Scientific computing and quantum research
Quantum simulations that know what they rest on.
A quantum simulation can converge cleanly and still be wrong: the noise model is stale, the Hamiltonian disagrees with the hardware, a log is missing. Tauss tracks what every result rests on and says when it no longer holds.
GammaLex is currently collaborating on quantum simulation research with quantum researchers at New Mexico State University. Together we are testing Tauss on computational physics and quantum hardware, where one stale noise model or calibration can change what a result means.
Every quantum simulation rests on three things
Each can break without changing a single number in the output.
The model
Hamiltonian, couplings, gate parameters
How it breaks
The control model says a gate takes 32 ns. The hardware spec says 28.
The noise
Calibration, decoherence times, drift
How it breaks
The noise model was fitted before the last pulse-sequence change.
The provenance
Code version, inputs, logs, vendor libraries
How it breaks
The cooldown this run depends on was never logged.
What Tauss does for a lab
Six things it does around every run, whichever tools the lab already uses.
Assumptions, tracked
Every result is tied to the model, noise and parameters it was computed under.
Drift, caught
When the device is recalibrated, the simulations fitted to the old state are flagged.
Model against hardware
Disagreements between the control model and the spec are named, parameter by parameter.
Provenance for every run
Code version, inputs and logs travel with the result, so it can be reproduced or retracted.
Restricted data respected
A vendor library this lab may not open is noted, and its contents stay out.
Any model, any stack
Works beneath whichever AI system and simulation tools the lab already uses.
Can today’s readout be trusted?
Any simulation fitted to this device inherits these five records. Four are broken, and the readout alone will never show it. Run the procedure and watch the answer change.
Today’s readout
Not yet. The readout cannot be trusted.
1/5records hold
Procedure
- 01Re-run calibration
- 02Log the cooldown
- 03Settle model vs spec
- 04Release the pulse library
Calibration drift
driftingReadout
loggedCooldown log
gapC7, this run’s cooldown, was never logged.
Model against spec
disagree| Parameter | Model | Spec |
|---|---|---|
| Drive amplitude | 0.42 | 0.42 |
| Readout frequency | 6.215 GHz | 6.215 GHz |
| Gate duration | 32 ns | 28 ns ≠ |
Vendor pulse library
heldRestricted
Contents withheld from this actor.
Running simulations or quantum hardware?
We are looking for labs that want every result to say what it rests on.