Struct LayerStackResidual 

pub struct LayerStackResidual {
    ks: Vec<[Complex; 2]>,
    q: Vec<[f64; 2]>,
    thickness: Vec<f64>,
    channels: Vec<Channel>,
    fixed_q: bool,
}

What layer_stack saves for its pullback: the solves of every transverse wavevector, each on its own, without a dense matrix over all of them.

The pullback adds the medium and thickness gradients of all wavevectors in chunks of consecutive wavevectors fixed by their count, and the chunk sums in chunk order (numerics::parallel::try_fold_ordered), so the thread count does not change them. Each q gradient belongs to one wavevector, which writes it in place.

Fields

ks: Vec<[Complex; 2]>

q: Vec<[f64; 2]>

thickness: Vec<f64>

channels: Vec<Channel>

fixed_q: bool

Implementations

impl LayerStackResidual

pub fn state_size(media: usize, channels: usize) -> Result<usize>

Fixed state byte count for the media and independent transverse channels.

impl LayerStackResidual

pub fn medium_count(&self) -> usize

Number of media, including the exterior half-spaces.

pub fn channel_count(&self) -> usize

Number of independent transverse channels.

pub fn pushforward( &self, ks: &[[Complex; 2]], zs: &[Complex], q: &[[f64; 2]], thickness: &[f64], ) -> Result<Vec<Blocks>>

Propagate one parameter direction through the independent two-polarization channels, reusing every saved interface and internal-field factorization.

pub fn pullback(&self, cotangent: Vec<Blocks>) -> Result<LayerStackGradient>

Reuse every two-polarization solve; accumulate shared medium and thickness derivatives.

Trait Implementations

impl Debug for LayerStackResidual

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl SavedState for LayerStackResidual

fn save_state(&self) -> Result<Vec<u8>>

Copy the saved numerical state into its fixed byte layout.

fn from_state(bytes: &[u8]) -> Result<Self>

Restore numerical state from bytes produced by this extension.