The observer is what separates AdS from de Sitter space, and I think it is the most important thing to understand about both. Anti-de Sitter space comes with a boundary, and the boundary supplies a Hamiltonian, a time and a fixed frame at infinity, so the theory is complete without anyone writing an observer down. Nobody calls the boundary an observer, but it does the job of one, and it is a very special one: infinitely heavy, classical, and untouched by what it measures. De Sitter space has no such boundary. Nothing supplies the clock, the algebra of a static patch is type III and has no well-defined entropy, and the theory only becomes well defined once an observer with a clock is put in by hand, at which point the algebra becomes type II and entropy appears. In AdS the observer is hidden and comes for free. In de Sitter it is explicit and unavoidable. That is why the title repeats the word.

That difference is also where the open questions live. If the AdS boundary is an ideal observer, what is the observer when the clock is finite, when it has to be paid for in energy, and when it is correlated with what it observes? The same question came up for minds in the earlier posts of this series. This post tries to find something concrete enough to compute, and computes it.

The short version: the literature on quantum reference frames makes precise what an observer is. The place where it stops being settled is narrow and specific, namely what happens when the observer’s clock is correlated with the rest of the system. I built two small models and worked through the entropy formula of one recent paper to see what that does. One of my first estimates came out wrong, in a way that made the result sharper. The result that survives is conditional on an assumption I could not check, and I will say exactly which.

The central idea of this post is:

The observer is what distinguishes AdS from de Sitter. In AdS it is an ideal clock hidden at the boundary. In de Sitter it is a finite clock that has to be written into the theory, and then its correlations with everything else become part of the entropy.

Before anything else, the status. This is a research note, not a paper. The numerical results come from small scripts on toy systems. Claims that are conjectures are marked (C), and should not be read as established. The weakest steps are collected in section 9.

A note on reading: terms followed by a reference like (A.3) are explained in plainer language in the appendix at the end.

1. Where the observer sits

In AdS: the boundary. In AdS/CFT1 the boundary theory supplies a Hamiltonian, a time coordinate and a fixed frame, all sitting at infinity. That is why AdS looks complete without any explicit observer. My first way of putting my complaint was that the boundary acts as a privileged observer: an infinitely heavy, non-gravitating, classical clock that never backreacts, and if anything, our universe doesn’t like privileged anything. AdS/CFT describes the world as seen by an ideal clock, and it has nothing to say about universes that have no boundary to put one on.

In de Sitter: the observer is explicit. Without a boundary, there is no such clock for free. Witten showed that gravity turns the type III algebra of a static patch into a crossed product2, and Chandrasekaran, Longo, Penington and Witten (CLPW) showed that adding an observer with a clock makes it type II, with a trace and a well-defined entropy3. Chen and Xu then extended this to observers that are covariant under the de Sitter isometries, so that the observer’s own worldline fluctuates, and showed that the resulting algebra is still type II, with an observer-dependent entropy4.

In a mind: the self. The previous post proposed that a mind that grows without bound has a type III algebra, and that what it needs in order to read its own time from the inside is a clock-bearing frame, with the self as the formal content of that frame. I will not repeat that argument here. I only want to point out that it uses the same mathematics as the de Sitter case.

Maldacena has recently been talking about what happens to the sphere partition function of de Sitter space when an observer with a clock is included5. The point it makes is the one above: in de Sitter space the observer is not optional.

2. What the reference-frame results settle, and what they do not

The results I take to be central, read by me at the level of abstracts and key sections and not every proof:

ResultWhat it showsCaveat as I read it
Page–Wootters67Time as conditioning on a clock, in a state with no time in itExact for an ideal clock
Trinity8Dirac quantization, Page–Wootters and gauge-fixed reduction are equivalentProven for a clock that does not couple to the evolving system
PW = CLPW9The CLPW observer is a temporal quantum reference frame; their constrained description matches Page–WoottersBounded clock energy is imposed after the constraint, so order of limits matters
Observer-dependent entropy10Different frames give different algebras and different entropies
Top-down crossed products11The crossed product is one particular choice of reference frame
Unsharp reference frames12The CLPW clock is an unsharp frame; finiteness of the algebra depends on the clock’s weight

So the chain is: observer, then reference frame, then frame plus constraint gives a crossed product, then the trace of the crossed product gives the entropy. That is a real structural result. It tells you what an observer is. It does not answer the AdS question, for four reasons.

  1. Non-ideal clocks. The equivalences are cleanest for ideal clocks. A finite observer has bounded energy, and that is exactly where the order of limits matters.
  2. The observer observes nothing. In CLPW the clock is not coupled to the fields. A realistic observer exchanges energy with the system and backreacts.
  3. The privilege may just move. Frame-relative descriptions put the frame-independent content into a perspective-neutral state, and if that state has no content without a frame, one privileged observer has been exchanged for another.
  4. AdS and de Sitter have different constraints (my own analysis, unverified). In de Sitter the isometries are pure gauge, so the clock’s time translation is a constraint. In AdS the boundary Hamiltonian is a physical charge, so a bulk clock’s evolution is not pure gauge, and there are two frames at once: boundary time and the bulk clock.

3. The boundary as an ideal clock

Put together, the above suggests a sharper form of the original complaint:

(C) The AdS boundary clock is the infinite-energy limit of a finite bulk observer’s clock. At finite energy, the algebra dressed to the bulk observer is a different algebra, with an observer-dependent entropy, and the difference is controlled by the clock’s energy fluctuations.

This turns “AdS has a privileged observer” into a precise statement: the privilege is a limit and not a feature of the theory. It also has a clear failure condition. If the difference between the finite-clock and boundary-dressed entropies vanishes identically, or can be absorbed in a relabeling, the conjecture is dead, and that would be worth reporting.

I have not done this calculation. The natural arena is JT gravity with matter and one massive particle carrying a quantum clock, because the boundary-dressed algebra is already known there131415. The rest of this post is about a smaller question that I could actually attack: how does the clock’s correlation with the rest of the system enter the entropy at all? Whatever happens at finite clock energy has to pass through that.

4. A toy model: two clocks and a system

The smallest model I could build that has the relevant features: two clocks \$A\$ and \$B\$ and a system \$S\$, with a single constraint. Each clock is an \$N\$-level system with Hamiltonian \$n\$, and the system has integer energies \$h_s\$. The gauge group is \$\mathbb Z_N\$, generated by the constraint

$$ C = n_A + n_B + h_S \equiv 0 \pmod N. $$

The clock readings come from the covariant time states \$|\tau_k\rangle=N^{-1/2}\sum_n e^{-in\tau_k}|n\rangle\$ with \$\tau_k=2\pi k/N\$. A physical state is the group average of “clock \$A\$ reads zero, the rest is in state \$\psi_{BS}\$”. Projecting \$A\$ onto its zero reading gives back \$\psi_{BS}\$, which is the perspective of \$A\$. Doing the same with \$B\$ gives \$\psi_{AS}\$, the perspective of \$B\$. This is the finite-dimensional version of the Page–Wootters mechanism, with the observer-dependence of Giacomini, Castro-Ruiz and Brukner built in1617.

Sanity. The physical states are gauge-invariant to machine precision. The distribution of the relative reading \$\tau_B-\tau_A\$, computed from a gauge-invariant POVM on the whole state, agrees with what each observer gets from their own reduction to \$10^{-16}\$.

Entanglement depends on the observer. Take \$B\$ to be sharp in energy and \$S\$ in a superposition of energies. Seen from \$A\$, \$B\$ and \$S\$ are in a product state, with entropy 0. Seen from \$B\$, the clock \$A\$ is maximally entangled with \$S\$: the entropy goes to \$\ln 4=1.386\$ and I measured 1.38. The reason is that \$S\$’s energy enters the constraint, so shifting \$B\$’s reading shifts \$A\$’s reading in an \$s\$-dependent way. This is the relativity of entanglement across quantum reference frames17. It is not new, and I include it because it is the finite analogue of the observer-dependent entropies in the crossed-product papers. In random states, the two perspectives disagree on the entropy by more than 0.1 in 66% of cases, though the entropies are positively correlated (0.67).

A mismatch between modular time and clock time. For a reduced clock state \$\rho\$ with clock Hamiltonian \$H=n\$, the modular flow is generated by \$-\log\rho\$ and the clock’s own evolution by \$H\$. They agree (up to a rescaling) exactly when \$\rho\$ is thermal for \$H\$. A natural measure of the failure is the relative entropy to the closest thermal state. If \$\Delta\$ is dephasing in the energy basis and \$\gamma_\beta=e^{-\beta H}/Z\$, then

$$ \mathrm{Mis}(\rho)\;=\;\min_\beta D(\rho\,\|\,\gamma_\beta) \;=\;\underbrace{S(\Delta\rho)-S(\rho)}_{\text{coherence}} \;+\;\underbrace{\min_\beta D(\Delta\rho\,\|\,\gamma_\beta)}_{\text{non-thermality}}. $$

The second equality is one line, because \$\gamma_\beta\$ is diagonal in the energy basis. The first term is the standard asymmetry measure of the resource theory of reference frames1819, so the coherence part of the mismatch is a known quantity, and the new content is only the way it is combined with the non-thermal part. Whether that combination does any work is the question.

The compact model has an artifact. Because the group is \$\mathbb Z_N\$, the relabeling \$n_A=-n_B\bmod N\$ wraps around, and a state that is thermal for \$B\$ looks scrambled for \$A\$ for a reason that has nothing to do with physics. I removed this by rebuilding the model with clocks bounded below (\$n\ge 0\$ with no wrap, an exact U(1) constraint, and the Susskind–Glogower time POVM). In the bounded model a thermofield-double-like state of \$B\$ and \$S\$ looks exactly Gibbs to \$A\$ (mismatch 0.000, fitted \$\beta^*=1.000\$ for \$\beta=1\$) and looks like a sharp energy eigenstate to \$B\$, with entropy 0. The same global state is thermal and entangled from one perspective, and sharp and unentangled from the other.

Two branches. Take a state \$\sqrt p\,|\chi_1\rangle|\phi_1\rangle +\sqrt{1-p}\,|\chi_2\rangle|\phi_2\rangle\$ in which the clock \$B\$ records which branch the system is in. From \$A\$’s side, the entropy of \$B\$ is the Holevo information of its clock about the branch. It saturates at \$H(p)=\ln 2\$ when the clock states are resolved in time, and falls (0.518 at time-width 1) as they overlap. From \$B\$’s side it stayed at \$\ln 2=0.693\$ for every time-width I tried, because the two branches differ in the system’s energy, so \$A\$’s states stay energy-distinguishable even when \$B\$ cannot tell the branches apart in time. Which clock “knows” the branch depends on the perspective.

What the toys do and do not do. They show that the clock-correlation term is real, computable, and perspective-dependent. They are finite and type I, with no horizon, no type II trace and no crossed product, so they say nothing yet about the actual entropy formula. That is the next section.

5. Where the correlation enters the entropy formula

I read the Chen–Xu paper4 through its entropy derivation. The structure, as I understand it: any density operator in the invariant algebra can be written as \$\rho=\mathcal D\!\left(e^{d/2}\hat P(d)\,e^{d/2}\right)\$, where \$d\$ is the clock energy and \$\hat P(d)\$ takes values in the matter algebra. This is exact. The semiclassical ansatz then assumes a sharply time-localized clock whose characteristic function approximately factorizes from matter expectation values (their condition 4.2), so that \$\hat P(d)\approx\tilde f(d)\,P(0)\$. Under that assumption the von Neumann entropy has the form (their 4.15)

$$ S\;\approx\;\text{(first-law term)}\;-\;\psi(\log\Delta_{\psi|\tilde\Psi}) \;+\;S_{\rm clock}(\psi), $$

where \$S_{\rm clock}\$ is, I take it, the differential entropy of the clock’s energy distribution. As printed it appears to lack a minus sign, which I read as a typo. The authors say explicitly that the density operator beyond semiclassical states can be obtained but is too complicated to discuss, and that the semiclassical ansatz assumes minimal entanglement between the clock and everything else. This is a gap that the paper itself marks, not an error.

The algebra also has a nontrivial center. For observer \$a\$ the center is generated by projections encoding whether another observer \$b\$ is inside \$a\$’s patch (A.3). These commute with everything, including the clock energy. That makes them ideal branch labels.

Derivation. Suppose the state is correlated with a central variable \$i\$ (for example, “\$b\$ is inside” or “\$b\$ is outside”) with probabilities \$p_i\$. A state on an algebra with a center decomposes into fibers, and the entropy splits exactly:

$$ S(\omega)\;=\;H(p)\;+\;\sum_i p_i\,S(\omega_i)\;\Big(+\sum_i p_i\log z_i\Big), $$

where the last term is the shift from the non-uniqueness of the trace, which depends only on the branch. If the semiclassical conditions hold fiber by fiber, with a branch-dependent clock energy density \$g_i(q)\$, then applying the formula fiber by fiber gives the conditional entropy \$h(q\,|\,i)=\sum_i p_i h(g_i)\$ in place of the marginal \$h(q)\$. The entropy with the correlation is therefore

$$ S\;=\;S_{\rm semi}\;-\;I(i\,;q),\qquad I(i\,;q)\;=\;h(q)-\sum_i p_i\,h(q\,|\,i), $$

where \$S_{\rm semi}\$ is the formula applied naively with the marginal clock distribution. The correction is the classical mutual information between the clock energy and the branch20. It is non-negative, at most \$H(p)\$, and independent of the trace choice, because the \$\log z_i\$ shift cancels. The formula is what a referee would call obvious once written down, which is also why I believe it. What is not obvious is whether it is nonzero in a physical state of the algebra.

For a small shift \$\Delta E\$ between the two clock energy distributions, each Gaussian with minimum-uncertainty width \$\sigma_q=1/(2\epsilon)\$ for time resolution \$\epsilon\$, I get \$I\simeq 2p(1-p)(\Delta E\,\epsilon)^2\$. Numerically for \$p=\tfrac12\$ and \$\Delta E=1\$:

\$\epsilon\$0.050.10.20.51.0
\$I\$ (exact, nats)0.001250.004980.019610.11140.337
\$2p(1-p)(\Delta E\epsilon)^2\$0.001250.005000.020000.12500.500

The formula is good for small \$\Delta E\,\epsilon\$ and it saturates at \$\ln 2\$ for large. Taken alone it would say that a sharp clock makes the correction second-order small, in agreement with the “neglect higher than \$\epsilon\$” convention of the semiclassical derivation. The next section shows why that estimate was too optimistic.

6. Two observers: when the clock records the other observer

Take observer \$a\$ at the origin, a second observer \$b\$, and the QFT in its vacuum. I use three facts from the paper’s setup.

  1. The projector onto “\$b\$ is visible to \$a\$” is a function of \$b\$’s relative geodesic, and \$d_{b,1}\$ generates left multiplication by the boost along \$a\$’s geodesic. That boost preserves \$a\$’s static patch, so it preserves whether \$b\$’s geodesic enters it. Hence the projector commutes with the clock energy, and the joint distribution of (inside or outside, clock energy) is well defined.
  2. With the QFT in the vacuum the constraint reads \$d_{a,1}+d_{b,1}=0\$, so \$a\$’s clock energy is \$q_a=-d_{b,1}\$.
  3. The paper gives the generator along a geodesic \$(\varphi,s)\$ as \$d_1\cos\varphi\cosh s+d_2\sin\varphi\cosh s+l\sinh s\$, and \$b\$ is visible to \$a\$ iff \$\cos\varphi\ge 0\$.

The assumption. If \$b\$ carries charge only along its own geodesic (call it \$E_b\$), then \$d_{b,1}=E_b\cos\varphi\cosh s\$, so

$$ q_a\;=\;-E_b\cos\varphi_b\cosh s_b. $$

For \$E_b>0\$ the sign of \$q_a\$ tells you which side of the horizon \$b\$ is on. This is the classical charge relation, and I have not verified it as an operator statement in the actual representation on \$L^2(SO(1,2))\$. Everything in the rest of this section depends on it.

I computed \$I(\text{inside};q_a)\$ for \$b\$’s geodesic distributed on either side of the patch, and for different clock-energy distributions of \$b\$.

\$E_b\$ distributionstd/mean\$I\$ (nats)
Gamma, shape 0.51.410.598
Gamma, shape 11.000.678
Gamma, shape 2 and above0.71 and below0.693 (\$=\ln 2\$)
symmetric about 0 (no lower bound)n/a0.000

With a lower-bounded clock energy for \$b\$, the clock energy of \$a\$ is, up to horizon-grazing geodesics, an exact record of whether \$b\$ is in its patch, and the correction is the full Shannon entropy \$H(p)\$. Without a lower bound, it vanishes. I had this wrong at first. My estimate at the end of section 5, that the correction is second-order small for a sharp clock, assumed that the two sectors’ clock energy distributions are shifted copies of each other. With a positive-energy \$b\$ they sit on opposite sides of zero and are nearly disjoint, so the correction does not scale like \$(\Delta E\,\epsilon)^2\$ at all.

Two consequences, both conditional on the classical charge relation:

  • The factorization condition of the semiclassical derivation fails in the very two-observer setting the paper builds, if the observers’ clock energies are bounded below. It holds if they are symmetric.
  • The paper says a spectral lower bound on the clock energy is not applicable with several covariant observers, and so does not impose it. Whether the effect exists therefore depends on a choice the authors left open. It also connects to the bound that separates type II\$_1\$ from type II\$_\infty\$ in CLPW, which is the same sign asymmetry seen from another side.

7. Back to AdS, de Sitter and the self

AdS. The conjecture of section 3 is untouched. What the toy models and section 5 add is a mechanism that any finite-clock correction has to pass through: the clock’s correlations with what it times, which depend on the perspective. The boundary clock is the case where those correlations are negligible because the clock is infinitely heavy.

De Sitter. For two or more observers there is a concrete candidate for a nonzero correction, and a clear dependence of its existence on an assumption the authors did not make. That is the part I think could become a paper, if the classical charge relation survives a proper treatment.

In a mind: the self. The previous post proposed that the arrow of time for a mind is a mismatch between modular time and clock time. Section 4 gives that mismatch a definite form (coherence plus non-thermality), shows that it is perspective-dependent, and shows that what looks like a thermal, entangled clock from one frame can be a sharp, unentangled one from another. That supports, weakly, the earlier claim that a copy of a state is not a copy of a frame. It does not support any more than that. The toy is not a model of a mind, and I would not read the cognitive claim off the physics.

8. What would count as evidence, and what is next

TestWhat it settlesStatus
Joint spectral measure of the projector \$\Theta(\cos\varphi)\$ and the left-boost generator on \$L^2(SO(1,2))\$Whether \$d_{b,1}\$ has disjoint spectra in the two sectors for \$E_b\ge 0\$, which is the classical charge relation at the quantum levelnot done; a well-posed harmonic analysis problem
An explicit two-observer state with clock energy densities \$g_{\rm in}\$, \$g_{\rm out}\$ computed from the constraintWhether \$I\$ is nonzero in a physical state, rather than in my illustrationnot done
Non-central correlations (clock with field modes)Whether \$-I\$ survives when the commutation of conditions 4.9 and 4.10 of the paper failsopen; I have no closed form
JT gravity with one massive clockThe conjecture of section 3not done
Literature check: Witten’s background-independent algebra21, the top-down crossed product work11, and anything on observer-dependent entropy in AdSNoveltynot done; this is the first thing to do

9. What this does and does not claim

It does not claim a result about AdS. The conjecture of section 3 is a conjecture, and no calculation in this post tests it. It does not claim that the classical charge relation holds for quantum observers. It does not claim anything about minds beyond the one structural point in section 7.

It does claim a specific, checkable statement about the de Sitter algebra: if a second observer’s clock energy is bounded below and the classical charge relation holds, the entropy of the semiclassical formula of Chen and Xu needs a correction equal to the mutual information between the first observer’s clock energy and whether the second observer is in its patch, which is independent of the trace choice and can be as large as the Shannon entropy of that variable.

The weakest steps, in the order I trust them least:

  1. The classical charge relation \$d_{b,1}=E_b\cos\varphi\cosh s\$. Section 6 stands or falls with it.
  2. Fiber-by-fiber semiclassicality. I assumed that the paper’s conditions hold inside each central sector. I did not check that they can.
  3. The setting. de Sitter in two dimensions, QFT in the vacuum, one second observer, ideal reference frame states. With field excitations the constraint adds a field-energy term to \$q_a\$.
  4. The toy models are finite, type I, and have no horizon. They illustrate the perspective dependence but do not touch the type II trace.
  5. Non-central correlations are not handled at all.
  6. Novelty. I have not finished the literature check, so some of this may already be known.

And a worry in the same spirit as the last post. I found that the same structure, a clock-bearing frame, fits AdS, de Sitter and a model of a mind, and a match across three domains looks like evidence for the framework. It is also exactly what you would see if “reference frame” were a generic concept that fits anything you ask it to. The safeguard I have is the narrowness of section 6: a specific formula that either is or is not nonzero in a specific algebra. That is the part that can fail.

If the previous posts ended on a proposal, this one ends on a question that has an answer. The observer is a clock, a clock can be correlated with what it times, and whether that correlation is large or negligible is a calculation I know how to set up and have not finished.

Appendix

Each entry says what the idea is, gives a small example, and says why it matters for the post.

A.1 Observer, clock, reference frame

A reference frame is whatever measurements are made relative to. Latitude needs an equator and a prime meridian. A clock is a reference frame for time: “the temperature when my clock reads noon”. A quantum reference frame is one that is itself a physical system, so it can be uncertain or entangled with what it measures. In this post an observer means exactly that, a reference frame with a clock and records. It does not need to be a mind, and a thermometer with a timer qualifies.

A.2 Constraints, gauge, and Page–Wootters

A constraint is a condition that physical states must satisfy, such as “the total energy is zero”. In general relativity there is no external time, so the Hamiltonian is a constraint. A state that satisfies it does not change, and yet the world around us does. Page and Wootters resolved this by picking a subsystem to be the clock: the rest of the world, conditioned on the clock reading, evolves normally. A good picture is a photo album of a dance. All the pages exist at once, and “time” is the page number you condition on. A physical state is obtained from any state by group averaging: summing it over all the shifts the constraint generates.

A.3 The center and branches

The center of an algebra is the set of operations that commute with every other operation in it. They behave like classical labels: they are definite whatever else you do. In the de Sitter algebra of an observer, the center contains projections for questions like “is the other observer inside my patch?”. If a state is spread over several values of such a label with probabilities \$p_i\$, the algebra splits into sectors, and the entropy of the state is the Shannon entropy of the \$p_i\$ plus the average entropy inside each sector. I call each value a branch.

A.4 Crossed products and types

Algebras are sorted into types by how they measure size: type I is ordinary matrices, type II has a trace but no smallest pieces, and type III has no trace at all. The algebra of a region of space in field theory is type III, so it has no absolute entropy. Adding a clock and keeping only what is invariant under shifting it produces a crossed product, which can be type II. The previous post explains this in more detail.

A.5 Mutual information between a clock and a label

Suppose a label takes the values “inside” and “outside” with equal probability, and a clock energy is measured. If the clock energy distribution is the same in both cases, the energy tells you nothing about the label, and the mutual information is zero. If the distributions are separated, as when “inside” always gives a negative energy and “outside” always a positive one, the energy reveals the label completely, and the mutual information equals the entropy of the label, \$\ln 2\$ for a fair coin. In between, it is the amount by which knowing the energy reduces your uncertainty about the label.

References


  1. Juan Maldacena, The Large N Limit of Superconformal Field Theories and Supergravity, Advances in Theoretical and Mathematical Physics 2, 1998, pp. 231-252. ↩︎

  2. Edward Witten, Gravity and the crossed product, 2021, https://arxiv.org/abs/2112.12828. ↩︎

  3. Venkatesa Chandrasekaran, Roberto Longo, Geoff Penington and Edward Witten, An Algebra of Observables for de Sitter Space, Journal of High Energy Physics 2023(2), 82, https://arxiv.org/abs/2206.10780. ↩︎

  4. Chen and Xu, An algebra for covariant observers in de Sitter space, arXiv:2511.00622, Journal of High Energy Physics 2026(9), 240, https://arxiv.org/abs/2511.00622. ↩︎ ↩︎

  5. Juan Maldacena, lecture at the Arnold Sommerfeld Center, LMU Munich, https://www.asc.physik.lmu.de/activities/lectures/archiv/thirty-first_series/maldacena_2/index.html. ↩︎

  6. Don N. Page and William K. Wootters, Evolution without evolution: Dynamics described by stationary observables, Physical Review D 27, 1983, pp. 2885-2892. ↩︎

  7. Vittorio Giovannetti, Seth Lloyd and Lorenzo Maccone, Quantum time, Physical Review D 92, 2015, 045033. ↩︎

  8. Philipp A. Höhn, Alexander R. H. Smith and Maximilian P. E. Lock, The Trinity of Relational Quantum Dynamics, Physical Review D 104, 2021, 066001, https://arxiv.org/abs/1912.00033. ↩︎

  9. Julian De Vuyst, Stefan Eccles, Philipp A. Höhn and Josh Kirklin, Crossed products and quantum reference frames: on the observer-dependence of gravitational entropy, Journal of High Energy Physics 2025(7), 063, https://arxiv.org/abs/2412.15502. ↩︎

  10. Julian De Vuyst, Stefan Eccles, Philipp A. Höhn and Josh Kirklin, Gravitational entropy is observer-dependent, 2024, https://arxiv.org/abs/2405.00114. ↩︎

  11. Ahmad et al., Quantum Reference Frames from Top-Down Crossed Products, 2024, https://arxiv.org/abs/2405.13884. ↩︎ ↩︎

  12. Christopher J. Fewster, David Janssen, Leon D. Loveridge, Kasia Rejzner and James Waldron, Quantum reference frames, measurement schemes and the type of local algebras in quantum field theory, 2024. ↩︎

  13. Geoff Penington and Edward Witten, Algebras and States in JT Gravity, 2023, https://arxiv.org/abs/2301.07257. ↩︎

  14. Chang-Han Chen and Geoff Penington, A clock is just a way to tell the time: gravitational algebras in cosmological spacetimes, 2024, https://arxiv.org/abs/2406.02116. ↩︎

  15. Sam Leutheusser and Hong Liu, Emergent times in holographic duality, 2021, https://arxiv.org/abs/2112.12156. ↩︎

  16. Esteban Castro-Ruiz, Flaminia Giacomini and Časlav Brukner, Entanglement of quantum clocks through gravity, Proceedings of the National Academy of Sciences 114(12), 2017, pp. E2303-E2312. ↩︎

  17. Flaminia Giacomini, Esteban Castro-Ruiz and Časlav Brukner, Quantum mechanics and the covariance of physical laws in quantum reference frames, Nature Communications 10, 2019, 494. ↩︎ ↩︎

  18. Gilad Gour and Robert W. Spekkens, The resource theory of quantum reference frames: manipulations and monotones, New Journal of Physics 10, 2008, 033023. ↩︎

  19. Iman Marvian and Robert W. Spekkens, Extending Noether’s theorem by quantifying the asymmetry of quantum states, Nature Communications 5, 2014, 3821. ↩︎

  20. Thomas M. Cover and Joy A. Thomas, Elements of Information Theory, 2nd edition, Wiley, 2006. ↩︎

  21. Edward Witten, A background-independent algebra in quantum gravity, 2023, https://arxiv.org/abs/2308.03663. ↩︎