Struct Layout 

pub struct Layout {
    pub non_slack_idx: Vec<usize>,
    pub pq_idx: Vec<usize>,
    pub n_angle: usize,
    pub n_unknowns: usize,
    pub n_buses: usize,
}

The Newton unknown layout, derived once per segment.

Reproduces solver::newton_raphson_cached’s ordering exactly — every non-slack bus contributes an angle unknown in bus order, then every PQ bus a magnitude unknown — because the corrector’s mismatch vector and the augmented Jacobian’s rows must agree with it term for term.

Fields

non_slack_idx: Vec<usize>

pq_idx: Vec<usize>

n_angle: usize

n_unknowns: usize

n_buses: usize

Implementations

impl Layout

pub fn analyze(buses: &[Bus]) -> Self

pub fn embed(&self, from: &Layout, t: &[f64]) -> Vec<f64>

Re-expresses a tangent written in from’s layout in this one.

A PV → PQ switch grows the unknown count by one, so the tangent that was steering the walk no longer even has the right length — but it still carries the direction of travel, which is the part that matters. Mapping it bus by bus (the newly-freed magnitude gets a zero component) gives the next tangent solve a bordering row that points the right way, so the walk continues forward across the event instead of doubling back.

pub fn get(&self, buses: &[Bus], lambda: f64, k: usize) -> f64

The value of unknown k, with k == n_unknowns meaning λ itself.

pub fn add(&self, buses: &mut [Bus], k: usize, delta: f64)

Adds delta to unknown k. λ is the caller’s own scalar, so k == n_unknowns is handled by the caller rather than here.

pub fn direction_rows(&self, d_p: &[f64], d_q: &[f64]) -> Vec<f64>

Δs in Newton row order: Δp on every non-slack P row, Δq on every PQ Q row. The same layout as the mismatch vector, which is what lets the λ column be appended row-for-row.