This project has moved. For the latest updates, please go here.
Many applications of matrix algebra result in eigen-problems. An eigen-problem gives us a matrix A and asks us to find vectors v such that A v = e v, that is vectors v that, when multiplied by the matrix A, do not produce some entirely new vector but instead merely scale by some number e. v is called an eigenvector of A and e its eigenvalue. In general, and N-dimensional square matrix will have N eigenvectors. For a generic square matrix, the eigenvalue and eigenvector components can be complex, but for symmetric matrices they are guaranteed to be real.

Meta.Numerics can solve eigen-problems for both symmetric and general square matrices. The solutions to eigen-problems are packaged as instances of the RealEigensystem or ComplexEigensystem classes, which can be queried for eigenvectors and eigenvalues.

For example, the following code gets the eigenvalues and eigenvectors of a symmetric matrix S.

// Define a symmetric matrix; note only half the entries need to be specified explicitly
SymmetricMatrix S = new SymmetricMatrix(3);
S[0,0] = 0.0; S[0,1] = 1.0; S[0,2] = 2.0;
S[1,1] = 2.0; S[1,2] = 3.0;
S[2,2] = 4.0;
// Find scalar a and vector v such that A v = a v
RealEigensystem E = S.Eigensystem();
// Process all the eigenvectors and eigenvalues found
for (int i = 0; i < E.Dimension; i++) {
  ColumnVector v = E.Eigenvector(i);
  double e = E.Eigenvalue(i);
  // now do something with these

It often occurs that you actually only require the eigenvalues of a matrix, and don't care about the eigenvectors. In that case, it's more efficient to not ask Meta.Numerics for the whole eigensystem solution, but just for the eigenvalue:

// Just the eigenvalues, please
double[] eigenvalues = S.Eigenvalues();

Last edited Apr 1, 2011 at 11:48 PM by ichbin, version 2


No comments yet.