The Picard map for rough differential equations #
The Picard map of dY = f(Y)·dX with initial condition y₀: a
controlled path Z is sent to the path based at y₀ with increments the
rough integral of the composed integrand f(Z), and Gubinelli derivative
f(Z.Y). On a control window normalised by ω ≤ 1 and
ω^α ≤ δα, the map preserves an explicit certificate box (for suitable
box constants), the first step towards the fixed-point construction of
solutions.
The ℝ≥0-valued germ constant of a controlled path.
Instances For
Basing an additive increment family at an initial point #
The path based at y₀ with increments I.
Instances For
The Picard map #
The chosen rough integral of the composed integrand.
Equations
- RoughPaths.picardIntegral V hX hω1 hfine hωne Z = Classical.choose ⋯
Instances For
The Picard map: base point y₀, increments the rough integral of
f(Z), Gubinelli derivative f(Z.Y), with explicit certificates.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Box invariance #
Membership of the certificate box.
Instances For
Box invariance of the Picard map: for box constants satisfying the three closure inequalities, the Picard map preserves the box.
The distance step #
Pointwise bound for a Gubinelli germ from a sup bound on the path data.
The difference of two additive families with sewing germ bounds for the composed integrands is bounded by the distance constants with the full window gain.
The folded form of the integral-difference germ bound on the window.
The distance step: certificates for the distance between two
Picard iterates, linear in the input distance with an explicit window
gain δα on the sup and remainder slots.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The distance step for a pair of solutions #
The distance step for two solutions: solutions of the same RDE from the same initial condition satisfy the Picard distance-step inequalities directly, with their own increment families.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Uniqueness of solutions #
Any two box-certified controlled paths with the same initial value admit a finite distance certificate on the window.
Equations
Instances For
Uniqueness of RDE solutions (Friz–Hairer Thm 8.4-type): two
box-certified solutions of dY = f(Y)·dX from the same initial value
coincide on the window, provided the distance step contracts a weighted
combination of the distance certificates.
Weakening certificates #
Enlarge the certificates of a controlled path to given box constants.
Equations
Instances For
Distance certificates transport across weakening (the paths are unchanged).
Equations
Instances For
The Picard iteration #
The Picard iterates: starting from the constant path, apply the
Picard map and re-certify to the box at every step. All iterates carry
literal box constants and start at y₀.
Equations
- One or more equations did not get rendered due to their size.
- RoughPaths.picardSeq V hX hω1 hfine hωne y₀ hδα hBb hBd hBy 0 = ⟨(RoughPaths.ControlledPath.const X ω α y₀).weaken ⋯ ⋯ ⋯, ⋯⟩
Instances For
Distance certificates between consecutive Picard iterates.
Equations
- One or more equations did not get rendered due to their size.
- RoughPaths.picardSeqDist V hX hω1 hfine hωne y₀ hδα hδα1 hBb hBd hBy 0 = RoughPaths.seedDist hX hω1 hδα ⋯ ⋯ ⋯
Instances For
The weighted distance decays geometrically along the iteration.
Shorthand for the weighted seed distance.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Consecutive Picard iterates are geometrically close, pointwise.
The Picard iterates converge pointwise: the limit path.
Consecutive Picard integrals are close, with the next distance constant.
The Picard integrals converge pointwise on ordered pairs.
Existence of RDE solutions (Friz–Hairer Thm 8.3-type): under the
box-closure and weighted-contraction window conditions, the Picard
iterates converge to a box-certified solution of dY = f(Y)·dX started
at y₀.