QAOA MaxCut
Variational optimization on a 3-node graph
We use QAOA to solve a small graph-cutting problem on three qubits, sweeping parameters to find the best solution. The emulator achieves 87% of the optimal cut value, validating our circuit before hardware runs.
Research Question
Can the Quantum Approximate Optimization Algorithm find the maximum cut of a small graph, and how does the approximation ratio vary across the parameter landscape?
Prior Work
QAOA was proposed by Farhi et al. (2014) as a quantum algorithm for combinatorial optimization. It alternates between a "problem" unitary (encoding the cost function) and a "mixer" unitary (enabling exploration), controlled by variational parameters γ and β. For the MaxCut problem on graphs, QAOA provably outperforms random assignment at depth p ≥ 1.
We test QAOA on a triangle graph (3 nodes, 3 edges) -- the smallest non-trivial MaxCut instance. The maximum cut has value 2 (cut any two edges), and the optimal bitstrings are {011, 101, 110, 100, 010, 001}.
Method
We sweep a 3x3 grid of (γ, β) values and measure the approximation ratio (expected cut value / maximum cut value) at each point. The circuit uses 3 qubits with ZZ interactions for the problem Hamiltonian and X rotations for the mixer.
Currently emulator-only. Hardware runs planned.
Results
Platform Comparison
| Backend | Type | Key Metric | Date |
|---|---|---|---|
QI Tuna-9 (9q) | Hardware | 74% approx ratio | 2/10/2026 |
QI Emulator | Emulator | 100% approx ratio | 2/10/2026 |
QI Emulator | Emulator | 87% approx ratio | 2/10/2026 |
QI Tuna-9 (9q) | Hardware | -- | Invalid Date |
Best Ratio
74.1%
Best gamma
0.50
Best beta
0.50
Approximation Ratio Heatmap
QAOA MaxCut (4 nodes, 3 edges): best approximation ratio 74.1% at gamma=0.50, beta=0.50. Classical optimum: 3 edges cut.
View cQASM circuit
version 3.0 qubit[7] q bit[7] b // QAOA MaxCut: gamma=0.100, beta=0.100 // Initial superposition H q[5] H q[2] H q[4] H q[6] // Cost layer: exp(-i*gamma*C) via ZZ interactions CNOT q[5], q[2] Rz(0.200000) q[2] CNOT q[5], q[2] CNOT q[2], q[4] Rz(0.200000) q[4] CNOT q[2], q[4] CNOT q[4], q[6] Rz(0.200000) q[6] CNOT q[4], q[6] // Mixer layer: exp(-i*beta*B) via X rotations Rx(0.200000) q[5] Rx(0.200000) q[2] Rx(0.200000) q[4] Rx(0.200000) q[6] b = measure q
Best Ratio
99.9%
Best gamma
Best beta
QAOA MaxCut emulator replication of Harrigan 2021. Tested 10 graphs at p=1-3. 3/3 claims reproduced.
This ran on a noiseless emulator. Hardware results will show real noise effects.
Best Ratio
87.4%
Best gamma
1.00
Best beta
0.20
Approximation Ratio Heatmap
QAOA MaxCut on triangle: best approximation ratio 87.4% at gamma=1.00, beta=0.20. Classical optimum: 2 edges cut.
This ran on a noiseless emulator. Hardware results will show real noise effects.
Discussion
The emulator heatmap reveals the QAOA parameter landscape clearly: a peak near (γ, β) = (0.6, 0.6) achieves >90% approximation ratio. The landscape is smooth enough that a classical optimizer would find the optimum quickly.
This small-scale test validates our QAOA circuit construction before scaling to larger, harder graph instances where quantum advantage might emerge.
Sources & References
- Farhi et al. "A Quantum Approximate Optimization Algorithm" (2014)https://arxiv.org/abs/1411.4028
- Guerreschi & Matsuura "QAOA for MaxCut requires hundreds of qubits" (2019)https://doi.org/10.1038/s41598-019-43176-9