A pragmatic introduction to the finite element

method for thermal and stress analysis with the

matlab toolkit SOFEA Petr Krysl pdf download

https://ebookgate.com/product/a-pragmatic-introduction-to-the-
finite-element-method-for-thermal-and-stress-analysis-with-the-
matlab-toolkit-sofea-petr-krysl/

Get Instant Ebook Downloads – Browse at https://ebookgate.com

 Get Your Digital Files Instantly: PDF, ePub, MOBI and More
Quick Digital Downloads: PDF, ePub, MOBI and Other Formats

The Finite Element Method An Introduction with Partial

Differential Equations 2nd Edition A. J. Davies

https://ebookgate.com/product/the-finite-element-method-an-
introduction-with-partial-differential-equations-2nd-edition-a-j-
davies/

An Introduction to Nonlinear Finite Element Analysis J.

N. Reddy

https://ebookgate.com/product/an-introduction-to-nonlinear-
finite-element-analysis-j-n-reddy/

The Finite Difference Time Domain Method For

Electromagnetics with MATLAB Simulations 2nd Edition
Atef Z. Elsherbeni

https://ebookgate.com/product/the-finite-difference-time-domain-
method-for-electromagnetics-with-matlab-simulations-2nd-edition-
atef-z-elsherbeni/

The Finite Element Method for Three Dimensional

Thermomechanical Applications 1st Edition Guido Dhondt

https://ebookgate.com/product/the-finite-element-method-for-
three-dimensional-thermomechanical-applications-1st-edition-
guido-dhondt/
The Scaled Boundary Finite Element Method 1st Edition
John P. Wolf

https://ebookgate.com/product/the-scaled-boundary-finite-element-
method-1st-edition-john-p-wolf/

Hydrothermal Analysis in Engineering Using Control

Volume Finite Element Method 1st Edition Mohsen
Sheikholeslami

https://ebookgate.com/product/hydrothermal-analysis-in-
engineering-using-control-volume-finite-element-method-1st-
edition-mohsen-sheikholeslami/

The Boundary Element Method for Plate Analysis 1st

Edition John T. Katsikadelis

https://ebookgate.com/product/the-boundary-element-method-for-
plate-analysis-1st-edition-john-t-katsikadelis/

Nonlinear Solid Mechanics for Finite Element Analysis

Statics 1st Edition Javier Bonet

https://ebookgate.com/product/nonlinear-solid-mechanics-for-
finite-element-analysis-statics-1st-edition-javier-bonet/

Finite Element Analysis New Trends and Developments 2nd

Edition Farzad Ebrahimi

https://ebookgate.com/product/finite-element-analysis-new-trends-
and-developments-2nd-edition-farzad-ebrahimi/
Petr Krysl

Thermal and Stress Analysis

with the
Finite Element Method

Accompanied by the MATLABr toolbox FAESOR

December 13, 2010

Pressure Cooker Press

San Diego
c 2010 Petr Krysl
Contents

1 Model of a Taut Wire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Deriving the PDE model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Balance equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Boundary conditions (in space) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Exercise 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Exercise 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Initial conditions (boundary conditions in time) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Initial Boundary Value Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Exercise 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Exercise 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Exercise 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 The Method of Galerkin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.1 Residual of the balance equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Integral test of the residual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Test function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Trial function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Shifting derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6 Essential boundary condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.7 Natural boundary condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.8 Stiffness matrix and load vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Exercise 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Exercise 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Exercise 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Exercise 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Exercise 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.9 Piecewise linear basis functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Exercise 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Exercise 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Exercise 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Exercise 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.10 Bookkeeping in the finite element method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.11 Finite element Galerkin method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.12 Element-by-element computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.12.1 Elementwise quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.13 Prescribed displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Exercise 15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4 Contents

2.14 Partitioned form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.14.1 Derivation of the partitioned form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.15 Principle of superposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3 Taut wire dynamics with the Galerkin method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.1 Residual of the balance equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Integral test of the residual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3 Weighted residual manipulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.4 Mass matrix and load vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.5 Elementwise mass matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.6 Initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Exercise 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.7 Free vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Exercise 17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4 Further refinements of the Galerkin finite element method . . . . . . . . . . . . . . . . . . . 63

4.1 Numerical quadrature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Exercise 18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Exercise 19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Exercise 20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2 Gauss quadrature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Exercise 21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 Derivatives of basis functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Exercise 22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5 More about Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5.1 Mixed essential and natural boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2 Essential boundary conditions only . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.3 Natural boundary conditions only . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4 Concentrated forces in the interior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Exercise 23 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.5 Elementwise stiffness matrix properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.6 Removing rigid body modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.6.1 Adding pin support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.6.2 Adding spring support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.7 Using springs to enforce essential boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Exercise 24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

6 Statics and Dynamics of Taut Wire with the FEM toolbox . . . . . . . . . . . . . . . . . . . 91

6.1 Statics: uniform load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Sparse matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.3 Free vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Exercise 25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Exercise 26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.4 Integration of transient motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.4.1 Using built-in Matlab solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.4.2 Using the Trapezoidal integrator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

7 Model of Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

7.1 Balance equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.2 Constitutive equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.3 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.3.1 On the sufficiency of boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.4 Example of Boundary Condition formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
 Contents 5

7.5 Initial condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

7.6 Summary of the PDE model of heat conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.7 Parallels between the taut wire and the heat conduction model . . . . . . . . . . . . . . . . . . . 113

8 Galerkin Method for the Model of Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . 117

8.1 Weighted residual formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.2 One-dimensional heat conduction model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.3 Comparison with the prestressed wire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.4 Heat conduction 1D FEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Exercise 27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Exercise 28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Exercise 29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.5 Reducing the model dimension to two . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.6 Test and trial functions: basis functions on triangulations . . . . . . . . . . . . . . . . . . . . . . . . 128
8.7 Basis functions on the standard triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Exercise 30 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
8.8 Direct construction of the T3 basis functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Exercise 31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
Exercise 32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
8.9 Discretizing the weighted residual equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.10 Derivatives of the basis functions; Jacobian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.11 Numerical integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Exercise 33 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
Exercise 34 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Exercise 35 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Exercise 36 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.12 Conductivity matrix and heat loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Exercise 37 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Exercise 38 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
Exercise 39 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
Exercise 40 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.13 Surface heat transfer matrix and load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Exercise 41 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Exercise 42 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Exercise 43 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Exercise 44 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Exercise 45 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Exercise 46 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

9 Steady-state Heat Conduction Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

9.1 Steady-state heat conduction equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
9.2 Thick-walled tube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
9.3 Orthotropic insert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
9.4 The T4 NAFEMS Benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Exercise 47 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Exercise 48 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

10 Transient Heat Conduction Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

10.1 Discretization in time for transient heat conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
10.2 The T3 NAFEMS Benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
10.3 Transient cooling in a shrink-fitting application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Exercise 49 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
6 Contents

11 Expanding the Library of Element Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

11.1 Quadratic triangle T6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Exercise 50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
11.2 Quadratic 1-D element L3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Exercise 51 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Exercise 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Exercise 53 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Exercise 54 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
11.3 Point element P1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
11.4 Integrating over m-dimensional domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
11.5 Tetrahedron T4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
11.6 Simplex elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
Exercise 55 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
Exercise 56 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
11.7 Quadrilateral Q4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
11.8 Hexahedron H8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Exercise 57 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Exercise 58 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Exercise 59 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
Exercise 60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
11.9 Extracting the mesh boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

12 Discretization Error, Error Control, and Convergence . . . . . . . . . . . . . . . . . . . . . . . . 225

12.1 Motivating example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Exercise 61 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Exercise 62 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
12.2 Interpolation errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
12.2.1 Interpolation error for temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
12.2.2 Interpolation error for temperature gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
12.2.3 Controlling the error; Convergence rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
12.3 Richardson extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Exercise 63 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Exercise 64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
12.4 The T4 NAFEMS Benchmark revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
12.5 Graded meshes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
12.6 Shrink fitting revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
12.7 Representing functions by interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Exercise 65 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
Exercise 66 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

13 Model of Elastodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

13.1 Balance of linear momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
13.2 Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Exercise 67 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
Exercise 68 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
13.2.1 Balance of angular momentum and stress symmetry. . . . . . . . . . . . . . . . . . . . . . . 259
Exercise 69 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Exercise 70 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
Exercise 71 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
13.3 Local equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
13.3.1 Change of linear momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
13.3.2 Stress divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
13.3.3 All together now . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
 Contents 7

13.4 Strains and displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265

13.5 Constitutive equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
13.6 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
13.6.1 Example: concrete dam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
13.6.2 Example: rigid punch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
13.6.3 Formal definition of the boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
13.6.4 Inadmissible “concentrated” boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . 270
13.6.5 Symmetry and anti-symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
13.6.6 Example: a pure-traction problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
13.6.7 Example: shaft under torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
13.6.8 Example: overspecified boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
13.7 Initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276

14 Galerkin Formulation for Elastodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277

14.1 Manipulation of the residuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
14.1.1 The first two steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
14.1.2 Step 3: Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
14.1.3 Step 3: Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
14.2 Method of weighted residuals as the principle of virtual work . . . . . . . . . . . . . . . . . . . . . 279
14.3 Discretizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
14.3.1 The trial function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
14.3.2 The test function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
14.3.3 Producing the requisite equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
14.4 The discrete equations: system of ODE’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
14.4.1 Inertial term: Mass matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Exercise 72 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
14.4.2 Body loads and traction loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Exercise 73 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Exercise 74 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
14.4.3 Resisting forces: Stiffness matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
14.4.4 Summary of the elastodynamics ODE’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
14.5 Constitutive equations of linearly elastic materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
14.5.1 General anisotropic material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
14.5.2 Orthotropic material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
14.5.3 Transversely isotropic material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
14.5.4 Isotropic material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
14.6 Imposed (thermal) strains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
14.7 Strain-displacement matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Exercise 75 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
Exercise 76 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
Exercise 77 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
14.8 Material directions and basis transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
14.9 Stiffness matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
14.10Pure-traction problems and singular stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304

15 Finite Elements for true 3-D Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

15.1 Modal analysis with the tetrahedron T4: the drum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
15.2 Modal analysis with the tetrahedron T4: the composite rod . . . . . . . . . . . . . . . . . . . . . . 307
15.3 Tetrahedron T10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
15.3.1 Example: the drum revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
15.4 The composite rod with the tetrahedron T10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
15.5 Static analysis with hexahedra H8 and H20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
15.5.1 Hexahedron H8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
8 Contents

15.5.2 Dilatational locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312

15.5.3 Shear locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
15.5.4 Thin clamped square plate with concentrated load . . . . . . . . . . . . . . . . . . . . . . . . 315
15.5.5 Quadratic element H20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
15.5.6 Quadratic element Q8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
15.5.7 Pinched cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
15.5.8 Pinched sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
15.5.9 Beam deflection revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
15.6 Errors, validation, and verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
15.6.1 Verification and Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
15.6.2 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
15.6.3 Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
15.6.4 Using modeling to make predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
15.6.5 Using benchmarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326

16 Analyzing the Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329

16.1 Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
16.2 Interpretation of stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
16.3 Stress concentrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
16.4 Adaptive refinement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336

17 Plane Strain, Plane Stress, and Axisymmetric Models . . . . . . . . . . . . . . . . . . . . . . . 341

17.1 Plane strain model reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
17.2 Plane stress model reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
17.3 Model reduction for axial symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
17.4 Material stiffness for two-dimensional models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
17.5 Strain-displacement matrices for two-dimensional models . . . . . . . . . . . . . . . . . . . . . . . . 347
17.6 Integration for two-dimensional models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
17.7 Thermal strains in two-dimensional models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
17.8 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
17.8.1 Thermal strains in a bimetallic assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
17.8.2 Orthotropic balloon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
17.9 Transient dynamic analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
17.9.1 Centered difference time stepping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
17.9.2 Example: stress wave propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
17.10Solved exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
Exercise 78 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
Exercise 79 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362

18 Consistency + Stability = Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

18.1 Consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
18.1.1 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
18.1.2 Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
18.2 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
18.2.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
1
Model of a Taut Wire

This chapter will formulate a relatively simple model for the so-called initial boundary value prob-
lem that describes the deflection or vibrations of a taut string. In the next chapter, we will seek
approximate solutions to this model with the Galerkin method.

1.1 Deriving the PDE model

Figure 1.1 illustrates an idealization of a taut wire. The wire is under prestress by the force P ,
assumed to be uniform along the length of the wire. The left-hand end is immovably fixed, while the
right-hand end is held in a fixture which can slide perpendicularly to the axis of the wire (occasionally
referred to as a “roller”). A transverse force FL is applied at the movable end. In addition, there
may be some distributed force q (in physical units of force per unit length) acting along the length
(for instance gravity). The transverse displacement is a function of both the axial coordinate x and
the time t, w = w(x, t) . The transverse displacement is assumed to be very small compared to the
length of the wire. The deformation in Figure 1.1 is highly magnified in order to be apparent.

Fig. 1.1. Schematic of taut wire

1.2 Balance equation

We take a segment of length ∆x of the wire (see Fig. 1.2). The forces acting on the segment are the
prestressing forces in either cross-section and the resultant of the distributed load. By assumption
the deflection is very small compared to the span, w ≪ L, and we also assume that the slope of the
wire is very small, w′ = ∂w/∂x ≪ 1. These geometrical features are introduced into the balance of all
the forces. In the horizontal direction we have just the two prestressing forces in opposite directions
and hence they cancel. In the vertical direction we add up the components of the prestressing forces
in the vertical direction with the transverse load, where we take the Taylor-series approximation for
the slope at x + ∆x

w′ (x + ∆x) ≈ w′ (x) + w′′ (x)∆x ,

and
2 Thermal and Stress Analysis with the FEM

∂ 2 w(x)
w′′ (x) = ,
∂x2
and we equate the resultant of the vertical forces to the inertial force (Newton’s law). This leads to
a balance equation for the taut wire
P w′′ + q = µẅ , (1.1)
∂2w
where ẅ = is the acceleration.
∂t2

Fig. 1.2. The forces acting on a segment of the taut wire

1.3 Boundary conditions

The function w that describes the transverse deflection takes two arguments, x, and t. It is defined
on a rectangular domain shown in Fig. 1.3: 0 ≤ x ≤ L, and 0 ≤ t ≤ t̄. The deflection function needs
to be determined to satisfy the balance equation (1.1). However, derivatives with the respect to the
variables x and t needs to be “integrated” in order to arrive at a solution, and that implies the
presence of integration “constants”. In order to determine the solution uniquely we need to resolve
the integrations, and for that we need additional equations. Indeed, there are other things we would
require a solution to satisfy, namely the conditions at the boundaries of the domain rectangle.
How many pieces of information do we need to know? A reasonable answer is, ‘Enough to make
the solution unique.’ Using the definitions
∂w ∂w
v= , θ= ,
∂t ∂x
we may rewrite the balance equation that involves the second derivatives of the function w as a
system of first order partial differential equations
∂θ ∂v
=
∂t ∂x
∂θ ∂v
P +q−µ = 0
∂x ∂t
∂v ∂θ
For each derivative , , one boundary condition (integration constant) will be needed. Similarly,
∂x ∂x
∂v ∂θ
for each of the time derivatives , and one boundary condition along the time axis will be
∂t ∂t
required.

1.4 Boundary conditions (in space)

The conditions on w along the edges of the domain rectangle parallel to the time axis are known
(for historical reasons) as the boundary conditions. (Perhaps also because they are applied along
the physical boundaries of the structure.)
 1.4 Boundary conditions (in space) 3

Fig. 1.3. The domain of the deflection function w. The function w that represents the response of a simply
supported wire going through slightly more than two cycles of vibration at the second natural frequency is
shown as a surface raised above the domain and level curves.

It needs to be realized that the domain of the wire, that is the interval 0 ≤ x ≤ L, has only
one boundary, namely the two endpoints, x = 0 and x = L. Since these two points are disjoint, the
boundary of the interval consists of two disjoint sets. As discussed in more detail in Chapter 5, we
are really prescribing a single boundary condition. Since it happens to be applied at two disjoint
points, we loosely use the plural “boundary conditions”.
In this example, at the left-hand end of the wire we are prescribing in general nonzero displace-
ment,
w(0, t) = w̄0 (t) . (1.2)
As we shall find out, there is a good reason why this kind of condition is commonly called the
essential boundary condition.
At the other end the boundary condition is of a different nature. It is also a bit more interesting,
as we have to derive it. Again, we take a short section of the wire of length ∆x (see Fig. 1.4). This
time there are terms that are multiplied by ∆x, but there are also others which are not. Only the
latter survive when we make ∆x go to zero
∂w
−P (L, t) + FL (t) = 0 . (1.3)
∂x
This boundary condition is simply the balance of forces at the end of the wire. Boundary conditions
of this kind are called natural boundary conditions.

Fig. 1.4. The forces acting on the right-hand end of the taut wire

Exercise 1.
4 Thermal and Stress Analysis with the FEM

Derive the force boundary condition at x = 0.

Solution: Take an infinitesimally short piece of wire starting at the left side end. Express the
balance of forces (the force in the wire P , and the force F0 applied at the left-hand side of the wire
as a load or reaction). From the geometry of the wire we have

Fig. 1.5. The forces acting on the left-hand end of the taut wire

α = arctan w′ (0) ,

and assuming the deflection w is very small compared to the span of the wire, and the slope w′ being
very small compared to one, we have

cos α ≈ 1 , sin α ≈ α

In the vertical direction the sum of forces yields

F0 + P sin α + q∆x = µẅ∆x

which gives with the geometrical simplifications

F0 + P α = F0 + P w′ (0) + q∆x = µẅ∆x

Since the piece of the wire is infinitesimally short, ∆x → 0, we can omit the transverse load and the
inertial force. This is then the boundary condition to control the slope at the left-hand side end

w′ (0) = −F0 /P

in dependence on the prestressing force and the given force F0 .

Exercise 2.
List possible combinations of force and displacement boundary conditions for prestressed cable.
Solution: At either end the transverse motion of the cable may be eliminated (known to be zero)
or prescribed at given nonzero value –pin support. Alternatively, at either end the transverse motion
of the cable may be unknown –roller support. When an end of the cable is supported by a pin, the
associated reaction will generally be unknown. On the other hand, at the end supported by the roller
we may apply a known force, zero or nonzero. Hence the possible combinations are as shown here:
 1.7 Examples 5

1.5 Initial conditions (boundary conditions in time)

Along the edges of the domain rectangle that are parallel to the space axis we also apply two
pieces of information. However, as we are all aware, the time direction is special. Therefore, it will
probably come to us naturally to expect to know something about the deflection at one point in
time, typically at t = 0. Because this is the initial point along the time axis, this condition is known
as the initial condition. We need two equations, one for each variable (i.e. for each derivative), in
order to compensate for prescribing the condition at one point only:
∂w
w(x, 0) = W̄ (x), (x, 0) = V̄ (x) , (1.4)
∂t
where W̄ (x) (the initial deflection) and V̄ (x) (the initial velocity) are known functions.

1.6 Initial Boundary Value Problem

The balance equation (1.1), the boundary conditions (1.2) and (1.3), and the initial conditions (1.4)
are all we need to fully define the model. It is an initial boundary value problem, and as such it is
quite typical of the models with which structural engineers deal in practice. In what follows, we shall
find out how to formulate an algorithm, the so-called Galerkin finite element method, which will find
an approximate solution to this problem. In the following exercises we will also solve some simple
static problems analytically, in order to familiarize ourselves with some aspects of the solutions.

1.7 Examples
Here are a few analytical solutions for selected statically loaded prestressed wire configurations.
Note on the solutions to the solved exercises: we describe here solutions developed with Matlab,
but using Matlab is not required. All the analytical solutions below can be obtained by hand, or
perhaps with another computer algebra system (Mathcad, Maple, or Mathematica).

Exercise 3.
Solve analytically for the static deflection of the shown prestressed pinned- pinned wire with
piecewise uniform transverse load.

Solution: Solving for the static deflection amounts to finding a solution of the following boundary
value problem:

P w′′ + q1 = 0 0 ≤ x ≤ αL
P w′′ + q2 = 0 αL ≤ x ≤ L
w(0) = 0
w(L) = 0

Note that the balance equation needs to be written for each interval separately since the transverse
load is discontinuous. A consequence of the discontinuous transverse load is that the curvature of the
cable (the second derivative of the deflection) is also discontinuous at x = αL. The first derivative
of the deflection however must be continuous at x = αL, otherwise the second derivative would
be infinite at that point. (Consider that the second derivative corresponds to the curvature of the
6 Thermal and Stress Analysis with the FEM

cable; also, the curvature is 1/(radius of the osculating circle); the radius of a sharp corner is zero
and therefore, the curvature at the sharp corner is infinite.) Therefore, we can supplement the above
BVP with two additional conditions: the deflection and the slope of the deflection curve must be
continuous at x = αL.
P w′′ + q1 = 0 0 ≤ x ≤ αL
P w′′ + q2 = 0 αL ≤ x ≤ L
w(0) = 0
w(L) = 0
w((αL)− ) = w((αL)+ )
w′ ((αL)− ) = w′ ((αL)+ )

where x = (αL)− and x = (αL)+ mean immediately to the left and immediately to the right of
x = αL. Because in each interval the transverse load is constant, we can deduce that in each interval
the deflection curve is going to be quadratic in x. Therefore we will write

w = A1 + B1 x + q1 C1 x2 , 0 ≤ x ≤ αL

and

w = A2 + B2 x + q2 C2 x2 , αL ≤ x ≤ L

Substituting into the balance equation we obtain

1
P w′′ + q1 = 2P q1 C1 + q1 = 0 ⇒ C1 = −
2P
1
and similarly C2 = − 2P . Therefore, there are four coefficients to be determined to complete the
solution, A1 , B1 , A2 , B2 . Fortunately, we have the boundary conditions and the conditions of conti-
nuity where the load changes. Four conditions for four unknown coefficients. Here is a short program
in Matlab to solve for the coefficients (it would be of course possible to solve the resulting system
by hand):
% Solve for the deflection of a prestressed cable
% with piecewise uniform distributed load
syms A1 B1 A2 B2 alpha L P q1 q2 x real
w1 = @(x)(A1+B1*x-q1*x^2/(2*P));
dw1 = @(x)(B1-q1*x/P);
w2 = @(x)(A2+B2*x-q2*x^2/(2*P));
dw2 = @(x)(B2-q2*x/P);

Solution =solve([char(w1(0)) ’=0’],...

[char(w2(L)) ’=0’],...
[char(w1(alpha*L)) ’=’ char(w2(alpha*L))],...
[char(dw1(alpha*L)) ’=’ char(dw2(alpha*L))],...
’A1’,’B1’,’A2’,’B2’);
The values of the constants in terms of the variables in the problem are:
pretty(Solution.A1)
0
pretty(simplify(Solution.B1))
2 2
L (q2 + 2 alpha q1 - 2 alpha q2 - alpha q1 + alpha q2)
--------------------------------------------------------
2 P
pretty(simplify(Solution.A2))
 1.7 Examples 7

2 2
L alpha (q1 - q2)
-------------------
2 P
pretty(simplify(Solution.B2))
2 2
L (q2 - alpha q1 + alpha q2)
------------------------------
2 P
Now we can pick some particular numbers for the parameters and plot the deflection curve. Note
that the two intervals are of the same length and the two distributed loads are equal in magnitude
but of opposite sign.
alpha = 0.5; q1 = 4; q2 = -4; L = 100; P = 25;
A1 =Solution.A1; B1 =Solution.B1;
A2 =Solution.A2; B2 =Solution.B2;
x= linspace(0,alpha*L, 20);
plot (x,eval((A1+B1*x-q1*x.^2/(2*P))));
hold on
x= linspace(alpha*L,L, 20);
plot (x,eval((A2+B2*x-q2*x.^2/(2*P))));
set(gca,’ytick’,-50:10:50)
xlabel (’x’)
ylabel (’w’)
grid on
left_handed_axes% positive deflection downwards
1
Note that the function left handed axes orients the axes of the figure in order to display the
positive deflection downwards.

−50
−40
−30
−20
−10
w

0
10
20
30
40
50
0 20 40 60 80 100
x

Exercise 4.
Solve analytically for the static deflection of the shown prestressed pinned- pinned cable with
piecewise uniform transverse load and support-settlement loads.

1
Folder: FAESOR/util
8 Thermal and Stress Analysis with the FEM

Solution: Solving for the static deflection amounts to finding a solution of the following boundary
value problem:
P w′′ + q1 = 0 0 ≤ x ≤ αL
P w′′ + q2 = 0 αL ≤ x ≤ L
w(0) = w0
w(L) = wL

The balance equation needs to be written for each interval separately since the transverse load
is discontinuous. Again, the BVP needs to be supplemented with two additional conditions: the
deflection and the slope of the deflection curve must be continuous at x = αL.
P w′′ + q1 = 0 0 ≤ x ≤ αL
P w′′ + q2 = 0 αL ≤ x ≤ L
w(0) = w0
w(L) = wL
w((αL)− ) = w((αL)+ )
w′ ((αL)− ) = w′ ((αL)+ )

where x = (αL)− and x = (αL)+ mean immediately to the left and immediately to the right of
x = αL. Because in each interval the transverse load is constant, we can deduce that in each interval
the deflection curve is going to be quadratic in x. Therefore we will write

w = A1 + B1 x + q1 C1 x2 , 0 ≤ x ≤ αL

and

w = A2 + B2 x + q2 C2 x2 , αL ≤ x ≤ L

Substituting into the balance equation we obtain

1
P w′′ + q1 = 2P q1 C1 + q1 = 0 ⇒ C1 = −
2P
1
and similarly C2 = − 2P . Therefore, there are four coefficients to be determined to complete the solu-
tion, A1 , B1 , A2 , B2 . Fortunately, we have the boundary conditions and the conditions of continuity
where the load changes. Four conditions for four unknown coefficients.
For expediency here is a short program in Matlab to solve for the coefficients:
% Solve for the deflection of a prestressed cable
% with piecewise uniform load + support settlement
syms A1 B1 A2 B2 alpha L P q1 q2 x w0 wL real
w1 = @(x)(A1+B1*x-q1*x^2/(2*P));
dw1 = @(x)(B1-q1*x/P);
w2 = @(x)(A2+B2*x-q2*x^2/(2*P));
dw2 = @(x)(B2-q2*x/P);

Solution =solve([char(w1(0)) ’=’ char(w0)],...

[char(w2(L)) ’=’ char(wL)],...
[char(w1(alpha*L)) ’=’ char(w2(alpha*L))],...
[char(dw1(alpha*L)) ’=’ char(dw2(alpha*L))],...
’A1’,’B1’,’A2’,’B2’);
 1.7 Examples 9

The values of the constants in terms of the variables in the problem are:
>> pretty(simplify(Solution.A1))

w0
>> pretty(simplify(Solution.B1))

2 2
L (q2 + 2 alpha q1 - 2 alpha q2 - alpha q1 + alpha q2) w0 - wL
-------------------------------------------------------- - -------
2 P L
>> pretty(simplify(Solution.A2))

2 2
L alpha (q1 - q2)
w0 + -------------------
2 P
>> pretty(simplify(Solution.B2))

2 2
L (q2 - alpha q1 + alpha q2) w0 - wL
------------------------------ - -------
2 P L
Now we can pick some particular numbers for the parameters and plot the deflection curve. Note
that the two intervals are of the same length and the two distribute loads are equal in magnitude
but of opposite sign.
alpha = 0.5; q1 = 4; q2 = -4; L = 100; P = 25;
w0 =-10; wL = 20;
A1 =Solution.A1; B1 =Solution.B1;
A2 =Solution.A2; B2 =Solution.B2;
x= linspace(0,alpha*L, 20);
plot (x,eval((A1+B1*x-q1*x.^2/(2*P))));
hold on
x= linspace(alpha*L,L, 20);
plot (x,eval((A2+B2*x-q2*x.^2/(2*P))));
set(gca,’ytick’,-50:10:50)
xlabel (’x’)
ylabel (’w’)
grid on
left_handed_axes% positive deflection downwards

−40
−30
−20
−10
0
w

10
20
30
40
50

0 20 40 60 80 100
x
10 Thermal and Stress Analysis with the FEM

Exercise 5.
Solve analytically for the static deflection of the shown prestressed roller- pinned cable with
piecewise uniform transverse load, concentrated force F0 at the left-hand side end, and support-
settlement loads.

Solution: Solving for the static deflection amounts to finding a solution of the following boundary
value problem:

P w′′ + q1 = 0 0 ≤ x ≤ αL
P w′′ + q2 = 0 αL ≤ x ≤ L
w′ (0) = − FP0
w(L) = wL

The balance equation needs to be written for each interval separately since the transverse load
is discontinuous. Again, the BVP needs to be supplemented with two additional conditions: the
deflection and the slope of the deflection curve must be continuous at x = αL.

P w′′ + q1 = 0 0 ≤ x ≤ αL
P w′′ + q2 = 0 αL ≤ x ≤ L
w′ (0) = − FP0
w(L) = wL
w((αL)− ) = w((αL)+ )
w′ ((αL)− ) = w′ ((αL)+ )

where x = (αL)− and x = (αL)+ mean immediately to the left and immediately to the right of
x = αL. Because in each interval the transverse load is constant, we can deduce that in each interval
the deflection curve is going to be quadratic in x. Therefore we will write

w = A1 + B1 x + q1 C1 x2 , 0 ≤ x ≤ αL

and

w = A2 + B2 x + q2 C2 x2 , αL ≤ x ≤ L

Substituting into the balance equation we obtain

1
P w′′ + q1 = 2P q1 C1 + q1 = 0 ⇒ C1 = −
2P
1
and similarly C2 = − 2P . Therefore, there are four coefficients to be determined to complete the
solution, A1 , B1 , A2 , B2 . Fortunately, we have the boundary conditions and the conditions of conti-
nuity where the load changes. Four conditions for four unknown coefficients. As a for way will be
well served by a short program in Matlab. Here is the symbolic solution for the coefficients from the
boundary conditions and continuity conditions. Note the change with the respect to the previous
codes due to the roller boundary condition.
 1.7 Examples 11

syms A1 B1 A2 B2 alpha L P q1 q2 x F0 wL real

w1 = @(x)(A1+B1*x-q1*x^2/(2*P));
dw1 = @(x)(B1-q1*x/P);
w2 = @(x)(A2+B2*x-q2*x^2/(2*P));
dw2 = @(x)(B2-q2*x/P);

Solution =solve([char(dw1(0)) ’=’ char(-F0/P)],...

[char(w2(L)) ’=’ char(wL)],...
[char(w1(alpha*L)) ’=’ char(w2(alpha*L))],...
[char(dw1(alpha*L)) ’=’ char(dw2(alpha*L))],...
’A1’,’B1’,’A2’,’B2’);
The values of the constants in terms of the variables in the problem are:
>> pretty(simplify(Solution.A1))

2 2 2 2 2 2 2
2 L alpha - L alpha L alpha - 2 L alpha + L F0 L
--------------------- q1 + -------------------------- q2 + ---- + wL
2 P 2 P P

>> pretty(simplify(Solution.B1))

F0
- --
P
>> pretty(simplify(Solution.A2))

L (2 F0 + L q2 + 2 L alpha q1 - 2 L alpha q2)

wL + ---------------------------------------------
2 P
>> pretty(simplify(Solution.B2))

F0 + L alpha q1 - L alpha q2
- ----------------------------
P
Now we can pick some particular numbers for the parameters and plot the deflection curve. Note
that the two intervals are of the same length. Note also that the distributed load in the left-hand
side interval is zero, and hence the deflection curve there has a zero curvature (straight line).
alpha = 0.5; q1 = 0; q2 = -4; L = 100; P = 25;
F0 =0; wL = 20;
A1 =Solution.A1; B1 =Solution.B1;
A2 =Solution.A2; B2 =Solution.B2;
x= linspace(0,alpha*L, 20);
plot (x,eval((A1+B1*x-q1*x.^2/(2*P))));
hold on
x= linspace(alpha*L,L, 20);
plot (x,eval((A2+B2*x-q2*x.^2/(2*P))));
set(gca,’ytick’,-180:20:20)
xlabel (’x’)
ylabel (’w’)
grid on
left_handed_axes% positive deflection downwards
12 Thermal and Stress Analysis with the FEM

−180
−160
−140
−120
−100
−80

w
−60
−40
−20
0
20

0 20 40 60 80 100
x
2
The Method of Galerkin

We continue working with the prestressed cable model. In this chapter we will begin to come to
grips with the possibility of satisfying the equations of the model not exactly but only approxi-
mately. There’s going to be an error in the balance equation (which we shall call a residual; another
appropriate label might be imbalance). Similarly, the natural (force) boundary condition may not
be satisfied exactly, and will also produce a residual.
In this book, the approximate solutions are obtained with the Galerkin method. Boris Grigorye-
vich Galerkin became a teacher of structural mechanics in St. Petersburg Polytechnical Institute in
1908. Among his contemporaries, also active in St. Petersburg, were I. G. Bubnov, A. N. Krylov, and
S. P. Timoshenko, all well-known names in various areas of mechanics. In 1915 Galerkin published
an article, in which he put forward an idea of an approximate method to solve differential boundary
value problems (he was working on plate and shell models at that time). Around that time Bubnov
developed similar variational approach, hence this method is also known as the Bubnov-Galerkin
method.
It is an extremely general and hence very valuable approach. In particular, it is applicable in
various areas of nonlinear computational mechanics, for instance for inelastic deformation of mate-
rials, large deflection and large strain deformation, and so on. In this book we will restrict ourselves
to linear models however.

2.1 Residual of the balance equation

For the moment we will reduce the complexity of the prestressed wire IBVP model by considering
only statics. The omission of dynamics will not invalidate the following developments, and including
dynamics will be shown to be straightforward. Thus in the following we will consider the statics
BVP, consisting of the balance equation

P w′′ (x) + q(x) = 0 , (2.1)

and the boundary conditions

w(0) = w̄0 . (2.2)

and

−P w′ (L) + FL = 0 . (2.3)

The balance equation (2.1) may be written in the residual form as

P w′′ (x) + q(x) = rB (x) . (2.4)

The residual rB is identically zero if w is the exact solution. For an approximate solution, the residual
rB varies from point to point, and is in general nonzero.
14 Thermal and Stress Analysis with the FEM

Fig. 2.1. Residual that integrates to zero, but is not identically zero

Checking that the balance residual is identically zero at each point x does not provide us with
anything we can use to talk about approximate solutions: the residual is either zero or it isn’t. So
how do we measure whether the approximate solution, for which the residual is not zero, is in some
sense satisfactory (or not)?

2.2 Integral test of the residual

One possible choice of a quality measure is to integrate the residual over the domain (length of the
wire). We could think of the integral
Z L
rB (x) dx (2.5)
0

as a test: if the residual is identically zero, this integral will also come out zero. However, Eq. (2.5)
may be zero even when the residual is not identically zero. In other words, if we wanted to prove
that the residual corresponded to an exact solution, this would be an incomplete and flawed test.
Consider Fig. 2.1: the integral (2.5) is zero, but the residual itself may be very large (for instance,
when rB = A sin(2πnx/L), with n = 1, 2, ...).

2.3 Test function

A remedy that addresses this blindness of (2.5) to the shape of the residual may be to use a “window”
(test) function η(x)
Z L
η(x)rB (x) dx . (2.6)
0

Note that η(x) is an arbitrary function. In particular, it could be a function of the shape shown
in Fig. 2.2, which is certainly going to give a nonzero value for (2.6) (the hatched area at the
bottom). Therefore, it correctly indicates that the residual does not correspond to the exact solution.
Equation (2.6) is known as the weighted residual statement, because each test function η applies
a variable weight to the residual in different parts of the domain. Approximate approaches that start
from the weighted residual statement are known as weighted residual methods.
Equation (2.6) is a reliable way of testing the residual, but computationally it seems hardly less
difficult than testing the residual at each point of the domain: equation (2.6) needs to be evaluated
for an infinite number of functions η in order to make sure there are no bumps in the residual. The
job will still take an infinite time.
Let us contemplate a tangible analogy of what we’re trying to do in Eq. (2.6). Imagine our job
is to hold an inflatable balloon in a box, so that it does not jut out anywhere. Use the fingers of one
hand to press down on the balloon, so that the balloon is at the top of the box in the spot where
it is being held by the finger. If we put down all five fingers, the situation is as shown on the left
in Fig. 2.3. Each of the fingers may be thought of as a single test function η that pushes down the
residual in some spots.
Evidently, the balloon bulges out a little bit in between the fingers, and a lot everywhere else.
However, we have the option of pressing down on the balloon with the fingers of our other hand, and
if we enrol our friends and relatives, and the chance passersby, and distribute the pressing fingers
 2.4 Trial function 15

Fig. 2.2. Nonzero residual which is detected in the integral (2.6)

Fig. 2.3. Stuffing a balloon into a box

wisely, we will manage to do a better and better job of stuffing the balloon into the box and holding
it so that it does not protrude very much. Indeed, with an infinite number of fingers, we can hold
the balloon so that it does not protrude at all. Note that we have to distribute the pressing fingers
in some sense densely and uniformly – no parts of the interval 0 ≤ x ≤ L may be left out, since
the residual could stay nonzero there.
In this way, we may begin to see how a trial-and-test approximate method may be formulated.
Selecting a finite number of suitable functions ηj (fingers), we may be able to control distribution
and magnitude of the residual (but, in general, it will remain nonzero). By applying larger numbers
of test functions, we will be able to reduce the error in the residual and get a better solution. Also,
for each ηj , j = 1, ..., N , we will make the integral (2.6) vanish
Z L
ηj (x)rB (x) dx = 0 , (2.7)
0

which provides us with the means of calculating N coefficients (numbers) from these N equations.

2.4 Trial function

The task of formulating the approximate solution requires describing the shape of the deflection w.
This can be done in a variety of ways, but for reasons that we shall give later, a piecewise linear
representation is a good choice. Figure 2.4 illustrates this concept by showing how the shape may
be defined by the N coefficients wj . The attentive reader will at this point fidget: the piecewise
linear shape of the deflection curve is not going to allow us to express the second order derivatives
w′′ = ∂ 2 w/∂x2 . At the corners, the first derivatives will be discontinuous, and hence the second
derivative will be a spike (the so-called Dirac delta function). We can choose either to abandon the
piecewise linear shape, and pass a smooth curve through the filled-circle points, or, we could change
16 Thermal and Stress Analysis with the FEM

the rules of the game by getting rid of the second-order derivatives. As we shall presently see, the
latter choice is commonly preferred.
In any case, the Eqs. (2.7) may be used to calculate the values of wj , j = 1, ..., N . The function
that describes the shape of the approximate solution (with the N free parameters) is known as the
trial function. It describes a possible (candidate, trial) shape of the approximate solution; which
becomes the solution once the values of the free parameters are known.

Fig. 2.4. Piecewise linear trial function

2.5 Shifting derivatives

Substituting for the balance residual, we get two terms
Z L Z L Z L
′′ (2.8)
ηj (x)rB (x) dx = ηj (x)P w (x) dx + ηj (x)q(x) dx .
0 0 0

Integration by parts may be applied to the first term on the right-hand side using the identity
′
(ηj P w′ ) = ηj ′ P w′ + ηj P w′′

and integrating from 0 to L we obtain for the left-hand side

Z L
′ L
(ηj P w′ ) dx = [ηj P w′ ]0 = ηj (L)P w′ (L) − ηj (0)P w′ (0)
0

Therefore we see that we can replace the term with the second derivatives as
Z L Z L
′′ ′ ′
ηj P w dx = ηj (L)P w (L) − ηj (0)P w (0) − ηj ′ P w′ dx . (2.9)
0 0

This does the trick: the second derivatives are gone from the trial function w. All that is left are
first derivatives, and the piecewise linear functions that we spoke about are now acceptable.

2.6 Essential boundary condition

We also must satisfy the essential boundary condition (2.2). We may realize that at the left-hand end
of the wire we can quite simply design the trial function to satisfy this condition identically. This is
not difficult, and we have therefore the following requirements on the trial function at this point: to
satisfy the essential boundary condition we require w(0) = w̄0 , and further the trial function w(x)
needs to be sufficiently smooth in x for the derivatives in the residual rB to exist.

2.7 Natural boundary condition

In general it will not be possible to come up with a trial function that would satisfy (2.3) at the
right-hand end of the wire (x = L) identically. We would have to control the slope of w at x = L,
and that is awkward at best. Therefore in addition to the residual rB along the length of the wire we
 2.7 Natural boundary condition 17

will need to put up with an imbalance (residual) rF at x = L where the natural boundary condition
is applied

rF = −P w′ (L) + FL . (2.10)

We may take a list from the book written for the balance residual: multiply the residual with a test
function and integrate. In this case integration means integrate over the boundary at x = L, that’s
where rF “lives”. Since we have taken ηj (x) for the balance residual, we may take the same function
here, evaluated on the boundary. Therefore the weighted residual for the natural boundary condition
may be written as

ηj (L)rF = ηj (L)(−P w′ (L) + FL )

Unfortunately, we cannot just set

ηj (L)rF = ηj (L)(−P w′ (L) + FL ) = 0 (2.11)

There are two ways in which this could happen: either the parenthesis is identically zero (which as
we said above is in general not going to happen), or we choose ηj (L) = 0. The latter is not helpful
however: all it means is that the force FL does not play any role in the solution, and that is definitely
not what we want. So we cannot keep the natural boundary condition in a weighted residual of its
own. Over the years, the following clever manipulation was developed to resolve this dilemma: note
that ηj (L)P w′ (L) appears both in (2.9) and in the natural boundary condition weighted residual
equation above. Therefore, we add the two equations (2.8) and (2.11), and introduce (2.9) to give
Z L
ηj (x)rB (x) dx + ηj (L)rF =
0 Z Z
L L
′ ′ ′ ′
ηj (L)P w (L) − ηj (0)P w (0) − ηj P w dx + ηj (x)q(x) dx + ηj (L)(−P w′ (L) + FL ) = 0 ,
0 0
(2.12)

where the underlined terms cancel, and the expression simplifies to

Z L
ηj rB dx + ηj (L)rF =
0 Z L Z L (2.13)
−ηj (0)P w′ (0) − ηj ′ P w′ dx + ηj q dx + ηj (L)FL = 0 ,
0 0

So far so good. We have combined the balance residual with the natural boundary condition residual
to obtain an expression which has only the first-order derivatives on the test and the trial functions,
and which incorporates the given force FL .
Unfortunately, there is one more snag: At the left-hand side end of the wire (at the pin support)
the value of the force (−P w′ (0)) is unknown – it is a reaction. We would prefer not to have this
force in the weighted residual. We do have the option of requiring that ηj vanish at x = 0, and thus
eliminate (−P w′ (0)). This will burden all the test ηj ’s with a condition, ηj (x = 0) = 0, but that is
something we can probably afford. At this point we will require all test functions to become zero
where the essential boundary conditions are prescribed (x = 0). Later we will relax this condition
since for some applications it is worthwhile to be able to compute reactions too.
In summary: We have satisfied the essential boundary condition by design of the trial function,
and the force boundary condition (2.10) was merged into the balance residual (which is by the way
why we call this the “natural” boundary condition: it appears naturally in the model equations).
Hence we will try to find the approximate solution w to satisfy the balance equation combined with
the natural boundary condition in the residual form
Z L Z L
ηj (L)FL − ηj ′ P w′ dx + ηj q dx = 0, j = 1, ..., N , (2.14)
0 0
18 Thermal and Stress Analysis with the FEM

where we subject the test and trial function to the conditions

ηj (x = 0) = 0, ηj ∈ C 0 , j = 1, ..., N ,
(2.15)
w(x = 0) = w̄0 , w ∈ C0.

We write for the trial function w ∈ C 0 and similarly for the test functions. This literally means that
the functions are continuous (C 0 denotes the set of functions that are continuous on the real line);
that is a substitute for a more precise mathematical statement, but one that nevertheless ensures
that the integrals in (2.14) exist.

2.8 Stiffness matrix and load vector

Let us revisit the choice of the test and trial functions. As advertised in Section 2.4, we have been
able to change the requirements on the test and trial function: Their derivatives are now balanced–
only the first-order derivatives are needed for either. Therefore, the piecewise linear interpolation
function of Fig. 2.4 is now a possibility. However, we can still forge ahead while keeping our options
open. In this section we therefore still do not commit to a specific form of the trial and test function.
To describe the trial function, we will resort to a common technique in interpolation and approx-
imation literature which is to write the trial function as a linear combination of basis functions.
Therefore, let us assume that the trial function is written as
N
X
w(x) = Ni (x)wi , (2.16)
i=1

where the wi ’s are the coefficients of the linear combination (real numbers); these coefficients are
also called degrees of freedom. The Ni (x)’s are known (suitably chosen) basis functions. Note that
the number of terms N in the trial function matches the number of the test functions used. That is
because the number of unknowns needs to be matched to the number of equations available.
Substituting into (2.14), we obtain
Z L N
X Z L
ηj (L)FL − ηj ′ P Ni ′ wi dx + ηj q dx = 0, j = 1, ..., N , (2.17)
0 i=1 0

which may be simplified to

N Z ! Z
X L L
′ ′
ηj (L)FL − ηj P Ni dx wi + ηj q dx = 0, j = 1, ..., N
i=1 0 0 (2.18)
0 0
ηj (x = 0) = 0, ηj ∈ C , j = 1, ..., N , w(x = 0) = w̄0 , w∈C ..

With the definitions

Z L
Kji = ηj ′ P Ni ′ dx , (2.19)
0

and
Z L
Lj = ηj (L)FL + ηj q dx (2.20)
0

we may write (2.18) in the form

N
X
Kji wi = Lj , j = 1, ..., N (2.21)
i=1
 2.8 Stiffness matrix and load vector 19

that makes it clear we have converted the original BVP to a linear-algebra problem of a system
of coupled linear algebraic equations. The coefficients Kji are usually referred to as the stiffness
matrix elements, and Lj are the elements of the load vector . Note well that the above describes
a Galerkin method, but it has nothing to do yet with finite elements.

Exercise 6.
Solve for the approximate deflection of a simply-supported prestressed cable with uniform load
using the Galerkin method. Take as the trial function N1 (x) = sin(πx/L), and set the test function
to coincide with the trial function, η1 = N1 . Compare the midpoint deflection computed analytically
and approximately.
Solution: We are solving the boundary value problem:

P w′′ + q = 0
w(0) = 0
w(L) = 0

The trial function is

w(x) = a1 N1 (x)

Note that the trial function satisfies the essential boundary conditions because

N1 (0) = N1 (L) = 0

The coefficient a1 is the only unknown. Therefore one test function is sufficient since we need only
one equation.
Equation (2.18) simplifies for no natural boundary conditions to
Z L Z L
∂ηj ∂w
− P dx + ηj (x)q(x) dx = 0 .
0 ∂x ∂x 0

Substituting the test and trial function we have

Z L Z L
∂N1 ∂N1
− P a1 dx + N1 (x)q dx = 0 .
0 ∂x ∂x 0

and simplifying further

Z L Z L
∂N1 ∂N1
−a1 P dx + q N1 (x) dx = 0 .
0 ∂x ∂x 0

Just five lines of Matlab symbolic algebra do the job:

syms L P q x real
N1=sin(pi*x/L);
K=int(diff(N1)*P*diff(N1),0,L)
F=q*int(N1,0,L)
a1=K\F
yielding
K =
(P*pi^2)/(2*L)
F =
(2*L*q)/pi
a1 =
(4*L^2*q)/(P*pi^3)
20 Thermal and Stress Analysis with the FEM

The analytical solution is

qx(L − x)
wex (x) =
2P
We can compare the midpoint deflection, for instance: analytical
qL2 0.125qL2
wex (L/2) = =
8P P
versus approximate
4qL2 0.129qL2
wapp (L/2) = ≈
π3 P P
The analytical (dashed black) and approximate (solid red) curves of the deflections scaled by
(P/q/L2 ) versus the nondimensional coordinate ξ = x/L are shown in the figure below.

0.02

0.04
(P/q/L )w

0.06
2

0.08

0.1

0.12

0.14
0 0.2 0.4 0.6 0.8 1
ξ

Exercise 7.
For the approximate solution computed in the exercise 6 evaluate the balance residual.
Solution: The Galerkin approximate solution to
P w′′ + q = 0 w(0) = 0, w(L) = 0
was found as
(4L2 q)
wapp (x) = sin(πx/L)
(P π 3 )
The balance residual is
rB = P w′′ + q
A bit of symbolic algebra
syms L P q x real
N1=sin(pi*x/L);
K=int(diff(N1)*P*diff(N1),0,L);
F=q*int(N1,0,L);
a1=K\F;
w=N1*a1;
rB =P*diff(diff(w))+q
 2.8 Stiffness matrix and load vector 21

yields the balance residual as a function in x

rB =
q - (4*q*sin((pi*x)/L))/pi
The balance the residual is not identical is identically zero – after all this is an approximate solution.
Here is the residual scaled by the transverse load:

−0.4

−0.2

0.2
r /q
B

0.4

0.6

0.8

1
0 0.2 0.4 0.6 0.8 1
ξ

Plotting the product N1 rB , that is the product of the test function and the residual, illustrates that
the solution was found by making the residual orthogonal to the test function: when the residual
multiplied by the test function is integrated over the length of the domain of the cable we obtain
zero, the areas above the horizontal axis canceling those below.
>> int(N1*rB,0,L)
ans =
0

−0.3

−0.2

−0.1
N r /q
1 B

0.1

0.2
0 0.2 0.4 0.6 0.8 1
ξ

Exercise 8.
Solve for the approximate deflection of a simply-supported prestressed cable with uniform load
using the Galerkin method. Take as the trial functions N1 (x) = sin(πx/L), and N2 (x) = sin(3πx/L)
and set the test functions to coincide with the trial functions, η1 = N1 , η2 = N2 . Compare the
midpoint deflection computed analytically and approximately.
22 Thermal and Stress Analysis with the FEM

Solution: We are solving the boundary value problem:

P w′′ + q = 0
w(0) = 0
w(L) = 0

The trial function is

w(x) = a1 N1 (x) + a2 N2 (x)

Note that the trial function satisfies the essential boundary conditions because

Nj (0) = Nj (L) = 0 for j = 1, 2

The coefficients a1 , a2 are the unknowns. Therefore two test functions are needed to generate to
equations from which to solve for the unknowns.
Equation (2.14) simplifies for statics (zero accelerations) and no natural boundary conditions to
Z L Z L
∂ηj (x) ∂w(x)
− P dx + ηj (x)q dx = 0 for j = 1, 2 .
0 ∂x ∂x 0

Substituting the test and trial function we have

Z L Z L Z L
∂N1 ∂N1 ∂N1 ∂N2
− P a1 dx + − P a2 dx + N1 (x)q dx = 0 ,
Z0 L ∂x ∂x Z0 L ∂x ∂x Z0 L
∂N2 ∂N1 ∂N2 ∂N2
− P a1 dx + − P a2 dx + N2 (x)q dx = 0 ,
0 ∂x ∂x 0 ∂x ∂x 0

and simplifying further

Z ! Z L ! Z L
L
∂N1 ∂N1 ∂N1 ∂N2
− P dx a1 + − P dx a2 + N1 (x)q dx = 0 ,
0 ∂x ∂x 0 ∂x ∂x 0
Z ! Z L ! Z L
L
∂N2 ∂N1 ∂N2 ∂N2
− P dx a1 + − P dx a2 + N2 (x)q dx = 0 ,
0 ∂x ∂x 0 ∂x ∂x 0

Here the terms

Z L
∂Ni ∂Nj
Ki,j = P dx
0 ∂x ∂x
are the elements of the stiffness matrix, and
Z L
Fi = Ni (x)q dx
0

are the elements of the load vector. These objects can be computed symbolically using only a few
lines of Matlab symbolic algebra (note that
>> diff([N1;N2])*P*diff([N1,N2])
ans =
[(P*pi^2*cos(pi*x/L)^2)/L^2, (3*P*pi^2*cos(pi*x/L)*cos(3*pi*x/L))/L^2]
[(3*P*pi^2*cos(pi*x/L)*cos(3*pi*x/L))/L^2, (9*P*pi^2*cos(3*pi*x/L)^2)/L^2]
is a 2 × 2 matrix):
>> syms L P q x real
N1=sin(pi*x/L);
N2=sin(3*pi*x/L);
 2.8 Stiffness matrix and load vector 23

K=int(diff([N1;N2])*P*diff([N1,N2]),0,L)
F=q*int([N1;N2],0,L)
a=K\F
K =
[ (P*pi^2)/(2*L), 0]
[ 0, (9*P*pi^2)/(2*L)]
F =
(2*L*q)/pi
(2*L*q)/(3*pi)
a =
(4*L^2*q)/(P*pi^3)
(4*L^2*q)/(27*P*pi^3)
yielding The analytical solution is

qx(L − x)
wex (x) =
2P
which can be compared with the approximate Galerkin solution

4L2 q 4L2 q
wapp (x) = sin(πx/L) + sin(3πx/L)
P π3 27P π 3
The midpoint deflections are: analytical

qL2 0.125qL2
wex (L/2) = =
8P P
versus approximate

104qL2 0.1242qL2
wapp (L/2) = ≈
27π 3 P P
The deflection error, that is the difference between the approximate and analytical deflections nor-
malized by wex (L/2) versus the nondimensional coordinate ξ = x/L is shown in the figure below.
Clearly the largest error is just a little bit over 1%.

−0.015
Normalized error of deflection

−0.01

−0.005

0.005

0.01
0 0.2 0.4 0.6 0.8 1
ξ

Exercise 9.
24 Thermal and Stress Analysis with the FEM

Solve for the approximate deflection of a prestressed table with uniform load, simply supported
at x = 0 and force-free boundary condition at x = L, using the Galerkin method. Take as the trial
function basis the to functions N1 (x) = x and N2 (x) = x2 . The test functions are Nj , j = 1, 2.
Solution: We are solving the boundary value problem:

P w′′ + q = 0
w(0) = 0
w′ (L) = 0

The trial function is

w(x) = a1 N1 (x) + a2 N2 (x)

Note that the trial function satisfies the essential boundary condition w(0) = 0 because

N1 (0) = N2 (0) = 0

The coefficients a1 , a2 are the unknowns.

Equation (2.14) simplifies for statics (zero accelerations) and a natural boundary conditions at
x = L to
Z L Z L
∂ηj ∂w
ηj (L)FL − P dx + ηj (x)q(x) dx = 0 .
0 ∂x ∂x 0

However, since FL = 0 (that is the meaning of “force-free”), there will be no contribution of the first
term in the final equations.
Substituting the test and trial function we have
Z L Z L Z L
∂N1 ∂N1 ∂N1 ∂N2
− P a1 dx + − P a2 dx + N1 (x)q dx = 0 ,
Z0 L ∂x ∂x Z0 L ∂x ∂x Z0 L
∂N2 ∂N1 ∂N2 ∂N2
− P a1 dx + − P a2 dx + N2 (x)q dx = 0 ,
0 ∂x ∂x 0 ∂x ∂x 0

The Matlab symbolic algebra is again helpful (note how minimal the needed changes are with
respect to the code in exercise 8– can you think of a better argument for using computers for the
grunt work?):
>> syms L P q x real
N1=x;
N2=x^2;
K=int(diff( [N1;N2])*P*diff([N1,N2]),0,L)
F=q*int([N1;N2],0,L)
a=K\F
The quantities of the discrete system are
K =
[ L*P, L^2*P]
[ L^2*P, (4*L^3*P)/3]
F =
(L^2*q)/2
(L^3*q)/3
a =
(L*q)/P
-q/(2*P)
This yields the approximate solution as
 2.8 Stiffness matrix and load vector 25

>> w=simplify(N1*a(1)+N2*a(2))
w =
(q*x*(2*L - x))/(2*P)
and since we obtained a quadratic polynomial we may begin to suspect that we arrived at the exact
solution. Firstly we check the balance equation
>> P*diff(diff(w))+q
ans =
0
and secondly we check the boundary condition at x = L
>> subs(diff(w),x,L)
ans =
0
Since those equations are satisfied, we have verified that the approximate solution is in fact exact.

Exercise 10.
Solve for the approximate deflection of a prestressed table without any distributed load, sim-
ply supported at x = 0 and nonzero concentrated force boundary condition at x = L, using
the Galerkin method. Take as the trial function basis the two functions N1 (x) = sin(πx/L) and
N2 (x) = (3/2) sin(πx/L). The test functions are Nj , j = 1, 2.
Solution: We are solving the boundary value problem:
P w′′ = 0
w(0) = 0
P w′ (L) = FL
The trial function is

w(x) = a1 N1 (x) + a2 N2 (x)

Note that the trial function satisfies the essential boundary condition w(0) = 0 because

N1 (0) = N2 (0) = 0

The coefficients a1 , a2 are the unknowns.

Equation (2.14) simplifies for statics (zero accelerations) and a natural boundary conditions at
x = L to
Z L Z L
∂ηj ∂w
ηj (L)FL − P dx + ηj (x)q(x) dx = 0 .
0 ∂x ∂x 0

However, this time FL 6= 0 and there will be a contribution of the first term in the final equations.
Also note that the two test functions Nj , j = 1, 2 must not be simultaneously zero at x = L. If
that was the case, all effects of the applied force FL would be erased from the formulation, and we
couldn’t possibly get a meaningful result.
Substituting the test and trial function we have
Z L Z L
∂N1 ∂N1 ∂N1 ∂N2
FL N1 (L) − P a1 dx − P a2 dx = 0 ,
Z0 L ∂x ∂x Z0 L ∂x ∂x
∂N2 ∂N1 ∂N2 ∂N2
FL N2 (L) − P a1 dx − P a2 dx = 0 ,
0 ∂x ∂x 0 ∂x ∂x
The Matlab symbolic algebra reads
26 Thermal and Stress Analysis with the FEM

>> syms L P q x FL real

q=0;
N1=sin(pi*x/L);
N2=sin(3/2*pi*x/L);
K=int(diff( [N1;N2])*P*diff([N1,N2]),0,L)
F=q*int([N1;N2],0,L)+ [FL*subs(N1,x,L);FL*subs(N2,x,L)]
a=K\F
K =
[ (P*pi^2)/(2*L), (9*pi*P)/(5*L)]
[ (9*pi*P)/(5*L), (9*P*pi^2)/(8*L)]
F =
0
-FL
a =
(80*FL*L)/(P*pi*(25*pi^2 - 144))
-(200*FL*L)/(9*P*(25*pi^2 - 144))
This yields the approximate solution as shown in the figure below.

−0.05

0.05
(P/FL/L)w

0.1

0.15

0.2

0.25

0.3
0 0.2 0.4 0.6 0.8 1
ξ

While the approximate solution is computed correctly, it is far from satisfactory. The curvature of the
cable is clearly nonzero, and that contradicts the balance equation which gives zero curvature when
the transverse load is zero. The force balance at the right hand side end of the cable is definitely not
satisfied: the force FL points downward, and so does the prestressing force P – they cannot balance.
The bad features of the approximate solution are the result of the choice of the basis functions.
Sinusoidal curves are wrong for the present purpose!

2.9 Piecewise linear basis functions

Let us recall the piecewise linear approximation proposed for the trial function in Section 2.4. The
broken line cannot be represented as a linear combination of linear functions that are all defined on
the whole interval 0 ≤ x ≤ L (only two such functions are linearly independent, and these functions
cannot represent the corners in the broken line). Therefore, we have to describe the piecewise linear
curve interval by interval.
The interpolant may be written as a linear combination of basis functions. In one dimension,
the piecewise linear basis function is called the hat function. The six functions that are shown in
Fig. 2.5, all are examples of hat functions. For reasons that will be discussed later, we would want
the hat functions in a linear combination to be able to reproduce an arbitrary linear function over
 2.9 Piecewise linear basis functions 27

the whole interval. Because of the way in which we construct the hat functions in Fig. 2.5, this
property is automatically available.

Fig. 2.5. Piecewise linear basis functions

Let us describe the construction of the piecewise linear basis functions. First, the length of the
wire is divided into disjoint subintervals. These subintervals are the finite elements for the one-
dimensional domain. The end-points of the finite elements are called nodes. Together, the finite
elements and the nodes are known as the finite element mesh: see Fig. 2.6 (the element numbers
are in the boxes; nodes are indicated by filled circles). In this book, the one-dimensional elements
with two nodes at the end points are going to be referred to as L2.
We are going to construct the basis functions so that they are piecewise linear, and they assume
values that are zero at all nodes of the mesh except one. We will associate basis functions with nodes
so that when a basis function is nonzero at node K the function will be called NK . We say basis
function NK is associated with node K.
Since all basis functions are constructed in the same way, we describe the procedure for the basis
function N3 (i.e. associated with node 3): as shown in the Fig. 2.5, it is nonzero over two elements,
2 and 3; zero everywhere else. To be able to write it down over the two adjacent elements, we have
to agree on the value of N3 at node 3 (i.e. N3 (x3 )), which is shared by elements 2 and 3. Choosing
N3 (x3 ) = 1 has certain advantages, which will be introduced momentarily. Using the concept of
Lagrange interpolation polynomials, we may write the function N3 within element 2 as
x − x2
N3 (x) = , x2 ≤ x ≤ x3 ,
x3 − x2
and within element 3 as
x − x4
N3 (x) = , x3 ≤ x ≤ x4 .
x3 − x4
Note that the basis function are “non-dimensional”: the formula above says length/length= no
dimension (we may say that the physical units are [ND])!
All the other functions Ni are expressed analogously. Putting them together in a linear combi-
nation for the trial function, we write
N
X
w(x) = Ni (x)wi , (2.22)
i=1

(for simplicity, we omit the time argument). The physical unit of the deflection w(x) is [length], and
since the basis functions themselves are non-dimensional, the degrees of freedom wi must have the
physical units of [length].
Evaluating w(x) at the node k, we obtain
N
X
w(xk ) = Ni (xk )wi ,
i=1
28 Thermal and Stress Analysis with the FEM

where the crucial expression is Ni (xk ): by definition, the basis function Nk has value +1 at xk , while
all other functions Ni , i 6= k are zero at xk . This property is usually expressed mathematically as

Ni (xk ) = δik , (2.23)

where the symbol δik is known as the Kronecker delta


1, if i = k;
δik =
0, otherwise.

Because of this property, the value of w(xk ) is

N
X N
X
w(xk ) = Ni (xk )wi = δik wi = wk ,
i=1 i=1

and we see that the parameters wi have the physical meaning of the value of the interpolated
function at the node i. The wi ’s are usually called the degrees of freedom, since, being the control
parameters of the trial function, they determine the shape of the actual solution from all the possible
shapes of the trial function. They are the objects that our numerical method solves for.

Fig. 2.6. The finite element mesh. Two different but completely equivalent numberrings of the nodes and
elements. The top mesh consists of elements 1:(1,2), 2:(2,3), 3:(3,4), 4:(4,5), 5:(5,6). The bottom mesh consists
of elements 1:(2,5), 2:(6,1), 3:(4,3), 4:(3,2), 5:(1,4).

Exercise 11.
For the shown finite element mesh express the finite element basis functions and their derivatives
as expressions in the independent variable x. Associate the basis functions with the nodes j whose
locations are xj .

Solution: We shall use the so-called Lagrange interpolation polynomials to construct the segments
of the individual basis functions. All the segments of all the basis functions are linear functions. For
instance, the segment of the basis function N2 over the element 1, 2, which we will call N21,2 passes
through the points (x1 , 0) and (x2 , +1).

The linear function

(x − x1 )
 2.9 Piecewise linear basis functions 29

becomes zero at x = x1 . It is not equal to +1 at x = x2 . In order to enforce this property we

normalize by the value of the function (x − x1 ) at x = x2 to obtain

(x − x1 )
N21,2 (x) =
(x2 − x1 )

The segment of the basis function N2 over the element 2, 3, which we will call N22,3 passes through
the points (x2 , +1) and (x3 , 0).

To design this segment we realize that the linear function

(x − x3 )

becomes zero at x = x3 , and since it is not equal to +1 at x = x2 we normalize by the value of the
function (x − x3 ) at x = x2 to obtain

(x − x3 )
N22,3 (x) =
(x2 − x3 )

All four basis functions are summarized in this figure:

Note that a basis function is nonzero only over the two elements which share the node at which the
basis function assumes value +1.
The derivatives follow by simple differentiation of the segments of the basis functions with respect
to x. Note that the derivatives are piecewise constant as they measure the slope of the linear segments
of the basis functions.
30 Thermal and Stress Analysis with the FEM

Exercise 12.
Interpolate the function cos(πx/L) on the interval 0 ≤ x ≤ L using a mesh of five L2 (i.e. linear)
finite elements of equal length.

Solution: Interpolation of the given function on the mesh is understood as the task of constructing
a linear combination of the basis functions defined on the mesh
X
wh (x) = Nj (x)wj
j

so that the linear combination is equal to the interpolated function g(x) at the nodes.

Mathematically, we say

g(xk ) = wh (xk ) for all k

which we call the interpolating condition. The coefficients of the linear combination wj need to be
determined from the interpolating conditions. For the finite element basis functions this is a breeze
because of the Kronecker delta property

+1, if j = k;
Nj (xk ) =
0, otherwise.

This gives
X
wh (xk ) = Nj (xk )wj = N1 (xk )w1 + . . . + Nk (xk )wk + . . . + Nn (xk )wn = wk
| {z } | {z } | {z }
j
0 +1 0
 2.9 Piecewise linear basis functions 31

so that wk = g(xk ). For our given function g(x) we can therefore write the interpolation function as
X
wh (x) = Nj (x) cos(πxj /L)
j

and numerically the coefficients of the linear combination are

>> cos(pi*[0:0.2:1.0])
ans =
1.0000 0.8090 0.3090 -0.3090 -0.8090 -1.0000
The construction of the linear combination is depicted in this figure:

And here are the interpolated (solid line) and interpolating (dashed line) functions.

Exercise 13.
Interpolate the function g(x) = Ax2 + Bx + C on the interval 0 ≤ x ≤ h using a single L2 finite
element mesh. Discuss the interpolation error.
Solution: Interpolation of the given function on the finite element mesh is understood as a linear
combination of the basis functions defined on the mesh
X
wh (x) = Nj (x)wj
j

so that the linear combination is equal to the interpolated function g(x) at the nodes. For a single-
element mesh there are only two basis functions that are nonzero within the element. As shown in
exercise 12, using the interpolating conditions to determine the coefficients of the linear combination
we can write

wh (x) = N1 (x)w1 + N2 (x)w2

where
32 Thermal and Stress Analysis with the FEM

N1 (x) = (x − x2 )/(x1 − x2 ), N2 (x) = (x − x1 )/(x2 − x1 )

and

w1 = g(x1 ) = Ax21 + Bx1 + C , w2 = g(x2 ) = Ax22 + Bx2 + C

Since the two basis functions are linear, their linear combination will also be linear. The error is the
difference between the interpolated function g(x) and the interpolating (linear) function wh (x)
error

Evidently, the error is zero at the nodes. The graphics below shows how the two basis functions are
multiplied by the coefficients of the linear combination and added together to form wh (x).

Since the error is zero at the nodes (the interpolating conditions!), and since a linear function is
defined uniquely by two points, we may suspect that wh (x) can exactly (without error) interpolate
(or as we say “reproduce”) an arbitrary linear function. In fact a little bit of symbolic manipulation
confirms this: We form symbolically the two functions g(x) and wh (x)
>> syms A B C x1 x2 x real
g=A*x^2+B*x+C;
w1 =subs(g,x,x1);
w2 =subs(g,x,x2);
wh=(x-x2)/(x1-x2)*w1+(x-x1)/(x2-x1)*w2;
and then we compute the error
simplify(g-wh)
ans =
A*(x - x1)*(x - x2)
The error term is a quadratic expression in x, and it is zero when A = 0. In other words, when the
interpolated function is linear g(x) = Bx + C, the interpolation on the mesh is without error. That
makes sense: Note that the coefficient A is related to the curvature: g ′′ = 2A. A piecewise linear
function will not be able to match a curve with nonzero curvature.

Exercise 14.
 2.9 Piecewise linear basis functions 33

Interpolate the function g(x) = −x2 +1.3x+1/3 on a two-element mesh on the interval 0 ≤ x ≤ 1.
Discuss the approximation of the derivative by the interpolating function.
Solution: For variety we will set up the mesh as shown: elements 1:(3,2), 2:(1,3).

The interpolating function is constructed element-by-element. The basis functions are associated
with nodes, basis function Nj with node j.
Within the extent of element 1 the interpolating function is

wh (x) = N3 (x)w3 + N2 (x)w2

where

N3 (x) = (x − x2 )/(x3 − x2 ), N2 (x) = (x − x3 )/(x2 − x3 )

and

w3 = −(1/2)2 + 1.3(1/2) + 1/3 = 22/30 , w2 = −(1)2 + 1.3(1) + 1/3 = 19/30

Within the extent of element 2 the interpolating function is

wh (x) = N1 (x)w1 + N3 (x)w3

where

N1 (x) = (x − x3 )/(x1 − x3 ), N3 (x) = (x − x1 )/(x3 − x1 )

and

w3 = −(1/2)2 + 1.3(1/2) + 1/3 = 22/30 , w1 = −(0)2 + 1.3(0) + 1/3 = 10/30

The interpolated and interpolating functions are shown here:

0.8
0.7
0.6
0.5
0.4
0.3

0.2
0.1

0
0 0.2 0.4 0.6 0.8 1

To compare their derivatives we just differentiate both the interpolated and interpolating function.
The derivative of the interpolated function is continuous, linear, g ′ (x) = −2x + 1.3. The derivative
of the interpolating function needs to be again computed element-by-element. Within the extent of
element 1 the interpolating function derivative is

wh′ (x) = N3′ (x)w3 + N2′ (x)w2

where

N3′ (x) = 1/(x3 − x2 ), N2′ (x) = 1/(x2 − x3 )

34 Thermal and Stress Analysis with the FEM

and the coefficients w3 , w2 are as computed above. Consequently we get

w2 − w3
wh′ (x) = = −1/5
x2 − x3
that is rise-over-run as expressed from the geometry of a straight line.
Within the extent of element 2 the derivative of the interpolating function is

wh′ (x) = N1′ (x)w1 + N3′ (x)w3

where

N1′ (x) = 1/(x1 − x3 ), N3′ (x) = 1/(x3 − x1 )

and w3 , w1 are as above. Substitution leads in element 2 to

w3 − w1
wh′ (x) = = 4/5
x3 − x1
Note well that the derivative lowered the order of the functions, the derivative of the quadratic inter-
polated function being linear, and the derivative of the piecewise linear interpolating function being
piecewise constant. As a consequence, the derivative of the interpolating function is discontinuous.
Therefore, taking the slope of the first derivative of wh′ (x) would be tricky. Usually the consequence
of negotiating the discontinuity at the node means following a positive or negative infinite slope.

1.5

0.5

−0.5

−1
0 0.2 0.4 0.6 0.8 1

2.10 Bookkeeping in the finite element method

In order to automate processing of the information in a finite element model we introduce some
bookkeeping devices.
All nodes are numbered sequentially from 1 to N , but in an arbitrary order. All elements connect
some nodes. In the cable problem, each element connects two nodes. The so-called element connec-
tivity are the numbers of the nodes, given in the order left to right (in other words, the nodes are
given in the order in which their coordinates increase).
Perhaps most importantly, all the degree of freedom parameters in the model will be numbered
sequentially. First the free degrees of freedom will be numbered (those degrees of freedom that
are unknown and need to be solved for), and only then the degrees of freedom that are prescribed
(known from the boundary conditions) will be numbered. Since eventually all the parameters are
associated with equations in which the parameters occur we also refer to this numbering as the
“equation numbering”.
 2.10 Bookkeeping in the finite element method 35

Figure 2.7 illustrates the bookkeeping information. Node numbers are in circles. Element numbers
are in boxes. Element 1 connects nodes 1 and 2, element 2 connects nodes 2 and 3. There are two
free degrees of freedom, 1 and 2. The deflections w1 , w2 are unknown. There is one prescribed degree
of freedom, w3 , which is given by the pinned boundary condition as w3 = 0.
The basis functions are always associated with nodes: basis function N1 with node 1 and so on.
On the other hand, the numbering of the unknowns does not necessarily correspond to the numbering
of the nodes. Therefore we resort to the following notation: the finite element expansion is written
as
N
X
wh (x) = Nj (x)w(j) (2.24)
j=1

where by w(j) we mean the unknown at node j. For the mesh in Figure 2.7 the finite element
expansion becomes

wh (x) = N1 (x)w(1) + N2 (x)w(2) + N3 (x)w(3) = N1 (x)w3 + N2 (x)w1 + N3 (x)w2

which corresponds to the degree of freedom 3 being at node 1 (so (1) ↔ 3), and so on (i.e. (2) ↔ 1,
(3) ↔ 2).
We might also need to find the node at which a given degree of freedom resides. We will use the
notation hji to mean the node with the degree of freedom j. For the mesh in Figure 2.7 we
have h1i ↔ 2, h2i ↔ 3, and h3i ↔ 1.

Fig. 2.7. Finite element mesh with bookkeeping information

Next we will introduce the finite element trial expansion (2.24) into the Galerkin method by
using it for the trial function, and by setting the test functions to be equal to the basis functions
themselves. We shall start with an example. We will solve for the deflection of the shown prestressed
cable using the finite element Galerkin method (see Figure 2.8). We will use a mesh of two L2 finite
elements of equal length.

Fig. 2.8. Prestressed cable configuration

The Galerkin weighted residual statement for the BVP for the static deflection of the pin-roller
cable is written as shown in equation (2.14) (the first term is eliminated since FL is zero)
Z L Z L
− ηj ′ P w′ dx + ηj q dx = 0, j = 1, ..., Nf ,
0 0

where we require for the test and trial function

36 Thermal and Stress Analysis with the FEM

ηj (x = 0) = 0, ηj ∈ C 0 , j = 1, ..., Nf ,
0
w(x = 0) = w̄0 , w∈C ,

and Nf is the number of free degrees of freedom.

The mesh consists of two L2 elements
Element Nodes
1 1,2
2 2,3

The degrees of freedom will be also numbered. We start the numbering with the unknowns
(deflections at nodes 2, 3), and we conclude the numbering with the deflection at node 1 which
is known. Sometimes we refer to the numbers of degrees of freedom interchangeably as “equation
numbers”.
Node Degree of freedom (equation) number
1 3
2 1
3 2
The basis (hat) functions are associated with the nodes.

The trial function is written as

3
X
w(x) = Nk (x)w(k)
k=1

where (k) means equation number (dof number) associated with node k, and therefore w(k) means
the deflection at node k. For instance, (1) = 3, and w(1) = w3 . Note however that it may also be
written as
3
X
w(x) = Nhki (x)wk
k=1

where k means equation member, which is associated with node hki, and therefore wk means the
deflection at node hki. In the first form of the trial function the sum was over all the nodes, whereas
in the second form the sum is over all the degrees of freedom.
As given in the BVP (2.14), (2.15), the trial function must satisfy the essential boundary
condition. Thus we a require
 2.10 Bookkeeping in the finite element method 37
3
X
w(0) = Nk (0)w(k) = 0
k=1

Since we have N1 (0) = 1 and N2 (0) = N3 (0) = 0 this condition is equivalent to

w(0) = w(1) = w3 = 0

We see that the essential boundary condition determines the value of the prescribed degree of freedom
at the pin w3 = 0.
We are using the finite element Galerkin method, hence the test functions are taken to be the hat
functions Nj on the mesh. The test functions must satisfy the condition ηj (x = 0) = 0, which means
that only N2 and N3 are allowed since they are both zero at the left-hand side end of the wire. Thus
we take η1 = N2 and η2 = N3 . Because the test functions are from now on always going to be the
finite element basis functions, we may just as well start writing the weighted residual equation as
Z L Z L
− Nhji ′ P w′ dx + Nhji q dx = 0, j = 1, ..., Nf ,
0 0

where Nhji is the basis function associated with node number hji which carries the degree of freedom
j.
The elements of the load vector Lj are now computed for j = 1, 2. We begin with L1 : first we see
that the integral should be split into integrals over each element, since the test function Nh1i = N2
has different definitions in different elements.
Z L Z x2 Z x3
L1 = Nh1i q dx = Nh1i q dx + Nh1i q dx
0 x1 x2

For element 1 we compute the contribution to L1 as

Z x2 Z x2
Nh1i q dx = N2 q dx = qL/4
x1 x1

and for element 2 we compute the contribution to L1 as

Z x3 Z x3
Nh1i q dx = N2 q dx = qL/4
x2 x2

yielding L1 = qL/2.
The load vector component L2 is computed as
Z L Z x2 Z x3
L2 = Nh2i q dx = Nh2i q dx + Nh2i q dx
0 x1 x2

were the contribution from element 1 is zero, since Nh2i = N3 = 0 in element 1, and the contribution
to L2 from element 2 is
Z x3
Nh2i q dx = qL/4
x2

The load vector is therefore

 
qL/2
[L] =
qL/4

The components ofPthe stiffness matrix are computed next. Substituting the second form of the trial
3
function, w(x) = k=1 Nhki (x)wk , into the weighted residual equation we obtain for the stiffness
term
Discovering Diverse Content Through
Random Scribd Documents
 NOTE THE NEW TITLES LISTED
Western Story Library
For Everyone Who Likes Adventure
Ted Strong and his band of broncho-busters have most exciting
adventures in this line of attractive big books, and furnish the reader
with an almost unlimited number of thrills.
If you like a really good Western cowboy story, then this line is
made expressly for you.
ALL TITLES ALWAYS IN PRINT

1—Ted Strong, Cowboy By Edward C. Taylor

2—Ted Strong Among the Cattlemen By Edward C. Taylor
3—Ted Strong’s Black Mountain Ranch By Edward C. Taylor
4—Ted Strong With Rifle and Lasso By Edward C. Taylor
5—Ted Strong Lost in the Desert By Edward C. Taylor
6—Ted Strong Fighting the Rustlers By Edward C. Taylor
7—Ted Strong and the Rival Miners By Edward C. Taylor
9—Ted Strong on a Mountain Trail By Edward C. Taylor
10—Ted Strong Across the Prairie By Edward C. Taylor
11—Ted Strong Out For Big Game By Edward C. Taylor
12—Ted Strong, Challenged By Edward C. Taylor
13—Ted Strong’s Close Call By Edward C. Taylor
14—Ted Strong’s Passport By Edward C. Taylor
15—Ted Strong’s Nebraska Ranch By Edward C. Taylor
16—Ted Strong’s Cattle Drive By Edward C. Taylor
17—Ted Strong’s Stampede By Edward C. Taylor
18—Ted Strong’s Prairie Trail By Edward C. Taylor
19—Ted Strong’s Surprise By Edward C. Taylor
20—Ted Strong’s Wolf Hunters By Edward C. Taylor
22—Ted Strong in Colorado By Edward C. Taylor
25—Ted Strong’s Search By Edward C. Taylor
26—Ted Strong’s Diamond Mine By Edward C. Taylor
27—Ted Strong’s Manful Task By Edward C. Taylor
 ROMANCE
ADVENTURE
MYSTERY
All types of stories are represented in this catalogue. The S & S
novels are the world’s greatest entertainment at a price that truly
entitles them to be termed:
THE RIGHT BOOKS AT THE RIGHT PRICE
 NICK CARTER STORIES
New Magnet Library
Not a Dull Book in This List
ALL BY NICHOLAS CARTER
Nick Carter stands for an interesting detective story. The fact that
the books in this line are so uniformly good is entirely due to the
work of a specialist. The man who wrote these stories produced no
other type of fiction. His mind was concentrated upon the creation of
new plots and situations in which his hero emerged triumphantly
from all sorts of troubles and landed the criminal just where he
should be—behind the bars.
The author of these stories knew more about writing detective
stories than any other single person.
Following is a list of the best Nick Carter stories. They have been
selected with extreme care, and we unhesitatingly recommend each
of them as being fully as interesting as any detective story between
cloth covers which sells at ten times the price.
If you do not know Nick Carter, buy a copy of any of the New
Magnet Library books, and get acquainted. He will surprise and
delight you.
ALL TITLES ALWAYS IN PRINT

850—Wanted: A Clew
851—A Tangled Skein
852—The Bullion Mystery
853—The Man of Riddles
854—A Miscarriage of Justice
855—The Gloved Hand
856—Spoilers and the Spoils
857—The Deeper Game
858—Bolts from Blue Skies
859—Unseen Foes
860—Knaves in High Places
861—The Microbe of Crime
862—In the Toils of Fear
863—A Heritage of Trouble
864—Called to Account
865—The Just and the Unjust
866—Instinct at Fault
867—A Rogue Worth Trapping
868—A Rope of Slender Threads
869—The Last Call
870—The Spoils of Chance
871—A Struggle with Destiny
872—The Slave of Crime
873—The Crook’s Blind
874—A Rascal of Quality
875—With Shackles of Fire
876—The Man Who Changed Faces
877—The Fixed Alibi
878—Out with the Tide
879—The Soul Destroyers
880—The Wages of Rascality
881—Birds of Prey
882—When Destruction Threatens
883—The Keeper of Black Hounds
884—The Door of Doubt
885—The Wolf Within
886—A Perilous Parole
887—The Trail of the Finger Prints
888—Dodging the Law
889—A Crime in Paradise
890—On the Ragged Edge
891—The Red God of Tragedy
892—The Man Who Paid
893—The Blind Man’s Daughter
894—One Object in Life
895—As a Crook Sows
896—In Record Time
897—Held in Suspense
898—The $100,000 Kiss
899—Just One Slip
900—On a Million-dollar Trail
901—A Weird Treasure
902—The Middle Link
903—To the Ends of the Earth
904—When Honors Pall
905—The Yellow Brand
906—A New Serpent in Eden
907—When Brave Men Tremble
908—A Test of Courage
909—Where Peril Beckons
910—The Gargoni Girdle
911—Rascals & Co.
912—Too Late to Talk
913—Satan’s Apt Pupil
914—The Girl Prisoner
915—The Danger of Folly
916—One Shipwreck Too Many
917—Scourged by Fear
918—The Red Plague
919—Scoundrels Rampant
920—From Clew to Clew
921—When Rogues Conspire
922—Twelve In a Grave
923—The Great Opium Case
924—A Conspiracy of Rumors
925—A Klondike Claim
926—The Evil Formula
927—The Man of Many Faces
928—The Great Enigma
929—The Burden of Proof
930—The Stolen Brain
931—A Titled Counterfeiter
932—The Magic Necklace
933—’Round the World for a Quarter
934—Over the Edge of the World
935—In the Grip of Fate
936—The Case of Many Clews
937—The Sealed Door
938—Nick Carter and the Green Goods Men
939—The Man Without a Will
940—Tracked Across the Atlantic
941—A Clew from the Unknown
942—The Crime of a Countess
943—A Mixed-up Mess
944—The Great Money-order Swindle
945—The Adder’s Brood
946—A Wall Street Haul
947—For a Pawned Crown
948—Scaled Orders
949—The Hate that Kills
950—The American Marquis
951—The Needy Nine
952—Fighting Against Millions
953—Outlaws of the Blue
954—The Old Detective’s Pupil
955—Found in the Jungle
956—The Mysterious Mail Robbery
957—Broken Bars
958—A Fair Criminal
959—Won by Magic
960—The Plano Box Mystery
961—The Man They Held Back
962—A Millionaire Partner
963—A Pressing Peril
964—An Australian Klondike
965—The Sultan’s Pearls
966—The Double Shuffle Club
967—Paying the Price
968—A Woman’s Hand
969—A Network of Crime
 970—At Thompson’s Ranch
971—The Crossed Needles
972—The Diamond Mine Case
973—Blood Will Tell
974—An Accidental Password
975—The Crook’s Double
976—Two Plus Two
977—The Yellow Label
978—The Clever Celestial
979—The Amphitheater Plot
980—Gideon Drexel’s Millions
981—Death in Life
982—A Stolen Identity
983—Evidence by Telephone
984—The Twelve Tin Boxes
985—Clew Against Clew
986—Lady Velvet
987—Playing a Bold Game
988—A Dead Man’s Grip
989—Snarled Identities
990—A Deposit Vault Puzzle
991—The Crescent Brotherhood
992—The Stolen Pay Train
993—The Sea Fox
994—Wanted by Two Clients
995—The Van Alstine Case
996—Check No. 777
997—Partners in Peril
998—Nick Carter’s Clever Protégé
999—The Sign of the Crossed Knives
1000—The Man Who Vanished
1001—A Battle for the Right
1002—A Game of Craft
1003—Nick Carter’s Retainer
1004—Caught in the Toils
1005—A Broken Bond
1006—The Crime of the French Café
1007—The Man Who Stole Millions
1008—The Twelve Wise Men
1009—Hidden Foes
1010—A Gamblers’ Syndicate
1011—A Chance Discovery
1012—Among the Counterfeiters
1013—A Threefold Disappearance
1014—At Odds with Scotland Yard
1015—A Princess of Crime
1016—Found on the Beach
1017—A Spinner of Death
1018—The Detective’s Pretty Neighbor
1019—A Bogus Clew
1020—The Puzzle of Five Pistols
1021—The Secret of the Marble Mantel
1022—A Bite of an Apple
1023—A Triple Crime
1024—The Stolen Race Horse
1025—Wildfire
1026—A Herald Personal
1027—The Finger of Suspicion
1028—The Crimson Clew
1029—Nick Carter Down East
1030—The Chain of Clews
1031—A Victim of Circumstances
1032—Brought to Bay
1033—The Dynamite Trap
1034—A Scrap of Black Lace
1035—The Woman of Evil
1036—A Legacy of Hate
1037—A Trusted Rogue
1038—Man Against Man
1039—The Demons of the Night
1040—The Brotherhood of Death
1041—At the Knife’s Point
1042—A Cry for Help
1043—A Stroke of Policy
1044—Hounded to Death
1045—A Bargain in Crime
1046—The Fatal Prescription
1047—The Man of Iron
1048—An Amazing Scoundrel
1049—The Chain of Evidence
1050—Paid with Death
1051—A Fight for a Throne
1052—The Woman of Steel
1053—The Seal of Death
1054—The Human Fiend
1055—A Desperate Chance
1056—A Chase in the Dark
1057—The Snare and the Game
1058—The Murray Hill Mystery
1059—Nick Carter’s Close Call
1060—The Missing Cotton King
1061—A Game of Plots
1062—The Prince of Liars
1063—The Man at the Window
1064—The Red League
1065—The Price of a Secret
1066—The Worst Case on Record
1067—From Peril to Peril
1068—The Seal of Silence
1069—Nick Carter’s Chinese Puzzle
1070—A Blackmailer’s Bluff
1071—Heard in the Dark
1072—A Checkmated Scoundrel
1073—The Cashier’s Secret
1074—Behind a Mask
1075—The Cloak of Guilt
1076—Two Villains in One
1077—The Hot Air Clew
1078—Run to Earth
1079—The Certified Check
1080—Weaving the Web
1081—Beyond Pursuit
1082—The Claws of the Tiger
1083—Driven from Cover
1084—A Deal in Diamonds
1085—The Wizard of the Cue
1086—A Race for Ten Thousand
1087—The Criminal Link
1088—The Red Signal
1089—The Secret Panel
1090—A Bonded Villain
1091—A Move in the Dark
1092—Against Desperate Odds
1093—The Telltale Photographs
1094—The Ruby Pin
1095—The Queen of Diamonds
1096—A Broken Trail
1097—An Ingenious Stratagem
1098—A Sharper’s Downfall
1099—A Race Track Gamble
1100—Without a Clew
1101—The Council of Death
1102—The Hole in the Vault
1103—In Death’s Grip
1104—A Great Conspiracy
1105—The Guilty Governor
1106—A Ring of Rascals
1107—A Masterpiece of Crime
1108—A Blow for Vengeance
1109—Tangled Threads
1110—The Crime of the Camera
1111—The Sign of the Dagger
1112—Nick Carter’s Promise
1113—Marked for Death
1114—The Limited Holdup
1115—When the Trap Was Sprung
1116—Through the Cellar Wall
1117—Under the Tiger’s Claws
1118—The Girl in the Case
1119—Behind a Throne
1120—The Lure of Gold
1121—Hand to Hand
1122—From a Prison Cell
1123—Dr. Quartz, Magician
1124—Into Nick Carter’s Web
1125—The Mystic Diagram
1126—The Hand that Won
1127—Playing a Lone Hand
1128—The Master Villain
1129—The False Claimant
1130—The Living Mask
1131—The Crime and the Motive
1132—A Mysterious Foe
1133—A Missing Man
1134—A Game Well Played
1135—A Cigarette Clew
1136—The Diamond Trail
1137—The Silent Guardian
1138—The Dead Stranger
1140—The Doctor’s Stratagem
1141—Following a Chance Clew
1142—The Bank Draft Puzzle
1143—The Price of Treachery
1144—The Silent Partner
1145—Ahead of the Game
1146—A Trap of Tangled Wire
1147—In the Gloom of Night
1148—The Unaccountable Crook
1149—A Bundle of Clews
1150—The Great Diamond Syndicate
1151—The Death Circle
1152—The Toss of a Penny
1153—One Step Too Far
1154—The Terrible Thirteen
1155—A Detective’s Theory
1156—Nick Carter’s Auto Trail
1157—A Triple Identity
1158—A Mysterious Graft
1159—A Carnival of Crime
1160—The Bloodstone Terror
1161—Trapped in His Own Net
1162—The Last Move in the Game
1163—A Victim of Deceit
1164—With Links of Steel
1165—A Plaything of Fate
1166—The Key Ring Clew
1167—Playing for a Fortune
1168—At Mystery’s Threshold
1169—Trapped by a Woman
1170—The Four Fingered Glove
1171—Nabob and Knave
1172—The Broadway Cross
1173—The Man Without a Conscience
1174—A Master of Deviltry
1175—Nick Carter’s Double Catch
1176—Doctor Quartz’s Quick Move
1177—The Vial of Death
1178—Nick Carter’s Star Pupils
1179—Nick Carter’s Girl Detective
1180—A Baffled Oath
1181—A Royal Thief
1182—Down and Out
1183—A Syndicate of Rascals
1184—Played to a Finish
1185—A Tangled Case
1186—In Letters of Fire
1187—Crossed Wires
1188—A Plot Uncovered
1189—The Cab Driver’s Secret
1190—Nick Carter’s Death Warrant
1191—The Plot that Failed
1192—Nick Carter’s Masterpiece
1193—A Prince of Rogues
1194—In the Lap of Danger
1195—The Man from London
1196—Circumstantial Evidence
1197—The Pretty Stenographer Mystery
1198—A Villainous Scheme
1199—A Plot Within a Plot
1200—The Elevated Railroad Mystery
1201—The Blow of a Hammer
1202—The Twin Mystery
1203—The Bottle with the Black Label
1204—Under False Colors
1205—A Ring of Dust
1206—The Crown Diamond
1207—The Blood-red Badge
1208—The Barrel Mystery
1209—The Photographer’s Evidence
1210—Millions at Stake
1211—The Man and His Price
1212—A Double-Handed Game
1213—A Strike for Freedom
1214—A Disciple of Satan
1215—The Marked Hand
1216—A Fight with a Fiend
1217—When the Wicked Prosper
1218—A Plunge into Crime
1219—An Artful Schemer
1220—Reaping the Whirlwind
1221—Out of Crime’s Depths
1222—A Woman at Bay
 1223—The Temple of Vice
1224—Death at the Feast
1225—A Double Plot

In order that there may be no confusion, we desire to say that the

books listed below will be issued during the respective months in
New York City and vicinity. They may not reach the readers at a
distance promptly, on account of delays in transportation.
To be published in January, 1928.

1226—In Search of Himself

1227—A Hunter of Men

To be published in February, 1928.

1228—The Boulevard Mutes

1229—Captain Sparkle, Pirate
1230—Nick Carter’s Fall

To be published in March, 1928.

1231—Out of Death’s Shadow

1232—A Voice from the Past

To be published in April, 1928.

1233—Accident or Murder?
1234—The Man Who Was Cursed

To be published in May, 1928.

1235—Baffled, But Not Beaten

1236—A Case Without a Clew

To be published in June, 1928.

1237—The Demon’s Eye

1238—A Blindfold Mystery
 BOOKS THAT NEVER GROW OLD
Alger Series
Clean Adventure Stories for Boys
The Most Complete List Published
The following list does not contain all the books that Horatio Alger
wrote, but it contains most of them, and certainly the best.
Horatio Alger is to boys what Charles Dickens is to grown-ups. His
work is just as popular to-day as it was years ago. The books have a
quality, the value of which is beyond computation.
There are legions of boys of foreign parents who are being helped
along the road to true Americanism by reading these books which
are so peculiarly American in tone that the reader cannot fail to
absorb some of the spirit of fair play and clean living which is so
characteristically American.
In this list will be included certain books by Edward Stratemeyer,
Oliver Optic, and other authors who wrote the Alger type of stories,
which are equal in interest and wholesomeness with those written by
the famous author after which this great line of books for boys is
named.
ALL TITLES ALWAYS IN PRINT

By HORATIO ALGER, Jr.

12—Chester Rand
13—Grit, the Young Boatman of Pine Point
14—Joe’s Luck
15—From Farm Boy to Senator
16—The Young Outlaw
17—Jack’s Ward
18—Dean Dunham
19—In a New World
20—Both Sides of the Continent
22—Brave and Bold
24—Bob Burton
26—Julius, the Street Boy
28—Tom Brace
29—Struggling Upward
31—Tom Tracy
32—The Young Acrobat
33—Bound to Rise
34—Hector’s Inheritance
35—Do and Dare
36—The Tin Box
37—Tom, the Bootblack
38—Risen from the Ranks
39—Shifting for Himself
40—Wait and Hope
41—Sam’s Chance
42—Striving for Fortune
43—Phil, the Fiddler
44—Slow and Sure
45—Walter Sherwood’s Probation
47—The Young Salesman
48—Andy Grant’s Pluck
49—Facing the World
50—Luke Walton
51—Strive and Succeed
52—From Canal Boy to President
53—The Erie Train Boy
54—Paul, the Peddler
55—The Young Miner
56—Charlie Codman’s Cruise
57—A Debt of Honor
58—The Young Explorer
59—Ben’s Nugget
62—Frank Hunter’s Peril
64—Tom Thatcher’s Fortune
 65—Tom Turner’s Legacy
66—Dan, the Newsboy
67—Digging for Gold
69—In Search of Treasure
70—Frank’s Campaign
71—Bernard Brook’s Adventures
73—Paul Prescott’s Charge
74—Mark Manning’s Mission
76—Sink or Swim
77—The Backwoods Boy
78—Tom Temple’s Career
79—Ben Bruce
80—The Young Musician
81—The Telegraph Boy
82—Work and Win
84—The Cash Boy
85—Herbert Carter’s Legacy
86—Strong and Steady
87—Lost at Sea
89—Young Captain Jack
90—Joe, the Hotel Boy
91—Out for Business
92—Falling in with Fortune
93—Nelson, the Newsboy
94—Randy of the River
96—Ben Logan’s Triumph
97—The Young Book Agent
168—Luck and Pluck
169—Ragged Dick
170—Fame and Fortune
171—Mark, the Match Boy
172—Rough and Ready
173—Ben, the Luggage Boy
174—Rufus and Rose
 By EDWARD STRATEMEYER

98—The Last Cruise of The Spitfire

99—Reuben Stone’s Discovery
100—True to Himself
101—Richard Dare’s Venture
102—Oliver Bright’s Search
103—To Alaska for Gold
104—The Young Auctioneer
105—Bound to Be an Electrician
106—Shorthand Tom
108—Joe, the Surveyor
109—Larry, the Wanderer
110—The Young Ranchman
111—The Young Lumberman
112—The Young Explorers
113—Boys of the Wilderness
114—Boys of the Great Northwest
115—Boys of the Gold Field
116—For His Country
117—Comrades in Peril
118—The Young Pearl Hunters
119—The Young Bandmaster
121—On Fortune’s Trail
122—Lost in the Land of Ice
123—Bob, the Photographer

By OLIVER OPTIC

124—Among the Missing

125—His Own Helper
126—Honest Kit Dunstable
127—Every Inch a Boy
128—The Young Pilot
129—Always in Luck
130—Rich and Humble
131—In School and Out
133—Work and Win
135—Haste and Waste
136—Royal Tarr’s Pluck
137—The Prisoners of the Cave
138—Louis Chiswick’s Mission
139—The Professor’s Son
140—The Young Hermit
141—The Cruise of The Dandy
142—Building Himself Up
143—Lyon Hart’s Heroism
144—Three Young Silver Kings
145—Making a Man of Himself
146—Striving for His Own
147—Through by Daylight
148—Lightning Express
149—On Time
150—Switch Off
151—Brake Up
152—Bear and Forbear
153—The “Starry Flag”
154—Breaking Away
155—Seek and Find
156—Freaks of Fortune
157—Make or Break
158—Down the River
159—The Boat Club
160—All Aboard
161—Now or Never
162—Try Again
163—Poor and Proud
164—Little by Little
165—The Sailor Boy
 166—The Yankee Middy
167—Brave Old Salt

175—Fighting for Fortune By Roy Franklin

176—The Young Steel Worker By Frank H. MacDougal
177—The Go-ahead Boys By Gale Richards
178—For the Right By Roy Franklin
179—The Motor Cycle Boys By Donald Grayson
180—The Wall Street Boy By Allan Montgomery
181—Stemming the Tide By Roy Franklin
182—On High Gear By Donald Grayson
183—A Wall Street Fortune By Allan Montgomery
184—Winning by Courage By Roy Franklin
185—From Auto to Airship By Donald Grayson
186—Camp and Canoe By Remson Douglas
187—Winning Against Odds By Roy Franklin
188—The Luck of Vance Sevier By Frederick Gibson
189—The Island Castaway By Roy Franklin
190—The Boy Marvel By Frank H. MacDougal
191—A Boy With a Purpose By Roy Franklin
192—The River Fugitives By Remson Douglas
193—Out For a Fortune By Roy Franklin
194—The Boy Horse Owner By Frederick Gibson
195—Always on Deck By Roy Franklin
196—Paul Hassard’s Peril By Matt Royal
197—His Own Master By Roy Franklin
198—When Courage Wins By Edward S. Ellis
199—Bound to Get There By Roy Franklin
200—Who Was Milton Marr? By Frederick Gibson
201—The Lost Mine By Roy Franklin
202—Larry Borden’s Redemption By Emerson Baker
 EVERY BOY
Knows

FRANK MERRIWELL
No other fiction character is half so well known.
Why? Well the books tell why in no uncertain
manner
 BOOKS OF QUALITY
Select Library
Big, Popular Standards
This line is truly named. It is Select because each title in it has
been selected with great care from among hundreds of books by
well-known authors.
A glance over the following list will show the names of Mary J.
Holmes, Marie Corelli, Rider Haggard, “The Duchess,” R. D.
Blackmore, and translations of some of the more famous French
authors, like Victor Hugo and Alphonse Daudet.
If you are looking for books which will add to your knowledge of
literature, a complete set of the Select Library, which is so
reasonably priced, will do more for you than a like amount expended
on ordinary fiction between cloth covers.
ALL TITLES ALWAYS IN PRINT

1—Cousin Maude By Mary J. Holmes

2—Rosamond Leyton By Mary J. Holmes
6—Beulah By Augusta J. Evans
10—The Homestead on the Hillside By Mary J. Holmes
14—East Lynne By Mrs. Henry Wood
16—A Romance of Two Worlds By Marie Corelli
17—Cleopatra By H. Rider Haggard
18—Maggie Miller By Mary J. Holmes
27—Under Two Flags By “Ouida”
28—Dora Deane By Mary J. Holmes
29—Ardath. Vol. I By Marie Corelli
30—Ardath. Vol. II By Marie Corelli
31—The Light That Failed By Rudyard Kipling
32—Tempest and Sunshine By Mary J. Holmes
35—Inez By Augusta J. Evans
36—Phyllis By “The Duchess”
42—Vendetta By Marie Corelli
43—Sapho By Alphonse Daudet
 Round the World Library
Stories of Jack Harkaway and His Comrades
Every reader, young and old, has heard of Jack Harkaway. His
remarkable adventures in out-of-the-way corners of the globe are
really classics, and every one should read them.
Jack is a splendid, manly character, full of life and strength and
curiosity. He has a number of very interesting companions—
Professor Mole, for instance, who is very funny. He also has some
very strange enemies, who are anything but funny.
Get interested in Jack. It will pay you.
ALL TITLES ALWAYS IN PRINT

By BRACEBRIDGE HEMYNG
1—Jack Harkaway’s School Days
2—Jack Harkaway’s Friends
3—Jack Harkaway After School Days
4—Jack Harkaway Afloat and Ashore
5—Jack Harkaway Among the Pirates
6—Jack Harkaway at Oxford
7—Jack Harkaway’s Struggles
8—Jack Harkaway’s Triumphs
9—Jack Harkaway Among the Brigands
10—Jack Harkaway’s Return
11—Jack Harkaway Around the World
12—Jack Harkaway’s Perils
13—Jack Harkaway in China
14—Jack Harkaway and the Red Dragon
15—Jack Harkaway’s Pluck
16—Jack Harkaway in Australia
17—Jack Harkaway and the Bushrangers
 18—Jack Harkaway’s Duel
19—Jack Harkaway and the Turks
20—Jack Harkaway in New York
21—Jack Harkaway Out West
22—Jack Harkaway Among the Indians
23—Jack Harkaway’s Cadet Days
24—Jack Harkaway in the Black Hills
25—Jack Harkaway in the Toils
26—Jack Harkaway’s Secret of Wealth
27—Jack Harkaway, Missing
28—Jack Harkaway and the Sacred Serpent
29—The Fool of the Family
30—Mischievous Matt
31—Mischievous Matt’s Pranks
32—Bob Fairplay Adrift
33—Bob Fairplay at Sea
34—The Boys of St. Aldates
35—Billy Barlow
36—Larry O’Keefe
37—Sam Sawbones
38—Too Fast to Last
39—Home Base
40—Spider and Stump
41—Out for Fun
42—Rob Rollalong, Sailor
43—Rob Rollalong in the Wilds

44—Phil, the Showman By Stanley Norris

45—Phil’s Rivals By Stanley Norris
46—Phil’s Pluck By Stanley Norris
47—Phil’s Triumph By Stanley Norris
48—From Circus to Fortune By Stanley Norris
49—A Gentleman Born By Stanley Norris
50—For His Friend’s Honor By Stanley Norris
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookgate.com