In this week I wrote a code that implements the augmented matrix described in theorem 2.1 of [HAM 11]. This matrix will be given as input to expmv and will allow us to implement the exponential integrators by evaluating a single exponential of this augmented matrix avoiding the need to compute any φ functions. The code is

function [Atilde, eta] = augmat(h,A,V)

p = size(V,2);

W = fliplr(V/diag(h.^(0:p-1)));

eta = 2^-ceil(log2(max(norm(W,1),realmin)));

Atilde = [ A , eta*W ; zeros(p,size(A)) , diag(ones(p-1,1),1) ];

in [HAM 11] eta is introduced to avoid rounding errors. I made a small change to avoid further errors, I added the max function so that smaller values than realmin will not be considered.

Now I summarize my work during this month.

Week 1: phi functions. I implemented four functions, based on [BSW 07] (phi1m.m, phi2m.m, phi3m.m, phi4m.m).

Week 2-3: general exponential schemes. I implemented the two schemes for a general exponential Runge-Kutta and Rosenbrock integrator (exprk.m, exprb.m).

As already mentioned, these schemes are not really fast and efficient, but these are working codes that I applied to four different exponential methods (exprk2.m, exprk3.m, exprb3.m, exprb4.m) and I will use them as a reference when I go to implement the official methods.

Week 4: augmented matrix (augmat.m).

My codes can be found here

https://bitbucket.org/ChiaraSegala/chiara-segala

## Nessun commento:

## Posta un commento