Facing a scheduling or routing puzzle with too many possible answers? Quantum annealing is a quantum method that encodes such a problem as an energy landscape and searches for its lowest point, though it is a heuristic, not a guaranteed solver.
This page is part of our coverage of quantum technology, where annealing sits beside other quantum approaches.
| Item | Value |
|---|---|
| Physical qubits per logical variable (minor embedding) | Usually 5 to 12 qubits |
| Effective problem capacity limited by embedding | 80-92% |
| Device used in the nurse scheduling study | D-Wave 2000Q |
| Qubit technology of that device | Superconducting flux qubits |
| Input format the device is designed to solve | QUBO (quadratic unconstrained binary optimization) |
These figures come from a review of quantum annealing for combinatorial optimization and a nurse scheduling study. Research figures can change as hardware and methods evolve, so check the sources before you rely on them.
How quantum annealing works
Quantum annealing is a heuristic method for combinatorial optimization problems. It is derived from the principles of quantum mechanics, and it aims for good answers rather than a proof of the best one.
The basic recipe has three steps:
- Encode the optimization problem as a problem Hamiltonian, an energy function whose lowest-energy state (the ground state) is the optimal solution.
- Start a quantum system in the easily prepared ground state of a separate driver Hamiltonian.
- Slowly move the system from the driver to the problem Hamiltonian, then read out the final state as a candidate solution.
If that evolution is carried out slowly enough, the adiabatic theorem says the system stays in its ground state and ends on the best solution. The nurse scheduling paper notes that in practice the adiabatic condition is rarely obtained, so the guarantee that the true solution will be recovered is lost.
The main contrast with classical simulated annealing is how barriers are crossed. Simulated annealing uses thermal activation to overcome energy barriers, while quantum annealing takes advantage of quantum tunnelling to get through them. The review describes the idea as taking shape in the late 1990s.
Problems that suit quantum annealing
The target is combinatorial optimization: picking the best configuration out of an enormous set of discrete choices. These problems scale exponentially, which makes trying every option infeasible. The review gives these examples:
- Scheduling manufacturing processes.
- Picking paths for delivery trucks.
- Creating molecular compounds with required properties.
- Assigning portfolio combinations of investments.
The same review describes three main classical strategies for such problems, and it is useful to see where annealing fits among them. You can read about them in the review.
- Exact algorithms such as branch-and-bound are guaranteed to provide optimal solutions, but their worst-case running time is exponential.
- Approximation algorithms run in polynomial time with provable performance, but only for particular problem structures.
- Metaheuristics such as simulated annealing, tabu search, genetic algorithms and particle swarm methods give up guarantees of optimality in favour of scalability.
Quantum annealing belongs with the heuristics. Optimization is only one family of use cases, and the page on quantum computing applications places it in a wider context.
From a real problem to QUBO and hardware
Before an annealer can touch your problem, it must be written as a QUBO. This framework offers a coherent mathematical abstraction of optimization problems, and it coincides formally with the Ising spin glass model, which is what links it to annealing hardware.
The D-Wave 2000Q in the nurse scheduling study is based on superconducting flux qubits and is designed to solve QUBO problems. The problem is encoded into a sparsely connected graph that expresses the hardware interactions, as the study explains.
Because the hardware graph is sparse, the problem has to be mapped onto it, and that costs qubits. The review says minor embeddings usually need between 5 and 12 physical qubits per logical variable, which limits effective problem capacity by 80-92%.
Its authors conclude that embedding and encoding overhead is the largest determinant of scalability and performance, not just the number of qubits. Chain-breaking errors can also compromise the quality of solutions.
Example: nurse scheduling on an annealer
The nurse scheduling problem (NSP) is the task of creating a rotating roster of nurses at a hospital while respecting constraints on their availability and level of effort. Solutions must respect hard constraints, such as days off and minimum availability, as well as softer ones, according to the study.
The authors start with a simple two-shift system of day duty and night duty. They apply two hard constraints: a minimum number of nurses on each shift, and a minimal rest period between shifts for each nurse.
After reducing the problem to an Ising-type Hamiltonian, they ran it on the D-Wave 2000Q. For the test problems explored, the results indicate that quantum annealing recovers satisfying solutions. They also found that reverse annealing, which refines returned results by using the annealing process a second time, can greatly improve solution quality.

Does quantum annealing beat classical solvers?
That question is still open. The nurse scheduling paper says it remains open when quasi-adiabatic behaviour can provide a computational advantage, and it states that its own results cannot settle the matter.
Theory is also unsettled. Problems like NSP are described as stoquastic, and the paper says it is unknown whether such problems can be solved efficiently by strictly classical methods. It sees a clear potential for advantage only for non-stoquastic problems, which do not correspond to NSP or other combinatorial optimization problems.
The review, posted to arXiv on 3 Feb 2026, argues that annealing shows measurable utility as a hybrid refinement method and not as a standalone solver. It also criticises benchmarks that use poor classical baselines and exclude preprocessing overhead, as the review sets out.
Annealing is a computing approach and not a communication one. The separate topic of the quantum internet deals with quantum networking.
What to check before trusting a claim
A quick checklist helps when you read a result about annealing:
- Did the comparison include a strong classical baseline?
- Was preprocessing overhead counted in the timing?
- How many physical qubits did embedding consume per variable?
- Was the annealer used alone, or as a hybrid refinement step?
A yes to the first two and honest answers to the others make a claim far easier to take seriously.
Frequently asked questions
What is quantum annealing in simple terms?
Show answer
It is a heuristic method that encodes an optimization problem as an energy function, starts a quantum system in an easy state and slowly steers it toward the lowest-energy state, which represents a good solution.
How is it different from simulated annealing?
Show answer
Simulated annealing uses thermal activation to overcome energy barriers. Quantum annealing takes advantage of quantum tunnelling instead, according to the review.
What is a QUBO?
Show answer
QUBO stands for quadratic unconstrained binary optimization. It is the problem format the D-Wave 2000Q is designed to solve, and it is closely tied to the Ising spin glass model.
Is quantum annealing proven faster than classical methods?
Show answer
No. The nurse scheduling paper says it remains open when such behaviour can provide a computational advantage, and the review warns that weak baselines and ignored overhead can distort comparisons.
What is reverse annealing?
Show answer
It is a way to refine returned results by using the annealing process a second time. In the nurse scheduling study it greatly improved solution quality.



