Suppose a linear system is given $$AX=B,$$ where $A\in\mathbb{R}^{n\times n}$ is a symmetric strictly diagonal matrix, and $X, B\in\mathbb{R}^{n\times 2}$. Therefore, the 2D Jacobi iterative solver is applicable here, $$X_{k+1}=D^{-1}(B-RX_k),$$ where the splitting corresponds to $A=D+R$, with $D$ containing only the diagonal entries of $A$. Could it be proven that the iterands $X_0, X_1, X_2, \dots$ are maintaining the line per each row of $X$, meaning that, for $x^i\in\mathbb{R}^2$ being the $i-$th row of $X$, the rows in sequence $x^i_0, x^i_1, x^i_2, \dots$, are colinear (on one line)?
-
$\begingroup$ My first instinct is no, but I don't have any sources to back it up. $\endgroup$ – Paul♦ Mar 7 '12 at 19:25
Let's take the difference between two successive iterations, and define $P$ and $Q$ this way: $$ \Delta_k = X_{k+1} - X_{k} = D^{-1}B - (D^{-1}R + I)X_{k} = P-QX_{k} $$ Then $$ \Delta_{k+1} = X_{k+2} - X_{k+1} = P-QX_{k+1} $$ To be row-wise collinear, we need $\Delta_{k+1} - \Delta_{k} = S \Delta_{k}$ for some diagonal matrix $S$. Doing this, we get $$ \Delta_{k+1} - \Delta_{k} = Q(X_{k+1}-X_k) = Q\Delta_k $$ Since $Q = D^{-1}R + I$ is not diagonal unless $R$ is diagonal, then in general these iterates will not be collinear.
-
$\begingroup$ Good argument. If you are aware of some other iterative solver that has a property of "line maintenance" in case of 2D, please let me know. $\endgroup$ – usero Mar 8 '12 at 8:05