Quantum Computing Research Programs
Q-Psi conducts open-science quantum compiler research, physical-QPU algorithm evaluation, and reproducible benchmark studies on real superconducting hardware (IBM Heron r2, 156 programmable qubits).
Primary Physical-QPU Evidence
COMPILER + GROVER QUERY ADVANTAGE
Demonstrated compiler-enabled physical quantum query advantage on structured search spaces on IBM Heron hardware: query reduction of 2.43x (N=4), 1.73x (N=8), and 1.19x (N=16) across 9/9 cases and 3/3 problem sizes.
DYNAMIC BERNSTEIN–VAZIRANI
Statistically significant physical quantum query complexity scaling advantage on N=4..16 qubits (slope 0.153 vs classical 0.696, t = -30.65, p = 3.47e-7).
MCM QUBIT REUSE
Qubit reuse compiler reduced register footprint by median 78.1% (up to 87.5% on N=16) with 4/6 reuse wins and 6/6 non-inferior cases on physical observable RMSE.
QUANTUM DATA FOUNDRY V2
332,768 physical QPU samples across 4 jobs sampled 65,138 unique candidate states, yielding 93 verifier-backed positive repair records (1.593 MB, ~417k tokens) with 100% QPU provenance.
MQT CROSS-COMPILER BENCHMARK
18/18 valid compilations across MQT suite; physical execution shows mixed competitiveness (disfavored vs Qiskit by -0.0067 delta, favored vs TKET by +0.0398 delta driven by vqe_su2_n8).
Hardware-Aware Compiler Studies & Baselines
DYNAMIC ROUTING CROSSOVER
Dynamic routing compressed 2Q depth from 73 to 2 on 13 hops; unitary routing won on short/medium distances (2/2 confirmed predictions); empirical physical crossover was not resolved.
STAGE 6F PHYSICAL OPTIMIZATION
4/4 exact optimum hits on small compiler graphs (N<=10), matching classical branch-and-bound; degraded to 0/4 hits on N=18..25 due to hardware noise.
QUANTUM DATA FOUNDRY V1
Initial feasibility demonstration of physical QPU selector on M=1024 space yielding 14 verified unique code repair records.
MULTI-BACKEND RETARGETING & SELECTION
Across six canonical workloads and three IBM Heron r2 backends, Q-Psi successfully retargeted all 18 workload-backend combinations (100.0%). Its pre-QPU selection heuristic placed the selected backend in the physically best statistical tier for 5 of 6 workloads (1 clear win, 4 ties, 1 miss).
Preserving Negative & Inconclusive Evidence
CALIBRATION-AWARE LAYOUT VS QISKIT
Q-Psi calibration layout achieved 1 win, 2 losses, 3 inconclusive vs Qiskit O3 best-of-8 baseline (mean delta -0.0106); hypothesis NOT SUPPORTED.
FRACTIONAL GATE LOWERING
Fractional pulse gate lowering achieved structural duration reduction and 2/6 physical wins (N4_B, N8_B), with 4/6 inconclusive cases; classified as mixed physical signal.
DIRECTED EXECUTION SELECTION
Selective Pauli twirling twirled only 41.1% of CZ layers on IBM Directed Execution beta mode; achieved 1 win vs unmitigated RAW baseline; comparison vs full twirling was inconclusive.
RESTRICTED SIMON STUDY
6/12 instances recovered on physical hardware (N=8..28) with polynomial query reductions, but recovery failures at higher N due to noise prevent establishing asymptotic speedup.
STATIC BV & MANTRA ENCODING
Exploratory demonstration of 5-qubit static Bernstein-Vazirani and state preparation on physical hardware.
STAGE 6C-R1 STATE PRUNING
Deterministic changed-file state compiler achieved 98.4% candidate reduction with zero ground truth leakage on developmental benchmark universe.
Strict Pre-Registration & Deterministic Reproducibility
All 14 completed physical campaigns were executed under strict pre-QPU protocol freezes, zero post-hoc parameter adjustments, and SHA-256 evidence hashing. 14/14 automated audit scripts reproduce all reported numbers with zero discrepancies.