Jacobi Iteration

In numerical linear algebra, the Jacobi method is an iterative algorithm for determining the solutions of a strictly diagonally dominant system of linear equations. Each diagonal element is solved for, and an approximate value is plugged in. The process is then iterated until it converges. This algorithm is a stripped-down version of the Jacobi transformation method of matrix diagonalization. The method is named after Carl Gustav Jacob Jacobi.

Description

Let

:$A\mathbf x = \mathbf b$

be a square system of ''n'' linear equations, where:
$A = \begin a_ & a_ & \cdots & a_ \\ a_ & a_ & \cdots & a_ \\ \vdots & \vdots & \ddots & \vdots \\a_ & a_ & \cdots & a_ \end, \qquad \mathbf = \begin x_ \\ x_2 \\ \vdots \\ x_n \end , \qquad \mathbf = \begin b_ \\ b_2 \\ \vdots \\ b_n \end.$

Then ''A'' can be decomposed into a diagonal component ''D'', a lower triangular part ''L'' and an upper triangular part ''U'':

:$A=D+L+U \qquad \text \qquad D = \begin a_ & 0 & \cdots & 0 \\ 0 & a_ & \cdots & 0 \\ \vdots & \vdots & \ddots & \vdots \\0 & 0 & \cdots & a_ \end \text L+U = \begin 0 & a_ & \cdots & a_ \\ a_ & 0 & \cdots & a_ \\ \vdots & \vdots & \ddots & \vdots \\ a_ & a_ & \cdots & 0 \end.$

The solution is then obtained iteratively via
:$\mathbf^ = D^ (\mathbf - (L+U) \mathbf^),$

where $\mathbf^$ is the ''k''th approximation or iteration of $\mathbf$ and $\mathbf^$ is the next or ''k'' + 1 iteration of $\mathbf$. The element-based formula is thus:

:$x^_i = \frac \left(b_i -\sum_a_x^_j\right),\quad i=1,2,\ldots,n.$

The computation of $x_i^$ requires each element in x^(''k'') except itself. Unlike the Gauss–Seidel method, we can't overwrite $x_i^$ with $x_i^$, as that value will be needed by the rest of the computation. The minimum amount of storage is two vectors of size ''n''.

Algorithm

Input: , (diagonal dominant) matrix ''A'', right-hand side vector ''b'', convergence criterion
Output:
Comments:

while convergence not reached do
for ''i'' := 1 step until n do

for ''j'' := 1 step until n do
if ''j'' ≠ ''i'' then

end
end

end
increment ''k''
end

Convergence

The standard convergence condition (for any iterative method) is when the spectral radius of the iteration matrix is less than 1:

:$\rho(D^(L+U)) < 1.$

A sufficient (but not necessary) condition for the method to converge is that the matrix ''A'' is strictly or irreducibly diagonally dominant. Strict row diagonal dominance means that for each row, the absolute value of the diagonal term is greater than the sum of absolute values of other terms:

:$\left , a_ \right , > \sum_ .$

The Jacobi method sometimes converges even if these conditions are not satisfied.

Note that the Jacobi method does not converge for every symmetric positive-definite matrix.
For example

:$A = \begin 29 & 2 & 1\\ 2 & 6 & 1\\ 1 & 1 & \frac \end \quad \Rightarrow \quad D^ (L+U) = \begin 0 & \frac & \frac\\ \frac & 0 & \frac\\ 5 & 5 & 0 \end \quad \Rightarrow \quad \rho(D^(L+U)) \approx 1.0661 \,.$

Examples

Example 1

A linear system of the form $Ax=b$ with initial estimate $x^$ is given by

:$A= \begin 2 & 1 \\ 5 & 7 \\ \end, \ b= \begin 11 \\ 13 \\ \end \quad \text \quad x^ = \begin 1 \\ 1 \\ \end .$
We use the equation $x^=D^(b - (L+U)x^)$, described above, to estimate $x$. First, we rewrite the equation in a more convenient form $D^(b - (L+U)x^) = Tx^ + C$, where $T=-D^(L+U)$ and $C = D^b$. From the known values

:$D^= \begin 1/2 & 0 \\ 0 & 1/7 \\ \end, \ L= \begin 0 & 0 \\ 5 & 0 \\ \end \quad \text \quad U = \begin 0 & 1 \\ 0 & 0 \\ \end .$
we determine $T=-D^(L+U)$ as
:$T= \begin 1/2 & 0 \\ 0 & 1/7 \\ \end \left\ = \begin 0 & -1/2 \\ -5/7 & 0 \\ \end .$
Further, $C$ is found as

:$C = \begin 1/2 & 0 \\ 0 & 1/7 \\ \end \begin 11 \\ 13 \\ \end = \begin 11/2 \\ 13/7 \\ \end.$
With $T$ and $C$ calculated, we estimate $x$ as $x^= Tx^+C$:
:$x^= \begin 0 & -1/2 \\ -5/7 & 0 \\ \end \begin 1 \\ 1 \\ \end + \begin 11/2 \\ 13/7 \\ \end = \begin 5.0 \\ 8/7 \\ \end \approx \begin 5 \\ 1.143 \\ \end .$
The next iteration yields
:$x^= \begin 0 & -1/2 \\ -5/7 & 0 \\ \end \begin 5.0 \\ 8/7 \\ \end + \begin 11/2 \\ 13/7 \\ \end = \begin 69/14 \\ -12/7 \\ \end \approx \begin 4.929 \\ -1.714 \\ \end .$
This process is repeated until convergence (i.e., until $\, Ax^ - b\,$ is small). The solution after 25 iterations is
:$x=\begin 7.111\\ -3.222 \end .$

Example 2

Suppose we are given the following linear system:

:$\begin 10x_1 - x_2 + 2x_3 & = 6, \\ -x_1 + 11x_2 - x_3 + 3x_4 & = 25, \\ 2x_1- x_2+ 10x_3 - x_4 & = -11, \\ 3x_2 - x_3 + 8x_4 & = 15. \end$

If we choose as the initial approximation, then the first approximate solution is given by

:$\begin x_1 & = (6 + 0 - (2 * 0)) / 10 = 0.6, \\ x_2 & = (25 + 0 + 0 - (3 * 0)) / 11 = 25/11 = 2.2727, \\ x_3 & = (-11 - (2 * 0) + 0 + 0) / 10 = -1.1,\\ x_4 & = (15 - (3 * 0) + 0) / 8 = 1.875. \end$

Using the approximations obtained, the iterative procedure is repeated until the desired accuracy has been reached. The following are the approximated solutions after five iterations.

The exact solution of the system is .

Python example

import numpy as np

ITERATION_LIMIT = 1000

# initialize the matrix
A = np.array(10., -1., 2., 0.
1., 11., -1., 3.
., -1., 10., -1.
.0, 3., -1., 8.)
# initialize the RHS vector
b = np.array( ., 25., -11., 15.

# prints the system
print("System:")
for i in range(A.shape :
row = *x".format(A[i,_j_j_+_1)_for_j_in_range(A.shape[1.html" ;"title=",_j.html" ;"title="*x".format(A[i, j">*x".format(A[i, j j + 1) for j in range(A.shape[1">,_j.html" ;"title="*x".format(A[i, j">*x".format(A[i, j j + 1) for j in range(A.shape[1]
print(f' = ')
print()

x = np.zeros_like(b)
for it_count in range(ITERATION_LIMIT):
if it_count != 0:
print("Iteration : ".format(it_count, x))
x_new = np.zeros_like(x)

for i in range(A.shape :
s1 = np.dot(A , :i x i
s2 = np.dot(A , i + 1: x + 1:
x_new = (b - s1 - s2) / A, i if x_new

x_new -1
break

if np.allclose(x, x_new, atol=1e-10, rtol=0.):
break

x = x_new

print("Solution: ")
print(x)
error = np.dot(A, x) - b
print("Error:")
print(error)

Weighted Jacobi method

The weighted Jacobi iteration uses a parameter $\omega$ to compute the iteration as

:$\mathbf^ = \omega D^ (\mathbf - (L+U) \mathbf^) + \left(1-\omega\right)\mathbf^$
with $\omega = 2/3$ being the usual choice.
From the relation $L + U = A - D$, this may also be expressed as
:$\mathbf^ = \omega D^ \mathbf + \left( I - \omega D^ A \right) \mathbf^$.

Convergence in the symmetric positive definite case

In case that the system matrix $A$ is of symmetric positive-definite type one can show convergence.

Let $C=C_\omega = I-\omega D^A$ be the iteration matrix.
Then, convergence is guaranteed for
:$\rho(C_\omega) < 1 \quad \Longleftrightarrow \quad 0 < \omega < \frac \,,$
where $\lambda_\text$ is the maximal eigenvalue.

The spectral radius can be minimized for a particular choice of $\omega = \omega_\text$ as follows
:$\min_\omega \rho (C_\omega) = \rho (C_) = 1-\frac \quad \text \quad \omega_\text := \frac \,,$
where $\kappa$ is the matrix condition number.

See also

*Gauss–Seidel method
*Successive over-relaxation
* Iterative method § Linear systems
* Gaussian Belief Propagation
*Matrix splitting

References

External links

*
*

Jacobi Method from www.math-linux.com

{{Authority control

Numerical linear algebra
Articles with example pseudocode
Relaxation (iterative methods)
Articles with example Python (programming language) code