How Noise Taught Quantum Computers to Tame Turbulence
09 Jun 2026, Yanjiang
By injecting noise into turbulent dynamics, a new quantum algorithm transforms exponential complexity into logarithmic cost, turning randomness into a computational resource.
Picture a river in flood — not the smooth, glassy surface of a postcard, but the roiling interior where eddies birth smaller eddies and every swirling packet of momentum feels every other. A classical computer that tried to track all those interactions would drown. The number of pairings grows with the square of the number of grid points, and for a turbulence-resolving simulation even a supercomputer chokes. For decades, this has been the grim boundary: simulate a high-dimensional, densely interacting nonlinear system, and you pay an exponential price.
A new quantum algorithm, laid out in a preprint (arXiv:2606.08349) by a team led by Sergiy Zhuk at IBM Quantum — Sergey Bravyi, Adam Byrne, Mykhaylo Zayats, and Zhuk — proposes to redraw that boundary. Their method tackles a broad class of stochastic nonlinear differential equations with all-to-all interactions, and it does so with a cost that climbs only logarithmically with the number of variables. Even more striking, it does not demand the smallness or sparsity assumptions that handcuffed earlier quantum approaches. The algorithm’s secret is counterintuitive: it needs noise to work. Without noise, the quantum advantage flatlines; with just enough randomness, the most punishing equations become tame.
This is a story about why noise, long treated as the enemy of computation, might become a quantum computer’s most trusted ally. To understand why, we have to look inside the machinery.
The Hidden Cost of Predicting Chaos
The equations that govern turbulent flows — the Navier-Stokes equations, their damped Euler cousins — are deterministic. In principle, if you know the velocity field at every point at time zero, you can march forward. In practice, to resolve the cascade of energy from large whorls to microscopic dissipation, you need an astronomical number of variables. A direct numerical simulation of a tiny patch of atmosphere already strains the largest classical clusters.
Quantum computers have been offered as a way out. The dream is that by encoding the velocity field into the amplitudes of qubits, one might exploit quantum parallelism to simulate N-dimensional systems with resources that scale as the logarithm of N. The difficulty has always been nonlinearity. Quantum mechanics is linear; nonlinear classical equations don’t map directly onto the Schrödinger equation without a heavy price. Previous quantum algorithms resorted to linearization tricks, such as the Carleman embedding, but those methods demanded that the nonlinear terms be weak or that the interaction network be sparse — only neighboring variables coupling, never all at once. Real turbulence is neither weak nor sparse. A single eddy tugs on distant parts of the flow, and the nonlinear drag can dominate the dynamics.
The IBM team’s insight was to change the question. Instead of trying to simulate the exact deterministic trajectory, they ask: what if we inject a tiny amount of random white noise into the system and compute the average of a physical observable over many noisy realizations? This is not a fudge. Turbulent flows, after all, are inherently chaotic; the precise microscopic state is unobservable, while statistical quantities — mean velocities, correlation functions — are what experiments measure. The noise acts as a mathematical regulator. It smears the deterministic delta-functions into smooth, well-behaved distributions that the quantum algorithm can handle.
Noise as a Computational Resource
How does the method actually work? Think of the original nonlinear equation for the state vector X(t) in N dimensions. The team adds a tiny random kick to each variable at every instant, turning the deterministic ordinary differential equation into a stochastic differential equation. The quantity they want — the noise-averaged value of a scalar observable u₀(X(t)) — satisfies a deterministic linear partial differential equation called the Kolmogorov backward equation. Linear in the sense of probability amplitudes, not in the underlying dynamics. And here is the first stroke of genius: they embed this Kolmogorov equation into the Hilbert space of N interacting quantum harmonic oscillators. Each oscillator stands in for one variable of the original system.
Adding noise turns a hard nonlinear equation into a problem quantum computers can solve efficiently. That means faster simulations of real-world systems like turbulent flows or financial markets. (Source: arXiv:2606.08349)
Naively, simulating N interacting harmonic oscillators still looks like a classical nightmare — the full Hilbert space grows exponentially with N. However, the team devised a subroutine that simulates this auxiliary oscillator system with a cost that is polylogarithmic in N and linear in the total evolution time. The trick relies on the fact that the initial state of the oscillators is a product of coherent states, and the observable is a simple combination of position and momentum operators. By harnessing techniques from continuous-variable quantum computation, they can track the necessary expectation values without ever representing the full wavefunction.
The result is an algorithm that, for a broad class of “norm-preserving” drifts — a condition satisfied by certain discretizations of the Navier-Stokes and damped Euler equations — delivers rigorous error bounds at an exponential speedup over any known classical method. The more variables, the deeper the quantum advantage. The price? The runtime scales inversely with the noise strength. If the noise is allowed to vanish, the cost blows up. This is not a bug; it is a fundamental trade-off. The noise provides the essential mixing that prevents the quantum simulation from getting snarled in the exponentially complex deterministic dynamics. In a strange inversion, the very randomness that would degrade a classical computation becomes the fuel for a quantum one.
Where Does the Advantage Fray?
An important question sharpened by earlier work on quantum simulation of noisy classical dynamics is whether the removal of sparsity and weak-nonlinearity conditions is as absolute as it first appears. In a 2025 preprint, Zhuk and colleagues themselves explored quantum simulation of noisy classical nonlinear systems, but their algorithm was limited to sparse interactions. The present paper claims to lift that restriction. However, a closer look reveals a nuance: the algorithm’s efficiency depends on the ratio of the initial kinetic energy to the noise intensity. For very small noise relative to the energy, the simulation still requires resources that grow with the energy-to-noise ratio — a parameter that can be enormous for realistic turbulent flows.
Similarly, work by other groups on Carleman-embedding-based quantum solvers required the nonlinearity to satisfy a strict smallness condition, otherwise the linearization error exploded. The IBM algorithm dodges that requirement by relying on the noise to regularize the nonlinear terms. Yet as the adversarial dialogue embedded in the review process probed, the algorithm does not entirely escape the curse of dimensionality in the deterministic limit; it converts it into a curse of noise. For the quantum simulation to be efficient, the noise must be large enough that the stochastic system is well-mixed, but that same noise may wash out the very fine-scale features that make turbulence interesting. The art will lie in choosing a noise level that is both physically meaningful and computationally tractable.
A more concrete limitation concerns the divergence-free condition. The Navier-Stokes equations for incompressible fluids enforce that the velocity field has zero divergence — fluid elements neither compress nor expand. The norm-preserving property that the quantum algorithm demands is satisfied by certain discrete formulations of the equations, but the standard direct discretization of Navier-Stokes does not produce a drift with this property. To apply the algorithm to a standard turbulence benchmark, one must either pre-filter the equation or use a specialized discretization. The team acknowledges this and suggests that filtering and turbulence modelling are part of the picture; indeed, they illustrate their method on a damped Euler–Bardina equation, a simplified model that already captures some of the statistical features of real turbulence. The gap between this toy model and a full three-dimensional Navier-Stokes simulation remains wide, but the door is now open.
The Turbulence in a Computer
The team did not merely prove theorems. They ran classical Monte Carlo simulations of the damped Euler–Burgers model to confirm that the quantum algorithm’s underlying physics is sound. Starting from a randomized initial velocity profile, they evolved the stochastic dynamics and measured the average velocity at selected spatial points over time. The results show that the noise-averaged expectation reproduces the expected smoothing of sharp gradients, with sample variances well-behaved and controllable. They tested different noise rates, different observables — single-point velocities, squared velocities, next-neighbor products — and the algorithm tracked the statistical expectations with the precision that the theory predicts.
Deterministic forcing amplifies the averaged velocity field in the stochastic model. This test confirms that quantum algorithms can accurately capture such nonlinear dynamics. (Source: arXiv:2606.08349)
These classical simulations are not, of course, the quantum speedup itself; they are a validation that the mathematical framework is correct. The next step, which the team is already pursuing, is to implement the subroutine for simulating the auxiliary harmonic oscillators on a real quantum processor. That will require continuous-variable quantum hardware — photonic systems, trapped-ion motional modes — that are still maturing. The road from a provably efficient algorithm to a machine that actually outperforms a classical cluster is long, but the theoretical foundation is now in place.
The Deeper Meaning of Necessary Noise
There is a philosophical tremor beneath this technical story. For a century, physicists have regarded noise as an adversary — a source of decoherence that destroys fragile quantum superpositions and robs quantum computers of their power. The industry has poured billions into error correction and fault tolerance, all to seal the quantum world from the environment’s hiss. Yet here we find a quantum algorithm that actively requires noise to function. The randomness is not an imperfection to be eliminated; it is a computational resource, a kind of lubricant that enables the quantum evolution to smoothly sample the statistical ensemble that the classical Kolmogorov equation describes.
This is not willfulness on the part of the machine, of course. It emerges from the mathematical structure: the probability distribution of the stochastic system is a smooth function that lives in a low-dimensional effective subspace, while the deterministic trajectory is a singular, high-complexity object. The noise broadcasts the solution across that tractable subspace. Quantum mechanics, with its natural affinity for wave-like propagation and its ability to simulate bosonic systems, becomes the ideal substrate. Nature already solves fluid turbulence without conscious effort — particles in a flow do not suffer algorithmic complexity — and perhaps the lesson here is that quantum simulation, by giving noise its due, can mimic nature’s own economies.
We are left with a deeper question. Is the exponential classical hardness of turbulence a fact about the physical world, or a fact about the classical computational model we inherited from Turing? If a quantum machine can dance with noise to extract statistical answers that are provably out of reach for non-quantum machines, then the boundary between the tractable and the intractable is not a property of the equations themselves but of the kind of logic we use to interrogate them. The IBM algorithm does not promise to solve turbulence tomorrow, but it shines a light on a path — one where noise, once the villain, becomes the key that unlocks a door we did not even know was locked.
— Yanjiang
Yanjiang is the founding editor of LoomSci.com, specializing in physics and science communication.
References
- Sergey Bravyi et al., Quantum algorithms for stochastic nonlinear differential equations, arXiv:2606.08349
- Sergiy Zhuk et al., Quantum simulation of a noisy classical nonlinear dynamics, arXiv:2507.06198
- Jennings et al., Quantum algorithms for general nonlinear dynamics based on the Carleman embedding, arXiv:2509.07155
- Ding et al., Quantum Algorithms for Nonlinear Differential Equations via Pivot-Shifted Carleman Linearization, arXiv:2605.20071