Quantum Adaptive Search (QAGS)

• QAGS is a hybrid quantum-classical algorithm that integrates quantum amplitude encoding with adaptive domain contraction for global optimization in high-dimensional, continuous and discrete spaces.
• The method employs quantum state preparation, probabilistic evaluations, and adaptive thresholding to focus computational resources on the most promising regions, significantly reducing run time and memory usage.
• Validated across benchmarks in optimization, QUBO problems, and genomics, QAGS demonstrates notable improvements over classical approaches through precise amplitude amplification and iterative refinement.

Quantum Adaptive Search (QAGS) is a family of hybrid quantum-classical algorithms designed for efficient global optimization and quantum-enhanced search in both continuous and discrete high-dimensional spaces. QAGS integrates quantum state preparation and amplitude amplification—often inspired by Grover’s algorithm or its generalizations—with adaptive contraction or filtering techniques to focus resources on the most promising regions of the solution space. By utilizing quantum probability amplitude distributions to guide domain reduction or solution space pruning, QAGS achieves substantial improvements in accuracy, run time, and resource efficiency compared to both classical and naively quantum approaches (Intoccia et al., 26 Jun 2025, Sano et al., 2023).

1. Theoretical Framework and Algorithmic Structure

QAGS targets high-dimensional global unconstrained optimization, which seeks $$\mathbf{x}^* = \operatorname{argmin}_{\mathbf{x} \in \Omega_0} f(\mathbf{x})$$ for black-box functions $f: \mathbb{R}^d \to \mathbb{R}$ over a bounded domain $\Omega_0 \subset \mathbb{R}^d$. The algorithm iteratively performs the following steps (Intoccia et al., 26 Jun 2025):

1. Grid Discretization: At iteration $k$, construct a uniform grid $G_k$ over the current search box $\Omega_k$, using $m$ qubits per dimension ($M = (2^m)^d$ points in total).
2. Quantum Amplitude Encoding: For each $x \in G_k$, compute $f(x)$. Define a Boltzmann-type mapping for each point's amplitude, $f: \mathbb{R}^d \to \mathbb{R}$0, where $f: \mathbb{R}^d \to \mathbb{R}$1 and $f: \mathbb{R}^d \to \mathbb{R}$2 are the minimum and standard deviation of $f: \mathbb{R}^d \to \mathbb{R}$3 values over $f: \mathbb{R}^d \to \mathbb{R}$4.
3. Quantum State Preparation: Prepare the quantum state $f: \mathbb{R}^d \to \mathbb{R}$5 using qubit-efficient qRAM-based amplitude loading.
4. Probability Measurement and Adaptive Domain Contraction: Measure $f: \mathbb{R}^d \to \mathbb{R}$6 in the computational basis, estimate the probabilities $f: \mathbb{R}^d \to \mathbb{R}$7, and select a high-probability quantile subset $f: \mathbb{R}^d \to \mathbb{R}$8 (e.g., top-$f: \mathbb{R}^d \to \mathbb{R}$9). Project this back to $\Omega_0 \subset \mathbb{R}^d$0 to contract the domain.
5. Classical Local Refinement: Apply a classical optimizer (e.g., L-BFGS-B) to each contracted set $\Omega_0 \subset \mathbb{R}^d$1.
6. Termination: Repeat until the search box diameter falls below threshold $\Omega_0 \subset \mathbb{R}^d$2 or a maximum number of iterations $\Omega_0 \subset \mathbb{R}^d$3 is reached.

For combinatorial or QUBO-formulated problems, QAGS is realized as a Grover Adaptive Search (GAS), where instead of continuous contraction, quantum amplitude amplification is repeatedly performed with adaptively updated cost thresholds, marking solution subsets of decreasing cost (Sano et al., 2023, Tariq et al., 6 Nov 2025).

2. Quantum Encoding and Probabilistic Evaluation

The core mechanism in QAGS is the quantum encoding of solution quality. For continuous problems, the amplitude of each grid point in $\Omega_0 \subset \mathbb{R}^d$4 is exponentially stacked according to its fitness, yielding a highly non-uniform quantum probability distribution biased toward lower $\Omega_0 \subset \mathbb{R}^d$5 values (Intoccia et al., 26 Jun 2025). In discrete settings, GAS circuits encode the cost polynomial $\Omega_0 \subset \mathbb{R}^d$6 as phase rotations on value qubits, and adaptive threshold oracles perform amplitude amplifications over sublevel sets, enabling iterative focus on improving bit strings (Sano et al., 2023, Tariq et al., 6 May 2025).

Measurement yields probability estimates, $\Omega_0 \subset \mathbb{R}^d$7 in the continuous case or binary cost oracle outputs in GAS variants, that are used to select subregions or bit-strings for further iteration or classical verification.

3. Adaptive Contraction and Solution Refinement

Contraction is achieved via a quantile selection rule in the continuous domain: after estimating $\Omega_0 \subset \mathbb{R}^d$8, $\Omega_0 \subset \mathbb{R}^d$9 is shrunk to the smallest hyperrectangle containing the top-$k$0 quantile $k$1 of grid points (Intoccia et al., 26 Jun 2025). In QUBO/GAS-type settings, contraction is realized by adaptively updating Grover thresholds based on the best-so-far function values, restricting the quantum search to increasingly smaller marked sets (sublevel sets of the cost function).

This hybrid mechanism guarantees the active region’s volume satisfies $k$2 per iteration in the continuous case, leading to geometric shrinkage and, under mild continuity assumptions, to convergence of solution sequences to global minima (Intoccia et al., 26 Jun 2025). In QUBO/GAS variants, the threshold schedule ensures the expected query complexity remains $k$3 but is reduced substantially with good initialization (e.g., via MMSE), especially when only a small fraction of solutions remain marked (Tariq et al., 6 May 2025, Tariq et al., 6 Nov 2025).

4. Complexity, Resource Analysis, and Implementation

Resource requirements and runtime scaling are determined by:

• Quantum Memory: $k$4 qubits for grid encoding (continuous); $k$5 qubits for QUBO encoding (GAS).
• Classical Memory: $k$6 for grid storage or amplitude/combinatorial mapping.
• Quantum Gate Depth: $k$7 (continuous) or $k$8 rotations depending on cost polynomial order and degree (GAS) (Sano et al., 2023).
• Iterations: Number of contraction steps scales logarithmically with the contraction rate; overall wall-clock complexity is $k$9 for $G_k$0 adaptive steps and $G_k$1 measurement shots per step (Intoccia et al., 26 Jun 2025).
• Grover/Amplitude Amplification Queries: QAGS/GAS achieves $G_k$2 scaling versus $G_k$3 classical exhaustive search for binary optimization, further reduced with threshold initialization and high-order circuit optimizations (Sano et al., 2023, Tariq et al., 6 Nov 2025).

Empirical results demonstrate that QAGS can achieve up to 88.5% reduction in run time and 87.7% in memory vs. classical adaptive grid search for quadratic benchmark functions in dimension 10, and higher-order circuit optimizations (e.g., polynomial factorization, order reduction) further reduce the number of qubits, gates, and T-counts for GAS compiled on surface code platforms (Intoccia et al., 26 Jun 2025, Sano et al., 2023).

5. Applications and Benchmarking

QAGS has been validated on both continuous and discrete problems:

• Continuous Global Optimization: On Rastrigin, Styblinski–Tang, Rosenbrock, and sphere functions, QAGS achieved either exact or near-exact minima, with errors down to $G_k$4 in high dimensions. It surpassed classical grid search in both speed and RAM usage (Intoccia et al., 26 Jun 2025).
• Discrete Optimization (QUBO/GSM/MLD): QAGS/GAS outperforms classical exhaustive search in maximum-likelihood detection for RIS-assisted single-carrier and generalized spatial modulation systems. With MMSE-initialized thresholds, GAS achieves near-optimal BER with substantially fewer oracle queries and maintains performance under depolarizing and readout noise in NISQ settings (Tariq et al., 6 Nov 2025, Yukiyoshi et al., 2024, Tariq et al., 6 May 2025).
• Genomics: In DNA alignment, a QAGS variant (QiBAM) realized $G_k$5 quantum scaling for read matching and was validated via OpenQL/QX simulation for bins up to ~133 qubits (Sarkar et al., 2019).
• Resource-Efficient Search on NISQ: The ReSaQuS framework implements QAGS in constrained environments, demonstrating up to 86% reduction in cumulative qubit consumption by iterative filtering and domain compression, enabling search with limited quantum resources (Jiang et al., 2024).

6. Hybrid and High-Order Extensions

The QAGS paradigm spans a spectrum of hybridizations:

• Quantum–Classical Partitioning: Initial quantum search/preprocessing followed by classical local refinement (continuous, (Intoccia et al., 26 Jun 2025)) or classical brute-force subsearch in reduced uncertainty regions (discrete, (Sousa et al., 2015)).
• High-Order Circuit Design for QUBO/GAS: Polynomial factorization (HUBO-PF) and order-halving strategies (HUBO-OR) compact the search encoding and reduce both ancilla/qubit and gate depth requirements, closing the gap to practical QUBO deployment on current and near-term hardware (Sano et al., 2023).
• Annealer-Assisted Adaptive Search: QAGA leverages Ising model marginals from quantum annealers to contract the variable set, employing greedy variable fixing based on spin-marginal uncertainty, and empirically yields lower-energy solutions than both pure quantum annealing and multi-qubit correction post-processing (Ayanzadeh et al., 2019).

7. Limitations, Assumptions, and Outlook

QAGS depends fundamentally on the ability to efficiently encode amplitude-discriminating quantum states and to accurately estimate their output probabilities or marginal biases. Challenges include:

• State-preparation circuit depth and qRAM feasibility for large grids.
• The need for efficient oracles in discrete QAGS (e.g., Hamming-distance or cost-threshold oracles).
• Hardware resource constraints, especially for deep circuits and fault-tolerant operation.
• Tradeoffs in grid resolution ($G_k$6) versus iteration count and convergence guarantee.

Notably, all practical advances reported involve substantial quantum–classical cooperation (quantum for search/pruning, classical for refinement or subproblem solving), and most QAGS variants are shown to outperform classical and quantum-naïve alternatives in both resource use and solution quality for challenging benchmark cases (Intoccia et al., 26 Jun 2025, Sano et al., 2023, Tariq et al., 6 May 2025). Empirical and theoretical analyses support QAGS as a leading candidate for scalable, high-dimensional quantum optimization on both current and future hardware platforms.