IBM Heron · ibm_fez · 46 qubits · Mastoparan I (IDWKKLLDAAKQIL)

A single-pass polycontextural folding circuit
fully audited

We compile Miyazawa-Jernigan folding physics into one polycontextural circuit and run it once per mode on IBM Fez (46 logical qubits). This page is not an energy record — it is a compilation, portability, and audit demonstration, published with every tier and a no-quantum control so results can be checked rather than trusted.

+13.5
QPC raw best (Tier A)
NISQ on heavy-hex
−4.19
IonQ raw mean (Tier A)
their per-shot signal
−9.74
QPC after repair (Tier D)
−10.65
Random control, same repair
no quantum — scores lower

Read the control first. Uniform random bitstrings through our same Tier D repair reach −10.65 — below both Eref (−8.698) and quantum −9.74. Post-repair energy is not a quantum differentiator on this instance; the repair routine does the work. We report −9.74 because hiding it would be dishonest, not because it demonstrates advantage.

What this run does and does not show

Supported by the data (what we claim)

  • Single-pass compilation. Geometry-contact coupling in one fixed schedule via transjunctions — not iterative BF-DCQO. Eight IBM job IDs are public.
  • Portability. The same polycontextural layer exports to IBM, IonQ, IQM, Pasqal, and Origin — an architecture property independent of this energy outcome.
  • Full auditability. Tiers A-D separately, plus random-repair control (IonQ Fig. 4 analogue), with public job IDs.

Not supported (what we do not claim)

  • No energy advantage. Random repair (−10.65) beats QPC repair (−9.74).
  • No raw-sample advantage. QPC Tier A best (+13.5) does not beat random Tier A best (+8.77, seed 7). IonQ raw mean (−4.19) is the genuine per-shot quantum signal here.
  • Pool concentration claim withdrawn where it compared our best to random mean; matched best-vs-best does not favor QPC on Tier A.

What the IonQ team did

Kipu Quantum and IonQ reported BF-DCQO for protein-inspired models on trapped-ion hardware (46–61 qubits), using multi-body HUBO encodings and a hybrid loop: repeated quantum executions plus classical consensus / repair over samples (their Fig. 4 addresses when repair alone erases the quantum signal). arXiv:2604.26861

IonQ stack (quantum + classical)

What QPC does in one pass

QPC does not implement BF-DCQO. It compiles the same Miyazawa–Jernigan physics into a fixed polycontextural circuit and submits it once per mode to IBM Runtime — geometry and contact structure encoded in one schedule, not discovered through iterative counterdiabatic rounds.

Circuit construction, transjunction wiring, and gate schedules are proprietary. Public evidence on this page is limited to tier energies, random-repair controls, and verifiable IBM job IDs.

Results with audit controls (40 960 Fez shots)

Tier / metric IonQ BF-DCQO (literature) QPC Fez (8 jobs merged) Random control (no quantum)
Tier A — best raw sample Mean ≈ −4.19 (strong per-shot ions) +13.5 (NISQ noise on heavy-hex) +8.77 best (seed 7); +8.8 to +15.8 range
Pool quality — mean of best 200 unique outcomes before repair Elite pool mean lower than uniform random (~31) — secondary signal only ~31 (uniform random; seeds 42, 7, 2026)
Tier D — same repair on sample pool Consensus ≈ −8.698 −9.74 −7.96 / −10.65 / −9.74 (seeds 42, 7, 2026)
Eref (reference fold) −8.698
Verifiability Published 8 job IDs Random Tier D control published with audit page
Audit (IonQ Fig. 4 analogue): We applied the same Tier D repair to 40 960 uniform random bitstrings (no IBM jobs). Best random seed reaches −10.65, so Tier D by itself is not the quantum differentiator. QPC's architectural claim on this page is single-pass polycontextural encoding + portable circuits + open controls — not beating IonQ on folding energy on Fez at this depth.

Honest summary: IonQ leads on raw quantum per shot (ions + iterative BF-DCQO). QPC contributes a compact, portable, single-pass encoding with every tier and random control published. Random repair (−10.65) undershoots QPC repair (−9.74) — report both at equal weight.

Six-sequence panel (paper Table I): The IonQ/Kipu study benchmarks six peptides (46–61 qubits). This release reports the full audited Fez campaign on IDWKKLLDAAKQIL only. The remaining five sequences were not run on IBM hardware here because of cloud cost (each additional peptide needs many Runtime jobs at full width). Classical random-repair controls for other sequences were not published without matched quantum shots.

Eref vs Tier D best (−9.74): IonQ’s arXiv:2604.26861 gives Eref = −8.698 from a converged classical genetic algorithm on the same Robert et al. tetrahedral MJ Hamiltonian — a strong benchmark, not a proof that no lower energy exists in our discrete encoding. Their conclusion calls for quantum workflows more aware of folding structure; that is a modeling goal, not a unique “golden” energy. Our Tier D score applies the same repair code to Fez bitstrings; a value below Eref can mean repair found a lower-cost state in the sampled subspace, not necessarily a new global fold better than the paper’s GA. The random audit (best Tier D −10.65 without quantum) shows classical repair alone can undershoot Eref on this instance — so sub-Eref Tier D is not by itself evidence of a better physical minimum.

Why QPC for complex optimization

QPC on IonQ hardware (already demonstrated)

This page compares energies on IBM Fez. QPC has already run verifiable workloads on IonQ Forte (Azure Quantum, Braket, portfolio optimization, RCS). A natural next step is the same 46Q polycontextural protein circuit on IonQ — where all-to-all coupling may improve Tier A raw scores while keeping the single-pass schedule.

Invitation to the IonQ / Kipu line of work

The trapped-ion BF-DCQO result is a major achievement. QPC proposes a complementary question: can the same folding physics be encoded as a fixed polycontextural circuit on any QPU, with open job IDs and a no-quantum control? We publish the random-repair control so reviewers can ask the same questions IonQ answered in Fig. 4. Joint benchmarks on IonQ using exported QPC circuits are welcome.

Artifacts & full data

Sequence IDWKKLLDAAKQIL · Backend ibm_fez

→ Full results tables, all 8 job IDs, audit seeds