Boundary handling

When a particle steps outside the search box, the optimizer brings it back. How it does that is controlled by bounds_handling; each strategy changes the search dynamics near the edges.

Strategy	What it does
"clamp" (default)	Clip the position to the boundary. Simple; particles can pile up on the edge.
"reflect"	Bounce the position back into range and flip the offending velocity component.
"wrap"	Wrap around toroidally to the opposite side — useful for periodic domains.
"reinit"	Re-sample the offending component uniformly within bounds and zero its velocity (adds diversity).

import turboswarm as pso

r = pso.minimize("rastrigin", bounds=[(-5.12, 5.12)] * 2,
                 bounds_handling="reflect", seed=1)

All strategies guarantee the reported best_position lies within the bounds. The default is "clamp", so existing behavior is unchanged unless you opt in.

From Rust

The strategy lives in PsoParams; the built-in spaces implement all four. A custom SearchSpace that does not override enforce_bounds falls back to clamping.

use turboswarm_core::prelude::*;

let params = PsoParams {
    bounds_handling: BoundaryHandling::Reflect,
    ..Default::default()
};