ENTRY FOR THE 23rd ANNUAL (2026) "HUMIES" AWARDS FOR HUMAN-COMPETITIVE RESULTS PRODUCED BY GENETIC AND EVOLUTIONARY COMPUTATION =========================================================== ITEM 1: PAPER TITLE -------------------- Evolving Hardware-Efficient Grover Circuits with Grammatical Evolution ITEM 2: AUTHOR CONTACT INFORMATION ----------------------------------- Author 1: Name: Arinze Obidiegwu Address: University of Limerick, Limerick, V94 T9PX, Ireland Email: arinzeobidiegwu@gmail.com Phone: +353 87 030 9305 Author 2: Name: Douglas Mota Dias Address: ATU Galway City, Dublin Road, Galway City, H91 T8NW, Ireland Email: douglas.motadias@atu.ie Phone: +353 83 147 4321 Author 3: Name: Emmanuel Obidiegwu Address: 9 Racecourse Park, Dublin 13, D13 F25F, Ireland Email: blaise.zaga@gmail.com Phone: +353 87 360 3339 Author 4: Name: Conor Ryan Address: University of Limerick, Limerick, V94 T9PX, Ireland Email: conor.ryan@ul.ie Phone: +353 86 241 8791 ITEM 3: CORRESPONDING AUTHOR ----------------------------- Arinze Obidiegwu ITEM 4: ABSTRACT ---------------- Canonical quantum algorithms are designed for generality on idealised hardware. On current NISQ devices, this generality incurs a substantial cost. Standard implementations of Grover's search, for example, produce deep, gate-heavy circuits that are highly susceptible to noise, resulting in low execution fidelity on real hardware. We take a different approach. Using Grammatical Evolution (GE), we automatically discover hardware-efficient, state-specific quantum circuits. We evolve bespoke circuits for all eight 3-qubit computational basis states and execute them on a 133-qubit IBM Heron processor. The results are clear. The evolved circuits achieve hardware fidelities of up to 96.9%, compared to 66.3% for the canonical baseline compiled using IBM's Qiskit pipeline. At the same time, circuit depth is reduced by 82.5-96.6%, and gate count is reduced by 77.4-94.6%. These results show that automated symbolic search can discover circuit structures that are significantly better aligned with real hardware constraints, and can outperform standard algorithmic implementations in practical NISQ settings. ITEM 5: CRITERIA CLAIMED ------------------------- C ITEM 6: STATEMENT OF WHY THE RESULT SATISFIES CRITERION C ---------------------------------------------------------- We claim criterion (C): the result is better than the most recent human-created solution to a long-standing problem of indisputable difficulty in its field. THE PROBLEM AND ITS DIFFICULTY: Executing quantum algorithms faithfully on real quantum hardware is a central, well-documented challenge in quantum computing. Current Noisy Intermediate-Scale Quantum (NISQ) devices suffer from short coherence times, limited qubit connectivity, and significant gate errors. These physical constraints cause textbook quantum algorithms designed for idealised, fault-tolerant hardware to produce output fidelities that are often indistinguishable from random noise when run on real devices. The severity of this gap is established by high-profile, independent benchmarks. Kim et al. (2023), in a landmark Nature paper, presented the first IBM-internal evidence that NISQ devices can produce useful results but only under carefully managed circuit depths. That work underscores the central constraint: on current hardware, circuit depth is the dominant bottleneck for computational utility. This fidelity gap is further documented as one of the primary obstacles to extracting value from near-term quantum hardware (Preskill, 2018; Bharti et al., 2022; Tannu and Qureshi, 2019). THE HUMAN-CREATED BASELINE: The baseline is not a naive implementation. It is the canonical Grover algorithm as implemented in IBM's Qiskit SDK, transpiled at optimization_level=3 the highest default setting in the standard compiler. This pipeline includes extensive gate cancellation, commutation, and routing optimisations, and uses the same workflow as IBM Quantum for benchmarking. It therefore represents a production-grade baseline rather than a simplified reference. When executed on the 133-qubit IBM Heron processor (ibm_torino) with 10,000 shots, these circuits achieved fidelities ranging from 66.3% to 79.5%. HOW THE EVOLVED RESULT IS BETTER: Grammatical Evolution discovers circuits that substantially outperform this baseline across all metrics. All comparisons are made on identical target states, executed on the same hardware under the same measurement conditions. - Fidelity: Evolved circuits achieved 88.4% to 96.9%, compared to 66.3% to 79.5% for the optimised canonical circuits. On state |110>, the canonical baseline achieved 66.3% fidelity; the evolved circuit on the same state achieved 96.9% a 30.6 percentage point improvement on identical hardware under identical measurement conditions. - Circuit depth: Reduced by 82.5% to 96.6%. For example, the canonical |000> circuit has depth 145 after full optimisation; the evolved circuit has depth 5. - Gate count: Reduced by 77.4% to 94.6%. The canonical |110> circuit uses 182 gates; the evolved circuit uses 10. - Statistical significance: The Mann-Whitney U test confirmed the performance gap is statistically significant (p = 2.3 x 10^-4 for |000>, p < 0.01 for all states with Hamming weight <= 2), ruling out measurement noise or evolutionary variance as explanations. These improvements are not incremental. The evolved circuits operate well within the coherence window of the hardware (execution time ~3 microseconds vs. T2 ~98 microseconds), whereas the canonical circuits approach or exceed the decoherence limit (~80 microseconds), which is the fundamental physical mechanism causing their failure. THE ARMS-LENGTH STANDARD: The difficulty of the problem is established entirely external to the authors. The NISQ execution fidelity gap is documented in peer-reviewed literature by independent research groups: Kim et al. (2023, Nature -- IBM Research), Preskill (2018 -- Caltech), Bharti et al. (2022 -- multi-institutional review), and Tannu and Qureshi (2019 -- Georgia Tech). The baseline compiler (Qiskit) was developed by IBM, not by the authors. The hardware validation was performed on IBM's ibm_torino processor, accessed through the IBM Quantum Network. Both the problem definition and the baseline are entirely independent of the authors' own work. ITEM 7: FULL CITATION ---------------------- Arinze Obidiegwu, Douglas Mota Dias, Emmanuel Obidiegwu, and Conor Ryan. 2026. Evolving Hardware-Efficient Grover Circuits with Grammatical Evolution. In Genetic and Evolutionary Computation Conference (GECCO '26), July 13-17, 2026, San Jose, Costa Rica. ACM, New York, NY, USA, 9 pages. https://doi.org/10.1145/3795095.3805176 ITEM 8: PRIZE MONEY DIVISION ----------------------------- Any prize money, if any, is to be divided equally among the co-authors. ITEM 9: STATEMENT OF WHY THIS ENTRY SHOULD BE CONSIDERED "BEST" ---------------------------------------------------------------- This entry merits consideration as the best for three reasons: real-hardware validation, the magnitude of improvement, and the broader methodological significance. 1. HARDWARE-VALIDATED, NOT SIMULATED: Every result in this paper was executed on a 133-qubit IBM Heron quantum processor (ibm_torino) with 10,000 measurement shots per experiment. The evolved circuits were not merely theoretically superior; they demonstrably worked on a physical device where the canonical algorithm demonstrably failed. This hardware validation distinguishes the entry from the majority of evolutionary quantum circuit work in the literature and establishes a direct, measurable impact on a real engineering problem. 2. MAGNITUDE AND CONSISTENCY OF IMPROVEMENT: The performance gains are not marginal. The best evolved circuit achieves 96.9% fidelity where the fully optimized canonical circuit achieves 66.3% a result that crosses the threshold from "computationally useless" to "near-perfect." Circuit depth reductions of 82-97% and gate count reductions of 77-95% were achieved consistently across all eight 3-qubit target states. The evolved circuits use as few as 1-2 two-qubit CZ gates (the dominant error source on this architecture) compared to 20+ in the canonical implementations. The evolutionary process achieved this implicit noise minimization without any explicit noise model in the fitness function the parsimony pressure on gate count alone was sufficient to drive the search toward hardware-efficient solutions. 3. ALGORITHM DISCOVERY, NOT PARAMETER TUNING: The GE framework does not operate within a fixed circuit template (as in variational approaches). Instead, it searches directly over circuit structure. As a result, it finds qualitatively different solutions rather than incremental optimisations of a given design. In several cases, the evolved circuits diverge significantly from the canonical Grover structure. For example, for the |000> target, evolution removes both the Hadamard initialisation layer and the diffuser, effectively replacing the Grover iteration with a direct state-preparation unitary. This is not a contradiction of Grover's algorithm. It highlights its cost. When the target state is fixed and known, the generality of Grover introduces overhead that is unnecessary on real hardware. The evolutionary process shows that this generality can be traded for much higher hardware efficiency. A similar pattern appears for separable states such as |010>. Here, the search independently discovers that no entanglement is required, reducing the circuit to a tensor product of single-qubit rotations. More broadly, this illustrates a key difference between compilation and search. Standard transpilers preserve the structure of a given algorithm and optimise its implementation. They do not attempt to discover alternative structures. In contrast, the evolutionary process operates at the level of representation, allowing it to uncover circuits that are structurally different and better aligned with hardware constraints. The resulting circuits are therefore not simply optimised versions of Grover's algorithm. They are alternative realisations of the same computational objective, but far better suited to NISQ execution. This work builds on earlier symbolic search approaches in quantum computing, beginning with Spector's (2004) foundational work on automatic quantum computer programming using genetic programming, and including Leier and Banzhaf (2003) and Stadelhofer, Banzhaf, and Suter (2008), which showed that evolutionary methods can discover quantum algorithms in noiseless simulation. Here, we show that these ideas still hold under real-device constraints. Instead of simulation, the evolved circuits are executed directly on NISQ hardware, where noise, not idealised fidelity, is the dominant limitation. The combination of these factors validated on real quantum hardware, large and consistent improvements over the best available human-engineered baseline, and the discovery of novel algorithmic structures that extend a well-established symbolic-search tradition makes this entry a strong candidate for the award. ITEM 10: TYPE OF EVOLUTIONARY COMPUTATION ----------------------------------------- GE (Grammatical Evolution) ITEM 11: PUBLICATION DATE AND IN-PRESS STATUS ---------------------------------------------- Publication date: July 13-17, 2026 (GECCO 2026 conference) This paper is "in press." It was unconditionally accepted as a full paper for GECCO 2026 on March 20, 2026. The camera-ready version has been submitted. Documentation of acceptance (GECCO 2026 Paper Decision Notification email) is included with this submission. =========================================================== END OF ENTRY