Oscillating quantum networks reveal where simulation complexity jumps

A mathematical proof demonstrates that vibrating particles can simulate quantum algorithms efficiently before unconstrained cross interactions trigger post-selected computational hardness.

When physical objects vibrate in synchrony, their collective motions can mirror computational problems that would otherwise require extraordinary effort to resolve. A network of pendulums or an array of masses connected by mechanical springs exchanges energy back and forth according to orderly physical laws. In the microscopic domain, particles such as photons or trapped ions behave like harmonic oscillators that follow comparable principles, fluctuating between position and momentum while preserving quantum traits such as vacuum fluctuations and squeezing. For decades, physicists treated these quadratic interactions as comparatively simple because the mathematical equations describing their average positions and momenta decouple cleanly from higher order complexities. Yet beneath that apparent simplicity lies a rich landscape of computational power, where subtle changes in how particles interact can cause a dramatic leap in the resources required to track their evolution.

Understanding where this physical behavior becomes difficult to calculate matters directly to the development of quantum algorithms and physical simulations. If a physical system can be mapped efficiently onto a quantum processor, researchers can predict the behavior of large optical networks, molecular vibrations, and acoustic arrays without tracking every individual particle on a classical computer. Conversely, identifying precisely when a model exceeds standard quantum capabilities reveals the limits of quantum algorithms themselves. The boundary separates physical processes that an ordinary quantum computer can simulate in reasonable time from those that demand hypothetical post-selection capabilities, where an algorithm must discard failed computational branches until an extraordinarily rare measurement outcome appears.

The intuitive explanation of this computational behavior follows a straightforward chain of events. A collection of physical particles begins in a known initial state, where only a modest number of average positions, momenta, or low order correlations have non-zero values. As time passes, physical interactions between these particles transfer momentum and displacement across the network, causing vibrations to ripple through neighboring modes according to linear differential equations. To extract a specific answer at the end of the process, an observer measures whether the accumulated energy or correlation across a designated target group exceeds a set threshold. When the interaction couples momenta to momenta and coordinates to coordinates, the wave of energy travels through the network in a unitary fashion that preserves total energy, allowing each step to be tracked sequentially without runaway amplification. When interactions simultaneously mix coordinates and momenta in an asymmetric manner, the mathematical stability of that wave changes, turning an otherwise manageable quantum tracking problem into an intractable calculation.

In a preprint posted to the arXiv server, Lilith Zschetzsche, Refik Mansuroglu, and Norbert Schuch at the University of Vienna demonstrate that this dividing line in quantum simulation complexity can be formalized with mathematical precision.¹ The authors introduce a broad class of physical systems known as inertially coupled bosons, where particles interact through both spatial stiffness and momentum couplings.¹ The preprint proves that simulating exponentially large networks of these inertially coupled bosons is complete for the complexity class known as BQP, or bounded-error quantum polynomial time.¹ This classification means that an ordinary quantum computer can simulate the physical network efficiently, while the network itself is capable of performing any calculation that a universal quantum machine can execute. Furthermore, the preprint shows that extending the interactions to include general mixed position and momentum terms causes the simulation problem to jump immediately into PostBQP-completeness, matching the power of quantum computers equipped with post-selection.¹

How do inertially coupled particles unify different quantum models?

Inertially coupled bosons unify classical oscillator networks and quantum walks into a single mathematical framework by describing how particles exchange energy through generalized spring constants and shared inertia.¹ In a conventional network of masses and springs, the interaction between two objects depends strictly on the difference between their physical positions.²¹ In the setting explored by Lilith Zschetzsche, Refik Mansuroglu, and Norbert Schuch, the Hamiltonian includes not only coordinate couplings through a symmetric stiffness matrix, but also momentum couplings through a symmetric inverse mass matrix.¹ Off-diagonal elements in this momentum matrix represent inertial couplings, where accelerating one particle exerts a direct dynamical force on the momentum of an adjacent particle, much like the linked pivots of a physical double pendulum.¹

Refik, a researcher at the University of Vienna, clarified the physical nature of these momentum couplings in response to questions from Primary. Refik stated: “I would not interpret this as a literal force being transmitted from one particle to another. An off-diagonal term in the momentum matrix means that the kinetic energy contains a term proportional to (p_jp_k), so the equations of motion couple the momenta of the two particles directly. There are classical analogues of this in constrained mechanical systems, such as a double pendulum, where the kinetic energy is not simply a sum of independent (p_j^2) terms.”^contributed

The authors demonstrate that this formulation encompasses earlier landmark findings in quantum computing as direct subcases. In 2023, Ryan Babbush and colleagues established that simulating networks of classical harmonic oscillators is BQP-complete when tracking kinetic energies across exponential numbers of modes.²¹ In late 2025, Lilith Zschetzsche, Refik Mansuroglu, Andras Molnar, and Norbert Schuch established a direct mathematical equivalence between classical harmonic oscillator networks and continuous-time quantum walks, which describe a single quantum excitation wandering across a graph.¹ The new preprint proves that a continuous-time quantum walk is mathematically identical to an inertially coupled bosonic system where the position stiffness matrix and the momentum stiffness matrix are set exactly equal to one another, operating in the specific sector where exactly one bosonic excitation is present.¹ By broadening this framework, the authors show that any particle-number-conserving quadratic Hamiltonian remains within BQP, even when complex hopping amplitudes are introduced.

Addressing this connection, Refik explained how the equivalence emerges naturally from the underlying operator algebra. Refik told Primary: “The connection becomes apparent when writing the bosonic hopping Hamiltonian in terms of position and momentum operators. For real hopping amplitudes, the Hamiltonian can be written in the form of our inertially coupled Hamiltonian with (A=C). In the sector containing exactly one boson, the resulting dynamics is precisely the continuous-time quantum walk on the corresponding graph. So rather than being a completely different model, the quantum walk appears as a particularly simple sector of the same quadratic bosonic framework.”^contributed

To simulate these expansive networks on a quantum processor without consuming prohibitive memory, the authors employ an amplitude encoding method rooted in the Heisenberg picture of quantum mechanics. For a physical system consisting of an exponential number of bosonic modes, tracking the full quantum wave function would demand an impossible amount of classical or quantum storage. However, because quadratic interactions produce equations of motion that preserve the algebraic structure of canonical observables, the time evolution of all position and momentum products up to a fixed order remains closed.¹ The authors map the expectation values of these observable products directly onto the probability amplitudes of an auxiliary quantum register, an approach closely related to shadow Hamiltonian simulation techniques developed by Rolando D. Somma and colleagues in 2024.¹ For a system with two to the power of m bosonic modes, storing all correlation values up to a constant order requires only a polynomial number of qubits proportional to m. The physical time evolution generated by the stiffness matrices then maps onto a sparse, effective Hamiltonian that acts unitarily on the auxiliary state, enabling standard quantum simulation algorithms to step the system forward using gate counts that scale polynomially with the number of qubits.¹

What causes the sudden jump into post-selected complexity?

The computational complexity jumps abruptly into PostBQP-completeness when quadratic interactions introduce mixed position and momentum terms that break the unitary structure of the compressed equations of motion.¹ In the general quadratic framework, the interaction matrix includes an off-diagonal block that directly pairs the spatial coordinate of one particle with the momentum coordinate of another. While certain symmetric pairings merely generate phase shifts or hopping, more general cross terms introduce squeezing effects that alter the phase space area of the fluctuations.¹ The authors show that when these unconstrained quadratic terms are permitted, the simulation problem can encode decision problems that are complete for PostBQP, an elevated complexity class that corresponds classically to the class PP.

Refik explained the underlying computational mechanism that elevates these mixed terms into post-selection power. Refik said: “The important point is not simply that the equations become non-unitary. The additional quadratic terms allow us to engineer dynamics in which one component of the encoded state is exponentially amplified relative to another. In our construction, this effectively selects the branch corresponding to a particular measurement outcome of a quantum computation, even though that outcome can have exponentially small probability. This is exactly the kind of conditioning that gives post-selection its additional computational power.”^contributed

This transition represents a qualitative shift in how hard it is to predict the behavior of a physical system. For inertially coupled bosons, a quantum computer can simulate the physical dynamics in polynomial time, whereas a classical computer requires exponential time to evaluate the sparse matrix dynamics across exponentially many degrees of freedom. In that regime, quantum hardware provides an exponential speedup over classical simulation. When arbitrary mixed position and momentum couplings are introduced, however, this exponential quantum advantage disappears. Simulating the resulting system becomes PostBQP-complete on a quantum processor, which matches the classical difficulty of simulating the exact same dynamics. In both cases, obtaining the outcome requires computational resources equivalent to searching through an exponential space of trajectories and conditioning on an exponentially improbable result.

The mathematical mechanism driving this leap lies in the properties of the linear differential equations that govern the operator moments.¹ In the BQP-complete setting, the stiffness matrices are positive semidefinite and diagonally dominant, which ensures that the effective Hamiltonian acting on the encoded amplitudes remains strictly Hermitian and unitary.¹ However, Refik noted that diagonal dominance is primarily a physical interpretation rather than a strict mathematical constraint for unitarity. Refik said: “Diagonal dominance is not essential for the unitarity of the encoded Hamiltonian. The relevant condition in our BQP-containment argument is that (A) and (C) are positive semidefinite, which allows us to factor them as (A=BB^ ) and (C=DD^ ). The resulting encoded Hamiltonian is then Hermitian. Diagonal dominance enters when we additionally want to interpret the matrices as arising from ordinary positive spring-like couplings.”

What are the boundaries of these simulation findings?

These theoretical results characterize the asymptotic computational complexity of abstract bosonic networks rather than delivering immediate blueprints for laboratory experiments. The proof relies on an idealized oracle model, in which an algorithm queries the non-zero matrix entries of the interaction graph in unit time, assuming that initial coordinates and stiffness couplings are known to high precision.¹ Real physical apparatuses, such as optical cavities or superconducting resonator arrays, inevitably suffer from experimental noise, dissipation, and calibration drift that could obscure the precise decision thresholds defined in the paper. The paper also focuses on the worst-case complexity of deciding bounded expectation values at a specific final time, which does not guarantee that typical or average physical systems will exhibit the same computational power. Furthermore, the preprint has been posted to the arXiv repository and has not yet completed formal peer review.¹

Refik emphasized the distinction between abstract complexity models and laboratory reality in response to questions about physical implementations. Refik said: “The oracle is part of the complexity-theoretic formulation of the problem; it is not meant to describe how an experimentalist literally accesses a physical system. Our result therefore establishes a statement about the computational complexity of the idealized simulation problem. Real experiments introduce additional issues such as noise, dissipation and finite precision, which are important for determining whether the theoretical scaling can actually be observed experimentally. Those questions are outside the scope of the present work.”^contributed

Whether intermediate interaction classes exist between the BQP-complete models and the PostBQP-hard Hamiltonians remains an open theoretical question.¹ Discussing this gap in the current classification, Refik said: “We do not currently know whether such intermediate families exist. One would need to identify a restricted class of quadratic Hamiltonians for which the simulation problem can be shown to lie strictly beyond BQP while also proving that it is not PostBQP-hard. Our results establish BQP-completeness for several broad subclasses and PostBQP-completeness once sufficiently general quadratic terms are allowed, but they do not classify all possible intermediate extensions.”^contributed

Where do these discoveries lead computational physics?

Classifying the exact computational boundaries of bosonic systems provides researchers with a rigorous map for identifying which physical simulations genuinely benefit from quantum processors. In quantum chemistry and materials science, vibrational modes in large molecular structures are frequently modeled as coupled harmonic oscillators. Knowing that inertially coupled systems reside firmly within the BQP complexity class confirms that quantum simulation algorithms can track large networks of molecular vibrations or phononic metamaterials with exponential advantages over classical numerical integrators, provided the interactions avoid destabilizing position momentum cross terms.

Looking forward, the results direct attention toward the unresolved space between standard quantum efficiency and post-selected hardness. Theorists are now tasked with determining whether every mathematical extension beyond inertially coupled and hopping Hamiltonians instantly triggers PostBQP-hardness, or whether subtle physical constraints can maintain BQP-completeness across a wider variety of mixed interactions. Clarifying that boundary will help computational physicists design robust quantum algorithms for analog systems while steering clear of dynamical regimes where classical and quantum simulation methods alike encounter an impenetrable computational wall.

What this rests on

39 sentences trace to 2 sources and 1 contributor.

All 3Documents 2Records 0Reporting 0People 1

1. 1 LZ Lilith Zschetzsche, Refik Mansuroglu, Norbert Schuch Paper · arxiv.org · 25 Sep 2026 Complexity of Quadratic Bosonic Hamiltonian Simulation: BQP-Completeness and PostBQP-Hardness Preprint · may not have been peer reviewed See the source
2. 2 A arxiv.org Paper · arxiv.org Exponential quantum speedup in simulating coupled classical oscillators Preprint · may not have been peer reviewed See the source
3. 3 R Refik Contributor · interview 30 Sep 2026 · in the article since 5 Oct 2026 Contribution — Refik 5 statements added to this article

Also consulted: 4 pages across 2 sites, none cited.

Article history

1. Published 5 Oct 2026, 13:31
   Assembled by the Primary desk from 2 sources · 1 contributor · 39 cited sentences
2. R Refik, a researcher at the University of Vienna 5 statements added 30 Sep 2026 6 days before publication

   1. Go to the sentence: Refik stated: “I would not interpret this as a literal force being transmitted from one particle to another. An off-diagonal term in the momentum matrix means that the kinetic energy contains a term proportional to (p_jp_k), so the equations of motion couple the momenta of the two particles directly. There are classical analogues of this in constrained mechanical systems, such as a double pendulum, where the kinetic energy is not simply a sum of independent (p_j^2) terms.”↑
   2. Go to the sentence: Refik told Primary: “The connection becomes apparent when writing the bosonic hopping Hamiltonian in terms of position and momentum operators. For real hopping amplitudes, the Hamiltonian can be written in the form of our inertially coupled Hamiltonian with (A=C). In the sector containing exactly one boson, the resulting dynamics is precisely the continuous-time quantum walk on the corresponding graph. So rather than being a completely different model, the quantum walk appears as a particularly simple sector of the same quadratic bosonic framework.”↑
   3. Go to the sentence: Refik said: “The important point is not simply that the equations become non-unitary. The additional quadratic terms allow us to engineer dynamics in which one component of the encoded state is exponentially amplified relative to another. In our construction, this effectively selects the branch corresponding to a particular measurement outcome of a quantum computation, even though that outcome can have exponentially small probability. This is exactly the kind of conditioning that gives post-selection its additional computational power.”↑
   4. Go to the sentence: Refik said: “The oracle is part of the complexity-theoretic formulation of the problem; it is not meant to describe how an experimentalist literally accesses a physical system. Our result therefore establishes a statement about the computational complexity of the idealized simulation problem. Real experiments introduce additional issues such as noise, dissipation and finite precision, which are important for determining whether the theoretical scaling can actually be observed experimentally. Those questions are outside the scope of the present work.”↑
   5. Go to the sentence: Refik said: “We do not currently know whether such intermediate families exist. One would need to identify a restricted class of quadratic Hamiltonians for which the simulation problem can be shown to lie strictly beyond BQP while also proving that it is not PostBQP-hard. Our results establish BQP-completeness for several broad subclasses and PostBQP-completeness once sufficiently general quadratic terms are allowed, but they do not classify all possible intermediate extensions.”↑