Nonlinear CFD on real quantum hardware: Inside the new breakthrough by Quanscient and Haiqu

Key takeaways

• Quanscient and Haiqu announced a new quantum algorithm for computational fluid dynamics (CFD), successfully running nonlinear flow simulations past an embedded airfoil-type object on real hardware.

• The one-step simplified lattice Boltzmann method combines collision and propagation steps, reducing circuit complexity, and making practical quantum simulations more feasible.

• The algorithm provides flexibility across different physics models, while retaining algorithmic efficiency.

• Validation against classical simulations confirms accurate results for CFD and multiphysics applications.

• Early applications include aerospace, automotive, and maritime simulations.

• This work marks a step toward industrially relevant quantum CFD and opens paths for AI-quantum convergence in simulation.

Introduction

Earlier this month, Quanscient and Haiqu announced a new algorithm designed to make CFD simulations more practical on quantum computers. The work demonstrated a nonlinear fluid-flow simulation with an embedded object running on real quantum hardware, a step toward more realistic engineering simulations on quantum systems.

CFD simulations are among the most computationally demanding workloads in engineering. They are widely used across industries such as aerospace, automotive, and energy, where accurately modeling fluid behavior is critical for product performance and safety. However, these simulations are extremely demanding for even today’s most powerful supercomputers, often taking days or even weeks to complete, if possible at all.

This latest collaboration focused on addressing one of the key limitations in quantum simulations: the high number of qubits and operations typically required to model complex physical systems. By introducing a more efficient algorithmic approach, the teams demonstrated a viable path toward scaling quantum simulations to more practical engineering problems in the future.

To better understand the technical challenges, the breakthrough itself, and what this result means for the future of engineering simulation, we spoke with our Chief Scientist, Industry Prof. Valtteri Lahtinen.

Quanscient Chief Scientist, Industry Prof. Valtteri Lahtinen

How did the collaboration with Haiqu begin, and what shared goal brought two teams together?

I remember meeting the Haiqu team at several events over the years. After meeting two or three times at different quantum events, we started thinking about whether there might be a way to collaborate. We also connected quite naturally on a personal level, which helped build trust early on.

From there, we began having more regular meetings. I think the first official collaboration was the BMW-Airbus challenge. We participated together and became finalists in that challenge, as a testimony to a successful collaboration. That was the first time we formally worked together, where we provided the lattice Boltzmann algorithms and they ran the circuits using their middleware tools, much like in this recent collaboration.

So that is how it started. There is also a strong personal connection between the teams. They are great people to work with. And, of course, their technology and professionalism are outstanding.

At the start of this work, what specific challenge were you trying to solve in quantum simulations of CFD?

In many ways, this work continues what we started during the BMW-Airbus challenge. In that project, we ran an airfoil problem, but it was executed on an emulator rather than on a real quantum device.

Here, we wanted to see whether an airfoil-like shape could actually be run on today’s real hardware. A major challenge in quantum algorithms is embedding geometric objects into the modeling domain. That process can be quite complex.

So our goal was to address that challenge and run this type of airfoil problem, even at a smaller scale, on an actual quantum device.

What problem in CFD makes it such a persistent bottleneck in engineering today?

First of all, running Navier–Stokes simulations is already a demanding problem in itself. The circuits become quite deep when you deal with more complicated physics.

With this new algorithm, we have been able to significantly reduce both the number of qubits and the number of gates required. We also found a more optimal way to represent the lattice Boltzmann method on a quantum computer. That is why we have now been able, with the help of Haiqu, to run this problem on a real device.

In particular, embedding an object into the simulation requires many multi-controlled quantum gates. When those gates are decomposed into one- and two-qubit gate circuits that can actually be run on real hardware, the circuits can become very deep.

Despite this challenge, the new algorithm allowed us to complete the simulation successfully. That is essentially the key point.

At what point did it become clear that classical computing alone would not be enough to keep scaling CFD simulations?

In this particular demonstration, we are not yet doing something that a classical computer could not do. In fact, we validate the results using classical computers to confirm correctness.

The difference lies in how the algorithm scales. Each time step of this quantum algorithm scales much more favorably with lattice size than the corresponding classical algorithm.

This means that once we reach sufficiently large lattice sizes, we will be able to run simulations that cannot be executed on classical devices. The computational scaling becomes significantly more efficient with the quantum algorithm compared with classical methods.

In simple terms, what makes the one-step simplified lattice Boltzmann method different from previous quantum approaches to fluid simulation?

In the traditional lattice Boltzmann method, you typically have two main steps: collision and propagation. First, particle densities collide, and then they propagate across the lattice.

While the one-step simplified Lattice Boltzmann Method originated in classical computing, we have pioneered its first-ever quantum implementation. In this method, by mathematically fusing the collision and propagation, both are executed in a single, unified operation, streamlining the algorithm for quantum architectures.

Why is reducing the number of qubits such an important step toward practical quantum simulation?

The fewer qubits we need, the sooner we will have hardware capable of running meaningful simulations.

In near-term and mid-term quantum hardware development, logical qubits will remain a limited resource. If we can reduce the number of required qubits as much as possible, we can reach industrially relevant simulations much sooner.

This experiment combined nonlinear flow, geometry, and repeated time stepping on real hardware. What made that combination meaningful from a simulation perspective?

Previous hardware demonstrations have mostly focused on relatively simple linear problems without embedded geometric objects.

That is why solving a nonlinear flow problem, embedding a geometric object into the simulation, and running multiple time steps with accurate results represents a significant milestone. This is the first airfoil-like flow simulation carried out on a real quantum computer, and as such, it demonstrates that our algorithmic framework charts a path towards industrially meaningful simulations as we scale up.

What was the most difficult technical challenge in getting this method to run successfully on today’s quantum systems?

From our perspective, this has been a long development process.

It has taken several years to design a flexible algorithm like this. We had to develop the one-step simplified quantum lattice Boltzmann method, refine the details, and ensure that it worked correctly, first on an emulator and now on a real device.

We also performed very rigorous validation to confirm that the method produces correct CFD and multiphysics results. This approach is not limited to CFD alone.

Overall, the most challenging part has been the persistence required to develop and validate the method step by step.

What does this result make possible that was not feasible before?

First, the new algorithm enables more flexible simulation of different multiphysics problems. By adjusting one part of the algorithm, we can apply it to different types of physics.

Second, as we demonstrated, it is now possible to run a nonlinear problem with complex geometry on today’s quantum hardware. That had not been achieved before.

How does this work move quantum CFD from theoretical research toward something engineers could eventually use in real workflows?

At Quanscient, we work directly with enterprises on quantum proof-of-concept projects.

In these collaborations, we demonstrate what is possible today and outline a roadmap toward quantum advantage, identifying when specific capabilities are expected to become practical within real engineering workflows.

This approach moves the work beyond purely theoretical research into real enterprise applications. We are already using the one-step simplified quantum lattice Boltzmann method in partner and customer projects.

We validate the results against well-established classical CFD tools. This helps organizations understand what performance gains to expect, when those gains may arrive, and how they should prepare for quantum computing in simulation workflows.

As the technology scales to larger problems, engineers will be able to solve simulations that are currently intractable using classical computers.

Which types of simulation problems are most likely to benefit first as this technology continues to develop?

For example, CFD problems in aerospace, automotive, maritime, and similar industries.

The common thread here is resource efficiency. At a large scale, these problems are traditionally memory-bound or compute-bound, that is, they run ouf or RAM, or they take too long. Our quantum approach effectively compresses the problem size, making these massive industrial tasks fit onto quantum hardware sooner.

What are the next major milestones that need to happen before quantum CFD becomes commercially relevant, and what signal should engineers and industry leaders take from this result today?

We have now demonstrated that a small-scale 2D problem can run on a real quantum device.

The next major milestone is scaling this approach to larger 2D problems and eventually moving to 3D simulations. That is something we are actively working on, with the goal of making progress possibly by the end of the year.

Hardware development is also critical. We will need robust enough error correction to solve problems large enough to deliver clear industrial value.

Based on current hardware roadmaps, we expect this capability to emerge at a sufficient level for quantum utility at a limited capacity within roughly the next four years. It could happen sooner, perhaps in two or three years, but that remains to be seen. In any case, the first utility demonstrations will be in well-defined, pinpointed simulation use cases, rather than universal advantage across engineering R&D.

The key signal for industry is that progress is steady. We are moving step by step toward the point where quantum systems can solve problems larger than any classical high-performance computer can handle. That is the moment when true quantum advantage becomes visible.

Where do you see the first meaningful convergence between quantum computing and AI in engineering simulation?

There is already significant work happening in simulation AI and engineering AI. At Quanscient, for example, we are developing MultiphysicsAI that can be trained on the large amounts of simulation data generated using our cloud-based solvers.

One of the earliest benefits from quantum computing may come from generating new types of data that are difficult or impossible to produce using classical computing, since for that to be beneficial, you may not need to simulate a full end-to-end multiphysics problem on a quantum computer, but some limited number of time steps might be enough.

For instance, quantum systems could generate simulation data representing complex physical behaviors, such as small-scale turbulence, that classical simulations cannot efficiently capture. That data could then be used to train AI models, improving their predictive accuracy.

This could represent one of the first practical points of convergence between quantum computing and artificial intelligence in engineering simulation.

Learn more about Quanscient and get in touch now at quanscient.com