Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M....
Transcript of Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M....
![Page 1: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/1.jpg)
Getting startedThe results ...
... and their impactsSummary
Abstract Perturbed Krylov MethodsJust another point of view?
Jens-Peter M. Zemke
Arbeitsbereich Mathematik 4-13Technische Universität Hamburg-Harburg
08.03.2005 / ICS of CAS / Prague
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 2: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/2.jpg)
Getting startedThe results ...
... and their impactsSummary
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 3: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/3.jpg)
Getting startedThe results ...
... and their impactsSummary
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 4: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/4.jpg)
Getting startedThe results ...
... and their impactsSummary
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 5: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/5.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 6: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/6.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstraction
Merriam-Webster Online: abstraction (noun)
1 a : the act or process of abstracting : the state of beingabstracted b : an abstract idea or term
2 absence of mind or preoccupation3 abstract quality or character4 a : an abstract composition or creation in art b :
abstractionism
We aim at 1a (possibly 3 and 4a), not 2.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 7: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/7.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstraction
Merriam-Webster Online: abstraction (noun)
1 a : the act or process of abstracting : the state of beingabstracted b : an abstract idea or term
2 absence of mind or preoccupation3 abstract quality or character4 a : an abstract composition or creation in art b :
abstractionism
We aim at 1a (possibly 3 and 4a), not 2.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 8: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/8.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstraction
Merriam-Webster Online: abstraction (noun)
1 a : the act or process of abstracting : the state of beingabstracted b : an abstract idea or term
2 absence of mind or preoccupation
3 abstract quality or character4 a : an abstract composition or creation in art b :
abstractionism
We aim at 1a (possibly 3 and 4a), not 2.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 9: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/9.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstraction
Merriam-Webster Online: abstraction (noun)
1 a : the act or process of abstracting : the state of beingabstracted b : an abstract idea or term
2 absence of mind or preoccupation3 abstract quality or character
4 a : an abstract composition or creation in art b :abstractionism
We aim at 1a (possibly 3 and 4a), not 2.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 10: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/10.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstraction
Merriam-Webster Online: abstraction (noun)
1 a : the act or process of abstracting : the state of beingabstracted b : an abstract idea or term
2 absence of mind or preoccupation3 abstract quality or character4 a : an abstract composition or creation in art b :
abstractionism
We aim at 1a (possibly 3 and 4a), not 2.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 11: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/11.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstraction
Merriam-Webster Online: abstraction (noun)
1 a : the act or process of abstracting : the state of beingabstracted b : an abstract idea or term
2 absence of mind or preoccupation3 abstract quality or character4 a : an abstract composition or creation in art b :
abstractionism
We aim at 1a (possibly 3 and 4a), not 2.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 12: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/12.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstract
Selected definitions for “abstract”
Merriam-Webster Online: abstract (verb)
2 to consider apart from application to or association with aparticular instance
Merriam-Webster Online: abstract (adjective)
1 a : disassociated from any specific instance2 expressing a quality apart from an object3 a : dealing with a subject in its abstract aspects
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 13: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/13.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstract
Selected definitions for “abstract”
Merriam-Webster Online: abstract (verb)
2 to consider apart from application to or association with aparticular instance
Merriam-Webster Online: abstract (adjective)
1 a : disassociated from any specific instance2 expressing a quality apart from an object3 a : dealing with a subject in its abstract aspects
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 14: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/14.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
abstract
Selected definitions for “abstract”
Merriam-Webster Online: abstract (verb)
2 to consider apart from application to or association with aparticular instance
Merriam-Webster Online: abstract (adjective)
1 a : disassociated from any specific instance2 expressing a quality apart from an object3 a : dealing with a subject in its abstract aspects
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 15: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/15.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
perturbed KRYLOV methods
We consider perturbed KRYLOV subspace methods that can bewritten in the form
AQk = Qk+1Ck−Fk , (1a)
Qk+1Ck = QkCk + Mk , (1b)
Mk = qk+1ck+1,keTk . (1c)
We refer to the set of equations (1) as a perturbed KRYLOV
decomposition.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 16: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/16.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
perturbed KRYLOV methods
We consider perturbed KRYLOV subspace methods that can bewritten in the form
AQk = Qk+1Ck−Fk , (1a)
Qk+1Ck = QkCk + Mk , (1b)
Mk = qk+1ck+1,keTk . (1c)
We refer to the set of equations (1) as a perturbed KRYLOV
decomposition.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 17: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/17.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
the main actors
In the perturbed KRYLOV decomposition:
A ∈ Cn×n is the system matrix from
Ax = b or Av = vλ
Qk ∈ Cn×k captures the “basis” vectors constructed
Ck ∈ Ck×k is unreduced upper HESSENBERG
Ck ∈ C(k+1)×k is extended upper HESSENBERG
Fk ∈ Cn×k is zero or captures perturbations (due to finiteprecision, inexact methods, both, . . . )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 18: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/18.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
the main actors
In the perturbed KRYLOV decomposition:
A ∈ Cn×n is the system matrix from
Ax = b or Av = vλ
Qk ∈ Cn×k captures the “basis” vectors constructed
Ck ∈ Ck×k is unreduced upper HESSENBERG
Ck ∈ C(k+1)×k is extended upper HESSENBERG
Fk ∈ Cn×k is zero or captures perturbations (due to finiteprecision, inexact methods, both, . . . )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 19: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/19.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
the main actors
In the perturbed KRYLOV decomposition:
A ∈ Cn×n is the system matrix from
Ax = b or Av = vλ
Qk ∈ Cn×k captures the “basis” vectors constructed
Ck ∈ Ck×k is unreduced upper HESSENBERG
Ck ∈ C(k+1)×k is extended upper HESSENBERG
Fk ∈ Cn×k is zero or captures perturbations (due to finiteprecision, inexact methods, both, . . . )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 20: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/20.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
the main actors
In the perturbed KRYLOV decomposition:
A ∈ Cn×n is the system matrix from
Ax = b or Av = vλ
Qk ∈ Cn×k captures the “basis” vectors constructed
Ck ∈ Ck×k is unreduced upper HESSENBERG
Ck ∈ C(k+1)×k is extended upper HESSENBERG
Fk ∈ Cn×k is zero or captures perturbations (due to finiteprecision, inexact methods, both, . . . )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 21: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/21.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
the main actors
In the perturbed KRYLOV decomposition:
A ∈ Cn×n is the system matrix from
Ax = b or Av = vλ
Qk ∈ Cn×k captures the “basis” vectors constructed
Ck ∈ Ck×k is unreduced upper HESSENBERG
Ck ∈ C(k+1)×k is extended upper HESSENBERG
Fk ∈ Cn×k is zero or captures perturbations (due to finiteprecision, inexact methods, both, . . . )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 22: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/22.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
crucial assumptions
given: A ∈ Cn×n
and q1 ∈ Cn
computed: unreduced HESSENBERG Ck ∈ Ck×k
unknown: properties of the “basis” Qk
“measurable”: the perturbation terms Fk
We treat the system matrix A, the starting vector q1 and theperturbation terms {fl}k
l=1 as input data and express everythingelse based on the computed Ck .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 23: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/23.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
crucial assumptions
given: A ∈ Cn×n and q1 ∈ Cn
computed: unreduced HESSENBERG Ck ∈ Ck×k
unknown: properties of the “basis” Qk
“measurable”: the perturbation terms Fk
We treat the system matrix A, the starting vector q1 and theperturbation terms {fl}k
l=1 as input data and express everythingelse based on the computed Ck .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 24: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/24.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
crucial assumptions
given: A ∈ Cn×n and q1 ∈ Cn
computed: unreduced HESSENBERG Ck ∈ Ck×k
unknown: properties of the “basis” Qk
“measurable”: the perturbation terms Fk
We treat the system matrix A, the starting vector q1 and theperturbation terms {fl}k
l=1 as input data and express everythingelse based on the computed Ck .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 25: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/25.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
crucial assumptions
given: A ∈ Cn×n and q1 ∈ Cn
computed: unreduced HESSENBERG Ck ∈ Ck×k
unknown: properties of the “basis” Qk
“measurable”: the perturbation terms Fk
We treat the system matrix A, the starting vector q1 and theperturbation terms {fl}k
l=1 as input data and express everythingelse based on the computed Ck .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 26: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/26.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
crucial assumptions
given: A ∈ Cn×n and q1 ∈ Cn
computed: unreduced HESSENBERG Ck ∈ Ck×k
unknown: properties of the “basis” Qk
“measurable”: the perturbation terms Fk
We treat the system matrix A, the starting vector q1 and theperturbation terms {fl}k
l=1 as input data and express everythingelse based on the computed Ck .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 27: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/27.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
crucial assumptions
given: A ∈ Cn×n and q1 ∈ Cn
computed: unreduced HESSENBERG Ck ∈ Ck×k
unknown: properties of the “basis” Qk
“measurable”: the perturbation terms Fk
We treat the system matrix A, the starting vector q1 and theperturbation terms {fl}k
l=1 as input data and express everythingelse based on the computed Ck .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 28: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/28.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 29: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/29.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
ARNOLDI
In the ARNOLDI method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has orthonormal columns
Ck ∈ Ck×k is unreduced HESSENBERG
Fk ∈ Cn×k is
(ask Miro about the details :- )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 30: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/30.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
ARNOLDI
In the finite precision ARNOLDI method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has “approximately” orthonormal columns
Ck ∈ Ck×k is unreduced HESSENBERG
Fk ∈ Cn×k is “small”
(ask Miro about the details :- )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 31: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/31.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
ARNOLDI
In the inexact ARNOLDI method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has orthonormal columns
Ck ∈ Ck×k is unreduced HESSENBERG
Fk ∈ Cn×k is “controlled by the user”
(ask Miro about the details :- )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 32: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/32.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
ARNOLDI
In the finite precision inexact ARNOLDI method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has “approximately” orthonormal columns
Ck ∈ Ck×k is unreduced HESSENBERG
Fk ∈ Cn×k is “small” plus “controlled by the user”
(ask Miro about the details :- )
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 33: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/33.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
LANCZOS
In the LANCZOS method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has bi-orthonormal columns
Ck ∈ Ck×k is unreduced tridiagonal
Fk ∈ Cn×k is
The error terms may grow unbounded . . .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 34: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/34.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
LANCZOS
In the finite precision LANCZOS method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has “locally” bi-orthonormal columns
Ck ∈ Ck×k is unreduced tridiagonal
Fk ∈ Cn×k is “small”
The error terms may grow unbounded . . .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 35: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/35.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
LANCZOS
In the inexact LANCZOS method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has bi-orthonormal columns
Ck ∈ Ck×k is unreduced tridiagonal
Fk ∈ Cn×k is “controlled by the user”
The error terms may grow unbounded . . .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 36: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/36.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
LANCZOS
In the finite precision inexact LANCZOS method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has “locally” bi-orthonormal columns
Ck ∈ Ck×k is unreduced tridiagonal
Fk ∈ Cn×k is “small” plus “controlled by the user”
The error terms may grow unbounded . . .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 37: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/37.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
power method
In the power method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has nearly dependent columns
Ck ∈ Ck×k is nilpotent unreduced HESSENBERG
Fk ∈ Cn×k is “small” compared to Qk
Columns of Qk may be dependent from the beginning.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 38: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/38.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
power method
In the finite precision power method:
A ∈ Cn×n is a general matrix
Qk ∈ Cn×k has nearly dependent columns
Ck ∈ Ck×k is nilpotent unreduced HESSENBERG
Fk ∈ Cn×k is “small” compared to Qk
Columns of Qk may be dependent from the beginning.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 39: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/39.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
a rather silly method
Consider any v 6= 0 such that Av = vλ with λ 6= 0
A ∈ Cn×n is a general matrix not identical zero
Qk ≡[v , . . . , v
]∈ Cn×k
Ck ∈ Ck×k should be unreduced HESSENBERG
Set
Ck ≡(
oTk−1 0
λIk−1 λek−1
)(2)
Then AQk = QkCk .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 40: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/40.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
a rather silly method
Consider any v 6= 0 such that Av = vλ with λ 6= 0
A ∈ Cn×n is a general matrix not identical zero
Qk ≡[v , . . . , v
]∈ Cn×k
Ck ∈ Ck×k should be unreduced HESSENBERG
Set
Ck ≡(
oTk−1 0
λIk−1 λek−1
)(2)
Then AQk = QkCk .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 41: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/41.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
a rather silly method
Consider any v 6= 0 such that Av = vλ with λ 6= 0
A ∈ Cn×n is a general matrix not identical zero
Qk ≡[v , . . . , v
]∈ Cn×k
Ck ∈ Ck×k should be unreduced HESSENBERG
Set
Ck ≡(
oTk−1 0
λIk−1 λek−1
)(2)
Then AQk = QkCk .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 42: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/42.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
a rather silly method
Consider any v 6= 0 such that Av = vλ with λ 6= 0
A ∈ Cn×n is a general matrix not identical zero
Qk ≡[v , . . . , v
]∈ Cn×k
Ck ∈ Ck×k should be unreduced HESSENBERG
Set
Ck ≡(
oTk−1 0
λIk−1 λek−1
)(2)
Then AQk = QkCk .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 43: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/43.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
a rather silly method
Consider any v 6= 0 such that Av = vλ with λ 6= 0
A ∈ Cn×n is a general matrix not identical zero
Qk ≡[v , . . . , v
]∈ Cn×k
Ck ∈ Ck×k should be unreduced HESSENBERG
Set
Ck ≡(
oTk−1 0
λIk−1 λek−1
)(2)
Then AQk = QkCk .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 44: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/44.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 45: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/45.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
eigenmatrices et al.
JORDAN form, eigenmatrices:
AV = VJΛ, CkSk = SkJΘ. (3)
left eigenmatrices:
V H ≡ V T ≡ V−1, SHk ≡ ST
k ≡ S−1k . (4)
JORDAN matrices (, boxes) and blocks:
JΛ = ⊕ Jλ, Jλ = ⊕ Jλι, JΘ = ⊕ Jθ. (5)
partial eigenmatrices:
V = ⊕Vλ, Vλ = ⊕Vλι, Sk = ⊕Sθ. (6)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 46: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/46.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
eigenmatrices et al.
JORDAN form, eigenmatrices:
AV = VJΛ, CkSk = SkJΘ. (3)
left eigenmatrices:
V H ≡ V T ≡ V−1, SHk ≡ ST
k ≡ S−1k . (4)
JORDAN matrices (, boxes) and blocks:
JΛ = ⊕ Jλ, Jλ = ⊕ Jλι, JΘ = ⊕ Jθ. (5)
partial eigenmatrices:
V = ⊕Vλ, Vλ = ⊕Vλι, Sk = ⊕Sθ. (6)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 47: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/47.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
eigenmatrices et al.
JORDAN form, eigenmatrices:
AV = VJΛ, CkSk = SkJΘ. (3)
left eigenmatrices:
V H ≡ V T ≡ V−1, SHk ≡ ST
k ≡ S−1k . (4)
JORDAN matrices (, boxes) and blocks:
JΛ = ⊕ Jλ, Jλ = ⊕ Jλι, JΘ = ⊕ Jθ. (5)
partial eigenmatrices:
V = ⊕Vλ, Vλ = ⊕Vλι, Sk = ⊕Sθ. (6)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 48: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/48.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
eigenmatrices et al.
JORDAN form, eigenmatrices:
AV = VJΛ, CkSk = SkJΘ. (3)
left eigenmatrices:
V H ≡ V T ≡ V−1, SHk ≡ ST
k ≡ S−1k . (4)
JORDAN matrices (, boxes) and blocks:
JΛ = ⊕ Jλ, Jλ = ⊕ Jλι, JΘ = ⊕ Jθ. (5)
partial eigenmatrices:
V = ⊕Vλ, Vλ = ⊕Vλι, Sk = ⊕Sθ. (6)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 49: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/49.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
characteristic matrix et al.
characteristic matrices:
zA ≡ zI − A, zCk ≡ zIk − Ck . (7)
the adjugate:P(z) ≡ adj(zCk ). (8)
characteristic polynomials:
χCk(z) ≡ det(zCk ), χCi:j
(z) ≡ det(zCi:j). (9)
reduced characteristic polynomial:
χCk(z) = (z − θ)αω(z). (10)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 50: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/50.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
characteristic matrix et al.
characteristic matrices:
zA ≡ zI − A, zCk ≡ zIk − Ck . (7)
the adjugate:P(z) ≡ adj(zCk ). (8)
characteristic polynomials:
χCk(z) ≡ det(zCk ), χCi:j
(z) ≡ det(zCi:j). (9)
reduced characteristic polynomial:
χCk(z) = (z − θ)αω(z). (10)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 51: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/51.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
characteristic matrix et al.
characteristic matrices:
zA ≡ zI − A, zCk ≡ zIk − Ck . (7)
the adjugate:P(z) ≡ adj(zCk ). (8)
characteristic polynomials:
χCk(z) ≡ det(zCk ), χCi:j
(z) ≡ det(zCi:j). (9)
reduced characteristic polynomial:
χCk(z) = (z − θ)αω(z). (10)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 52: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/52.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
characteristic matrix et al.
characteristic matrices:
zA ≡ zI − A, zCk ≡ zIk − Ck . (7)
the adjugate:P(z) ≡ adj(zCk ). (8)
characteristic polynomials:
χCk(z) ≡ det(zCk ), χCi:j
(z) ≡ det(zCi:j). (9)
reduced characteristic polynomial:
χCk(z) = (z − θ)αω(z). (10)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 53: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/53.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 54: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/54.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
HESSENBERG eigenvalue-eigenmatrix relations
Definition (off-diagonal products)
We denote the products of off-diagonal elements by
ci:j ≡∏j
`=ic`+1,`. (11)
Definition (polynomial vectors ν and ν)
We define vectors of (scaled) characteristic polynomials by
ν(z) ≡(
χCl+1:k(z)
cl:k−1
)k
l=1, ν(z) ≡
(χCl−1
(z)
c1:l−1
)k
l=1. (12)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 55: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/55.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
HESSENBERG eigenvalue-eigenmatrix relations
Definition (off-diagonal products)
We denote the products of off-diagonal elements by
ci:j ≡∏j
`=ic`+1,`. (11)
Definition (polynomial vectors ν and ν)
We define vectors of (scaled) characteristic polynomials by
ν(z) ≡(
χCl+1:k(z)
cl:k−1
)k
l=1, ν(z) ≡
(χCl−1
(z)
c1:l−1
)k
l=1. (12)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 56: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/56.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
HESSENBERG eigenvalue-eigenmatrix relations
Definition (matrices of derivatives)
We define rectangular matrices collecting the derivatives by
Sα−1(θ) ≡
[ν(θ), ν ′(θ),
ν ′′(θ)
2, . . . ,
ν(α−1)(θ)
(α− 1)!
](13)
Sα−1(θ) ≡
[ν(α−1)(θ)
(α− 1)!, . . . ,
ν ′′(θ)
2, ν ′(θ), ν(θ)
](14)
Observation
These matrices gather complete left and right JORDAN chains.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 57: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/57.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
HESSENBERG eigenvalue-eigenmatrix relations
Definition (matrices of derivatives)
We define rectangular matrices collecting the derivatives by
Sα−1(θ) ≡
[ν(θ), ν ′(θ),
ν ′′(θ)
2, . . . ,
ν(α−1)(θ)
(α− 1)!
](13)
Sα−1(θ) ≡
[ν(α−1)(θ)
(α− 1)!, . . . ,
ν ′′(θ)
2, ν ′(θ), ν(θ)
](14)
Observation
These matrices gather complete left and right JORDAN chains.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 58: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/58.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
HESSENBERG eigenvalue-eigenmatrix relations
Theorem (HEER)
HESSENBERG eigenmatrices satisfy
P(α−1)(θ)
(α− 1)!= Sθ ω(Jθ) ST
θ = c1:k−1 Sα−1(θ) Sα−1(θ)T . (15)
Proof.
Proof based on comparison of TAYLOR expansions of theadjugate P(z) as inverse divided by determinant and thepolynomial expression for the adjugate in terms ofcharacteristic polynomials of submatrices (Zemke 2004,submitted to LAA).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 59: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/59.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
HESSENBERG eigenvalue-eigenmatrix relations
Theorem (HEER)
HESSENBERG eigenmatrices satisfy
P(α−1)(θ)
(α− 1)!= Sθ ω(Jθ) ST
θ = c1:k−1 Sα−1(θ) Sα−1(θ)T . (15)
Proof.
Proof based on comparison of TAYLOR expansions of theadjugate P(z) as inverse divided by determinant and thepolynomial expression for the adjugate in terms ofcharacteristic polynomials of submatrices (Zemke 2004,submitted to LAA).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 60: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/60.jpg)
Getting startedThe results ...
... and their impactsSummary
the name of the gamea few examplesbasic notationsHESSENBERG structure
HESSENBERG eigenvalue-eigenmatrix relations
Lemma (HEER)
We can choose the partial eigenmatrices such that
eT1 Sθ = eT
α (ω(Jθ))−T , (16a)
STθ el = c1:l−1χCl+1:k
(Jθ)T e1. (16b)
Tailored to diagonalizable Ck :
sljs`j =χC1:l−1
(θj)cl:`−1χC`+1:k(θj)
χ′Ck(θj)
∀ l 6 `. (17)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 61: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/61.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 62: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/62.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
basic definitions
Definition (basis polynomials)
We define the (trailing) basis polynomials by
Bk (z) ≡χCk
(z)
c1:k= νk+1(z), (18)
Bl+1:k (z) ≡χCl+1:k
(z)
cl+1:k=
cl+1,l
ck+1,kνl(z), ∀ l = 1, . . . , k . (19)
Observation
The trailing basis polynomials are the basis polynomials of thetrailing submatrices Cl+1:k .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 63: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/63.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
basic definitions
Definition (basis polynomials)
We define the (trailing) basis polynomials by
Bk (z) ≡χCk
(z)
c1:k= νk+1(z), (18)
Bl+1:k (z) ≡χCl+1:k
(z)
cl+1:k=
cl+1,l
ck+1,kνl(z), ∀ l = 1, . . . , k . (19)
Observation
The trailing basis polynomials are the basis polynomials of thetrailing submatrices Cl+1:k .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 64: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/64.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
“basis” vectors
Theorem (the “basis” vectors)
The “basis” vectors of a KRYLOV method are given by
qk+1 = Bk (A)q1
+k∑
l=1
Bl+1:k (A)fl
cl+1,l
. (20)
Observation
The perturbed “basis” vectors can be interpreted as an additiveoverlay of exact “basis” vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 65: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/65.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
“basis” vectors
Theorem (the “basis” vectors)
The “basis” vectors of a perturbed KRYLOV method are given by
qk+1 = Bk (A)q1 +k∑
l=1
Bl+1:k (A)fl
cl+1,l. (20)
Observation
The perturbed “basis” vectors can be interpreted as an additiveoverlay of exact “basis” vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 66: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/66.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
“basis” vectors
Theorem (the “basis” vectors)
The “basis” vectors of a perturbed KRYLOV method are given by
qk+1 = Bk (A)q1 +k∑
l=1
Bl+1:k (A)fl
cl+1,l. (20)
Observation
The perturbed “basis” vectors can be interpreted as an additiveoverlay of exact “basis” vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 67: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/67.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a rough sketch of a short proof
Proof.Introduce variable z:
Mk = Qk (zI − Ck ) + (zI − A)Qk + Fk
Mkadj(zCk ) = QkχCk(z) + (zI − A)Qkadj(zCk ) + Fkadj(zCk ).
HEER: adj(zCk )e1 = c1:k−1ν(z). Insert A into
ck+1,kqk+1 =q1χCk
(z)
c1:k−1+ (zI − A)Qkν(z) + Fkν(z).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 68: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/68.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a rough sketch of a short proof
Proof.Introduce variable z:
Mk = Qk (zI − Ck ) + (zI − A)Qk + Fk
Mkadj(zCk ) = QkχCk(z) + (zI − A)Qkadj(zCk ) + Fkadj(zCk ).
HEER: adj(zCk )e1 = c1:k−1ν(z). Insert A into
ck+1,kqk+1 =q1χCk
(z)
c1:k−1+ (zI − A)Qkν(z) + Fkν(z).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 69: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/69.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a rough sketch of a short proof
Proof.Introduce variable z:
Mk = Qk (zI − Ck ) + (zI − A)Qk + Fk
Mkadj(zCk ) = QkχCk(z) + (zI − A)Qkadj(zCk ) + Fkadj(zCk ).
HEER: adj(zCk )e1 = c1:k−1ν(z).
Insert A into
ck+1,kqk+1 =q1χCk
(z)
c1:k−1+ (zI − A)Qkν(z) + Fkν(z).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 70: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/70.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a rough sketch of a short proof
Proof.Introduce variable z:
Mk = Qk (zI − Ck ) + (zI − A)Qk + Fk
Mkadj(zCk ) = QkχCk(z) + (zI − A)Qkadj(zCk ) + Fkadj(zCk ).
HEER: adj(zCk )e1 = c1:k−1ν(z). Insert A into
ck+1,kqk+1 =q1χCk
(z)
c1:k−1+ (zI − A)Qkν(z) + Fkν(z).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 71: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/71.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a closer & deeper look
Theorem (the “basis” vectors revisited)
Let Ck be diagonalizable and suppose that λ 6= θj for all j : k∑j=1
c1:k
χ′Ck(θj)(λ− θj)
vHqk+1 = vHq1
+k∑
l=1
k∑j=1
c1:lχCl+1:k(θj)
χ′Ck(θj)(λ− θj)
vH flcl+1,l
.
Remark
Generalization to the non-diagonalizable case exists.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 72: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/72.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a closer & deeper look
Theorem (the “basis” vectors revisited)
Let Ck be diagonalizable and suppose that λ 6= θj for all j : k∑j=1
c1:k
χ′Ck(θj)(λ− θj)
vHqk+1 = vHq1
+k∑
l=1
k∑j=1
c1:lχCl+1:k(θj)
χ′Ck(θj)(λ− θj)
vH flcl+1,l
.
Remark
Generalization to the non-diagonalizable case exists.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 73: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/73.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a closer & deeper look
Theorem (the “basis” vectors revisited)
Let Ck be diagonalizable and suppose that λ 6= θj for all j : k∑j=1
c1:k
χ′Ck(θj)(λ− θj)
vHqk+1 = vHq1
+k∑
l=1
k∑j=1
c1:lχCl+1:k(θj)
χ′Ck(θj)(λ− θj)
vH flcl+1,l
.
Remark
Generalization to the non-diagonalizable case exists.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 74: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/74.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 75: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/75.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
eigenvalues, JORDAN block, partial eigenmatrix
Unreduced HESSENBERG matrices Ck are non-derogatory.
Notations
In the following,
(generic) eigenvalue: denoted by θ = θ(k),
(algebraic) multiplicity: denoted by α = α(θ),
JORDAN block: denoted by Jθ = J(k)θ ,
partial eigenmatrix: Sθ = S(k)θ .
The matrices are such that
CkSθ = SθJθ, where Jθ ∈ Cα×α. (21)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 76: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/76.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
eigenvalues, JORDAN block, partial eigenmatrix
Unreduced HESSENBERG matrices Ck are non-derogatory.
Notations
In the following,
(generic) eigenvalue: denoted by θ = θ(k),
(algebraic) multiplicity: denoted by α = α(θ),
JORDAN block: denoted by Jθ = J(k)θ ,
partial eigenmatrix: Sθ = S(k)θ .
The matrices are such that
CkSθ = SθJθ, where Jθ ∈ Cα×α. (21)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 77: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/77.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
eigenvalues, JORDAN block, partial eigenmatrix
Unreduced HESSENBERG matrices Ck are non-derogatory.
Notations
In the following,
(generic) eigenvalue: denoted by θ = θ(k),
(algebraic) multiplicity: denoted by α = α(θ),
JORDAN block: denoted by Jθ = J(k)θ ,
partial eigenmatrix: Sθ = S(k)θ .
The matrices are such that
CkSθ = SθJθ, where Jθ ∈ Cα×α. (21)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 78: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/78.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
eigenvalues, JORDAN block, partial eigenmatrix
Unreduced HESSENBERG matrices Ck are non-derogatory.
Notations
In the following,
(generic) eigenvalue: denoted by θ = θ(k),
(algebraic) multiplicity: denoted by α = α(θ),
JORDAN block: denoted by Jθ = J(k)θ ,
partial eigenmatrix: Sθ = S(k)θ .
The matrices are such that
CkSθ = SθJθ, where Jθ ∈ Cα×α. (21)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 79: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/79.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
eigenvalues, JORDAN block, partial eigenmatrix
Unreduced HESSENBERG matrices Ck are non-derogatory.
Notations
In the following,
(generic) eigenvalue: denoted by θ = θ(k),
(algebraic) multiplicity: denoted by α = α(θ),
JORDAN block: denoted by Jθ = J(k)θ ,
partial eigenmatrix: Sθ = S(k)θ .
The matrices are such that
CkSθ = SθJθ, where Jθ ∈ Cα×α. (21)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 80: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/80.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ pairs, RITZ residuals
Definition (RITZ pair)
Define RITZ pair by
(Jθ, Yθ ≡ QkSθ). (22)
Not necessarily a “true” RITZ pair, since there need to be noRITZ projection associated with it.
Observation
A backward expression for the RITZ residual is given by
AYθ − YθJθ = qk+1ck+1,keTk Sθ − FkSθ. (23)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 81: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/81.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ pairs, RITZ residuals
Definition (RITZ pair)
Define RITZ pair by
(Jθ, Yθ ≡ QkSθ). (22)
Not necessarily a “true” RITZ pair, since there need to be noRITZ projection associated with it.
Observation
A backward expression for the RITZ residual is given by
AYθ − YθJθ = qk+1ck+1,keTk Sθ − FkSθ. (23)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 82: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/82.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ pairs, RITZ residuals
Definition (RITZ pair)
Define RITZ pair by
(Jθ, Yθ ≡ QkSθ). (22)
Not necessarily a “true” RITZ pair, since there need to be noRITZ projection associated with it.
Observation
A backward expression for the RITZ residual is given by
AYθ − YθJθ = qk+1ck+1,keTk Sθ − FkSθ. (23)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 83: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/83.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ residuals (generic case)
Theorem (generic RITZ residuals)
The RITZ residual for an (arbitrarily chosen) RITZ pair:
AYθ − YθJθ =
(χCk
(A)
c1:k
)q1eT
k Sθ
+k∑
l=1
(χCl+1:k
(A)
cl:k−1
)fle
Tk Sθ − fle
Tl Sθ. (24)
Proof.
Backward expression and Theorem on the “basis” vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 84: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/84.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ residuals (generic case)
Theorem (generic RITZ residuals)
The RITZ residual for an (arbitrarily chosen) RITZ pair:
AYθ − YθJθ =
(χCk
(A)
c1:k
)q1eT
k Sθ
+k∑
l=1
(χCl+1:k
(A)
cl:k−1
)fle
Tk Sθ − fle
Tl Sθ. (24)
Proof.
Backward expression and Theorem on the “basis” vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 85: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/85.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ residuals (special case)
Use (unique) choice for the partial eigenmatrix Sθ (HEER):
Theorem (special RITZ residuals)
The RITZ residual for the special partial eigenmatrix fromHEER is given by
AYθ − YθJθ = χCk(A)q1eT
1
+k∑
l=1
c1:l−1
(χCl+1:k
(A)fleT1 − fle
T1 χCl+1:k
(Jθ)). (25)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 86: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/86.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
bivariate adjugate polynomials
Definition (bivariate adjugate polynomials)
We define the bivariate adjugate polynomials by
Ak (θ, z) ≡{ (
χCk(θ)− χCk
(z))(θ − z)−1, z 6= θ,
χ′Ck(z), z = θ.
(26)
Trailing bivariate adjugate polynomials Al+1:k are defined usingCl+1:k in place of Ck , l = 1, . . . , k .
Observation
Even with an eigenvalue θ: Ak (θ, Ck ) = adj(θIk − Ck ) = P(θ).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 87: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/87.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
bivariate adjugate polynomials
Definition (bivariate adjugate polynomials)
We define the bivariate adjugate polynomials by
Ak (θ, z) ≡{ (
χCk(θ)− χCk
(z))(θ − z)−1, z 6= θ,
χ′Ck(z), z = θ.
(26)
Trailing bivariate adjugate polynomials Al+1:k are defined usingCl+1:k in place of Ck , l = 1, . . . , k .
Observation
Even with an eigenvalue θ: Ak (θ, Ck ) = adj(θIk − Ck ) = P(θ).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 88: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/88.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ vectors
Theorem (the RITZ vectors)
The RITZ vectors of a KRYLOV method are given by
vec(Yθ) =Ak (θ, A)A′
k (θ, A)...
A(α−1)k (θ, A)
(α− 1)!
q1
+k∑
l=1
c1:l−1
Al+1:k (θ, A)A′
l+1:k (θ, A)...
A(α−1)l+1:k (θ, A)
(α− 1)!
fl
. (27)
(derivation with respect to “shift” θ)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 89: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/89.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
RITZ vectors
Theorem (the RITZ vectors)
The RITZ vectors of a perturbed KRYLOV method are given by
vec(Yθ) =Ak (θ, A)A′
k (θ, A)...
A(α−1)k (θ, A)
(α− 1)!
q1 +k∑
l=1
c1:l−1
Al+1:k (θ, A)A′
l+1:k (θ, A)...
A(α−1)l+1:k (θ, A)
(α− 1)!
fl . (27)
(derivation with respect to “shift” θ)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 90: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/90.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: basics
The proof utilizes the following general aspects:
The adjugate of a matrix is defined as matrix of cofactors.
The adjugate is linked to eigenvectors and, more general,principal vectors.
The adjugate is linked to the inverse and the determinant.
The problem: the definition of the bivariate adjugatepolynomials given here is not “adequate”, we need anotherform.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 91: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/91.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: basics
The proof utilizes the following general aspects:
The adjugate of a matrix is defined as matrix of cofactors.
The adjugate is linked to eigenvectors and, more general,principal vectors.
The adjugate is linked to the inverse and the determinant.
The problem: the definition of the bivariate adjugatepolynomials given here is not “adequate”, we need anotherform.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 92: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/92.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: basics
The proof utilizes the following general aspects:
The adjugate of a matrix is defined as matrix of cofactors.
The adjugate is linked to eigenvectors and, more general,principal vectors.
The adjugate is linked to the inverse and the determinant.
The problem: the definition of the bivariate adjugatepolynomials given here is not “adequate”, we need anotherform.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 93: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/93.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: basics
The proof utilizes the following general aspects:
The adjugate of a matrix is defined as matrix of cofactors.
The adjugate is linked to eigenvectors and, more general,principal vectors.
The adjugate is linked to the inverse and the determinant.
The problem: the definition of the bivariate adjugatepolynomials given here is not “adequate”, we need anotherform.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 94: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/94.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: basics
The proof utilizes the following general aspects:
The adjugate of a matrix is defined as matrix of cofactors.
The adjugate is linked to eigenvectors and, more general,principal vectors.
The adjugate is linked to the inverse and the determinant.
The problem: the definition of the bivariate adjugatepolynomials given here is not “adequate”, we need anotherform.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 95: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/95.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: HESSENBERG basics
To derive this peculiar form we use the first adjugate identity:
Lemma (first (HESSENBERG) adjugate identity)
First adjugate identity:
(z − θ)adj(zA)adj(θA) = det(zA)adj(θA)− det(θA)adj(zA). (28)
Specialized to HESSENBERG matrices:
(z − θ)k∑
j=1
χC1:j−1(z)χCj+1:k
(θ) = χCk(z)− χCk
(θ). (29)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 96: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/96.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: HESSENBERG basics
To derive this peculiar form we use the first adjugate identity:
Lemma (first (HESSENBERG) adjugate identity)
First adjugate identity:
(z − θ)adj(zA)adj(θA) = det(zA)adj(θA)− det(θA)adj(zA). (28)
Specialized to HESSENBERG matrices:
(z − θ)k∑
j=1
χC1:j−1(z)χCj+1:k
(θ) = χCk(z)− χCk
(θ). (29)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 97: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/97.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: HESSENBERG basics
To derive this peculiar form we use the first adjugate identity:
Lemma (first (HESSENBERG) adjugate identity)
First adjugate identity:
(z − θ)adj(zA)adj(θA) = det(zA)adj(θA)− det(θA)adj(zA). (28)
Specialized to HESSENBERG matrices:
(z − θ)k∑
j=1
χC1:j−1(z)χCj+1:k
(θ) = χCk(z)− χCk
(θ). (29)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 98: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/98.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: gluing results together
The last line implies the following representations (` > 0):
A(`)l+1:k (θ, z) =
k∑j=l+1
χCl+1:j−1(z)χ
(`)Cj+1:k
(θ) ∀ l = 0, 1, . . . , k . (30)
This together with
the special choice of the partial eigenmatrix Sθ
the representation of the “basis” vectors
are the building blocks for the proof.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 99: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/99.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: gluing results together
The last line implies the following representations (` > 0):
A(`)l+1:k (θ, z) =
k∑j=l+1
χCl+1:j−1(z)χ
(`)Cj+1:k
(θ) ∀ l = 0, 1, . . . , k . (30)
This together with
the special choice of the partial eigenmatrix Sθ
the representation of the “basis” vectors
are the building blocks for the proof.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 100: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/100.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: gluing results together
The last line implies the following representations (` > 0):
A(`)l+1:k (θ, z) =
k∑j=l+1
χCl+1:j−1(z)χ
(`)Cj+1:k
(θ) ∀ l = 0, 1, . . . , k . (30)
This together with
the special choice of the partial eigenmatrix Sθ
the representation of the “basis” vectors
are the building blocks for the proof.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 101: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/101.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
sketch of proof: gluing results together
The last line implies the following representations (` > 0):
A(`)l+1:k (θ, z) =
k∑j=l+1
χCl+1:j−1(z)χ
(`)Cj+1:k
(θ) ∀ l = 0, 1, . . . , k . (30)
This together with
the special choice of the partial eigenmatrix Sθ
the representation of the “basis” vectors
are the building blocks for the proof.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 102: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/102.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 103: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/103.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the approach
Suppose that Ck is invertible and that q1 = r0/‖r0‖. Let zk
denote the solution to the linear system of equations
Ckzk = e1‖r0‖. (31)
Define the k th (Q)OR iterate xk by
xk = Qkzk (32)
and the k th (true) (Q)OR residual by
rk = r0 − Axk . (33)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 104: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/104.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the approach
Suppose that Ck is invertible and that q1 = r0/‖r0‖. Let zk
denote the solution to the linear system of equations
Ckzk = e1‖r0‖. (31)
Define the k th (Q)OR iterate xk by
xk = Qkzk (32)
and the k th (true) (Q)OR residual by
rk = r0 − Axk . (33)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 105: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/105.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the approach
Suppose that Ck is invertible and that q1 = r0/‖r0‖. Let zk
denote the solution to the linear system of equations
Ckzk = e1‖r0‖. (31)
Define the k th (Q)OR iterate xk by
xk = Qkzk (32)
and the k th (true) (Q)OR residual by
rk = r0 − Axk . (33)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 106: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/106.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a backward expression for the (Q)OR residual
Observation
A backward expression for the (Q)OR residual is given by
rk = r0 − Axk = (QkCk − AQk )C−1k e1‖r0‖
= (−qk+1ck+1,keTk + Fk )zk
= −qk+1ck+1,kzkk +k∑
l=1
flzlk
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 107: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/107.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a backward expression for the (Q)OR residual
Observation
A backward expression for the (Q)OR residual is given by
rk = r0 − Axk = (QkCk − AQk )C−1k e1‖r0‖
= (−qk+1ck+1,keTk + Fk )zk
= −qk+1ck+1,kzkk +k∑
l=1
flzlk
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 108: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/108.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
a backward expression for the (Q)OR residual
Observation
A backward expression for the (Q)OR residual is given by
rk = r0 − Axk = (QkCk − AQk )C−1k e1‖r0‖
= (−qk+1ck+1,keTk + Fk )zk
= −qk+1ck+1,kzkk +k∑
l=1
flzlk
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 109: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/109.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, determinant
Express the inverse of Ck as adjugate by determinant:
−zlk
‖r0‖= eT
l (−Ck )−1e1
=eT
l adj(−Ck )e1
det(−Ck )
=c1:l−1χCl+1:k
(0)
χCk(0)
.
Utilize
rk = qk+1ck+1,k (−zkk )−k∑
l=1
fl(−zlk ). (34)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 110: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/110.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, determinant
Express the inverse of Ck as adjugate by determinant:
−zlk
‖r0‖= eT
l (−Ck )−1e1 =eT
l adj(−Ck )e1
det(−Ck )
=c1:l−1χCl+1:k
(0)
χCk(0)
.
Utilize
rk = qk+1ck+1,k (−zkk )−k∑
l=1
fl(−zlk ). (34)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 111: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/111.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, determinant
Express the inverse of Ck as adjugate by determinant:
−zlk
‖r0‖= eT
l (−Ck )−1e1 =eT
l adj(−Ck )e1
det(−Ck )
=c1:l−1χCl+1:k
(0)
χCk(0)
.
Utilize
rk = qk+1ck+1,k (−zkk )−k∑
l=1
fl(−zlk ). (34)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 112: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/112.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, determinant
Express the inverse of Ck as adjugate by determinant:
−zlk
‖r0‖= eT
l (−Ck )−1e1 =eT
l adj(−Ck )e1
det(−Ck )
=c1:l−1χCl+1:k
(0)
χCk(0)
.
Utilize
rk = qk+1ck+1,k (−zkk )−k∑
l=1
fl(−zlk ). (34)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 113: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/113.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the residuals
This backward expression plus Theorem on the “basis” vectors:
Theorem (the (Q)OR residual vectors)
The residual vectors of a
perturbed
(Q)OR KRYLOV method aregiven by
rk =χCk
(A)
χCk(0)
r0
+ ‖r0‖k∑
l=1
c1:l−1χCl+1:k
(A)− χCl+1:k(0)
χCk(0)
fl
. (35)
The perturbation terms remind of adjugate polynomials . . .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 114: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/114.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the residuals
This backward expression plus Theorem on the “basis” vectors:
Theorem (the (Q)OR residual vectors)
The residual vectors of a perturbed (Q)OR KRYLOV method aregiven by
rk =χCk
(A)
χCk(0)
r0 + ‖r0‖k∑
l=1
c1:l−1χCl+1:k
(A)− χCl+1:k(0)
χCk(0)
fl . (35)
The perturbation terms remind of adjugate polynomials . . .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 115: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/115.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the residuals
This backward expression plus Theorem on the “basis” vectors:
Theorem (the (Q)OR residual vectors)
The residual vectors of a perturbed (Q)OR KRYLOV method aregiven by
rk =χCk
(A)
χCk(0)
r0 + ‖r0‖k∑
l=1
c1:l−1χCl+1:k
(A)− χCl+1:k(0)
χCk(0)
fl . (35)
The perturbation terms remind of adjugate polynomials . . .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 116: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/116.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, interpolation (I)
Definition (univariate adjugate polynomials)
We define univariate adjugate polynomials by
Ak (z) = (−1)k (χCk(0)− χCk
(z))z−1
By CAYLEY-HAMILTON: Ak (Ck ) = adj(Ck )
ObservationUnivariate and bivariate adjugate polynomials are related by
Ak (z) = (−1)k+1Ak (z, 0) = (−1)k+1Ak (0, z)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 117: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/117.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, interpolation (I)
Definition (univariate adjugate polynomials)
We define univariate adjugate polynomials by
Ak (z) = (−1)k (χCk(0)− χCk
(z))z−1
By CAYLEY-HAMILTON: Ak (Ck ) = adj(Ck )
ObservationUnivariate and bivariate adjugate polynomials are related by
Ak (z) = (−1)k+1Ak (z, 0) = (−1)k+1Ak (0, z)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 118: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/118.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, interpolation (II)
Notations
We define and denote the LAGRANGE interpolation of theinverse by
Lk [z−1](z) =Ak (z)
det(Ck )=
(1 −
χCk(z)
χCk(0)
)z−1
Notations
We define and denote the LAGRANGE interpolation of aperturbed identity by
L0k [1 − δz0](z) = Lk [z−1](z)z =
χCk(0)− χCk
(z)
χCk(0)
.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 119: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/119.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
adjugate, inverse, interpolation (II)
Notations
We define and denote the LAGRANGE interpolation of theinverse by
Lk [z−1](z) =Ak (z)
det(Ck )=
(1 −
χCk(z)
χCk(0)
)z−1
Notations
We define and denote the LAGRANGE interpolation of aperturbed identity by
L0k [1 − δz0](z) = Lk [z−1](z)z =
χCk(0)− χCk
(z)
χCk(0)
.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 120: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/120.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
trailing {adjugate, inverse, interpolation}
We expand all notations to the trailing submatrices Cl+1:k .
Then,
c1:l−1χCl+1:k
(0)− χCl+1:k(A)
χCk(0)
=
χCl+1:k(0)− χCl+1:k
(A)
χCl+1:k(0)
·c1:l−1χCl+1:k
(0)
χCk(0)
=
L0l+1:k [1 − δz0](A)
zlk
‖r0‖
(36)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 121: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/121.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
trailing {adjugate, inverse, interpolation}
We expand all notations to the trailing submatrices Cl+1:k .Then,
c1:l−1χCl+1:k
(0)− χCl+1:k(A)
χCk(0)
=
χCl+1:k(0)− χCl+1:k
(A)
χCl+1:k(0)
·c1:l−1χCl+1:k
(0)
χCk(0)
=
L0l+1:k [1 − δz0](A)
zlk
‖r0‖
(36)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 122: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/122.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
trailing {adjugate, inverse, interpolation}
We expand all notations to the trailing submatrices Cl+1:k .Then,
c1:l−1χCl+1:k
(0)− χCl+1:k(A)
χCk(0)
=
χCl+1:k(0)− χCl+1:k
(A)
χCl+1:k(0)
·c1:l−1χCl+1:k
(0)
χCk(0)
=
L0l+1:k [1 − δz0](A)
zlk
‖r0‖
(36)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 123: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/123.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
trailing {adjugate, inverse, interpolation}
We expand all notations to the trailing submatrices Cl+1:k .Then,
c1:l−1χCl+1:k
(0)− χCl+1:k(A)
χCk(0)
=
χCl+1:k(0)− χCl+1:k
(A)
χCl+1:k(0)
·c1:l−1χCl+1:k
(0)
χCk(0)
=
L0l+1:k [1 − δz0](A)
zlk
‖r0‖
(36)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 124: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/124.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
trailing {adjugate, inverse, interpolation}
We expand all notations to the trailing submatrices Cl+1:k .Then,
c1:l−1χCl+1:k
(0)− χCl+1:k(A)
χCk(0)
=
χCl+1:k(0)− χCl+1:k
(A)
χCl+1:k(0)
·c1:l−1χCl+1:k
(0)
χCk(0)
=
L0l+1:k [1 − δz0](A)
zlk
‖r0‖
(36)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 125: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/125.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
trailing {adjugate, inverse, interpolation}
We expand all notations to the trailing submatrices Cl+1:k .Then,
c1:l−1χCl+1:k
(0)− χCl+1:k(A)
χCk(0)
=
χCl+1:k(0)− χCl+1:k
(A)
χCl+1:k(0)
·c1:l−1χCl+1:k
(0)
χCk(0)
=
L0l+1:k [1 − δz0](A)
zlk
‖r0‖(36)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 126: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/126.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
trailing {adjugate, inverse, interpolation}
We expand all notations to the trailing submatrices Cl+1:k .Then,
c1:l−1χCl+1:k
(0)− χCl+1:k(A)
χCk(0)
=
χCl+1:k(0)− χCl+1:k
(A)
χCl+1:k(0)
·c1:l−1χCl+1:k
(0)
χCk(0)
=
L0l+1:k [1 − δz0](A)
zlk
‖r0‖(36)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 127: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/127.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the residuals
Theorem (the (Q)OR residual vectors)
Suppose that all submatrices Cl+1:k are nonsingular.Then the residual vectors can be written as
rk =χCk
(A)
χCk(0)
r0 −k∑
l=1
zlkL0l+1:k [1 − δz0](A) fl . (37)
This occurs frequently, consider e.g. CG for HPD A.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 128: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/128.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, regular A
What about the error vectors?
Theorem (the (Q)OR error vectors, regular A)
Suppose that A is invertible and let x = A−1r0 denote theunique solution of the linear system Ax = r0.Then the error vectors are given by
(x − xk ) =χCk
(A)
χCk(0)
(x − 0) + ‖r0‖k∑
l=1
c1:l−1Al+1:k (0, A)
χCk(0)
fl . (38)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 129: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/129.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, regular A
What about the error vectors?
Theorem (the (Q)OR error vectors, regular A)
Suppose that A is invertible and let x = A−1r0 denote theunique solution of the linear system Ax = r0.Then the error vectors are given by
(x − xk ) =χCk
(A)
χCk(0)
(x − 0) + ‖r0‖k∑
l=1
c1:l−1Al+1:k (0, A)
χCk(0)
fl . (38)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 130: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/130.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, regular A
What about invertible submatrices?
Theorem (the (Q)OR error vectors, regular A and Cl+1:k )
Suppose that all trailing submatrices Cl+1:k are nonsingular.Then the error vectors can be written as
(x − xk ) =χCk
(A)
χCk(0)
(x − 0)−k∑
l=1
zlkLl+1:k [z−1](A) fl . (39)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 131: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/131.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, regular A
What about invertible submatrices?
Theorem (the (Q)OR error vectors, regular A and Cl+1:k )
Suppose that all trailing submatrices Cl+1:k are nonsingular.Then the error vectors can be written as
(x − xk ) =χCk
(A)
χCk(0)
(x − 0)−k∑
l=1
zlkLl+1:k [z−1](A) fl . (39)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 132: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/132.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, singular A
What about singular A?
Theorem (the (Q)OR error vectors, singular A)
When A is singular, with x ≡ ADr0, where AD denotes theDRAZIN inverse of A,
(x − AADxk ) =χCk
(A)
χCk(0)
(x − 0)
+ ‖r0‖k∑
l=1
c1:l−1Al+1:k (0, A)
χCk(0)
AAD fl . (40)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 133: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/133.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, singular A
What about singular A?
Theorem (the (Q)OR error vectors, singular A)
When A is singular, with x ≡ ADr0, where AD denotes theDRAZIN inverse of A,
(x − AADxk ) =χCk
(A)
χCk(0)
(x − 0)
+ ‖r0‖k∑
l=1
c1:l−1Al+1:k (0, A)
χCk(0)
AAD fl . (40)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 134: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/134.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, singular A
What about invertible submatrices?
Theorem (the (Q)OR error vectors, singular A, regular Cl+1:k )
When A is singular, with x ≡ ADr0,
(x − AADxk ) =χCk
(A)
χCk(0)
(x − 0)
−k∑
l=1
zlkLl+1:k [z−1](A)AAD fl . (41)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 135: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/135.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the errors, singular A
What about invertible submatrices?
Theorem (the (Q)OR error vectors, singular A, regular Cl+1:k )
When A is singular, with x ≡ ADr0,
(x − AADxk ) =χCk
(A)
χCk(0)
(x − 0)
−k∑
l=1
zlkLl+1:k [z−1](A)AAD fl . (41)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 136: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/136.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the iterates
The iterates xk can be composed like the RITZ vectors.
Theorem (the (Q)OR iterates)
xk = Lk [z−1](A)r0 − ‖r0‖k∑
l=1
c1:l−1Al+1:k (0, A)
χCk(0)
fl . (42)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 137: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/137.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the iterates
The iterates xk can be composed like the RITZ vectors.
Theorem (the (Q)OR iterates)
xk = Lk [z−1](A)r0 − ‖r0‖k∑
l=1
c1:l−1Al+1:k (0, A)
χCk(0)
fl . (42)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 138: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/138.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the iterates
The case of invertible Cl+1:k :
Theorem (the (Q)OR iterates, regular Cl+1:k )
Suppose that all Cl+1:k are regular.Then
xk = Lk [z−1](A)r0 +k∑
l=1
zlkLl+1:k [z−1](A)fl . (43)
Observation
This is a linear combination of k + 1 approximations fromdistinct KRYLOV subspaces, spanned by the same matrix A, butdistinct starting vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 139: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/139.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the iterates
The case of invertible Cl+1:k :
Theorem (the (Q)OR iterates, regular Cl+1:k )
Suppose that all Cl+1:k are regular.Then
xk = Lk [z−1](A)r0 +k∑
l=1
zlkLl+1:k [z−1](A)fl . (43)
Observation
This is a linear combination of k + 1 approximations fromdistinct KRYLOV subspaces, spanned by the same matrix A, butdistinct starting vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 140: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/140.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)OR: the iterates
The case of invertible Cl+1:k :
Theorem (the (Q)OR iterates, regular Cl+1:k )
Suppose that all Cl+1:k are regular.Then
xk = Lk [z−1](A)r0 +k∑
l=1
zlkLl+1:k [z−1](A)fl . (43)
Observation
This is a linear combination of k + 1 approximations fromdistinct KRYLOV subspaces, spanned by the same matrix A, butdistinct starting vectors.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 141: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/141.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 142: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/142.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: the approach
Let zk denote the minimal-norm solution of the least-squaresproblem
‖Ckzk − e1‖r0‖‖ = min . (44)
Define the k th (Q)MR iterate xk by
xk = Qkzk (45)
and the k th quasi-residual by
rk = e1‖r0‖ − Ckzk = (Ik − CkC†k )e1‖r0‖. (46)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 143: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/143.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: the approach
Let zk denote the minimal-norm solution of the least-squaresproblem
‖Ckzk − e1‖r0‖‖ = min . (44)
Define the k th (Q)MR iterate xk by
xk = Qkzk (45)
and the k th quasi-residual by
rk = e1‖r0‖ − Ckzk = (Ik − CkC†k )e1‖r0‖. (46)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 144: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/144.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: the approach
Let zk denote the minimal-norm solution of the least-squaresproblem
‖Ckzk − e1‖r0‖‖ = min . (44)
Define the k th (Q)MR iterate xk by
xk = Qkzk (45)
and the k th quasi-residual by
rk = e1‖r0‖ − Ckzk = (Ik − CkC†k )e1‖r0‖. (46)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 145: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/145.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: a backward expression for the residual
Observation
The residual r k of the (Q)MR iterates has the followingbackward expression:
r k = r0 − Axk = Qk+1e1‖r0‖ − AQkzk (47)
= Qk+1(e1‖r0‖ − Ckzk ) + Fkzk = Qk+1rk +k∑
l=1
flz lk . (48)
Observation
To express the residual r k as polynomial in A, we “only” need“polynomial” expressions for rk and zk .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 146: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/146.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: a backward expression for the residual
Observation
The residual r k of the (Q)MR iterates has the followingbackward expression:
r k = r0 − Axk = Qk+1e1‖r0‖ − AQkzk (47)
= Qk+1(e1‖r0‖ − Ckzk ) + Fkzk = Qk+1rk +k∑
l=1
flz lk . (48)
Observation
To express the residual r k as polynomial in A, we “only” need“polynomial” expressions for rk and zk .
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 147: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/147.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: HESSENBERG rewritings
Definition (the scalar vectors µ, µ and µ)
We define pairs of vectors µj , µj ∈ Cj and µj ≡ µj ∈ Cj :
µ ≡
((−1)l+1 det(Cl+1:j)
cl:j−1
)j
l=1
, (49)
µ ≡(
(−1) j−l det(Cl−1)
c1:l−1
)j
l=1. (50)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 148: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/148.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: HESSENBERG rewritings
Lemma (MOORE-PENROSE inverse of extended HESSENBERG)
The MOORE-PENROSE inverse of the extended HESSENBERG
matrix Ck is given by
C†k =
k∑j=1
|cj+1:k |2(
det(Cj)adj(Cj) c1:jadj(Cj)µj Oj,k−j
Ok−j,j ok−j Ok−j
)∑k
j=0 |cj+1:k |2|det(Cj)|2.
(51)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 149: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/149.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: HESSENBERG rewritings
Lemma (the minimal norm solution)
The minimal norm solution zk is given by
zk
‖r0‖=
k∑j=1
|cj+1:k |2(
det(Cj)c1:j−1µj
ok−j
)∑k
j=0 |cj+1:k |2|det(Cj)|2(52)
= (−1)k+1
(ok adj(C4
k+1))
adj(CHk+1)ek+1∑k
j=0 |cj+1:k |2|det(Cj)|2. (53)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 150: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/150.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: HESSENBERG rewritings
Lemma ((Q)MR and (Q)OR)
Suppose all leading Cj are regular. Then the relation betweenthe kth (Q)MR solution zk and all prior (Q)OR solutions zj isgiven by
zk =
∑kj=0 |det(Cj)|2|cj+1:k |2
(zj
ok−j
)∑k
j=0 |det(Cj)|2|cj+1:k |2, (54)
where z0 is the empty matrix with dimensions 0 × 1.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 151: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/151.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: HESSENBERG rewritings
Lemma (the quasi-residual)
The quasi-residual rk is given by
rk
‖r0‖= c1:k
((−1)l−1cl:k det(Cl−1)∑k
j=0 |cj+1:k |2|det(Cj)|2
)k+1
l=1
. (55)
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 152: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/152.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: the residuals, errors and iterates
The (Q)MR residuals, errors and iterates can be composed liketheir (Q)OR counterparts . . .
Lacking is the “right” interpretation.
This is currently work in progress.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 153: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/153.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: the residuals, errors and iterates
The (Q)MR residuals, errors and iterates can be composed liketheir (Q)OR counterparts . . .
Lacking is the “right” interpretation.
This is currently work in progress.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 154: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/154.jpg)
Getting startedThe results ...
... and their impactsSummary
“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
(Q)MR: the residuals, errors and iterates
The (Q)MR residuals, errors and iterates can be composed liketheir (Q)OR counterparts . . .
Lacking is the “right” interpretation.
This is currently work in progress.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 155: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/155.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 156: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/156.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
general comments
The results . . .
do not prove anything about convergence.
do explain certain observations.
help in understanding the intrinsic behavior.
are well suited for classroom introduction.
are useful in connection with results on particular methods.
are aiding the design of particular finite precision/inexactmethods.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 157: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/157.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
general comments
The results . . .
do not prove anything about convergence.
do explain certain observations.
help in understanding the intrinsic behavior.
are well suited for classroom introduction.
are useful in connection with results on particular methods.
are aiding the design of particular finite precision/inexactmethods.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 158: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/158.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
general comments
The results . . .
do not prove anything about convergence.
do explain certain observations.
help in understanding the intrinsic behavior.
are well suited for classroom introduction.
are useful in connection with results on particular methods.
are aiding the design of particular finite precision/inexactmethods.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 159: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/159.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
general comments
The results . . .
do not prove anything about convergence.
do explain certain observations.
help in understanding the intrinsic behavior.
are well suited for classroom introduction.
are useful in connection with results on particular methods.
are aiding the design of particular finite precision/inexactmethods.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 160: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/160.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
general comments
The results . . .
do not prove anything about convergence.
do explain certain observations.
help in understanding the intrinsic behavior.
are well suited for classroom introduction.
are useful in connection with results on particular methods.
are aiding the design of particular finite precision/inexactmethods.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 161: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/161.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
general comments
The results . . .
do not prove anything about convergence.
do explain certain observations.
help in understanding the intrinsic behavior.
are well suited for classroom introduction.
are useful in connection with results on particular methods.
are aiding the design of particular finite precision/inexactmethods.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 162: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/162.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
general comments
The results . . .
do not prove anything about convergence.
do explain certain observations.
help in understanding the intrinsic behavior.
are well suited for classroom introduction.
are useful in connection with results on particular methods.
are aiding the design of particular finite precision/inexactmethods.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 163: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/163.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 164: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/164.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
descriptions
We know that finite precision CG/Lanczos methods
compute clusters of RITZ values resembling (simple)eigenvalues.
tend to show a “delay” in the convergence.
We can use the theorem(s)
on the “basis” vectors to explain the occurrence of multipleRITZ values.
on the RITZ residuals and vectors to understand the sizesof the RITZ vectors.
on the (Q)OR iterates to understand the “delay”.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 165: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/165.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
descriptions
We know that finite precision CG/Lanczos methods
compute clusters of RITZ values resembling (simple)eigenvalues.
tend to show a “delay” in the convergence.
We can use the theorem(s)
on the “basis” vectors to explain the occurrence of multipleRITZ values.
on the RITZ residuals and vectors to understand the sizesof the RITZ vectors.
on the (Q)OR iterates to understand the “delay”.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 166: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/166.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
descriptions
We know that finite precision CG/Lanczos methods
compute clusters of RITZ values resembling (simple)eigenvalues.
tend to show a “delay” in the convergence.
We can use the theorem(s)
on the “basis” vectors to explain the occurrence of multipleRITZ values.
on the RITZ residuals and vectors to understand the sizesof the RITZ vectors.
on the (Q)OR iterates to understand the “delay”.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 167: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/167.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
descriptions
We know that finite precision CG/Lanczos methods
compute clusters of RITZ values resembling (simple)eigenvalues.
tend to show a “delay” in the convergence.
We can use the theorem(s)
on the “basis” vectors to explain the occurrence of multipleRITZ values.
on the RITZ residuals and vectors to understand the sizesof the RITZ vectors.
on the (Q)OR iterates to understand the “delay”.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 168: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/168.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
descriptions
We know that finite precision CG/Lanczos methods
compute clusters of RITZ values resembling (simple)eigenvalues.
tend to show a “delay” in the convergence.
We can use the theorem(s)
on the “basis” vectors to explain the occurrence of multipleRITZ values.
on the RITZ residuals and vectors to understand the sizesof the RITZ vectors.
on the (Q)OR iterates to understand the “delay”.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 169: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/169.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
descriptions
We know that finite precision CG/Lanczos methods
compute clusters of RITZ values resembling (simple)eigenvalues.
tend to show a “delay” in the convergence.
We can use the theorem(s)
on the “basis” vectors to explain the occurrence of multipleRITZ values.
on the RITZ residuals and vectors to understand the sizesof the RITZ vectors.
on the (Q)OR iterates to understand the “delay”.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 170: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/170.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
descriptions
We know that finite precision CG/Lanczos methods
compute clusters of RITZ values resembling (simple)eigenvalues.
tend to show a “delay” in the convergence.
We can use the theorem(s)
on the “basis” vectors to explain the occurrence of multipleRITZ values.
on the RITZ residuals and vectors to understand the sizesof the RITZ vectors.
on the (Q)OR iterates to understand the “delay”.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 171: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/171.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
Outline
1 Getting startedthe name of the gamea few examplesbasic notationsHESSENBERG structure
2 The results ...“basis” transformationseigenvalue problemslinear systems: (Q)ORlinear systems: (Q)MR
3 ... and their impactsgeneral commentsfinite precision issuesinexact KRYLOV methods
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 172: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/172.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
choices
In the inexact methods we have to chose the magnitudes of theerrors fl ≡ ∆lql such that convergence is not spoiled.
Example (inexact (Q)OR, e.g., inexact CG)
We have proven
xk = Lk [z−1](A)r0 +k∑
l=1
zlkLl+1:k [z−1](A)fl . (56)
Based on the behavior of the solution vectors zk and/or theLAGRANGE interpolations we can allow the perturbation vectorsfl to grow (in certain directions).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 173: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/173.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
choices
In the inexact methods we have to chose the magnitudes of theerrors fl ≡ ∆lql such that convergence is not spoiled.
Example (inexact (Q)OR, e.g., inexact CG)
We have proven
xk = Lk [z−1](A)r0 +k∑
l=1
zlkLl+1:k [z−1](A)fl . (56)
Based on the behavior of the solution vectors zk and/or theLAGRANGE interpolations we can allow the perturbation vectorsfl to grow (in certain directions).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 174: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/174.jpg)
Getting startedThe results ...
... and their impactsSummary
general commentsfinite precision issuesinexact KRYLOV methods
choices
In the inexact methods we have to chose the magnitudes of theerrors fl ≡ ∆lql such that convergence is not spoiled.
Example (inexact (Q)OR, e.g., inexact CG)
We have proven
xk = Lk [z−1](A)r0 +k∑
l=1
zlkLl+1:k [z−1](A)fl . (56)
Based on the behavior of the solution vectors zk and/or theLAGRANGE interpolations we can allow the perturbation vectorsfl to grow (in certain directions).
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 175: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/175.jpg)
Getting startedThe results ...
... and their impactsSummary
Summary
Our abstraction
can not be used to directly prove convergence.
does not predict the behavior of the RITZ values.
expresses RITZ vectors and (Q)OR quantities in terms ofthe computed RITZ values.
establishes and promotes a new point of view:
perturbed abstract KRYLOV methodsas additive overlay of
exact abstract KRYLOV methods.
(Q)MR case has to be investigated more thoroughly.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 176: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/176.jpg)
Getting startedThe results ...
... and their impactsSummary
Summary
Our abstraction
can not be used to directly prove convergence.
does not predict the behavior of the RITZ values.
expresses RITZ vectors and (Q)OR quantities in terms ofthe computed RITZ values.
establishes and promotes a new point of view:
perturbed abstract KRYLOV methodsas additive overlay of
exact abstract KRYLOV methods.
(Q)MR case has to be investigated more thoroughly.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 177: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/177.jpg)
Getting startedThe results ...
... and their impactsSummary
Summary
Our abstraction
can not be used to directly prove convergence.
does not predict the behavior of the RITZ values.
expresses RITZ vectors and (Q)OR quantities in terms ofthe computed RITZ values.
establishes and promotes a new point of view:
perturbed abstract KRYLOV methodsas additive overlay of
exact abstract KRYLOV methods.
(Q)MR case has to be investigated more thoroughly.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 178: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/178.jpg)
Getting startedThe results ...
... and their impactsSummary
Summary
Our abstraction
can not be used to directly prove convergence.
does not predict the behavior of the RITZ values.
expresses RITZ vectors and (Q)OR quantities in terms ofthe computed RITZ values.
establishes and promotes a new point of view:
perturbed abstract KRYLOV methodsas additive overlay of
exact abstract KRYLOV methods.
(Q)MR case has to be investigated more thoroughly.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 179: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/179.jpg)
Getting startedThe results ...
... and their impactsSummary
Summary
Our abstraction
can not be used to directly prove convergence.
does not predict the behavior of the RITZ values.
expresses RITZ vectors and (Q)OR quantities in terms ofthe computed RITZ values.
establishes and promotes a new point of view:
perturbed abstract KRYLOV methodsas additive overlay of
exact abstract KRYLOV methods.
(Q)MR case has to be investigated more thoroughly.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 180: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/180.jpg)
Getting startedThe results ...
... and their impactsSummary
Summary
Our abstraction
can not be used to directly prove convergence.
does not predict the behavior of the RITZ values.
expresses RITZ vectors and (Q)OR quantities in terms ofthe computed RITZ values.
establishes and promotes a new point of view:
perturbed abstract KRYLOV methodsas additive overlay of
exact abstract KRYLOV methods.
(Q)MR case has to be investigated more thoroughly.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 181: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/181.jpg)
Getting startedThe results ...
... and their impactsSummary
Summary
Our abstraction
can not be used to directly prove convergence.
does not predict the behavior of the RITZ values.
expresses RITZ vectors and (Q)OR quantities in terms ofthe computed RITZ values.
establishes and promotes a new point of view:
perturbed abstract KRYLOV methodsas additive overlay of
exact abstract KRYLOV methods.
(Q)MR case has to be investigated more thoroughly.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods
![Page 182: Abstract Perturbed Krylov Methods - cs.cas.cz · 08.03.2005 / ICS of CAS / Prague Jens-Peter M. Zemke Abstract Perturbed Krylov Methods. Getting started The results ..... and their](https://reader033.fdocuments.us/reader033/viewer/2022050114/5f4b272966ea1221ff5306a9/html5/thumbnails/182.jpg)
Getting startedThe results ...
... and their impactsSummary
that’s all . . .
Dekuji.
Jens-Peter M. Zemke Abstract Perturbed Krylov Methods