Mo Logo [Home] [Lexikon] [Aufgaben] [Tests] [Kurse] [Begleitmaterial] [Hinweise] [Mitwirkende] [Publikationen] [Suche] Englische Flagge

Mathematik-Online-Lexikon:

Lineares Gleichungssystem des Ausgleichsproblems


A B C D E F G H I J K L M N O P Q R S T U V W X Y Z Übersicht

Bei einem Ausgleichsproblem

$\displaystyle \vert Ax-b\vert \to \min
$

sind die Normalengleichungen

$\displaystyle A^{\operatorname t}A x = A^{\operatorname t}b
$

zu lösen. Die dabei auftretende Matrix $ S = A^{\operatorname t}A$ ist symmetrisch und positiv semidefinit, d.h.

$\displaystyle x^{\operatorname t}S x = (Ax)^{\operatorname t}(Ax) = \vert y\vert^2 \ge 0\,
.
$

Hat die $ (m\times n)$-Matrix $ A$ ($ m>n$) maximalen Rang $ n$, so ist

$\displaystyle 0 = y = Ax
$

nur für $ x=0$ möglich. In diesem Normalfall ist $ S$ positiv definit und die Gauß-Seidel-Iteration konvergent.

Die Gauß-Seidel-Iteration ist ebenfalls anwendbar, wenn $ A$ linear abhängige Spalten hat ( $ \operatorname{Rang} A <n$). Man muss lediglich fordern, dass $ A$ keine Nullspalten $ a_i$ enthält, also die Diagonalelemente

$\displaystyle s_{i,i} = a_i^{\operatorname t}a_i
$

positiv sind. In diesem Fall ist zwar die Folge der Approximationen einer Lösung $ x$ nicht mehr konvergent, aber das Residuum $ Ax - b$ strebt gegen Null.
(Autoren: App/Höllig)

[Verweise]

  automatisch erstellt am 25.  1. 2006