Surrogate prediction and NK correction
Compare a configurable Surrogate prediction, Newton–Krylov correction, and an optional cold-start ADflow reference on the same mesh.
Public research preview over HTTP. Traffic is not encrypted. Do not upload confidential, proprietary, personal, or otherwise sensitive geometries or data.
Timing on this shared research server. Limited hardware and queueing can make this public Demo slower than the formal benchmark setting. Cold-start ADflow convergence is case-dependent and its wall time may fluctuate substantially, typically from 10–100 s; Surrogate + NK typically takes 3–10 s. Optimization speedup is likewise strongly dependent on the case and optimization trajectory. Queue waiting time is reported separately from solver wall time.
Define the case
Select, upload, or deform a two-dimensional airfoil.
Airfoil geometry
Custom mode exposes draggable deformation points.
Compare predictions
Completed methods remain visible on common axes and colour scales.
Surface pressure coefficient
Raw wall-cell values are shown without curve smoothing so shock discontinuities remain visible. The shared trailing-edge closure is the mean of the final upper/lower values.
Run the Surrogate to populate the comparison.
Surface skin-friction coefficient
Raw wall-face Cf from the ADflow-aligned viscous-traction operator; positive is downstream wall shear.
Run the Surrogate to populate the comparison.
Case metrics
Cm is referenced at x/c = 0.25 with nose-up positive. Signed Δ values and absolute relative errors appear after ADflow finishes; expand CD to inspect pressure and viscous components.
| Metric | Surrogate | Surrogate + NK | ADflow |
|---|
Pressure fields
Switch among the five predicted channels; all methods use the same configurable scale.
Awaiting result
Awaiting result
Awaiting result
Absolute pressure error against ADflow
Surrogate and corrected fields use the same configurable error scale.