Quantum-Enhanced Memetic Tabu Search

• Quantum-Enhanced Memetic Tabu Search (QE-MTS) is a hybrid algorithm that combines fixed-depth quantum routines (DCQO or QAOA) with classical memetic tabu search to warm-start and accelerate combinatorial optimization.
• It leverages high-quality quantum-generated candidate solutions to achieve empirical scaling improvements, with reported reductions in circuit depth (up to ~6×) and superior performance on benchmarks like LABS.
• The approach employs rigorous statistical methods including replicate averaging and bootstrap confidence intervals to ensure robust convergence and validate its enhanced performance over classical methods.

Quantum-Enhanced Memetic Tabu Search (QE-MTS) is a class of hybrid quantum–classical algorithms for combinatorial and higher-order unconstrained binary optimization (HUBO) problems. QE-MTS frameworks combine structured quantum routines—typically efficient, non-variational samplers such as digitized counterdiabatic quantum optimization (DCQO) or the quantum approximate optimization algorithm (QAOA)—with robust classical metaheuristics, most notably population-based memetic tabu search (MTS). By leveraging quantum-generated high-quality candidate bitstrings to “warm-start” a classical search, QE-MTS achieves state-of-the-art empirical scaling and substantial circuit-depth reductions, establishing competitive performance for paradigmatic problems such as the low-autocorrelation binary sequence (LABS) benchmark and dense HUBO instances (Cadavid et al., 6 Nov 2025, Chandarana et al., 7 Oct 2025, Moussa et al., 2020).

1. Foundational Principles and Algorithmic Structure

QE-MTS operates by tightly integrating quantum sampling routines with classical memetic tabu search. The typical workflow contains the following components:

• Quantum Subroutine: A fixed-depth quantum circuit, such as DCQO or QAOA, is employed to sample bitstrings likely to have low cost with respect to the problem Hamiltonian. In DCQO, counterdiabatic corrections derived from first-order nested commutators are used to suppress diabatic transitions, yielding circuits highly expressive at shallow depth (Cadavid et al., 6 Nov 2025, Chandarana et al., 7 Oct 2025).
• Population Initialization: The best quantum-generated bitstring is identified via classical postprocessing and used to seed the entire metaheuristic population (“quantum warm-start”) (Cadavid et al., 6 Nov 2025, Chandarana et al., 7 Oct 2025).
• Classical MTS Loop: Global exploration is performed via recombination (crossover) and mutation, while a local intensification stage is implemented through a tabu search subroutine. Short-term memory is maintained in the form of a tabu list to avoid cycling and promote diversification (Cadavid et al., 6 Nov 2025, Chandarana et al., 7 Oct 2025, Moussa et al., 2020).
• Solution Update: Offspring are generated, mutated, and locally improved; populations are updated based on elitist selection. The process iterates until convergence or a known optimum is reached.

This architecture enables quantum sampling to guide the classical search toward promising regions of the landscape, while classical heuristics ensure robust local refinement and global convergence.

2. Quantum Sampling Subroutines: DCQO and QAOA

The effectiveness of QE-MTS derives in large part from the use of advanced quantum subroutines.

• Digitized Counterdiabatic Quantum Optimization (DCQO): DCQO is based on counterdiabatic (CD) driving, where an auxiliary term constructed from the commutator expansion is introduced to suppress diabatic transitions during the adiabatic interpolation between an initial easy Hamiltonian $H_i$ and the target problem Hamiltonian $H_f$. The practical implementation truncates the nested-commutator series at first order, significantly reducing circuit depth (Cadavid et al., 6 Nov 2025, Chandarana et al., 7 Oct 2025). In the impulse regime, CD terms dominate, further lowering the quantum resource requirements, with a single Trotterized DCQO step utilizing roughly the same resources as two layers of QAOA.
• Quantum Approximate Optimization Algorithm (QAOA): When combined with Tabu Search as a quantum neighborhood sampler (Moussa et al., 2020), the QAOA is applied to small k-bit subproblems. At each TS iteration, a QAOA instance of depth $p$ is locally optimized in parameter space, and samples are mapped back to the global bitstring. This allows exploration of classically intractable Hamming-ball neighborhoods while requiring only limited quantum resources.

A central feature is that both DCQO and QAOA are operated at fixed or shallow depths, keeping gate counts and circuit depths within or below the tolerance of near-term quantum devices. DCQO in QE-MTS, for example, achieves $\sim 6\times$ reduction in total two-qubit gate count compared to standalone twelve-layer QAOA when solving LABS (Cadavid et al., 6 Nov 2025).

3. Classical Memetic Tabu Search Framework

The classical backbone in most QE-MTS instantiations is a population-based memetic tabu search. This framework consists of:

• Population Management: A fixed-size population $K$ maintains the current candidate solutions. Offspring are generated either via recombination of parents (with probability $p_{comb}$) or via random initialization.
• Tabu Search Subroutine: Each offspring undergoes a one-spin-flip tabu search, where local improvement steps are guided by a short-term memory (tabu tenure) to prevent cycles. The tabu list is typically randomized in length to encourage diversification (Cadavid et al., 6 Nov 2025, Chandarana et al., 7 Oct 2025).
• Region Intensification: Warm-starting the population with quantum-evolved bitstrings positions the search in high-quality landscape basins.
• Mutation and Crossover: Standard genetic operators are employed, with bit-flip mutation rate $p_{mut}$, typically inversely proportional to problem size. Decaying mutation schedules may be used (Chandarana et al., 7 Oct 2025).

The MTS terminates upon finding an optimal or sufficiently low-energy solution, or after exhausting a pre-specified iteration/generation budget.

4. Empirical Scaling and Performance Benchmarks

QE-MTS demonstrates quantifiable scaling advantages over both classical MTS and quantum-only heuristics.

Method	Scaling Exponent $\kappa$	Time-to-Solution $T(N)$	Circuit Depth
Classical MTS	$1.34$	$O(1.34^N)$	Classical
QE-MTS (w/ DCQO)	$1.24$	$O(1.24^N)$	$\sim 1/6$ of QAOA
QAOA (12 layers, LABS)	$1.46$	$O(1.46^N)$	Baseline

For LABS on $N \in [27,37]$, QE-MTS achieves a median scaling exponent $\kappa \in [1.23, 1.25]$, outperforming classical MTS ($\kappa \in [1.36, 1.37]$), and surpasses QAOA in both scaling and circuit resource consumption (Cadavid et al., 6 Nov 2025). A two-stage bootstrap analysis confirms that for $N \gtrsim 47$, QE-MTS crosses a threshold where it requires fewer function evaluations than even the best-tuned classical MTS (crossover $N_{\times} \sim 46.6$, 95% CI $[44.9, 48.9]$) (Cadavid et al., 6 Nov 2025). In HSQC pipelines, warm-started QE-MTS achieves up to $9\times$ speedup over standalone MTS and $700\times$ over simulated annealing on N=156 HUBO benchmarks (Chandarana et al., 7 Oct 2025).

5. Statistical Methodology and Robustness

Performance assessment for QE-MTS employs rigorous statistical procedures:

• Replicate and Seed Averaging: For each sequence length, 100 independent initial populations (replicates) are generated, each running 100 MTS instances (seeds) (Cadavid et al., 6 Nov 2025).
• Bootstrap Confidence Intervals: A two-stage bootstrap ($B=5,000$ draws) is used, resampling both replicates and seeds, then fitting the empirical quantiles $Q_p(N, m)$ for log-linear scaling models, producing tight confidence intervals on scaling exponents and crossover points (Cadavid et al., 6 Nov 2025).
• Convergence Metrics: Both solution quality (e.g., best energy, ground-state recovery) and time-to-solution (number of function evaluations) are tracked. In hybrid sequential settings, time-to-99% confidence convergence $\tau$ is also computed (Chandarana et al., 7 Oct 2025).

This methodological rigor establishes the statistical significance of observed speedups and scaling improvements.

6. Circuit-Depth Reduction and Practical Considerations

The resource efficiency of QE-MTS is directly attributed to shallow-depth quantum circuits:

• DCQO Circuit Compression: First-order truncated CD terms and impulse-regime operation yield a factor of $\sim 6$ savings in entangler count relative to high-depth QAOA (Cadavid et al., 6 Nov 2025). This makes quantum-enhanced warm-starts feasible on early fault-tolerant or heavy-hex superconducting hardware (Chandarana et al., 7 Oct 2025).
• Trade-offs: For small $N \lesssim 30$, classical MTS may outperform QE-MTS in wall-clock time due to quantum sampling overhead. However, the lower scaling exponent ensures asymptotic advantage at larger $N$.
• Population Diversity: Multiple independent quantum runs or adaptive CD schedules can diversify the initial population, further improving early-stage convergence (Cadavid et al., 6 Nov 2025).

A plausible implication is that hybrid quantum-classical pipelines are likely to remain essential for practical quantum advantage in combinatorial optimization, as circuit depth and parameter noise remain limiting factors on current hardware.

7. Extensions, Limitations, and Research Directions

Recent developments highlight both the flexibility and the current boundaries of QE-MTS:

• Hybrid Sequential Frameworks: Workflow pipelines integrating classical annealing, quantum optimization (e.g., BF-DCQO), and memetic tabu search generalize QE-MTS concepts and have demonstrated consistent speedups on dense, higher-order problems (Chandarana et al., 7 Oct 2025).
• Parameter and Memory Adaptation: Extensions include online adaptation of quantum circuit parameters, subproblem sizes in QAOA-based samplers, tabu tenures, and acceptance criteria for probabilistic move selection (Moussa et al., 2020).
• Robustness to Noise: The classical metaheuristic backbone provides resilience to quantum device imperfections and parameter mis-tuning, a crucial attribute for near-term deployments (Moussa et al., 2020).
• Current Limitations: Overhead from quantum subroutine optimization and finite-sample estimation on noisy devices can blunt gains at modest instance sizes. Greedy move acceptance in classical TS may limit exploration depth unless hybrid probabilistic schemes are incorporated.

It is also suggested that QE-MTS methodologies can be extended and adapted to other problem classes, such as MAXCUT and graph coloring, by reformulating the subproblem Hamiltonians and adjusting classical search components (Moussa et al., 2020). Empirical validation on state-of-the-art quantum hardware remains an active area of investigation, especially regarding error mitigation and hardware-specific scheduling.

• "Scaling advantage with quantum-enhanced memetic tabu search for LABS" (Cadavid et al., 6 Nov 2025)
• "Hybrid Sequential Quantum Computing" (Chandarana et al., 7 Oct 2025)
• "Tabu-driven Quantum Neighborhood Samplers" (Moussa et al., 2020)