How Classical Tensor Networks Reproduced D‑Wave’s Quantum Annealing Results

In 2025, D‑Wave reported a striking milestone: its Advantage2 quantum annealing processor could simulate the dynamics of complex spin‑glass systems far faster than classical computers. The company argued that reproducing the same results on a classical supercomputer would take nearly a million years, positioning the experiment as a demonstration of “beyond‑classical” computation.

But follow‑up work by physicists at the Flatiron Institute’s Center for Computational Quantum Physics and collaborators at Boston University showed that the same physics could be reproduced with carefully designed classical algorithms. Their approach—based on tensor‑network simulations—was efficient enough to run on modest hardware, including a personal laptop in some demonstrations.

The result does not invalidate quantum computing. Instead, it reveals how subtle the classical‑versus‑quantum boundary really is.

The Original Claim: Quantum Annealing Outperforms Classical Simulation

D‑Wave’s study focused on the quantum dynamics of disordered spin systems, specifically transverse‑field Ising models that represent magnetic materials and spin‑glass physics. These systems are notoriously difficult to simulate because the quantum state space grows exponentially with the number of interacting spins.

Using its superconducting Advantage2 quantum annealer, D‑Wave generated samples that closely matched the dynamics predicted by the Schrödinger equation for these systems.

The company reported that:

• The quantum processor could perform the simulation in minutes.
• Classical simulations using conventional approaches would require extremely large supercomputing resources and impractically long runtimes.

These results were presented as evidence that the quantum processor had achieved a practical form of quantum computational advantage for simulating complex materials.

The Classical Counter‑Result

Researchers at Flatiron and Boston University revisited the same class of spin‑glass dynamics using tensor‑network methods, a family of algorithms that compress quantum states into structured mathematical objects.

Their study showed that two‑ and three‑dimensional tensor networks can accurately and efficiently simulate the quantum annealing dynamics of Ising spin glasses across multiple lattice geometries.

The team adapted tensor‑network evolution methods and incorporated belief‑propagation techniques to keep up with the entanglement produced during the time evolution of the system. This allowed the algorithm to maintain an accurate but compressed representation of the quantum state.

Because the state representation remained compact, the simulations could run on relatively modest hardware—demonstrating that the supposedly intractable task did not necessarily require a large quantum computer.

Why Tensor Networks Can Work So Well

The key insight lies in entanglement structure.

D‑Wave’s own analysis reported that the simulated systems exhibited area‑law scaling of entanglement in their quench dynamics.

This detail matters enormously for simulation complexity.

When a quantum system obeys an area law:

• Entanglement grows with the boundary of regions rather than their volume.
• The quantum state can often be compressed efficiently.
• Tensor‑network representations remain tractable even for fairly large systems.

Tensor networks are specifically designed to exploit this structure, allowing classical computers to simulate certain many‑body quantum systems that would otherwise appear exponentially complex.

What This Means for “Quantum Supremacy”

The result doesn’t prove that classical computers can simulate every quantum process efficiently. Instead, it demonstrates that the presence of many qubits alone does not guarantee a classical barrier.

The real dividing line is subtler:

• If a quantum system generates highly complex, volume‑law entanglement, classical simulation becomes extremely difficult.
• If its entanglement remains structured or limited (for example, area‑law scaling), tensor‑network techniques may still simulate it efficiently.

In other words, the advantage of quantum hardware depends less on qubit count and more on the structure of the quantum state being generated.

The Bigger Lesson for Quantum Computing

This episode illustrates a recurring pattern in quantum‑computing research. Claims of quantum advantage often trigger rapid improvements in classical algorithms designed to reproduce the same results.

Rather than weakening the field, this dynamic helps clarify where quantum devices truly outperform classical methods. Each challenge pushes researchers to identify the precise physical regimes where classical compression techniques fail.

For now, the Flatiron–Boston University work suggests that the boundary between classical and quantum advantage is narrower—and more dependent on entanglement structure—than simple hardware comparisons might imply.

As quantum hardware continues to improve, the next generation of experiments will likely focus on regimes where tensor‑network compression breaks down, providing a clearer demonstration of genuinely hard quantum dynamics.