| 1 |
Jacobi and Gauss Seidel give a formula for xk in terms of x(k-1) call this xbasick
|
| 2 |
Overrelaxation forms xSORk = ? xbasick + (1- ?)xbasic(k-1)
|
| 3 |
Typically only 0 < ? < 2 is sensible and ? < 1 is relaxation 1 < ? < 2 is over relaxation
|
| 4 |
It is "over" because if ? > 1, you go "further" in direction of new estimate than calculated
|
| 5 |
In "relaxation", you conservatively, average prediction with old value
|
| 6 |
Unfortunately, the best value of ? cannot be calculated except in simple cases and so SOR is not so reliable for general problems
|
