Dynamic Data Analysis: Modeling Data with Differential Equations

ISBN-13: 9781493971886 / Angielski / Twarda / 2017 / 230 str.

Jim Ramsay; Giles Hooker

Recenzje

"This book is intended both for first year graduate students and for researchers in applied mathematics and/or statistics who want to check models with differential equations in data science. These kinds of models have a mechanistic approach, enlarging the classes of models for statisticians, and giving techniques for estimation of parameters, assessing the adequacy of models and planning experiments for applied mathematicians." (Sylvie Viguier-Pla, Mathematical Reviews, August, 2018)

Spis treści

1. Introduction to Dynamic Models

1.1 Six Examples of Input/Output Dynamics

1.1.1 Smallpox in Montreal

1.1.2 Spread of Disease Equations

1.1.3 Filling a Container

1.1.4 Head Impact and Brain Acceleration

1.1.5 Compartment models and pharmacokinetics

1.1.6 Chinese handwriting

1.1.7 Where to go for More Dynamical Systems

1.2 What This Book Undertakes

1.3 Mathematical Requirements

1.4 Overview

 

2 DE notation and types

2.1 Introduction and Chapter Overview

2.2 Notation for Dynamical Systems

2.2.1 Dynamical System Variables

2.2.2 Dynamical System Parameters

2.2.3 Dynamical System Data Configurations

2.2.4 Mathematical Background

2.3 The Architecture of Dynamic Systems

2.4 Types of Differential Equations

2.4.1 Linear Differential Equations

2.4.2 Nonlinear Dynamical Systems

2.4.3 Partial Differential Equations

2.4.4 Algebraic and Other Equations

2.5 Data Configurations

2.5.1 Initial and Boundary Value Configurations

2.5.2 Distributed Data Configurations

2.5.3 Unobserved or Lightly Observed Variables

2.5.4 Observational Data and Measurement Models

2.6 Differential Equation Transformations

2.7 A Notation Glossary

 

3 Linear Differential Equations and Systems

3.1 Introduction and Chapter Overview

3.2 The First Order Stationary Linear Buffer

3.3 The Second Order Stationary Linear Equation

3.4 The mth Order Stationary Linear Buffer

3.5 Systems of Linear Stationary Equations

3.6 A Linear System Example: Feedback Control

3.7 Nonstationary Linear Equations and Systems

3.7.1 The First Order Nonstationary Linear Buffer

3.7.2 First Order Nonstationary Linear Systems

3.8 Linear Differential Equations Corresponding to Sets of Functions

3.9 Green’s Functions for Forcing Function Inputs

 

4 Nonlinear Differential Equations

4.1 Introduction and Chapter Overview

4.2 The Soft Landing Modification

4.3 Existence and Uniqueness Results

4.4 Higher Order Equations

4.5 Input/Output Systems

4.6 Case Studies

4.6.1 Bounded Variation: The Catalytic Equation

4.6.2 Rate Forcing: The SIR Spread of Disease System

4.6.3 From Linear to Nonlinear: The FitzHugh-Nagumo Equations

4.6.4 Nonlinear Mutual Forcing: The Tank Reactor Equations

4.6.5 Modeling Nylon Production

 

5 Numerical Solutions

5.1 Introduction

5.2 Euler Methods

5.3 Runge-KuttaMethods

5.4 Collocation Methods

5.5 Numerical Problems

5.5.1 Stiffness

5.5.2 Discontinuous Inputs

5.5.3 Constraints and Transformations

< 

6 Qualitative Behavior

6.1 Introduction

6.2 Fixed Points

6.2.1 Stability

6.3 Global Analysis and Limit Cycles

6.3.1 Use of Conservation Laws

6.3.2 Bounding Boxes

6.4 Bifurcations

6.4.1 Transcritical Bifurcations

6.4.2 Saddle Node Bifurcations

6.4.3 Pitchfork Bifurcations

6.4.4 Hopf Bifurcations

6.5 Some Other Features

6.5.1 Chaos

6.5.2 Fast-Slow Systems

6.6 Non-autonomous Systems

6.7 Commentary

 

7 Trajectory Matching

7.1 Introduction

7.2 Gauss-Newton Minimization

7.2.1 Sensitivity Equations

7.2.2 Automatic Differentiation

7.3 Inference

7.4 Measurements on Multiple Variables

7.4.1 Multivariate Gauss-Newton Method

7.4.2 VariableWeighting using Error Variance

7.4.3 Estimating s2

7.4.4 Example: FitzHugh-NagumoModels

7.4.5 Practical Problems: Local Minima

7.4.6 Initial Parameter Values for the Chemostat Data

7.4.7 Identifiability

7.5 Bayesian Methods

7.6 Multiple Shooting and Collocation

7.7 Fitting Features

7.8 Applications: Head Impacts

 

8 Gradient Matching

8.1 Introduction

8.2 Smoothing Methods and Basis Expansions

8.3 Fitting the Derivative

8.3.1 Optimizing Integrated Squared Error (ISSE)

8.3.2 Gradient Matching for the Refinery Data

8.3.3 Gradient Matching and the Chemostat Data

8.4 System Mis-specification and Diagnostics

8.4.1 Diagnostic Plots

8.5 Conducting Inference

8.5.1 Nonparametric Smoothing Variances

8.5.2 Example: Refinery Data

8.6 Related Methods and Extensions

8.6.1 Alternative Smoothing Method

8.6.2 Numerical Discretization Methods

8.6.3 Unobserved Covariates

8.6.4 Nonparametric Models

8.6.5 Sparsity and High Dimensional ODEs

8.7 Integral Matching

8.8 Applications: Head Impacts

 

9 Profiling for Linear Systems

9.1 Introduction and Chapter Overview

9.2 Parameter Cascading

9.2.1 Two Classes of Parameters

9.2.2 Defining Coefficients as Functions of Parameters

9.2.3 Data/Equation Symmetry

9.2.4 Inner Optimization Criterion J

9.2.5 The Least Squares Cascade Coefficient Function

9.2.6 The Outer Fitting Criterion H

9.3 Choosing the Smoothing Parameter r

9.4 Confidence Intervals for Parameters

9.4.1 Simulation Sample Results

9.5 Multi–Variable Systems

9.6 Analysis of the Head Impact Data

9.7 A Feedback Model for Driving Speed

9.7.1 Two-Variable First Order Cruise Control Model

9.7.2 One-Variable Second Order Cruise Control Model

9.8 The Dynamics of the Canadian Temperature Data

9.9 Chinese Handwriting

9.10 Complexity Bases

9.11 Software and Computation

9.11.1 Rate Function Specifications

9.11.2 Model Term Specifications

9.11.3 Memoization

 

10 Nonlinear Profiling

10.1 Introduction and Chapter Overview

10.2 Parameter Cascading for Nonlinear Systems

10.2.1 The Setup for Parameter Cascading

10.2.2 Parameter Cascading Computations

10.2.3 Some Helpful Tips

10.2.4 Nonlinear Systems and Other Fitting Criteria

10.3 Lotka-Volterra

10.4 Head Impact

10.5 Compound Model for Blood Ethanol

10.6 Catalytic model for growth

10.7 Aromate Reactions

References

Glossary

Index

O autorze

Jim Ramsay, PhD, is Professor Emeritus of Psychology and an Associate Member in the Department of Mathematics and Statistics at McGill University. He received his PhD from Princeton University in 1966 in quantitative psychology. He has been President of the Psychometric Society and the Statistical Society of Canada. He received the Gold Medal in 1998 for his contributions to psychometrics and functional data analysis and Honorary Membership in 2012 from the Statistical Society of Canada.

Giles Hooker, PhD, is Associate Professor of Biological Statistics and Computational Biology at Cornell University. In addition to differential equation models, he has published extensively on functional data analysis and uncertainty quantification in machine learning. Much of his methodological work is inspired by collaborations in ecology and citizen science data.

Blurb

This text focuses on the use of smoothing methods for developing and estimating differential equations following recent developments in functional data analysis and building on techniques described in Ramsay and Silverman (2005) Functional Data Analysis. The central concept of a dynamical system as a buffer that translates sudden changes in input into smooth controlled output responses has led to applications of previously analyzed data, opening up entirely new opportunities for dynamical systems. The technical level has been kept low so that those with little or no exposure to differential equations as modeling objects can be brought into this data analysis landscape. There are already many texts on the mathematical properties of ordinary differential equations, or dynamic models, and there is a large literature distributed over many fields on models for real world processes consisting of differential equations. However, a researcher interested in fitting such a model to data, or a statistician interested in the properties of differential equations estimated from data will find rather less to work with. This book fills that gap. 

• Offers an accessible text to those with little or no exposure to differential equations as modeling objects 
  
• Updates and builds on techniques from the popular Functional Data Analysis (Ramsay and Silverman, 2005)
  
• Opens up new opportunities for dynamical systems and presents additional applications for previously analyzed data

Jim Ramsay, PhD, is Professor Emeritus of Psychology and an Associate Member in the Department of Mathematics and Statistics at McGill University. He received his PhD from Princeton University in 1966 in quantitative psychology. He has been President of the Psychometric Society and the Statistical Society of Canada. He received the Gold Medal in 1998 for his contributions to psychometrics and functional data analysis and Honorary Membership in 2012 from the Statistical Society of Canada.

Giles Hooker, PhD, is Associate Professor of Biological Statistics and Computational Biology at Cornell University. In addition to differential equation models, he has published extensively on functional data analysis and uncertainty quantification in machine learning. Much of his methodological work is inspired by collaborations in ecology and citizen science data.