The Axiom of Algorithmic Closure: a proposed law of physics
Dr. Ilan Kreitmann · Ra’anana · November 2026
No. No algorithm executed on a quantum computer has ever delivered the answer to a pure quantum real-world task, and none will. This article states the reason as a physical axiom, the Axiom of Algorithmic Closure: whatever an algorithm brings out of a quantum computer is either reachable by classical means or carries no verified meaning about any real physical system. The quantum content that exceeds classical reach stays closed inside the device.
The axiom is not derived; it is postulated, as every axiom is. It earns its place on three grounds. It rests on proven theorems at its foundation. It is consistent with every published experiment to date, including the strongest advantage claims of 2025 and 2026. And it names the single experiment that would refute it. A law of physics that cannot fail is not a law. This one can, and so far it has not.
1. The category, defined precisely
An algorithm is a finite classical description of operations, written, compiled, optimised and scheduled by classical machines, followed by a measurement whose output is a string of classical bits. Shor, Grover, HHL, VQE, QAOA, quantum singular value transformation, Quantum Echoes: every one of them begins as classical text and ends as classical bits. The quantum computer sits between two classical interfaces. That double interface is the object of this article.
A pure quantum real-world task has four properties at once. Its object is a real physical quantum system: a named molecule, material, catalyst or nucleus, not a model Hamiltonian chosen for being hard. Its answer is a quantum-native property of that system. It is needed at a size and precision no classical method reaches. And the answer is delivered as a usable result and verified independently of the device that produced it.
Two things are excluded by definition. Benchmarks are excluded: random circuits and sampling problems whose only meaning is that they are hard. Analog quantum simulation is excluded too: a cold-atom lattice that reproduces the physics of a material is not running an algorithm. It is nature computing itself, with no classical description at the input and no algorithmic readout at the output. That exclusion is not a loophole. It is the point. The axiom places the obstacle exactly where the algorithm enters.
2. Wall one: the answer cannot be read out
This wall is a theorem, not an opinion. The Holevo bound limits what n qubits can deliver to the classical world to at most n classical bits per measurement. The state that holds the computation lives in a space of dimension d = 2ⁿ. To recover that state, which is precisely the quantum-native answer, requires full state tomography, and Haah, Harrow, Ji, Wu and Yu proved in 2016 that this needs on the order of d²/δ copies of the state for fidelity 1 − δ, with a matching lower bound up to logarithmic factors. With d = 2ⁿ, the cost is exponential in the number of qubits. There is no engineering escape from a lower bound.
The consequence is decisive. A quantum computer can hold the full answer to a quantum problem, and no algorithm can hand it over. What reaches the world is a polynomial handful of statistics: expectation values, samples, a few correlators. The quantum computation is exponentially rich inside and polynomially thin outside. Gottesman–Knill sharpens the picture further: circuits built only from Clifford gates are efficiently simulable on a classical computer, so whatever quantum resource exists is confined to the non-Clifford part of the circuit. The resource is real. Its export is rationed by physics.
3. Wall two: what can be read out, classical methods capture
If only a thin slice of the computation reaches the world, the question becomes whether that slice is beyond classical reach. The published record answers it, claim after claim.
IBM, 2023. The 127-qubit “utility” experiment in Nature measured expectation values of a kicked Ising model and presented them as beyond brute-force classical computation. Within weeks, Begušić and Chan reproduced the observables with sparse Pauli dynamics on a single laptop core, orders of magnitude faster than the quantum run. Tindall, Fishman, Stoudenmire and Sels produced a tensor-network simulation more accurate than the device in the verifiable regime. Later work extended classical simulation of the same experiment to 433 and 1,121 qubits.
Google, 2019. The random circuit sampling claim of 10,000 classical years was cut to days and then further by improved classical algorithms.
D-Wave, 2025–2026. D-Wave’s Science paper claimed spin-glass dynamics beyond classical simulation on its Advantage2 annealer. EPFL and the Flatiron Institute answered with classical simulations; on 21 May 2026 a Flatiron and Boston University team published in Science a belief-propagation tensor-network method reproducing the dynamics. D-Wave disputes the scope and points to full-state and fourth-order observables the classical method did not compute. Note what is being argued about: observables of a benchmark lattice, not a property of any real material.
Photonic boson sampling, 2026. An April 2026 positive-P phase-space algorithm reproduced the measured statistics of flagship Gaussian boson sampling experiments, concluding they had not achieved unequivocal advantage over optimised classical simulation.
Quantum machine learning. Ewin Tang’s 2018 result dequantized the quantum recommendation-systems algorithm, then considered one of the best candidates for exponential speedup. The framework of Chia, Gilyén, Li, Lin, Tang and Wang (Journal of the ACM, 2022) generalised it to essentially all quantum linear-algebra algorithms in the QRAM input model: principal component analysis, clustering, support vector machines, low-rank regression, semidefinite programming. The speedup was in the input assumptions, not in the quantum mechanics.
The pattern is not a series of accidents. It is Wall one seen from the other side. Because only a few statistics leave the device, those statistics have low informational complexity, and low-complexity outputs are exactly what classical compression, tensor networks, Pauli propagation and sampling methods are built to reach.
4. Wall three: the surviving claims are not real-world tasks
Some claims have not been captured. Every one of them fails the definition in Section 1.
Google Quantum Echoes, October 2025. Willow measured second-order out-of-time-order correlators on 65 qubits in about 2.1 hours against an estimated 3.2 years of classical supercomputer time. A 2026 study confirmed that tensor networks with belief propagation fail on these circuits. The hardness is credible. But the object is a random circuit. The real-world companion, molecular geometry from NMR spin echoes, studied small molecules that are easy to analyse classically, and Google’s own Tom O’Brien said those results are not beyond classical yet. Hard and meaningless; meaningful and classical.
IBM and the University of Chicago, July 2026. 70 logical qubits on 97 physical qubits, 468 T gates, about 15 minutes of runtime, with a fidelity lower bound derived from embedded error detection. The task is doped Clifford sampling: a structured sampling problem, not a scientific or commercial application. The fidelity bound holds under stated noise assumptions and post-selection, and the work was released as a preprint.
IBM with Algorithmiq and Qedma, July 2026. These come closest to physics. Algorithmiq ran 56-qubit circuits of an operator Loschmidt echo in a disordered model, and three classical groups produced conflicting predictions. Conflicting classical predictions do not verify the quantum answer; they leave it unverified. The experiment did not simulate a particular catalyst or battery. Qedma’s Floquet dynamics resisted 700,000 CPU hours on Fugaku and were cross-checked on a Quantinuum machine, which shows that two quantum devices agree with each other, not that either agrees with any real material.
The sharpest finding of this article: in the entire published record, classical hardness and real-world meaning have never appeared in the same output. When the output describes a real physical system, classical methods reach it. When classical methods cannot reach it, the output describes no real physical system. That is not a temporary gap in the literature. It is the signature of a law.
5. Wall four: the flagship algorithm has never run
Shor’s algorithm is the reason the field exists. The largest number factored with it on quantum hardware is 21, a record set in 2012 and still standing; attempts at 35 failed. Meanwhile, a classical GPU simulator at Jülich, using 2,048 GPUs, executed Shor’s algorithm on the 39-bit number 549,755,813,701. The classical simulation of the quantum algorithm is ten orders of magnitude ahead of the quantum hardware running it. Factoring is a classical question, so it does not test the axiom directly. It tests the claim that algorithms are the road to quantum usefulness, and after thirty-two years that road has carried the number 21.
6. The axiom
Axiom of Algorithmic Closure. For every computation that is specified by a finite classical algorithm and returned to the world through finite measurement, the information that reaches the world is closed under classical reach: either it is attainable by classical means at comparable cost, or it carries no verified meaning about any real physical system. The quantum content that exceeds classical reach remains closed inside the device.
Three corollaries follow.
- Readout. An answer that is a quantum state is never delivered. This corollary is already a theorem: Holevo and the tomography lower bound.
- Capture. Every real-world observable extracted from an algorithmic quantum computation will be matched by a classical method. Quantum advantage is a moving frontier that classical algorithms keep reaching, not a line that has been crossed.
- Decoupling. Classical hardness and physical meaning are mutually exclusive properties of the output of an algorithmic quantum computation.
The physical reading is simple. A molecule does not run code. Nature evolves its own quantum states with no classical description at the input and no algorithmic measurement at the output, and it pays no readout tax. An algorithm imposes a classical bottleneck at both ends: classical description in, classical bits out. The axiom says that bottleneck is not an engineering inconvenience but a physical closure, in the same sense that the second law of thermodynamics is not an engineering inconvenience for perpetual-motion machines. Work cannot be extracted from a single heat bath; classical-unreachable meaning cannot be extracted from a quantum computer through an algorithm.
7. Where the axiom stands against established theory
A critical article must evaluate its own thesis by the same standard it applies to others. Three points are decisive.
It is a minority position, and it says so. Mainstream complexity theory conjectures that BQP is strictly larger than BPP, and the fault-tolerance threshold theorem says sufficiently good hardware can run arbitrarily long quantum computations. Tang herself notes that BQP-complete problems cannot be dequantized unless BQP equals BPP. The axiom is a bet against the operational consequence of that conjecture for real-world quantum tasks. It is not a refutation of the conjecture; no one has one.
It has a precedent, and it goes further. Gil Kalai of the Hebrew University has argued since the 2000s that noise sensitivity confines noisy quantum devices to a very low complexity class and blocks good-quality error correction. Kalai says the hardware will fail. The Axiom of Algorithmic Closure says something stronger and narrower: even hardware that works delivers nothing beyond classical reach through an algorithm when the task is a real physical system. July 2026’s logical-qubit results put pressure on Kalai’s version. They do not touch this one, because they are benchmarks.
Factoring is outside its scope. If a fault-tolerant machine factors a 2,048-bit RSA modulus with Shor’s algorithm, that refutes the claim that algorithmic quantum computation is useless; it does not refute the axiom, because factoring is a classical question with a classical answer. The axiom is about pure quantum tasks. That is its strength and its boundary.
8. The experiment that refutes it
The axiom falls the day one publication meets all of the following conditions together:
- A named, real physical system: a specific molecule, catalyst, material or nucleus, not a model chosen for hardness.
- A quantum-native property of that system, of scientific or industrial consequence.
- A stated machine, qubit count, circuit depth, and published job IDs.
- A stated best classical baseline, and a result that classical methods fail to reach after an open public challenge, such as the Quantum Advantage Tracker, over a sustained period.
- Independent verification against nature itself, by laboratory measurement, not merely agreement between two quantum devices.
This standard binds every vendor and every software layer that sits above the hardware, including hardware-agnostic layers such as the author’s own Quantum Polycontextural Computing. No layer is exempt from an axiom about the algorithmic interface, and an article that applied it to others but not to itself would be advocacy, not physics.
Conclusion
Thirty-two years after Shor, the record is unambiguous. The quantum answer cannot be read out, by theorem. What can be read out has been captured classically, claim after claim. What has not been captured describes no real system. And the flagship algorithm has factored 21. The Axiom of Algorithmic Closure states the pattern as a law: an algorithm can run on a quantum computer, but it cannot bring a pure quantum real-world answer out of it. The burden of proof now lies with the experiment in Section 8. Until it is published, the answer to the question in the title is no.
Sources
Haah, Harrow, Ji, Wu, Yu, “Sample-optimal tomography of quantum states,” STOC 2016; IEEE Trans. Inf. Theory 63 (2017). arxiv.org/abs/1508.01797
Kim et al. (IBM), “Evidence for the utility of quantum computing before fault tolerance,” Nature 618, 500 (2023). nature.com/articles/s41586-023-06096-3
Begušić, Chan, “Fast classical simulation of evidence for the utility of quantum computing before fault tolerance” (2023). arxiv.org/abs/2306.16372
Begušić, Gray, Chan, “Fast and converged classical simulations of evidence for the utility of quantum computing” (2023). arxiv.org/abs/2308.05077
Tindall, Fishman, Stoudenmire, Sels, “Efficient tensor network simulation of IBM’s kicked Ising experiment” (2023). arxiv.org/abs/2306.14887
Patra et al., “Efficient tensor network simulation of IBM’s largest quantum processors” (2023). arxiv.org/abs/2309.15642
Tang, “Quantum principal component analysis only achieves an exponential speedup because of its state preparation assumptions.” arxiv.org/abs/1811.00414
Chia, Gilyén, Li, Lin, Tang, Wang, “Sampling-based sublinear low-rank matrix arithmetic framework for dequantizing quantum machine learning,” J. ACM 69(5) (2022). arxiv.org/abs/1910.06151
Tang, Quantum Machine Learning Without Any Quantum, PhD thesis, University of Washington (2023). hdl.handle.net/1773/50751
Google Quantum AI, “A verifiable quantum advantage” (22 Oct 2025). research.google/blog/a-verifiable-quantum-advantage
O. Ezratty, “Understanding Google Echoes” (Oct 2025). oezratty.net/wordpress/2025/understanding-google-echoes
IEEE Spectrum, “Quantum Echoes” (2025). spectrum.ieee.org/quantum-echoes
“Tensor networks with belief propagation: limitations in simulating Google’s Quantum Echoes experiment” (2026). arXiv:2604.15427
IBM Research, “Researchers demonstrate quantum advantage through trusted quantum computation” (30 Jul 2026). ibm.com/quantum/blog/quantum-advantage
PostQuantum, “IBM declares a quantum advantage era. The three papers make different claims” (Jul 2026). postquantum.com/industry-news/ibm-trusted-quantum-advantage-three-papers
The New Stack, on the IBM–Algorithmiq result (Jul 2026). thenewstack.io/ibm-quantum-advantage-verification
Quantum Computing Report, “D-Wave systematically rebuts Flatiron claims” (May 2026). quantumcomputingreport.com
Physics World, “D-Wave Systems claims quantum advantage, but some physicists are not convinced.” physicsworld.com
“Gaussian boson sampling: benchmarking quantum advantage” (Apr 2026). arXiv:2604.12330
Willsch et al., “Large-scale simulation of Shor’s quantum factoring algorithm” (2023). arxiv.org/abs/2308.05047
OpenSSL Conference 2025 talk: the largest number factored with Shor’s algorithm is 21, a record since 2012. cfp.openssl-conference.org
G. Kalai, “The argument against quantum computers” (2019). arxiv.org/abs/1908.02499