Inverse Design API

Coming in V4

The inverse design API is planned for MetoSim V4 (Q4 2026). This page documents the planned interface.

Overview

The inverse design API provides endpoints for running adjoint-based topology optimization on cloud GPUs. It extends the simulation API with iterative optimization loops.

Planned Endpoints

Method	Endpoint	Description
POST	/v1/optimizations	Start an optimization run
GET	/v1/optimizations/{id}	Check optimization status + current iteration
GET	/v1/optimizations/{id}/results	Download optimized geometry
POST	/v1/optimizations/{id}/stop	Early-stop a running optimization
GET	/v1/optimizations/{id}/history	Iteration-by-iteration loss history

Planned Request Format

{
  "target": {
    "monitor": "transmission",
    "wavelength": 1.55e-6,
    "target_value": 0.95,
    "objective": "maximize"
  },
  "design_region": {
    "center": [0, 0, 0],
    "size": [4e-6, 4e-6, 0.22e-6],
    "materials": ["Air", "Si"],
    "min_feature_size": 100e-9
  },
  "optimization": {
    "method": "adjoint_gradient",
    "max_iterations": 100,
    "learning_rate": 0.01,
    "convergence_threshold": 1e-4
  },
  "simulation": {
    "domain_size": [8e-6, 8e-6, 4e-6],
    "resolution": 20e-9,
    "fdtd_settings": { "time_steps": 20000 }
  }
}

Planned Response

Each iteration produces:

{
  "optimization_id": "opt-abc123",
  "status": "RUNNING",
  "current_iteration": 42,
  "max_iterations": 100,
  "current_loss": 0.0823,
  "best_loss": 0.0712,
  "convergence_history": [0.45, 0.32, 0.21, ...],
  "estimated_remaining_time": 1200
}

WebSocket Streaming

For real-time optimization monitoring, a WebSocket endpoint will stream iteration updates:

wss://api.metosim.io/v1/optimizations/{id}/stream

The SDK will handle this automatically with a progress callback:

result = metosim.inverse_design(
    target=target,
    design_region=region,
    on_iteration=lambda i, loss: print(f"Iter {i}: loss={loss:.4f}"),
)

Cost Estimation

Each iteration requires 2 FDTD simulations (forward + adjoint). A typical 100-iteration optimization on a 200³ grid:

• ~200 FDTD runs × ~4 min each = ~13 hours GPU time
• Estimated cost: ~$325 at $25/hour

Surrogate model acceleration (V4) can reduce this by 10-100×.