A quantum pulse compiler operating in a new category
Every tool in production today optimizes pulses per qubit instance. We compile for entire families of qubits from a single training run. Generation time per pulse: 0.6 microseconds, at infidelity 10-11 — three orders below the hardware noise floor.
instant generation for any qubit in the family
What is shown in this dossier is a presentation of measured results. The technology itself is transferred only under an agreed technological cooperation, accepted by all parties involved. What can be made available under such an agreement extends beyond what is presented here.
Per-instance optimization vs. parametric compilation
Every tool in use today (Q-CTRL Boulder Opal, QuTiP, Krotov, Quandary, Quanlse) works inside the paradigm one qubit → one optimization → one pulse. Every new qubit means re-optimizing from scratch. Nobody has a parametric compiler that separates the learning phase (once) from the generation phase (instant, per instance).
| Metric | Q-CTRL / QuTiP / Krotov | Parametric compiler |
|---|---|---|
| Compilation time per pulse | 458 ms | 0.6 µs |
| Recompilation for a new qubit | from scratch | instant |
| 1,000 distinct qubits | ~7.6 minutes | 0.6 ms |
| Infidelity, unitary propagation | 10⁻¹³ | 10⁻¹¹ |
| With decoherence (realistic T1/T2) | 2,269 ms | 0.48 µs |
| Mental model | optimizer | compiler |
Three of four physical parameters cost nothing to change
Of the four parameters that describe a device in this family, three can be varied at no computational cost once the family has been compiled. The fourth is the only one that carries any work at all, and it is represented exactly rather than approximated. What follows is measured consequence, not argument.
| Consequence | What it replaces | Measured |
|---|---|---|
| A pulse for a device not seen before costs no optimization | a full GRAPE run | 0.6 µs |
| Adding decoherence does not change the cost | a Lindblad optimization | 0.48 µs |
| Drift between experiments needs no offline recalibration | a recalibration cycle | instant |
| T1/T2 rates can be varied like any other parameter | re-solving per rate | no cost |
| Interpolated devices below experimental detection threshold | — | 300 / 300 |
| The one-off offline cost is repaid after | — | 12 devices |
The construction rests on seven structural identities, each verified numerically against direct computation. The largest deviation across all seven is below 10⁻¹² — machine precision. Four of them were verified a second time under an open-system master equation with realistic decoherence rates, with no loss of precision. Everything downstream of the offline phase runs without an optimizer, without iterations and without gradients, which is the structural difference from any method built on per-instance numerical optimization. The identities themselves, and the construction they support, are held as trade secret and are communicated within a cooperation.
The same approach applies to each hardware family
The pulse compiler is the first visible application, not the last. The approach applies in principle to every type of quantum hardware that carries an analogous parameter family — photons, 2D Rydberg atoms, silicon spin qubits, majorana. Each one is a new family to compile.
Conditions of validity
- What is established here covers one superconducting device family, end to end, from the parameter description to the generated control pulse.
- Measured under unitary propagation and, separately, under an open-system master equation with realistic decoherence rates.
- The construction is closed-form. It has no optimizer inside it and no hyperparameters that have to be tuned per device.
- Transfer to a different hardware family requires the corresponding preparatory work for that family. The construction is general; the preparation is per-family.
Established by measurement, not by argument
- Seven structural identities verified at machine precision, the largest deviation below 10⁻¹².
- The one-off offline cost is repaid after 12 devices.
- 300 of 300 interpolated devices below 10⁻⁶ infidelity.
- Validation on physical quantum processors is not part of this dossier.