Mathematische Methoden der Unsicherheitsquantifizierung

Similar documents
4. Solvability of elliptic PDEs

NOTES ON SOME EXERCISES OF LECTURE 5, MODULE 2

REAL ANALYSIS II TAKE HOME EXAM. T. Tao s Lecture Notes Set 5

Markov Chains. Andreas Klappenecker by Andreas Klappenecker. All rights reserved. Texas A&M University

Parallel Domain Decomposition Strategies for Stochastic Elliptic Equations Part A: Local KL Representations

Variational inequality formulation of chance-constrained games

Mariusz Jurkiewicz, Bogdan Przeradzki EXISTENCE OF SOLUTIONS FOR HIGHER ORDER BVP WITH PARAMETERS VIA CRITICAL POINT THEORY

Basic Operator Theory with KL Expansion

Approximation in the Zygmund Class

Singular integral operators and the Riesz transform

Functional Analysis Review

arxiv: v2 [math.ca] 13 May 2015

Faster Kernel Ridge Regression Using Sketching and Preconditioning

On principal eigenpair of temporal-joined adjacency matrix for spreading phenomenon Abstract. Keywords: 1 Introduction

CHAPTER VIII HILBERT SPACES

arxiv: v1 [math.ca] 4 Apr 2017

Mathematische Methoden der Unsicherheitsquantifizierung

On compact operators

arxiv: v2 [math.pr] 27 Oct 2015

REAL ANALYSIS II HOMEWORK 3. Conway, Page 49

Analysis Preliminary Exam Workshop: Hilbert Spaces

Approximation of Weighted Local Mean Operators

a P (A) f k(x) = A k g k " g k (x) = ( 1) k x ą k. $ & g k (x) = x k (0, 1) f k, f, g : [0, 8) Ñ R f k (x) ď g(x) k P N x P [0, 8) g(x)dx g(x)dx ă 8

FMM Preconditioner for Radiative Transport

C.6 Adjoints for Operators on Hilbert Spaces

arxiv: v1 [math.fa] 17 Feb 2014

Ground States are generically a periodic orbit

DS-GA 1002: PREREQUISITES REVIEW SOLUTIONS VLADIMIR KOBZAR

A revealed preference test for weakly separable preferences

Karhunen-Loève Approximation of Random Fields Using Hierarchical Matrix Techniques

08a. Operators on Hilbert spaces. 1. Boundedness, continuity, operator norms

Math 246B - Partial Differential Equations

Random Variables. Andreas Klappenecker. Texas A&M University

Section 3.9. Matrix Norm

J. M. ALMIRA, L. SZÉKELYHIDI

Electronic Companion Dynamic Pricing of Perishable Assets under Competition

SOLUTIONS Math B4900 Homework 9 4/18/2018

Karhunen-Loève decomposition of Gaussian measures on Banach spaces

DS-GA 3001: PROBLEM SESSION 2 SOLUTIONS VLADIMIR KOBZAR

NOTES WEEK 04 DAY 1 SCOT ADAMS

NOTES WEEK 15 DAY 1 SCOT ADAMS

MATH 387 ASSIGNMENT 2

NONLINEAR EVOLUTION EQUATIONS SHORT NOTES WEEK #1

LECTURE 16: UNITARY REPRESENTATIONS LECTURE BY SHEELA DEVADAS STANFORD NUMBER THEORY LEARNING SEMINAR FEBRUARY 14, 2018 NOTES BY DAN DORE

NOTES WEEK 13 DAY 2 SCOT ADAMS

Cartan subalgebras in uniform Roe algebras

Lectures 9 & 10: Fredholm operators

Geometry on Probability Spaces

FURSTENBERG S THEOREM ON PRODUCTS OF I.I.D. 2 ˆ 2 MATRICES

NOTES WEEK 11 DAY 2 SCOT ADAMS

A Concise Course on Stochastic Partial Differential Equations

Lecture 15: Quadratic approximation and the second variation formula

Homework for MATH 4604 (Advanced Calculus II) Spring 2017

Review and problem list for Applied Math I

Minimizing the Number of Function Evaluations to Estimate Sobol Indices Using Quasi-Monte Carlo

Theorem The simple finite dimensional sl 2 modules Lpdq are indexed by

ADVANCE TOPICS IN ANALYSIS - REAL. 8 September September 2011

PART IV Spectral Methods

AN ELEMENTARY PROOF OF THE SPECTRAL RADIUS FORMULA FOR MATRICES

Chapter 8 Integral Operators

Entropy and Ergodic Theory Lecture 28: Sinai s and Ornstein s theorems, II

Theory of Positive Definite Kernel and Reproducing Kernel Hilbert Space

The spectral zeta function

Partial Differential Equations and Semigroups of Bounded Linear Operators

Extensive Form Abstract Economies and Generalized Perfect Recall

The Riemann Roch theorem for metric graphs

Math Solutions to homework 5

Entropy and Ergodic Theory Lecture 27: Sinai s factor theorem

Review of Linear Algebra

Some Thoughts on Guaranteed Function Approximation Satisfying Relative Error

LECTURE 7. k=1 (, v k)u k. Moreover r

Solutions of exercise sheet 3

Convergence of Eigenspaces in Kernel Principal Component Analysis

Implementation of Sparse Wavelet-Galerkin FEM for Stochastic PDEs

Tempered Distributions

MATH34032 Mid-term Test 10.00am 10.50am, 26th March 2010 Answer all six question [20% of the total mark for this course]

EXAMINATION SOLUTIONS

5 Compact linear operators

Linear Algebra and Dirac Notation, Pt. 2

Some Concepts of Uniform Exponential Dichotomy for Skew-Evolution Semiflows in Banach Spaces

On Borel maps, calibrated σ-ideals and homogeneity

Karhunen-Loève decomposition of Gaussian measures on Banach spaces

S. K. Mohanta, Srikanta Mohanta SOME FIXED POINT RESULTS FOR MAPPINGS IN G-METRIC SPACES

HILBERT-SCHMIDT AND TRACE CLASS OPERATORS. 1. Introduction

BOUNDARY VALUE PROBLEMS ON A HALF SIERPINSKI GASKET

Last time: Recall that the fibers of a map ϕ : X Ñ Y are the sets in ϕ 1 pyq Ď X which all map to the same element y P Y.

Basic Calculus Review

New Developments in LS-OPT - Robustness Studies

Determinant lines and determinant line bundles

Your first day at work MATH 806 (Fall 2015)

Entropy and Ergodic Theory Notes 22: The Kolmogorov Sinai entropy of a measure-preserving system

Monte Carlo Methods for Uncertainty Quantification

Self-Adjointness of the Atomic Hamiltonian Operator

Singular Value Decomposition and its. SVD and its Applications in Computer Vision

Sparse polynomial chaos expansions in engineering applications

Solution 4. hence the Cauchy problem has a differentiable solution for all positive time y ą 0. b) Let hpsq ups, 0q, since hpsq is decreasing in r π

APPROXIMATE HOMOMORPHISMS BETWEEN THE BOOLEAN CUBE AND GROUPS OF PRIME ORDER

The following definition is fundamental.

Background on Metric Geometry, I. Facundo Mémoli Math and CSE Departments, OSU.

Cartan subalgebras in uniform Roe algebras

Transcription:

Mathematische Methoden der Unsicherheitsquantifizierung Oliver Ernst Professur Numerische Mathematik Sommersemester 2014

Contents 1 Introduction 1.1 What is Uncertainty Quantification? 1.2 A Case Study: Radioactive Waste Disposal 2 Monte Carlo Methods 2.1 Introduction 2.2 Basic Monte Carlo Simulation 2.3 Improving the Monte Carlo Method 2.4 Multilevel Monte Carlo Estimators 2.5 The Monte Carlo Finite Element Method 3 Random Fields 3.1 Introduction 3.2 Karhunen-Loève Expansion 3.3 Regularity of Random Fields 3.4 Covariance Eigenvalue Decay 4 Stochastic Collocation 4.1 Introduction 4.2 Collocation 4.3 Analytic Parameter Dependence 4.4 Convergence Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 6 / 315

Contents 5 Probability Theory 6 Elliptic Boundary Value Problems 7 Collection of Results from Functional Analysis 8 Miscellanea Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 296 / 315

Contents 5 Probability Theory 6 Elliptic Boundary Value Problems 7 Collection of Results from Functional Analysis 7.1 Hilbert-Schmidt Operators 8 Miscellanea Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 297 / 315

For normed linear spaces X and Y, we denote by L px, Y q the set of all bounded linear operators from X to Y. Definition C.1 Let X and Y be separable Hilbert spaces with norms } } X and } } Y and let tx j u npn denote a CONS of X. A linear operator L : X Ñ Y for which 8ÿ 1{2 }L} HSpX,Y q : }Lx j } 2 Y ă 8 is called a Hilbert-Schmidt operator. We shall write }L} HS if X Y. Proposition C.2 j 1 The mapping } } HSpX,Y q is a norm, called the Hilbert-Schmidt norm, on the space of all Hilbert-Schmidt operators from X to Y, which we denote by HSpX, Y q. In addition, phspx, Y q, } } HSpX,Y q q is Banach space. Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 298 / 315

Examples Example C.3 For X Y R n with the Euclidean norm } }, the Hilbert-Schmidt norm of a matrix A P R nˆn coincides with the Frobenius-norm }A} 2 F ř n i,j 1 a2 i,j. Example C.4 Define L P L pl 2 p0, 1qq by pluqpxq ż x 0 upyq dy, u P L 2 p0, 1q, x P p0, 1q. For the CONS tf j pxq? 2 sinpjπxq : j P Nu, we have? 2 plf j qpxq p1 cospjπxqq. jπ L is a Hilbert-Schmidt operator since }Lf j } L 2 p0,1q ď 2? 2 jπ. Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 299 / 315

Integral operators Lemma C.5 Let H be a separable Hilbert space. If L P HSpHq, then }L} L phq ď }L} HS. In particular, Hilbert-Schmidt operators are bounded. Definition C.6 For a domain D Ă R d and k P L 2 pd ˆ Dq, the integral operator with kernel function k is defined as the linear operator ż K : u ÞÑ pkuqpx q : kpx, yqupyq dy, x P D. (C.1) Theorem C.7 D An integral operator with kernel function k P L 2 pd ˆ Dq is a Hilbert-Schmidt operator on L 2 pdq. Conversely, any Hilbert-Schmidt operator K on L 2 pdq can be written in the form (C.1) with }K} HS }k} L 2 pdˆdq. Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 300 / 315

Compact operators Definition C.8 A set B in a Banach space X is said to be compact if every sequence u n Ă B has a convergent subsequence u nk with limit u P B. Definition C.9 A linear operator L : X Ñ Y, where X and Y are Banach spaces, is said to be compact if the image of any bounded set B Ă X has compact closure in Y, i.e., if LpBq } } Y Theorem C.10 is a compact set in Y for all bounded B Ă X. For k P L 2 pd ˆ Dq the associated integral operator K on L 2 pdq with kernel function k is a compact operator. Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 301 / 315

Selfadjoint operators, eigenvalues Definition C.11 An operator L P L phq on a Hilbert space H is said to be selfadjoint if plu, vq pu, Lvq @u, v P H. Proposition C.12 For a domain D Ă R d, if k P L 2 pd ˆ Dq is symmetric, i.e., kpx, yq kpy, x q for all x, y P D, then the integral operator with kernel function k is selfadjoint with respect to the L 2 pdq inner product. Definition C.13 If L P L phq, λ P C is called an eigenvalue of L if there exists nonzero φ P H such that Lφ λφ. The element φ is called an eigenvector or eigenfunction of L. Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 302 / 315

Spectral theorem Theorem C.14 (Spectral theorem for selfadjoint compact operators) Let H be a separable Hilbert space and K Ă L phq be selfadjoint and compact. Denote the eigenvalues of K by tλ j u jpn ordered such that λ j`1 ď λ j and denote the associated eigenfunctions by tφ j u. Then (i) All eigenvalues are real and λ j Ñ 0 as j Ñ 8. (ii) The sequence tφ j u can be chosen as a CONS of the range KpHq of K and, (iii) for any u P H, Ku 8ÿ λ k pu, φ j qφ j. j 1 (C.2) Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 303 / 315

Nonnegative functions, operators Definition C.15 A function k : D ˆ D Ñ R is nonnegative definite if for any set of points x 1,..., x n P D and numbers a 1,..., a n P R there holds nÿ j,k 1 a j a k kpx j, x k q ě 0. A linear operator L P L phq on a Hilbert space H is nonnegative definite if pu, Luq ě 0 @u P H and positive definite if pu, Luq ą 0 @u P H. Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 304 / 315

Nonnegative functions, operators, trace class operators Lemma C.16 For a domain D Ă R d and a nonnegative definite function k P CpD ˆ Dq, the integral operator K on L 2 pdq with kernel function k is nonnegative. Lemma C.17 (Dini) For a bounded domain D let f n P CpDq be such that f n px q ď f n`1 px q for n P N and f n px q Ñ fpx q as n Ñ 8 for all x P D. Then }f f n } 8 Ñ 0 as n Ñ 8. Definition C.18 Let H be a separable Hilbert space. A nonnegative definite operator L P L phq is said to be of trace class if traceplq ă 8, where the trace of L is defined as 8ÿ traceplq : plψ j, ψ j q for any CONS tψ j u jpn of H. j 1 Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 305 / 315

Mercer s theorem Theorem C.19 (Mercer) For a bounded domain D, let k P CpD ˆ Dq be a symmetric and nonnegative definite function and let K be the integral operator with kernel function k. There exist eigenfunctions φ j of K with eigenvalues λ j ą 0 such that φ j P CpDq and 8ÿ kpx, yq λ j φ j px qφ j pyq, x, y P D, j 1 where the series converges in CpD ˆ Dq. Furthermore, n sup ˇ ˇkpx, yq ÿ 8ÿ λ j φ j px qφ j pyq ˇ ď sup x,ypd j 1 The operator K is of trace class and ż tracepkq D x PD j n`1 kpx, x q dx. λ j φ j px q 2. (C.3) Oliver Ernst (Numerische Mathematik) UQ Sommersemester 2014 306 / 315