Expand description
Validated integration of a polynomial differential equation x' = f(x).
Every enclosure this module returns contains the exact solution: nothing in it is an approximation that is merely believed accurate.
One step of length h from an interval box X₀ of start points runs in two halves:
- An a priori box. A box
BwithX₀ + [0, h]·f(B) ⊆ Bholds every solution fromX₀over the whole step (Picard–Lindelöf through Schauder’s fixed point theorem). It is found by iteratingB ← X₀ + [0, h]·f(B)fromX₀, inflating each round. - A Taylor model. The solution’s Taylor coefficients satisfy
(j + 1)·x_[j+1] = f(x)_[j]. Each coefficient of a monomial offis a Cauchy product of the coefficients already known, soKorders costO(K²)products per factor. Lagrange’s remainder givesx(h) ∈ Σ_{j<K} x_[j](X₀)·h^j + x_[K](B)·[0, h^K]. The coefficient atBenclosesx^{(K)}(ξ)/K!for everyξin the step, because the solution stays inBthere.
The driver walks the steps at a working precision q. It halves the step when no a priori
box closes. A solution that leaves every bound, and a step pushed below 2^-cap, ends the walk
with FlowError::Escapes.
A scalar field that blows up is decided apart from the walk. When x' = p(x) with
p > 0 on [x₀, ∞) and deg p ≥ 2, the solution is increasing and leaves every bound at
T* = ∫_{x₀}^∞ dx/p(x). escape_time_bound bounds T* from above by a rational number:
a positive interval lower bound of p on each piece of a grid over [x₀, b], and past
b ≥ 1, where every lower term is at most S·x^{d−1} for S the sum of their magnitudes,
p(x) ≥ (lc − S/b)·x^d.
Structs§
- Field
- A polynomial vector field: component
ioff(x)is the sum ofcomponents[i]. - Monomial
c·∏ xᵢ^{eᵢ}.
Enums§
- Flow
Error - Why a walk stopped before its time.
Functions§
- escape_
time_ bound - For
x' = p(x)withp > 0on[x₀, ∞)anddeg p ≥ 2, a rational number at least the time the solution takes to leave every bound;Nonewhenphas a root there, grows too slowly to leave every bound in finite time, or no grid finer than2^cappieces bounds it. - flow
- An enclosure of the solution at time
t ≥ 0from every start instart, at working precisionq;capbounds how small a step may become.