The problem
Mining capacity is fixed. Where it earns most keeps moving with coin prices, difficulty, transaction fees and network conditions, and the objective combines several competing factors, so it is non-convex.
The project looked at whether a quantum-inspired metaheuristic handled that better than the classical alternatives.
Enhanced Adaptive Quantum Genetic Algorithm
- Problem
- Allocate a fixed hash rate across four proof-of-work chains under a capacity constraint, against an objective balancing profit against risk.
- Approach
- EAQGA — superposition-based initialisation, quantum-guided selection and adaptive mutation, with repair operators keeping every individual feasible rather than tuning penalty coefficients.
- Result
- I derived the closed-form optimum and validated the optimiser against it: a 0.00% gap in 1.8 s, converging by generation 14. A reduced search budget reproduced a 9.4% gap, which showed the metric was sensitive rather than saturated.
How it was put together
Four independent layers: synthetic market data, the mining objective, the optimisation algorithms, and the quantum simulation backend. Keeping the economic model separate from the search meant either could be swapped without touching the other.
One adapter interface sits over IBM Qiskit and Google Cirq with three tiers of fallback, so the full test suite runs with no quantum dependencies installed.
Modelling the market and the objective
720 hourly observations per chain, from correlated geometric Brownian motion over a positive-semidefinite equicorrelation matrix, with Beta and lognormal processes for the bounded and strictly positive variables.
Fitness folded expected profit, a Sharpe-style risk-adjusted return and risk into a single score, under the allocation constraint.
Benchmarking, and what it caught
EAQGA against a classical GA and a quantum-inspired GA, compared on convergence, solution quality, runtime and population diversity across repeated runs.
Per-generation diversity and coherence telemetry surfaced a 6× scale bias in the benchmark methodology itself, found by comparing predicted against measured values.