most_aitken Struct Reference

#include <ERF_MOSTUtils.H>

Collaboration diagram for most_aitken:

Public Member Functions
AMREX_GPU_HOST_DEVICE AMREX_FORCE_INLINE amrex::Real	update (amrex::Real u_new, int iter)

Public Attributes
int	start {25}
bool	extrapolated {false}
	whether the last update returned an extrapolated value More...
int	n {0}
amrex::Real	u0 {zero}
amrex::Real	u1 {zero}
amrex::Real	u2 {zero}

Detailed Description

Aitken delta-squared acceleration of the relaxed u* iteration in the prescribed-flux kernels.

Near the stable fold the slope of the fixed-point map approaches 1 and the plain iteration needs hundreds of passes. From pass start on, every third iterate is replaced by the Aitken extrapolation of the last three (kept only within a factor of 2 of them), so points that converge before start are unchanged. The map is not stationary when w* or a u*-dependent z0 also changes from pass to pass, so the kernels do not stop on an extrapolated pass: the plain pass after it, with zeta and psi from the same u*, decides convergence.

Member Function Documentation

update()

AMREX_GPU_HOST_DEVICE AMREX_FORCE_INLINE amrex::Real most_aitken::update ( amrex::Real  u_new, int  iter  )

inline

Return the iterate to continue from.

Parameters

[in]	u_new	iterate after pass iter
[in]	iter	number of passes done

    {
        extrapolated = false;
        if (iter < start) { return u_new; }
        u0 = u1;
        u1 = u2;
        u2 = u_new;
        if (++n < 3) { return u_new; }
        const amrex::Real d = u2 - two * u1 + u0;
        const amrex::Real ua = (d != zero) ? u2 - (u2 - u1) * (u2 - u1) / d : u_new;
        const amrex::Real lo = myhalf * amrex::min(u0, amrex::min(u1, u2));
        const amrex::Real hi = two * amrex::max(u0, amrex::max(u1, u2));
        if (d == zero || !(ua > lo && ua < hi)) {
            n = 0;
            return u_new;
        }
        // The extrapolated value starts the next triple
        u2 = ua;
        n  = 1;
        extrapolated = true;
        return ua;
    }

Referenced by surface_flux_charnock::iterate_flux(), surface_flux_mod_charnock::iterate_flux(), surface_flux_donelan::iterate_flux(), surface_flux_wave_coupled::iterate_flux(), surface_flux::iterate_flux(), and surface_flux_eb::iterate_flux().

Here is the caller graph for this function:

Member Data Documentation

extrapolated

bool most_aitken::extrapolated {false}

whether the last update returned an extrapolated value

Referenced by surface_flux_charnock::iterate_flux(), surface_flux_mod_charnock::iterate_flux(), surface_flux_donelan::iterate_flux(), surface_flux_wave_coupled::iterate_flux(), surface_flux::iterate_flux(), surface_flux_eb::iterate_flux(), and update().

n

int most_aitken::n {0}

Referenced by update().

start

int most_aitken::start {25}

Referenced by update().

u0

amrex::Real most_aitken::u0 {zero}

Referenced by update().

u1

amrex::Real most_aitken::u1 {zero}

Referenced by update().

u2

amrex::Real most_aitken::u2 {zero}

Referenced by update().

---

The documentation for this struct was generated from the following file:

• Source/BoundaryConditions/ERF_MOSTUtils.H