GitHub

Metrics

Loss functions for inverse design optimization.

Overview

The Metrics module provides:

  • Field overlap: DotProduct and PowerCoupling metrics
  • Field matching: SquaredFieldDifference for complex field matching
  • Intensity matching: SquaredIntensityDifference for intensity-based objectives
  • AD-compatible: Automatic differentiation support via custom gradients

Examples

Power Coupling

using FluxOptics

# Create fields
xv, yv = spatial_vectors(128, 128, 2.0, 2.0)
u_current = ScalarField(Gaussian(15.0)(xv, yv), (2.0, 2.0), 1.064)
u_target = ScalarField(Gaussian(20.0)(xv, yv), (2.0, 2.0), 1.064)

# Power coupling metric
metric = PowerCoupling(u_target)

# Evaluate coupling (returns power in Watts)
power_coupled = metric(u_current)
power_coupled[]
0.9215999999999991

Field Matching

using FluxOptics

# Create fields with different phases
xv, yv = spatial_vectors(128, 128, 2.0, 2.0)
gaussian = Gaussian(15.0)(xv, yv)
u_current = ScalarField(gaussian, (2.0, 2.0), 1.064)
u_target = ScalarField(gaussian .* cis.(0.01 .* (xv.^2 .+ yv'.^2)), (2.0, 2.0), 1.064)

# Squared field difference
metric = SquaredFieldDifference(u_target)
loss = metric(u_current)
loss[]
0.27931034482758627

Intensity Matching

using FluxOptics

# Create field and target intensity pattern
xv, yv = spatial_vectors(128, 128, 2.0, 2.0)
u_current = ScalarField(Gaussian(35.0)(xv, yv), (2.0, 2.0), 1.064)

# Target: ring pattern
r = sqrt.(xv.^2 .+ yv'.^2)
target_intensity = exp.(-(r .- 30).^2 / 100)
target_intensity ./= sum(target_intensity) * 2.0 * 2.0

# Intensity matching metric
metric = SquaredIntensityDifference((u_current, target_intensity))
loss = metric(u_current)
loss[]
5.9310733173905715e-5

Key Types

Key Functions

See Also

Index