Agent-Drivable Quantum Layer with Trajectory Diagnostics on a Consumer Mobile GPU
We present CUDANT_VQPC, a virtual quantum computer that runs as a Model Context Protocol (MCP) server on a consumer mobile workstation (Lenovo P53, NVIDIA Quadro RTX 4000, 8 GB VRAM), exposing 10 tools to AI agents. The system provides two complementary backends — a geometric "cage" model (intuitive, shows trajectory and wobble) and an exact 2^n state-vector "engine" (1–16 qubits, full entanglement) — with an agent-decidable selection rule. We introduce a wobble metric (maximum path deviation from the chord on the Bloch sphere) as a circuit-quality diagnostic that gives the agent a confidence signal independent of the measurement outcome. The system supports pre-built algorithms (Grover search, Bell pairs, GHZ states, QFT, phase estimation) and a differentiable learning layer where gate angles are nn.Parameters trained by gradient descent. All results run in ≤512 KB VRAM. We demonstrate 16-qubit Grover search (P(\text{target})=0.999989), exact phase estimation (P(\text{answer})=1.0), and a 16-qubit differentiable layer converging to P(\text{target})=1.0 in 200 steps — all on a laptop GPU.
Authors
- Jose Luis Van Houtven Escobar (ORCID: https://orcid.org/0009-0000-7806-2082)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24270248
- Primary Topic
- Quantum Computing Algorithms and Architecture
- Type
- preprint