Numerical Methods Contents - SAM
Numerical Methods Contents - SAM
Numerical Methods Contents - SAM
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
eplaced with κ(A) !<br />
4.4.2 Iterations with short recursions<br />
Iterative solver for Ax = b with symmetric system matrix A:<br />
Iterative methods for general regular system matrix A:<br />
MATLAB-functions:<br />
• [x,flg,res,it,resv] = minres(A,b,tol,maxit,B,[],x0);<br />
• [. ..] = minres(Afun,b,tol,maxit,Binvfun,[],x0);<br />
Idea: Given x (0) ∈ R n determine (better) approximation x (l) through Petrov-<br />
Galerkin condition<br />
Computational costs : 1 A×vector, 1 B −1 ×vector per step, a few dot products & SAXPYs<br />
Memory requirement: a few vectors ∈ R n<br />
Extension to general regular A ∈ R n,n :<br />
x (l) ∈ x (0) + K l (A,r 0 ): p H (b − Ax (l) ) = 0 ∀p ∈ W l ,<br />
with suitable test space W l , dim W l = l, e.g. W l := K l (A H ,r 0 ) (→ biconjugate<br />
gradients, BiCG)<br />
Idea: Solver overdetermined linear system of equations<br />
x (l) ∈ x (0) + K l (A,r 0 ): Ax (l) = b<br />
in least squares sense, → Chapter 6.<br />
x (l) = argmin{‖Ay − b‖ 2 : y ∈ x (0) + K l (A,r 0 )} .<br />
Ôº¿ º<br />
Zoo of methods with short recursions (i.e. constant effort per step)<br />
Ôº¿ º<br />
Memory requirements: 8 vectors ∈ R n<br />
MATLAB-function: • [x,flag,r,it,rv] = bicgstab(A,b,tol,maxit,B,[],x0)<br />
• [...] = bicgstab(Afun,b,tol,maxit,Binvfun,[],x0);<br />
Computational costs :<br />
2 A×vector, 2 B −1 ×vector, 4 dot products, 6 SAXPYs per step<br />
➤<br />
GMRES method for general matrices A ∈ R n,n<br />
MATLAB-function: • [x,flag,r,it,rv] = qmr(A,b,tol,maxit,B,[],x0)<br />
• [...] = qmr(Afun,b,tol,maxit,Binvfun,[],x0);<br />
MATLAB-function: • [x,flag,relr,it,rv] = gmres(A,b,rs,tol,maxit,B,[],x0);<br />
• [. ..] = gmres(Afun,b,rs,tol,maxit,Binvfun,[],x0);<br />
Computational costs :<br />
Memory requirements:<br />
2 A×vector, 2 B −1 ×vector, 2 dot products, 12 SAXPYs per step<br />
10 vectors ∈ R n<br />
Computational costs : 1 A×vector, 1 B −1 ×vector per step,<br />
: O(l) dot products & SAXPYs in l-th step<br />
Memory requirements: O(l) vectors ∈ K n in l-th step<br />
Remark 4.4.1 (Restarted GMRES).<br />
little (useful) covergence theory available<br />
stagnation & “breakdowns” commonly occur<br />
Example 4.4.2 (Failure of Krylov iterative solvers).<br />
After many steps of GMRES we face considerable computational costs and memory requirements for<br />
every further step. Thus, the iteration may be restarted with the current iterate x (l) as initial guess →<br />
rs-parameter triggers restart after every rs steps (Danger: failure to converge).<br />
△<br />
Ôº¿ º<br />
⎛<br />
0 1 0 · · · · · · 0<br />
0 0 1 0 .<br />
. . .. ... ...<br />
A =<br />
... .<br />
⎜<br />
. ... ... 0<br />
⎝0 0 1<br />
1 0 · · · · · · 0<br />
⎞<br />
⎟<br />
⎠<br />
⎛ ⎞<br />
0.<br />
, b =<br />
⎜<br />
.<br />
⎟<br />
⎝0⎠<br />
1<br />
x = e 1 .<br />
Ôº¿ º