| name | pubo-mst-falqon |
| description | PUBO formulation for Minimum Spanning Tree using FALQON quantum optimization methodology. Reformulates MST as Polynomial Unconstrained Binary Optimization without auxiliary variables, reducing qubit requirements. Use when solving graph optimization problems (MST, OPF classifiers, network design) on quantum or quantum-inspired hardware, especially when qubit count is constrained. Applies to quantum machine learning pipelines needing efficient combinatorial optimization, graph-based classifiers, and prototype selection. |
PUBO-MST via FALQON
Methodology from Pexe et al. (arXiv:2605.20637, May 2026).
Problem
Minimum Spanning Tree (MST) computation becomes prohibitive for large-scale datasets, especially in graph-based classifiers like Optimum-Path Forest (OPF) where MST-derived prototypes define decision boundaries.
Solution
Reformulate MST as a Polynomial Unconstrained Binary Optimization (PUBO) problem and solve using FALQON (Feedback-Based Quantum Optimization) — a feedback-driven quantum algorithm that doesn't require classical outer-loop optimization.
Key Innovation
- No auxiliary variables needed (unlike QUBO formulations)
- Fewer qubits required — direct binary encoding of edge selection
- FALQON replaces variational optimization with real-time feedback
PUBO Formulation
The MST problem is encoded as minimizing:
H = H_connectivity + λ · H_cardinality + μ · H_cycle
Where:
H_connectivity: ensures all nodes are connected (sum of edge weights)
H_cardinality: enforces exactly n-1 edges for n nodes
H_cycle: penalizes cycles in the selected subgraph
Binary Variables
x_e ∈ {0, 1} for each edge e — 1 if edge is included, 0 otherwise.
Unlike QUBO, PUBO allows higher-order terms directly, eliminating the need for auxiliary variables to reduce polynomial degree.
FALQON Algorithm
FALQON is a feedback-based quantum optimization algorithm:
For each layer k:
β_k = ⟨ψ_k| [H_mix, H_prob] |ψ_k⟩ (feedback signal)
|ψ_{k+1}⟩ = e^{-iβ_k H_mix Δt} e^{-i H_prob Δt} |ψ_k⟩
Key properties:
- No classical optimizer needed (unlike QAOA/VQE)
- Feedback signal β_k is measured from quantum state
- Converges toward ground state of problem Hamiltonian
Implementation Workflow
1. Formulate PUBO
def mst_pubo(graph):
"""Convert MST to PUBO formulation.
Returns: H = connectivity + λ·cardinality + μ·cycle_penalty
Variables: x_e for each edge
"""
n = len(graph.nodes)
edges = list(graph.edges)
H_connect = sum(w_e * x_e for e, w_e in graph.edge_weights())
H_card = (sum(x_e for e in edges) - (n - 1)) ** 2
H_cycle = sum(cycle_penalty(cycle, x_vars)
for cycle in find_all_cycles(graph))
return H_connect + λ * H_card + μ * H_cycle
2. Execute FALQON
def falqon(H_problem, H_mix, n_layers, dt=0.01):
"""Feedback-Based Quantum Optimization."""
state = uniform_superposition(n_qubits)
betas = []
for k in range(n_layers):
beta_k = expectation_value(state, commutator(H_mix, H_problem))
betas.append(beta_k)
state = evolve(state, H_problem, dt)
state = evolve(state, H_mix, beta_k * dt)
return state, betas
3. Extract Solution
def extract_mst(measurement_counts, edges):
"""Extract MST from FALQON measurement results."""
best_bitstring = max(measurement_counts, key=measurement_counts.get)
selected_edges = [e for i, e in enumerate(edges)
if best_bitstring[i] == '1']
return selected_edges
Results
- FALQON-optimized MST achieves accuracies comparable to classical Prim's algorithm
- Maintains prototype quality for OPF classification
- Occasionally reaches local minima but accuracy impact is minimal
- Reduces qubit count vs. QUBO formulations (no auxiliary variables)
When to Use
- Graph optimization on quantum/quantum-inspired hardware
- Qubit count is a bottleneck for QUBO formulations
- Need variational-free quantum optimization
- OPF classifiers on large datasets
- Network design, clustering, or prototype selection
Comparison
| Method | Qubits | Classical Optimizer | Accuracy |
|---|
| QUBO + QAOA | n_edges + n_aux | Yes (outer loop) | High |
| PUBO + FALQON | n_edges only | No | Comparable |
| Classical Prim | N/A | N/A | Exact |
Pitfalls
- FALQON may converge to local minima — monitor convergence
- Cycle penalty formulation grows exponentially with graph size
- Best suited for moderate-sized graphs on current hardware
- Parameter tuning (λ, μ, Δt, n_layers) affects convergence
Activation
quantum MST, PUBO optimization, FALQON, quantum graph algorithms, optimum-path forest, quantum combinatorial optimization, QML classifier, feedback-based quantum optimization, graph prototype selection