Skip to main content
Proceedings of the National Academy of Sciences of the United States of America logoLink to Proceedings of the National Academy of Sciences of the United States of America
. 2026 Mar 16;123(12):e2523350123. doi: 10.1073/pnas.2523350123

Rational quantum mechanics: Testing quantum theory with quantum computers

Tim Palmer a,1
PMCID: PMC13012021  PMID: 41838912

Significance

Is there a fundamental reason why quantum computers cannot factor large integers used for encryption today? We introduce a theory of quantum physics based on the notion that the continuum nature of quantum mechanics’ state space approximates something inherently discrete, and argue that the reason for such discreteness is gravity. From this we predict that quantum computers will never break realistic RSA-encrypted messages, for fundamental rather than practical reasons. This predicted breakdown of quantum theory may be falsifiable in a few years. If verified, quantum computers’ biggest impact may be in the development of new finite theories which synthesize quantum and gravitational physics, potentially with commercial benefits (albeit for future generations) eclipsing those derived from quantum mechanics alone.

Keywords: discrete Hilbert Space, quantum information, testing quantum mechanics, foundations of quantum physics, limit to quantum computing

Abstract

Motivated in part by John Wheeler’s assertion that the continuum nature of Hilbert Space conceals the “it-from-bit” information-theoretic character of the quantum wavefunction, a theory of quantum physics (Rational Quantum Mechanics–RaQM) is proposed based on a specific discretization of complex Hilbert Space. The Schrödinger equation is not modified in RaQM, even during measurement. However, the bases in which the quantum state is defined must satisfy certain rational-number constraints. These constraints lead to the notion of finite qubit information capacity Nmax: For any N>Nmax qubit state, there is insufficient information in the N qubits (linearly growing in N) to allocate even one bit to each of all 2N+1−2 continuum degrees of freedom (exponentially growing in N) associated with quantum mechanics/theory (QM, where Nmax=∞). It is proposed that the discretization of Hilbert Space in RaQM is due to gravity, hence QM is the (singular) continuum limit of RaQM at G=0. On this basis, it is estimated that Nmax lies between about 200 and 400 for current qubit technologies, and will never exceed 1,000. While QM and RaQM are experimentally indistinguishable for small numbers of qubits, RaQM predicts that the exponential advantage of quantum algorithms which, like Shor’s, require bases with maximal N-qubit superposition/entanglement, will have saturated at 1,000 perfect qubits. Hence, insofar as a classical computer will never factor a 2,048-bit RSA integer, RaQM predicts that a quantum computer will not either. This predicted breakdown of QM could be testable in less than 5 y.


It has been reported (1) that 2,048-bit RSA integers could be factored in under a week by a quantum computer with less than a million noisy qubits. But could a quantum computer with such capability ever be built? Unitary quantum mechanics/theory (QM) does not itself limit the number of qubits that can be coherently entangled in quantum computers. In addition, environmental decoherence can be minimized by ensuring a quantum computer’s qubits are sufficiently isolated from their environmental, e.g., by positioning them, as has been suggested, in a deep crater on the dark side of the Moon.

However, perhaps there are other constraints on what a quantum computer can in principle achieve. For example, gravitationally induced state collapse would certainly limit multiqubit coherence (2–4). However, for a million entangled quantum-computing qubits, collapse timescales are longer than the age of the universe and therefore irrelevant as a constraint on near-future quantum computing capability. Another source of error might be the indirect dissipative effects of state collapse (5). However, no dissipative effects have been observed (6).

Hence, it would appear that limitations on quantum computing performance are technical and not fundamental. However, here we propose a type of fundamental constraint that is not directly related to state collapse and involves no dissipative effects: the notion that a qubit has a finite information capacity. To be clear about what is being proposed here, consider a quantum mechanical qubit state

|ψ(θ,ϕ)⟩= cosθ2|1⟩+eiϕsinθ2|−1⟩ [1]

in the eigenvector basis (|1⟩,|−1⟩) of some hermitian operator, with 2 continuum degrees of freedom represented by θ,ϕ∈R. Certainly in QM, cosθ and ϕ/2π could be irrational numbers. Indeed, they might be noncomputable numbers whose full description would require the specification of unbounded amounts of information. QM does not itself limit the number-theoretic complexity associated with these two degrees of freedom.

Does this matter? In classical physics, we routinely discretize differential equations to solve them numerically on a digital computer. However, unless these equations are pathological, theorems exist (7) which assert that for fine-enough discretization, the numerical solution will be as close as we like to the continuum solution. In this sense, the differential equations define a smooth limit for the discretized models as the discretization scale goes to zero. However, we cannot assume the continuum limit is smooth if we discretize QM. In particular, one of Lucien Hardy’s axioms for QM (8)—the single most important in his estimation—asserts the continuity of QM’s state space: complex Hilbert Space. This raises the possibility that QM is a singular limit (9) of some fundamentally new theory of quantum physics in which Hilbert Space is discretized (10–12), as the discretization scale goes to zero. Although such a theory may have similar observational and experimental consequences to QM, Hardy’s axiom suggests it will have an extremely different conceptual underpinning to QM (e.g, as general relativity is to Newtonian gravity). The question we consider here is whether such a theory has testable consequences, different from QM. By way of example, consider a model of fluid turbulence based on the viscous Navier–Stokes equations at very large but finite Reynolds number Re, and a model based on the inviscid Euler equations. The two theories generate similar types of turbulent motion. However, in going from Navier–Stokes to Euler, the boundary condition at an aerofoil fundamentally changes, from no slip to no normal motion. This has testable consequences: Aircraft cannot fly in an inviscid fluid!

With this in mind, we will suppose that the continuum nature of complex Hilbert Space is merely an idealization for—and hence an approximation to–a deeper, inherently discrete theory of quantum physics. Such a possibility was advocated by the eminent physicist John Wheeler, who famously coined the aphorism “It from Bit” (i.e., reality from information), writing (13):

  • The familiar probability function or functional, and wave equation or functional wave equation, of standard quantum theory provide mere continuum idealizations and by reason of this circumstance conceal the information-theoretic source from which they derive.

As such, we present a theory of quantum physics “Rational Quantum Mechanics–RaQM” based on a discretization of Hilbert Space, itself based on a discretization of the Riemann Sphere of (extended) complex numbers. Specifically, in RaQM, permitted qubit bases, i.e., bases where |ψ⟩ in [1] is mathematically defined, must be associated with rational values of cosθ and ϕ/2π (more precise restrictions are defined below). Of course, we cannot assume that Wheeler would have agreed with this particular choice of discretization, however, as discussed below, it does indeed reveal the information-theoretic nature of the qubit state, concealed by QM. As part of this revelation, we define a finite quantity Nmax referred to as Qubit Information Capacity (QIC): When N>Nmax, there is insufficient bitwise information contained in an N-qubit system to allocate even one bit of information to each of the 2N+1−2 Hilbert-space dimensions demanded by QM. In QM itself, Nmax=∞, hence QM’s unbounded QIC poses no fundamental constraint on experimental quantum physics.

A sufficiently large but finite Nmax implies that RaQM is experimentally indistinguishable from QM for systems comprising a small number of qubits. However, due to the exponential growth of Hilbert Space with N, any finite Nmax could constrain the behavior of a quantum system with sufficiently many entangled qubits, providing the entanglement is fully spread across the available state space. The purpose of this paper is to propose an experimental test (which might be achievable in as few as 5 y) of the existence of such a constraint.

But why take RaQM seriously, finiteness issues notwithstanding? As discussed in a companion paper (14), RaQM provides a wholly novel way of framing and hence understanding the so-called “weird” properties of quantum physics, such as complementarity, noncommutativity of observables and violation of Bell’s inequality. This reframing illustrates the singular nature of QM as a limiting case of RaQM. By abandoning continuity even slightly, one can arrive at a theory of quantum physics with radically different mathematical and conceptual properties to QM itself. Not least, by violating the Measurement Independent assumption in Bell’s Theorem (15), not for nominal measurement settings under the full control of the experimenter, but for exact measurement settings which were never under their control anyway (consider a gravitational wave passing through an interferometer at the time of measurement), RaQM can be shown to describe a locally realistic theory of quantum physics.

Information-theoretic considerations notwithstanding, it is proposed that the key physical reason for eschewing the state-space continuum [and hence the infinitesimal (16–18)] is gravity. It is generally accepted that gravitational effects may to a breakdown of the continuum structure of space-time; a fortiori we propose such effects lead to a breakdown of the continuum structure of state space. As such, we claim that gravity should play a fundamental role in the very structure of theories of quantum physics, and hence is not some additional field “to be quantized.” Consistent with this, we use existing models of gravitized QM (19) to provide quantitative estimates of the discretization scale and hence QIC, for different types of qubit.

The specific framework for discretizing Hilbert Space is outlined in Section 1. Based on a constructive discretization of the Riemann Sphere of extended complex numbers described in SI Appendix, an explicitly information-theoretic representation of the N-qubit state in discretized Hilbert Space is presented in Section 2. As discussed in Section 3, state reduction/measurement in discretized Hilbert Space is represented geometrically by a state-space magnification of a Cantor Set, and arithmetically by a chaotic shift map on 2-adic representations of the Cantor Set. By equating the information-theoretic reduction time with the Diósi-Penrose collapse time (2, 4), we arrive in Section 4 at a simple formula for QIC. Nmax for different types of qubit in existence today are estimated and an upper bound of Nmax≈1,000 that will never be exceeded is found. The proposed test of RaQM and hence of the breakdown of QM itself is described in Section 5. Concluding remarks are made in Section 6, where we collect together the assumptions made in developing RaQM.

1. Discretized Hilbert Space

RaQM invokes a restriction to the Dirac-von Neumann axioms of QM. In particular, a qubit state [1] in RaQM, evolved using the Schrödinger equation without modification, is only mathematically defined in bases {|1⟩,|−1⟩} where

cos2θ2=mL∈Q;ϕ2π=nL∈Q. [2]

L∈N is the single most important variable in this paper (a function of the mass/energy of the qubit as defined in Sections 3 and 4), defining the degree of granularity of discretized Hilbert Space; 0≤m,n≤L are whole numbers. Qubits in quantum computers will be found (Section 4) to have L⋙1. By contrast, the classical limit corresponds to “maximal discretization” at L=1. QM (where |ψ⟩ is mathematically defined in the bases of all hermitian operators) corresponds to the singular limit of RaQM at L=∞. As mentioned in the Introduction, a model with finite L leads to a complete reimagining of the conceptual foundations of quantum physics. For example, as discussed below, Born’s Rule is emergent in RaQM and hence not needed as a separate axiom.

By construction, |ψ⟩ is undefined in a basis where the rationality constraint [2] is not satisfied—for example, where cosθ=2cos2(θ/2)−1 and/or ϕ/2π are irrational numbers. Such hypothetical bases are found to be associated ubiquitously with counterfactual measurements which arise when analyzing nonclassical properties of quantum systems such as complementarity, noncommutativity of observables, and entanglement (e.g., when considering the outcome of a simultaneous counterfactual which-way experiment, when in reality one has performed an interferometric experiment on some specific photon) (14).

Eq. 2 is straightforwardly generalized for multiqubit states. In QM, a general N-qubit state can be written

α1|1,…,1,1⟩+α2|1,…,1,−1⟩+ α3|1,…,−1,1⟩++…+α2N|−1,…,−1,−1⟩, [3]

where αi∈C. Eq. 3 can be written in an explicitly normalized form; e.g. for N=3, as

cosθ12|1⟩⏟1×cosθ22|1⟩⏟2×(cosθ42|1⟩+eiϕ4sinθ42|−1⟩⏟3)++eiϕ2sinθ22|−1⟩⏟2×(cosθ52|1⟩+eiϕ5sinθ52|−1⟩⏟3)++eiϕ1sinθ12|−1⟩⏟1×cosθ32|1⟩⏟2×(cosθ62|1⟩+eiϕ6sinθ62|−1⟩⏟3)++eiϕ3sinθ32|−1⟩⏟2×(cosθ72|1⟩+eiϕ7sinθ72|−1⟩⏟3), [4]

where

α1= cos(θ1/2)cos(θ2/2)cos(θ4/2)α2= cos(θ1/2)cos(θ2/2)sin(θ4/2)eiϕ4…α7= sin(θ1/2)sin(θ3/2)sin(θ7/2)ei(ϕ1+ϕ3+ϕ7) [5]

The form [4] is represented schematically in Fig. 1, where the 3 underbraced nested qubits in [4] exhibit 2, 4, and 8 degrees of freedom respectively, totaling 14. Continuing to larger values of N, each new qubit adds 2N extra degrees of freedom to the quantum state, yielding 2+4+8+…+2N=2N+1−2 in total [=2N complex degrees of freedom, less 2 for normalization and global phase as in (3)]. In RaQM, we generalize [2] so that it applies to each of the cos2θi/2 and each of the ϕi/2π.

Fig. 1.

Diagram shows a branching structure with labels from one to three and coordinate pairs theta sub one, phi sub one to theta sub seven, phi sub seven.

A schematic representation of [4] corresponding to [3], explicitly normalized, for N=3. The self-similar generalization of this representation to arbitrary N is straightforward, giving 2+4+8+…2N=2N+1−2 degrees of freedom in total. Here the discretization ansatz [2] is applied to each pair (θi,ϕi).

As suggested by Fig. 1, the form [4] is similar to that for a photon (say) passing through a set of nested beamsplitters. The notion of using multiple beam splitters as a way of testing the author’s model of discretized Hilbert Space is described in ref. 20. The estimates below, e.g., of Nmax≈1,000 seem so large that building a one-off laboratory apparatus appears impractical. On the other hand, piggy-backing on the quantum-tech revolution, building such an apparatus may not be inconceivable in the future. In any case, we argue below that quantum computers (no matter what qubit technology is used) provide a testbed for discretized Hilbert Space, potentially in a few years.

Before continuing, we consider a question whose answer may seem esoteric, but will be central, not only to the mathematical viability of RaQM, but also to the way in which the notion of measurement is described in RaQM. Since rational and irrational numbers can be infinitesimally close, is not RaQM (where states in bases with irrational squared amplitudes and/or phases are undefined) necessarily extremely fine-tuned, and hence not robust to small perturbations? To see that this is not the case, first note that irrational real numbers are limit points of Cauchy sequences of rational numbers with specific respect to the Euclidean metric. However, it is well known that Cauchy sequences of rationals can also be defined using the so-called p-adic metric (21), with respect to which the irrational reals are simply undefined (and hence infinitely far away). The p-adic metric is to fractal geometry as the Euclidean metric is to Euclidean geometry. In particular, the set of 2-adic integers is homeomorphic to the self-similar Cantor Set C (21, 22). With this in mind, when we come to discuss state reduction as a state-space instability (Section 3), it will be done with respect to a magnification of C, corresponding to a shift map on the 2-adic numbers which label pieces of C. All this goes to reinforce the idea that QM is the singular continuum limit of RaQM at L=∞.

2. Information-Theoretic Description of the N-Qubit State in RaQM

As already discussed, discretizing Hilbert Space, even when L⋙1, leads to a very different theoretical description of the quantum wavefunction than in QM. Specifically, the qubit state [1] in a basis satisfying the rationality constraints [2], can be expressed as a finite length-L bit string (23). The key to this result, as described explicitly in SI Appendix, is a discretization of the Riemann Sphere of (extended) complex numbers as permutation/negation operators acting on length-L bit strings. On the discretized Riemann Sphere, i=−1 is not introduced axiomatically as in C, but constructively, as a permutation/negation operator acting on 2-bit strings {a1,a2} where ai∈{1,−1}. Specifically, i{a1,a2}={−a2,a1} so that i2{a1,a2}=−{a1,a2}, where −{a1,a2}={−a1,−a2}. The full 3D discretized Riemann Sphere is built up from i, using constructive quaternionic/spinorial generalizations of i, as described in SI Appendix. The discretized Riemann Sphere has nontrivial arithmetic properties which underpin RaQM’s unique physical interpretation (e.g., quantum physics without randomness or spooky action at a distance; see Section 6 and ref. 14).

The discretized Riemann Sphere leads directly to the discretized Bloch Sphere as L-bit strings. Specifically,

|ψ(θ,ϕ)⟩≡{1,1,1,⋯1,1,1,−1,−1,−1,…,−1,−1,−1⏟θ,ϕ}modξ, [6]

where the bits “1” and “−1” are symbolic labels for measurement outcomes (24). Here the fraction of 1s in |ψ⟩ is equal to cos2(θ/2) and ϕ/2π=n/L encodes n cyclic permutations of the L-bit string. ξ is a specific but unknown permutation operator, representing a specific but unknown relationship of a particular quantum system with respect to the rest of the universe. In RaQM, ξ acts as a hidden variable, accounting for what in QM would be described as the inherent randomness of quantum measurement. Just as |ψ(θ,ϕ)⟩ in QM is invariant under a global phase transformation, so |ψ(θ,ϕ)⟩ in [6] is invariant under a generic permutation ξ. As such, |ψ⟩ in [6] describes an unordered ensemble of L possible binary measurement outcomes for identically prepared states, with frequencies consistent with Born’s rule.

Importantly, however, for some specific ξ, RaQM allows the state of a particular quantum system to be defined (something problematic in QM) as the specific bit string

|ψ(θ,ϕ)⟩ξ=ξ{1,1,1,⋯1,1,1,−1,−1,−1,…,−1,−1,−1⏟θ,ϕ} [7]

which can be written as the difference between two complementary integers in standard base-2 representation (for example, {1,−1,−1,1}≡1001.−0110.). As will be discussed elsewhere, such a representation allows one to describe single-particle interference in a Mach-Zehnder interferometer—the key point being that the finite bit-string representation of that single quantum particle is the resource that can describe its wave-like properties.

A system comprising N entangled qubits is represented as N correlated length-L bit strings, where the same permutation ξ applies to each string (consistent with ξ as a global phase in QM). Importantly, N qubits comprise N×L bits of information. For example, with N=3, in some particular rational basis, the 3 length-L bit strings corresponding to [4] can be written as

|ψ⟩≡{1,1,1,⋯1,1,1,−1,−1,−1,…,−1,−1,−1⏟θ1,ϕ1}modξ{1,1,…,1,−1,−1,…,−1⏟θ2,ϕ21,1,…,1,−1,−1…,−1⏟θ3,ϕ3}modξ{1,…,−1⏟θ4,ϕ41,…,−1⏟θ5,ϕ51,…,−1⏟θ6,ϕ61,…,−1⏟θ7,ϕ7}modξ, [8]

where the variables in the underbraces correspond directly to the continuum degrees of freedom in [4] and Fig. 1, but now subject to the rationality constraints [2]. For example, cos2θ2/2 denotes the fraction of 1 bits in the second bit string which correspond to 1 bits in the first bit string, and ϕ2 represents a cyclic permutation of these specific bits in the second bit string. As above, these bit strings can be interpreted as an ensemble of measurement outcomes described by the 8 symbolic labels (1,1,1), (1,1,−1), …, (−1,−1,−1), i.e., as belonging to the set {±1,±2,±3,±4}.

As with a single qubit, it is also possible to interpret a set of bit strings, for a specific ξ, as representing the state

|ψ⟩ξ=ξ{1,1,1,⋯1,1,1,−1,−1,−1,…,−1,−1,−1⏟θ1,ϕ1}ξ{1,1,…,1,−1,−1,…,−1⏟θ2,ϕ21,1,…,1,−1,−1…,−1⏟θ3,ϕ3}ξ{1,…,−1⏟θ4,ϕ41,…,−1⏟θ5,ϕ51,…,−1⏟θ6,ϕ61,…,−1⏟θ7,ϕ7} [9]

of an individual 3-qubit quantum system. These can be represented, as above, as a sum of complementary integers in base-4 (e.g., {2,−3,−1,4}≡2004.−0310.).

At L=∞ (the QM limit), arbitrarily many degrees of freedom can be encoded in the infinitely long bit strings, and there would be no information-theoretic constraint on the number of qubits that could be entangled. However, if L has some finite value, there will be a finite limit to the number of degrees of freedom that can be encoded in these bit strings. In particular, since the total number of the number of bits in N bit strings equals N×L, and the number of degrees of freedom in an N-qubit state in QM is 2N+1−2, then when

2N+1−2>N×L [10]

there simply are not enough bits to allocate even one bit to each QM degree of freedom. By way of illustration, suppose L=16. A 5-qubit state in QM has 62 degrees of freedom. Since 5 length-16 bit strings contain 80 bits in total, there are just enough bits in the 5 bit strings to allocate at least 1 bit to each QM degree of freedom. However, a 6-qubit state in QM has 126 degrees of freedom. Since 6 length-16 bit strings contain 96 bits in total, there are not enough bits to allocate even 1 bit to each degree of freedom. Hence, with L=16, Nmax=5.

When L≫1, [10] implies

Nmax≈ log2L [11]

In the classical limit L=1 there are not enough bits for even Nmax=1: The classical limit is not quantum-like at all! Consistent with this, as shown in SI Appendix, the discretized Riemann Sphere contains no complex structure when L=1.

3. State Reduction/Measurement

Although state reduction plays no direct role in defining QIC (and hence in constraining quantum computers), state reduction in RaQM is analyzed here as it allows us to estimate L numerically (Section 4). In RaQM, state reduction can be defined in two equivalent ways. First, it describes a steady loss of information in a quantum system as that system becomes entangled with the environment. Second, it describes a chaotic state-space instability, separating initially close state-space trajectories until they are classically separated (see e.g., figure 30.19 of ref. 25). To illustrate these processes, we return to the discussion at the end of Section 1 and consider the Cantor Set C (Fig. 2). The iterates of C are themselves labeled by L, where coarser iterations correspond to smaller values of L. This is consistent with the representation of pieces of C by 2-adic numbers—the coarser the iteration of C, the fewer the number of bits needed to represent the corresponding 2-adic integers. We use an initial value L=4 in Fig. 2 purely for illustrative convenience—the generalization to much larger values of L is entirely straightforward. The two pieces of C at L=1 are to be considered attractors to which the qubit state is attracted. The fractal structure of C implies that the corresponding basins of attraction are profoundly intertwined (26).

Fig. 2.

Diagram with levels L equals zero to four showing binary values and arrows indicating relationships between values.

State reduction as both loss of information and as a chaotic state-space instability, describable geometrically as a magnification of a Cantor Set C, or arithmetically as a shift map on the 2-adic representation of a rational number. Here, for illustrative simplicity, we consider a qubit state S [relative to a basis satisfying (2)] initially as a piece of C at the L=4th iteration. The iterate at L=3 shows a magnification of the Left half of the iterate at L=4 containing the state S, and the iterate at L=2 is a magnification of the Right half of the iterate at L=3, and so on. This process is represented arithmetically by the shift map [12]. While S=1010. reduces to 1., the neighboring (2-adically close) state S′=0010. reduces to 0., illustrating the nature of the instability. The hypothetical state represented by “X” to the immediate right of S, but not lying on C, is completely undefined. X might correspond to the qubit state in a counterfactual basis where squared amplitudes or complex phases (in degrees) are irrational numbers. Such bases typically occur when considering nonclassical properties of quantum systems (14).

Here a qubit state S is represented by a small segment of C at the L=4th iterate, or equivalently by a 2-adic integer with 4 bits (Fig. 2). State reduction to the classical limit L=1 occurs in 3 timesteps t∗. During the first timestep, from L=4 to L=3, the left half of the Cantor Set (containing S) is magnified back to C’s original size. During the second timestep, the right half of the L=3rd iterate of C (containing S) is again magnified back to C’s original size. Finally, during the third timestep, the left half of the L=2nd iterate of C is again magnified back to C’s original size. S is now “reduced” to the right-hand piece of C at L=1. Shown is the corresponding evolution of the state S′ initially neighboring S at L=4, to the left-hand piece of C at L=1, illustrating geometrically the nature of the instability.

Equivalently, such state reduction is given by a simple chaotic shift map (27) on the 2-adic representation of S, each step corresponding to a division by 2. In particular, under the 3 timesteps, S is mapped

1010.↦101.↦10.↦1., [12]

consistent with Fig. 2 and highlighting measurement as a loss of information (to the environment). With respect to the 2-adic metric, S′=0010. is 2-adically close to S=1010. and hence neighbors S on S (Fig. 2). Under the 3 timesteps t∗, S′ is mapped to

0010.↦001.↦00.↦0., [13]

This is consistent with S and S′ ‘reducing’ to the two classically distinguishable attractors at the L=1 iterate. The evolution to L=1 is associated with a decrease over 3t∗ from the initial value L=4. For a general L, the collapse time τ* to L=1 is given by

τ*=L−1t∗ [14]

i.e., L−1 individual magnifications/shift maps, terminating at L=1.

But what corresponds to the timestep t∗? This question can be answered by assuming that the physical process associated with Hilbert Space discretization is gravity. It is already widely assumed that the space-time continuum will break down at the Planck Scale. Perhaps, a fortiori a synthesis of quantum and gravitational physics will lead to a breakdown in the Hilbert Space continuum. Since QM has been such experimentally well-tested theory, it is clear that any such breakdown of the Hilbert Space continuum must occur on very fine scales (in the sense that discretization is undetectable for small numbers of qubits). This is consistent with gravity being so weak. However, as stressed above, if the continuum limit is singular, there will surely be some observable consequences of Hilbert Space discretization, no matter how fine the discretization.

With this assumption, we draw on the Diósi-Penrose (2, 4) model of quantum state reduction with collapse timescale

τDP=ħEG [15]

Here, EG is the Diósi-Penrose gravitational self-energy associated with two instances of a mass in QM superposition. As discussed e.g., in ref. 19, EG can be viewed as the energy needed to move one instance of the mass away from the other by a distance b taking account of the gravitational field of the masses. It can be noted that when M equals the Planck mass, τDP=tP, the Planck time.

We now let τ*=τDP and put t∗=tP. Then L is described by the simple formula

L=⌈EPEG⌉, [16]

where EP is the Planck energy (≈109J) and ⌈x⌉ denotes the ceiling function mapping the real number x to the nearest integer greater than x. The identification t∗=tP is consistent with the Diósi-Penrose model, since if EG were very slightly less than EP, so that L=2, reduction to the classical limit would occur in one timestep t∗=tP according to the Diósi-Penrose model. From [16], we see that QM corresponds to the singular (unphysical) limit EG=0, i.e. G=0.

It is important to note that in RaQM, state collapse is not described by modifying the Schrödinger equation. In particular, this state-reduction process does not itself imply any local space-time energy dissipation, a feature of some collapse models (e.g., refs. 3 and 28). Hence RaQM is wholly consistent with a failure to detect the local effects of energy dissipation during measurement (6).

As L decreases to 1, we expect the quantum system to become more and more entangled with its environment. As discussed in Section 2, an N-qubit system in RaQM is described by N partially correlated bit strings, or equivalently as 1 string comprising 2N labels. Hence, the state of the fully entangled system in the classical limit L=1 is describable by a single one of these 2N labels, i.e. as a single integer p (typically ≫0) lying between 0 and 2N. Since these are now single integers, and not strings of integers, in describing how different systems described by different integers p1, p2 etc interact when L=1, the arithmetic would necessarily be Euclidean rather than p-adic. In this sense, the classical limit corresponds not only to L=1, but to a transition from p-adic to Euclidean arithmetic. We will pursue this conclusion elsewhere.

The state reduction process described here plays no direct role in the quantum computing constraint described in this paper. Rather, we “merely” invoke state reduction in order to estimate QIC numerically. To see the irrelevance of state reduction as a direct physical constraint, consider our estimate (Section 4) L=1070 for a quantum dot qubit. After 1 billion years, L will have decreased by ≈3 × 1060 which is a negligible fraction of 1070. By contrast, if the qubit had sufficient mass that L≈1040, still ≫1, then collapse to L=1 would take less than 1ms.

4. Estimating L and Nmax

For a system of mass M with characteristic size R, in QM superposition over a distance b, we make use of formulae for EG already available in the literature. In particular (29):

EG=6GM25R53β2−54β3+16β5if0≤β≤1=6GM25R1−512βifβ≥1, [17]

where β=b/(2R).

1) Consider a typical quantum dot associated with an electron of mass 10−30kg in QM superposition over b=5nm. A question immediately rises as to whether we take R to be the size Re of the electron, 10−15m, or the size Rp of the electron’s QM probability cloud, say 10−9m. We will use this ambiguity to provide an estimate of uncertainty in EG. In both cases β≥1 and, for order-of-magnitude estimates we use the approximate expression

EG≈GM2R≈10−55J [18]

when R=Re and EG≈10−61J when R=Rp. Using [16], then L≈1064≈2212 when R=Re and L≈1070≈2232 when R=Rp. In both cases, we have Nmax≈200, showing that the calculation is not especially sensitive to the choice of R.

2) With t∗=tP, the state-reduction model for discretized Hilbert Space necessarily incorporates the speed of light c. We will take advantage of this to estimate L for photonic qubits in a quantum computer, in superposition over a distance given by some fraction β of the photon wavelength λ. Although the Diósi-Penrose model was developed within a Newtonian gravitational framework, we assume it will continue to hold in a post-Newtonian framework where gravity couples to energy. For a photon, we let M in [17] denote the relativistic mass h/cλ in the computer’s rest frame. Consider an infrared photon of wavelength λ=103nm, as used in a photonic quantum computer. With β=10−3 say, [17] becomes

EG≈β2Gh2c2λ3≈10−82J [19]

so that L≈1091≈2302. Hence, Nmax≈300, somewhat bigger than for the quantum dot example. To be consistent with special relativity, both L and τ∗ are frame dependent.

3) Consider an ion in an ion-trap where the qubit basis states are based on hyperfine atomic line splitting with an energy difference ΔE≈10−9eV. Writing [19] with β≈1, then

EG≈GΔE3hc5≈10−100J [20]

and hence L≈10109≈2400. Hence Nmax≈400, somewhat higher than the estimates above.

4) Undoubtedly, technological developments in qubit precision will increase Nmax further. However, is there an upper limit to Nmax that technology, no matter how precise, cannot exceed? Consider the energy 10−32eV of a photon with the lowest possible frequency that can be accommodated within the age of the universe (taken as 1011 years), giving a wavelength λ=1025m. Now consider two such ultra-low frequency photons in QM superposition over a distance of one Planck length 10−35m so that β2≈10−120. Using [19], EG≈10−289J and hence L≈10298, giving Nmax≈1,000.

These estimates are based on single qubit mass/energy values. In the limit where the qubits in a quantum computer are spatially well separated, then the relevant EG for the composite system will be equal to that for the individual qubits. On the other hand, if we were to consider the N qubits as a single “conglomerated” mass, this would slightly reduce the estimates of Nmax but not by much. For example, if M↦300M in [18] then EG≈217EG, and the estimate of Nmax only decreases by 17.

This limit does not change the more a quantum system is shielded from its nongravitational environment; positioning a quantum computer in a deep crater on the dark side of the moon will not help to overcome the constraints of a finite Nmax. By the Principle of Equivalence, a quantum computer will always be gravitationally coupled to the rest of the universe.

5. An Experimental Test of RaQM

From the discussion above, and the values obtained for L, RaQM will be experimentally indistinguishable from QM for small numbers of entangled qubits, and indeed for large numbers N>Nmax qubits where superposition/entanglement is restricted to a low-dimensional subspace of the full 2N+1−2 Hilbert Space dimension. On the other hand RaQM will be experimentally distinguishable from QM for N>Nmax for quantum systems where superposition/entanglement is maximally distributed across the full 2N+1−2 dimensions of Hilbert Space.

The Quantum Fourier Transform is such an algorithm (30). Hence a candidate for testing RaQM (and hence the breakdown of QM) is Shor’s algorithm. For example, the quantum circuit described in ref. 31, implementing Shor’s algorithm, needs N qubits to factor an N−1-bit semiprime. Here N=2,049 exceeds even our largest conceivable estimate of Nmax. Hence, insofar as it is impossible to factor a 2,048-bit integer (within the age of the universe) using a classical computer, RaQM predicts it will be impossible to factor such an integer with a quantum computer. Indeed, according to estimates described in Section 4, a quantum computer based on current qubit technology will already have ceased having an advantage over a classical computer in factoring a 1,024-bit RSA integer.

In practice, the performance of quantum computers is limited by qubit imperfections and environmental noise. Error-correction algorithms can correct for such factors, utilizing additional qubits. Hence, RaQM does not predict that the performance of noisy qubits will saturate at 1,000 qubits. However, suppose, based on QM, it will be possible to factor a 2,048-bit RSA integer in a quantum computer using some large number N2,048>2,048 of noisy qubits [say a million (1)], then we predict, based on RaQM, that the performance of the computer in factoring such an integer will be no better than a classical computer as the number of qubits approaches N2,048. We therefore have a test of RaQM that could be falsified in a few years if quantum technology roadmaps are accurate.

Perhaps it might be argued that macroscopic quantization contradicts predictions of the breakdown of QM when N>Nmax. For example, a superconducting fluid comprising Avogadro’s number Ne of electrons is very well described by QM (e.g., using the Bardeen, Cooper Schrieffer BCS model) where Ne≫Nmax. On the other hand, while such a system has the potential for each of its component parts to be describable by individual wave functions so that the system is a product state of a very large number of individual subsets, in practice these subsets lock into a single collective wave function, having, for example, the same phase. In this way, one can talk about the phase of the system’s wave function independent of the individual electrons. In the BCS model, therefore, the system evolves in a very small subset of the available Hilbert Space. That is to say, unlike the Quantum Fourier Transform, the BCS model does not involve bases in which the system’s state is maximally spread across the available Hilbert Space. A classical analogy of the point being made here might be the nonlinear synchronization of a large number No of oscillators. The fact that the oscillators are synchronized means that the phase portrait of the system can be described in a very much smaller dimensional space than that of the Cartesian product of all No individual oscillators. As such, macroscopic quantization does not contradict RaQM’s predictions.

QM is not predicted to break down “suddenly’ at N=Nmax qubits. For N<Nmax, then NL/(2N+1−2) bits can be allocated to each Hilbert Space dimension. For example, with Nmax≫1 and N=Nmax−1, then 21 qubits can be allocated to each Hilbert State dimension. When N is 5 qubits short of Nmax, then 25=32 bits can be allocated to each Hilbert Space dimension, and so on. Hence, if the computer can be completely shielded from environmental noise, the output from the computer (in factoring integers with Shor’s algorithm) will appear to be subject to an increasingly strong type of discretized “Shot” noise as N approaches Nmax.

6. Conclusions and Discussion

A theory, Rational Quantum Mechanics (RaQM), of quantum physics has been outlined. In RaQM, the complex Hilbert state of an entangled N-qubit system, evolved using the unmodified Schrödinger equation, is only mathematically defined in bases where squared amplitudes and complex phases (in degrees) are rational numbers of the form m/L where L∈N and m is a whole number with m≤L. Using a constructive discretization of the Riemann Sphere of complex numbers, the complex Hilbert state in such a basis can be described as N correlated length-L bit strings, where L is a fundamental variable of RaQM describing the degree of granularity of Hilbert Space. QM is itself the singular continuum limit of RaQM at L=∞.

There are a number of compelling reasons for positing such a theory, only some of which are discussed in this paper. First, as discussed, it allows the information theoretic nature of the quantum system to be described explicitly (13). Second, as also discussed, it provides an explicit description of state reduction to the classical limit L=1 in terms of a chaotic instability which can be described both geometrically and arithmetically. Third, not discussed here but see the companion paper (14), see also below, discrete Hilbert Space provides interpretations of inherently quantum properties, such as complementarity, noncommutativity of observables, and entanglement (and hence EPR/Bell nonlocality). Finally, as discussed, discretization provides a route for incorporating the effects of gravity in quantum physics, not as some field to be quantized by independently derived quantization rules, but as an inherent ingredient of such quantization rules.

Based on RaQM, we predict a breakdown of QM that may be experimentally falsifiable in a few years using quantum computers. The assumptions made in making this prediction are summarized here:

  • RaQM assumes the discretization ansatz [2] for a single qubit, thus introducing the finite discretization parameter L, also named Qubit Information Capacity (QIC).

  • RaQM assumes a straightforward generalization for multiqubit states, implying that the QIC of N entangled qubits equals N×L. Hence, for given L, QIC grows linearly with N. Since in QM the Hilbert space dimension of N entangled qubits equals 2N+1−2 and hence grows exponentially with N, there must exist a maximum value Nmax≈ log2L of N, above which it is not possible to allocate even one bit of information to each QM degree of freedom. At this point, a quantum algorithm such as Shor’s, which utilizes Hilbert Space maximally, can no longer continue to have an advantage over a corresponding classical algorithm.

  • We estimate QIC by studying state reduction. It is assumed that state reduction is associated with a chaotic state-space instability whereby initially nearby states diverge exponentially until they have separated sufficiently to be classically distinguishable. This process corresponds to a decrease in L by a value of 1 per timestep t∗ until the classical limit at L=1 is reached. Physically, we can suppose that this reduction in information is lost to the environment as the quantum system decoheres.

  • It is assumed that gravity sets the discretization scale L, so that the state reduction timescale becomes τ∗=(L−1)tP, where tP is the Planck time. Consistent with this, τ∗=τDP, the Diósi-Penrose collapse timescale, so that L=⌈EP/EG⌉, where EP is the Planck energy, EG is the Diósi-Penrose gravitational self-energy difference, and ⌈…⌉ is the ceiling function of number theory.

  • Existing formulae for τDP are used to estimate L for a quantum dot qubit in a quantum computer, giving Nmax≈200.

  • It is assumed that the Diosi-Penrose formula can be extended to describe photonic qubits by replacing M with the relativistic mass–energy h/cλ in the computer’s rest frame. This leads to a value Nmax≈300 for a photonic qubit, and Nmax≈400 for an ion trap qubit based on hyperfine line splitting.

  • Considering a photon with frequency given by the inverse age of the universe, put into phase superposition over one Planck length, we show that Nmax can never exceed 1,000 and hence 2,048-bit RSA integers will never be factored.

The notion that there may be an information-theoretic limitation to the capability of a quantum computer is not entirely new (e.g., ref. 32). In particular Davies (33) has argued, using Lloyd’s (34) estimate Iuniverse≈10122 of bits of information (in bosons, fermions, and gravitational degrees of freedom) encoded in the universe, that the Hilbert Space dimension of a quantum state which entangles more than NDavies= log2Iuniverse qubits is larger than Iuniverse. Davies suggests that this may signal a fundamental physical limit to quantum computation.

Whether one is convinced by a cosmological argument of this type or not, the value NDavies≈400 is remarkably similar to the ones deduced here, made without explicit reference to the rest of the universe. Is this a coincidence? The author thinks not. The original motivation for developing a model of quantum physics based on discretized Hilbert Space (23) was as a way to understand the violation of Bell’s inequality. For example, the arithmetic properties of the discretization of the Riemann Sphere (SI Appendix) imply that the Measurement Independence postulate (15) in Bell’s Theorem is violated, not for nominal measurement settings that are always under the full control of experimenters, but for exact measurement settings that were never under their control anyway. Hence, as discussed in ref. 14, the violation of Bell’s inequality need not imply abandonment of local realism. However, it must imply some notion of holism. Indeed, one can take inspiration from the development of general relativity, motivated by Mach’s holistic but not nonlocal principle that inertia “here” is due to mass “there.” The author’s Invariant Set Postulate (35), that the universe be considered a chaotic dynamical system evolving precisely on a dynamically invariant measure-zero fractal, is also an example of a holistic but not EPR/Bell nonlocal concept—and is consistent with the p-adic state representations discussed in this paper. In this regard, the close numerical agreement between Nmax and NDavies may well be evidence of the holistic nature of the laws of physics at their most fundamental, as discussed above. This surely must be important when attempting to synthesize quantum and gravitational physics.

Much of the development of quantum computers has been motivated by their potential commercial applications. If the proposed test does indeed support RaQM, it will stymie the development of such applications. On the other hand, if quantum computers can aid the development of an improved understanding of the fundamental laws of physics, not least in helping to synthesize quantum and gravitational physics, then currently unforeseeable commercial opportunities may arise for future generations.

Supplementary Material

Appendix 01 (PDF)

Acknowledgments

My thanks to Simon Benjamin, Stephen Blundell, Paul Davies, Jonte Hance, Chris Heunen, Sabine Hossenfelder, Richard Howl, Roger Penrose, Felix Tennie, and the PNAS reviewers for extremely helpful comments and inputs. My thanks also to Brendan Palmer for assistance with the graphics in the SI. The research described in this paper was supported by the Leverhulme Trust Grant Number EM-2023-060.

Author contributions

T.P. designed research; performed research; and wrote the paper.

Competing interests

The author declares no competing interest.

Footnotes

Reviewers: I.F., University of Southampton, UK; N.G., University of Geneva; L.H., Perimeter Institute; and S.H., Michigan State University.

Data, Materials, and Software Availability

There are no data underlying this work.

Supporting Information

References

  • 1.C. Gidney, How to factor 2048 bit RSA integers with less than a million noisy qubits. arXiv [Preprint] (2025). http://arxiv.org/abs/2505.15917 (Accessed 1 August 2025).
  • 2.Diósi L., Models for universal reduction of macroscopic quantum fluctuations. Phys. Rev. A 40, 1165–1174 (1989). [DOI] [PubMed] [Google Scholar]
  • 3.Ghirardi G., Rimini A., Weber T., Continuous-spontaneous-reduction model involving gravity. Phys. Rev. A A42, 1057–1064 (1990). [DOI] [PubMed] [Google Scholar]
  • 4.Penrose R., On gravity’s role in quantum state reduction. Gen. Relativ. Gravit. 28, 581–600 (1996). [Google Scholar]
  • 5.Vischi M., Ferialdi L., Trombettoni A., Bassi A., Possible limits on superconducting quantum computers from spontaneous wave-function collapse models. Phys. Rev. B 106, 174506 (2022). [Google Scholar]
  • 6.Donaldi S., Underground test of gravity related wave-function collapse. Nature 17, 74–78 (2021). [Google Scholar]
  • 7.Strogatz S., Nonlinear Dynamics and Chaos (Westview Press, 2000). [Google Scholar]
  • 8.L. Hardy, “Why is nature described by quantum theory?” in Science and Ultimate Reality: Quantum Theory, Cosmology and Complexity, J. D. Barrow, P. C. W. Davies, C. L. Harper Jr., Eds. (Cambridge University Press, 2004).
  • 9.Berry M., Singular limits. Phys. Today 55, 10–11 (2002). [Google Scholar]
  • 10.Buniy R., Hsu S., Zee A., Is Hilbert space discrete? Phys. Lett. B 630, 68–72 (2005). [Google Scholar]
  • 11.Hsu S., Discrete Hilbert Space, the born rule and quantum gravity. Mod. Phys. Lett A36, 2150013 (2021). [Google Scholar]
  • 12.Carroll S., Completely discretised, finite quantum mechanics. Found. Phys. 53, 90 (2013). [Google Scholar]
  • 13.J. Wheeler, “Information, physics, quantum: The search for links” in 3rd International Symposium Foundations of Quantum Mechanics, Tokyo (1989), pp. 354–368.
  • 14.T. Palmer, Solving the mysteries of quantum mechanics: why nature abhors a continuum. arXiv [Preprint] (2026). http://arxiv.org/abs/2602.16382 (Accessed 19 February 2026).
  • 15.Hossenfelder S., Palmer T., Rethinking superdeterminism. Front. Phys. 8, 139 (2020). [Google Scholar]
  • 16.G ’tHooft, Models on the boundary between classical and quantum mechanics. Philos. Trans. A A373, 20140236 (2015). [DOI] [PubMed] [Google Scholar]
  • 17.Ellis G., Meissner K. A., Nicolai H., The physics of infinity. Nature 14, 770–772 (2018). [Google Scholar]
  • 18.Gisin N., Indeterminism in physics, classical chaos and Bohmian mechanics: Are real numbers really real? Erkenntnis 86, 1469–1481 (2021). [Google Scholar]
  • 19.Penrose R., On the gravitization of quantum mechanics 1: Quantum state reduction. Found. Phys. 44, 557–575 (2014). [Google Scholar]
  • 20.Hance J. R., Palmer T. N., Rarity J., Experimental tests of invariant set theory. Phys. Scr. 100, 065123 (2025). [Google Scholar]
  • 21.Katok S., P-Adic Analysis Compared with Real (American Mathematical Society, 2007). [Google Scholar]
  • 22.Robert A. M., A Course in P-Adic Analysis (Springer ISBN 0-387-98660-3, 2000). [Google Scholar]
  • 23.T. Palmer, Discretization of the Bloch sphere, fractal invariant sets and Bell’s theorem. Proc. R. Soc. 476, 20190350 (2020), 10.1098/rspa.2019.0350. [DOI] [PMC free article] [PubMed]
  • 24.Schwinger J., Quantum Mechanics: Symbolism of Atomic Measurements (Springer, 2001). [Google Scholar]
  • 25.Penrose R., The Road to Reality: A Complete Guide to the Laws of the Universe (Jonathan Cape, London, 2004). [Google Scholar]
  • 26.Palmer T., A local deterministic model of quantum spin measurement. Proc. R. Soc. A451, 585–608 (1995). [Google Scholar]
  • 27.Woodcock C., Smart N. P., P-adic chaos and random number generation. Exp. Math. 7, 333–342 (1998). [Google Scholar]
  • 28.Pearle P., Reduction of the state vector by a nonlinear Schrödinger equation. Phys. Rev. D 13, 857–868 (1976). [Google Scholar]
  • 29.Howl R., Penrose R., Fuentes I., Exploring the unification of quantum theory and general relativity with a Bose-Einstein condensate. New J. Phys. 21, 043047 (2019). [Google Scholar]
  • 30.Tyson J., Operator-Schmidt decomposition of the quantum Fourier transform on c1n⊗c2n. J. Phys. A. 36, 6813 (2003). [Google Scholar]
  • 31.D. Willsch et al., The state of factoring on quantum computers. arXiv [Preprint] (2024). http://arxiv.org/abs/2410.14397 (Accessed 1 August 2025).
  • 32.G. ’tHooft, The Cellular Automaton Interpretation of Quantum Mechanics (Springer, 2016).
  • 33.Davies P., The implications of a cosmological information bound for complexity, quantum information and the nature of physical law. Fluctuation Noise Lett. 07, C37–C50 (2007). [Google Scholar]
  • 34.Lloyd S., Computational capacity of the universe. Phys. Rev. Lett 88, 237901 (2002). [DOI] [PubMed] [Google Scholar]
  • 35.Palmer T., The invariant set postulate: A new geometric framework for the foundations of quantum theory and the role played by gravity. Proc. R. Soc. A465, 3165–3185 (2009). [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Appendix 01 (PDF)

Data Availability Statement

There are no data underlying this work.


Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences

RESOURCES