Proof of Stationary Values for Ritz Values in Lanczos Iteration

  • Context: Graduate 
  • Thread starter Thread starter redlegend
  • Start date Start date
Click For Summary
SUMMARY

The discussion centers on the proof of the lemma stating that the Ritz values at step n of the Lanczos iteration are the stationary values of the Rayleigh quotient r(x) = (xT A x) / (xT x) when x is confined to the Krylov subspace Kn. Participants express a shared interest in obtaining this proof, indicating its significance in understanding the relationship between Ritz values and the Rayleigh quotient within the context of Lanczos iteration.

PREREQUISITES
  • Understanding of Lanczos iteration and its application in numerical linear algebra.
  • Familiarity with the concept of Ritz values in the context of eigenvalue problems.
  • Knowledge of the Rayleigh quotient and its role in optimization problems.
  • Basic comprehension of Krylov subspaces and their significance in iterative methods.
NEXT STEPS
  • Research the derivation of the Rayleigh quotient and its properties in optimization.
  • Study the Lanczos algorithm in detail, focusing on its convergence and stability.
  • Explore the relationship between Ritz values and eigenvalues in iterative methods.
  • Investigate existing proofs or literature regarding the lemma in question to gain deeper insights.
USEFUL FOR

Mathematicians, numerical analysts, and researchers in computational mathematics focusing on eigenvalue problems and iterative methods for large-scale systems.

redlegend
Messages
1
Reaction score
0
I was wondering whether you have the proof to the following lemma:

"The Ritz values at step n of the Lanczos iteration are the stationary values of the Rayleigh quotient r(x)=(xT A x)/(xT x) if x is restricted to Kn "
 
Physics news on Phys.org
Did you ever find the proof, because I have the same exact question?
 

Similar threads

  • · Replies 22 ·
Replies
22
Views
3K
  • · Replies 1 ·
Replies
1
Views
3K
  • · Replies 2 ·
Replies
2
Views
2K
  • · Replies 5 ·
Replies
5
Views
3K
  • · Replies 6 ·
Replies
6
Views
4K
  • · Replies 9 ·
Replies
9
Views
2K
  • · Replies 11 ·
Replies
11
Views
2K
  • · Replies 5 ·
Replies
5
Views
2K
  • · Replies 11 ·
Replies
11
Views
3K
  • · Replies 29 ·
Replies
29
Views
5K