Struct MultiDirJet 

pub struct MultiDirJet {
    pub coeffs: Vec<f64>,
}

Fields

coeffs: Vec<f64>

Implementations

impl MultiDirJet

pub fn zero(n_dirs: usize) -> Self

pub fn constant(n_dirs: usize, value: f64) -> Self

pub fn linear(n_dirs: usize, base: f64, first: &[f64]) -> Self

pub fn with_coeffs(n_dirs: usize, coeffs: &[(usize, f64)]) -> Self

pub fn coeff(&self, mask: usize) -> f64

pub fn add(&self, other: &Self) -> Self

pub fn scale(&self, scalar: f64) -> Self

pub fn mul(&self, other: &Self) -> Self

Subset-convolution product out[mask] = Σ_{sub ⊆ mask} a[sub]·b[mask^sub].

Bit-identical to the shared [crate::jet_algebra::leibniz_product] walker (the submasks are enumerated in the same ascending order — the walker’s compacted subset index is a monotone bit-deposit of the submask) while dropping its per-subset SlotBuf/closure/mask_of overhead. The scalar n_dirs == 0 case keeps the shared walker live as its reference.

pub fn compose_unary(&self, derivs: [f64; 5]) -> Self

Exact (order-4 truncated) unary composition f(self) from the Taylor stack [f, f', f'', f''', f''''] at self.coeff(0).

Computed by the truncated-Taylor reassociation (see the module note): f(self) = Σ_{k=0}^{4} (f^{(k)}/k!)·v^{⊛k} with v the non-constant part of self. The three subset-convolution powers v², v³, v⁴ are compensated (Dot2) and the per-mask combine is Neumaier-compensated and vectorised, so the result is more accurate vs. the true real-arithmetic value than the prior naive partition sum (proven against a double-double oracle in tests). The scalar n_dirs == 0 case keeps the shared Faà di Bruno walker live as its reference.

impl MultiDirJet

pub fn bilinear(base: f64, d1: f64, d2: f64, d12: f64) -> Self

pub fn sub(&self, other: &Self) -> Self

Trait Implementations

impl Clone for MultiDirJet

fn clone(&self) -> MultiDirJet

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.