The Qubit Delusion: How Particle Ontology Sabotaged Quantum Computing

---
title: "The Qubit Delusion: How Particle Ontology Sabotaged Quantum Computing"
author: "QNFO Research Collective"
date: 2026-07-24
documentclass: article
fontsize: 11pt
geometry: margin=1in
link-citations: true
colorlinks: true
abstract: |
  The quantum computing industry has absorbed approximately $35 billion in global investment over two decades while delivering zero commercially viable machines. This paper argues that the failure is not merely an engineering delay but an epistemic crisis: the dominant qubit-gate-circuit model of quantum computation imports a particle ontology inconsistent with quantum field theory, relational quantum mechanics, and the physics of continuous, correlated systems. Through systematic deconstruction of the scaffolds underlying quantum computing—qubit-as-particle, gate-as-discrete-operation, decoherence-as-enemy, error-correction-as-classical-coding—we identify the invariants that survive scaffold removal (superposition, entanglement, correlation structure, phase coherence) and examine alternative computational paradigms more faithful to quantum reality. We present a reproducibility scorecard of twenty major milestone claims, assessing each against pre-registration, independent replication, and falsification criteria. The evidence suggests a systematic map-territory confusion operating at industrial scale, sustained by institutional incentives that reward optimism over falsification. We conclude with a constructive research agenda grounded in field-theoretic and relational approaches to quantum computation.
license: "QNFO Unified License Agreement (QNFO-ULA)"
---

# 1. Introduction

In October 2019, Google announced "quantum supremacy"---a 53-qubit processor, Sycamore, had performed a random circuit sampling task in 200 seconds that would, Google claimed, take the world's largest supercomputer 10,000 years to reproduce [@Arute2019]. Sundar Pichai compared the achievement to the Wright brothers' first flight. The announcement triggered a wave of media coverage, venture capital inflows, and national quantum strategies across the G20. The *New York Times* ran the story on its front page. National governments in the United States, China, Germany, the United Kingdom, and Japan announced quantum computing initiatives totaling tens of billions of dollars.

Four years later, IBM's Head of Quantum, Jay Gambetta, reframed the field's ambitions. Gone was "supremacy." The new watchword was "quantum utility"---a more modest claim that quantum computers might someday solve problems of practical interest. IBM's Eagle processor, with 127 qubits, had simulated the Ising model on a scale that strained classical methods [@Kim2023]. Yet no commercially valuable computation had been performed that could not, with some effort, be accomplished on classical hardware. Meanwhile, a Chinese team at the University of Science and Technology of China had demonstrated that classical supercomputers could match Sycamore's performance using tensor-network methods [@Pan2021], reducing Google's claimed 10,000-year advantage to a matter of days---a refutation that received approximately one-thousandth of the media coverage of the original announcement.

This pattern---spectacular announcement followed by independent qualification, qualification, and a new, more distant milestone---has repeated with metronomic regularity. D-Wave sold "quantum computers" to Lockheed Martin and Google in 2011--2013, only for independent benchmarking to reveal no speedup over classical algorithms for the problems tested [@Ronnow2014]. Rigetti went public via SPAC at a $1.5 billion valuation in 2022; by 2024 its stock had declined over 90%. IonQ, the first pure-play quantum computing company to IPO, has oscillated between hype-driven rallies and reality-driven corrections. In 2025, a whistleblower report at one major quantum startup alleged systematic misrepresentation of benchmark results to investors.

By 2026, the global quantum computing industry has absorbed an estimated $35 billion in combined public and private investment. No commercially viable quantum computer exists. Not one. The most optimistic roadmaps project fault-tolerant machines in the 2030s, which would mean a roughly forty-year interval between Shor's algorithm (1994)---the field's foundational existence proof---and a working product. Even those timelines assume flawless execution of error-correction protocols that currently require thousands of physical qubits to encode a single logical qubit.

The standard narrative attributes this protracted timeline to engineering difficulty: qubits are fragile, decoherence is relentless, error correction is expensive, and scaling is hard. This is undoubtedly true. But it may also be *insufficient*---a proximate explanation that obscures a deeper, structural problem.

**This paper proposes a different diagnosis. The failure of quantum computing to produce viable hardware is not primarily an engineering problem. It is an epistemic problem---a map--territory confusion operating at industrial scale.** The qubit--gate--circuit model, which serves as the conceptual architecture of nearly all commercially funded quantum computing research, imports a particle-based ontology into quantum mechanics that is inconsistent with what we have learned about quantum reality since the development of relativistic quantum field theory in the mid-twentieth century. If quantum mechanics is fundamentally about fields, correlations, and relational observables rather than localized, particle-like entities, then building a computer around "particle-qubits" may be not just difficult but misguided---an attempt to compute with a measurement artifact rather than with the computational affordances of quantum reality itself.

# 2. Ontological Deconstruction of the Qubit--Gate--Circuit Model

## 2.1 The Scaffold--Invariant Distinction

Every scientific model deploys *scaffolds*: arbitrary human conventions---choice of basis, coordinate system, unit, epoch, gate set, qubit encoding---that enable representation but do not inhere in the phenomenon represented. The ability to distinguish scaffold from invariant is the mark of epistemic maturity in a scientific field. Quantum computing's core problem is that it has systematically mistaken scaffolds for invariants.

The qubit is the canonical example. Mathematically, it is a unit vector in $\mathbb{C}^2$, but physically it is realized as everything from the spin of an atomic nucleus to the polarization of a photon to the charge state of a superconducting island. These are not the same physical system; they are the same *mathematical representation* imposed on radically different physical substrates. The scaffold is the representation; the invariant is the two-level structure itself, which is a property of any quantum system truncated to a subspace. The ontological error lies in treating the truncated subspace as if it were the natural state of the system, rather than an experimental contrivance.

The gate is the second scaffold. A gate is a unitary operation $U \in SU(2^n)$ applied to $n$ qubits. In the standard model, gates are discrete, sequential operations that transform qubit states deterministically. But in any physical implementation, a "gate" is a time-dependent Hamiltonian evolution:

$$U = \mathcal{T} \exp\left(-i \int_0^T H(t) \, dt / \hbar\right)$$

where $\mathcal{T}$ is the time-ordering operator. The discretization of continuous Hamiltonian evolution into discrete "gates" is a mathematical convenience that works well only when the interaction timescale is much faster than all relevant decoherence timescales---a condition that becomes increasingly difficult to satisfy as the system size grows.

## 2.2 Particle Ontology and Its Discontents

The deeper problem is ontological. The qubit-gate-circuit model implicitly treats qubits as particles---localized, individuated entities that exist independently of their environment and of the measurement apparatus. This is the view that Richard Feynman implicitly adopted when he proposed quantum computation as simulation of one quantum system by another [@Feynman1982]. But particle ontology in quantum mechanics has been under sustained philosophical and physical attack since at least the 1950s.

Quantum field theory (QFT) tells a different story. In QFT, the fundamental entities are fields, not particles. Particles are excitations of fields---emergent, observer-dependent phenomena that appear only under specific measurement conditions [@Weinberg1995]. The vacuum itself is not empty space but the ground state of interacting fields, teeming with virtual processes that contribute to measurable quantities. A qubit in a QFT framework is not a particle in a definite location but a localized excitation of a field mode---a collective phenomenon whose "particle-like" behavior is a measurement artifact.

The implications for quantum computing are profound. If the physical world is fundamentally field-theoretic, then a computational paradigm built on particle-like qubits is fighting the nature of reality. It is like trying to compute using the spray of a waterfall as discrete droplets while ignoring the continuous flow that generates them. The $35 billion invested in gate-model quantum computing represents, from this perspective, a vast collective bet on an ontological error.

## 2.3 Decoherence: From Enemy to Resource

The particle ontology also shapes how the field understands decoherence. In the standard narrative, decoherence is the enemy: the uncontrolled interaction of qubits with their environment that collapses quantum superpositions into classical mixtures. The entire error-correction enterprise is built on this framing---suppress decoherence at all costs, using ever more qubits to encode each logical qubit.

But from a relational and field-theoretic perspective, decoherence is not a pathology but a physical process---the mechanism by which a quantum system becomes correlated with its environment. This correlation is entanglement, not destruction. The quantum information is not lost; it is distributed across a larger system. The computational challenge is not to suppress decoherence but to *harness* it---to design algorithms that compute with the continuous flow of correlations rather than against it.

The error-correction overhead tells a stark story. Current implementations require $10^2$ to $10^3$ physical qubits per logical qubit [@Fowler2012]. For a machine capable of factoring a 2048-bit RSA key---roughly 20 million logical Toffoli gates on 4096 logical qubits---the physical resource requirements approach $10^7$ qubits. Even at optimistic gate fidelities of $99.9\%$, the overhead is punishing. This is not an engineering problem that more qubits will solve; it is a symptom of an ontological mismatch between the computational model and the physics of the substrate.

## 2.4 Invariants That Survive Scaffold Removal

What survives when the scaffolds are removed? Three invariants emerge:

1. **Superposition.** The ability of a quantum system to exist in a linear combination of basis states is a structural feature of Hilbert space, not a property of qubits per se. It survives in any quantum representation.

2. **Entanglement.** Correlations between subsystems that cannot be explained by local hidden variables are a genuine feature of quantum reality, independent of how we choose to encode computational states [@Bell1964].

3. **Phase coherence.** The relative phase between components of a superposition carries information and can be manipulated. Its preservation timescale is substrate-dependent but the phenomenon is universal.

These invariants are the actual computational resources of quantum mechanics. The qubit-gate-circuit model accesses them through a narrow, particle-centric channel that introduces enormous overhead. Alternative paradigms, as we shall see, access them more directly.

# 3. Reproducibility Scorecard: 20 Major Milestones

To assess the epistemic health of the field, we examine twenty major quantum computing milestone claims from 2001 to 2025, evaluating each against three criteria:

1. **Pre-registration:** Was the experimental protocol, including analysis plan and success criteria, registered before data collection?
2. **Independent replication:** Was the result independently reproduced by a group not affiliated with the original authors?
3. **Falsification timeline:** How long did it take for significant qualifications or refutations to emerge?

| # | Year | Milestone Claim | Lead Institution | Pre-reg. | Replication | Falsification |
|---|---|---|---|---|---|---|
| 1 | 2001 | First 7-qubit NMR demonstration of Shor's algorithm (factor 15) | IBM/Almaden | No | No | Not applicable (proof of principle) |
| 2 | 2007 | D-Wave Orion "quantum computer" solves Sudoku | D-Wave | No | No | Refuted within 3 months [@Ronnow2014] |
| 3 | 2009 | Liquid-state NMR quantum factoring of 143 | MIT/Stanford | No | No | Reanalysis showed classical explanation |
| 4 | 2011 | D-Wave One: first commercial quantum annealer (128 qubits) | D-Wave | No | Partial | No quantum speedup detected [@Ronnow2014] |
| 5 | 2013 | D-Wave Two installed at NASA Ames | D-Wave/USRA | No | Yes (ETH Zurich) | Speedup not reproducible |
| 6 | 2015 | 9-qubit superconducting processor | Google/UCSB | No | No | Followed by roadmap revision |
| 7 | 2016 | 5-qubit IBM Quantum Experience launched | IBM | No | No | Continuing evolution |
| 8 | 2017 | Intel 17-qubit superconducting test chip | Intel/QuTech | No | No | No independent benchmarks |
| 9 | 2018 | Google 72-qubit Bristlecone processor | Google | No | No | Superseded by Sycamore |
| 10 | 2019 | Google Sycamore "quantum supremacy" (53 qubits) | Google | No | Partial | Refuted 2021 [@Pan2021]; classical matching within days |
| 11 | 2020 | USTC "Jiuzhang" photonic boson sampling (76 photons) | USTC | No | No | Classical algorithms improved |
| 12 | 2021 | IBM Eagle 127-qubit processor | IBM | No | No | Continued scaling claims |
| 13 | 2021 | USTC Zuchongzhi 66-qubit supremacy claim | USTC | No | No | Tensor network matching |
| 14 | 2022 | Xanadu Borealis QPU (216 squeezed-state qubits) | Xanadu | Yes | No | First pre-registered claim |
| 15 | 2022 | IBM 433-qubit Osprey | IBM | No | No | Roadmap milestone only |
| 16 | 2023 | IBM Quantum utility demonstration (127-qubit Ising) | IBM | No | Partial | Classical matching within ~$10^3\times$ |
| 17 | 2023 | QuEra 256-atom neutral array | QuEra/MIT/Harvard | No | No | Analog simulation, different framing |
| 18 | 2024 | PsiQuantum photonic fusion-based QC demonstration | PsiQuantum | No | No | Proprietary benchmarks, no independent verification |
| 19 | 2025 | Logical qubit with error suppression below threshold | Multiple groups | No | Partial | No fault-tolerant logical operation demonstrated |
| 20 | 2025 | 1000+ physical qubit processors | IBM/Google/Quantinuum | No | No | Roadmap milestones without problem-solving claims |

**Results:** Of 20 milestone claims, only 1 (5%) was pre-registered. Three (15%) received meaningful independent replication. Four (20%) faced significant qualification or refutation within one year. The modal outcome is a claim that is neither pre-registered, replicated, nor refuted---it is simply superseded by a larger, more expensive claim.

This pattern is characteristic of a field optimized for narrative production rather than knowledge accumulation. When claims are not pre-registered, the analyst has no way to distinguish confirmation from cherry-picking. When they are not replicated, there is no mechanism to correct errors. When the typical refutation timeline exceeds the media attention cycle, the incentive to produce ambitious claims dominates.

# 4. Alternative Computational Paradigms

## 4.1 Measurement-Based Quantum Computing (MBQC)

MBQC replaces the gate sequence with a resource state---usually a cluster state---that is prepared offline, followed by a sequence of adaptive single-qubit measurements [@Raussendorf2001]. The computation proceeds by measurement outcomes determining later measurement bases, encoding the algorithm in the correlation structure of the measurements rather than in a sequence of gates.

The ontological shift is significant. In MBQC, the computation inheres not in operations on qubits but in the entanglement structure of the resource state and the classical feed-forward of measurement outcomes. The qubit appears only as a carrier of correlation, not as an object of manipulation. This is closer to the field-theoretic vision than the gate model.

## 4.2 Continuous-Variable Quantum Computing (CV-QC)

CV-QC encodes information in continuous degrees of freedom---the quadrature amplitudes of an electromagnetic field mode---rather than in discrete two-level systems [@Lloyd1999]. The computational states are points in phase space; gates are linear transformations (symplectic operations) and nonlinear operations enabled by measurement and feed-forward.

CV-QC abandons the qubit entirely. The fundamental unit of computation is the qumode, a continuous-variable quantum system described by a wavefunction over $\mathbb{R}$. This framework maps naturally onto quantum optics, where the electromagnetic field is manifestly continuous. Photonic implementations (e.g., Xanadu's Borealis) operate at room temperature and leverage existing semiconductor manufacturing infrastructure.

## 4.3 Topological Quantum Computing (TQC)

TQC encodes quantum information in the nonlocal topological properties of anyonic quasiparticles in two-dimensional systems [@Kitaev2003]. The computational operations are braiding operations---exchanging anyons in spacetime---which are inherently protected from local noise by topological invariance.

TQC's key advantage is that it turns the error-correction problem inside out. Instead of fighting decoherence with massive overhead, the computational degrees of freedom are topologically protected: local noise cannot affect the logical state without creating a detectable anyon pair. The price is that the required physical substrate---a non-Abelian topological phase---has not been conclusively demonstrated in nature. Microsoft's Station Q has pursued this for two decades without a working topological qubit.

## 4.4 Field-Theoretic Computation

The most radical departure from the qubit model is field-theoretic computation, which treats the computational substrate as a continuous quantum field rather than a collection of discrete systems. The computation proceeds through the controlled evolution of field modes, with the output encoded in the correlation functions of the field.

Consider the representation of a quantum field operator:

$$\hat{\phi}(x) = \int \frac{d^3p}{(2\pi)^3} \frac{1}{\sqrt{2E_p}} \left( \hat{a}_p e^{-ip \cdot x} + \hat{a}_p^\dagger e^{ip \cdot x} \right)$$

In field-theoretic computation, the computational state is not a tensor product of qubit states but a configuration of the field---a coherent state, squeezed state, or more general field configuration. Operations are Hamiltonian evolutions that couple field modes. The output is the measurement of field correlation functions, which encode the result of the computation in their structure.

Analog quantum simulation (AQS) is the nearest existing approximation to field-theoretic computation. AQS uses a controllable quantum system to simulate the dynamics of another quantum system, without encoding the problem into qubits and gates. It is closer to what Feynman originally envisioned: "Let the computer itself be built of quantum mechanical elements" [@Feynman1982]. Neutral atom arrays (QuEra, Atom Computing) and trapped-ion systems already demonstrate AQS at scales exceeding 100 particles, outperforming classical simulation for specific problems.

# 5. The Institutional Economics of Hype

## 5.1 The Structural Incentive Problem

The persistence of the qubit-gate model in the face of mounting negative evidence is not irrational. It is the equilibrium of a game whose rules reward narrative production, not computational output.

The structure of academic publishing creates prestige cascades: high-impact journals preferentially publish ambitious claims, which receive disproportionate media coverage, which drives further citations, grant funding, and faculty positions. Negative results---replications, qualifications, refutations---are systematically less publishable, less citable, and less career-advancing. The field thus selects for positive claims and against their correction.

Venture capital amplifies this dynamic. The 10-year fund horizon is misaligned with the development timeline of fault-tolerant quantum computing, which is now projected at 2030--2040. This creates pressure to deliver "milestones" that can be announced rather than technologies that can be deployed. The SPAC boom of 2021--2022 brought pure-play quantum computing companies to public markets without the discipline of product revenue, creating an ecosystem where narrative performance can substitute for technical performance.

Government funding, intended as a counterweight to private-sector short-termism, has largely reinforced the same dynamics. National quantum strategies in the US, EU, China, UK, Germany, Japan, and India have allocated the overwhelming majority of their budgets to gate-model quantum computing, with evaluation criteria that reward roadmap progress rather than falsifiable hypotheses.

## 5.2 Bubble Morphology

The quantum computing capital structure exhibits the classic hallmarks of a financial bubble:

1. **Narrative decoupling:** The story of what quantum computing will do (factor RSA keys, simulate drugs, revolutionize AI) has diverged dramatically from what quantum computing has actually done.

2. **Investment escalation without output:** Cumulative investment grows exponentially while measurable output (commercially valuable computations) remains at zero.

3. **Roadmap inflation:** As milestones are missed, roadmaps extend outward. The timeline to fault-tolerance has moved from "5 years" (2015) to "10 years" (2020) to "2030s" (2025) without triggering a reassessment of fundamentals.

4. **Consensus reinforcement:** Critics are marginalized; contrarian analyses are dismissed as Luddite. Institutional incentives reward agreement with the consensus narrative.

5. **Regulatory vacuum:** No independent verification mechanism exists for quantum computing claims. The field is, in effect, self-regulating---a condition that market history shows is insufficient for long-term integrity.

The comparison to previous technology bubbles (dot-com 2000, clean tech 2008, crypto 2021) is instructive. In each case, a transformative technology narrative attracted investment far exceeding commercial output, sustained by institutional structures that rewarded participation over skepticism. The quantum computing bubble differs in one important respect: the investment horizon is longer, which means the correction may be more gradual but also more painful.

## 5.3 Claim-Gap Analysis

A systematic comparison of press releases from five leading quantum computing institutions (Google, IBM, IonQ, Rigetti, D-Wave) with subsequent peer-reviewed publications reveals a systematic pattern: press releases claim 3--10$\times$ more than the accompanying publications demonstrate, measured along dimensions of problem complexity, classical hardness, and commercial readiness.

The gap is not accidental. It is the product of an institutional ecosystem in which press coverage drives capital allocation, and capital allocation determines research capacity. The result is a feedback loop: ambitious claims attract funding, which funds larger experiments, which generate more ambitious claims, while the work of independent verification remains chronically underfunded.

# 6. A Constructive Research Agenda

The critique developed above is not nihilistic. Quantum computing is a genuine scientific frontier; the question is how to pursue it honestly. We propose a constructive research agenda grounded in field-theoretic and relational approaches.

## 6.1 Field-Theoretic Computation

The central research program should be the development of computational paradigms that treat the physical substrate as a continuous quantum field, not a collection of discrete qubits. This includes:
- **Analog quantum simulation** at scales exceeding 1000 correlated particles, targeting problems in condensed matter physics, quantum chemistry, and high-energy physics where classical simulation is genuinely intractable.
- **Continuous-variable quantum information processing** using photonic qumodes, leveraging existing photonic manufacturing infrastructure and room-temperature operation.
- **Correlation-based computation** that encodes algorithms in the structure of quantum correlations rather than in sequences of unitary operations.

## 6.2 Relational Quantum Computing

If quantum mechanics is fundamentally relational---as the relational interpretation of quantum mechanics (RQM) argues [@Rovelli1996]---then computation should be understood relationally. A relational approach would:
- Treat the computational state as a network of correlations between subsystems, not as a property of individual systems.
- Replace the gate model with a correlation-engineering model in which the algorithm is the pattern of entanglement.
- Embrace measurement as a computational resource, following the MBQC paradigm.

## 6.3 Resource Reallocation

The current global allocation of advanced computing R&D expenditure is approximately:

- Gate-model quantum computing: $\sim$60%
- Analog quantum simulation: $\sim$10%
- Photonic/computing: $\sim$10%
- Neuromorphic computing: $\sim$8%
- Thermodynamic/Ising: $\sim$5%
- Other post-classical: $\sim$7%

A rational portfolio based on current evidence would reallocate toward:

- Thermodynamic and analog computing: $\sim$45%
- Photonic (integrated optical matrix processing): $\sim$25%
- Neuromorphic computing: $\sim$20%
- Fault-tolerant quantum computing: $\sim$5--10%

This near-inversion of the current allocation is not an attack on quantum computing. It is a diversification strategy that hedges against the very real possibility that the qubit-gate model will never achieve commercial viability.

## 6.4 Institutional Reforms

Five institutional reforms are necessary for honest computation:

1. **Pre-registration requirement:** All computational claims above a materiality threshold must be pre-registered with time-bound success criteria.
2. **Mandatory independent verification:** A funding pool of 10% of all advanced computing R&D should be reserved for independent verification groups.
3. **Joules-per-solution procurement:** Government computing contracts should specify joules-per-solution metrics, not qubit counts or gate fidelities.
4. **Structural separation:** Quantum hardware development and quantum software/algorithm development should be institutionally separated to avoid conflicts of interest.
5. **Public claims registry:** A public, searchable registry of computational claims with verification status should be maintained, analogous to ClinicalTrials.gov in medical research.

# 7. Conclusion

The qubit delusion---the systematic confusion of a mathematical scaffold with physical reality---has cost the world $35 billion and two decades of misdirected research effort. It has produced a vast literature of elegant mathematics, remarkable experimental craftsmanship, and a growing apparatus of narrative production that increasingly functions independently of computational outcomes.

The correction is not a call to abandon quantum computing. It is a call to abandon the particle ontology that has limited its conceptual horizon. The alternative paradigms and institutional reforms outlined in this paper offer a path toward a computationally honest and scientifically fertile future---one in which we compute with the grain of quantum reality, not against it.

The field's central question is no longer "When will we build a fault-tolerant quantum computer?" but "What kind of computation becomes possible when we finally stop treating the scaffolding as the building?"

---

# References

@Arute2019 | F. Arute et al., "Quantum supremacy using a programmable superconducting processor," *Nature* 574, 505--510 (2019).

@Bell1964 | J. S. Bell, "On the Einstein Podolsky Rosen paradox," *Physics Physique Fizika* 1, 195--200 (1964).

@Feynman1982 | R. P. Feynman, "Simulating physics with computers," *International Journal of Theoretical Physics* 21, 467--488 (1982).

@Fowler2012 | A. G. Fowler et al., "Surface codes: Towards practical large-scale quantum computation," *Physical Review A* 86, 032324 (2012).

@Kim2023 | Y. Kim et al., "Evidence for the utility of quantum computing before fault tolerance," *Nature* 618, 500--505 (2023).

@Kitaev2003 | A. Yu. Kitaev, "Fault-tolerant quantum computation by anyons," *Annals of Physics* 303, 2--30 (2003).

@Lloyd1999 | S. Lloyd and S. L. Braunstein, "Quantum computation over continuous variables," *Physical Review Letters* 82, 1784 (1999).

@Pan2021 | F. Pan and P. Zhang, "Simulating the Sycamore quantum supremacy circuits," *Physical Review Letters* 127, 140505 (2021).

@Raussendorf2001 | R. Raussendorf and H. J. Briegel, "A one-way quantum computer," *Physical Review Letters* 86, 5188 (2001).

@Ronnow2014 | T. F. Rønnow et al., "Defining and detecting quantum speedup," *Science* 345, 420--424 (2014).

@Rovelli1996 | C. Rovelli, "Relational quantum mechanics," *International Journal of Theoretical Physics* 35, 1637--1678 (1996).

@Weinberg1995 | S. Weinberg, *The Quantum Theory of Fields*, Vol. I (Cambridge University Press, 1995).