|
|
IDL Reference Guide: Procedures and Functions |
|
The SVDC procedure computes the Singular Value Decomposition (SVD) of a square (n x n) or non-square (n x m) array as the product of orthogonal and diagonal arrays. SVD is a very powerful tool for the solution of linear systems, and is often used when a solution cannot be determined by other numerical algorithms.
The SVD of an (n x m) non-square array A is computed as the product of an (n x m) column orthogonal array U, an (n x n) diagonal array SV, composed of the singular values, and the transpose of an (n x n) orthogonal array V: A = U SV VT
SVDC is based on the routine svdcmp described in section 2.6 of Numerical Recipes in C: The Art of Scientific Computing (Second Edition), published by Cambridge University Press, and is used by permission.
| Note If you are working with complex inputs, use the LA_SVD procedure instead. |
SVDC, A, W, U, V [, /COLUMN] [, /DOUBLE] [, ITMAX=value]
The square (n x n) or non-square (n x m) single- or double-precision floating-point array to decompose.
On output, W is an n-element output vector containing the "singular values."
On output, U is an n-column, m-row orthogonal array used in the decomposition of A.
On output, V is an n-column, n-row orthogonal array used in the decomposition of A.
Set this keyword if the input array A is in column-major format (composed of column vectors) rather than in row-major format (composed of row vectors).
Set this keyword to force the computation to be done in double-precision arithmetic.
Set this keyword to specify the maximum number of iterations. The default value is 30.
To find the singular values of an array A:
; Define the array A: A = [[1.0, 2.0, -1.0, 2.5], $ [1.5, 3.3, -0.5, 2.0], $ [3.1, 0.7, 2.2, 0.0], $ [0.0, 0.3, -2.0, 5.3], $ [2.1, 1.0, 4.3, 2.2], $ [0.0, 5.5, 3.8, 0.2]] ; Compute the Singular Value Decomposition: SVDC, A, W, U, V ; Print the singular values: PRINT, W
IDL prints:
8.81973 2.65502 4.30598 6.84484
To verify the decomposition, use the relationship A = U ## SV ## TRANSPOSE(V), where SV is a diagonal array created from the output vector W:
sv = FLTARR(4, 4) FOR K = 0, 3 DO sv[K,K] = W[K] result = U ## sv ## TRANSPOSE(V) PRINT, result
IDL prints:
1.00000 2.00000 -1.00000 2.50000
1.50000 3.30000 -0.500001 2.00000
3.10000 0.700000 2.20000 0.00000
2.23517e-08 0.300000 -2.00000 5.30000
2.10000 0.999999 4.30000 2.20000
-3.91155e-07 5.50000 3.80000 0.200000
This is the input array, to within machine precision.
CHOLDC, LA_SVD, LUDC, SVSOL, Linear Systems.
IDL Online Help (March 06, 2007)