Add optimization direction to invdes.Optimizer

CHANGELOG.md

@@ -13,6 +13,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Support for `dilation` argument in `JaxPolySlab`.
 - Support for autograd differentiation with respect to `Cylinder.radius` and `Cylinder.center` (for elements not along axis dimension).
 - `Cylinder.to_polyslab(num_pts_circumference, **kwargs)` to convert a cylinder into a discretized version represented by a `PolySlab`.
+- `tidy3d.plugins.invdes.Optimizer` now accepts an optional optimization direction via the `maximize` argument (defaults to `maximize=True`).
 
 ### Changed
 - `PolySlab` now raises error when differentiating and dilation causes damage to the polygon.

tidy3d/plugins/invdes/design.py

@@ -39,10 +39,12 @@ class AbstractInverseDesign(InvdesBaseModel, abc.ABC):
     )
 
     def make_objective_fn(
-        self, post_process_fn: typing.Callable
+        self, post_process_fn: typing.Callable, maximize: bool = True
     ) -> typing.Callable[[anp.ndarray], tuple[float, dict]]:
         """construct the objective function for this ``InverseDesignMulti`` object."""
 
+        direction_multiplier = 1 if maximize else -1
+
         def objective_fn(params: anp.ndarray, aux_data: dict = None) -> float:
             """Full objective function."""
 
@@ -52,7 +54,7 @@ def objective_fn(params: anp.ndarray, aux_data: dict = None) -> float:
             post_process_val = post_process_fn(data)
 
             penalty_value = self.design_region.penalty_value(params)
-            objective_fn_val = post_process_val - penalty_value
+            objective_fn_val = direction_multiplier * post_process_val - penalty_value
 
             # store things in ``aux_data`` passed by reference
             if aux_data is not None:

tidy3d/plugins/invdes/optimizer.py

@@ -31,6 +31,12 @@ class AbstractOptimizer(InvdesBaseModel, abc.ABC):
         description="Step size for the gradient descent optimizer.",
     )
 
+    maximize: bool = pd.Field(
+        True,
+        title="Direction of Optimization",
+        description="If ``True``, the optimizer will maximize the objective function. If ``False``, the optimizer will minimize the objective function.",
+    )
+
     num_steps: pd.PositiveInt = pd.Field(
         ...,
         title="Number of Steps",
@@ -103,7 +109,7 @@ def continue_run(
         history = deepcopy(result.history)
 
         # use autograd to take gradient the objective function
-        objective_fn = self.design.make_objective_fn(post_process_fn)
+        objective_fn = self.design.make_objective_fn(post_process_fn, maximize=self.maximize)
         val_and_grad_fn = ag.value_and_grad(objective_fn)
 
         # main optimization loop