Quantum-Assisted Recursive Algorithm (QARA)

• Quantum-Assisted Recursive Algorithm (QARA) is a hybrid method that alternates between deterministic classical pruning and quantum-guided variable selection to solve complex combinatorial problems.
• It employs classical reduction rules and QAOA-based routines to achieve up to 60% greater exact cover success rates while reducing parameter optimization iterations by roughly 81%.
• The framework incorporates local verification and rollback safeguards to maintain solution integrity and scalability on NISQ devices.

A Quantum-Assisted Recursive Algorithm (QARA) is a hybrid quantum-classical algorithmic framework in which recursive control alternates between deterministic classical simplification and quantum-information-guided pruning. Designed specifically for combinatorial search problems with complex constraints, QARA exploits quantum subroutines—such as the output of the Quantum Approximate Optimization Algorithm (QAOA)—together with classical inference rules to efficiently reduce the problem size, maximizing solution quality even at shallow quantum circuit depths (Ni et al., 13 Sep 2025).

1. Hybrid Recursive Algorithm Structure

QARA addresses recursive combinatorial problems by leveraging both classical and quantum means to progressively prune the search space. The workflow comprises:

• Classical Pruning: Initially applies deterministic reduction rules derived from problem constraints (e.g., uniqueness and completeness for the exact cover problem). For example, if an element is found in only one subset, the corresponding subset variable must be fixed, and subsets violating uniqueness are iteratively removed.
• Quantum Pruning: When classical pruning can no longer simplify the instance (i.e., every element appears in at least two subsets), a quantum routine—concretely a QAOA circuit at shallow depth—is run. Quantum output is used as statistical guidance to bias the selection of variables for fixing.
• Alternation: This process alternates. After quantum pruning modifies the instance, classical pruning is invoked again, recursively reducing the problem until all variables are fixed or the search exhausts permissible choices.

This alternation is central: the recursion proceeds through repeated application of reduction (classical or quantum), shrinking the problem at each step until a solution is either found or all paths are exhausted.

2. Classical and Quantum Pruning Methodologies

Classical Pruning in QARA uses deterministic and problem-specific reduction rules:

• If an element $e_j$ is contained in a unique subset $S_t$, then the corresponding variable $x_t$ is set to 1 (selected for the cover). All subsets $S_k$ that overlap with $S_t$ are then fixed to 0 (excluded), and redundant elements/subsets are removed from consideration.

Quantum Pruning occurs when no deterministic inferences can be drawn:

• A parameter-optimized QAOA circuit is executed to produce an output state $|\psi\rangle$.
• For each subset $S_i$, the measurement bias is quantified as

$$M_i = \langle\psi|Z_i|\psi\rangle = \sum_{x\in\{0,1\}^m} |a_x|^2 (-1)^{x_i}$$

with $a_x$ the probability amplitudes of each bitstring assignment.

• The variable with the strongest bias $|M_i|$ is identified ($S_t$0), and reduced according to its sign:
  • If $S_t$1, then $S_t$2, and all subsets sharing elements with $S_t$3 are set to 0, enforcing uniqueness.
  • If $S_t$4, only $S_t$5 is rejected ($S_t$6).

This process is recursively iterated—with each reduction subject to subsequent classical pruning—producing a hybrid search tree with quantum-guided variable assignments.

3. Local Verification and Rollback Safeguards

Recognizing the stochastic and approximate nature of quantum output, QARA incorporates robust correctness mechanisms:

• Local Verification: After each quantum-determined reduction, QARA checks that the reduced instance remains viable—specifically, that every uncovered element in $S_t$7 is still covered by at least one remaining subset in $S_t$8. This verification has polynomial overhead ($S_t$9) and preserves the completeness requirement.
• Rollback: If the reduction renders the instance unsatisfiable (e.g., if any element is uncovered), the reduction is reverted and a different quantum pruning is attempted. Rollback is capped (e.g., up to $x_t$0 attempts for instance size $x_t$1), trading off computational cost and solution completeness. Even after exhausting rollbacks, QARA may conservatively accept the partially reduced instance to avoid getting stuck.

This safeguard structure ensures that quantum-induced errors in variable selection do not irreversibly eliminate potential solutions—a critical property for reliable performance.

4. Performance and Empirical Evaluation

QARA has been benchmarked comprehensively on the exact cover problem and exhibits marked improvements:

• Solution Probability: In tests with 140 instances and subset sizes from 8 to 20, QARA finds an exact cover about 60% more often than both standard QAOA and Recursive QAOA (RQAOA) at equivalent circuit depths.
• Efficiency: QARA achieves this improvement with roughly 81% fewer parameter optimization iterations compared to RQAOA. By pruning multiple variables per quantum reduction, the search tree is more aggressively reduced at each recursive step.
• Stability and Success Rate: Compared to a classical random pruning variant (Classical Recursive Reduction Algorithm, CRRA), QARA's quantum-guided reductions yield 16.4% higher single-run solution probabilities and improved mean stability in the objective value function.
• Exactness: The average objective function value in QARA solutions remains near zero, indicating that the constructed solutions satisfy the exact cover criterion.
• Scalability: QARA's recursion depth and iteration count remain manageable as problem size increases, and complexity is dominated by classical and quantum device calls per reduction.

5. Applications and Generalization

While QARA is developed and tested for exact cover—a prototypical NP-complete problem relevant to areas such as wireless network coverage—the algorithmic paradigm is portable:

• Broad NP-complete Applicability: Any combinatorial problem with local deterministic reduction rules and the ability to encode variable selection bias as expectation values can be addressed with QARA's hybrid recursion.
• NISQ-Era Suitability: QARA is expressly compatible with noisy intermediate-scale quantum (NISQ) hardware: it achieves improved performance with only shallow quantum circuits (typically depth-1 QAOA), minimizing hardware demands and noise sensitivity.
• Classical-Quantum Hybridization: The alternating structure exploits classical inference to maximal extent, resorting to quantum computation only when classical information is exhausted. This reduces quantum resource consumption and increases practical viability.
• Potential for Broader Hybrid Paradigms: The demonstrated improvement suggests hybrid, recursion-driven strategies could generalize to other graph covering, partitioning, or constraint satisfaction problems where classical and quantum techniques can systematically cooperate.

6. Algorithmic Formulation

The recursive algorithm and its core reduction rules can be expressed as:

Step	Procedure	Purpose
Classical prune	For each unique element $x_t$2 in subset $x_t$3: fix $x_t$4; invalidate overlapping subsets	Deterministic reduction using problem constraints
Quantum prune	Run QAOA; measure $x_t$5; fix $x_t$6 with strongest bias	Quantum-guided variable selection when no deterministic reductions
Rollback	After reduction, verify completeness; if necessary, revert and retry with another variable	Correctness and solution completeness

7. Broader Significance

QARA represents a rigorous advancement in quantum-classical hybrid algorithms. The data indicate that recursive alternation of classical and quantum reductions, combined with verification and controlled backtracking, can materially extend the capabilities of NISQ devices on structured combinatorial search problems beyond "standalone" QAOA or its naive recursive extensions.

These findings provide a foundation for further research into integrating tailored classical simplification heuristics with quantum-guided bias extraction, as well as formalizing correctness and efficiency guarantees for hybrid recursive frameworks in quantum optimization contexts (Ni et al., 13 Sep 2025).