Recent Advances in Algorithmic Differentiation

Name: Recent Advances in Algorithmic Differentiation
Brand: Springer
Price: 96.29 EUR
Availability: OnlineOnly

Shaun Forth Paul Hovland Eric Phipps Jean Utke Andrea Walther(Editor)

Springer (Publisher)

1st Edition

Published on 30. July 2012

XVIII, 362 pages

E-Book

PDF with digital watermarking

System requirements

978-3-642-30023-3 (ISBN)

€96.29incl. 7% vat

System requirements

for PDF with digital watermarking

E-Book Single Licence

Available for download

Description

More details

Other editions

Content

Intro
Recent Advances in Algorithmic Differentiation
Preface
Contents
Contributors
A Leibniz Notation for Automatic Differentiation
1 Historical Background
2 The Leibniz Notation
2.1 The Forward Mode
2.2 The Reverse Mode
2.3 Forward over Forward
2.4 Forward over Reverse
3 Second Order Approaches Involving Reverse Mode
3.1 Forward over Reverse
3.2 Reverse over Forward
3.3 Reverse over Reverse
4 The Equivalence Theorem
References
Sparse Jacobian Construction for Mapped Grid Visco-Resistive Magnetohydrodynamics
1 Introduction
1.1 Model
1.2 Implicit Solver Framework
2 Preconditioner Construction
2.1 Code Reconfiguration
2.2 OpenAD Usage
2.3 Code Integration
3 Results
3.1 Conclusions and Future Work
References
Combining Automatic Differentiation Methods for High-Dimensional Nonlinear Models
1 Introduction
2 Methodology
3 Implementation
3.1 Gradients with OpenAD
3.2 Higher-Order Derivatives with Rapsodia
4 Test Cases
References
Application of Automatic Differentiation to an Incompressible URANS Solver
1 Introduction
2 Incompressible URANS Equations and Flow Solver
3 Generation of a Discrete Adjoint Solver
3.1 Reversal of the Time Loop
3.2 Adjoining the Outer Iterations
3.3 Parallelization of the Adjoint Solver
4 Validation of the Adjoint Solver
References
Applying Automatic Differentiation to the Community Land Model
1 Introduction
2 Background
3 Automatic Differentiation and OpenAD
4 AD Development Process
4.1 Code Comprehension
4.2 Preprocessing
4.3 Transformation
4.4 Postprocessing
5 Results
6 Conclusion
References
Using Automatic Differentiation to Study the Sensitivity of a Crop Model
1 The Application Domain: The Agronomic Crop Model STICS
2 Sensitivity Analysis
3 Automatic Differentiation of STICS
3.1 The TAPENADE Automatic Differentiation Tool
3.2 STICS Adjoint : The Pains and Sufferings of an AD End-User
3.3 Validation of the Adjoint Model
4 Results: Sensitivity Analysis of STICS
4.1 Selection of Input Parameters for Sensitivity Analysis of Output Variables
4.2 Sensitivity Results of LAI and Biomass
5 Conclusion and Outlook
References
Efficient Automatic Differentiation of Matrix Functions
1 Introduction
1.1 Terminology
1.2 Matrix Derivatives in Matrix Form
2 Kronecker Products
3 Box Products
3.1 Box Products of Identity Matrices
4 Differentiation Rules
4.1 A Simple Example
4.2 The Hessian and Newton's Method
4.3 An Example Taylor Series
5 Future Work
References
Native Handling of Message-Passing Communication in Data-Flow Analysis
1 Introduction
2 Context-Sensitive and Flow-Sensitive Data-Flow Analysis
3 Impact of Message-Passing on Data-Flow Analysis
4 Data-Flow Analysis with Flow Graph Local Restart
5 Performance Discussion
6 Choosing a Good Set of Channels
7 Implementation and Outlook
References
Increasing Memory Locality by Executing Several Model Instances Simultaneously
1 Introduction
2 Cloning
3 Applications
4 Conclusions
References
Adjoint Mode Computation of Subgradients for McCormickRelaxations
1 Introduction
2 Tangent-Linear Mode Computation of Subgradients of McCormick Relaxations
3 Adjoint Mode Computation of Subgradients of McCormick Relaxations
4 Numerical Results
5 Conclusion and Outlook
References
Evaluating an Element of the Clarke Generalized Jacobian of a Piecewise Differentiable Function
1 Introduction
2 Mathematical Background
2.1 Piecewise Differentiable Functions
2.2 Automatic Differentiation
3 Generalized Jacobian Element Evaluation
4 Concluding Remarks
References
The Impact of Dynamic Data Reshaping on Adjoint Code Generation for Weakly-Typed Languages Such as Matlab
1 Introduction
2 Implicit Scalar Expansion and Array Reshape
2.1 Binary Component-Wise Operators
2.2 General Expressions
2.3 Matrix Multiplication
2.4 One-to-Many Index Assignment and Implicit Array Reshape
3 Performance of the Adjred Mechanism
4 Implicit Array Enlargement
5 Performance of the push_index/pop_index Mechanism
6 Conclusion
References
On the Efficient Computation of Sparsity Patterns for Hessians
1 Introduction
2 Algorithms for Exact and Overestimated Sparsity Patterns
3 New Data Structures and Corresponding Set Operations
4 Numerical Examples
5 Conclusions and Outlook
References
Exploiting Sparsity in Automatic Differentiation on Multicore Architectures
1 Introduction
2 Parallel Derivative Computation in ADOL-C
3 Parallel Sparse Derivative Computation
3.1 Sparsity Pattern Detection
3.2 Seed Matrix Determination
3.3 Derivative Calculation
4 Experimental Results
4.1 Test Cases
4.2 Runtime Results
5 Conclusion
References
Automatic Differentiation Through the Use of Hyper-Dual Numbers for Second Derivatives
1 Introduction
2 First Derivative Calculations
3 Second Derivative Calculations
4 Hyper-Dual Number Implementation and Results
4.1 Application to a Computational Fluid Dynamics Code
4.2 Computational Cost
5 Conclusion
References
Connections Between Power Series Methods and Automatic Differentiation
1 Introduction
2 Applying PSM
2.1 Example 1: PSM and AD Applied to an IVODE
2.2 Example 2: Root-Finding
2.3 Example 3: The Maclaurin Polynomial for an Inverse Function
3 PSM Theory and AD
4 Conclusion
References
Hierarchical Algorithmic Differentiation A Case Study
1 Introduction
2 Problem Description
3 Structure of JURASSIC2
4 Performance Analysis
5 Numerical Results
6 Summary and Conclusion
References
Storing Versus Recomputation on Multiple DAGs
1 Introduction
2 A Use Case for Storing Edge Labels
3 Computational Graphs Share Value Restoration
4 Problem Formulation
4.1 A Cost Function
4.2 A Search Strategy
4.3 Algorithm
5 Observations and Summary
References
Using Directed Edge Separators to Increase Efficiency in the Determination of Jacobian Matrices via Automatic Differentiation
1 Introduction
1.1 Automatic Differentiation and the Edge Separator
2 On Finding Separators to Increase Efficiency in the Application of Automatic Differentiation
2.1 Minimum Weighted Separator
2.2 Natural Order Edge Separator
2.3 Multiple Separators
3 Experiments
3.1 Thin Computational Graphs
3.2 Fat Computational Graphs
4 Accelerating the Calculation of the Jacobian matrix
5 Concluding Remarks
References
An Integer Programming Approach to Optimal DerivativeAccumulation
1 Introduction
2 Integer Programming Formulations
2.1 IP Formulation
2.2 Computational Consideration for Solving (MinFlops)
3 Computational Experiments
4 Conclusion
References
The Relative Cost of Function and Derivative Evaluations in the CUTEr Test Set
1 Introduction
2 Numerical Results
2.1 Problem Dimension and Ratio
2.2 Run-Time and Run-Time Ratios
2.3 Sparsity
3 Conclusion
References
Java Automatic Differentiation Tool Using Virtual OperatorOverloading
1 Introduction
1.1 Java AD Tool
1.2 Requirements for System Design Optimization
2 Virtual Operator Overloading
2.1 Forward Model Principle
2.2 Using Jacobian and Partial Derivatives
2.3 Management of External Functions
3 JAP Architecture
3.1 Lexical and Syntax Analysis
3.2 Source Class Decomposition
3.3 Adding Forward Propagation Method
3.4 JAP Limits
4 JAP Performance
5 Conclusions and Perspectives
References
High-Order Uncertainty Propagation Enabled by Computational Differentiation
1 Introduction
2 Computational Differentiation
3 State Transition Tensor Models
4 Probability Distribution Function
4.1 Approximate Solutions to the Liouville's Equation
4.2 Data Structures for Numerical Integration
5 Applications
5.1 Unforced Duffing Oscillator
5.2 Two-Body Problem
6 Summary and Conclusion
References
Generative Programming for Automatic Differentiation
1 Introduction
2 Related Work
3 Implementation
3.1 Implementation Details
3.1.1 Expression Tree
3.1.2 Code Generation
4 Experimental Results
5 Conclusion and Future Work
References
AD in Fortran: Implementation via Prepreprocessor
1 Introduction
2 Language Extensions
2.1 Extension 1: AD Syntax
2.2 Extension 2: Nested Subprograms
2.3 Concrete Example
3 Implementation
3.1 Nested Subprograms in Detail
3.2 AD Syntax in Detail
3.3 Performance
4 Conclusion
References
An AD-Enabled Optimization ToolBox in LabVIEWT
1 Introduction
2 Implementation of AD in LabVIEW: The LVAD Package
2.1 LVAD Class
2.2 Operator Overloading
2.3 Examples
3 Implementation of a LabVIEW Optimization Toolbox
3.1 Root Finding: The fzero subVI
3.2 Minimization: The fmin subVI
4 Results
4.1 Simple Example
4.2 Performance Testing
5 Conclusions
References
CasADi: A Symbolic Package for Automatic Differentiation and Optimal Control
1 Introduction and Motivation
2 Background: Eight Flavors of Automatic Differentiation
2.1 Operator Overloading versus Source Code Transformation
2.2 Scalar versus Matrix-Valued Atomic Operations
3 Structure of CasADi
3.1 Two Different Expression Graph Representations
3.1.1 Scalar-Valued Atomic Operations
3.1.2 Sparse, Matrix-Valued Atomic Operations
3.2 Three Different Symbolic Types
3.3 Function Objects
3.4 Sparsity Pattern Detection and Graph Coloring for Jacobians
3.5 External Solver Interfaces
3.6 Python and Octave Front-Ends
4 Numerical Results
5 Conclusion
References
Efficient Expression Templates for Operator Overloading-Based Automatic Differentiation
1 Introduction
2 Expression Templates for Forward Mode AD
3 Improving Performance of Expression Templates
4 Application to Differentiation of a Fluid Dynamics Simulation
5 Concluding Remarks
References
Computing Derivatives in a Meshless Simulation Using Permutations in ADOL-C
1 Introduction
2 Finite Pointset Method
3 Permutations and Selections
3.1 Active rvalue and lvalue Subscripting
3.2 Permutations in Finite Pointset Method
4 Numerical Simulation with Finite Pointset Method
5 Other Applications
6 Conclusions
References
Lazy K-Way Linear Combination Kernels for Efficient Runtime Sparse Jacobian Matrix Evaluations in C++
1 Introduction
1.1 This Work
2 Lazy Evaluation Techniques for Dynamic Sparsity
3 K-Way SPLC Evaluation Kernels
3.1 K-Way SPLC Kernel 1: Caching Nodal Binary Tree
3.2 K-Way SPLC Kernel 2: Prolong-and-Restrict Dense Accumulation
3.3 Summary
4 Computational Examples
4.1 Model SPLC Numerical Experiments
4.2 Model Problem Simulation Jacobian
5 Summary
References
Implementation of Partial Separability in a Source-to-Source Transformation AD Tool
1 Introduction
1.1 ADIC2
1.2 ColPack
2 Related Work
3 Implementation
4 Experimental Results
5 Conclusion
References
Editorial Policy
Lecture Notesin Computational Science and Engineering
Monographs in Computational Science and Engineering
Texts in Computational Science and Engineering

Content (PDF)

System requirements

Save as PDF Copy link into clipboard

Schweitzer Fachinformationen

Recent Advances in Algorithmic Differentiation

Description

More details

Other editions

Additional editions

Content

System requirements