,B ◦ RH,A tach the reference frame label A or B to the Heisenberg |ψBS|A |ψAS|B reduction map and to the clock reading τ in the case of Temporal frame change in the symmetry reduced quantum theory the Schr¨odinger reduction map. A change of temporal reference frames is performed by FIG. 4. The temporal frame change (TFC) maps in the acting on the state of BS relative to A with the inverse relational Schr¨odinger picture (Page-Wootters formalism), A→B A→B reduction map associated with A, followed by the clock ΛS , and the relational Heisenberg picture, ΛH , as well B reduction map. The composition of these two maps as TFC maps acting in-between them, as given in Eq. (62). yields the TFC maps which take states relative to A to To transform the state of clock B and system S and with re- A B phys phys spect to clock A, to the state of A and S with respect to B, states relative to B, that is, Λ → : HB ⊗ HS → we must first pass to the physical Hilbert space via the inverse phys phys,28 and where, depending on which rela- HA ⊗ HS of the reduction map, indicated by the arrows pointing from tional picture we work in and whether we also change the top and bottom left corners to the center, followed by the relational picture, the application of the reduction map, depicted by the arrows pointing from the center to the top and bottom right corners. A B 1 ΛS → := S(τ ) S− (τ ) R B ◦ R A A B 1 ΛH→ := H H− R ,B ◦ R ,A to another coordinate description of the local physics. In- A B ΛH→ S := S(τB ) H,A, deed, here we can think of Eq. (62) as defining a “quan- → R ◦ R A B tum coordinate change”. The temporal reference frames ΛS →H := H,B S(τA), (62) → R ◦ R A and B define two possible descriptions in the coordi- The structure of these four ways of changing frame from nates τA and τB for the quantum evolution of the remain- ing degrees of freedom. The physical Hilbert space , A to B is depicted in Fig. 4. Hphys Thanks to the compositional structure in Eq. (62), defined here by Eq. (60), assumes the analogous role of A B the TFC map Λ → always passes through the physi- the manifold since it is independent of the choice of which cal Hilbert space phys. For instance, in the relational subsystem is used as a temporal reference system. The H 1 physical Hilbert space encodes a multitude of such tem- Schr¨odinger picture S− (τA) ψBS A(τA) phys as shown in Sec. VA2,R and similarly| | fori the ∈ relational H poral frame choices (clock perspectives), not just A and B. This is why we may think of as defining a clock- Heisenberg picture. The TFC map thereby has the Hphys compositional structure analogous to coordinate changes neutral [25, 26], rather than timeless quantum theory; it 1 is a quantum description prior to having chosen a tempo- ϕB ϕA− on a manifold. For example, in general relativ- ity these◦ pass from one coordinate description of the local ral quantum reference frame. The framework developed physics via the reference frame independent (i.e. coordi- here thereby contributes to the more general perspective- nate independent) description of the spacetime manifold, neutral approach to both spatial and temporal quan- tum reference frames introduced in [25, 26, 39, 40, 80]. Changes of perspective (i.e. quantum reference frame) in this approach always proceed via the perspective-neutral 27 With two clocks, as described by the constraint in Eq. (60), one physical Hilbert space; see Fig. 5 for more discussion. can apply a second reduction map to the state yielding twice The TFC map, defined in Eq. (62), transforms states conditioned state of S in the relational Schr¨odinger picture as

|ψ (τ , τ )i := RS(τ ) ◦ RS(τ ) |ψ i . S|AB A B A B phys A B phys phys phys phys ΛS → : B S A S , Note that RS(τA)RS(τB ) = RS(τB )RS(τA). An expansion of H ⊗ H → H ⊗ H A B a physical state may be specified in terms of this twice reduced ψBS A(τA) ψAS B(τB ) =ΛS → ψBS A(τA) , state | | i 7→ | | i | | i A B |ψ i = dτ dτ |τ i|τ i|ψ (τ , τ )i . where ΛS → is the operator phys Z A B A B S|AB A B A B 1 ΛS → := S(τB ) S− (τA) 28 phys phys Note that HB ⊗ HS is the physical subspace of HB ⊗ HS , R ◦ R i.e. the subspace permitted by the constraint Eq. (60). = ( τ I ) δ(Cˆ ) ( τ I ) , (63) h B |⊗ AS H | Ai⊗ BS 27

The perspective-neutral physical Hilbert space of Dirac quantization 1. Observable transformations in the relational Schr¨odinger picture

Hphys Consider in the relational Schr¨odinger picture the ob- servable Oˆphys ( phys phys) associated with BS A ∈ L HB ⊗ HS BS ‘seen’ from| the perspective of A. Demanding the expectation value of Oˆphys with the untrans- R R BS A A B | formed state ψBS A(τA) be equal to the expectation | | i The perspective of The perspective of value of the transformed observable, which we denote quantum reference frame A!B 1 quantum reference frame phys phys phys Λ = RB ◦ R Oˆ (τ , τ ) ( ) on the transformed A is described by the A A is described by the AS B A B A S | ∈ L H ⊗ H Hilbert space Hrest;BjA Hilbert space Hrest;SjB state implies that ˆphys ψBS A(τA) OBS A ψBS A(τA) FIG. 5. A change of quantum frame perspective has the same h | | | | | i compositional structure as coordinate changes on a manifold. ˆphys = ψAS B(τB ) OAS B(τA, τB) ψAS B(τB ) . The ‘quantum coordinate maps’ RA and RB take as their in- h | | | | | i put the perspective-neutral physics on H and map it to a phys The appearance of the evolution parameters τ , τ in B’s description relative to the perspective of either quantum ref- A B Schr¨odinger picture wil be clarified shortly. It then fol- erence frame A or B. The quantum coordinate maps RA, RB are maps between Hilbert spaces (quantum reduction maps). lows that the observables transform between perspectives A B Just like coordinates on a manifold, a perspective need not be under conjugation with the TFC map ΛS → globally valid (due to the Gribov problem) [25, 26, 39, 40]. ˆphys A B ˆphys A B † OAS B(τA,τB)=ΛS → OBS A ΛS → (65) | | τ phys 1 S A ˆ  = (τB ) S OBS A S− (τB) and IAS denotes the identity on A S and similarly R ◦E | ◦ R H ⊗ H   phys for IBS . In the relational Heisenberg picture the state ˆ ˆ ˆ = τB δ(CH ) τA τA OBS A δ(CH ) τB , transforms as h | | ih |⊗ | | i   where we have made use of Eqs. (43) and (63). It is A B phys phys phys phys phys ΛH→ : , thus seen that the observable Oˆ transforms from A’s HB ⊗ HS → HA ⊗ HS BS A A B perspective to B’s perspective by| first acting on it with ψBS A ψAS B =ΛH→ ψBS A , | | i 7→ | | i | | i the operator τ τ associated with clock A reading the | Aih A| time τ , yielding τ τ Oˆphys . This operator is then where ΛA B is the operator A | Aih A|⊗ BS A H→ projected onto the physical Hilbert| space via the operator δ(CˆH ) and conditioned on clock B reading the time τB. A B 1 ΛH→ := H,B H−,A This procedure yields the transformed observable on AS R ◦ R as seen from the perspective of B. = τ U † (τ ) δ(Cˆ ) ( τ U (τ )), (64) h B |⊗ AS B H | Ai⊗ BS A Crucially, notice that in line with the perspective-   neutral approach [25, 26, 39, 40] alluded to above, these observable transformations from one ‘clock-perspective’ i (HˆA+HˆS ) τB where UAS† (τB ) = e and similarly for to another always proceed via the algebra of Dirac ob- UBS(τA). We emphasize that in the sequel we will al- servables on . Indeed, adapting Theorem 3 to the Hphys ways assume the TFC operators in Eqs. (63) and (64) τA ˆphys phys phys present case implies that the encoding S (OBS A) inside to act on , so that we may use, e.g., the E | HA ⊗ HS Eq. (65) corresponds to the relational Dirac observable simpler form Eq. (53) for H. ˆ R FOBS|A,TA (τA) on phys. This is the observable ana- log of the ‘quantumH coordinate changes’ described be- fore, which map reduced states from one perspective al- ways via phys to reduced states of another perspective B. Observable transformations (cf. Fig. 5H). In order to understand the meaning of the state and A change of temporal reference frame also induces a observable transformations, it is important to note that transformation of observables. Under a change of tem- we are always describing the same physics (encoded in poral reference frame, the expectation value of the un- the clock-neutral phys), just from different (clock) per- H transformed observable with the untransformed state is spectives. In particular, just as we always describe the equal to the expectation value of the transformed observ- same clock-neutral physical state ψphys in reduced form | i able with the transformed state. We examine transfor- relative to different clocks, we also always describe the mations of observables in the relational Schr¨odinger and same Dirac observable from phys (in Eq. (65) this is ˆ H Heisenberg pictures in the following two subsections. FOBS|A,TA (τA)) in the respective reduced theories. It is 28 precisely these clock-neutral structures of states and ob- with some external influence (e.g. classical control of a servables on phys that provide the consistent link be- magnetic field). Here the situation is different. Theo- tween the differentH reduced descriptions relative to dif- ˆ rem 7 shows that if the observable FOBS|A,TA (τA) being ferent choices of clock. transformed contains degrees of freedom of the new clock It is seen fromEq. (65) that the transformed observable B that evolve non-trivially with respect to the old clock may depend on both τA and τB , even though the untrans- A, this observable will have an explicit τB dependence formed observable was independent of both τA and τB. even when described in the Schr¨odinger picture relative The explicit dependence of the transformed observable to the new clock B. on the evolution parameter τA from the old perspective This is, in fact, an indirect instance of self-reference should not surprise because, as just observed, we are now ˆphys by clock B: the transformed observable OAS B(τA, τB ) describing the relational Dirac observable Fˆ (τ ) | OBS|A,TA A is the description of the relational Dirac observ- from the perspective of clock B, and this observable in- able Fˆ (τ ) from the perspective of B. But cludes a description of how system degrees of freedom OBS|A,TA A Fˆ (τ ) describes how B (and S) degrees of free- evolve relative to clock A. Loosely speaking, this is analo- OBS|A,TA A ˆphys gous to how in relativity an observer B may describe from dom evolve relative to A. Hence, OAS B(τA, τB ) indi- their reference frame how a system S evolves relative to rectly describes how B degrees of freedom| evolve rel- the clock of some other observer A. The τB dependence, ative to B. This becomes particularly evident when, ˆ ˆ ˆ by contrast, is more subtle. The following theorem states e.g., OBS A = TB IS and so FOBS A,TA (τA) | ⊗ | ≡ the necessary and sufficient conditions under which the ˆ ˆphys FTB IS ,TA (τA). In that case, O (τA, τB) encodes how transformed observable is independent of τ . ⊗ AS B B the first moment of the clock B |evolves relative to the Theorem 7. Consider an operator on BS from the per- clock A and describes these relations from the perspective spective of A described by Oˆphys ( phys phys). of B. It should be no surprise that this observable must BS A ∈ L HB ⊗ HS From the perspective of B, this| operator is independent depend on τB even in the Schr¨odinger picture relative to ˆphys ˆphys phys B, despite the evolution generator being τB independent. of τB , so that OAS B(τA, τB )= OAS B(τA) ( A | | ∈ L H ⊗ Theorem 7 clarifies that such an indirect clock self- phys) if and only if HS reference will in general manifest itself in the shape of observables in the relational Schr¨odinger picture of this ˆphys ˆphys ˆphys OBS A = OB A fS A , clock, which explicitly depend on its own evolution pa- | | i ⊗ | i i X     rameter. We note that this observation is only possible ˆphys ˆphys thanks to the clock-neutral picture on phys, which en- where (fS A )i is an operator on S and (OB A )i is a con- H | | codes many clock choices at once. ˆphys ˆ phys stant of motion, [(OB A )i, HB]=0. Furthermore, in this From A’s perspective, if it is the case that Oˆ = | BS A case phys ˆphys | IB A fS A , it follows immediately from Eq. (66) that | ⊗ | ˆphys ˆphys the transformed observable on AS from the perspective OAS B(τA) = ΠσABS AS τA τA fS A | G | ih |⊗ | i of B is " i X     ˆphys ˆphys ˆphys ˆ OAS B(τA)=ΠσABS AS τA τA fS A ΠσABS . tB OB A δ(CH ) tB ΠσABS , (66) | G | ih |⊗ | × h | | i | i #   (67)   where Π is a projection onto the subspace of σABS HA ⊗ This can also be seen to follow from the shape of S spanned by energy eigenstates whose energy lies in ˆ ˆ ˆ FO ,TA (τA) = FIB fS ,TA (τA) = FfS ,TA (τA) IB on H BS|A ⊗ σABS := Spec(HˆA + HˆS) Spec( HˆB), tB is an arbi- ⊗ phys, by adapting Eq. (27) to the constraint Eq. (60). trary clock state of B, and∩ −is the G|-twirli over the H GAS The G-twirl appearing in Eq. (67) has the effect of re- group generated by Hˆ + Hˆ . phys A S moving any the operator τ τ fˆ may | Aih A|⊗ S A Proof. The proof is given in Appendix C. | have across the eigenspaces of HˆA + HˆS; that is, the Adapting Eq. (37) to the present case, it follows that transformed observable is superselected with respect to the charge sectors induced by Hˆ + Hˆ [37, 61]. Hˆ + Hˆ is the Hamiltonian which generates the time A S A S Equation (67) implies the following corollary, which evolution in the Schr¨odinger picture relative to clock B. provides the necessary and sufficient conditions for a sys- This Hamiltonian is τB independent. Observables in a ˆphys tem observable fS A to be invariant under a change of Schr¨odinger picture with a time independent Hamilto- | nian are usually time independent themselves. Theo- temporal frame. rem 7 shows that this is the case in the new perspective Corollary 4. Consider an observable seen from the per- when the observable being transformed does not encode spective of A that acts nontrivially only on S, any evolving degrees of freedom of the new clock B. When Schr¨odinger picture observables are nevertheless ˆphys phys ˆphys OBS A = IB A fS A . explicitly dependent on time, one often associates this | | ⊗ | 29

Under a temporal frame change to the perspective of B, if and only if such an observable transforms to ˆphys ˆphys ˆphys phys phys phys OBS A(τA)= OB A fS A (τA) , ˆ ˆ | | i ⊗ | i OAS B = IA B fS B , i | | ⊗ | X     where fˆphys = fˆphys if and only if fˆphys is a constant of ˆphys ˆ ˆphys S B S A S A and OB A is a constant of motion, [HB, OB A ]=0. | ˆphys ˆ| | | | motion, [fS A , HS]=0. | Proof. The proof is given in Appendix C. Proof. The proof is given in Appendix C. The interpretation of these observable transformations Hence, whenever an observable is not a constant of is of course analogous to those between different rela- motion, it will appear differently relative to the different tional Schr¨odinger pictures. In particular, when there is clocks. an explicit τB dependence in the relational Heisenberg equations of motion relative to clock B, this can be in- terpreted as a manifestation of a clock B self-reference. 2. Observable transformations in the relational Heisenberg picture C. Temporal localization is frame dependent Similarly, in the relational Heisenberg picture we de- mand the following criterion between untransformed and We now consider two explicit examples of temporal transformed states and observables frame changes in the relational Schr¨odinger picture. In ˆphys the first example, we change from the perspective of A to ψBS A OBS A(τA) ψBS A h | | | | | i the perspective of B, when the state of B seen by A at H ˆ clock A time τA has support localized around the clock = ψAS B OAS B(τA, τB) ψAS B , h | | | | | i state τA . In this case, we find that the evolution of where for distinction we write the transformed observ- AS seen| i by B is temporally local in the sense that the ˆH able as OAS B(τA, τB ) as this will in general not coin- evolution of AS is described by a single time evolution | ˆphys operator UAS(τB ) generated by HˆA + HˆS . In the second cide with the transformed observable OAS B(τA, τB) of the relational Schr¨odinger picture above. Again,| in the example, we change to the perspective of B, when B is relational Heisenberg picture observables transform be- seen by A to be in a superposition of two states localized around different clock states τ ∆ . In this case, we tween perspectives under conjugation with the TFC map | A ± i ΛA B find that the evolution of AS is temporally nonlocal, by H→ which we mean that the evolution of AS is described by a H ˆ A B ˆphys A B † superposition of the time evolution operators UAS(τB OAS B(τA, τB)=ΛH→ OBS A(τA) ΛH→ ± | | ∆). These examples are depicted in Fig. 6, and illustrate phys 1 H H ˆ  that temporal localization is frame dependent. = ,B OBS A(τA) H−,B R ◦E | ◦ R Consider again two clocks A and B and a system S ˆphys  = UAS† (τB )OAS B(τA, τB)UAS(τB ). (68) described by a physical state satisfying Eq. (60). For | simplicity we assume that the associated clock states are ˆphys In the last line, OAS B(τA, τB) is the transformed ob- orthogonal. Suppose that in the relational Schr¨odinger servable from the relational| Schr¨odinger picture. Again, picture the state of BS from the perspective of A is a the transformation between different reduced descrip- product of pure states of B and S tions of observables proceeds via Dirac observables on the ψBS A(τA) = ψB A(τA) ψS A(τA) . (69) clock-neutral Hilbert space phys. The above equation | | i | | i | | i and Theorem 7 imply the followingH corollary that spec- As constructed, Eq. (69) is temporally local in the ifies the necessary and sufficient conditions under which ˆ ˆ ˆH evolution generated by HB + HS as it can be writ- OAS B(τA, τB) evolves in clock B time τB according to | ten in the form UBS(τA) ψA ψS , where UBS(τA) := the Heisenberg equation of motion with no explicit τB | i | i i(HˆB +HˆS )τA A B dependence. e− . Application of the TFC map ΛS → yields the state of AS from the perspective of B (see Ap- Corollary 5. Consider an operator on BS from the per- pendix D) ˆphys phys phys spective of A described by OBS A(τA) ( B S ). | ∈ L H ⊗H A B Under a temporal frame change to the perspective of B, ψAS B(τB) =ΛS → ψBS A(τA) ˆH | | i | | i this operator transforms to OAS B(τA, τB) that satisfies dt | = ψB A(τB t) t A ψS(t) , (70) the Heisenberg equation of motion in clock B time τB R 2π | − | i | i Z without an explicitly τB dependent term, where ψB A(τB t) := τB ψB A(t) is the | − h | | i d ˆH ˆ ˆ ˆH of clock B in the clock state basis. In the description rel- OAS B(τA, τB )= i HA + HS, OAS B(τA, τB) , dτB | | ative to clock A, the wave function ψB A rather depends h i | 30

A B Temporal frame change ΛS →

⊗ ⊗

dt φB τA ψS τA φB t τB t |ψS A(t) | ( ) | ( )  2π ( − ) |  | BS from the perspective of A AS from the perspective of B

(a) Temporally local evolution seen by A and B.

A B Temporal frame change ΛS →

+ ⊗ ⊗ + ⊗

dt φB (t τB ) |φB (τA − ∆) |φB (τA + ∆) |ψS (τA)  − |t − ∆A |ψS (t − ∆) |t + ∆A |ψS (t + ∆) 2π √2N BS from the perspective of A AS from the perspective of B

(b) Temporally local evolution seen by A and temporally nonlocal evolution seen by B.

FIG. 6. Clock A and B are depicted in blue and red respectively, and the system S in green. (a) The evolution of the state |ψBS|A(τA)i of BS seen by A is temporally local (left). Since clock B is localized in its clock state basis as seen by A, transforming to the perspective of B yields a temporally local evolution of the state |ψAS|B (τB)i of AS (right) described by Eq. (71). (b) The evolution of the state |ψBS|A(τA)i of BS seen by A is again temporally local (left). Since clock B is in a superposition of two states localized in its clock state basis as seen by A, transforming to the perspective of B yields a temporally nonlocal evolution of the state |ψAS|B(τB )i of AS (right) described by Eq. (72). While BS appears unentangled from the perspective of A, from B’s perspective AS appears as an entangled state comprised of a superposition of two branches localized at different times, t ± ∆. This is the temporal analog of the observation in that spatial entanglement depends on the quantum frame perspective [38, 39] and complements the recent discussion in [65].

on τA, but the TFC map replaces this by a dependence on Next, suppose instead that B is seen by A to be in τB (see Appendix D). We note that from the perspective a superposition of two states localized around different of A, BS is in a product state, while from the perspective clock states of B, AS is entangled. 1 First, suppose that from the perspective of A the state ψB A(τB)= φB (τB ∆) + φB (τB + ∆) , of B is a localized Gaussian wave packet of width σ, | √2N −   2 2 2 2 (τB t) /2σ ∆ /σ e− − where N :=1+ e− . Then the state of AS from the ψB A(τB t)= =: φB(τB t). | − π1/4√σ − perspective of B is

The parameter σ quantifies the degree of localization of 1 ψAS B(τB) = UAS(τB ∆) + UAS(τB + ∆) | B around the clock state τB . In this case, the state of | i √2N − AS seen by B is | i dt  φB (t) t A ψS(t) , (72) × R 2π | i | i dt Z ψAS B(τB ) = UAS(τB ) φB (t) t A ψS(t) , (71) | | i R 2π | i | i From Eq. (72) we conclude that the state of AS as seen by Z B is in a superposition of wave packets localized around i(HˆA+HˆS )τB where U (τ ) := e− . We conclude that AS τ ψ (τ ) translated forward and backward in clock AS B | Bi | S B i is seen by B to be localized around τB ψS A(τB ) , since B time τB by ∆. We thus conclude that the evolution | i | | i φB (t) is peaked around t = 0, and that the evolution of of AS is temporally nonlocal because it corresponds to AS is temporally local because its evolution is written in a superposition of time evolutions separated in clock B terms of a single time evolution operator UAS(τB). This time by an amount 2∆, see Fig. 6(b). This is an example situation is shown in Fig. 6(a). of a superposition of time evolutions [125]. 31

The particular form of entanglement in the state of AS We now show that the clock changes of [65] are in- in Eq. (72) implies that the reduced state of S is mixed cluded in the class of temporal frame changes devel- relative to B oped above, which pass through the clock-neutral phys- 1 ical Hilbert space. For example, adapted to our nota- ρS B(t) ψS(τB ∆) ψS (τB ∆) | ≈ 2 | − ih − | tion and normalization, the quantum clock transforma-  tion Eq. (25) of [65] reads + ψ (τ + ∆) ψ (τ + ∆) , | S B ih S B | A B dt  S → = ( τB =0 IAS ) t A UBS(tA) , where we have assumed σ 1 (and that ψS(t) is not S h |⊗ R 2π | i ⊗ 2∆ periodic); note that here,≪ in contrast| to thei rest of Z ≈ A B the article, does not denote a weak equality but rather where S → transforms states from those with respect S approximate equality. The above can be to A to those with respect to B, and the clock states explained as S being temporally localized at either τ ∆ are assumed to be orthogonal for different values of t. B − or τB + ∆, but from the perspective of B it is indefinite Comparing with Eqs. (40), (41), (53) and (C10), it is as to which of these two possibilities is realized. Thus, B easy to see that sees the temporal locality of S as indefinite. A B 1 S → = S(τ = 0) S− (τ = 0) The lesson of these examples is that temporal locality S R B ◦ R A 1 is frame dependent. From the perspective of A the evolu- = H,B (τB = 0) H−,A , (73) tion of BS was temporally local. From the perspective of R ◦ R B, which depends on the state of B as seen by A, the evo- which is an example of the TFC maps (in the case of lution of AS can either be temporally local or nonlocal. ideal clocks) as defined in Eq. (62), i.e. This complements the discussion in [65] where likewise A B A B A B S → ΛS → (τ =0, τ = 0) ΛH→ , an interesting temporal non-locality was reported that S ≡ A B ≡ depends on the clock perspective. where for clarity we have included the times between A B which ΛS → translates Schr¨odinger-picture states. D. Connection with past work on quantum For completeness, we note that we can decompose our temporal frame changes TFC map in Eq. (63) as follows: A B ΛS → = ( τB IAS) δ(CˆH ) ( τA IBS) The first systematic method for changing quantum h |⊗ | i⊗ 1 clocks [27–29] was developed at a semiclassical level = UA(τA) IBS dt t A t B US(t) ⊗ 2π R | i ⊗ h− | ⊗ using so-called effective techniques for constraint sys-  Z  tems. This approach already featured what we may call UB† (τB ) IAS. a perspective-neutral structure (a constraint surface in ⊗ a quantum phase space) that contained all clock per- The term in the square brackets can be further decom- spectives at once. The perspective-neutral approach to posed as quantum frame changes was then generalized to a full ∞ n 1 i (n) ˆ n quantum method for switching clock perspectives for the dt t A t B US(t)= A B HS , 2π R | i ⊗ h− | ⊗ n! P → ⊗ parametrized particle [25] and for a model which can be Z n=0 interpreted either as a quantum cosmological model or X where we define as a relativistic particle [26]. These two examples were discussed in the relational Heisenberg picture (which in (n) 1 n A B := dt ( t) t A t B those models is equivalent to reduced phase space quan- P → 2π R − | i ⊗ h− | Z tization) and illustrate specific realizations of the TFC th A B as the n -moment -swap operator between clocks map ΛH→ for both states and relational observables. th In these two models, the various clock operators are self- A and B. This generalizes the (‘0 -moment’) parity- swap operator, which was originally introduced in [38] for adjoint on kin and thus have orthogonal clock states. However, inH both models one also has to deal with de- spatial quantum reference frames, appeared in [25, 26, 65] generate clock Hamiltonians. for quantum clocks also, and which applies to self-adjoint Recently, temporal frame changes for the Page- reference frame degrees of freedom, to covariant clock Wootters formalism were derived independently from the POVMs. Indeed, in the special case that the clock states present work in [65], offering an example of the TFC map are orthogonal, in which case they are eigenstates of a A B self-adjoint first moment operator TˆB, we can simplify ΛS → , although observable transformations were not ex- plored. The clocks considered in [65] are of the ideal, non- the above expression to ˆ ˆ degenerate case Spec(HC ) = R when T is a self-adjoint 1 (0) i TˆB HˆS dt t A t B US(t)= A B e ⊗ , operator with orthogonal clock eigenstates on kin. The 2π R | i ⊗ h− | ⊗ P → authors of [65] explore how an indefinite causalH order of Z (0) quantum events may arise through gravitationally inter- where A B is the standard parity-swap operator; cf. acting quantum clocks. Eq. (26)P of→ [65], see also [25, 26]. 32

ˆphys ˆphys Thanks to the equivalence established through the fS (τ) and fS , so that the encoding maps Eq. (56) trinity, the present article thus unifies and extends both and (43), previous methods to a much larger class of models in phys 1 phys which the clock need not be quantized as a self-adjoint H fˆ (τ) = H− fˆ (τ) H E S R S R operator, but is rather encoded in the more general no- τ ˆphys 1 ˆphys tion of a covariant clock POVM. In this manner, we are S f = S− (τ) f S(τ) , (74) E S R S R able to go beyond the assumption of ideal clocks, includ-   ing those which may classically feature pathological be- constitute the quantum analog of the ‘gauge-invariant haviour as illustrated in the example of the exponential extension of gauge-fixed quantities’ procedure. Indeed, potential (cf. Sec. IIIA). In a companion article [83] we as established in Theorems 3 and 5, the encoded ob- servables coincide weakly, i.e. on , with the power extend the ability of the TFC maps in [25, 26] to deal Hphys with the subtleties arising in the presence of the clock series quantization Eq. (27) of the relational Dirac ob- energy degeneracies in relativistic systems to covariant servables FfS ,T (τ) of Eq. (5). In line with all this, we phys clock POVMs. have also shown in Theorem 2 that the map fˆ S 7→ Fˆ phys (τ) is weakly an algebra homomorphism with re- fS ,T spect to addition, multiplication and the commutator VIII. IMPLICATIONS OF THE TRINITY (see also Sec. VI C). This is precisely the quantum ana- log of the corresponding classical weak algebra homomor- A. Quantum analog of gauge-invariant extension of phism f Ff,T (τ) established in [14], which relied on gauge-fixed quantities the notion7→ of ‘gauge-invariant extension of gauge-fixed quantities.’ As explained in Sec. II A, the classical relational Dirac Recall that the power-series quantization of the classi- ˆ observables Ff,T (τ) are so-called gauge-invariant exten- cal relational Dirac observables yields FfS ,T (τ) as the sions of gauge-fixed quantities [14–17, 79]. Ff,T (τ) cor- G-twirl ( τ τ fˆ ), i.e. an integral over the one- G | i h | ⊗ S responds to the value that the function f takes on the parameter group generated by the constraint CˆH . Hence, intersection of the gauge fixing surface T = τ with the on phys, we may alternatively think of the relational constraint surface (cf. Fig. 2). In particular, this in- DiracH observables as G-twirls of the reduced observables C tersection of T = τ with corresponds to a gauge-fixed together with the ‘projector’ τ τ onto the clock time τ. C reduced phase space (cf. Sec. IVB1). Conversely, Eq. (74) provides| i a h new| way to understand So far, the quantum analog of the notion of ‘gauge- the G-twirl: it is weakly equal to a conjugation with sym- invariant extension of gauge-fixed quantities’ has been metry reduction maps. This seems to have been unknown lacking in the literature. One reason is that, within before. These observation thereby offer a novel system- the canonical Dirac quantization procedure, there is atic construction procedure for quantum relational Dirac no gauge-fixing:29 the physical Hilbert space — i.e. the observables. quantum constraint surface — is already gauge-invariant While completing this work, we became aware of a re- in contrast to the classical constraint surface which con- cent complementary article [30] which also carefully de- tains all the gauge orbits. Another is that the quantum velops a quantum version of ‘gauge-invariant extension analog of ‘gauge-fixed’ reduced phase space seems to have of gauge-fixed quantities’. In contrast to us, this work been missing. begins with integral representations of relational observ- For the class of systems defined by the constraint ables [13, 19], rather than the power-series expansions Eq. (23), we have clarified in this article precisely the [14–17], which we have employed. The approach in [30] quantum versions of both ‘gauge-invariant extensions of can be viewed as a canonical operator analog of Faddeev- gauge-fixed quantities’ and ‘gauge-fixed reduced phase Popov gauge-fixing [126]. Interestingly, this construc- spaces’. The canonical quantum analog of ‘phase space tion also yields what we call the G-twirl (compare with reduction through gauge-fixing’ is given by the reduction Eqs. (36) and (46) in [30]) and, in fact, a systematic con- maps S(τ) and H, especially the latter, as it gives struction procedure for relational Dirac observables for a R R rise to the relational Heisenberg picture in analogy to the wider class of systems with a Hamiltonian constraint (the classical relational Hamiltonian equations of motion on restriction Eq. (23) is not assumed, while a monotonic the reduced phase spaces (cf. Sec. IVB1). The quantum clock is implicitly assumed). However, the advantage of analog of the reduced phase space is the physical sys- our procedure for the class of systems considered is that phys tem Hilbert space S . Accordingly, the quantum ana- we do not rely on a (kinematical) self-adjoint quantiza- log of a ‘gauge-fixed’H quantity are the system observables tion of classical gauge-fixing conditions unlike [30]. In our case the classical gauge fixing conditions are T = τ and, as described in Sec. IIIB, we instead quantize T more generally as a covariant clock POVM. This enables us to 29 Clearly, at the path integral formulation there is the well-known consider a much wider class of clocks. Furthermore, the Faddeev-Popov gauge fixing [126] and its generalization, the relation with quantum symmetry reduction and the alge- Batalin-Vilkovisky formalism [127]. bra homomorphism were not discussed in [30], which we 33

believe elucidates clearly the quantum analog of ‘gauge- gauge-fixings and thus different reduced theories, inter- invariant extensions of gauge-fixed quantities.’ It would preted as the descriptions of the same dynamics, but be very interesting to combine the techniques developed relative to different choices of temporal reference frame. in [30] with the results established in this manuscript. In The same is true in the quantum theory: different clock particular, the shape of Eq. (27) suggests that our con- choices yield different families of relational observables struction of quantum Dirac observables in terms of the G- and different reduction maps S(τ) and H, and hence twirl holds for general Hamiltonian constraints including different relational Schr¨odingerR and HeisenbergR pictures interactions, as also observed in [30] (see also Eq. (3.1.10) with different conditional inner products. However, these in [3]). Note, however, that for non-integrable systems different reduced quantum theories, i.e. descriptions of this G-twirl expression will be formal as in that case the quantum dynamics relative to different choices of a quantum representation problem of Dirac observables temporal reference frame, are all equivalent by being arises [128, 129] (see also [29]). different quantum gauge-fixings of the manifestly gauge- invariant clock-neutral picture on .30 Hphys B. Conditional inner product as quantum gauge-fixed physical inner product C. Resolving Kuchaˇr’s three criticisms

The quantum reduction maps S(τ) and H and R R Kuchaˇrraised a serious challenge to the Page-Wootters their inverses thus give rise to the quantum analogs of formalism in his seminal review on the problem of both gauge-invariantly extending gauge-fixed quantities time [10]. He presented three distinct criticisms to the and the converse, gauge-fixing gauge-invariant quanti- proposal, which we paraphrase here: ties for both observables and states. The relational phys Schr¨odinger and Heisenberg pictures on are the 1. Inappropriate for Klein-Gordon systems: When HS ‘quantum gauge-fixed’ descriptions of the clock-neutral applied to a relativistic particle in Minkowski space, picture on the manifestly gauge-invariant phys. the conditional probability for the position of the H In line with this, the physical inner product in Eq. (26) particle as a function of Minkowski time differs is clock-neutral: its definition does not depend on a tem- from the accepted Klein-Gordon probability den- poral reference frame and is compatible with a multitude sity for the localization of a relativistic particle. of different clock choices. Accordingly, we can regard it as a description of the theory’s inner product prior to 2. Violation of the constraints: The Page-Wootters having chosen a temporal frame. By contrast, it is now formalism postulates the conditional probability in clear that the conditional/Page-Wootters inner product Eq. (35), which is motivated by applying the Born in Eq. (38), originally introduced in [50, 51], is a quan- rule to a measurement corresponding to the effect tum gauge-fixed version of the physical inner product, operator e (τ) e (f). Such an effect operator T ⊗ fS thanks to Corollary 1. The definition of the conditional does not commute with the constraint operator CˆH , inner product requires a specific clock choice and a spe- and thus the measurement throws ψphys out of the cific reading of that clock. Classically, any fixed clock physical Hilbert space. The Page-Wootters| i formal- reading corresponds to a choice of gauge. Consistent with ism would thus be based on a postulate that vio- the interpretation that the conditional inner product is a lates the constraint. gauge-fixed version of the gauge-invariant physical inner product, one finds that it is actually independent of the 3. Wrong propagators: When applied to answering clock reading because the reduced dynamics is unitary. the fundamental dynamical question — ‘If one finds As such, we can view the conditional inner product as the system at position q at time τ, what is the prob- the description of the inner product relative to a choice ability of finding it at position q′ at time τ ′?’ — the of temporal reference frame. conditional probability in Eq. (35), interpreted in Classically different clock choices lead to different the two-time case as

ψphys eT (τ) eT (τ ′) eT (τ) eqS (q) eqS (q′) eqS (q) ψphys kin Prob(q′ when τ ′ q when τ)= h | · · ⊗ · · | i , (75) | ψ e (τ) e (q) ψ h phys| T ⊗ S | physikin

where eqS (q) is an improper projector associated 30 Global equivalence requires that the different choices of temporal with the particle located at q, yields the wrong an- reference frame correspond each to monotonic clocks. For a non- monotonic clock the equivalence will not be global on the physical Hilbert space (see [27–29, 40] for a related discussion). 34

swer, and prohibits time to flow. This amounts to of freedom. a reductio ad absurdum. As Kuchaˇremphasized [10], the problem has to do with

the fact that the (improper) projector eT (τ ′) eqS (q′) In a companion article [83], in which we treat rela- inside Eq. (75) acts on a state that no longer resides⊗ in tivistic settings, we address the first criticism by again phys. For this two-time conditioning, we have not es- choosing a clock POVM, in that case chosen covariant tablishedH gauge-invariance, since Theorems 3 and 4 ap- with respect to quadratic clock Hamiltonians, and ap- ply only to the one-time conditioning scenario. In fact, propriately adapting the Page-Wootters inner product, Eq. (75) is simply the wrong way to express a conditional Eq. (38), introduced in [50]. We show that conditioning probability from the point of view of Dirac quantization; on the covariant clock POVM instead of the Minkowski it evaluates kinematical operators in kinematical states. time operator results in a Newton-Wigner type localiza- It is impossible to express Eq. (75) purely in terms of tion probability commonly used in relativistic quantum gauge-invariant objects. However, the trinity establishes mechanics. By extending the trinity to relativistic sys- an equivalence between the gauge-invariant quantum the- tems, this also connects with the treatment of the Klein- ory on and the relational Schr¨odinger picture on Gordon system in [26, 110]. Hphys phys, suggesting that there must be an alternative. The second criticism above has been resolved in the S H Indeed, we now propose a new two-time conditional present manuscript. Theorems 3 and 4 show that, while probability at the level of phys, inspired by the usual the individual kinematical operators eT (τ) efS (f) in- H deed are not Dirac observables on ⊗, the entire expression for conditional probabilities [131]. Through phys the trinity, this proposed conditional probability induces conditional probability in Eq. (35) is Hmanifestly gauge- invariant and coincides with the expectation value of the an expression for the two-time conditional probability in terms of the Page-Wootters conditional state, from which corresponding Dirac observables (through the encoding map) in the physical inner product. Hence, the condi- we recover the correct propagator. To this end, recall ˆphys ˆ Theorem 2, which establishes that fS Ff phys,T (τ) is tional probability in Eq. (35) does not actually violate 7→ S any constraints. It is just the reduced form (having un- an algebra homomorphism. This permits us to generalize dergone the quantum analog of gauge-fixing) of a gauge- Kuchaˇr’s conditional probability question above to: “If invariant expression. one finds the system in the state corresponding to the The third criticism is also completely resolved by the observable Aˆ taking the value a at clock time τ, what trinity. This criticism has previously been discussed is the probability of finding it in the state correspond- ˆ and proposals for its resolution were put forward in ing to observable B taking the value b at clock time τ ′?” [21, 49, 64, 130] (see also the recent exposition of the In particular, if ΠA=a is the (possibly improper) projec- phys ˆ different proposals in the context of the Wigner’s friend tor on S corresponding to the system observable A scenario [71]). However, the proposed resolution in [21] takingH the value a, then the relational Dirac observable ˆ relies on approximations in the limit of ideal clocks, while FΠA=a,T (τ) too will act as a (possibly improper) projec- 31 the proposal in [49] hinges on auxiliary ancilla systems. tor on phys, however, this time associating the system The trinity established in this paper offers a different observableH reading a with the clock reading τ. This sug- route and resolves the two-time conditioning problem gests the following two-time conditional probability on arising from Eq. (75) exactly, and without extra degrees Hphys

ˆ ˆ ˆ ψphys FΠA=a,T (τ) FΠB=b,T (τ ′) FΠA=a,T (τ) ψphys phys Prob(B = b when τ ′ A = a when τ) := h | · · | i , (76) | ψ Fˆ (τ) ψ h phys| ΠA=a,T | physiphys where we note the evaluation of the expectation values is done using the physical inner product. In Appendix E2 we show that this probability can be rewritten as ˆ ˆ ψphys (eT (τ) ΠA=a)δ(CH )(eT (τ ′) ΠB=b)δ(CH )(eT (τ) ΠA=a) ψphys kin Prob(B = b when τ ′ A = a when τ)= h | ⊗ ⊗ ⊗ | i . | ψ (e (τ) Π ) ψ h phys| T ⊗ A=a | physikin (77) Interestingly this is the generalization of Dolby’s two-time conditional probability to the case of constraints which have zero in the continuous part of their spectrum [64].32 In Appendix E2, we further demonstrate that this expression

31 This criticism was also discussed in [95], however the authors ob- 32 In the special case of ideal clocks, this expression was recently tained incorrect propagators. This is a consequence of evaluating studied in the context of the Wigner friend scenario [71]. invariant observables on kinematical states. 35 simplifies to

ψS (τ) ΠA=a US† (τ ′ τ) ΠB=b US(τ ′ τ) ΠA=a ψS(τ) Prob(B = b when τ ′ A = a when τ)= h | − − | i . (78) | ψ (τ) Π ψ (τ) h S | A=a | S i

This is the correct propagator associated with transition- clare a choice of relational Dirac observable as a gauge- ing from the system state corresponding to the observable invariant clock and then to ask how other relational ob- Aˆ reading a at Schr¨odinger time τ to the system state cor- servables behave when the gauge-invariant clock has a responding to the observable Bˆ reading b at Schr¨odinger particular value. In order for this to be possible one has time τ ′. Note that the projectors ΠA=a and ΠB=b need to introduce a second clock system in contrast to our not necessarily be one-dimensional projectors and that setup which thus amounts to a modification of the orig- the two-time conditional probability Eq. (78) holds for inal problem posed by Kuchaˇr. In their construction of the entire class of models considered in this manuscript. conditional probabilities directly on the physical Hilbert Moreover, Eq. (78) holds in the more general case where space, the authors in [21] then integrate out the evo- ΠA=a and ΠB=b are replaced with effect operators corre- lution parameter τ owing to its alleged unobservability. phys sponding to outcomes of a POVM on S . This leads to decoherence effects and modified transition Let us now specialize to the case consideredH by Kuchaˇr, probabilities that only approximate the standard text- where the system S is some particle and Aˆ = Bˆ =q ˆS is book ones for ideal clocks and Gaussian states. phys We take a distinct approach, avoiding such an integra- simply the position operator on S . Equation (78) then becomes H tion because τ corresponds to the reading of a dynamical clock. While its kinematical time observable is not gauge- 2 Prob(q′ when τ ′ q when τ)= q′ US(τ ′ τ) q , invariant, the values it can take in fact are in the follow- | | h | − | i | ing sense: in the classical theory the evolution parameter which is precisely the correct expression for the transition τ corresponding to a kinematical clock function T also probability of a non-relativistic particle. labels the outcomes of gauge-invariant relational observ- It is compelling to observe the conceptual difference ables FT,T ′ (τ ′) asking for the value of T when another 33 between the conditional probabilities in Eq. (76) at the kinematical time observable T ′ reads τ ′. In particular, level of the clock-neutral physical Hilbert space and the it can also be understood as the relational Dirac observ- equivalent expression Eq. (78) at the level of the reduced able FT,T (τ)= τ. Gauge invariance thus does not offer a theory. The latter includes the obvious time evolution reason per se to deem τ unobservable in principle, nor to in-between the conditionings expected in a Schr¨odinger integrate it out.34 Instead, we see that only invoking our picture. These are two conditionings separated by an manifestly gauge-invariant equivalence of the relational ‘external’ time. By contrast, the former does not include observable and Page-Wootters formalism necessarily re- an evolution operator in-between the conditionings, in covers the standard transition probabilities without any line with the often emphasized ‘timelessness’ of what we approximations and additional clock or state choices, nor call the clock-neutral physical Hilbert space phys. In- does a fundamental decoherence mechanism result as a stead, the double conditioning in Eq. (76) canH rather be consequence of using realistic (i.e. bounded Hamiltonian) regarded as the probability for “the event a when τ AND clocks. the event b when τ ′” in the clock-neutral physical state Moreover, unlike [49] our resolution (i) does not neces- ψ . It makes sense to compute such a two-time joint sitate auxiliary ancilla systems, (ii) does not depend on | physi probability from the physical state ψphys as it contains ideal clocks, and (iii) is manifestly gauge-invariant thanks the entire history of the relational dynamics| i of the com- to the relational conditional probability in Eq. (76). The posite system CS at once. Recall that the physical state proposal in [49] extends to an arbitrary number of condi- is a description of physics prior to having chosen a tem- tionings of the physical state, however, crucially requiring poral reference frame. We are thus asking for the proba- the addition of extra ancilla systems for every new condi- bility that a history contains the two events above, each tioning. As such, one has to modify the total composite being a coincidence between two dynamical degrees of system described by the Hamiltonian constraint with ev- freedom. ery new conditioning by adding new degrees of freedom We emphasize that our resolution of Kuchaˇr’s third in order to describe the corresponding measurement pro- criticism is qualitatively different from the proposal in cess. While this is an option for (effective) laboratory [21] and does not rely on approximations and ideal clocks. While the authors of [21] also evaluate relational Dirac observables in the physical inner product in order to de- 33 This statement can also be extended to the quantum theory, fine conditional probabilities, they do so in a very differ- however, is more complicated to phrase due to the observations ent manner, arguing that the evolution parameter τ is in [132]. physically unobservable because it is associated with a 34 However, one may justify integrating out clock readings based kinematical observable. This leads them to instead de- on epistemic grounds when an observer has partial knowledge. 36 situations, it is unsatisfactory for more fundamental de- ner product.35 However, the authors remark that this scriptions in quantum gravity and cosmology where the approach is not fully satisfactory because the normaliza- solution to the Wheeler-DeWitt equation is the quantum tion procedure is completely arbitrary. Indeed, it should state of the entire Universe. In this context, it is not be clear from Sec. IV A that this procedure cannot suc- appropriate to keep adding effective ancilla degrees of ceed when physical states are improper eigenstates of the freedom to the fundamental description. By contrast, constraint: either one violates the constraint or one ob- it is clear that our conditional probabilities Eqs. (76) tains a divergent inner product. In [50, 51] this issue was and (78) can be extended to an arbitrary number of con- avoided by introducing the Page-Wootters inner prod- ditionings without adding new degrees of freedom and uct, as defined in Eq. (38), and demanding the physical one will still always get the correct result, consistent with states are normalized with respect to this inner product standard quantum theory. Given the general validity of as opposed to the kinematical inner product. Eq. (78), we thus regard Eq. (76) as the proper resolution By establishing the trinity, in particular Corollary 1, of Kuchaˇr’s third criticism. we prove that the Page-Wootters inner product is equiv- alent to the standard physical inner product on de- It is interesting to note that Eq. (77) is what Kuchaˇr Hphys had warned against in [10]: fined by group averaging techniques. This completely re- solves the issue of how the physical states should be nor- malized within the Page-Wotters formalism: they should Of course, one can try to modify the con- be normalized with respect to the physical inner prod- ditional probability interpretation, say, by uct, in line with standard methodology used in constraint projecting the state back into the physical quantization [3, 107, 109, 110]. This is further corrobo- [Hilbert] space [...] each time the measure- rated in the companion article [83], where we extend the ment of the projector A,ˆ B,ˆ C,...ˆ brings it out Page-Wootters inner product of [50, 51] to the relativistic of the physical space. I better abstain from case, showing that it again agrees with the physical inner analyzing the shortcomings of such a scheme product obtained through group averaging. before someone seriously proposes it.

IX. CONCLUSION As noted above, Dolby [64] had used the analogous ex- pression to Eq. (77) in the context of discrete spec- trum constraints (which was criticized in [130]), and de- The central result of the manuscript is the establish- spite also considering continuous-spectrum constraints in ment of the trinity of relational quantum dynamics: the his paper, did not actually extend his considerations to dynamics defined by relational Dirac observables, the Eq. (77) in that case. Both Kuchaˇrand Dolby were thus Page-Wootters formalism, and the relational Heisenberg agonizingly close to recovering the correct propagator. picture obtained via symmetry reduction are all man- ifestations of the same relational quantum theory. The Finally, we note that Eq. (76) is an expression involv- trinity has been established for clocks whose Hamiltonian ing only objects from (i.e. Dynamics I of the trinity Hphys has a non-degenerate continuous spectrum, and can be in Sec. V), while Eq. (78) is written purely in terms of ob- extended to clocks with degenerate spectrum, including jects from the reduced theory on phys (i.e. Dynamics II HS a class of relativistic models [83], and periodic (discrete- of the trinity in Sec. V). Both of these expressions can be spectrum) clocks [84]. easily justified within either formulation of the relational To establish the equivalence of the relational dynam- quantum dynamics. By contrast, Eq. (77) is somewhat ics comprising the trinity, we described the kinematical of a hybrid expression, involving structures from both time observable associated with the clock as a covari- Dynamics I and II, and is difficult to fully justify with- ant POVM. This constitutes a more general notion of a out Eqs. (76) and (78). This is presumably the origin of (kinematical) time observable than that of a self-adjoint Kuchaˇr’s criticism above. In line with the trinity, we thus operator canonically conjugate to the clock’s Hamilto- propose that the Page-Wootters formalism should really nian, which is often employed in the context of relational be interpreted in the sense of the reduced Dynamics II quantum dynamics. In Sec. III we described in detail the alone and not in the hybrid way of conditioning physical properties of such covariant POVMs for clocks with con- states with kinematical operators. tinuous and discrete Hamiltonian spectra, and how their spectral properties relate to clock choices in classical re- lational dynamics. This notion of a time observable allowed us to resolve D. There is no normalization ambiguity in the the apparent non-monotonicity issue of self-adjoint ob- Page-Wootters formalism

In further developing the Page-Wootters formalism, 35 In Ref. [49] this was not explicitly stated, but this observation it was suggested in [49] that the physical states ψphys follows from the authors’ choice to normalize their Eq. (23) in should be normalized with respect to the kinematical| in-i the kinematical inner product induced by HC and HS . 37 servables associated with realistic quantum clocks which equivalence principle put forward in [54, 133] can be for- Unruh and Wald described in [77] and used to argue mulated within this program of quantum reference frame against a relational approach to the problem of time. In- changes. deed, thanks to the covariance property the covariant Finally, we discussed three implications of the trinity clock POVM is monotonic even for bounded Hamiltoni- in Sec. VIII. The encoding maps in Eqs. (56) and (43) ans and still admits a consistent probability interpreta- establish the quantum analog of the gauge-invariant ex- tion. The price we pay for giving up the orthodox no- tension of a gauge-fixed quantity [79], a concept central tion of self-adjointness of the time observable is that the to the classical construction of relational Dirac observ- possible clock readings over which the probability distri- ables [14–17] (see also [30]). We then resolved Kuchaˇr’s bution is defined need not necessarily be perfectly distin- criticisms of the Page-Wootters formalism, in particular, guishable. This is, however, common to many quantum by recovering the correct propagator via a conditioning measurements and thus does not constitute a fundamen- of physical states on outcomes of relational Dirac observ- tal obstacle. Hence, using dynamical clocks is a viable ables. This resolution does not require auxiliary ancilla approach to address the problem of time. systems, ideal clocks, or state dependent approximations In Sec. IV A the Dirac quantization procedure was ap- in contrast to previous proposals [21, 49]. Lastly, we plied to the the class of theories introduced in Sec. II, pointed out that the normalization issue with physical which are described by a Hamiltonian constraint associ- states in the Page-Wootters formalism reported in [49] ated with a clock and system that do not interact with does not arise. each other. Using covariant POVMs, we constructed Apart from the extension to relativistic models [83] a new quantization of relational Dirac observables via and periodic clocks [84], the most pressing generalization the G-twirl operation [37], and described their associ- of our work is to explore the validity of the trinity in the ated relational dynamics (Dynamics I). In addition to context of interactions between the chosen clock and the being crucial for establishing the trinity, this construc- evolving system. As we have emphasized in Appendix A, tion allowed us to prove in Theorem 2 the quantum ana- interactions will appear in generic models, particularly log of the classical weak algebra homomorphism between so in quantum gravity. However, this may lead to serious Dirac observables and phase space functions established challenges for relational quantum dynamics, as pointed in [14]. In Sec. V we introduced the Page-Wootters for- out in the context of Dynamics I in [12, 19, 29, 128, 129, malism (Dynamics II) and a relational Heisenberg picture 134]. The issue is essentially that interactions will lead obtained via symmetry reduction (Dynamics III), and to clocks which are non-monotonic, i.e. feature turning demonstrated their equivalence with each other, as well points. This is known as the global problem of time and as with Dynamics I. In Sec. VI we identified the clock- leads to a non-unitarity of the relational dynamics in the system entanglement appearing in the Page-Wootters turning regime of the clock [10, 11, 27–29]. formalism as a kinematical structure, and demonstrated Given the trinity, these challenges must also appear in that the same relational dynamics can be obtained us- the Page-Wootters formalism and the relational Heisen- ing the same conditioning procedure, but without such berg picture of the quantum symmetry reduced theory. kinematical entanglement. As shown in [50], certain interactions will lead to a mod- In establishing the trinity, we constructed invertible re- ified Schr¨odinger equation in the Page-Wootters formal- duction maps between the clock-neutral physical Hilbert ism, which still generates an isometry. In more generic space and the reduced Hilbert space associated with situations the global problem of time must also feature Dynamics II and III. This allowed us to extend the in the Page-Wootters formalism and it will be of inter- perspective-neutral approach to changing quantum ref- est to investigate how it further modifies the Schr¨odinger erence frames [25, 26, 39, 40] to a more general class picture. The results in [27–29], while using semiclassi- of clocks, namely those described by covariant POVMs. cal methods, suggest that the quantum reduction maps These temporal frame changes pass through the clock- from the clock-neutral to the relational Schr¨odinger and neutral physical Hilbert space, and thereby are the quan- Heisenberg pictures will need to separate the branches of tum analog of coordinate changes on a manifold. Such the relational dynamics before and after a clock’s turning a form of frame changes is a prerequisite for exploring point encoded in the physical state. In general, these can a quantum notion of general covariance [25, 26, 38–40]. be anticipated to only produce approximate Schr¨odinger Specifically, we illustrated how both states and observ- equations for each branch that fail on approach to the ables transform in the relational Schr¨odinger and Heisen- turning point. Such clock pathologies may then be nav- berg pictures naturally arising in Dynamics II and III. igated by an intermediate change to another choice of This allowed us to demonstrate a clock-dependent tem- clock and thereby ‘patching up’ the relational history poral nonlocality effect, complementing the recent discus- contained in the physical state with different temporal sion of the frame dependence of temporal localization in reference frames, in analogy to covering a manifold with [65]. The temporal nonlocality discussed above stemmed coordinate charts [27–29]. from transforming to the perspective of a clock in a su- It will also be interesting to explore the connections perposition of reading different times. In this regard, it with a recent algebraic approach to the problem of will be interesting to investigate whether the quantum time [119], which similarly seeks to establish a quantum 38 version of symplectic reduction. In particular, the re- Disciplinary Engagement at Dartmouth for an ICE Fel- lation between our trivialization map and their reduc- lowship, which facilitated a visit to Dartmouth College tion procedure warrants further investigation. In light during the final stages of this work. ARHS acknowledges of the trinity, another line of investigation will be to ex- support from the Natural Sciences and Engineering Re- plore the fundamental decoherence mechanism put for- search Council of Canada and the Dartmouth Society ward in [46, 47, 135], which originates in the observation of Fellows. MPEL acknowledges financial support by that there is a limit to how well one can measure the time the ESQ (Erwin Schr¨odinger Center for Quantum Sci- indicated by a physical clock. ence & Technology) Discovery programme, hosted by the Austrian Academy of Sciences (OAW),¨ as well as from the Austrian Science Fund (FWF) through the ACKNOWLEDGMENTS START project Y879-N27. This project was made pos- sible through the support of a grant from the John Tem- PAH is supported by the Simons Foundation through pleton Foundation. The opinions expressed in this pub- an ‘It-from-Qubit’ Fellowship and the Foundational lication are those of the authors and do not necessarily Questions Institute through Grant number FQXi-RFP- reflect the views of the John Templeton Foundation. 1801A. He further thanks the Institute for Cross-

[1] C. Rovelli, Quantum Gravity (Cambridge University [25] P. A. H¨ohn and A. Vanri- Press, Cambridge, 2004). etvelde, New J. Phys. 22, 123048 (2020), [2] A. Ashtekar, Lectures on Non-Perturbative Canonical arXiv:1810.04153 [gr-qc]. Gravity, Physics and Cosmology, Vol. 6 (World Scien- [26] P. A. H¨ohn, Universe 5, 116 (2019). tific, Singapore, 1991). [27] M. Bojowald, P. A. H¨ohn, and A. Tsobanjan, [3] T. Thiemann, Modern Canonical Quantum General Rel- Class. Quant. Grav. 28, 035006 (2011). ativity (Cambridge University Press, 2008). [28] M. Bojowald, P. A. H¨ohn, and A. Tsobanjan, [4] B. S. DeWitt, Phys. Rev. 160, 1113 (1967). Phys. Rev. D 83, 125023 (2011). [5] C. Rovelli, in Conceptual Problems of Quantum Gravity, [29] P. A. H¨ohn, E. Kubalova, and A. Tsobanjan, edited by A. Ashtekar and J. Stachel (Birkhauser, 1991) Phys. Rev. D 86, 065014 (2012). pp. 126–140. [30] L. Chataignier, Phys. Rev. D 101, 086001 (2020). [6] C. Rovelli, Phys. Rev. D 42, 2638 (1990). [31] K. Giesel and T. Thiemann, [7] C. Rovelli, Phys. Rev. D 43, 442 (1991). Class. Quant. Grav. 27, 175009 (2010). [8] C. Rovelli, Class. Quant. Grav. 8, 297 (1991). [32] M. Domagala, K. Giesel, W. Kaminski, and [9] C. Rovelli, Class. Quant. Grav. 8, 317 (1991). J. Lewandowski, Phys. Rev. D 82, 104038 (2010). [10] K. V. Kuchaˇr, Int. J. Mod. Phys. D 20, 3 (2011). [33] V. Husain and T. Pawlowski, [11] C. J. Isham, in Integrable Systems, Quantum Groups, Phys. Rev. Lett. 108, 141301 (2012). and Quantum Field Theories, edited by L. A. Ibort [34] K. Giesel and A. Vetter, and M. A. Rodr´ıguez (Springer Netherlands, Dordrecht, Class. Quant. Grav. 36, 145002 (2019). 1993) pp. 157–287. [35] Y. Aharonov and L. Susskind, Phys. Rev. 155, 1428 [12] D. Marolf, Class. Quant. Grav. 12, 2469 (1995). (1967). [13] D. Marolf, Class. Quant. Grav. 12, 1199 (1995). [36] Y. Aharonov and T. Kaufherr, Phys. Rev. D 30, 368 [14] B. Dittrich, Gen. Relativ. Gravit. 39, 1891 (2007). (1984). [15] B. Dittrich, Class. Quant. Grav. 23, 6155 (2006). [37] S. D. Bartlett, T. Rudolph, and R. W. Spekkens, [16] B. Dittrich and J. Tambornino, Rev. Mod. Phys. 79, 555 (2007). Class. Quant. Grav. 24, 757 (2007). [38] F. Giacomini, E. Castro-Ruiz, and C.ˇ Brukner, [17] B. Dittrich and J. Tambornino, Nat. Commun. 10, 494 (2019). Class. Quant. Grav. 24, 4543 (2007). [39] A. Vanrietvelde, P. A. H¨ohn, F. Giacomini, and [18] J. Tambornino, SIGMA 8, 017 (2012). E. Castro-Ruiz, Quantum 4, 225 (2020). [19] S. B. Giddings, D. Marolf, and J. B. Hartle, [40] A. Vanrietvelde, P. A. H¨ohn, and F. Giacomini, (2018), Phys. Rev. D 74, 064018 (2006). arXiv:1809.05093 [quant-ph]. [20] A. Ashtekar, T. Pawlowski, and P. Singh, [41] F. Giacomini, E. Castro-Ruiz, and C.ˇ Brukner, Phys. Rev. D 73, 124038 (2006). Phys. Rev. Lett. 123, 090404 (2019). [21] R. Gambini, R. A. Porto, J. Pullin, and S. Torterolo, [42] M. C. Palmer, F. Girelli, and S. D. Bartlett, Phys. Rev. D 79, 041501 (2009). Phys. Rev. A 89, 052121 (2014). [22] J. Pons, D. Salisbury, and K. Sundermeyer, [43] P. A. H¨ohn and M. P. M¨uller, Phys. Rev. D 80, 084015 (2009). New J. Phys. 18, 063026 (2016). [23] W. Kaminski, J. Lewandowski, and T. Pawlowski, [44] D. N. Page and W. K. Wootters, Phys. Rev. D 27, 2885 Class. Quant. Grav. 26, 035012 (2009). (1983). [24] W. Kaminski, J. Lewandowski, and T. Pawlowski, [45] W. K. Wootters, Int. J. Theor. Phys. 23, 701 (1984). Class. Quant. Grav. 26, 245016 (2009). [46] R. Gambini and J. Pullin, Found. Phys. 37, 1074 (2007). 39

[47] R. Gambini, R. Porto, and J. Pullin, [78] A. R. H. Smith, M. Piani, and R. B. Mann, Gen. Relativ. Gravit. 39, 1143 (2007). Phys. Rev. A 94, 012333 (2016). [48] R. M. Angelo, N. Brunner, S. Popescu, [79] M. Henneaux and C. Teitelboim, Quantization of Gauge A. J. Short, and P. Skrzypczyk, Systems (Princeton University Press, Princeton, 1992). J. Phys. A: Math. Theor. 44, 145304 (2011). [80] P. A. H¨ohn, Proceedings, 8th International Workshop: [49] V. Giovannetti, S. Lloyd, and L. Maccone, Phys. Rev. Spacetime - Matter - Quantum Mechanics (DICE2016): D 79, 945933 (2015). Castiglioncello, Tuscany, Italy, September 12-16, 2016, [50] A. R. H. Smith and M. Ahmadi, J. Phys. Conf. Ser. 880, 012014 (2017). Quantum 3, 160 (2019). [81] P. A. M. Dirac, Lectures on Quantum Mechanics (Belfer [51] A. R. H. Smith and M. Ahmadi, Graduate School of Sciencem Yeshiva University, New Nat. Commun. 11, 5360 (2020). York, 1964). [52] R. Gambini, L. P. Garc´ıa-Pintos, and J. Pullin, [82] M. Bojowald, Canonical Gravity and Applications: Cos- Stud. Hist. Philos. Mod. Phys. 42, 256 (2011). mology, Black Holes and Quantum Gravity (Cambridge [53] L. Loveridge, T. Miyadera, and P. Busch, University Press, Cambridge, 2010). Found Phys 48, 135 (2018). [83] P. A. H¨ohn, A. R. H. Smith, and M. P. E. Lock, (2020), [54] L. Hardy, (2019), arXiv:1903.01289 [quant-ph]. arXiv:2007.00580 [gr-qc]. [55] E. C. Ruiz, F. Giacomini, and C.ˇ Brukner, PNAS 114, [84] L. Chataignier, P. A. H¨ohn, and M. P. E. Lock, in E2303 (2017). preparation (2021). [56] S. D. Bartlett, T. Rudolph, R. W. Spekkens, and P. S. [85] A. Ashtekar, R. Tate, and C. Uggla, Turner, New J. Phys. 11, 063013 (2009). Int. J. Mod. Phys. D 2, 15 (1993). [57] M. Ahmadi, D. Jennings, and T. Rudolph, [86] A. Ashtekar and P. Singh, New J. Phys. 15, 013057 (2013). Class. Quant. Grav. 28, 213001 (2011). [58] I. Marvian and R. W. Spekkens, [87] K. Banerjee, G. Calcagni, and M. Mart´ın-Benito, New J. Phys. 15, 033001 (2013). SIGMA 8, 016 (2012). [59] I. Marvian and R. W. Spekkens, [88] M. Bojowald, Canonical Gravity and Applications: Cos- Nat Commun 5, 1 (2014). mology, Black Holes and Quantum Gravity (Cambridge [60] D. Safr´anek,ˇ M. Ahmadi, and I. Fuentes, University Press, Cambridge, 2011). New J. Phys. 17, 033012 (2015). [89] C. E. Misner, K. S. Thorne, and J. A. Wheeler, Gravi- [61] A. R. H. Smith, Phys. Rev. A 99, 052315 (2019). tation (W. H. Freeman and Company, 1973). [62] R. L. Arnowitt, S. Deser, and C. W. Misner, [90] W. Pauli, Handbuch der Physik 5, 1 (1958). Gen. Relativ. Gravit. 40, 1997 (2008). [91] M. A. Nielsen and I. L. Chuang, Quantum Computa- [63] E. Anderson, The Problem of Time, Fundamental The- tion and Quantum Information (Cambridge University ories of Physics, Vol. 190 (Springer International Pub- Press, Cambridge, 2010). lishing, 2017). [92] P. Libermann and C.-M. Marle, “Symplec- [64] C. E. Dolby, (2004), arXiv:gr-qc/0406034 [gr-qc]. tic manifolds and poisson manifolds,” in [65] E. Castro-Ruiz, F. Giacomini, A. Belenchia, and Symplectic Geometry and Analytical Mechanics C.ˇ Brukner, Nat. Commun. 11, 2672 (2020). (Springer Netherlands, Dordrecht, 1987) pp. 89–184. [66] A. Boette and R. Rossignoli, [93] V. I. Arnol’d, Mathematical methods of classical me- Phys. Rev. A 98, 032108 (2018). chanics, Graduate Texts in Mathematics, Vol. 60 [67] N. L. Diaz, J. M. Matera, and R. Rossignoli, (2019), (Springer-Verlag, New York, 1989). arXiv:1910.04004 [quant-ph]. [94] G. Chiribella, G. M. D’Ariano, and M. F. Sacchi, Phys- [68] J. Leon and L. Maccone, Found. Phys. 47, 1597 (2017). ical Review A 73, 062103 (2006). [69] C. Marletto and V. Vedral, [95] L. Loveridge and T. Miyadera, Phys. Rev. D 95, 043510 (2017). Found. Phys. 49, 549 (2019). [70] A. Nikolova, G. Brennen, T. J. Osborne, G. Milburn, [96] M. H. Stone, PNAS 16, 172 (1930). and T. M. Stace, Phys. Rev. A 97, 030101 (2018). [97] S. L. Braunstein, C. M. Caves, and G. J. Milburn, [71] V. Baumann, F. D. Santo, A. R. H. Smith, F. Gi- Ann. Phys. 247, 135 (1996). acomini, E. Castro-Ruiz, and C. Brukner, (2019), [98] L. E. Ballentine, Quantum Mechanics: A Modern De- arXiv:1911.09696 [quant-ph]. velopment (World Scientific, Singapore, 1998). [72] T. Thiemann, Class. Quant. Grav. 23, 1163 (2006). [99] J. C. Garrison and J. Wong, [73] A. S. Holevo, Probabilistic and Statistical Aspects of J. Math. Phys. 11, 2242 (1970). Quantum Theory, Statistics and Probability, Vol. 1 [100] S. Khandelwal, M. P. E. Lock, and M. P. Woods, Quan- (North-Holland, Amsterdam, 1982). tum 4, 309 (2020). [74] P. Busch, M. Grabowski, and P. J. Lahti, [101] P. Bocchieri and A. Loinger, Operational Quantum Physics, Lecture Notes in Phys. Rev. 107, 337 (1957). Physics Monographs, Vol. 31 (Springer-Verlag, Berlin, [102] I. C. Percival, J. Math. Phys. 2, 235 (1961). Heidelberg, 1995). [103] L. S. Schulman, Phys. Rev. A 18, 2379 (1978). [75] P. Busch, P. Lahti, J.-P. Pellonp¨a¨a, and K. Ylinen, [104] P. Lahti and J.-P. Pellonp¨a¨a, Quantum measurement, Vol. 23 (Springer, 2016). J. Math. Phys. 40, 4688 (1999). [76] R. Brunetti, K. Fredenhagen, and M. Hoge, [105] A. Galindo, Lett. Math. Phys. 8, 495 (1984). Found. Phys. 40, 1368 (2010). [106] K. V. Kuchaˇr, Phys. Rev. D 34, 3044 (1986). [77] W. G. Unruh and R. M. Wald, [107] D. Marolf, (2000), arXiv:gr-qc/0011112 [gr-qc]. Phys. Rev. D 40, 2598 (1989). [108] V. Guillemin and S. Sternberg, Symplectic techniques in physics (Cambridge Uni- 40

versity Press, Cambridge, 1990). [124] B. C. Hall, Quantum Theory for Mathematicians [109] D. Marolf, (1995), arXiv:gr-qc/9508015. (Springer, New York, NY, 2013). [110] J. B. Hartle and D. Marolf, [125] Y. Aharonov, J. Anandan, S. Popescu, and L. Vaidman, Phys. Rev. D 56, 6247 (1997). Phys. Rev. Lett. 64, 2965 (1990). [111] A. Ashtekar and G. t. Horowitz, [126] L. D. Faddeev and V. N. Popov, Phys. Rev. D 26, 3342 (1982). Phys. Lett. B 25, 29 (1967). [112] R. Loll, Phys. Rev. D 41, 3785 (1990). [127] I. A. Batalin and G. A. Vilkovisky, [113] C. Isham, in Relativity, Groups and Topology II, Les Phys. Lett. B 69, 309 (1977). Houches Summer School, 1983, edited by B. DeWitt [128] B. Dittrich, P. A. H¨ohn, T. A. Koslowski, and M. I. and R. Stora (North Holland, Amsterdam, 1984) pp. Nelson, Phys. Lett. B 769, 554 (2017). 1059–1290. [129] B. Dittrich, P. A. H¨ohn, T. A. Koslowski, and M. I. [114] K. Schleich, Class. Quant. Grav. 7, 1529 (1990). Nelson, (2015), arXiv:1508.01947 [gr-qc]. [115] J. D. Romano and R. S. Tate, [130] F. Hellmann, M. Mondragon, A. Perez, and C. Rovelli, Class. Quant. Grav. 6, 1487 (1989). Phys. Rev. D 75, 084033 (2007). [116] G. Kunstatter, Class. Quant. Grav. 9, 1469 (1992). [131] C. J. Isham, Lectures on Quantum Theory: Mathemati- [117] D. N. Page, NSF-ITP-89-18 (1989). cal and Structural Foundations (Imperial College Press, [118] D. N. Page, in Physical Origins of Time Asymmetry, 1995). edited by J. J. Halliwell, J. P´erez-Mercader, and W. H. [132] B. Dittrich and T. Thiemann, Zurek (Cambridge University Press, Cambridge, 1994). J. Math. Phys. 50, 012503 (2009). [119] M. Bojowald and A. Tsobanjan, [133] L. Hardy, (2018), arXiv:1807.10980 [quant-ph]. Commun. Math. Phys. 382, 547 (2021). [134] C. Kiefer, Phys. Rev. D38, 1761 (1988). [120] E. Moreva, G. Brida, M. Gramegna, V. Gio- [135] R. Gambini, R. A. Porto, and J. Pullin, Phys. Rev. vannetti, L. Maccone, and M. Genovese, Lett. 93, 24041 (2004). Phys. Rev. A 89, 052122 (2014). [136] D. N. Page, Class. Quant. Grav. 1, 417 (1984). [121] K. Noui and A. Perez, [137] A. Y. Kamenshchik, I. Khalatnikov, and A. Toporen- Class. Quant. Grav. 22, 1739 (2005). sky, Int. J. Mod. Phys. D 6, 673 (1997). [122] K. V. Kuchaˇr, Phys. Rev. D 34, 3031 (1986). [138] N. Cornish and E. Shellard, [123] B. Dittrich and P. A. H¨ohn, Phys. Rev. Lett. 81, 3571 (1998). J. Math. Phys. 54, 093505 (2013).

Appendix A: Comment on the validity of the absence of interactions

In the quantum theory, it has been shown that if a tensor factorization of the total Hilbert space of the clock and system exists in which the interaction term in the Hamiltonian constraint vanishes, then this factorization is unique [69]. In the context of the Page-Wootters formalism, this has been used as an argument against the ‘clock ambiguity problem’ (related to the ‘multiple choice problem’ in quantum gravity [10, 11]). According to the argument, that clock-system decomposition, which leads to a tensor factorization without interactions (and which is unique if it exists), singles out a preferred clock among a choice of infinitely many. One might thus wonder whether such a tensor factorization is always possible. For example, such an interaction-free factorization of the total Hilbert space is possible for homogeneous vacuum cosmologies, leading to CH in the form of Eq. (4). This has previously been exploited to simplify solving the quantum constraints [85]. However, for generic systems such an interaction free decomposition of the total Hilbert space is not possible. The classical analog of a unitary transformation changing the tensor product structure is a symplectic transformation on , leading to (under our assumptions) a new decomposition ′ ′ (possibly only Pkin ≃ PC ×PS Pkin ≃ PC ×PS locally). Now suppose dim ′ = dim ′ = 2, so that dim = 4, which is the smallest phase space dimension PC PS Pkin in which chaos can appear for autonomous systems. (For a general relativistic example, see [29, 136–138].) If CH did generate chaotic dynamics, it would have to include a non-vanishing interaction term, say HCS, in the original partition because all Hamiltonians of the form of Eq. (4) are completely integrable in four phase space (they decouple the dynamics of the two-dimensional , , which, being autonomous, are completely integrable). If PC PS a symplectic transformation existed that leads to HC′S′ = 0 in the new partition, it would change the dynamics from being chaotic to being integrable, which is impossible. This is a strong indication that for chaotic, or more generally, non-integrable systems (and these are generic), one cannot find a partition such that the interaction term vanishes globally, neither classically, nor in the quantum theory. This resonates with the criticism raised in [77] on the grounds of complex dynamics against the decompositions used in the Page-Wootters formalism. Note, however, that it may still be possible to define a relational dynamics in non-integrable systems (see [128, 129] for developments in this direction). Clock-system interactions have recently been consider within the Page-Wootters formalism [50], leading to a time non-local Schr¨odinger equation satisfied by 41 the system S with respect to the clock C d i ψ (t) = H ψ (t) + dt K(t,t′) ψ (t′) , dt | S i S | S i | S i Z where the second term on the right hand side is a self-adjoint integral operator, the of which K(t,t′) := t H t′ h | int| i depends on an interaction Hamiltonian Hint appearing in a Hamiltonian constraint.

Appendix B: Freedom of choice in classical and quantum time observables

For a given classical or quantum system, there is a freedom in choosing the time observable (assuming that one exists). In the classical case, given a time observable T satisfying the condition T,H = 1, an equivalent time { C} observable can be constructed by T˜ := T + h(HC ) for an arbitrary real function h(HC ). In the quantum case, the freedom of choice is represented by the arbitrary real function g(ε) in Eqs. (14) and (19). We now demonstrate the equivalence of these two freedoms when the quantum clock’s Hamiltonian has a continuous spectrum. First, let us assume that g(ε) is an analytic function, so that g(HˆC ) can be defined via its Taylor series. Now consider two covariant POVMs; the first, denoted ET , with time operator Tˆ, corresponds to the choice g(ε) = 0, and the second, ˜ˆ denoted ET˜, with time operator T , corresponds to an arbitrary choice of g(ε). Using Eqs. (8) and (14) one can see ˆ ˆ ˆ ˆ ig(HC ) ig(HC ) ˜ˆ ig(HC ) ˆ ig(HC ) that ET˜ = e ET e− , and therefore T = e Te− . Using the Baker-Campbell-Hausdorff formula, the latter expression can be written as

ˆ ∞ in T˜ = g(HˆC ), Tˆ . (B1) n! n n=0 X h i Expressing g(HˆC) via its Taylor series and using the canonical commutation relation in Eq. (17), after some calculation one finds

∞ g(n+1)(0) ∞ h(n)(0) g(Hˆ ), Tˆ = i Hˆ n = i Hˆ n = ih(Hˆ ), C − n! C − n! C − C n=0 n=0 h i X X (n) th (1) where g (ε) denotes the n derivative of g(ε), and we have defined h(ε) := g (ε). Consequently, [g(HˆC ), Tˆ]n =0 ˆ for n> 1, and then Eq. (B1) gives T˜ = Tˆ + h(HˆC ), which is exactly the quantization of the classical time observable T˜ above. In other words, the quantum freedom in choosing g(ε) is equivalent to the classical freedom in choosing h(HC ), the two functions being related by differentiation/integration.

Appendix C: Proofs of lemmas and theorems of Secs. IV and V

ˆ ˆ ˆ Theorem 1. FfS ,T (τ) is a (strong) Dirac observable, that is, FfS ,T (τ) commutes with the constraint operator CH ˆ ˆ CH , FfS ,T (τ) =0. h i Proof. To prove the first part of the theorem, consider ˆ UCS(s)FfS ,T (τ) 1 ˆ = dt UCS(t + s) τ τ fS UCS† (t) 2π R | ih |⊗ Z   1 ˆ = dt UCS(t) τ τ fS UCS† (t s) 2π R | ih |⊗ − Z ˆ   = FfS ,T (τ) UCS (s), where in the first and third equality we used Eq. (27) and the second equality follows from changing the integration variable, t t + s. It follows that → U (s), Fˆ (τ) =0, s R. (C1) CS fS ,T ∀ ∈ h i Differentiating both sides of Eq. (C1) with respect to s yields Eq. (28), as desired. 42

Lemma 1. Let Π be the projector from to its subspace spanned by all system energy eigenstates E with σSC HS | iS E σ , i.e. those permitted upon solving the constraint. The quantum relational Dirac observables Fˆ (τ) and ∈ SC fS ,T FˆΠ f Π ,T (τ) are weakly equal, i.e. coincide on phys. Hence, the relational Dirac observables associated to sys- σSC S σSC H ˆ ˆ ˆ tem observables form equivalence classes where FfS ,T (τ) and FgS ,T (τ) are equivalent if ΠσSC fS ΠσSC = ΠσSC gˆS ΠσSC .

Proof. Since I Π ψ = ψ and [Π , Hˆ ] = 0, we can write C ⊗ σSC | physi | physi σSC S Fˆ (τ) ψ = (I Π ) Fˆ (τ) (I Π ) ψ fS ,T | physi C ⊗ σSC fS ,T C ⊗ σSC | physi 1 ˆ ˆ itCH ˆ itCH = dt e− τ τ ΠσSC fS ΠσSC e ψphys 2π R | ih |⊗ | i Z   = FˆΠ f Π ,T (τ) ψphys . σSC S σSC | i

Theorem 2. Let fˆ ( ) and denote by fˆphys := Π fˆ Π its projection to phys. The map S ∈ L HS S σSC S σSC HS F (τ): phys ( ) T L HS → L Hphys  ˆphys ˆ fS Ffˆphys,T (τ) 7→ S is weakly an algebra homomorphism with respect to addition, multiplication and the commutator. That is, the following holds: ˆ ˆ ˆ ˆ Ff phys+gphys hphys,T (τ) Ff phys,T (τ)+ Fgphys,T (τ) Fhphys,T (τ) S S · S ≈ S S · S ˆ ˆ ˆ Ff phys,T (τ), Fgphys,T (τ) F[f phys,gphys],T (τ) , S S ≈ S h i where is the quantum weak equality of Eq. (29). ≈

Proof. That the map FT (τ) is a homomorphism with respect to addition is evident from the linearity of Eq. (27) in fˆS. Let us now check multiplication. Recalling Eqs. (15) and (16), we have

ˆ ˆ 1 phys ˆphys Fgphys,T (τ) Fhphys,T (τ)= dt ds UCS(t) τ τ gˆS UCS(s t) τ τ hS UCS† (s) S · S (2π)2 R | i h |⊗ − | i h |⊗ Z     1 phys ˆphys = dt ds UCS(t) τ τ (ˆgS χ(t s) US(s t) hS ) UCS† (s) . (2π)2 R | i h |⊗ − − Z  

Since U † (s) ψ = ψ , we can write SC | physi | physi

ˆ ˆ 1 phys ˆphys Fgphys,T (τ) Fhphys,T (τ) ψphys = dt ds UCS(t) τ τ (ˆgS χ(t s) US(s t) hS ) ψphys S · S | i (2π)2 R | i h |⊗ − − | i Z   1 phys ˆphys = dt ds UCS(t) τ τ (ˆgS χ∗(s) US(s) hS ) ψphys , (C2) (2π)2 R | i h |⊗ | i Z   upon a shift of integration variable. Next, we show that the operator

1 ΠσSC := dt χ∗(t)US(t) 2π R Z 1 iEt = dt χ∗(t) e− E E 2π R | ih | Z ZE 1 P iEt = dt χ∗(t)e− E E (C3) 2π R | ih | ZE  Z  P 43 is, in fact, the projector onto the HˆS eigenstates compatible with the constraint Eq. (23). The integration over t may be performed case by case by using Eq. (16)

2πδ(t), σ =R, 1 1 c iEt iEt iεmint 1 dt χ∗(t)e− = dt e− e− πδ(t) iP t , σc =(εmin, ), R R iε t iε t 2π 2π e− min e− −max ∞ Z Z  i − , σc =(εmin,εmax), −  t  1, σc =R, 1 =  2 [1 sgn(εmin + E)] , σc =(εmin, ), 1 − ∞  [sgn(εmax + E) sgn(εmin + E)] , σc =(εmin,εmax), 2 − 1, σc =R, = θ( ε E), σ =(ε , ),  − min − c min ∞ θ( εmin E) θ( εmax E), σc =(εmin,εmax). − − − − −  Hence, 

ΠσSC = E E , E σSC | ih | Z ∈ P is precisely the projector from the system Hilbert space S used in kinematical quantization to its subspace compatible with the constraint Eq. (23), i.e. to its physical subspace.H Accordingly, Eq. (C2) becomes

ˆ ˆ 1 phys ˆphys Fgphys,T (τ) Fhphys,T (τ) ψphys = dt UCS(t) τ τ (ˆgS ΠσSC hS ) ψphys S · S | i 2π R | i h |⊗ | i Z   1 phys ˆphys = dt UCS(t) τ τ (ˆgS hS ) ψphys 2π R | i h |⊗ · | i Z   1 phys ˆphys = dt UCS(t) τ τ (ˆgS hS ) UCS† (t) ψphys 2π R | i h |⊗ · | i Z ˆ   = Fgphys hphys,T (τ) ψphys . S · S | i

ˆphys ˆphys In the second step we used that ΠσSC hS = hS . Recalling the definition of the quantum weak equality in Eq. (29) yields the desired result.

Since the commutator involves only multiplication and subtraction, the above also implies that FT (τ) is a homo- morphism with respect to the commutator.

ˆ ˆ Theorem 3. Let fS ( S). The quantum relational Dirac observable FfS ,T (τ) acting on phys, Eq. (27), reduces ∈ L H H phys under S(τ) to the corresponding projected observable in the relational Schr¨odinger picture on , R HS

1 S (τ) Fˆ (τ) S− (τ) = Π fˆ Π , R fS ,T R σSC S σSC where Π is the projector so that phys = Π ( ). Conversely, let fˆphys phys . The encoding operation σSC HS σSC HS S ∈ L HS in Eq. (43) of system observables coincides on the physical Hilbert space phys with the quantum relational Dirac observables in Eq. (27), i.e. H

τ ˆphys ˆ S fS Ff phys,T (τ), E ≈ S   where is the quantum weak equality of Eq. (29). ≈ 44

Proof. Suppose fˆ is any linear operator on . The first statement is proved by direct computation S HS

1 S (τ) Fˆ (τ) S− (τ) = ( τ I ) τ τ fˆ δ(Cˆ ) ( τ I ) R fS ,T R h |⊗ S G | ih |⊗ S H | i⊗ S   1 itCˆH itCˆH 1 isCˆH = ( τ IS) dt e− τ τ fˆSe ds e− ( τ IS) h |⊗ 2π R | ih |⊗ 2π R | i⊗  Z  Z 1 ˆ = dtds τ t + τ τ τ + s t US(t)fS US† (t s) (2π)2 R h | ih | − i − Z 1 ˆ = dtds χ∗(t)χ(t s)US(t)fSUS† (t s) (2π)2 R − − Z 1 ˆ = dtds χ∗(t)χ(s)US(t)fS US† (s) (2π)2 R Z ˆ = ΠσSC fSΠσSC , where in the last step we have made use of Eq. (C3), which defines precisely the projector from the system Hilbert phys space S used in kinematical quantization to the one after Page-Wootters reduction S . This proves the first statement.H H The second statement is proved by recalling Eqs. (40) and (41) and the observation that

τ phys 1 phys S fˆ = S− (τ) fˆ S(τ) E S R S R   1 ˆphys = dt t τ US(t τ) fS 2π R | ih |⊗ − Z 1 ˆphys = dt UCS(t) τ τ fS , 2π R | ih |⊗ Z   where we used Eq. (10) and a shift of the integration variable. Since U † (t) ψ = ψ we can write CS | physi | physi

τ ˆphys 1 ˆphys S fS ψphys = dt UCS(t) τ τ fS ψphys E | i 2π R | ih |⊗ | i   Z   1 ˆphys = dt UCS(t) τ τ fS UCS† (t) ψphys 2π R | ih |⊗ | i Z   = τ τ fˆphys ψ , G | ih |⊗ S | physi   where is the G-twirl operation. Comparing with Eq. (27) proves the claim. G

Theorem 4. Let fˆ ( ) and fˆphys = Π fˆ Π be its associated operator on phys. Then S ∈ L HS S σSC S σSC HS phys τ phys φ Fˆ (τ) ψ = φ (τ) fˆ ψ (τ) = φ S(fˆ )) ψ , h phys | fS ,T | physiphys h S | S | S i h phys |E S | physiPW where ψ (τ) = S(τ) ψ . | S i R | physi

Proof. Using the definition of the physical inner product Eq. (26), Lemma 1 and Eq. (44), we have

τ phys φ Fˆ (τ) ψ = φ S(fˆ ) δ(Cˆ ) ψ h phys | fS ,T | physiphys h kin |E S H | kinikin ˆ ˆphys ˆ = φkin δ(CH ) τ τ fS δ(CH ) ψkin (43) h | | ih |⊗ | ikin   ˆphys = φphys τ τ fS ψphys h | | ih |⊗ | ikin  ˆphys  = φS (τ) fS ψS(τ) . (39) h | | i

To show also equivalence with the expectation value in the Page-Wootters inner product Eq. (38), we insert an identity 45 in the first line above, yielding τ phys φ Fˆ (τ) ψ = φ S(fˆ ) δ(Cˆ ) ψ h phys | fS ,T | physiphys h kin |E S H | kinikin τ phys = φ I S(fˆ ) δ(Cˆ ) ψ h kin | phys E S H | kinikin ˆ τ ˆphys ˆ = φkin δ(CH ) ( τ τ IS) S(fS ) δ(CH ) ψkin kin (42) h | | ih |⊗ E | i τ phys = φ ( τ τ I ) S(fˆ ) ψ h phys | | ih |⊗ S E S | physikin τ ˆphys = φphys S(fS ) ψphys PW . (38) h |E | i

Lemma 2. The trivialization map given in Eq. (45) trivializes the constraint to the clock degrees of freedom

1 T CˆH − = HˆC ε IS , T TT − ∗ ⊗   1 for any ε R. Furthermore, for ε Spec(HˆC ), − is the left inverse of T on physical states, ∗ ∈ ∗ ∈ TT T 1 − I , (C4) TT ◦ TT ≈ phys and the trivialization transforms physical states into product states with a fixed and redundant clock factor

i g(ε∗) i g( E) T ψphys = e ε e− − ψkin( E, E) E . (C5) ∗ C S T | i | i ⊗ E σSC − | i Z ∈ P

Proof. First note that after a shift of integration variables

ˆ(n) 1 n UC (s) T UC† (s)= dt (t s) t t . 2π R − | ih | Z Differentiation with respect to s and subsequently setting s = 0 gives

(n) (n 1) [Tˆ , HˆC ]= in Tˆ − .

Accordingly,

∞ n n i (n) [ T , HˆC ]= [Tˆ , HˆC ] HˆS + ε T n! ⊗ ∗ n=0 X   = IC HˆS + ε T . − ⊗ ∗ T   Recalling Eq. (23), this directly implies

1 T CˆH − = HˆC ε IS . T TT − ∗ ⊗   Note that so far we have not made any assumption about the value of ε . Next, we find ∗

1 1 i(t s)(HˆS +ε∗) T− T = dt ds χ(t s) t s e− − T · T (2π)2 R − | ih |⊗ Z 1 i(t s)ε∗ = dt ds χ(t s) ( t t IS ) UCS(t s)e− − (2π)2 R − | ih |⊗ − Z 1 isε∗ = dt ds χ(s) ( t t IS ) UCS(s)e− , (2π)2 R | ih |⊗ Z upon a change of integration variable. Since U (s) ψ = ψ , CS | physi | physi

1 1 isε∗ T− T ψphys = dt ds χ(s) ( t t IS ) e− ψphys . T · T | i (2π)2 R | ih |⊗ | i Z 46

Now we invoke the assumption that ε Spec(HˆC ) to find ∗ ∈

1 isε 1 is(ε ε) ds χ(s) e− ∗ = dε ds e− ∗− =1 . (C6) 2π R 2π R Z Zσc Z Recalling that the clock states form a resolution of the identity, Eq. (9), yields Eq. (C4). Finally, using Eq. (24), we have

1 i t(E+ε∗) T ψphys = ψkin( E, E) dt e t t E C E S . (C7) T | i E σSC − 2π R | i h |− i | i Z ∈ Z P Invoking Eq. (14) yields

1 1 ′′ ′ ′′ ′ i t(E+ε∗) i t(E+ε∗) i[g(ε ) g(ε )] i(ε ε )t dt e t t ε C = dt e dε′ dε′′ e − e− − ε′′ C ε′ ε C 2π R | i h | i 2π R | i h | i Z Z Zσc 1 ′′ ′′ i[g(ε ) g(ε)] i(E+ε∗ ε +ε)t = dt dε′′ e − e − ε′′ C 2π R | i Z Zσc i[g(ε′′) g(ε)] = dε′′ e − δ(E + ε ε′′ + ε) ε′′ ∗ − | iC Zσc i [g(E+ε∗+ε) g(ε)] e − E + ε + ε if E + ε + ε Spec(HˆC ), = | ∗ iC ∗ ∈ (C8) (0 otherwise.

This makes it clear that T cannot be a unitary (conditional) shift operator of the clock energy if Spec(HˆC ) = R, which is also when the clockT states are non-orthogonal and T (n) are not self-adjoint. But this is not a problem for6 us, as we need for much more restricted purposes. Indeed, applying Eq. (C8) to Eq. (C7), directly yields Eq. (C5), TT provided ε Spec(HˆC ). ∗ ∈

Lemma 3. On physical states, the quantum symmetry reduction map is equal to

H τ U † (τ) R ≈ h |⊗ S while its inverse can also be written as

1 H− = δ(Cˆ ) ( τ U (τ)) . R H | i⊗ S Moreover, the two maps are the appropriate inverses of one another:

1 H− H = I , R ◦ R phys 1 H H− = Π . R ◦ R σSC

Proof. Invoking the definition Eq. (45), we find

i ε∗ τ iε∗τ 1 it(HˆS +ε∗) H = e− ( τ IS ) T = e− dt χ(τ t) t e R h |⊗ T 2π R − h |⊗ Z 1 iε∗t = ( τ US† (τ)) dt χ(t)∗ e UCS† (t), (C9) h |⊗ 2π R Z upon also performing a change of integration variable. Noting that UCS† (t) ψphys = ψphys and using Eq. (C6), yields | i | i

H ψ = τ U † (τ) ψ . R | physi h |⊗ S | physi 47

Next, employing Eq. (52) and the definition Eq. (46) of the inverse trivialization, we compute

1 1 1 iε∗t H− = T− dt e t IS R (18) T 2π R | i⊗ Z

1 iε t is(HˆS +ε ) = ds dt χ(s t) e ∗ s e− ∗ (2π)2 R − | i⊗ Z 1 isε is(HˆS +ε ) = ds dε δ(ε ε ) e s e− ∗ (C6) 2π R σ − ∗ | i⊗ Z Z c 1 = ds s US(s) 2π R | i⊗ Z ′ 1 iHˆC (s τ ) = ds e− − τ ′ US(s) 2π R | i⊗ Z = δ(Cˆ ) ( τ ′ U (τ ′)) , (C10) H | i⊗ S where in the last line we have changed integration variables, s s τ ′. 1 7→ − Since H− is independent of the choice of τ ′, we can set τ ′ = τ so that R 1 H− H ψ = δ(Cˆ ) ( τ τ I ) ψ . R ◦ R | physi H | i h |⊗ S | physi 1 It is thus clear from Eq. (42) that H− H = Iphys for any τ R. Conversely, R ◦ R ∈

1 1 iε∗t 1 H H− = ( τ US† (τ)) dt χ(t)∗ e UCS† (t) ds UCS(s) ( τ ′ US(τ ′)) R ◦ R (C9) h |⊗ 2π R 2π R | i⊗ Z Z

1 iε∗t = dt ds χ∗(t) e τ UCS(s t) τ ′ US(τ ′ τ) (2π)2 R h | − | i − Z 1 iε∗t = dt ds χ∗(t) e χ(τ τ ′ s + t) US(s t + τ ′ τ) (2π)2 R − − − − Z 1 iε∗t = dt du χ∗(t) e χ∗(u) US(u) (2π)2 R Z = ΠσSC . (C6,C3)

ˆ ˆ Theorem 5. Let fS ( S ). The quantum relational Dirac observables FfS ,T (τ) on phys, Eq. (27), reduce under ∈ L H H phys H to the corresponding projected evolving observables of the relational Heisenberg picture on , Eq. (55), i.e. R HS 1 H Fˆ (τ) H− = Π fˆ (τ) Π . R fS ,T R σSC S σSC Conversely, let fˆphys(τ) phys be any evolving observable, Eq. (55). In analogy to Eq. (44), S ∈ L HS   phys H ˆ ˆ phys fS (τ) Ff ,T (τ). E ≈ S  

Proof. Direct computation yields for any τ ′ ′ ˆ 1 i ε∗ τ ˆ ˆ H FfS ,T (τ) H− = e− ( τ ′ IS) T FfS ,T (τ) δ(CH ) ( τ ′′ US(τ ′′)) R R (C10) h |⊗ T | i⊗

1 ˆ ′ it(HS +ε∗) iτ ε∗ ˆ = dt ds du χ(τ ′ t) t e − UCS(s) τ τ fS UCS† (s + u) ( τ ′′ US(τ ′′)) (2π)3 R − h |⊗ | i h |⊗ | i⊗ Z   ′′ ′ 1 i(t s)HˆS i(s+u τ )HˆS iε∗(t τ ) = dt ds du χ(τ ′ t)χ(t τ s)χ(τ + s τ ′′ + u) e − fˆS e − e − (2π)3 R − − − − Z 1 = dt ds du χ(τ ′ t)χ(t τ s)χ(τ + s τ ′′ + u) (2π)3 R − − − − Z ′′ ′ i(t s τ)HˆS i(s+u τ +τ)HˆS iε (t τ ) e − − U † (τ) fˆ U (τ) e − e ∗ − . × S S S 48

Performing now in sequence the variable shifts v = s u + τ ′′ τ, w = τ + s t and x = t τ ′, then recalling the − − − − − definition of the projector ΠσSC in Eq. (C3) and using Eq. (C6), one finally obtains 1 H Fˆ (τ) H− = Π fˆ (τ) Π . R fS ,T R σSC S σSC Conversely, employing Lemma 3, we find for any τ ′ in H R phys 1 phys H fˆ (τ) ψ = H− fˆ (τ) H ψ E S | physi R S R | physi   phys = δ(Cˆ ) ( τ ′′ U (τ ′′)) fˆ (τ) τ ′ U † (τ ′) ψ . H | i⊗ S S h |⊗ S | physi   1 Next, we recall that H− is independent of the choice of τ ′′ and that likewise τ ′ U † (τ ′) ψ is independent R h |⊗ S | physi of the choice of τ ′. In particular, we are therefore free to set τ ′′ = τ ′ = τ. In conjunction with Eq. (55), this yields

phys phys H fˆ (τ) ψ = δ(Cˆ ) τ τ fˆ ψ E S | physi H | i h |⊗ S | physi   ˆ   = Ff phys,T (τ) ψphys , S | i where in the last line we have made use of Eq. (43) and Theorem 3.

phys iτHˆ S iτHˆS Theorem 6. Let fˆ ( ) and fˆ (τ) = e Π fˆ Π e− be its associated evolving Heisenberg S ∈ L HS S σSC S σSC operator on phys. Then HS φ Fˆ (τ) ψ = φ fˆphys(τ) ψ , (C11) h phys| fS ,T | physiphys h S | S | S i phys where ψ = H ψ . | S i R | physi ∈ HS Proof. Using the second result of Theorem 5, Lemma 1 and the definition of the physical inner product Eq. (26), one finds

phys φ Fˆ (τ) ψ = φ H fˆ (τ) ψ h phys| fS ,T | physiphys h phys| E S | physiphys phys  = φ H fˆ (τ) ψ h kin| E S | physikin 1 phys  = φ H− fˆ (τ) H ψ h kin| R S R | physikin 1 phys = φ H− fˆ (τ) ψ . h kin| R S | Si Invoking Eqs. (22) and (C10), yields

1 1 φkin H− = dε φkin∗ (ε, E) C ε S E dt t US(t) h | R h | h | 2π R | i⊗ Zσc ZE Z P i g( E) = φkin∗ ( E, E) e − S E (14) E σSC − h | Z ∈ P = φS , (C12) (51) h | where the latter is a dual reduced state on phys. Hence, HS φ Fˆ (τ) ψ = φ fˆphys(τ) ψ . (C13) h phys| fS ,T | physiphys h S | S | S i

phys Corollary 3. The relational Heisenberg picture on S , obtained through the quantum symmetry reduction H, is only equivalent to the relational Heisenberg pictureH of reduced phase space quantization described in Sec. IVR B if red σCS = σS , i.e. if

Spec(Hˆ red) = Spec(Hˆ ) Spec( Hˆ ). S S ∩ − C Specifically, in this case, 49

phys (i) red := H ( ), HS ≃ HS R Hphys

red phys 1 (ii) Hˆ Hˆ := H Hˆ H− , and S ≡ S R S R

phys 1 ˆ H ˆ phys (iii) The set of quantum symmetry reduced evolving observables, Eq. (55), fS (τ) = Ff ,T (τ) H− coincides R S R ˆred with the set of evolving observables fS (τ), Eq. (34), from reduced phase space quantization. In particular, phys under the appropriate identifications, ψred ψ = H ψ and fˆ (τ) fˆred(τ), we have | S i ≡ | Si R | physi S ≡ S

red ˆred red ˆphys ˆ φS fS (τ) ψS φS fS (τ) ψS = φphys Ff phys,T (τ) ψphys phys . h | | i ≡ h | | i h | S | i

Proof. red contains all wave functions ψred(E) which are square-summable/integrable over the spectrum σred, as HS S S evident from Eq. (33). Similarly, phys contains all wave functions ψ (E) which are square-summable/integrable HS S over the spectrum σCS, as shown by Eqs. (C12), (C13), (26) and (51). These two sets of wavefunctions coincide if red red σS = σCS . Under the identification ψS (E)= ψS(E) (and possibly a redefinition of the integration/sum measure in phys red red one of the representations depending on whether E E′ is normalized identically on and ), where ψ (E) h | iS HS HS S is taken from the expansion Eq. (32) and ψS (E) is the wave function of the quantum reduced state given in Eqs. (50) red red red and (51), we have ψS ψS . Then by corollary 2 and Eqs. (26) and (33), it follows that φS ψS = φS ψS . This proves (i). | i ≡ | i h | i h | i Given that red and phys admit the same energy eigenstates, (ii) immediately follows, HS HS

red phys 1 Hˆ Hˆ := H Hˆ H− . S ≡ S R S R

phys ˆ phys phys ˆ phys ˆ red phys ˆ red ˆ i HS τ ˆ i HS τ i HS τ ˆ i HS τ Lastly, invoking (ii), note that by Eq. (55) fS (τ) = e fS e− = e fS e− , for any ˆphys phys ˆred ˆred red observable fS on S , while fS (τ) is given in Eq. (34) and requires fS to be any observable on S . Since red phys H ˆred ˆphys ˆphys ˆred H S S , we have fS (τ) fS (τ) for the appropriate identification of fS fS at τ = 0. The rest of Hstatement≃ H (iii) is now a direct consequence≡ of Theorem 6. ≡

ˆphys phys phys Theorem 7. Consider an operator on BS from the perspective of A described by OBS A ( B S ). From | ∈ L H ⊗ H the perspective of B, this operator is τ independent so that Oˆphys (τ , τ )= Oˆphys (τ ) ( phys phys) if and B AS B A B AS B A ∈ L HA ⊗ HS only if | |

ˆphys ˆphys ˆphys OBS A = OB A fS A , (C14) | | i ⊗ | i i X     ˆphys ˆphys ˆphys ˆ where (fS A )i is an operator on S and (OB A )i is a constant of motion, [(OB A )i, HB]=0. Furthermore, in this case | | |

ˆphys ˆphys ˆphys ˆ OAS B(τA) = ΠσABS AS τA τA fS A tB OB A δ(CH ) tB ΠσABS , (C15) G | ih |⊗ i h | i | i | " i | | # X       where Π is a projection onto the subspace of spanned by energy eigenstates whose energy lies in σABS HA ⊗ HS σ := Spec(Hˆ + Hˆ ) Spec( Hˆ ), t is an arbitrary clock state of B, and is the G-twirl over the group ABS A S ∩ − B | Bi GAS generated by HˆA + HˆS.

Proof. For simplicity, we drop the ‘phys’ labels on the operators in the following proof, implicitly assuming that we 50

phys phys phys ˆ ˆ ˆ always work with operators on A , B and S . Suppose now that OBS A = OB A OS A. Then H H H | | ⊗ | ˆ ˆ ˆ ˆ ˆ OAS B = τB δ(CH ) τA τA OB A OS A δ(CH ) τB | h | | ih |⊗ | ⊗ | | i 1  ˆ ˆ ˆ  ˆ ˆ ˆ it(HA+HB +HS ) ˆ ˆ is(HA+HB +HS ) = ds dt τB e τA τA OB A OS A e− τB (2π)2 R R h | | ih |⊗ | ⊗ | | i Z Z   1 ˆ ˆ ˆ ˆ ˆ it(HA+HS ) ˆ is(HA+HS ) = ds dt τB + t OB A τB + s e τA τA OS A e− (2π)2 R R h | | | i | ih |⊗ | Z Z   1 ˆ ˆ ˆ ˆ ˆ it(HA+HS ) ˆ is(HA+HS ) = du dv u OB A v ds dt χ(τB + t u)χ(v s τB )e τA τA OS A e− (2π)4 R h | | | i R R − − − | ih |⊗ | Z Z Z   1 ˆ ˆ ˆ ˆ ˆ i(t+u τB )(HA+HS ) ˆ i(v s τB )(HA+HS ) = du dv u OB A v ds dt χ(t)χ(s)e − τA τA OS A e− − − (2π)4 R h | | | i R R | ih |⊗ | Z Z Z   1 ˆ ˆ ˆ ˆ ˆ i(u τB )(HA+HS ) ˆ i(v τB )(HA+HS ) = du dv u OB A v ΠσABS e − τA τA OS A e− − ΠσABS (2π)2 R h | | | i | ih |⊗ | Z   1 ˆ ˆ ˆ ˆ ˆ iu(HA+HS ) ˆ iv(HA+HS ) = du dv τB u OB A τB v ΠσABS e− τA τA OS A e ΠσABS . (2π)2 R h − | | | − i | ih |⊗ | Z   In the sixth line we have adapted the definition of the projector Eq. (C3) to our case ΠσABS . It is seen from the above ˆ ˆ expression that OAS B is independent of τB if and only if τB u OB A τB v is independent of τB. ˆ | h − | | | − i If [OB A,HB] = 0, then | ˆ iHB τB ˆ iHB τB τB u OB A τB v = u e OB Ae− v h − | | | − i h− | | |− i ˆ = u OB A v , h− | | |− i ˆ ˆ and thus OAS B is independent of τB . If OAS B is independent of τB, then | | d ˆ 0= τB u OB A τB v dτB h − | | | − i d ˆ ˆ iHB τB ˆ iHB τB = u e OB Ae− v h− |dτB | |− i ˆ ˆ iHBτB ˆ ˆ iHB τB = i u e OB A, HB e− v , − h− | | |− i h i ˆ ˆ ˆ which vanishes only if OB A is a constant of motion, OB A, HB = 0. By linearity, it follows that the most general | | operator relative to clock A which leads to τB independenceh relativei to clock B is given in Eq. (C14). ˆ If OB A is a constant of motion, then | 1 ˆ ˆ ˆ ˆ ˆ ˆ iu(HA+HS ) ˆ iv(HA+HS ) OAS B = du dv 0 OB A u v ΠσABS e− τA τA OS A e ΠσABS | (2π)2 R h | | | − i | ih |⊗ | Z   1 ˆ ˆ ˆ ˆ ˆ i(u+v)(HA+HS ) ˆ iv(HA+HS ) = du dv 0 OB A u ΠσABS e− τA τA OS A e ΠσABS (2π)2 R h | | | i | ih |⊗ | Z   1 ˆ ˆ ˆ iu(HA+HS ) ˆ = ΠσABS du 0 OB A u e− AS τA τA OS A ΠσABS 2π R h | | | i G | ih |⊗ |  Z    ˆ ˆ ˆ = ΠσABS 0 OB Aδ(CH ) 0 AS τA τA OS A ΠσABS h | | | i G | ih |⊗ | ˆ ˆ  ˆ  = ΠσABS tB OB Aδ(CH ) tB AS τA τA OS A ΠσABS , h | | | i G | ih |⊗ |   where t is any clock state of B. By linearity, this extends to Eq. (C15). | Bi Corollary 4. Consider an observable seen from the perspective of A that acts nontrivially only on S, ˆphys phys ˆphys OBS A = IB A fS A . | | ⊗ | Under a temporal frame change to the perspective of B, such an observable transforms to ˆphys phys ˆphys OAS B = IA B fS B , | | ⊗ | ˆphys ˆphys ˆphys ˆphys ˆ where fS B = fS A if and only if fS A is a constant of motion, [fS A , HS]=0. | | | | 51

ˆ ˆphys Proof. If [HS, fS A ] =0, then Eq. (67) yields | ˆphys ˆphys OAS B = ΠσABS AS τA τA fS A ΠσABS | G | ih |⊗ |  phys  ˆphys = ΠσABS AS τA τA IS A IA B fS A ΠσABS G | ih |⊗ | | ⊗ |  ˆphys  = ΠσABS IA B fS A ΠσABS | ⊗ | phys ˆphys = IA B fS A , | ⊗ | ˆphys ˆphys from which it follows that fS B = fS A . ˆphys ˆphys | | If fS B = fS A , then | | phys ˆphys phys ˆphys IA B fS B = IA B fS A | ⊗ | | ⊗ | ˆphys = ΠσABS IA B fS A ΠσABS | ⊗ | phys ˆphys = ΠσABS AS τA τA IS A IA B fS A ΠσABS . G | ih |⊗ | | ⊗ |   However, from Eq. (67) we also have that

phys ˆphys ˆphys IA B fS B = ΠσABS AS τA τA fS A ΠσABS . | ⊗ | G | ih |⊗ |   Upon comparison of this equation with the previous equation, together with the definition of the G-twirl we conclude ˆphys phys that [fS A ,US(t)]=0 [fS A ,HS] = 0, as desired. | ⇐⇒ | Corollary 5. Consider an operator on BS from the perspective of A described by Oˆphys (τ ) ( phys phys). BS A A ∈ L HB ⊗ HS ˆH | Under a temporal frame change to the perspective of B, this operator transforms to OAS B(τA, τB) that satisfies the | Heisenberg equation of motion in clock B time τB without an explicitly τB dependent term,

d ˆH ˆ ˆ ˆH OAS B (τA, τB )= i HA + HS, OAS B(τA, τB ) , dτB | | h i if and only if

ˆphys ˆphys ˆphys OBS A(τA)= OB A fS A (τA) , | | i ⊗ | i i X     ˆphys ˆ ˆphys and OB A is a constant of motion, [HB, OB A ]=0. | | Proof. From Eq. (68), it follows that ˆH ˆphys OAS B(τA, τB)= UAS† (τB )OAS B(τA, τB)UAS(τB ). | |

Differentiating the above expression with respect to τB yields

d ˆH ˆ ˆ ˆH d ˆphys OAS B(τA, τB)= i HA + HS, OAS B(τA, τB) + UAS† (τB ) OAS B(τA, τB ) UAS(τB). dτB | | dτB | h i   Theorem 7 then implies that the second term vanishes if and only if

Oˆphys = Oˆphys fˆphys , BS A B A i ⊗ S A i | i | | X     ˆphys where OB A are constants of motion. Equivalently, this is true if and only if in the relational Heisenberg picture | i   phys phys phys Oˆ (τA)= Oˆ fˆ (τA) . BS A B A i ⊗ S A i | i | | X     52

Appendix D: Derivation referenced in Sec. VII C

Suppose that from the perspective of A the state of BS is in a product state

ψBS A(τA) = ψB A(τA) ψS A(τA) . | | i | | i | | i A B The action of the TFC map ΛS → on BS yields the state of AS from the perspective of B

A B ψAS B(τB ) =ΛS → ψBS A(τA) | | i | | i ˆ = ( τB IAS) δ(CH ) ( τA IBS) ψB A(τA) ψS A(τA) . h |⊗ | i⊗ | | i | | i 1 it(HˆA+HˆB +HˆS ) = ( τB IAS) dt e− τA ψB A(τA) ψS A(τA) h |⊗ 2π R | i | | i | | i Z 1 = ( τB IAS) dt τA + t ψB A(τA + t) ψS A(τA + t) h |⊗ 2π R | i | | i | | i Z

Changing integration variables to t′ := τA + t and defining ψB A(t t′) := t ψB A(t′) yields | − h | | i 1 ψAS B(τB ) = ( τB IAS ) dt′ t′ A ψB A(t′) ψS A(t′) | | i h |⊗ 2π R | i | | i | | i Z 1 1 = ( τB IAS ) dt′ t′ A dt′′ ψB A(t′′) t′′ + t′ B ψS A(t′) h |⊗ 2π R | i 2π R | | i | | i Z  Z  1 = dt′ dt′′ ψB A(t′′)δ(τB t′′ t′) t′ A ψS A(t′) (2π)2 R R | − − | i | | i Z Z 1 = dt′ ψB A(τB t′) t′ A ψS A(t′) 2π R | − | i | | i Z as stated in Eq. (70).

Appendix E: Mathematical details

1. Canonical transformation separating gauge and gauge-invariant degrees of freedom

We now demonstrate that the transformation TT introduced in Sec. VI B is a canonical transformation. Firstly, we know that T, C = 1 are a canonical pair. It also follows from [14] that { H } f F (τ) S 7→ fS ,T is a strong Poisson-algebra homomorphism on for the special form of Eq. (5). Hence, recalling that Pkin i j Q (τ)= Fqi ,T (τ) P (τ)= F j (τ), S S S pS ,T we have

Qi (τ), P j (τ) = qi ,pj = δij . { S S } { S S} From Eq. (5) it is furthermore obvious that T, F (τ) = 0. Finally, we find that the Dirac observables Eq. (5) { fS ,T } strongly commute with the constraint CH , since

n 1 n ∞ (τ T ) − (T τ) F (τ), C = − f ,H + − f ,H { fS ,T H } − (n 1)! { S S}n n! { S S}n+1 n=0 X  −  =0 .

We thus conclude that T is a canonical transformation on . T Pkin 53

2. Correct propagator from gauge-invariant conditional probability

In this appendix we show how to arrive at the correct propagator from the gauge-invariant conditional probability proposed in Eq. (76): ˆ ˆ ˆ ψphys FΠA=a,T (τ) FΠB=b,T (τ ′) FΠA=a,T (τ) ψphys phys Prob(B = b when τ ′ A = a when τ) := h | · · | i (E1) | ψ Fˆ (τ) ψ h phys| ΠA=a,T | physiphys Firstly, recall Theorem 3 and that Π , Π ( phys) by assumption (otherwise we would have to conjugate A=a B=b ∈ L HS these two projectors by ΠσSC ). Since we are always acting on physical states, we can replace every instance of the relational Dirac observables above by the Page-Wootters encoding, Eq. (43), of the corresponding reduced observables and projections onto the respective clock readings. Invoking the definition of the physical inner product, Eq. (26), this puts Eq. (E1) into the following form:

ˆ ˆ ψphys (eT (τ) ΠA=a)δ(CH )(eT (τ ′) ΠB=b)δ(CH )(eT (τ) ΠA=a) ψphys kin Prob(B = b when τ ′ A = a when τ) h | ⊗ ⊗ ⊗ | i . | ≡ ψ (e (τ) Π ) ψ h phys| T ⊗ A=a | physikin We note that this is the generalization of Dolby’s two-time conditional probability to the case of constraints which have zero in the continuous part of their spectrum [64]. It is clear that the denominator can be rewritten as ψ (e (τ) Π ) ψ = ψ (τ) Π ψ (τ) . h phys| T ⊗ A=a | physikin h S | A=a | S i Let us next rewrite the numerator as

ψ (e (τ) Π ) δ(Cˆ ) (e (τ ′) Π ) δ(Cˆ ) (e (τ) Π ) ψ h phys| T ⊗ A=a H T ⊗ B=b H T ⊗ A=a | physikin 1 1 = ψS(τ) ΠA=a dt χ(τ τ ′ + t) US† (t) ΠB=b ds χ(τ ′ τ s)US(s) ΠA=a ψS (τ) h | 2π R − 2π R − − | i Z Z = ψS(τ) ΠA=a ΠσSC US† (τ ′ τ) ΠB=b ΠσSC US(τ ′ τ) ΠA=a ψS(τ) . (C3) h | − − | i

Recalling that Π Π = Π , since by assumption Π ( phys), we thus obtain in conjunction σSC A=a A=a A=a ∈ L HS

ψS (τ) ΠA=a US† (τ ′ τ) ΠB=b US(τ ′ τ) ΠA=a ψS(τ) Prob(B = b when τ ′ A = a when τ)= h | − − | i . | ψ (τ) Π ψ (τ) h S | A=a | S i This is the correct propagator for transitioning from the system state corresponding to the observable A reading a at Schr¨odinger time τ to the system state corresponding to the observable B reading b at Schr¨odinger time τ ′.