ապ DPP Mathematical Software
ström procedure for regularizing generalized eigenproblems of matrix pencils
Understanding Generalized Eigenproblems
A generalized eigenproblem involves finding the values (eigenvalues) and vectors (eigenvectors) that satisfy the equation Av = λBv, where A and B are square matrices with the same dimensions, v is an eigenvector, and λ is an eigenvalue. The problem arises in various fields, including engineering, physics, and computer science.
To regularize this problem, we can use the Ström procedure, which introduces a small shift or modification to make the pencils (A - λB) regular, i.e., nonsingular for all λ. This allows for a well-conditioned linear system to solve and obtain the desired eigenvalues and eigenvectors.
Shifted Inverse Iteration Method
The Ström procedure can be implemented using the shifted inverse iteration method. This involves iteratively computing the inverse of a shifted matrix pencil and multiplying it by the initial estimate of the eigenvector. Convergence is achieved by monitoring the reduction in the norm of the residual vector.
Here's the mathematical representation of the method: 1. Choose an initial vector v₁ and a small shift σ. 2. For k = 1, 2, ..., compute v_{k+1} = σ(Aᵗ - λBᵗ)⁻¹(B - λA)vᵏ, where ᵗ denotes the matrix transpose. 3. If ∥vᵏ - v_{k+1}∥/∥vᵏ∥ < tolerance, then vᵏ is the eigenvector and Avᵏ = λBvᵏ; otherwise, go to step 2.
Practical Considerations
Some practical aspects to consider when implementing the Ström procedure include:
1. **Choosing the shift**: A small positive value for the shift σ often works well. Larger shifts may lead to quicker convergence but could potentially miss some eigenvalues.
2. **Initial estimate**: A good initial estimate of the eigenvector can hasten convergence. One common choice is to start with the first standard basis vector, e₁ = (1, 0, ..., 0)ᵗ.
3. **Tolerance**: The tolerance decides when to stop the iteration. A smaller tolerance allows for more accurate solutions but requires more iterations.
Examples of the Ström Procedure
To illustrate the Ström procedure, let's consider a simple 2x2 example:
A = [2 1; 1 2], B = [-1 0; 0 -1]. We want to find the generalized eigenvalues, λ. Using the Ström procedure with σ = 0.1, we find that the generalized eigenvalues are approximately λ₁ = -1.7321 and λ₂ = 0.2679.
For this problem, the initial estimate of the eigenvector was (1, 0)ᵗ, and convergence was achieved after 8 iterations for λ₁ and 7 iterations for λ₂. This example demonstrates that the Ström procedure can efficiently compute generalized eigenvalues for small, well-conditioned problems.
In practice, the Ström procedure is a powerful tool for regularizing generalized eigenproblems, but it may not always provide the most accurate or efficient solution for large or ill-conditioned problems. Therefore, it's essential to use this method judiciously and combine it with other techniques when appropriate.