Choosing Solution Methods - Linear Algebra
- How is solved most efficiently when is an orthogonal matrix?
- , since .
- To solve once for a large matrix, why is elimination preferred over computing ?
- Elimination takes about operations, roughly a third of the cost of forming , and is usually more accurate.
- Which method fits a system whose matrix is singular or not square, to describe all solutions?
- Row reduce the augmented matrix to (reduced) echelon form. No inverse exists to use.
- Which method is efficient for solving with one matrix and many different vectors ?
- Factor once, then for each do one forward and one back substitution.
- Why is QR factorization usually preferred over the normal equations for least squares in floating point?
- Forming squares the condition number, which can lose accuracy; QR works with directly.
- What is the purpose of partial pivoting in Gaussian elimination?
- Choosing the largest available pivot in absolute value avoids dividing by small numbers and limits round-off growth.
- Why is Cramer's rule impractical for solving a large system ?
- It needs determinants of size ; it is useful only for small systems or theoretical formulas.
- How is solved when is diagonal with nonzero diagonal entries?
- Divide each entry: .
- About how many operations does solving a triangular system by substitution take, in order of magnitude?
- About , which is far cheaper than the order cost of elimination.
- Which factorization is the efficient choice when is symmetric positive definite?
- The Cholesky factorization , which needs about half the work of LU.
- How should be solved when is upper triangular with nonzero diagonal?
- By back substitution directly. No elimination is needed.