1
Glossary
2
Acknowledgement
This contribution has been prepared in the frame of the project MSM0021630519 Progressive
reliable and durable civil engineering structures.
GLOSSARY........................................................................................................................................................... 7
1 PRINCIPLE OF FINITE ELEMENT METHOD .................................................................................. 13
1.1 MATHEMATICAL DEFINITION OF FEM................................................................................................. 13
1.2 DEFORMATION VARIANT OF FEM USED IN PRACTICAL STATICS.......................................................... 14
1.3 MORE GENERAL FORM OF FEM........................................................................................................... 15
1.4 MATHEMATICAL FORMULATION OF BOUNDARY ELEMENT METHOD BEM........................................ 18
1.5 PARTIAL DISCRETISATION OF THE PROBLEM  FINITE LAYER METHOD ............................................... 18
1.6 IMPACTS OF CURRENT DIVISION OF LABOUR ON FEM IN PRAKTICE..................................................... 19
2 BASIC TERMS AND ALGORITHMS OF FINITE ELEMENT METHOD....................................... 22
2.1 EXPLANATION OF USED TERMINOLOGY............................................................................................... 22
2.2 EXPLANATION OF THE PROCEDURE ON AN EXAMPLE DEFORMATION VARIANT OF FEM...................... 23
2.3 INDIVIDUAL STEPS OF DEFORMATION VARIANT OF FEM..................................................................... 27
2.4 SPECIFICATION OF SELECTED OPERATIONS THAT ARE USEFUL TO UNDERSTAND FEM TERMS............. 29
2.5 PRINCIPLE OF VIRTUAL WORK APPLIED IN FEM PROGRAMS ................................................................ 36
2.6 MAIN OUTCOME FOR THE USERS OF FEM PROGRAMS.......................................................................... 41
2.6.1 Selecting the elements of the FEM analysis model ........................................................................ 41
2.6.2 Interpretation of FEM output data ................................................................................................ 45
3 PHYSICAL AXIOMS AND VARIATIONAL PRINCIPLES OF MORE COMPLEX FEM
PROBLEMS......................................................................................................................................................... 48
3.1 PRINCIPLES OF APPROXIMATION OF THE SOUGHT DISTRIBUTION OF A QUANTITY IN FEM................... 48
3.2 ELEMENTARY PRINCIPLES OF PHYSICAL NATURE OF FEM................................................................... 55
3.3 VARIATIONAL PRINCIPLES OF MECHANICAL PROBLEMS OF FEM......................................................... 59
3.3.1 Position of variational principles in mechanics ............................................................................ 59
3.3.2 Scalar, vector and tensor field in FEM inputs and outputs ........................................................... 62
3.3.3 General variational principle ........................................................................................................ 69
3.3.4 Consequences for some variants of FEM procedures.................................................................... 73
3.3.5 Special configurations used in FEM.............................................................................................. 76
3.3.6 Example of evaluation of elements by means of the variational principle..................................... 80
3.3.7 Buflers variational principles....................................................................................................... 81
3.3.8 Inverse variational principles........................................................................................................ 84
4 FINITE ELEMENTS................................................................................................................................. 85
4.1 GEOMETRIC PROPERTIES OF ELEMENTS ............................................................................................... 85
4.1.1 Differential elements and finite elements....................................................................................... 86
4.1.1.1 How many differential elements are there?..........................................................................................87
4.1.2 Advantages and disadvantages of finite elements.......................................................................... 89
4.1.3 How not to get lost in the collection of finite elements .................................................................. 91
4.1.4 1D elements ................................................................................................................................... 93
4.1.4.1 Hermite 1D polynomials in FEM.........................................................................................................93
4.1.4.2 The most known 1D elements............................................................................................................100
4.1.4.3 Thinwalled beams of open cross section...........................................................................................102
4.1.5 2D elements ................................................................................................................................. 105
4.1.5.1 Triangular elements............................................................................................................................105
4.1.5.2 Triangular elements with polynomials in L
1
, L
2
, L
3
...........................................................................113
4.1.5.3 Quadrilateral elements with polynomials in x,y.................................................................................117
4.1.5.4 Iso, hypo and hyperparametric elements ........................................................................................121
4.1.5.5 Surface elements recommended by the authors..................................................................................127
4.1.6 3D elements ................................................................................................................................. 147
Glossary
3
4.1.6.1 Tetrahedron........................................................................................................................................147
4.1.6.2 Bricks .................................................................................................................................................150
4.1.6.3 Toroid.................................................................................................................................................157
4.1.6.4 Special 3D elements...........................................................................................................................157
4.1.6.5 Solid elements recommended by the authors .....................................................................................158
4.2 PHYSICAL PROPERTIES OF ELEMENTSE .............................................................................................. 174
4.2.1 Physical models of materials of elements .................................................................................... 174
4.2.2 What is the effect of physical properties in FEM algorithms....................................................... 180
4.2.2.1 3D constitutive laws...........................................................................................................................180
4.2.2.2 Reduction of the dimension of a problem..........................................................................................182
4.2.2.3 Description of deformation in a reduced problem..............................................................................186
4.2.2.4 Components of deformation in a reduced problem............................................................................193
4.2.2.5 Physical constants of 2D FEM elements ............................................................................................197
4.2.2.6 Physical constants of 1D FEM elements ............................................................................................202
4.2.2.7 Physical constants of toroids..............................................................................................................208
4.2.2.8 Gas elements ......................................................................................................................................209
5 MODELLING OF STRUCTURES FOR FEM ANALYSIS................................................................ 211
5.1 INTRODUCTION TO THE THEORY AND PRACTICE OF CREATION OF FEM MODELS............................. 211
5.1.1 Presentday Approach to Modelling of Structures and Soil Environment ................................... 211
5.1.1.1 Objects and Terms .............................................................................................................................211
5.1.1.2 The Selection of an Effective FEM Model in Practice.......................................................................218
5.1.2 Dimensions of the Model for FEM analysis................................................................................. 221
5.1.2.1 The 1D Models ..................................................................................................................................226
5.1.2.2 2D Models..........................................................................................................................................234
5.1.2.3 Systems Consisting of 1D and 2D Elements ......................................................................................241
5.1.3 Numerical Stability of the Calculation of FEM Models............................................................... 250
5.1.3.1 Defective FEM Results Due to Arithmetics .......................................................................................250
5.1.3.2 Presentday Possibilities of Improving Arithmetics in FEM Calculations .........................................258
5.1.4 Modelling of Nonlinear Behaviour of Structures by means of FEM Algorithms........................ 260
5.1.4.1 User Approach to Nonlinear FEM Problems ....................................................................................260
5.1.4.2 Assembly of Equation Systems in Nonlinear FEM problems...........................................................263
5.1.4.3 Users interventions into the execution of nonlinear FEM programs................................................266
5.1.4.4 More Complicated Constitutive Relations and Projects Depending on the Path................................270
5.1.4.5 The Selection of the Number of Increments and the Course of the Equilibrium Iteration .................280
5.1.4.6 NewtonRaphson Method and its Modifications................................................................................281
5.1.5 Transformation of Physical Quantities........................................................................................ 291
5.1.5.1 Transformation of Tensors of Stress, Deformation and Physical Constants ......................................292
5.1.5.2 Design Stress and Internal Forces ......................................................................................................309
5.2 NOTES CONCERNING THE PROBLEMS OF MODELLING OF CERTAIN STRUCTURES IN THE ENGINEERING
PRACTICE ........................................................................................................................................................ 311
5.2.1 Introductory note ......................................................................................................................... 311
5.2.2 Modelling of Stiffeners in Planar Structures ............................................................................... 312
5.2.3 Modelling of ColumnSupports of Floor Slabs............................................................................ 314
5.2.4 Boundary Effects in Slab Models................................................................................................. 317
5.2.5 Singularities in the Analyses of Structures .................................................................................. 318
5.2.6 Modelling of the Interactions between Foundation Grids and Subsoil........................................ 319
5.2.7 The Density of the Mesh............................................................................................................... 320
5.2.8 Modelling of a Double Beam Bridge with wide Beams........................................................... 322
5.3 PHYSICAL AND SHAPE ORTHOTROPY OF PLATES............................................................... 323
5.3.1 Purpose of the guide .................................................................................................................... 323
5.3.2 Method......................................................................................................................................... 323
5.3.3 Core principal of solution............................................................................................................ 323
5.3.4 Stress components in physically orthotropic plates..................................................................... 325
5.3.5 Internal forces in physically orthotropic plates........................................................................... 328
5.3.5.1 Technical theory of plates with the effect of transverse shear not taken into account........................328
5.3.5.2 Plates with the effect of transverse shear taken into account..............................................................331
5.3.6 Shape orthotropy of plates........................................................................................................... 332
5.3.6.1 Main principles of the transformation into physical orthotropy.........................................................332
5.3.6.2 Simple types of orthotropic plates......................................................................................................333
5.3.6.3 Plates with the effect of transverse shear taken into account..............................................................348
Glossary
4
5.3.6.4 Boxsections.......................................................................................................................................352
5.3.6.5 Multicell slabs with linear hinges in longitudinal direction ..............................................................356
5.3.6.6 Other plate types ................................................................................................................................358
6 MODELLING OF STRUCTURESOIL INTERACTION.................................................................. 360
6.1 INTRODUCTION............................................................................................................................ 360
6.1.1 Origin and Development of the Efficient Subsoil Model ............................................................. 360
6.1.2 The Main Ideas of the Efficient Subsoil Model ............................................................................ 362
6.1.3 The Efficient StructureSoil Interaction Model Assuming an Arbitrary Shape of StructureSoil
Interface ..................................................................................................................................................... 364
6.1.4 Some Remarks about SoilFoundationStructure Interaction...................................................... 366
6.2 ENERGY DEFINITION AND GENERAL THEORY OF THE EFFICIENT SUBSOIL MODEL.. 371
6.2.1 Reduction of the Threedimensional Model to the Twodimensional Model ............................... 371
6.2.1.1 Threedimensional Models in Geomechanics ....................................................................................371
6.2.1.2 Twodimensional Efficient Subsoil Model.........................................................................................373
6.2.2 Onedimensional Efficient Subsoil or Soil Medium Model.......................................................... 391
6.2.2.1 Introduction........................................................................................................................................391
6.2.2.2 An Example of the Relation Between the Constants of One and Twodimensional Models ............392
6.2.2.3 Basic Onedimensional Relations ......................................................................................................394
6.2.3 Threedimensional Efficient Subsoil Model as an Improvement on the Twodimensional Model396
6.2.3.1 Main Idea of the Improvement of the Twodimensional Efficient Subsoil Model .............................396
6.2.3.2 Basic Geometrical and Physical Relations .........................................................................................397
6.2.3.3 Some Special Cases ...........................................................................................................................402
6.3 THEORY OF PLATES ON THE EFFICIENT SUBSOIL MODEL................................................. 405
6.3.1 Introductory Comment................................................................................................................. 405
6.3.2 Variational Problem of the Plates on Efficient Subsoil Model .................................................... 406
6.3.2.1 Total Virtual Work of the StructureSoil System...............................................................................406
6.3.2.2 Potential Energy of the PlateSoil System..........................................................................................409
6.3.2.3 Potential Energy of the Improved Subsoil Model ..............................................................................415
6.3.2.4 Variational Principles of StructureSoil Interaction ...........................................................................416
6.3.2.5 Advantages of Lagrange's Variational Principle and Principle of Total Virtual Work.......................418
6.3.3 Implementation of the SoilStructure Interaction Model Using the Finite Element Technique... 420
6.3.3.1 Dimension and Compatibility of Finite Elements ..............................................................................420
6.3.3.2 Kirchhoff's Plate on the 2DEfficient Model of the Subsoil...............................................................423
6.3.3.3 General Remarks on the Finite Element Technique...........................................................................427
6.3.3.4 Conclusions for the Solution of the PlateSubsoil Interaction............................................................431
6.3.3.5 Mindlin's Plate on the 2DEfficient Model of the Subsoil.................................................................432
6.3.3.6 Mindlin's Plate on the 3DEfficient Model of the Subsoil..................................................................435
6.3.4 Nonlinear Analysis of StructureSoil Interaction using the 2D Efficient Subsoil Model ............. 448
6.3.4.1 Introduction........................................................................................................................................448
6.3.4.2 Stress in subsoil..................................................................................................................................448
6.3.4.3 Physical model of soil based on the formula stated in CSN 73 1001 .................................................449
6.3.4.4 Physical model of soil according to DIN 4019...................................................................................452
6.3.4.5 Physical model of soil according to Eurocode 7 ................................................................................455
6.3.4.6 Variability of subsoil input data.........................................................................................................455
6.3.4.7 Reduction of the dimension of the interactive problem......................................................................459
6.3.4.8 Surface model of subsoil....................................................................................................................459
6.3.4.9 The effect of subsoil outside of the structure .....................................................................................464
6.3.4.10 Implementation into SCIAESA PT system.......................................................................................465
6.3.4.11 Statistical analysis of the structuresoil interaction ............................................................................469
6.3.4.12 Conclusion .........................................................................................................................................470
7 NONLINEAR MECHANICS OF CONTINUA AND STRUCTURES............................................... 471
7.1 INTRODUCTION.................................................................................................................................. 471
7.1.1 Selected Mathematical Concepts and Notations.......................................................................... 471
7.1.1.1 Index, tensor and matrix notations .....................................................................................................471
7.1.1.2 Voigt notation ....................................................................................................................................474
7.1.1.3 Voigt rule for higher order tensors .....................................................................................................476
7.1.1.4 Tensors...............................................................................................................................................477
7.1.1.5 Transformation of finite elements matrices........................................................................................484
7.1.2 Classification of Nonlinearity...................................................................................................... 489
7.1.3 Basic Equations, Eulerian and Lagrangean Elements ................................................................ 492
Glossary
5
7.2 GEOMETRICAL NONLINEARITY............................................................................................... 494
7.2.1 Foundational Concepts................................................................................................................ 494
7.2.1.1 Systems of coordinates in nonlinear mechanics .................................................................................494
7.2.1.2 Deformation gradient .........................................................................................................................494
7.2.1.3 Rate of deformation ...........................................................................................................................497
7.2.2 Strain Measures ........................................................................................................................... 498
7.2.2.1 Green Lagrange strain tensor E....................................................................................................500
7.2.2.2 Euler  Almansi strain tensor ( e ) ......................................................................................................501
7.2.2.3 Logarithmic strain measure (
n
)......................................................................................................502
7.2.2.4 Infinitesimal strain tensors ( ), ( e ) .................................................................................................502
7.2.2.5 Other strain measures.........................................................................................................................503
7.2.2.6 Comparison of strain tensors..............................................................................................................504
7.2.3 Stress Measures ........................................................................................................................... 506
7.2.3.1 Cauchy stress ( ) .............................................................................................................................508
7.2.3.2 Nominal stress ( N ), First Piola Kirchhoff stress ( P) ..................................................................508
7.2.3.3 Second Piola Kirchhoff stress ( S) .................................................................................................509
7.2.3.4 Corotation stress ( ) ........................................................................................................................509
7.2.3.5 Kirchhoff stress ( ) ..........................................................................................................................509
7.2.3.6 Biot stress ( T)..................................................................................................................................510
7.2.3.7 Transformations between different types of stress .............................................................................510
7.2.3.8 Objective stress rate ...........................................................................................................................510
7.2.4 Energetically Conjugate Stress And Strain Measures ................................................................. 511
7.2.5 Two Formulations of Geometrical Nonlinearity in FEM ............................................................ 514
7.2.5.1 Formulation based on current configuration (updated Lagrangean)...................................................515
7.2.5.2 Formulation based on reference configuration (total Lagrangean).....................................................522
7.3 MATERIAL NONLINEARITY....................................................................................................... 527
7.3.1 Uniaxial Stress............................................................................................................................. 527
7.3.1.1 Uniaxial nonlinear elasticity...............................................................................................................531
7.3.2 General Stress.............................................................................................................................. 533
7.3.2.1 SaintVenant Kirchhoff material .....................................................................................................534
7.3.2.2 Hyperelastic materials.......................................................................................................................535
7.3.2.3 Hypoelastic Materials .......................................................................................................................536
7.4 SOLUTION METHODS FOR NONLINEAR ALGEBRAIC EQUATIONS................................... 537
7.4.1.1 Picard Iteration Method......................................................................................................................538
7.4.1.2 Newton Raphson Iteration Method .................................................................................................539
7.4.1.3 Riks Method.......................................................................................................................................542
7.5 LINEAR AND NONLINEAR STABILITY, POSTCRITICAL ANALYSIS ................................. 547
7.5.1 Introduction ................................................................................................................................. 547
7.5.2 Linear stability............................................................................................................................. 548
7.5.3 Nonlinear Stability....................................................................................................................... 550
7.5.4 Postcritical Analysis................................................................................................................... 551
8 LINEAR AND NONLINEAR DYNAMICS OF STRUCTURES........................................................ 553
8.1 VARIATIONAL FORMULATION OF THE INERTIAL PROBLEM............................................................... 553
8.2 DYNAMICS OF FOUNDATION PLATES................................................................................................. 555
8.2.1 Consistent Mass Matrix of the Plate on the 2D Subsoil Model ................................................... 555
8.2.1.1 Consistent Mass Matrix of the Plate...................................................................................................555
8.2.1.2 Consistent Mass Matrix of the Subsoil...............................................................................................556
8.2.1.3 Resulting Consistent Mass Matrix of the Plate on the 2D Subsoil Model..........................................557
8.2.2 Consistent Mass Matrix of the Plate on the 3D Subsoil Model ................................................... 558
8.2.2.1 Damping Properties of the PlateSoil System....................................................................................559
8.3 LINEAR SOLUTION OF STRUCTURES SUBJECTED TO VIBRATION ......................................................... 562
8.3.1 The decomposition into eigenmodes method ............................................................................... 562
8.3.1.1 Calculation of seismic effects from response spectrum......................................................................563
8.3.2 Numerical methods of direct integration ..................................................................................... 565
8.3.3 Explicit methods........................................................................................................................... 565
8.3.4 Method of central differences ...................................................................................................... 565
8.3.5 Implicit methods........................................................................................................................... 566
Glossary
6
8.3.5.1 Newmark method...............................................................................................................................566
8.3.5.2 Wilson method...................................................................................................................................568
8.4 NUMERICAL METHODS FOR NONLINEAR SOLUTION OF STRUCTURE MODELS SUBJECTED TO DYNAMIC
LOAD 569
8.4.1 Modification of relations in motion equations............................................................................. 569
8.4.2 Modification of relations between displacement, velocity and acceleration vectors................... 570
8.4.3 Variants of connection of methods............................................................................................... 572
8.4.3.1 Algorithm of linear solution...............................................................................................................573
8.4.3.2 Variant I .............................................................................................................................................574
8.4.3.3 Variant II............................................................................................................................................579
8.4.3.4 Variant III ..........................................................................................................................................580
9 BENCHMARKS AND ILLUSTRATIVE EXAMPLES....................................................................... 585
9.1 BENDING WITH THE SHEAR DEFORMATION........................................................................................ 585
9.1.1 General remarks .......................................................................................................................... 585
9.2 GEOMETRIC NONLINEARITY.............................................................................................................. 586
9.2.1 General remarks .......................................................................................................................... 586
9.2.2 Axially and transversally loaded cantilever beam....................................................................... 589
9.3 SUBSOIL ............................................................................................................................................ 594
9.3.1 General remarks .......................................................................................................................... 594
9.4 CABLES ............................................................................................................................................. 605
9.5 MEMBRANES..................................................................................................................................... 610
9.6 MECHANISMS.................................................................................................................................... 612
9.7 STABILITY (BUCKLING)..................................................................................................................... 617
9.8 DYNAMICS ........................................................................................................................................ 618
LITERATURE................................................................................................................................................... 621
1.1 Mathematical definition of FEM
7
Glossary
SYMBOLS
&
f For a field, the superposed dot denotes the material time
derivative, i.e. ( , ) ( , )
&
f t f t t X X ; for a function of time
only, it is the ordinary time derivative, i.e. ( ) ( )
&
f t df t dt
' x
f Derivative with respect to the variable x ; when the comma is
followed by an index, such as , , i j k , to s , it is the derivative
with respect to the corresponding spatial coordinate, i.e.
'
i i
f f x
As in a b indicates contraction of inner indices; for vectors,
a b is the scalar product
i i
a b ; if one or more of the variables
are tensors of second order or higher, the contraction is on the
inner indices, i.e. A B represents
ij jk
A B , A a represents
ij j
A a
: As in : A B indicates double contraction of inner indices: : A B
is given by
ij ij
A B , : C D is
ijkl kl
C D ; note the order of the
indices! Note also that if A or B is symmetric :
ij ji
A B A B
As in a b indicates crossproduct, or vector product; in indicial
notation,
ijk j k
e a b a b
As in a b indicates matrix product of vectors, or Kronecker
product of matrices; in indicial notation,
i j
a b a b ; in matrix
notation, { }{ }
T
a b a b
symbol of partial derivative
1.1 Mathematical definition of FEM
8
VARIABLES
, , ,
x y z
A A A A Cross section area of a beam,
x
A full area, ,
y z
A A shear area in
the y and z directions
,
I
B B Matrix of spatial derivatives of shape functions in Voigt notation
arranged so that { } e B d or { }
&
I I
D B d ; B is a
rectangular matrix [ ]
1 2
, , , K
n
B B B
B bandwiceth of the overall (global) stiffness matrix
,
I
B B Matrix of material derivatives of shape functions in Voigt
notation arranged so that { } E B d or
{ }
& &
I I
E B d ; B is a
rectangular matrix
1 2
, , , 1
]
K
n
B B B
C Cauchy Green tensor,
T
C F F ; it is distinguished from the
material response matrices which follow by absence of a
superscript, or damping matrix
, , 1
]
SE SE SE
ijkl
C C C Material tangent moduli relating
&
S to
&
E
, ,
1
]
ijkl
C C C Material tangent moduli relating convected rate of Kirchhoff
stress
c
to D
, ,
1
]
J J J
ijkl
C C C Material tangent moduli relating Jaumann rate of Cauchy stress
J
to D
, ,
1
]
ijkl
C C C Material tangent moduli relating Truesdell rate of Cauchy
stress
to D
1 2 2
, , ,
S S S S
x y
C C C C Stiffness of subsoil,
S
C subsoil stiffness matrix,
1 2 2
, ,
S S S
x y
C C C
subsoil stiffness parameters
{ } , ,
ij
D D D Rate of deformation, velocity strain, ( ) sym D v
,
ij
E E Green strain tensor,
( )
1
2
T
E F F I
,
ij
F F Deformation gradient,
ij i j
F x X
1.1 Mathematical definition of FEM
9
J Determinant of Jacobian between spatial and material
coordinates, det 1
]
i j
J x X
0
K Linear stiffnes matrix
T
K Tangent stiffnes matrices
,
M
K K Material and geometric tangent stiffness, respectively
,
ij
L L Spatial gradient of velocity field
M Mass matrix
I
N Shape functions
,
ij
P P Nominal stress (transpose of first PiolaKirchhoff stress)
,
ij
R R Rotation matrix (rotation tensor)
,
ij
S S Second PiolaKirchhoff (PK2) stress
T Transformation matrix
n
logarithmic strain tensor
Total potential energy
int
e
potential energy of one element
int
potential energy of internal forces
ext
potential energy of external forces
0
, Current and original density
Stress matrix from the relation d , in Section 6 stress
tensor, where each component is multiplied by the unit diagonal
matrix I
{ } , ,
ij
Cauchy (physical) stress tensor
{ }
, ,
ij
Corotational stress tensor
{ } , ,
ij
Kirchhoff stress tensor
( , ) t X Mapping from the initial configuration
0
to the current or
spatial configuration
to the spatial
configuration
0
, Domain of current (deformed), initial (undeformed)
e
Domain of one element
Matrix of differential operators, its application to vector u yield
tensor { } : u ; is matrix of the type ( , ) s d , where s is
the number of components of the tensor { } in Voigt notation
and d is dimension of the model, i.e. number of the components
of the displacement vector
1.1 Mathematical definition of FEM
12
1.1 Mathematical definition of FEM
13
1 Principle of Finite Element Method
1.1 Mathematical definition of FEM
When speaking about a general concept of finite element methods including FEM,
BEM, finite layer method and strip method the mathematical nature inheres in what is
termed discretisation of the problem. The term discrete is the opposite of continuous. To
be clear: searching for unknown functions in domain with boundary is replaced by
searching for a finite number of values of these functions or parameters d from which an
approximate solution can be formulated, as explained later. Older methods used the same
approach: classical variational methods (W. Ritz, 1908) searched for coefficients of pre
selected functions with generally nonzero values across the whole domain . The well
known sieving method, known also as a differential method, replaced derivatives by
differences or, more generally, by combinations of several functionvalues in the nodes of a
mesh. The collocation method limited itself to the requirement of satisfying the given
conditions (roughly) in several selected points of and . Formally, the analytical solution
of differential equations was always transformed to the solution of systems of algebraic
linear equations. The same applies to FEM. The improvement is in the way this
transformation is carried out, or mathematically speaking, in the selection of base functions
into which the sought functions are decomposed. The decomposition is closely related to the
division of domain (or in BEM) into subdomains
e
, briefly called finite elements,
contrary to infinitely small differentials d, d of the exact analysis.
In the beginning, mathematicians were not interested into this approach. The first
approximation of the function of two variables by a combination of linear functions over
triangular elements (R. Courant 1943, termed plated surface) remained completely unnoticed
and even significant accomplishments of engineers in 19561965 who started to use
polynomials of second, third and even higher degree over the elements, attracted no attention
of analysts. Only after the first international conference where the FEM was presented (1st
Conference on Matrix Methods in Engineering, Ohio, 1965) even mathematicians noticed
what engineers had thought for long that a qualitatively new mechanism was created and
that it should be thoroughly researched. And in around 1968, a quite exact mathematical
definition of FEM was already given.
FEM is a generalised RitzGalerkin variational method that uses base functions defined
in a small compact domains closely linked to the selected division of the whole analysed
domain to finite elements.
In the same period it was shown that FEM generates systems of linear equations that
are numerically significantly better conditioned than the still commonly used sieving method;
formally flawless definitions of many useful terms were presented, hierarchies of various 1D,
2D and 3D finite elements were established according to conditions of continuity, etc. For
detailed information refer to [3] to [5]. Presentday engineers benefit from this extensive
research in the following way: they do not have to doubt about the mathematical
unobjectionability of FEM and can rely on the convergence towards the exact solution with
gradual refinement of the mesh in FEM programs that employ what is termed correct or
1.2 Deformation variant of FEM used in practical statics
14
compatible elements (brief overview in [5], Appendix). In order to understand the core
principle of FEM, it is convenient to use the engineering explanation of its most often used
variant: deformation method, which employs what is called Lagrangean finite elements. This
explanation will be made in the following paragraph. It forms the basis of almost all (in 1995
estimated 90 to 95%) commercially successful FEM systems.
1.2 Deformation variant of FEM used in
practical statics
This method can be easily programmed and produces well conditioned equation
systems. The simplicity lies in the energetic concept of the problem, generally in the
variational formulation of the problem where we search for an extreme of an operator
(a functional in mechanics) that is of additive nature. It means that its value is for the whole
system (domain) equal to the sum of values in the parts or elements of the system
(subdomains finite elements). This nature is characteristic especially for all the quantities
defined by means of any bounded integral in the domain. Thus, for example total potential
energy
int ext
+ of internal and external forces in the body is according to the
Lagrange variational principle minimal just for the real state of the body ( , , ) u . The FEM
equation can be in this particular situation obtained through the differentiation of with
respect to individual deformation parameters
1 2
, , , , ,
m N
d d d d K K . For example, the mth
equation is:
( )
0
int ext
int ext
m m m m
m
d d K d f
d
+
+
(1.2.1)
and can be obtained through the versatile addition theorem, the principle of which lies in the
additivity of energy as a scalar, i.e. in the additivity of energy derivations:
( ) ( )
1
1, ,1
1
P
int
e P
int int e
m e m
N N
e
m
d d
d
K d (1.2.2)
1
1
P
ext
e P
ext ext e
m e m m
e
m
d d f
d
(1.2.3)
Notice that the geometric dimension of the elements 1, 2, , e P K is not important.
Only their stiffness matrices
e
K are involved together with vectors of load parameters
e
f
that
may be of different dimension for different elements of the same system ( , )
e e
n n , ( ,1)
e
n . The
elements may be one, two, or threedimensional with no negative impact on the principle of
the solution. One tobesolved system can generally include e.g. beam, plate, wall and/or shell
elements, as well as 3D elements. Before they can be summed, all the matrices
e
K and
e
f
must be formally transformed to the same dimension ( , ) N N and ( ,1) N , if N is the total
1.3 More general form of FEM
15
number of deformation parameters. This applies to matrices
er
K ,
er
f that are established
when elements of matrices
e
K ,
e
f are written to those positions in matrices ( , ) N N , ( ,1) N
where they belong to according to their global code numbers that define the topology of the
system. The remaining elements of matrices ( , ) N N , ( ,1) N are zero. The whole system of
linear equations for the calculation of unknown parameters
1 2
[ , , , , , ]
T
m N
d d d d d K K is
formed in a simple way:
( , ) ( ,1) ( ,1) N N N N
K d f (1.2.4)
(N,N) ( ,1)
( , ) ( ,1)
1 1
P P
er er
N
N N N
e e
f f (1.2.5)
As the global code numbers (list of element nodes) hold complete information about
the position of the matrix elements in the ( , ) N N grid, it is not necessary to factually establish
the extended matrices
er
K ,
er
f , which would consume a considerable portion of the memory,
but the algorithm employs only the original matrices
e
K ,
e
f . And it is just this simplicity and
universality together with the fact that the system of equations is wellconditioned, that
represents the practical advantage of FEM in comparison with classical approaches and that is
the main reason why it is so universally and widely used in the present advanced era of
computer development. The generalisation through the orthogonalisation principle or through
the weighed residua method is explained in the following text.
1.3 More general form of FEM
After a more detailed investigation, we can easily find that the additivity of the
functional
e
e
or potentially of any bounded integral in domain
e
e
e
e
e
d d
K K (1.3.1)
is the only condition for the application of a powerful apparatus for establishing the system of
equations (i.e. the calculation of coefficients of unknown and absolute elements) developed in
FEM. Therefore, even in the early period of FEM development, it became used for problems
where no nature quantity could be defined, but we know differential equation
1
( ) L u 0 in
domain and boundary conditions
2
( ) L u 0 on its boundary that must be met by the
sought function u. This may be a set of functions [ , , ]
T
u v u K , shortly a vector of functions,
1.3 More general form of FEM
16
or a vector function. In that situation we also have more equations
1
L and conditions
2
L ,
which is indicated by the used matrix notation. Then, the problem may be parameterised by
the substitution u Ua or directly by u Nd (more exact matrix notation will be discussed
later) with unknown coefficients a or parameters d. After substituting a set of values a or d
into the abovementioned equations, they will not be fulfilled for obvious reasons and the
righthand side will not give identical zeroes, but certain functions in and that can be
called residuum
1
and
2
; which, in general, represent vector functions:
1 1
in L u 0 (1.3.2)
2 2
on L 0 (1.3.3)
Now, we can use the orthogonalisation principle, termed recently also the principle
of weighed residua, that was in fact used already by Bubnov (1910) and Galerkin (1915) as
the main improvement of the Ritz method (1908  1909), but without the modern
parameterisation of the problem by a numerically suitable base functions with a small
compact support ( u Nd ), in the classical Ritz form u Ua . From this point of view, FEM
can be seen as a generalised Ritz  Galerkin method resulting in significantly better
conditioned systems of linear equations. The fundamental idea is simple. First, it will be
explained for one unknown function. For the exact solution of u, the residua (1.3.2), (1.3.3)
are identical zeros, and consequently, also products
1
1
g in and
2
2
g on , for
arbitrary weight functions
1
g in and
2
g on , are identical zeroes. The bounded
integral from an identical zero is exactly zero, and thus, the following must be satisfied for an
exact solution of u and arbitrary weights
1
g ,
2
g :
1 1 2 2
0 R g d g d
+
(1.3.4)
Condition (1.3.4) is not satisfied by an approximate solution. If we had enough time and could
gradually substitute various sets d (in modern approaches) or a (in classical approaches) into
(1.3.4), then, with preselected weights
1
g ,
2
g , we would obtain different values of the total
error R in comparison against zero. For example, sets
1, 2 3
, , ,
m
d d d d K would give
6385.51, 800.75, 1263.04, , 38.51 R K . And R would be absolutely smallest, let us say,
for set
83615
d , and it its value would be 0.0445996 R . We might even come across a set that,
for the given accuracy, would give 0 R B . This set of parameters can be then declared the best
with respect to meeting condition (1.3.4) for the preselected
1
g ,
2
g and the corresponding
approximate solution u can be used in further steps. The procedure is clear, but completely
useless in practice. Even if we (i) abstracted away from the continuous possibility of changes
of individual parameters
j
d in the set
1 2 3 1
, , , , , , ,
T
j N N
d d d d d d
1
]
d K K (1.3.5)
(ii) limited ourselves to certain nodal points of the number axis, (iii) eliminated technically
irrational values
j
d , (iv) managed to get over the impact of the preselection of
1
g ,
2
g , etc.,
we would have such a huge number of sets (1.3.5), that even the fastest stateoftheart
computers would not find the best set in a reasonable time.
1.3 More general form of FEM
17
The way out of this is that after the substitution of (1.3.5) into (1.3.4) and after integration
(over the elements
e
,
e
), R becomes a function of N parameters
1 2
, , ,
N
d d d K :
1 2 3
( , , , , ) 0
N
R d d d d (1.3.6)
If equations (1.3.2), (1.3.3), respectively their operators
1
L ,
2
L , are linear, then also formula
(1.3.4) is linear in parameters
1 2
, , ,
N
d d d K . It is therefore useful to establish for them
N algebraic linear equations, which can be achieved if we write equation (1.3.4)  N times
with N different weight functions
1 2
( , ) , 1, 2, ,
j
g g j N K :
1 2
( , , , ) 0 1, 2, ,
j N
R d d d j N K K (1.3.7)
The sought parameters
1 2
, , ,
N
d d d K , can be found from system of N equation (1.3.7), which
gives the approximate solution of the problem. If operators
1
L ,
2
L are not linear, also
equations (1.3.7) are nonlinear, which on one hand results in (i) wellknown complications of
solution algorithms (e.g. NewtonRaphson method), (ii) problem of ambiguity, etc., but, on
the other hand, does not mean a principal limitation of the application of the approach. In
problems of mechanics, this mathematical algorithm can be sometimes clearly mechanically
interpreted. For example, already Bubnov (1910) and Galerkin (1915) considered it to be a
generalised principle of virtual work, which was elaborated on by a number of other authors,
best by V. Z. Vlasov in 1950  1960. Weight functions g are in this sense considered to be a
generalised virtual displacements (states of deformation) of the system and L are
considered to be a generalised force quantities. As most of the authors did not reproduce the
original ideas correctly, we recommend that readers should study the original publications
[10] to [14] and the literature cited in [3] to [5].
It is clear from the formal derivation of (1.3.7) that different weight functions g would have
different corresponding equation systems (1.3.7) and thus different solutions d and
subsequently different internal forces, stresses and deformation in the analysed structure. If
the solution d was strongly dependent on the selection of g , it would not be reliable for
technical applications. Bubnov, Galerkin and also Vlasov tried to clear this issue through the
mechanical interpretation of g . We can briefly say: if we are not able to respect all the
possible virtual displacements, let us respect at least N mutually independent and, for the
given system and its connections, most characteristic displacements, i.e. functions g .
Relation to other methods was scrutinised as well. For example, in mechanical problems with
a potential energy, if we select the functions g to be gradually equal to all base functions of
set U , and if we fulfil in advance boundary conditions
2
L 0 on , we obtain equations
(1.3.7) identical to the Ritz method. If we fulfil in advance conditions
1
L 0 in the analysed
domain, the procedure is identical to Trefftz method, which is the historically oldest
boundary method preceding the nowadays BEM that will be briefly described in the
following paragraph.
1.4 Mathematical formulation of boundary element method BEM
18
1.4 Mathematical formulation of boundary
element method BEM
The most concise description of BEM reads: We select such weight functions
1
g ,
2
g
and perform such per partes domain integral integrations (Gauss  Ostrogradsky theorem), so
that equation (1.3.4) contains only integrals in and on that can be numerically
calculated from the input data and so that the only unknowns are the distributions of the
quantities along boundary . In general, this can be achieved in problems of mechanics with
boundary
p u
+ (stress vector p is given on
p
, displacement vector u on
u
)
through the selection of
1 2 2
, ,
j p j u j
g u g u g p
(1.4.1)
where
j
u
its reaction on
u
. The stated
distributions along the boundary are parameterised the usual way: we select a finite number of
elements on boundary , define finite boundary elements and their nodal parameters, in total
N values for set (1.3.5). If we gradually select the source functions for individual boundary
nodes to be the weight functions, we obtain such a number of algebraic equations that is just
required to solve the parameters. This rather mnemonic and encapsulated overview should be
elaborated at least in the following: If there is just one unknown function e.g. the deflection
of a membrane w, temperature T , torsion function F , etc. then each boundary node has
just one unknown parameter , , d w T F etc. If there are two or three unknown functions
e.g. , , ( ) u v w in a 2D (or 3D) elasticity problem then each boundary node has two or three
parameters d , marked locally
1 2 3
, , d d d , e.g. , , u v w in a pure deformation variant with
reaction components , ,
x y z
p p p in the fixed nodes. Similarly to FEM, also other variants are
possible. Also notations (1.3.2), (1.3.3), (1.3.4) then represent two or three equations, so that
also the number of conditions for the solution increases correspondingly as well. For the
weight functions are the used what is termed fundamental functions, under special conditions
the exact source functions of individual displacement components , , u v w. These are, in
accordance with the general influence principle, identical with the distribution of
displacement components caused by singular loads 1
x
P , 1
y
P or 1
z
P acting on the
analysed system in boundary nodes. The method has developed from the older method of
integral equations (Boundary Integral Equation Method, BIEM) and the presentday common
international name is BEM (Boundary Element Method).
1.5 Partial discretisation of the problem  Finite
Layer Method
The discretisation of the problem i.e. the substitution of unknown functions defined
over the continuum of an domain and along its boundary by a countable, even finite, set of
1.6 Impacts of current division of labour on FEM in praktice
19
parameters (parameterisation of the problem) does not have to be complete. If we can in a
certain direction, let us say in the  x direction, make a very good estimate of the character of
the course of functions ( , ) f x y e.g. by means of trigonometric components, it is sufficient to
divide just the  y interval (the front arch of a prismatic folded plate, support edge of a bridge
deck, etc.) and we get elements of a strip method that reduces a 2D problem into a 1D task.
The reduction in the dimension of the problem can also be made in advance through the
following. Instead of unknown function f
D
of several variables in domain
D
we introduce its
projection to function
D s
f
of fewer variables in domain
D s
, where D is the dimension of
the original domain and s is the reduced dimension practically, 1or 2 s in common
transformations (Fourier, Laplace, Hankel and others). It is an integral transformation with
different weight functions
D
g , defined in the original domain
D
. After bounded integral
D s D D
f f g d
(1.5.1)
is introduced, the only remaining variables are those from domain
D s
. One of the oldest
technical applications deals with a layered continuum (Bufler, NikitinShapiro, Falk and
others) and reduces a 2D symmetrical problem or a 3D general problem into a 1D problem
within the interval 0 z H , divided into layers of thickness
i
H , 1, 2, , i n K . The
procedure is known as a finite layer method = FLM. The reduction in the dimension of type
(1.5.1) is also intensively exploited for the reduction of time variable t , e.g. in viscoelasticity
problems. The solution then deals with what is termed assigned elastic problem.
The reduction in the dimension of a problem from a 3D construction subsoil massif
into a 2D surface problem in the footing surface is the fundamental precondition for the
effectiveness of FEM programs in the field of common foundation engineering; see [8, 9].
1.6 Impacts of current division of labour on
FEM in praktice
It is a wellknown fact that in the current period the extent of practically useful
scientific/technical knowledge doubles within 10 years, while the halflife of scientific
knowledge (replacement by new, more accurate, more economic with regard to scientific
thought, and more generally valid pieces of knowledge) is about 5 years. This gives rise to the
question what part of the presentday information explosion is supposed to penetrate down to
the engineerdesigner, what part should be understood and used actively in their design
practice, what part they are supposed to know as existing in order to be able to find the
details, etc. This is not a simple task as the human brain competes with the most powerful
computers as far as the structure is concerned, but it dramatically lags behind in terms of the
(i) speed of performed operations, (ii) scanning of information (concentration about 6 bits per
second), and (iii) time over which the information is stored (some data, even most of them,
are erased immediately). An erudite specialist just before retirement holds in their brain, i.e. in
their operational memory, approximately 10
9
 10
10
bits of information, unless they develop
sclerosis (through bad diet and insufficient mental gymnastics) when they are about 40 that
1.6 Impacts of current division of labour on FEM in praktice
20
keeps progressing and increases the natural handicap of hundreds of thousands of neurons
dying every day. Even under optimal circumstances, it is illusionary to require that the
engineerdesigner fully understands everything they use for their work, e.g. that they have
acquired comprehensive knowledge of methods of analysis, their numerical algorithms and
programs, that they know everything about the work with computer and its peripheries,
scanners, printers, plotters, hardcopy generators, digitizers, etc. Similarly, they can hardly be
capable of citing (by heart) even a fraction of various standards and regulations. We have to
accept the fact that also the division of labour has increased dramatically and that engineering
and design institutions now have specialists to tackle this issue: mathematiciansanalysts,
programmes, electronics engineers, operators, specialists in technical fields, sometimes even
documentalists of technical standards and their amendments, etc.
What is thus left for the engineer, what burden cannot be taken from their
shoulders and what cannot be done by anyone else?
There is quite a bit of it, as the everyday practice of structural engineers, production
boards, site engineers and other noninterchangeable roles proves. In this text, we will focus on
structural engineering, and in this field it is the structural engineer (which includes a team of
structural engineers in the case of larger projects) who must (i) define the analysis model, (ii)
find all the related material in applicable standards or request a corresponding survey, (iii)
prepare input data for the program that will be used to analyse the model, (iv) perform the
solution at the computer terminal themselves or with the help of operators, (v) correctly
interpret the results for further steps in the design process and (vi) during all phases carry
out effective checks of all input data and outputs.
The structural engineer has to safely sort all the obtained information relating to the
designed structure into categories: geometrical data (lengths, angles, shapes, topology),
statical (external impulses, loads, action conditions) and physical (in general, rules for the
behaviour of substances, or Hookes law in the simplest terms), as they are sorted into these
groups by software input interfaces. They need to have a clear idea about the conditions of
continuity and equilibrium, both in the structure and at its boundary (support conditions).
To sum up: the intrinsic task of the structural engineer is to possess the knowledge of
mechanics to the extent that is required by the selected analysis model.
As far as the calculation methods (that form the basis for the applied programs) are
concerned, the structural engineer has to be familiar with their core principle to the extent
relevant to their reliable application under standard circumstances. When FEM programs are
used, it is necessary to know what elements are implemented, i.e. what approximations of
functions are assumed on them. This must be taken into account when the density (size) of
finite elements is being chosen. In addition, also accepted assumptions concerning the
internal forces with the resulting limitations in the practical application must be known.
Moreover, one must understand the style of expression used by manuals and become
familiar with the terminology employed, in order to be able to communicate with the
programmers when unavoidable errors in input and output have to be clarified, quickly
located and corrected. One also needs to have a correct idea of the extent of application of
the selected program or calculation method. With regard to complex problems that appear
only occasionally in engineering practice, one must be aware of the programs and technical
literature suitable for resolving the problem and must be able to study these material on ones
own.
Basically, a civil engineer has been prepared for all these key tasks already at
1.6 Impacts of current division of labour on FEM in praktice
21
university, even though to the extent that is proportional to the (i) possibilities of the
curriculum, (ii) study plan, (iii) teaching staff, and (iv) university facilities. The present era
makes lifelong learning in practice vitally important. And the aim of this book is to provide a
part of the knowledge required.
2.1 Explanation of used terminology
22
2 Basic terms and algorithms of
finite element method
2.1 Explanation of used terminology
Element is a part of the whole that is either physically composed of the elements, or that is
divided into the elements in our theorization.
Infinitely small element, also infinitesimal, dx , dxdy , dxdydz etc. is a limit of a finite size
element whose dimensions approach zero. It is used in differential and integral calculus and is
commonly used in the technical practice as it has its place in a university curriculum.
Finite element is an element of finite dimensions, contrary to the infinitely small element. For
example, a rectangle a b can be divided into 4 finite elements with dimensions 2 2 a b , or
a triangle can be split into several triangles on condition that we create the additional vertices
through a kind of triangulation, etc.
A domain is a connected set of points (open, unless we consider also its boundary; closed
including the points of the boundary). It can be simply or multiply connected (with openings).
In technical practice, this means socalled bodies (3D) that are in fact threedimensional in
Euclidian space with coordinates , , x y z in fact a set of points ( , , x y z ). And this is the way
that they are handled in dams, thickwalled blocks, soil massifs, etc. where none of the
dimensions is significantly smaller. Often, however, simpler analysis models are introduced:
Twodimensional domain (2D) of walls, plates, shells, box structures, etc., planar or spatial of
more or less complex structure, the points of which are assigned certain physical properties
including those depending on thickness, i.e. sectional dimension that does not exist in a two
dimensional domain.
One dimensional domain (1D) of beams, frames, truss girders, grids, netting, etc., in general
in structures composed of what is termed beams (the beam is thus a onedimensional model of
a body that is in fact threedimensional).
Subdomain is a set of points of the domain that have the same properties, i.e. nonzero
measure, in the corresponding dimension 3 or 2 or 1 d in spatial, planar and linear,
respective three, two, onedimensional domains. From this we get the shortest definition:
Finite element is a subdomain.
A connected set of finite elements, which is usually treated as what is termed substructure or
segment, can also be a subdomain. This is also related to the formation of more complex finite
elements, see further in the text.
Element is the English term that gave the name to the Finite Element Method.
Subelement is an element that forms a part of a larger element. For example, four triangular
subelements can form one quadrilateral element, or five tetrahedrons can form one
hexahedron (called brick), etc.
2.2 Explanation of the procedure on an example deformation variant of FEM
23
Superelement is an element created from two or more subelements.
Substructure is a part of the structure that can be analysed separately usually with the aim to
simplify the solution of a larger structure.
Load is in FEM perceived in a generalised sense as all impacts (external and internal,
gravitational, thermal, hygroscopic) that produce internal forces and displacement of the
structure. In input they are sorted into force load (common forces and moments in points or
distributed over a certain 3D, 2D, or 1D domain) and deformation load (similar deformation
impulses) and, quite often, thermal and similar impacts (shrinkage analogy to cooling, etc.)
are extracted from the deformation ones.
Support (support conditions) is a technical term for boundary geometric or, as the case may
be, kinematical, conditions. Basically, it means the reduction of degrees of freedom of a 0D
(point), 1D (linear) or 2D (planar support) figure. Inputs strictly distinguish (i) fixed
supports in which all degrees of freedom, or more precisely deformation parameters, are a
priori nullified, (ii) support with given values of deformation parameters (if zero values are
input, this type coincides with the previous one) and (iii) flexible supports.
Elastic support is the most frequent type of flexible support. It is assumed that the size of
reactions depends only on the magnitude of deformation parameters in the state when the
reactions are analysed. In general, the relation may be nonlinear, but the term elastic
eliminates possible dependence on other factors, especially on the history of the loading
process. That means that the relation between reactions and deformation parameters is always
unambiguous.
Physically linear elastic support is a type of support where the relation between reactions and
deformation parameters is approximated by a linear law of the following type: r kd, which
is similar to Hookes law. In general, the set of reactions can contain also parameters d of
other nodes. Usually, however, we assume in one node a linear nature of this relation. In the
simplest example with just one reaction r the matrix of various connections k is converted
into what is termed a spring constant k in relation r kd . The linkage between quantities r
and d is that virtual work r on d is the full product rd (principle of net virtual work).
Element node list is a sorted list of numbers of the element nodes.
2.2 Explanation of the procedure on an
example deformation variant of FEM
The finite element method is a method based on dividing the analysed domain into
subdomains, or illustratively into finite elements. In the widest meaning of the word, this
name can be used for any calculation that exploits finite elements. For example, already in the
years before Christ, in order to determine the area of a planar figure U , ancient
mathematicians divided it into finite elements
i
U of a simple shape (rectangles, sectors) and
applied rule
2.2 Explanation of the procedure on an example deformation variant of FEM
24
i
i
U U
(2.2.1)
that is actually commonly used until now, and even for other quantities relating to planar
figures, e.g. crosssections of beams. In addition, it is sufficient that the quantity is additive in
nature, i.e. that its magnitude for U is the sum of magnitudes for
i
U . This is for example the
second moment of area (moment of inertia) about the same x axis:
2
2
i
i i
i
J y dxdy
J y dxdy J J
(2.2.2)
It can be easily understood that this applies to any quantity defined by the value of a bounded
integral over domain U . Whatever the dimension of this domain, the theorem on the
calculation of bounded integral (only one integration symbol is used in the notation here, the
type of the integral depends on the dimension of domain U ) whose value equals V holds for
all continuous functions (this assumption can be even weakened):
i
i
U i
U U
fdU fdU V V
(2.2.3)
For example, in a onedimensional domain 0 x L we have: After the domain is divided
into finite elements  intervals
0 1
( , ) x x ,
1 2
( , ) x x , ...,
1
( , )
n n
x x
for
0
0 x ,
n
x L in dividing
points (termed nodes) , 1, 2, , 1
j
x j n K , the mesh has 2 endnodes and 1 n internal
nodes plus n finite elements and applies to any continuous function ( ) f x :
1
1
0
( ) ( )
j
j
x
L
n
j
x
f x dx f x dx
(2.2.4)
Function ( ) f x can be for example the density of potential energy of internal forces in a
beam:
2 2
2 2 2 2
( ) ( )
( ) ( ) ( ) 1 ( )
( )
2
y y y
x z z z
k y z
Q x M x
M x Q x M x N x
f x
EA GA GA GJ EJ EJ
1
+ + + + +
1
1
]
(2.2.5)
where sectional characteristics and modules can even be for beams of variable crosssection
functions of x . The additive nature of the energy, which is scalar, is evident: the energy
accumulated in the beam as what is called deformation work of an elastic state is the sum of
energies of its parts, i.e. finite elements into which the beam is divided. This applies to all,
even multidimensional, elastic bodies (structures):
int int
j
j
(2.2.6)
Here, j is the summation index. The potential energy of external forces, i.e. of the given load
(index z ) and reactions, resulting usually from gravitation and other sources outside of the
body, is marked
z
and is of the same additive nature, e.g. for a general body with
boundary :
2.2 Explanation of the procedure on an example deformation variant of FEM
25
ext T T
d d +
X u p u (2.2.7)
Also the singular load by concentrated forces and moments is included into the integration,
i.e. elements of type
(P u M )
k k k k
k
+
(2.2.8)
and we sum over all the loaded points k . Using the oldest and most general principle of
virtual works (used in simple machines already by Archytas of Tarentum and Archimedes in
fourth and third century B.C.), we can derive the Lagrangean variational principle of the
minimum (for details see [5]) total potential energy of the system (body + its load) and of the
minimum of its internal and external forces.
min.
int ext
+ (2.2.9)
This principle means that the algebraically lowest value of is achieved in a real state of the
body when related to the initial state in which the zero energetic levels of ,
int
and
ext
are defined. Following from the Clapeyron theorem,
1
2
int ext
in linearly elastic bodies,
and therefore
int
(there is always a decline from the initial state). If we calculate the
value of
ext
from any other than the real system of stresses and deformations, we always get
on condition that we meet all geometrical supporting and continuity conditions
algebraically higher value of . Instead of finding the stressstate and deformation of the
body from the equilibrium and continuity conditions which are in the form of differential
equations with enormous demands on the smoothness of the function (continuity up to the
third derivative) we can find the solution directly from the abovementioned condition of
the minimum, without having to transform to Euler differential equations of this variational
problem, that would lead us back to the equilibrium and continuity conditions.
The finite element method in a narrower meaning of the word is the oldest form of FEM, only
for elements
e
of domain . Methods that introduce finite elements on the boundary, finite
strips and layers (see art. 1.5) have special names. In the period of early development between
19561965 FEM used to be usually connected with the problem of finding the extreme of an
operator (in mechanics it is also called functional) . Contrary to classical variational
methods that do not divide domain U and that narrow down the class of allowable functions
(among which the extremal meeting the requirement of the variational problem is sought) to a
linear combination
k k
k
f a g
ga (2.2.10)
where base functions
k
g are nonzero throughout almost the whole domain U (except the set
of points of zero measure in which the coordinate axes or planes are possibly intersected),
the finite element method uses the following new means:
a) Base functions
k
g , closely related to the division of the domain into finite elements
and nonzero only in the elements that contain one common node of the mesh (this set
of elements is a small compact support of function
k
g ). These functions have a
typical pyramidal character and can be named in a popular way as source functions
of the sought function in terms of coefficients
k
a , as they define the distribution of
2.2 Explanation of the procedure on an example deformation variant of FEM
26
function f in the case that only 1
k
a and other 0
k
a , e.g. for
1 2 3
1, 0 a a a K we have
1
f g ,
b) Particular meaning of decomposition coefficients f (it is an analogy to the
decomposition of a vector into components, hence the name base functions) in
technical problems: the original decomposition of f within one element is replaced
by the decomposition that uses as coefficients certain parameters of nodes in the
domain mesh contained in this element and common to all adjacent elements. The
relation between these parameters d and original coefficients a in matrix notation is
1
d Sa a S d (2.2.11)
The original decomposition of f within one element is thus transformed to
1 1
f
ga gS d Nd N gS (2.2.12)
with base functions N with which it is significantly easier to satisfy the continuity
between elements. Once again, these are source functions, this time in terms of
parameters d , and e.g. for
1
1 d , other 0 d we have
1
f N etc., which gives an
illustrative and technically feasible meaning of FEM base functions. Expressed in
popular terms, they can be compared to influence lines.
c) with regard to the additive nature of the bounded integral (2.2.3), e.g. energy (2.2.9),
the general algorithms for the calculation of coefficients of the equations with
unknown nodal parameters can be based on the summation of corresponding
expressions (integrals) over elements, which can be conveniently programmed using
code numbers as explained later. Therefore, FEM algorithms do not depend on the
shape of the analysed domain, but on the shape and type of applied elements.
Similarly, it is not necessary to distinguish between internal and boundary points of
the domain and conditions (connections) for parameters can be prescribed the same
way in internal and external nodes, i.e. various supporting conditions can be met,
which was unmanageable in classical variational methods.
d) The character of FEM base functions means that the equations have a sparse and
under certain circumstances even band matrix of coefficients of lefthand sides and
are wellconditioned for numerical solution. The most significant band character of the
matrix would occur for mutually orthogonal base functions, i.e. for
0
k m
q q d k m
(2.2.13)
0
k m
N N d k m
(2.2.14)
which was used even in classical methods, e.g. in the analysis of plates using double
series. In that case, only diagonal coefficients of the equations remain nonzero and
the system splits into separate equations of one unknown. This ideal configuration can
be achieved in classical methods only at the cost of a dramatic increase in the
stringency of the conditions, e.g. for plates we have to limit ourselves to rectangular
plates without openings that are simply supported around the whole circumference. In
FEM, the base functions have the already mentioned pyramidal character and their
small compact supports have nonzero intersection only for closely neighbouring
functions, i.e. functions relating to the nodes that are common to one element. For a
fine mesh the stated condition of orthogonality is satisfied almost for all pairs of base
functions, and thus their system is almost orthogonal. How to achieve this favourable
2.3 Individual steps of deformation variant of FEM
27
condition through a suitable numbering of nodes or through other means will be
explained later. The aim is always the same: to get the narrowest band of nonzero
coefficients around the diagonal. The stateoftheart program systems perform this
minimisation internally and automatically.
In conclusion, let us emphasise the following: FEM is a method of analysis of a given model
of reality the behaviour of which is described by a variational problem. The problem can be
from the field of mechanics, electromagnetics, etc. FEM cannot improve the properties of the
used model of reality. On the other hand, it allows for the input of models that are closer to
the reality than the models that could be analysed by means of classical methods.
2.3 Individual steps of deformation variant of
FEM
For the sake of brevity and clarity, we limit ourselves mainly to (i) one area of FEM
application mechanical problems from the field of civil and mechanicalengineering and to
(ii) the most frequent solution of such problems Lagrange variational principle. In this
standard situation we can keep a unified approach according to the enclosed summary, which
can be applied with small modifications even in other FEM applications.
o.
2. Quantity: 3. Formulas:
4.
.
5. Matrix of selected functions
of coordinates U (mononorms) 
type ( ,1) d
6. d = number of unknown function in
the problem, d = 1 to 6 for FEM programs in
common structures
7.
.
8. Deformation parameters 
type ( ,1) n
9.
( ,1) ( , ) ( ,1) n n n n
d S a
10.
.
11. Vector of unknown
coefficients type ( ,1) n
12.
1
( ,1) ( , ) ( ,1) n n n n
a S d
13.
.
14. Distribution of displacement
components and possibly rotation
components type ( ,1) d
15.
1
( ,1) ( , ) ( ,1) ( , ) ( , ) ( ,1) ( , ) ( ,1) d d n n d n n n n d n n
u U a U S d N d
16.
.
17. Matrix depending on
element shape  type ( , ) n n
18. S
19. Note: 1 to 5 = selection of the finite element
20.
.
21. Distribution of strain
components or other strain
characteristics type ( ,1) s
22.
1
( , ) ( , ) ( ,1) ( ,1) s n s n s n
B a BS d B d
2.3 Individual steps of deformation variant of FEM
28
23.
.
24. Matrix derived
25.
( , ) s d
G through derivations
from U
26.
( , ) ( , ) ( , ) s n s d d n
B G U
27.
( ,1) ( , ) ( ,1) s s d d
1
1
]
G u
28.
.
29. Distribution of stress
components or internal forces 
type ( ,1) s :
30. Space: 6 s ,
31. plane: 3 s ,
32. plate: 5 s ,
33. shell: 8 s etc.
34.
1
( , ) ( , ) ( ,1) ( , ) ( ,1) ( , ) ( , ) ( ,1) ( ,1) s n s n s s s s s s n n n n
C C B S d d
35.
.
36. Matrix of physical constants
 type ( , ) s s
37. C
38.
0.
39. Matrix of stresses type
( , ) s n
40.
1
( , ) ( , ) ( , ) ( , ) ( , ) ( , ) s n s n s n s s n n s s
C B S C B
41.
1.
42. Stiffness matrix of element
e,
43. = analysed domain
44.
e
= element domain
45.
1 1
( , ) ( , ) ( , ) ( , ) ( , ) ( , )
( , ) ( , ) ( , )
e
e
T
T
e
n n n s s n n n s s n n
T
e
n s s n s s
d
d
K S B C B S
B C B
46.
1a
47. Introduction of global
deformation parameters
48.
e eg
T
eg e
d Td
d T d
49.
2.
50. Element stiffness matrix in
global parameters, extended to
dimension ( , ) N N
51.
( , )
( , ) ( , )
( , )
( , )
T
eg e
N N
T
er eg
N n n N
N N
n n
K T K T
K L K L
52.
3.
53. Structure stiffness matrix,
type ( , ) N N (created through
addition
54. theorem)
55.
56.
( , )
( , )
er
N N
N N
e
K K
57.
4.
58. Vector of load parameters
( ,1) n of element e.
59. X volume load,
60. p surface load,
61.
i
P concentrated impulses.
63. Type X, p, P, is ( ,1) d
64.
65.
e e
T T
e
T T
ui i i i
i i
d d
N P N M
+ +
+ +
f N X N p
2.4 Specification of selected operations that are useful to understand FEM terms
29
62.
i
M moments if u does
not contain rotations , is
differentiated from u
66.
67. Types: ( ,1) ( , ) ( ,1) n n d d
68.
5.
69. Matrix of influence
functions
70.
1
1
( , )
( , )
type d n
type s n
N US
B GN BS
71.
6.
72. Vector of load parameters of
element in global coordinates,
extended to dimension ( ,1) N
73.
( ,1)
( , )
( ,1)
( ,1)
T
eg e
n
T
er eg
N n
N
n
f T f
f L f
74.
7.
75. Vector of righthand sides
of equations, type ( ,1) N
76.
77.
e
e
f f
78. All types ( ,1) N
79.
80.
8.
81. System of linear equations
for unknown deformation
parameters of the body ( ,1) N d
82.
( , ) ( ,1) ( ,1) N N N N
K d f
83.
9.
84. Calculation of stresses and
displacements in elements once d
has been solved, type ( ,1) s , type
( ,1) d U
85.
( , ) ( ,1) ( ,1) s n s n
d 86.
( ,1) ( , ) ( ,1) d d n n
U N d
2.4 Specification of selected operations that
are useful to understand FEM terms
Detailed derivation of the procedure is given in common textbooks [1 to 5]. FEM is
nowadays a part of the curriculum at civil engineering and mechanical engineering faculties.
Consequently, the graduates are familiar with its fundamentals. Other users will however find
it useful to describe some operations the understanding of which facilitates the handling of
inputs and outputs in FEM practice. Also users familiar with the method may benefit from
this brief summary.
The deformation variant of FEM, used almost in all FEM programs, uses displacement
components u, sometimes together with rotation components , as unknowns. In general,
the unknowns are put together into one matrix vector d. For the simplest onedimensional
problem ( 1 d ), it is a single deflection function ( ) w x of a beam in plane ( , ) x z . For bending
of plates according to the Kirchhoff theory it is ( , ) w x y in the direction of z axis. For planar
2.4 Specification of selected operations that are useful to understand FEM terms
30
elasticity problems [ , ]
T
u v u : there are two functions ( , ) x y , dimension of the problem
2 d . For 3D bodies we have three functions , , u v w with variables , , x y z etc. Normally, we
deal at most with six components of vector d displacements , , u v w and rotations , ,
x y z
for planar and beam elements of Mindlin type (for details see art. 4). If we can find u
accurately enough (including required derivatives), calculation of other quantities necessary
for the design practice is a routine work.
The distribution of u across one element is approximated polynomially according to
no. 4 of art. 2.3 with coefficients a , i.e. expression u Ua , where U is matrix (no.1 of
art. 2.3), containing just zeros (0), ones (1) and monomials, i.e. expressions of type
, , x y z
with integer nonnegative , , . We select values u, or derivatives,
exceptionally also linear combinations of components of u as deformation parameters d.
And we do it in what is called nodal points or element nodes. These are in fact nodes of the
mesh that represents the division of the domain into elements called corner (or vertex)
nodes or sometimes also some boundary (midside) nodes or even inner nodes, e.g. in the
centroid. If we introduce such parameters, band width BW of the global stiffness matrix K
increases while time
2
T a N BW necessary for the solution of the system Kd f (no. 18
of art. 2.3) increases quadratically with BW . Therefore, we often eliminate these parameters
through suitable procedures, which will be explained later in the text. However, it is always
true that when substituting nodal coordinates into the expression u Ua or corresponding
derivatives of components u we can obtain relation (2.2.11) and (2.2.12) between the
deformation parameters d and coefficients a in the form d Sa (no.2, art. 2.3), i.e. also its
inversion
1
a S d (no.3, art. 2.3.). Consequently, following from no. 4, art. 2.3, the distribution of u
across the element can be expressed only through its deformation parameters:
1
( ,1) ( , ) ( , ) ( ,1) ( , ) ( ,1) d d n n n n d n n
u U S d N d (2.4.1)
Also the second form of (2.4.1) is for many elements obtained directly. This introduces a
slight shortening of the form of writing the formulas. For better understanding, we keep both
forms of (2.4.1).
The first one is indispensable for classical elements, which use polynomial coefficients that
are only in the second part of the reasoning replaced by the deformation parameters.
This represents what is termed parameterisation of the problem: instead of unknown
functions, we search only for a finite number of parameters d. In the given variant, we would
use equilibrium conditions for the classical solution, e.g. in the theory of elasticity the Lams
differential equations of second order. This approach is however applicable only to special
homogenous and isotropic problems and due to its complexity was replaced in practical
problems by the variational problem of finding the minimum of the total potential energy
of internal forces (
int
) and external forces, i.e. loads (
ext
) acting on the body. This is the
Lagrange variational principle (2.2.9) that is superior to all equilibrium conditions that prove
to be only its Euler differential equations. We omit the mathematically crucial, but for the first
explanation less important, consideration about the definition of the solution of the original
and variational problem, the term generalised solution and relating questions concerning the
function spaces, equivalence, and convergence. We present only the algorithm of the
procedure.
2.4 Specification of selected operations that are useful to understand FEM terms
31
The potential energy has in fact always the form of work, i.e. the product of force and
distance. The original, nondeformed, state is chosen as the zero energetic level. External
forces X in body and p on its boundary acting in the real final deformed state of the
body have always negative potential energy when related to this level
ext T T
d d
u X u p (2.4.2)
because they would not do any work during the return into the original state, on the contrary,
they would consume it, it would have to be supplied to them, the loads would have to be
lifted, etc. On the other hand, the internal forces would do some work, similarly to a wound
spring in a watch that supplies energy to overcome friction and to move the hands. This
energy is generated during the static deformation of the body gradually, following the
increase in the action of external forces from zero to final values. Therefore, potential energy
int
of internal forces equals to a half of the product of internal forces and corresponding
strains , generally in the form
1
2
int T
d
(2.4.3)
What we understand under internal forces and strains is briefly summarised under
symbols and . See also matrix G in the procedure overview stated under no. 7, art. 2.3.
In general, components and components are mutually linked through the requirement
that their product represents the virtual work, or in the case of reversible problems that do not
depend on the trajectory (conservative, scleronomic constraints, etc.) the potential energy
arising from them.
Thus, in terms of physics, everything is ready for the formulation of mathematical algorithms.
It is enough to substitute formulas (2.4.1) and no. 6 to 8, art. 2.3 into (2.4.2) and (2.4.3):
( ) ( ) ( )
1 1
1
2
e e e
T
T T
ext int
e e e e e e
d d d
+ +
Nd X Nd p BS d CBS d
For the transposition of the product we can apply the wellknown rule ( )
T
T T
Nd d N ,
( )
1 1
T
T T
BS d d S B , and thus once the parameters are factored out of the integrals
(parameters are constants, even though still unknown) and the notation according to no. 11
and 14, art. 2.3 is introduced, we get a lucid expression for the total potential energy
e
of
one element:
1
2
T T
e e e e e e
+ d f d K d (2.4.4)
For clarity reason, we introduced element subscript e for all quantities in (2.4.4). In the whole
body
e
e
the potential energy is the sum of formulas (2.4.4) in individual elements.
Before we can perform the summation, we have to swap from parameters
e
d ,
e
f , defined in
2.4 Specification of selected operations that are useful to understand FEM terms
32
element coordinates, to parameters
eg
d ,
eg
f in global coordinates that are common to all
elements in the system. Practically, this represents a transformation displacement components
of the same vector u and small rotation and components of force P and moment M in
nodal points from the element coordinates into global ones. Usually, these are rightangle
coordinates. Then we have a wellknown operation of rotation of the coordinatetrihedral
from the element position , ,
e e e
x y z into the global position , ,
g g g
x y z . For any vector, e.g. u,
we can apply the projection formulas with direction cosines that are independent on the order
of axes, as e.g. cos( ) cos( )
e g g e
x x x x , which follows from the identity cos cos( ) .
In manuals for FEM programs the user can come across both directions of the transformation,
one of which is chosen as the basic one, e.g. from global to element components:
cos( ) cos( ) cos( )
cos( ) cos( ) cos( )
cos( ) cos( ) cos( )
e g e g e g eg
e
e e g e g e g eg
e
e g e g e g eg
x x x y x z u
u
v y x y y y z v
w
z x z y z z w
1 1
1
1 1
1
1 1
1
1 1
1
]
] ]
Brief matrix notation with matrix
3
(3, 3) T :
3 e eg
u Tu
This direction of transformation, the calculation of element components from the global ones,
is usually applied at the end of calculation before the stresses in elements are determined. It
can be simply proved that multiplication
3 3
T
T T results in unit matrix 1 , it is sufficient to
apply common trigonometric theorems. Therefore, we have
1
3 3
T
T T
and the opposite
direction of the transformation is defined by the transposed matrix
3
(3, 3)
T
T :
3
T
eg e
u T u
The same matrix
3
T applies also to the transformation of vectors , P, M. The
transformation is usually written for all parameters of both the element deformation
e
d and
external load
e
f . Thus, for example, a tetrahedron with linear polynomials (art. 4.1.6.1., Fig.
4.19a) has in each of the four vertices 1, 2, 3, 4 three components , , u v w of vectors
1
u ,
2
u ,
3
u ,
4
u . Its elementcoordinates can follow some of the distinctive directions of the
tetrahedron, generally different from the global axes. Then we can write
1 1 3
2 2 3
3 3 3
4 4 3
0 0 0
0 0 0
0 0 0
0 0 0
eg e
eg e
eg e
eg e
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1
] ] ]
u u T
u u T
u u T
u u T
in the matrix form
12
12
e eg
T
eg e
d T d
d T d
with transformation matrix
12
(12,12) T :
2.4 Specification of selected operations that are useful to understand FEM terms
33
[ ]
12 3 3 3 3
DIAG , , , T T T T T
Similarly, a 1D beam element with two endnodes 1 and 2 has in general 2 6 12
parameters in space combined in vector
1 1 2 2
[ , , , ]
T
e
d u u . Their components are
transformed by means of a similar matrix
12
T , from global axes , ,
G G G
x y z into central ones
, ,
C C C
x y z and back by means of matrix
12
T
T . Components of force parameters
e
f are
transformed in the same way as
e
d .
In general, the transformation can be even more complex, it can also include other
geometrical operations, which is usually presented in theoretical manuals. The important
thing is that it is always possible to write the matrix relation:
T
eg e e eg
T
eg e e eg
d T d d Td
f T f f Tf
(2.4.5)
What remains is numbering of all deformation and force parameters globally in the whole
structure by numbers 1, 2, , N K , which is done by a special subprogram which remains
hidden from the user. The aim is to sum properly formulas (2.4.4), i.e. just those items that
belong to one particular parameter. They represent the contribution of elements sharing the
same node and common value of one parameter in it. This is done through what is termed
code number, which will be explained later. Formally, we can introduce what is called
extended quantities from element dimension n to dimension N and place the nonzero
elements to the positions where they belong by their code numbers. This can be written using
localisation matrix L whose size is ( , ) n N . In total, n elements of the matrix are equal to 1
and the remaining ones are zero. When the extended quantities are being abandoned in favour
of the element ones, i.e. during narrowing from N to n , matrix L can be called the selection
matrix:
( , ) ( , ) ( ,1) ( ,1)
( ,1) ( ,1)
eg er eg er
n N n N N N
n n
d L d f L f (2.4.6)
The element stiffness matrices
e
K are first defined in global components of the parameters:
T
eg e
K T K T (2.4.7)
and then extended to dimension ( , ) N N :
T
er eg
K L K L (2.4.8)
which will be described in detail in formula (2.4.14). That will allow for clear summation of
potential energy (2.4.4) of all elements of the system.
1
min
2
T T
e e e e e e
e e e
+
d f d K d (2.4.9)
Only this formula is minimal according to (2.2.9), not some partial items of (2.4.4). Further
modification is based on simple reasoning: dimensions of the matrices in (2.4.9) are ( ,1)
e
n d ,
( ,1)
e
n f , ( , )
e
n n K , where n is the number of deformation parameters of one element. In
practical applications of FEM to beam structures n equals to 4 (planar truss girders) up to 12
2.4 Specification of selected operations that are useful to understand FEM terms
34
(spatial frames), in structures with slabs n is 6 to 48, and even in 3D bodies it is always rather
small number, e.g. in 3D bricks it is 24 (trilinear), 60 (triquadratic), 96 (tricubic), mostly
less then 100. On the other hand, the total number of all deformation parameters of the
analysed system, structure or body is usually even for middlesize practical problems
significantly larger, approximately
3 5
10 10 N . These parameters are globally
(comprehensively) numbered from 1 to N by what is termed global indices or code numbers
1 to N , see art. 2.1:
[ ]
1 2 3 1
, , , , ,
T
N N
d d d d d
d K (2.4.10)
The definition of the state of each element e requires only n of these parameters that do not
have to go in the sequence in the global numbering as, in general, the nodes of the domain
mesh (from the numbers of which the global indices of the parameters are derived) cannot be
numbered this way. However, we can always define a relation between the local indices
1, 2, , n K and global indices
1 2
, , ,
n
J J J K belonging to the same deformation parameters,
which are the first time perceived as a set of deformation parameters of element e , the second
time as a part of all deformation parameters of the whole system. The information contained
in selection matrix L in formula (2.4.6) can be simply coded in the following way:
[ ]
1 2 3 1
1 2 3 1
, , , , , , , , , ,
n n
T
T
eg n n J J J J J
e
GLOB
d d d d d d d d d d
1
]
d K K (2.4.11)
What is this primitive, almost entirely administrative commonplace in notation, good for?
We will show that after a certain amendment or extension, we get the fundamental
assumption for the success of FEM algorithms that gives the right to apply what is called
addition theorem (assemblage), which is exceptionally suitable for computers.
The extension consists of the following: the matrix vector ( ,1) n according to (2.4.11) is
written into a grid ( ,1) N in such a way that positions to which no value is attributed are set to
zero:
1 2 3 4 1
0, 0, , 0, , , 0, 0, , , 0, , 0, , , 0, , 0, 0
n n
T
er J J J J J J
d d d d d d
1
]
d K K K K (2.4.12)
In the presented example, the nonzero elements emerge in pairs, which is, for instance, the
situation in planar elasticity problems with components u, v, but it is just a sample, not a rule.
The appearance can be fully arbitrary. In practice, especially after the optimisation of the band
width of global matrix K , the appearance of nonzero elements has cumulative nature they
form a kind of an island or a group of islands in the sea of zeros. Expression (2.4.12) is called
extended or localised (positioned) vector of element deformation parameters
er
d . In addition
to information (2.4.11) it holds also important topological information about the position of
the element in the system. That is, however, already specified by n global indices
1
J to
n
J
written precisely in the order of local element indices 1, 2, , n K :
1 2 3 1
, , , , ,
n n
J J J J J
K (2.4.13)
We briefly call expression (2.4.13) the global code number of the set of deformation
parameters of element e . Each element has different number (2.4.13). Strictly speaking, it is
not a number, but a string composed of n integers.
Let us go back to formula (2.4.9). It is quite clear that vector of load parameters ( ,1)
e
n f can
2.4 Specification of selected operations that are useful to understand FEM terms
35
be similarly extended to ( ,1)
er
N f as well. Each
1 2
, , d d K is virtually associated with just one
parameter
1 2
, , f f K on which the virtual or possibly the real work (potential energy
z
is
work that must be done to return the load to the position in the original nondeformed state) is
done. But also square matrix ( , )
eg
n n K , see the overview, no. 11 to 13, art. 2.3., can be
extended to the order ( , ) N N and marked as ( , )
er
N N K , localised stiffness matrix of element
e . It is sufficient to write its elements k in rows 1, 2, , i n K and columns 1, 2, , j n K
symbolically into rows
1 2
, , ,
n
J J J K and columns
1 2
, , ,
n
J J J K of square matrix ( , ) N N :
1 2 3
11 12 13 1 1
21 22 23 2 2
31 32 33 3 3
( , )
1 2 3
1
1
n
n
n
er n
N N
n n n nn n
J J J J N
k k k k J
k k k k J
K k k k k J
k k k k J
N
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
]
K K K K K
M
M
M
M
M
(2.4.14)
What are all these extended notations good for? They enable us to write potential energy
(2.4.4) of one element e first in a completely equivalent form
1
2
T T
e er er er er er
+ d f d K d (2.4.15)
from which there is only one small step towards the final form. That is obtained when
er
d is
replaced by the vector (column matrix) of all deformation parameters of the analysed system
d:
1
2
T T
e er er
+ d f d K d (2.4.16)
This replacement is possible for a simple reason: from the whole vector d only nonzero
elements
er
K act in mentioned multiplication operations, while other elements of vector d
are only multiplied by zeros, which does not influence the value of energy (2.4.16). The
simplicity of this reasoning, easily verifiable in practical examples with a small number of
parameters N , is the reason why it is not in fact explicitly mentioned in any FEM text. This,
however, is annoying for curious readers as the whole vector d suddenly appears in the
derivation of the procedure. Even the famous monographs [1], [2] use incorrect formulation
such as " K is the sum of
e
K ", even though the summation operation is defined only for
matrices of the same order, i.e. summation of matrices ( , ) n n cannot result in matrix ( , ) N N !
2.5 Principle of virtual work applied in FEM programs
36
The detailed explanation that is important not only for programmers, but also for anybody
who wants to understand principal advantages of FEM in comparison with other methods is
given in books [4]. Adding the elements of the stiffness matrix into the appropriate positions
in the global matrix was briefly named the addition theorem. Even though that this term
penetrated into the international literature thanks to the translation of [5] already in 1975, the
presentday FEM texts still use vague formulations such as arithmetic sum of matrices, or,
most often, just the term assemblage. This applies to almost all theoretical manuals and
users guides of FEM programs as well.
Expression (2.4.16) is the final form of potential energy
e i z
+ of one element of the
system. Next steps represent just a simple summation of the energies of all the elements as
scalar values into total energy of the analysed system with common d factored out:
1
2
T T
e er e
e e
_
+
,
d f d K d (2.4.17)
marking of global stiffness matrix K as vector of load parameters f :
er er
e e
K K f f (2.4.18)
1
2
T T
+ d f d Kd (2.4.19)
application of the condition of the minimum:
+ f Kd 0 (2.4.20)
and rewriting of the obtained system of equations in usual form with vector of righthand
sides and matrix of coefficients K :
Kd f (2.4.21)
This reasoning is in full applicable to arbitrary functionals (operators) that are additive in
their nature
e
e
2.5 Principle of virtual work applied in FEM
programs
First, let us focus on the most frequent FEM applications in technics, i.e. mechanical
problems, that are in general nonlinear but finite, which means that the solution is
independent on the history, i.e. on the conditions that the analysed system was exposed to
prior to the analysed state. Or more precisely: we will assume that the system of external and
internal forces is conservative and that the geometrical constraints are scleronomous. The
following text brings just brief information for the users of FEM programs.
2.5 Principle of virtual work applied in FEM programs
37
The abovementioned examples of mechanical problems allow for a simple application
of the oldest known mechanical principle of virtual work (Archytas of Tarentum 400, Stevin
1550, Galileo 1600, Newton 1678, Huygens 1700, Mach 1880). Its axiomatic extension to
nonrigid bodies and systems (Navier 1830, Lam 1860, SaintVenant 1870, Mohr 1900)
includes also the work of internal forces into the virtual work.
Let's define a virtual work of external forces:
virt virt
P Pi i i
i i
W W P r
(2.5.1)
and internal forces
virt virt
S S j j j
j j
W W P s
(2.5.2)
where the sign (minus) in (2.5.2) is introduced axiomatically, i.e. from the experience
with the accumulation of a part of the work (energy) into the body. The principle of virtual
work then axiomatically says (without any proof) that for every system of forces { , }
i j
P S that
are in equilibrium and for every virtual displacement { , }
i j
r s the total virtual work of
external and internal forces equals to zero:
0
virt virt virt
P S i i j j
i j
W W W P r P s +
(2.5.3)
Several popularisation formulations have been created with the aim to explain the principle in
an intelligible way. E. Mach analyses them in a voluminous chapter of his famous
Mechanics (7
th
edition [24]) and he sums up them in: ES GESCHIEHT NICHTS, WENN
NICHTS GESCHEHEN KANN, i.e. if nothing can happen (i.e. there is no motion in any
direction under which the impulse of force Pt could change into momentum mv, etc.), then
nothing really happens (i.e. there is no motion, it is an equilibrium steady state).
If we have distributed loads and internal forces, the sums from (2.5.3) become
integrals, which is nothing important, just a formal modification of the notation of the
principle of virtual work that may in general contain both sums and integrals depending on
the type of loads (point, linear, surface, volume) and internal forces (1D, 2D, 3D models,
singularity) and we may have a component, tensor or matrix notation.
As the work is scalar, we can add together work done in parts
e
(subdomains,
elements) of domain into the total work to which the principle applies. Should it happen
that a system contains parts that have no connection to the remainder of the system, we can
simply find out through the application of a virtual displacement that is nonzero only in
such parts that the principle of virtual work applies to every such unconnected part
separately, which is correct as the bodies (domains) are independent. Therefore, we can limit
ourselves only to systems (domains) where each part
e
has at least one connection with
another part . There is an unlimited number of such connections in classical continuum
divided into elements
e
. They are geometrically described by the SaintVenant equations of
compatibility. We will, however, consider more general systems consisting of elements of
various dimension and nonclassical continuum. The parameterisation of deformation degrees
of freedom of such systems ensures that the analysis of their behaviour is integrated into the
2.5 Principle of virtual work applied in FEM programs
38
standard procedure.
The principle of virtual work (2.5.3) applies to arbitrary nonrigid bodies and was
axiomatically extended also to elastic, or more generally deformable, continuum of 1D, 2D,
3D dimension. We employ symbols from the overview in art. 2.3. and from notation:
i
P : External volume and surface forces b, p ( ,1) d
i
r : Displacement
( , ) ( ,1) d N N
u N d ( ,1) d
j
S : Internal forces
( , ) ( , ) ( ,1) s N s s N
C B d ( ,1) s
j
s : Strains on which work
( , ) ( ,1) s N N
B d ( ,1) s
All the terms are perceived as generalised in such a way that they contain not only the
classical 3D continuum 3D ( 3, 6 d s ), but also e.g. planar 2D problems ( 2, 3 d s ),
axially symmetrical problem ( 2, 4 d s ), Mindlin shells ( 6, 8 d s ), etc. Statical and
also geometrical quantities are written in the form of matrix vectors, i.e. column matrices, the
dimension of which is added in brackets. External forces b, p are given. Other quantities are
assumed as depending only on N parameters d, similarly to the derivation of the FEM
procedure, i.e. after the parameterisation of the system has been carried out through the
selection of base functions that is closely related to the division into finite elements and to the
chosen polynomial distribution over elements. We have already put down the global forms of
these dependencies on d, that is for the whole analysed system, using matrices N , B
extended to dimension ( , ) d N , ( , ) s N . Let us remind that in dimension ( , ) a b the a is the
number of lines and b the number of columns of the matrix.
For linear problems it is sufficient to take the final state of the system after loads and
deformations have been applied as the equilibrium system { , , } b p . The following N
displacements of the system are sequentially chosen as the virtual displacement:
1
2
(1) 0 0 1
(2) 0 0 2
( ) 0 0
( ) 0 0
i
i
j i
N i
d d i
d d i
j d d i j
N d d i N
M
M
(2.5.4)
Let us call them unit parametric states as just one parameter is nonzero in these
states. In textbooks we often see its size set to one, which is neither necessary nor illustrative,
as it soon disappears from the derivation and the physical meaning is not apparent. The
following virtual displacements and strains correspond to the abovementioned parameters:
2.5 Principle of virtual work applied in FEM programs
39
1 1 1 1 1 1
2 2 2 2 2 2
( ,1) ( ,1) (1,1) ( ,1) ( ,1) (1,1)
(1) u
(2) u
( ) u
( )
j j j j j j
N N N N N N
d d s s
d d
d d
j d d
N u d d
N B
N B
N B
N B
M
M
(2.5.5)
if we use subscript j to mark the row submatrices of which the global matrix shapes N are
composed
1 1
2 2
( , ) ( , ) d N d N
j j
N N
1 1
1 1
1 1
1 1
1 1
1 1
1 1
1 1
1 1
] ]
N B
N B
N B
N B
N B
M M
M M
(2.5.6)
With the selected notation, the principle of virtual work (2.5.3) now gets for the continuum in
domain with boundary the form:
(1,1)
( ,1) ( ,1) ( ,1) (1, ) (1, ) (1, )
T T T
v v v
d d s d d s
V d d d
+
u b u p (2.5.7)
where
v
u is an arbitrary virtual displacement of the points of the continuum that may be
subjected to volume and surface external forces b, p and
v v
u is the virtual deformation
derived from it. If we choose the parametric state 0
i
d as
v
u , principle (2.5.7) produces one
equation with the defined notation:
( ) ( ) ( )
T T T
j j j j j j
d d d d d d
+
N b N p B CBd (2.5.8)
After the transposition of products in the form ( )
T
T T
AC C A and factoring of d out of the
integrals, while for scalar
j
d (one number, matrix (1,1) ) we have
T
j j
d d , we get:
( ,1)
( ,1) ( ,1) ( , ) ( , )
(1, ) (1, ) (1, )
0
T T T
j j j j
N
d d s s s N
d d s
d d d d
1 _ _
1 +
1
, , ]
( ( (
N b N p B C B d (2.5.9)
This can be satisfied for nonzero
j
d only if the expression in the box brackets equals to zero.
When comparing with no. 11, 14, art. 2.3 of the overview, it can be found that the expression
in the first round brackets is the j th parameter
j
f
of vector f of all load parameters and the
expression in the second round brackets is the j th line
j
K of global stiffness matrix K of
the system. Therefore, after rearrangement to both sides of the equals sign, equation (2.5.9)
reads:
2.5 Principle of virtual work applied in FEM programs
40
j j
K d f
(2.5.10)
If we gradually select as virtual displacements all N parametric states, i.e. 1, 2, , j N K , we
obtain N equation of type (2.5.10), in other words the whole system
Kd f (2.5.11)
that is identical with system (2.4.21) derived through the Lagrangean variational principle in
the previous art. 2.4. from the condition of min .
Why is the system of equations, i.e. also the solution, identical? There are two reasons
for this: mechanical and mathematical. In terms of mechanics, the principle of virtual work is
superior to all equilibrium conditions and it can generate an arbitrary number of such
conditions as there exists an arbitrary (infinite) number of selectable virtual displacements. It
is superior to the Lagrangean variational principle min , or, as the case may be, 0
that is one of its consequences. In terms of mathematics, we have limited in both situations
the deformability of the system to states that are a linear combination of the same unit
parametric states, which reduces the unlimited freedom of the continuum to N degrees of
(deformation) freedom that are the same in both procedures.
Is there any advantage in the derivation through the principle of virtual work in comparison
with the Lagrangean variational principle? Yes, and the advantage is significant, as during the
derivation we do not have to limit ourselves just to problems of classical equilibrium that is
independent on the trajectory (= history proceeding to the analysed state). The system
{ , , } b p must be in equilibrium in every time instant and thus also the increment at time dt ,
{ , , } d d d b p is in equilibrium and the principal of virtual work can be applied to it. This
allows for the application of FEM solution in steps, increments of load using what is
called an incremental method. The main problem is then how to add together increments d,
d, du in a way that is consistent with the laws of physics if deformations are large. Instead of
details, let us give only brief notes that make the modelling of structures for inputs needed
in FEM programs easier:
1. If body is subjected to singular forces concentrated in figures of lower dimension
than that of (e.g. for 3D body on 1D lines or in 0D points "i"), then we can
use directly the definition of such loads: the measure of the area on which it acts
converges to zero but the resultant remains constant, as the intensity increases in
indirect proportion to the measure. Consequently, instead of integrals in (2.5.9) se use
directly the sums of products of the resultants with the virtual displacements (or
moments with rotations), i.e. what is termed concentrated impact, see overview no. 15,
art. 2.3. For linear loads we use line integrals, always with the aim to get the virtual
work in the form of a sum or integral of products of the force and distance. It is thus
possible to calculate load elements f , i.e. vector f of the righthand sides of the set of
equations Kd f , for a very wide set of loads. FEM program are usually limited to :
Loads distributed over elements, of general direction, decomposed into components
, ,
x y z
p p p or , ,
x y z
b b b ,
Concentrated forces with components , ,
x y z
F F F , or also moments , ,
x y z
M M M in mesh
nodes with deformation parameters , , u v w, or also , ,
x y z
,
Linear, mainly straightline, loads on element boundaries with components into the direction
2.6 Main outcome for the users of FEM programs
41
of axes , , x y z . The stateoftheart programs allow for the selection of various coordinates,
e.g. global or planar (element) axes. The given load is approximated by this modelled load,
meeting the requirement that the statical resultants must be identical. There should be a node
in the point of action of the concentrated load or moment, which is taken into account already
in the phase of division of the domain (generation of the mesh).
2. The global (localised, extended) forms of matrices N and B used in the derivation
hold, apart from the information on element matrices
e
N ,
e
B , only the topological
data about the position of the element in the globally numbered mesh of nodes that is
stored in element code numbers (2.4.13). These can be created from the sorted global
numbers of its nodes with the inadvancenullified deformation parameters taken into
account, as explained in art. 2.4. This mass notation allowing for mathematically
correct summation of matrices of the same type does not mean that the huge
matrices are really created and stored in the memory. The program employs the
addition theorem and sends the corresponding items of individual matrices of
elements e where they really belong to according to their global code number and
thus gradually creates 1
st
, 2
nd
, ..., to N th equation. As we almost always have band
matrices K with nonzero elements only in a band with the width of BW on the
left and righthand side of the diagonal, and as matrix K is symmetrical, only this
band is stored in the memory. Some programs combine the assembly with the
solution, or presolution or with the transformation into a triangular matrix, which
means that no equation is stored in the original form. Other programs first assemble
and store the whole system Kd f .
3. Some programs do not use optimisation (almost minimisation) of the band width BW ,
but take advantage of a small density of matrix K (schwach besetze Matrix, sparse
matrix) instead in order to speed up the solution of the system of equations. The
presentday FEM programs enable the users to number the nodes by themselves or to
renumber the nodes in such order that suits their particular needs. Any potential
renumbering is performed just internally and the communication with the user is done
in the users numbering.
2.6 Main outcome for the users of FEM
programs
2.6.1 Selecting the elements of the FEM analysis model
I. In each FEM program the corresponding analysis model is defined at the beginning of
the manual. This is usually done in the description of the applied finite element.
Each real object is threedimensional, which means that 3D elements do not cause
major problems. The most frequent type today is what is called bricks, or more
precisely hexahedrons or dodecahedrons, with planar or curved sides and straight or
2.6 Main outcome for the users of FEM programs
42
curved edges. Usually, they represent shapes that are formed through a transformation
of a cube with sides equal to 2 with vertices (1, 1, 1). Other spatial elements with
details about the distribution of displacement components ( , , ) u x y z , ( , , ) v x y z ,
( , , ) w x y z in various elements can be found e.g. in chapter 3 of title [5]. The main
issue that need to be addressed in practice is the problem of accurate physical
constants, unless we deal with elementary examples of homogenous and isotropic
matter with two easily determinable constants E , . Physical constants for general
configurations of nonhomogenous and anisotropic bodies are specified by means of
21 functions ( , , )
ik
d x y z . As FEM programs perform the integration within one
element numerically, typically using the Gauss procedure, only values
ik
d (in general
any physical quantity) in the nodes of the integration mesh apply. For the simplest
situation of onepoint integration (in 3D space it is an analogy of a staircase
diagram) only the value in the centroid of the element applies, which is the point
where the origin (0,0,0) in the original cube is located. Depending on the required
accuracy, various 3D bricks are nowadays used: especially what is termed brick 24,
60, and 96, art. 4.1.6.2, fig. 4.19, with trilinear, triquadratic and tricubic distribution
of displacement components , , u v w in one element (for details see formulas (4.1.65) 
(4.1.82)).
II. Analysis models of FEM programs for what is termed planar structures (i.e. walls,
plates and shells) are of 2D dimension. It means that one of the geometrical
dimensions, thickness h, is missing in the geometry. This is the source of traditional
troubles in the creation and selection of the analysis model. A whole hierarchy of
options how to tackle the missing thickness h, or more precisely how to tackle the
distribution of displacement components in the interval 2 2 h z h along h , has
been established over the years. This includes the Reissner sequences that expand the
displacement components , , u v w into Taylors series in variable z , see art. 4.1.6.
Thus, a countable set of plate or shellmodels can be created, of which mainly two
appear in the presentday FEM programs:
e) The oldest known model is the Kirchhoff one. It assumes that component w along h
is constant (common deflection of a rigid normal h ) and that components , u v use
one member of the Reissner sequence, linear in z , which introduces a physically rigid
mass normal h . Moreover, it is assumed that this normal remains normal even after
the deformation of the midplane (plate) or midsurface (shell) into a general
deflection surface. Therefore, the deformation of an element is uniquely defined by
means of a single function ( , ) w x y . The difficulty in FEM programs is that the
compatibility between elements (no breaks allowed) requires that not only function w,
but also derivatives w x , w y are continuous. FEM terminology uses term class
of function
1
C
(subscript 1 means continuity up to the 1
st
derivative). Hundreds of
publications that deal with this problem for triangular and quadrilateral elements were
presented between 1956 and 1976. They introduce what is called rational correction
functions for correction of common polynomials, etc. An overview of the hierarchy of
elements without correction can be found in [5]. For further reading about correction
functions refer to [2], and the literature cited there. Some companies today treat these
functions as a business secret, in order to hide them from competitors. Their
importance significantly decreased as new elements come into play see point (b).
2.6 Main outcome for the users of FEM programs
43
What is important for users is that internal forces are differentiated out of function
( , ) w x y using rule no. 19 from the overview in art. 2.3. According to [31] chapter 3,
pp. 212 223, bending moments
x
m ,
y
m and torque
xy
m are determined by second
derivatives, shear forces
x
q ,
y
q and reactions r by third derivatives of w with respect
to , x y . The lowest degree of polynomial ( , ) w x y is thus equal to 3 if it is supposed to
generate nonzero transverse shear. Classical compatible elements, as they were
called, used degree 5 (21 coefficients) on a triangle and Ahlins bicubic polynomials
on a quadrilateral. It was this disharmony between the 10 coefficients of a polynomial
of third degree in x,y and 3 triangle vertices (which means that for the selection of
parameters w, w x , w y there is only 9 parameters) that started explosion of
publications in 1960 1970 where also correction functions were developed. As also
these functions apply to derivatives and as they are rational, not every modification
results in reasonable approximations of
x
m ,
y
m ,
xy
m and especially
x
q ,
y
q , r . After a
decade of experience authors of these publications also admit that transverse shear is
obtained more reliably when elements from paragraph (b) are applied.
f) A newer model, the Mindlin one, seems to be only a small generalisation of the
Kirchhoffs model: It does not include the requirement that the rigid normal h remains
normal in the deflection surface. The right angles between directions ( , ) x z and ( , ) y z
can alter by slope
xz
,
yz
that are permanently zero in the Kirchhoff approach, which
can be demonstrated using what is called pin model. Pins jabbed up to their centre
point into a rubber midplane represent normals h , the heads 2 z h represent the
top face and the peaks 2 z h are the lower face of the element. In the Mindlin
approach these normals have, in addition to deflection w, two more degrees of
freedom
x
,
y
components of rotation about the x axis and y axis that are
considered a vector, which is possible only for very small rotations, i.e. in
geometrically linear mechanics of small deformations, both displacements and
rotations (classical linear theory). The disadvantage that three unknowns
( , ) w x y , ( , )
x
x y , ( , )
y
x y
are introduced is outweighed by the fact that instead of one differential equation of
fourth order we get one equation of second order (equilibrium of forces in the z
direction) and two equations of first order (moment equilibrium about x axis and y
axis). The impact on the FEM theory is that the requirement on the continuity of base
functions is reduced by one order, i.e. from class
1
C to class
0
C , which represents
formulas that are continuous only in function values (0
th
derivative = the function
itself). This eliminates complications that are solved, among others, also through
correction functions. Moreover, it is sufficient to choose all three functions as linear
polynomials in the form a bx cy + + . As a result, we get 9 unknown coefficients on a
triangle and a regular expression with 9 deformation parameters w,
x
,
y
in its
vertices. This basic element can be then improved in various ways, e.g. through the
rule of Kirchhoffstyle permanence, with the tendency to have the polynomial for
x
and
y
one degree lower that the polynomial for w (as in the Kirchhoff approach we
have
x
w y ,
y
w x ). The first improvement (still used in presentday
2.6 Main outcome for the users of FEM programs
44
FEM programs) was presented already by Lynn and Dhillon at the 2
nd
Conference on
Matrix Methods Japan  USA in Tokyo in 1971. Another substantial improvement
relating to the extension to spatial shell systems was made in 1977 by Kolar and
Nemec. Similar elements are now implemented in all prestigious FEM programs.
The most important fact for users is that internal forces of the Mindlin model
x
m ,
y
m ,
xy
m ,
x
q ,
y
q depend only on the first derivatives of functions w,
x
,
y
(at
x
q ,
y
q we
have in addition to w x , w y only the functions themselves
x
,
y
, i.e. not
differentiated!). This ensures better numerical reliability, especially when no rational
correction functions are necessary. For the abovementioned elementary triangle all
the moments in the element are constant and shear forces have linear distribution. It is
an inverted ratio in comparison with Kirchhoffs elements (without correction
functions) where, primarily, the distribution of shear forces is one degree lower than
the distribution of moments.
III. Similarly to paragraph II, 1D beam elements can be defined in the Kirchhoff style
(more precisely using the BernoulliNavier hypothesis about the preservation of planar
sections in normal planes of deflection curves) termed classical beams (classical 1D
continuum)) or in the Mindlin style i.e. with the transverse shear taken into
account. For the first approach, the distribution of all the three displacement
components , , u v w in the body of the beam can be derived from four functions of one
variable s on the (generally curved) beam axis.
( ), ( ), ( ), ( ) u s v s w s s
and their derivatives with respect to s . In the second approach, we get, in terms of
geometry, a Cosserat 1D continuum in which each point has 6 degrees of freedom
( ), ( ), ( ), ( ), ( ), ( )
x y z
u s v s w s s s s .
The users can see the following differences (see also art. 4.1.4.):
In general, a classical straight beam element in space has in the central (principal
centroidal) axes , , y z x at each end 6 parameters of type , , , , , u v w dv dx dw dx , in
total 12 parameters, with corresponding end components of force and moment vectors
, , , , ,
x y z x y z
P P P M M M . Axial effects , u (tension/compression and torsion) are
distributed linearly over the element and transverse effects , v w (bending and shear)
by polynomials of third degree.
A similar Mindlin element has at its ends 6 parameters, independent degrees of
freedom of endsections , , , , ,
x y z
u v w . The requirements of FEM applications
would be satisfied if the six functions were selected in the form of linear polynomials.
The element is however significantly improved if we take into account the well
proven Kirchhoff principle, i.e. if we select a one degree higher polynomial for , v w
the polynomial of second degree. In order to guarantee compatibility with 2D
elements in space it is necessary to rise the degree of polynomial u accordingly. This
results in an element with three nodes ( 0) I x , ( ) J x L , ( 2) K x L , six
parameters in , I J and just three parameters , , u v w in K , in total 15 parameters. It
can be reduced through energetic elimination of internal parameters in node K into an
element with the same number and location of parameters at ends , I J as the classical
2.6 Main outcome for the users of FEM programs
45
element. Practical applications have proven that, provided the mesh is of normal
density, the compatibility between 1D and 2D elements only in nodes is sufficient. On
the other hand, it is useful not to neglect the effect of transverse shear.
Some FEM programs introduce the effect of shear in 1D beam elements through a
systematic procedure. Slopes
xy
,
xz
due to shear forces ,
y z
Q Q are considered to be
13
th
and 14
th
parameter added to the 12 common abovementioned parameters. The
energetic elimination then modifies the stiffness matrix into the classical dimension
(12,12) and with elements that include the effect of shear in what is termed shear
factors , see also chapter 2.
The important factor in modelling of a real structure is not the shear factor of elements, but
of the whole structural members that can be, of course, divided into several elements (curved
beams, beams with haunches, etc.). The same applies to plates. Socalled thin and thick plates
are considered depending on the ratio h L , where h is the thickness and L the characteristic
dimension of the whole slab (diameter, shorter side of a rectangle, etc.), not the dimension of
its element. The limit value is usually the ratio 1 5 h L . Thinner plates can be analysed by
FEM programs with Kirchhoff elements, which may be thick themselves if the mesh is
dense enough, it may even happen that h L > . For thicker plates it is better to apply FEM
analysis using Mindlin elements. Most such elements feature a builtin numericallystabilising
test that allows for convergence to a reasonable result even for thin plates. The feature is
termed shear locking. A numerical limit value over which the shear no longer increases, must
be introduced.
2.6.2 Interpretation of FEM output data
In order to meet the needs of design practice, sizing, checking of resistance and
assessment of serviceability, the outputs of the FEM procedure are further processed either
manually or in numerical or graphical postprocessors. The following principle is valid:
FEM programs with elements whose dimension is n D hold in the outputs only such
an amount of information of dimension mD, m n > that is embedded into n D finite elements
through hypotheses that reduce the dimension from m to n . In particular: FEM programs
with 2D elements cannot be used for a detailed 3D stressstate analysis in such places of 2D
structures (walls, plates, shells) that require such an analysis (the vicinity of column heads in
flat slabs, bridge bearings, concentrated impulses acting on area smaller than
2
h , where h is
the thickness of the planar structure, etc.). Even less applicable to the 3D analysis are the
outputs from FEM programs that use 1D elements. For example, such programs are not able
to tell on which particular part of the crosssection the load or reaction acts (top or bottom
flange, etc.), as only the resultant over the crosssection is known. Therefore, it is not possible
to perform an accurate analysis of the stressstate in e.g. fixed ends where the specific design
of the fixing cannot be taken into account or in connections where the exact welding, bolting
or gluing characteristics cannot be considered, etc.
If the real design is modelled through a kind of abstraction, e.g. if a foundation strip is
2.6 Main outcome for the users of FEM programs
46
modelled by a rigid 1D line, the singular outputs from the FEM analysis (that converges to
exact singularities) cannot give an accurate idea about the detailed stressstate in the vicinity
of the singularities. A typical example: If we define a rigid line support of a bridge deck,
either perpendicular or skewed, we get nondesignable moments in the corners of the deck,
which is a correct trend towards the singularity of the corner. If we want to have realistic
outputs, it is sufficient to specify the real stiffness of the supporting beam that can never be
infinite, which is included in the definition of a rigid support with 0 w , moreover with a
zero area measure
z
. Even the lowest estimate of the stiffness of the line support or
the lowest estimate of the spring constant k in older programs that did not allow for such a
support type produces usable moments. Hundreds of similar situations can be summed up to
express the following principle:
If we want correct FEM results, it is sufficient to exploit the wide capabilities of FEM
in order to express correctly the real design and realisation of the structure and its
supports and the real distribution of loads.
In this context, mistakes are made by both younger engineers (due to the classical
tuition at faculties) and experienced specialists who were during their long practice forced
to simplify everything (loads, supports, material properties, etc.) as program with the power
of presentday FEM systems were missing.
Many engineers learn about the effectiveness of FEM at various courses, but they
doubt about the perplexity of input data for complex spatial systems consisting of thousands
of 1D and 2D (and possibly also 3D) elements, which usually prevented the full utilisation of
FEM in the previous era of development (in the Czech Republic approximately until 1980).
The main idea of the division of a domain into finite elements is however much stronger and
it includes also the following capabilities:
g) The analysed structure is first divided into smaller substructures, which makes the
division more understandable. The substructures may have technological and
production functions, but may also be parts selected only on a formal basis, in order to
improve the handling of the model. Sometimes, it is possible to combine (i) parts
made of 1D elements and other parts composed of 2D elements, sometimes (ii)
individual walls and ceilings, foundation slabs, pile foundations, etc. If we use this
division just with the aim to make the handling of inputs more comfortable, it is better
to talk about macroelements and sometimes only about segments. If the program
creates for these parts separate global stiffness matrices
e
K and vectors of load
parameters
e
f and if the addition theorem is applied to them (no. 13 and 17 of the
overview in art. 2.3.), then we can refer to real substructures in terms of FEM. What is
remarkable about this procedure is that the substructure can have much more internal
nodes than boundary nodes through which it is connected to other substructures.
Only the parameters of these joint nodes then appear in the global stiffness matrix K
and in the vector of load parameters f of the whole structure. This reduces the
number of unknowns N and also the band width BW at the cost of certain
preparatory works resembling preelimination which is just the formation of
matrices
e
K and
e
f .
h) Creation of what is called superelements by assembling simpler elements, called
subelements, is not in fact too different from (a), but it has slightly different technical
meaning. One of the oldest superelements is a quadrilateral composed of four triangles
2.6 Main outcome for the users of FEM programs
47
or a hexahedron (i.e. a brick with the shape of a cube, block, or parallelepiped with
either planar or curved faces) composed of five tetrahedrons. The meaning is the
following: we apply simple base functions U (no. 1 and 4 in the table in art. 2.3.) in
the subelement. These functions usually impose a significant deformation
restriction, physical strengthening of the element by means of fictive connections that
impose prescribed displacements. This restriction is somewhat relieved for the
superelement. Its internal parameters are regularly (in compliance with the energetic
principle) eliminated and the result is an element that is softer in terms of
deformation than it would be after a simple addition of the subelements. Let us
present one illustrative example: if we use for a triangle linear functions U , i.e.
monomials 1, , x y , or for a tetrahedron 1, , , x y z , then we force it to displace along the
plane or hyperplane inserted into 3D space. The created quadrilateral can move along
four partial planes (tetrahedron along five partial hyperplanes), which can already
approximate more complex, outofplane deformation. The reduction of the number of
unknowns N and band width BW through the eliminated parameters is also
welcome.
In addition to the abovementioned elimination, also a simple interpolation of parameter
values is sometimes used, mainly with the aim to exclude undesired nodes in the middle of
the sides. On side 123 we thus set ( )
2 1 3
2 d d d + (similarly for more complex geometry)
and linear or quadric interpolation is applied. The principle of compatibility between
adjacent elements must be satisfied, i.e. the elimination must generate the same values of
deformation parameters in both elements. The users are advised to pay attention to the
definition of the element in the manual as it can make clear some characteristics of the
element that may show up in the output interpretation of the results. Let us describe a simple
example: let us have d w slab deflection without other parameters (angles or derivatives).
If
2
d is eliminated through a linear interpolation, then for supporting conditions
1
0 d ,
3
0 d , side 123 is straight with all possible consequences for the internal forces in the
element. If
2
d is eliminated energetically, its value may be different and the boundary of the
element can be curved, which produces different internal forces. The situation is different if
additional nodal parameters are defined. The requirement of 0 d in nodes does not have to
mean 0 w at the edge of the plate, i.e. exact linear support, but just skewering of the plate
on the vertex nodes of the element. This has an impact on the output internal forces and must
be interpreted appropriately with respect to real support conditions required, see chapter 5.
3.1 Principles of approximation of the sought distribution of a quantity in FEM
48
3 Physical axioms and variational
principles of more complex FEM
problems
3.1 Principles of approximation of the sought
distribution of a quantity in FEM
Every single day, professionals in engineering practice face the problem how to
reliably forecast the behaviour of various objects under different conditions, i.e. the response
of the analysed system to external influences that cause changes the mechanical state of the
system, or simply: how to reliably forecast the behaviour of the object and its surroundings
(soil environment, subsoil, etc.) in the production phase (construction, assembly), in service,
in exceptional situations (accidents, failures) and in what is termed limit states (ultimate,
serviceability, deformation, cracks, fatigue).
With regard to the needs of practice, this is not a mathematical problem and no theory
is required. The only necessary thing is to comply with all the clauses of valid standards that
concentrate manyyear experience and that reflect the production possibilities in a particular
state. If this is feasible for common objects without any special calculations, i.e. if it is
possible to prove that the design is safe, economical and even optimal and if it can be
evidenced that the assessment is reliable without any additional analysis exploiting modern
methods, then no engineer will do that. In practice, there is no time and there are no means for
such mathematical amusement. In todays competitive environment, it is becoming more
and more a must for companies to be innovative and implement new things. Also increasing is
the number of cases when it is necessary to perform calculations using such methods that are
appropriate to the nature of these new things, if the organisation is supposed to do everything
that can be reasonably required in order to prevent any damage. No methods are used in
practice just for fun, but only due to the instinct of self preservation. It is a certain tool for the
fight against nature, even though it is different from the famous stick of a primeval man,
however, with the same purpose in the engineering field where only the mustbesolved, and
not the canbesolved, is solved (contrary to the realm of mathematics where, and it is right,
the canbesolved is solved).
Art. 1. presented a brief overview of different FEM procedures that are available in
nowadays practice for the analysis of various engineering problems. The typical feature of
these problems is that we seek the distribution of certain quantities and their extreme
values, which means that we work with functions.
Briefly and in popular terms, we can say that the final aim of almost every numerical
method is to represent the sought solution, in general a set of functions (e.g. , , u v w, etc.) of
several variables (e.g. , , , x y z t , etc.), as a vector function, e.g.
[ ] ( ) ( , , , ), ( , , , ),
T
U x u x y z t v x y z t K (3.1.1)
3.1 Principles of approximation of the sought distribution of a quantity in FEM
49
approximately in the form of a linear combination (sums of products) of preselected
functions ( )
i
u x , 1, 2, , i n K with coefficients
i
a :
1
(x) ( )
n
i i
i
aU x
u (3.1.2)
which can be most concisely written in matrix notation according to no. 4 from the overview
in art. 2.3:
u Ua (3.1.3)
Functions u can have an arbitrary physical meaning, in mechanics they may represent any
geometrical or statical quantities, in the deformation variant of FEM (used to illustrate finite
elements methods in art.1.2. and 2.2.4) usually the displacement components (or their
derivatives) and small rotations. Their values in one point of the solution of the system are
written in a column matrix, called also matrix vector.
The historically oldest vector is vector of displacement [ ] , ,
T
u v w u
r r r r
in Euclidian 3D space
(D denotes the dimension of the space). Let us mark the unit vectors in the direction and
orientation of x , y , and z axis by symbols , , u v w
r r r
and the magnitude of the components
by
1 1 1
, ,
u v w
u a u v a v w a w
r r r r r r
We can apply the decomposition of the vector into three
components, in other words the vector sum
1 1 1 u v w
a u a v a w + + u
r r r r
(3.1.4)
which can be symbolised by a diagonal in a parallelepiped with the sides equal to the
components. This approach was used to add vectors already in antiquity, either axiomatically
or with attempts for proofs. Only at the end of the 19
th
century, the correctness of this
operation was consistently axiomatically analysed using the principles of independence and
superposition (E. Mach: Die Mechanik, 7
th
edition. 1912), see art. 3.2.
Analogically, formula (3.1.3) can be called the decomposition of function, or a set of
functions (i.e. vector function, with an arbitrary number of components, it is a mathematical
vector column matrix, not a physical vector), into components U . If these components are
known, or chosen, then function u is determined only by the set of coefficients a , which is
practically always a finite number of numbers and even for n it is a countable set of the
lowest cardinality, no continuum. Through this the discretisation of the problem has been
performed (discrete is the opposite of continuous, we could also say point or isolated).
Analogically to the base, or the base for the decomposition of vector u
r
(3.1.4), i.e. to base
components , , u v w
r r r
, functions in U can be called base function. The same way as the end
points of vector u fill in a certain space (Euclidean 3D), it can be said that all possible
functions u that can be written in the form (3) fill in a certain function space the base of
which is represented by functions U . There are no physical concepts related to this reasoning.
We have just mathematical terms that are very useful for further abstraction and mutual
understanding.
Seeking the unknown function is a significantly more complex problem then seeking a
(even large) number of unknown values, because even for the simplest situation of one
function of one variable it represents a problem in which we have to find function values
assigned to the continuum of an independent variable. In popular terms: it is a problem of
seeking the infinite number of values, and we do not deal with a countable infinity, but with
3.1 Principles of approximation of the sought distribution of a quantity in FEM
50
greater cardinality. In a smallest possible interval, the cardinality of unknown function values
assigned to the independent variable from this interval is greater than countable infinity.
Therefore, if we (e.g. on a powerful computer) degrade the solution of a problem to the
solution (for nonlinear problems the repetitive solution) of a system of algebraic equations
with 10
5
10
6
variables, it is really a significant degradation represented by such
discretisation.
After the degradation of a problem through its discretisation it becomes clear that the success
of a numerical method depends primarily on the selection of base functions of
decomposition (3.1.3). Even a trivial decomposition of vector (3.1.4), which gave the name to
the base functions, could cause numerical difficulties if the selected base was not
orthogonal, but oblique with a very small angle between a pair of the base, which could lead
to the requirement of high precision of the function of the projected cosines, etc.
Already the selection of the base functions could in the classical methods satisfy the
boundary conditions on (RITZ) or the differential equation in (TREFFTZ)  see table
(3.1.5) for the overview.
The original versions of corresponding methods were published in [10] to [14].
Unfortunately, the clear formulations were distorted by later interpreters. Another obstacle for
normal users was that most interpretations focused on the mathematical side of the problem
and completely neglected the physical and technical needs of practice that initiated the
method. An illustrative example is given in [3, 5]. In our text, let us present this simple but
rather exact explanation of mathematical advantages:
Differential equation Boundary conditions Type of method
(0) satisfied satisfied exact solution
(1) not satisfied satisfied original RITZ
(2) not satisfied not satisfied BubnovGalerkin
(3) satisfied not satisfied TREFFTZ
3.1 Principles of approximation of the sought distribution of a quantity in FEM
51
Figure 3.1:
a) Analysed domain and its boundary .
b) Elements
e
of the domain and elements
e
of its boundary.
c) Division of the analysed domain into layers (FLM).
d) Division of the analysed domain into strips (strip method).
3.1 Principles of approximation of the sought distribution of a quantity in FEM
52
The advantage of the modern variational FEM is in the support of the base function, i.e. in
domain
b
in which the base function has nonzero value, with the exception of points where
their measure
0
0 which are usually points in which the diagrams of base functions (e.g.
trigonometric ones) intersect the basic coordinate planes. In the classical variational methods,
the support is the whole analysed domain
b
(3.1.6)
which results in the wellknown numerical difficulties (especially due to the fact that the
equation systems are illconditioned), impossibility to adapt the base functions to the shape
and modification of the analysed domain. Any opening or cut eliminates the possibility to
apply e.g. Fourier solution of plates by means of trigonometric series, etc.
Base functions in the finite element method are much more expedient. First, let us note that,
for historical reasons, the term finite element method is nowadays used only for the solution
based on the division of a domain with dimension D n (fig. 3.1a). It was established for the
elements with dimension D n (fig. 3.1b) also at the national FEM course [6], even though all
other methods that use the elementprinciple work with finite elements too  FEM  finite
element method = only elements of a domain with dimension D n , BEM = boundary element
method = elements only at the boundary of the domain, BIEM = boundary integral equation
method, FLM = finite layer method, FSM = finite strip method. In FEM we use
decomposition (3.1.3) in such a way that we define base functions U with a small compact
support that is always created only by a few elements that share a certain node (vertex, mid
side). For example, we use pyramidal functions with the value equal to one in the given node.
If the mesh is fine, these functions are equal to zero over the major part of the domain and
create nonzero graphs only on the mentioned small compact support. Mathematics [3, 5]
adopted term finite functions. The integral of the product of two such functions is equal to
zero for most functions (as the integrand itself is zero), except when the intersection of the
supports is not an empty set. Just this ensures that the analysed algebraic equations of FEM
are wellconditioned. The equations have a distinctly band (in general sparse) matrix of the
coefficients in contrast to the full matrix for BEM.
Historically, a similar progress in numerical methods was made for the first time 140 years
ago when Clapeyron was analysing a continuous beam and used as unknowns the support
(hogging) moments M instead of (at those times) usual reactions R . The base functions of
deflection on the basic system that represents a single long simple beam (removal of
intermediate supports (constraints) of the continuous beam) use the whole analysed domain as
the support and lead to the full matrices of the coefficients of the lefthand sides of the linear
equations for unknown reactions, in contrast to wellknown threemoment equations with the
band width (halfwide, for symmetrical equations) equal to 2 regardless of the number of
spans of the beam. The mechanical meaning of the presentday base functions in FEM can be
better expressed if we, in addition to the discretisation, introduce also convenient
parameterisation, i.e. if we choose as unknowns the (technically illustrative) quantities in
nodes (displacements, rotations, moments, stress components, etc.). This illuminates the
relation with source functions of the discretised system (influence lines of the parameters).
The base functions in the boundary element method (BEM) (fig. 1a, b) become clearer if we
realise that it is in fact a finite treatment of the Trefftz method, i.e. we choose such base
functions that precisely satisfy the differential equations inside domain and if we try to
achieve that their linear combination (3.1.3) satisfies also, in a certain exactly defined sense
3.1 Principles of approximation of the sought distribution of a quantity in FEM
53
(i.e. in what is termed norm, e.g. energetic, minimum of the sum of the squares of variations
or just collocations), the boundary conditions. With this aim in mind we divide boundary
(= edge of the domain) into finite elements
e
and all operations are then carried out only on
the boundary, which is a onedimensionlower figure than domain ( ... D n , ... ( 1)D n ,
see fig. 1b). It is indisputable that this significantly reduces the number of unknown
parameters and even though we get equations with full (nonband) matrices of the coefficients
of the lefthand sides, which may result in saving of totaltime and memorysize. Method
called BIEM (boundary integral equation method) appeared already in the prehistory of this
procedure. Hundreds of articles were describing the method in 1965  1980. Those who are
seriously interested in BEM are recommended to study [34], original sources [15] to [22].
Similarly to FEM, a vast number of unoriginal authors explain the BEM inaccurately and
usually in a completely intelligible way for engineers. They only lay the emphasis on the
exact fundamental functions and used symbols that are all Greek to a normal engineer.
However, the principle is in fact fairly simple:
If we know the exact solution for a certain domain and given internal load obtained through
influence functions and if this solution holds for certain boundary conditions (usually for
infinitely large domain), then the sought solution can be obtained in the following way: we
cut out the given domain and situate it into the given boundary conditions. Apparently, we
intervene only on boundary . This can be demonstrated on a rubbermade model, e.g. a
rubber band with the dimensions of 500 500 2 mm , that serves the purpose of the infinite
domain. Let us put the band on a table and draw, somewhere close to its centre, boundary
of the analysed domain, e.g. circle 80 mm d , ring etc. Inside this domain we apply e.g. one
horizontal force acting in a certain point (centre, etc.). The boundary deforms. Next, let us
imagine that distributed loads are distributed along and these loads cause that the boundary
takes the original shape. The same happens with defined geometrical conditions. Or we can
imagine that one section is defined on and both of its banks are subjected to such
compressive stresses (doublevectors, actions and reactions) that the given domain is
subjected to the required stress component, which represents statical boundary conditions.
There exist also mixed situations when we require a certain given displacement (e.g. zero) on
one part of and certain given stress components (e.g. zero) on another part of . We may
even require a certain mix of geometrical and statical conditions be met on a part of , i.e. in
total boundary can be divided into three parts, symbolically (they do not have to be
continuous):
u p up
+ +
It is quite obvious that throughout the 30year history of BEM its principles have
continuously developed and nowadays conferences and publications with new pieces of
knowledge are intended rather for professionals who are following this development. Practical
engineers cannot usually understand these works, but they should be able to use the
appropriate software. Generally, the abovementioned popularisation, which is definitely not
intended for mathematicians, is sufficient for them.
The appearance of each new method is followed by a period of uncertainty in the engineering
practice and doubting questions rise such as: what is it in fact, does it have any negative
impact on the old software package, has it been already verified, what are the
advantages, etc. All this is quite logical as the engineer has to add their signature to the
design and is legally liable. Senior engineers certainly remember the confusion about FEM
3.1 Principles of approximation of the sought distribution of a quantity in FEM
54
from the years 1967 1972. These difficulties still increase with the continuing development
as (i) the structural engineers no longer play the roles of mathematicians analysts and
programmers (as was the case in the past) and (ii) long past is the time when new methods
were presented to engineers in an understandable way with the subject matter of mechanics.
This used to be common in the prehistory of BEM when crystal clear information was
presented [14, 15 to 19]. It is beyond the capabilities of a practical engineer to read or even
browse through these publications and, consequently, he/she has to rely on widespread
rumours that are quite often commercially oriented. Let us add a few conclusions concerning
both the theory and practice based on our own experience and drawn from international
resources:
i) Contrary to FEM, BEM is advantageous in simple problems dealing with homogenous
domains in which we know the source function of the solution (e.g. influence surface
of plate deflection, displacement of a wall halfspace, etc.). The method results in a
lower number of equations with full matrices of coefficients. Another positive feature
is that infinite domains can be handled without any special elements.
j) In practice however, the analysed domain is often nonhomogenous (e.g. fig. 3.1c), but
it is possible to define not too many homogenous domains where, once again, a certain
known unit solution holds, i.e. the source function is known. Then, the difficulties
can be overcome by the introduction of additional, internal segments of the boundary
(
1 2
,
2 3
, etc.) and their boundary elements. On the other hand, this increases the
number of unknowns in the nodes of all boundaries and, because matrices of
coefficients are not band matrices, also the time needed for the solution of the
equation system increases. Using the wellknown formula for the determination of the
time that is required to solve the system of equations:
2
(minutes) T a N BW (3.1.7)
where N is the number of unknowns and BW the band width, we can easily estimate
under which configuration the advantage of BEM over FEM vanishes. It is rather
soon, as BW is squared in (3.1.7). The time estimate (3.1.7) can be used
approximately also for what is termed sparse matrices where the nonzero elements
are concentrated in a band around the main diagonal. BW can be set equal to
approximately a half of the average number of nonzero elements in one equation.
Quite often, these elements form a kind of islands in the sea of zeros and are spread
almost over the whole matrix, especially in large problems with 100,000 or more
deformation parameters. The time estimates can never be performed accurately in
advance, but numerous programs tell them to users as a tentative initial information
after the input has been completed. Moreover, the progress of the solution is shown in
the screen for individual phases of the solution and, therefore, the user can at any
moment see the current status. Also the solution of the system of equations can be
followed in the same way, even though it is usually more time demanding.
k) Any change of physical properties, thickness, reinforcement, etc. prevents application
of the source function of the homogenous configuration and there is no chance that it
could be possible or useful to create source functions for such situations.
Consequently, if an engineering office is supposed to be able to analyse anything, it
must own (as a sine qua non) a certain software package. And this basis can be
extended by a BEM package. As a result, homogenous domains can be analysed more
effectively. In particular, parametric studies of the influence of the shape of a planar
3.2 Elementary principles of physical nature of FEM
55
domain are effective in BEM, as only the boundary is changing.
The reduction in the dimension of the problem can also be achieved through an integral
transformation (Fourier, Hankel, Laplace, etc.), which is used in the finite layer method fig.
1c. Or alternatively, it can be done using the assumption about the distribution of the solution
along one of the coordinates, e.g. x (see fig. 3.1d) and discretisation of a crosssection, e.g. in
the yzplane. This strip method was introduced already in 1965 by Y. K. Cheung.
3.2 Elementary principles of physical nature of
FEM
The contents of this article belongs to the sphere of human thinking that always
remains reserved to the human engineer, as it represents learning, decisionmaking,
intuition, invention and sensation processes and other functions of a human brain that cannot
be turned to an algorithm and that cannot be reduced to the seven basic operations of a Turing
machine. It means that as the performance capacity of computers is growing, also this
component must increase otherwise the human would become an insignificant peripheral
device of a computer. Unfortunately, this is in fact frequently happening due to lack of
education and related ethics and aesthetics. And it is accompanied by the wellknown impacts
on the real standard of life, ecology of nature in whole, social catastrophes, etc. What was
enough for the engineer equipped just by a handheld calculator cannot be sufficient for the
same engineer if they are fitted with a fourthgeneration computer. The speed of numerical
operations has grown (approximately) from 10
1
op./s in antiquity and 10
0
op./s in the era of
Pascals calculator (called the Pascaline or the Arithmtique) in modern times up to 10
10
op./s
in presentday computers. Throughout this historical period, the homo sapiens sapiens
(doublewise or civilised human according to the scientific terminology of the theory of
species) remains all the same, with the maximum speed of the receipt of information at
around 6 bit/s, maximum usable memory about 1 10MB, which is slower than a normal PC
by a factor of 10
9
, continuously forgetting most of the information received within several
hours and doing in average 1 to 2 errors per 100 elementary operations (physiologically
normal state). The human can compete with the machine only as its creator in the above
mentioned processes that cannot be turned into algorithms. This includes e.g. a deep
understanding and further development of principles of physics. Misapplication of these
principles, if it leads to a logically possible machine, cannot be revealed by the machine. And
it cannot be sensitive to fine ecological connections of the production process controlled by
CAD  CAM  CAE  CIM algorithms. It means that an inaccurate but possible analysis FEM
or BEM model cannot be corrected by a machine.
The engineer meets in practice a lot of problems whose correct solution is based on
physical principles. Principles of mechanics are prevailing and the most often used (even
though sometimes only subconsciously) ones are summarised in fig. 3.2.
The principles are of axiomatic nature, i.e. they cannot be proved by decomposition
into simpler theorems or pieces of knowledge.
3.2 Elementary principles of physical nature of FEM
56
87. PRINCIPLE OF
ISOLATION
88. EFFECTS
89. PRINCIPLE OF
SUPERPOSITION
90. PRINCIPLE OF
RELEASE
91. OBJECT BODIES
92. PRINCIPLE OF
INTERACTION
96. MULTIPOLAR
MODELS
93. CLASSICAL
BOLTZMANN:
94. 3 DEGREES OF
FREEDOM IN A POINT
95. MODELS OF
CONTINUUM
97. COSSERAT:
98. 6 DEGREES OF
FREEDOM
99. MODELS OF PARTICULATE MATERIALS
100. DEFINITION OF MODEL BEHAVIOUR
101. GEOMETRY 102. PHYSICS
103. STATICS
104. DYNAMICS
105. REOLOGY
Figure 3.2:
Some principles and model used in the practical application of finite methods in the field of mechanics
The academic curriculum usually covers only the commonly used principle of superposition
of the effects (fig. 3.3), but the principle of isolation, which must precede it, is rarely
explained. In a real world, each object is under influence of many effects and, based on
human experience, e.g. m effects may be important for that analysed behaviour of a structure.
Let us assume a simple example shown at the top of fig. 3.3 where there are two effects: '1'
e.g. gravity and the load following from it and '2' e.g. thermal changes in the vicinity of the
object, magnetic field, etc. The human brain has just a limited number of neurons
(approximately 120 billion) and their connections (each neuron is connected to roughly 8 000
of others) and is able to think about a problem of limited capacity, e.g. only about the effect
'1' or only about the effect '2'. Therefore, we have to isolate individual effects and analyse
them separately (as isolated).
The philosophical background for this is the belief (i.e. unproven axiom) that this conduct has
some purpose, that if we analyse the effect '1' we can abstract away from the effect '2' if we
have no other choice due to the limited capacity of our brain. If we accept this axiomatically
as the principle of isolation, it can be extended by the wellknow axiom of the superposition
of effects that is common in linear mechanics.
In a nonlinear mechanics we always have to carefully determine if these principles can be
applied. It is usually an exception if it is possible, e.g. the Timoshenkos beam under
compression subjected to a constant force N and the superposition of transverse load. On the
other hand, the effects of load P and thermal load in fig. 3.4 must be analysed together and
3.2 Elementary principles of physical nature of FEM
57
cannot be separated.
It goes similarly with the nowadayspopular principle of interaction. Already in the years
B.C. people (Chinese mechanics, Greek Archytas of Tarentum, Archimedes, Heron, etc.)
knew the preceding principle of release, without which the principle of interaction is
meaningless. Let us have an object (fig. 3.5) and let us be interested in a part of it, e.g. D or E.
We must require axiomatically, i.e. we have to believe, that following from the human
experience acquired so far it is possible to release the analysed part and consider it
separately on its own and that, at the same time, the impact of the neighbouring parts and
other surroundings J can be somehow replaced. Mechanics uses this approach to introduce
all the fundamental terms of what is called internal forces in beams, plate structures and, in
general, stress in bodies. The action of the surroundings is replaced by forces acting on the
boundary of the extracted part.
Particularly topical is the application of this principle through the introduction of what is
termed boundary connections. For example, the system allows for the solution of an arbitrary
foundation limited to a small domain
1
with boundary
1
, which brings considerable
savings of computer time. The surrounding of the foundation is modelled by energetically
equivalent connections and the subsoil itself from a certain depth H (within the extent of
which there are 3D elements) is replaced by a surface model. This approach would be
impossible without the initial axiom that the foundation can be released.
Both the classical and modern conception of forces and force fields is closely related to the
abovementioned four principles and today used models of continuum and particular materials
(fig. 3.2). Before Isaac Newton, forces were known from the physiological analogy as tensile
or compressive forces directly acting on objects. Newton opened the era of forces acting
remotely. Even though he remained bothered throughout the whole life by how they can be
transferred, he stuck to his formulation hypotheses non fingo, i.e. to the description of how
it works without faking the hypotheses on why it is so. The remote force (Fernkraft) became
so popular with later physicists that even contact forces were explained by remote forces,
direct contacts were abandoned and were replaced by remote transfer, even though of course
in the atomic or, at most, molecular scale. Today we know four groups of forces
(gravitational, electromagnetic, of weak and strong interaction) with a problematic fifth group
of what is termed antigravitational forces. However, any relation to these four fundamental
physical forces is still missing in the mechanics in the explanation of the stressstate of
different models of continuum and discontinuum. Those who are interested in these problems
are recommended to study [24 to 31] containing both popularising and original formulations.
3.2 Elementary principles of physical nature of FEM
58
Figure 3.3:
Principle of isolation. Principle of superposition of the results.
Figure 3.4:
Example of a strut or suspension subjected to force load and thermal load when the principles of isolation and
superposition cannot be applied. Force F is transferred by internal forces R. This behaviour can be expressed by
a simple of compound rheological model (see art. 4.2.1, fig. 4.23 and 4.24). The picture shows the frequent
Kelvins model, i.e. parallel connection of spring H a damper N.
3.3 Variational principles of mechanical problems of FEM
59
Figure 3.5:
Principle of release. Principle of interaction. The problem of the size of the surroundings J that should be
together with the analysed cutout AB...H released from the objective reality. Subjective conception of the
substitution of the action of removed parts by forces F.
3.3 Variational principles of mechanical
problems of FEM
3.3.1 Position of variational principles in mechanics
The elementary explanation of articles 1 and 2 is sufficient for a general idea about
how the finite element method provides for the substitution of unknown functions for their
approximate shapes a linear combination of base functions that are closely related to the
division of the analysed body (FEM) or (BEM) into finite elements. To sum up: how
the discretisation of the project and the related finite number of unknown parameters is
carried out in FEM. The solution of the systems of algebraic equations is the most effective
procedure the stateoftheart computers can offer. Nonlinear problems must be solved by the
repetitive application of the same procedure too. The given explanation focused on the most
often used variant called deformation variant of FEM, in which the unknown parameters d
have the nature of geometrical quantities: components of displacement and small rotations, or
their derivatives. In article 1.3 we showed that the application of the firstrate FEM apparatus
is possible in all the situations in which the functional (whose extreme (minimum or
maximum) is being sought) is of additive nature. We also explained that formulas (1.3.2) 
(1.3.4) clearly show how such a functional can be created in problems where we lack any
physical definition for it through the principle of weighed residua (1.3.4). This can be
done also in nonmechanical, even nonphysical, purely mathematical problems. It is clear
that for engineers especially mechanical problems are of great importance. For these, the
development of FEM has offered not only the deformation, but also force and mixed
3.3 Variational principles of mechanical problems of FEM
60
variants for problems of varying content and extent. Some FEM systems contain them in
the set of implemented finite elements. Therefore, it is useful to be familiar with the
variational principles that represent the basis of the particular solution in order to be able to
evaluate their practical advantages and disadvantages. There exist many illustrious original
resources dealing with these issues, e.g. the literature cited in [39]. For the first acquaintance
with the subject let us mention certain sine qua non for the engineers.
The principles of mechanics belong to the oldest principles of physics. This includes such
statements that can be used to derive the equations of motion or under special conditions of
statics the equations of mechanical systems, both in arbitrary coordinates. Depending on the
mathematical formulation and mechanical meaning we can distinguish two types of
principles:
Differential principles examine an arbitrary instantaneous state of the system and compare it
with the nearby state.
Integral principles are in general variational principles that examine such states which the
system passes during a limited time interval and compare them with certain nearby states.
Some principles can exist in both formulations and are usually closely related to each other.
The differential principles include e.g. Gausss principle of least constraint and Hertzs
principle of least curvature. The typical integral principles include Hamiltons, Euler
Maupertuiss and Jacobis principles called together the principle of least effect (effect =
energy time) which differ from each other in the assignment of the comparative
trajectories. DAlemberts principle and the principal of virtual work are not in fact extremal
principles, but also represent statements from which the equations of motion or, as the case
may be, equilibrium, can be derived.
When dealing with problems in the field of statics of elastic bodies, we can start either with
the system of differential equations of with energetic reasoning, which leads either to the
classical boundary problem or to the variational problem from the theory of elasticity. There
is no general answer to the question which formulation of the problem should be considered
the primary one, unless we take into account some additional, e.g. methodological, positions.
Each basic principle is actually an unprovable axiom and can be replaced only by another
axiom (or derived from it). The basis for the finite element is however formed by the
fundamental variational (integral) principles.
The oldest principles of mechanics focus only on what is termed mechanical systems of mass
points or rigid bodies that have finite number of degrees of freedom and such connections
whose reactions do no work for any virtual displacement. On the other hand, the virtual work
of stresses appears in elastic bodies.
The most general principle is Gausss principle of least constraint: the real motion of the
system is such that in every time instant the resistance Z is minimal:
2
1 2
1
1
( , , , ) ( ) min
N
n i i i
i
i
Z a a a P ma
m
K (3.3.1)
on condition that we take into account all possible motion states of the system with the same
positions and speeds but different accelerations
i
a ( 1, 2, , i N K ) that meet the given
conditions.
i
P means external forces and m
i
the mass of N points of the system. Under
special circumstances, this principle is equivalent to d'Alemberts principle, but makes it
3.3 Variational principles of mechanical problems of FEM
61
possible to take into account more general (nonlinear, nonhomogeneous) conditions. It is
based on the method of the least sum of squares and for a free motion without forces ( 0
i
P )
it can be harmonised with the Hertzs principle of least curvature.
The most important principle for the field of statics is the principle of virtual works, which
can be traced back to Aristotle (383 322 B.C.) and which was generally formulated by
Bernoulli in 1717 in a letter to Varignon who published it in 1725. In the original form it is
applicable just to the systems of rigid bodies, noncompressible liquids without friction and
nonexpansible ropes and reads: a mechanical system is in equilibrium only if the virtual work
V of given forces
i
P ( 1, 2, , i N K ) is for any virtual displacement
i
r ( 1, 2, , i N K ) equal
to zero
1
0
N
i i
i
V P r
(3.3.2)
A virtual displacement is an arbitrary infinitesimal change in the position of the mechanical
system that is assumed in a certain time instant and that complies with the given geometrical
connections. Forces
i
P (contrary to the reactions that are subsequently determined by the
analysis) must be defined in advance. They are of physical origin and, in general, depend on
physical constants (gravitation, friction, electrical or magnetic forces, etc.). They can act both
outside and inside of the system. The principle of virtual works can be extended to elastic
continuum only axiomatically. All the mechanical (statical) conditions of equilibrium in the
body and on its boundary are then based on it. On the other hand, if we accept these
conditions as an axiom, the principle of virtual work can be derived from them. The
statements are actually equivalent.
In order to understand the wider context, let us add the following facts. We deal only with
what is termed scleronomous mechanical systems with scleronomic constraints in which time
t is not explicitly included (contrary to rheonomic constraints). In addition, we take into
account only such forces P whose potential is independent on time or more exactly that
do not follow the deformation of the bodies and keep their direction, so that their potential
energy is unequivocally defined by the deformation of the body. We term such system
shortly as conservative. The following statement applies to them: A scleronomous mechanical
system with given conservative forces moves in such a way that the sum of kinetic ( K ) and
potential ( ) energy remains constant during the full movement ( k is the energetic
constant).
K k + (3.3.3)
It must be stated that in technical practice we may encounter also rheonomic constraints and,
in particular, given forces that are not conservative and their potential energy is not
independent on time, e.g. external forces follow the deformation of the structure (fluid
pressure according to Pascals law), are influenced by its vibration (various aerodynamic
effects), etc. Similarly, internal forces may follow more complex constitutive law in the case
of ideally elastic bodies. The system is then nonconservative (dissipative) and the mechanical
energy is partially transformed into other forms of energy, e.g. thermal, etc.
3.3 Variational principles of mechanical problems of FEM
62
3.3.2 Scalar, vector and tensor field in FEM inputs and
outputs
Texts dealing with FEM often use the term field as the opposite to the term single
value. Users sometimes misunderstand it and consider it identical to the term function, which
is known to be a relation that relates one set if independent variables to another set of
dependant variables, regardless of whether the relation is analytical (formulas), numerical
(tables) or graphical (graphs). A field is just a description of the state in all points
1 2
( , , , )
N
x x x x K of a body with dimension N D in particular in FEM in points of beams
and structures composed of them (1D), walls, plates, shells and their systems (2D), spatial
figures (3D), and occasionally even 4D figures in what is called spacetime elements in space
( , , , ) x y z t . Such elements were developed for certain dynamic problems and were applied in
practice. In general, however, another discretisation of time factor t has proved useful. In free
vibration problems it is the harmonic analysis, in forced vibration it is either modal analysis
(decomposition into the free vibration shapes) or what is termed direct integration in time t ,
e.g. over reliably small differences t , which allows for a good estimate of the response of
the analysed system to the given dynamic excitation. Consequently, we stick to three
dimensions 1, 2, 3 N (as the configuration 1 N and 2 can always be obtained from the
basic configuration 3 N ), because this is the way they are defined by means of various
static and in particular geometric restrictions concerning the nature of quantities acting in real
3D bodies (other bodies do not exist in physical terms). Therefore, the explanation can be
limited to 3D body and its points
1 2 3
( , , ) ( , , ) x x x x y z x with the illustrative notation of
coordinate axes ( 1, 2, 3)
i
x i using letters , , x y z .
The subscript notation i (or j if two different axes play a certain role) is in FEM
common also for other quantities than coordinates. For example, displacement components
( 1, 2, 3)
i
u i instead of , , u v w, components of volume forces in
i
X instead of , , X Y Z ,
components of surface forces on (boundary of )
i
p instead of , ,
x y z
p p p , components of
stress tensor
ij
instead of
11 x
,
12 21 xy yx
,
13 31 xz zx
,
22 y
,
23 32 yz zy
,
33 z
(in total 6 components of a symmetrical stress tensor of the
classical Boltzmann continuum), components of strain tensor
ij
instead of
11 x
,
12 21
2 2
xy yx
, etc. as for
ij
, is the change of the originally right angle between
the directions specified by the subscripts. The fraction 1 2 has been introduced to ensure that
quantity has the character of tensor, i.e. that it transforms during the rotation of the
coordinate trihedral following the same rules as any other tensor, e.g. . In addition, this
coefficient is important for the unification of the notation of the geometrical link between
displacements u
i
and components of strain
ij
. The following notation is introduced in which
the subscript following the comma denotes the partial derivative with respect to this index:
1
, 1,2
2
for example :
i
i j
j
u u u
u u
x x y
(3.3.4)
Then the wellknown relation from textbooks dealing with the theory of elasticity (six
3.3 Variational principles of mechanical problems of FEM
63
relations, three for , three for ) has the following form:
, ,
1
( ) 1, 2, 3 1, 2, 3
2
ij i j j i
u u i j + (3.3.5)
e.g.
( )
( )
11 1,1 1,1 1,1
12 1,2 2,1
1
2
1 1 1
2 2 2
x
xy
u
u u u
x
u v
u u
y x
_
+ +
,
Components of rotation that were earlier neglected for many reasons were implemented into
modern FEM programs (approximately from 1980 onwards). Formally, they differ from the
components of strain only by the minus sign in the general formula that is similar to (3.3.5):
( )
, ,
1
2
ij j i i j
u u i j (3.3.6)
In common engineering notation in detail:
1
2
1
2
1
2
x yz
y zx
z xy
w v
y z
u w
z x
v u
x y
_
,
_
,
_
,
(3.3.7)
The notation with one subscript denotes the axis of rotation, the notation with two subscripts
the plane of rotation. It represents the real physical rotation. If it is small enough, it is exactly
satisfied for the limit approaching zero, then the rotation can be considered a vector and the
three quantities (3.3.7) its components in the x , y , and z axis. As a preliminary point it
should be observed that they can be accepted as parameters of deformation in nodes of
finite elements, which makes it possible to concentrate all the parameters into their vertices.
This eliminates the intermediate side parameters that worsen the good numerical
conditionality of the equation systems for the solution. They increase the band width BW or
occupy more elements in matrix K .
A scalar field is defined in each point of body by a single value (e.g. by specific weight
or density , temperature
0
T etc.) as a function of coordinate
i
x , i.e. the position of the
point. Whether we know the values (e.g. definition of the density of the body) or not (e.g.
temperature
0
T ) plays no role in the nature of the field. On the other hand, a scalar field (i)
can vary over time t , which is a common situation towards the rheological process, or (ii) can
be constant over time t , e.g. in the case of stationary (stable) thermal flow through the body
each point has permanently constant temperature.
A vector field is in each point of body by one vector, which is a quantity representable by
an arrow the origin of which is in the analysed point, the length expresses the magnitude and
the arrow defines the direction and orientation of the vector. The way of input is once again
unimportant. A useful aid is an idea of a dense system of arrows each of which starts in one
point of a body.
3.3 Variational principles of mechanical problems of FEM
64
The way these arrows are documented is of no primary significance. In practice, it is always
by means of what is termed vector components in the selected coordinate axes, of which
there is a wide range of selections and which can be the global axes common to the whole
body ( , , ; , , ; , , x y z r z r etc.), or local. For example, for shells it is possible to select
one axis in the normal n to the midsurface and the other two in the tangents to the normal
sections that have the extreme, i.e. principle, radii of curvature (max. a min.).
Users are recommended not to overlook in which axes the components of the vector field are
defined. It is usually displayed in a help line on the status bar in the input graphical
environment or sometimes only mentioned in the manual. This relates mainly to the following
vector fields: (i) displacement components
i
u , both in the input of wanted displacements
and in the output of sought displacements, etc; (ii) components of volume forces
i
X , usually
defined in what is termed gravitational global system , , x y z , with the vertical z + axis
following the gravitation of the Earth, but sometimes also differently (centrifugal forces!);
(iii) components of surface forces
i
p on boundary , defined only for this 2D area; (iv)
speed components often in the product with the density (scalar field) , i.e. the quantities of
shape
*
i i i
v u u t (3.3.8)
The same applies to components of acceleration that are often also in the product with the
density for use in dAlemberts principle, i.e. what is termed volume fictitious forces (inertial
forces) in the meaning of the second Newtons law, or more precisely the density of these
forces in a point:
2
*
2
i i
i i i
u v
P a v
t t
(3.3.9)
Components
i
n of unit vector (length=1) of the normal to the boundary , or
i
m of unit
vector m of the normal to another surface, usually internal boundary between elements, form
a vector field that is of purely geometrical nature in a 2D area.
A tensor field is defined in every point of body by one tensor, which is a physical
quantity originally related to the perception of a general stress (tension) and later also general
deformation of a small neighbourhood of a point. Today, it is formally defined in
mathematical terms through the requirement on the transformation of its components during
the rotation of the coordinate trihedral. In practice, in FEM we come across stress tensor
and strain tensor . With the exception of Cosserats continuum and some other tensors in
nonlinear mechanics, we deal with symmetrical tensors, i.e.
ij ji
,
ij ji
, which
reduces the number of independent components to six. As a default, the components of
normal stress , 1, 2, 3
ii
i are marked , ,
x y z
and components of shear stress ( )
ij
i j
are denoted , ,
yz zx xy
. Arranged in this order, they are often written in the form of single
row matrices always as column matrices in order to save the notation as a transposed
row with one subscript defining the position of the element:
[ ]
1 2 3 4 5 6
, , , , , , , , , ,
T
T
x y z yz zx xy
1
]
(3.3.10)
The matrix algebra calls the singlerow matrices shortly row or column vector, which is
3.3 Variational principles of mechanical problems of FEM
65
justified by the generalisation from the original dimension of 3 in Euclidian space to the
dimension of N in the N dimensional space. This rather innocent choice of terms has caused
considerable troubles to the users of FEM programs. In addition to notation (3.3.10), FEM
algorithms and manuals use (inevitably) also the original matrix notation
11 12 13
21 22 23
31 32 33
x xy xz
ij yx y yz
zx zy z
1
1
1
1
1
1
1
1
]
]
(3.3.11)
in all operations where the simplification (3.3.10) would result in nothing useful. The
subscript operations in algorithms cause no difficulties in the current languages even if a
larger number of indices are used. Similarly, also the strain tensor uses both the vector
notation:
[ ]
1 2 3 4 5 6
, , , , , , , , , ,
2 2 2
T
T
yz xy
zx
x y z
1
1
]
(3.3.12)
and the original matrix notation:
11 12 13
21 22 23
31 32 33
2 2
2 2
2 2
xy
xz
x
yx yz
ij y
zy
zx
z
1
1
1
1
1
1
1
1
1
1
]
1
1
]
(3.3.13)
Components of rotation (3.3.6), (3.3.7) form a skewsymmetric (antisymmetric) matrix
12 13
21 23
31 32
0
0
0
ij ji ij
1
1
1
1
]
(3.3.14)
The antisymmetry is due to the fact that the rotation is in the limit case of geometrical
linearity (for very small rotations) an oriented vector. The rotation in a plane with the
orientation ji is positive if the j axis transforms into the i axis. Apparently, it is the same
vector k ij , but of opposite direction. A plane has an oriented normal that together with
the plane axes creates a right handed system. If axes i and j are swapped, k must change its
orientation in order to keep the system right handed.
The physical formula known as a generalised Hookes law can be thus written in the original
form termed tensor notation
3 3
1 1
ij ijkl kl
k l
C
(3.3.15)
where most of FEM manuals (for the reason of brevity) omit the summation symbols, which
is what is called Einsteins summation convention (the summation is carried out with respect
to the repeating subscripts , k l ):
3.3 Variational principles of mechanical problems of FEM
66
ij ijkl kl
C (3.3.16)
or a typical FEM notation, called matrix notation, is used
(6,1) (6,6) (6,1)
C = (3.3.17)
with matrix of physical constants C of type (6,6), which, thanks to the symmetry, contains
only 21 different elements d , double subscripted , , 1, 2, , 6
ij
c i j K .
Let us note that the stress components depend only on the components of what is termed pure
strain . The physical law contains no rotations.
It can be proved that quantities are the exact average of rotations of all elementary line
segments passing the investigated point, having all the possible directions, measured e.g. from
a certain base direction with the angle ranging from 0
o
to 360
o
or from 0 to 2. For better
understanding, can be imagined to be a rotation of close surroundings of a point as a rigid
compact whole, without giving rise to any stress. This nature of the rotation components is the
root of the peculiar behaviour of finite elements with rotational parameters of
deformation, which will be explained later in the appropriate paragraphs about elements. As
a preliminary point it should be observed that in such elements there exist states when the
elements are not deformed at all, even though they embody a certain set of nonzero rotational
parameters. These states are in technical literature called zero displacement modes. A
certain amount of work of deformation accumulates in the elements so that it represents one
of the zero energy modes. Zero work of deformation can be accepted only when the
element moves as a rigid whole termed rigid body displacement, otherwise it is an unwanted
situation that must be eliminated, usually through an artificial stiffening of the element. It is
perceived as a kind of penalty for undue behaviour hence the term penalty stiffness and
penalty functions for the corresponding additional base functions. These terms are
commonly used in the nowadays FEM manuals without detailed explanations which normal
users may not be interested in.
Important notice for all FEM users:
l) A vector is not equivalent to its three components, i.e. a vector field is different from
three scalar fields, because (i) there exist an infinite number of triplets of components
for every vector, depending on the selection of coordinate axes, but namely (ii) due to
physical reasons. There is an irreconcilable physical difference between a quantity
defined by a single value (e.g. temperature
0
T ) and a vector representable by an arrow
of a certain size and direction. This difference cannot be reduced to any scalar
reasoning, it can only be documented in the form of scalars with a comprehensive
explanation of what is going on.
m) A tensor is not equivalent to its nine (nonsymmetrical) or six (symmetrical)
components neither to three vectors as it is often popularised, e.g. stress tensor = three
components of principal stress
1 2 3
, , , etc. It must be somehow numerically
documented, usually by six components (3.3.10) for stress and (3.3.12) for strain. The
fact that we deal with a completely different physical quantity can be shown on what
is termed uniaxial stress states
33 z
(remaining components equal to zero) or
strain
33 z
(remaining components equal to zero) in the direction of the z axis,
e.g. gravitation, which is a state recommended by most standards for subsoil. Values
3.3 Variational principles of mechanical problems of FEM
67
z
can be formally treated as a scalar field. It is however enough to rotate the
coordinate system into another orientation
* * *
, , x y z where no axis is neither parallel
nor perpendicular to z . All six components of tensor are generally nonzero in
these axes! it means that a pure compression of elementary element dxdydz in the z 
direction results in a general change of the element
* * *
dx dy dz into a parallelepiped,
including sloping of right angles, what can be simply found after substitution into the
wellknown transformation formulas. The situation is similar for stress tensor . Both
tensors describe a physical state in a point neighbourhood, which can be documented
by means of certain values (scalars) with the explanation of what is documented, but
such a phenomenon cannot be reduced to any reasoning that uses scalars or vectors.
n) Internal forces in 1D and 2D FEM elements, i.e. six values for 1D elements of
beams with a solid crosssection
, , , , ,
x y z x y z
N Q Q M M M (3.3.18)
and 3 membrane and 5 bending components for 2Delements of surfaces:
, , , , , , ,
x y xy x y xy x y
n n q m m m q q (3.3.19)
have the character of what is called double vectors. For beams these are force vectors
with components , ,
x y z
N Q Q and moments , ,
x y z
M M M acting in the section across
the 1D element on both banks of the section within the meaning of the principle of
release (art. 3.2.). If it is a section
0
x x in an element 0 x L , then one set of
vectors replaces the action of the part
0
x x > on the part
0
x x < and the other set of
vectors the action of the part
0
x x < on the part
0
x x > . They have the same
magnitude but opposite signs, which is expressed in classical definitions of
summation by the wellknown phrase if we proceed from the other side, we change
the sign. The principle is the same for plated structures, but quantities (3.3.19)
represent only local point intensities of forces and moments. If these intensities
remained unchanged along a specified length of the section c, it would result in a
similar situation as in beams. A planar element
x y
c c , in planar coordinates cut by
sections ,
p p
x
x x c + , ,
p p
y
y y c + , would be subjected to forces and moments (kN,
kNm):
, , , ( )
, , , , , ( )
y x x y y xy x yx xy yx
y x x y y xy x yx y x x y xy yx
c n c n c q c q q q
c m c m c m c m c q c q m m
(3.3.20)
Such a state is very rare in practice, it is what is termed elementary stressstate, and
formulas (3.3.20) are applicable only to the situation 0
y
c dy , 0
x
c dx and
only for distributed internal forces without singular external loads. If, in such
situations, we need forces and moments over a certain length of sections ,
x y
c c , in
particular in the design of reinforcement in reinforced concrete structures where a unit
length is usually chosen, the resultant must be obtained through the integration along
the sections:
3.3 Variational principles of mechanical problems of FEM
68
y x
y x
y x
y x
y x
x x y y
c c
xy xy yx yx
c c
x x y y
c c
xy xy yx yx
c c
x x y y
c c
N n dy N n dx
N q dy N q dx
M m dy M m dx
M m dy M m dx
Q q dy Q q dx
(3.3.21)
Let us mention the school definition of the meaning of the subscripts: the first one
denotes the direction of the normal to the face against which the force is acting, the
second one (if required as the symbol of the quantity itself is insufficient) determines
the direction and orientation of the action. Therefore, the statement about the
reciprocity of shear stresses
xy yx
, resulting in the equalities
xy yx
q q and
xy yx
m m , implies the equality of forces
xy yx
N N and moments
xy yx
M M only for
identical lengths of sections
x y
c c in (3.3.20). This is rare in a general configuration
(3.3.21), as we deal with distributions in completely different points of a planar
structure, which means that these distribution should be identical, which practically
happens only for elementary stressstates (pure shear, pure torsion).
o) It follows, among others, from formulas (3.3.21) that local point values of internal
forces are not decisive for the strength of the structure, but their distribution over
lengths ,
x y
c c are. These lengths are always finite, even though not necessarily always
equal to 1 meter (reinforced concrete), but e.g. values corresponding roughly to
thickness h . If a FEM program prints in the output document singular data about
internal forces (3.3.19), either enormously large numbers (e.g. 10
9
against 10
3
) or
symbols of computerinfinity XXXXX, i.e. overflow in the output format , there is no
reason to worry about the safety of the structure. The output just gives a signal about
proper convergence of FEM towards the exact solution of the problem. This infinity
becomes more apparent with refinement of the mesh. The exact solution often gives
some internal forces of infinitely large magnitude, or more precisely increases
beyond all limits, e.g. under concentrated load P in a Kirchhoff plate the deflection
w is finite, but the bending moments are infinitely large, similarly to a planar
problem, where, however, already the term stress from elementary technical
mechanics defined as ratio
P
A
(3.3.22)
necessarily increases beyond all limits for 0 A and a permanent concentrated
force P .
What can be derived from the stated above for FEM users is explained on an example of
modelling of primary (load) and secondary (reactions) external forces in chapter 5. For the
time being, let us only repeat the rule from art. 2.6 in a slightly modified wording:
3.3 Variational principles of mechanical problems of FEM
69
4. If we want that FEM gives correct results, we must model correctly both loads and
reactions, i.e. on 2D shapes whose planar measure is not zero in the way they actually
act on the structure.
5. If we apply the useful SaintVenant principle and replace the real distribution of loads
and reactions by singular abstractions (point or linear resultants) on area with size d,
then the FEM outputs can be directly used only in the distance nd from the edge or
from the action point of load or reaction. Strict authors admit only n =3 to 5 (e.g. A.
Fppl even in classical calculations). In FEM applications with normally required
accuracy of famous standarddefined two percent, the use of n =1 to 2 is sufficient.
6. If we use userfriendly inputs in the form of concentrated forces and line loads
(nothing like this does not exist in the real world), then we have to interpret the
outputs in domain d correctly within the meaning of formulas (3.3.21). Usually, we
have limited integrals with acceptable (designable) magnitude. Graphically, it is
demonstrated by cutting off of infinite peaks as recommended already by A.
Pucher in the evaluation of influence areas for moments in plates or rounding of
extreme values of the curves which is in fact recommended in standards.
7. A question of a user: what would really cause a concentrated P of linear q can
be answered surprisingly simply: Nothing at all! It would pass through the atomic
space between particles and with regard to the dimension of 0 d there is even no
chance that it would hit any of the particles. Just this strict explanation can discourage
persistent users. This explanation is in fact in another dimension proven
experimentally by tests made by C . Bach at the beginning of the 20
th
century in
Dresden. He applied load to very thin glass plates (window panes) by means of
gradually sharper spike. A special device ensured that the holder was stopped before
hitting the pane. From a certain value of sharpness (defined by a contact area with the
diameter lower than the thickness of the plate) the plate did not break but was
punched. An elementary thought [47] can be used to find out when the failure due to
shear on the surface of the pressed out little cylinder happens earlier than the rupture
of the plate. This is now implemented also in regulations of technical standards
dealing with what is called punching shear in flat slabs. These slabs require different
arrangement of reinforcement. It is clear that the Bachs spike is not a real
concentrated load P , but something close to distributed load over a small circle and
thus it represents a dimension far different from the theoretical limit 0 d . It is
however a valuable source of guidance. The presentday FEM would provide, on
condition of appropriate modelling of the load and implementation of relevant strength
characteristics, Bachs results even without any experiment.
3.3.3 General variational principle
Simple, though a bit formal is the derivation of the general variational principle in
mechanics of solid the phase using orthogonalisations (1.3.2) to (1.3.4) art. 1.3, that will be
extended to eight components. The lefthand sides of nullified equations of dynamic
equilibrium, compatibility of body matter, continuity between adjacent elements (that may be
broken in nonconformal elements) static and geometric boundary or support conditions will
3.3 Variational principles of mechanical problems of FEM
70
be selected as residua
1
to
8
. Distributions of displacement, deformation, stress
components, dislocations and step changes between adjacent elements, reactions or
displacements of supports (or their variations) will be chosen for weight functions
1
g to
8
g .
This is an approach used in FEM literature that is intended mainly for mathematicians and
programmers. For engineers users of FEM programs another derivation will be more
illustrative the derivation to which a physical content can be assigned, which enables the
engineers not to get lost in the vast number of offers by various companies that usually
emphasize which conditions are perfectly satisfied, but at the same time keep quiet about
what remains unsatisfied and to what extent. This lesson became topical already around 1970
when, in addition to improved Lagrangean elements (art. 2.2 to 2.6), Castiglianos elements
(force method) and HellingerReissners elements (mixed method) appeared. Later,
around 1980, rigid finite elements emerged that concentrated all the elasticity or plasticity
into their contact (rigid element method) together with various transitional and semi
rigid elements, etc.
The aim of these historical notes is to persuade the users of FEM programs that after
twenty years of further development any attempt to prepare an overview of existing finite
elements would be just a hopeless and useless, neverending, task. Therefore, it is necessary
to make the users familiar with something else: (i) with the overview of principles on which
the elements are based and (ii) with what can be expected from them in engineering practice,
why they are applicable to something and why in different situations they completely fail to
meet the expectations promoted by dishonest advertisements appearing in the fierce
competition of commercial companies. Suitable for this aim is the following formulation of
the general variational principle:
First, let us define the following fields (art. 3.3.2.):
8. Vector field of components of displacement, matrix notation
1 2 3
[ , , ] [ , , ]
T T
u u u u v w u , tensor notation ( 1, 2, 3)
i
u i .
9. Tensor field of components of strain, matrix notation (3, 3) see art. 3.3.2., tensor
notation ( , 1, 2, 3)
ij
i j , assumption: symmetry
ij ji
, i.e. classical Boltzmann
continuum with the axiom about the reciprocity of shear components
ij
i
ij
.
10. Tensor field of components of dislocation stresses, matrix notation (3, 3) see art.
3.3.2., tensor notation ( , 1, 2, 3)
ij
i j , the choice of subscript symbols is apparently
not important, if required, it is possible to use e.g.
kl
, similarly for the previous field
kl
etc., if it is useful for overall understandability.
Warning: Fields u, , are defined in a completely independent way. No relation
is a priori expected among them. This is emphasised also by the adjective
dislocation in field no. 3, as even meeting of the compatibility conditions of body
is not preassumed (SaintVenant relations between u and ).
11. Vector field of the dislocation boundary load acting on part
u
of boundary , i.e.
in a 2D area of a 3D body on which the distribution of displacement components u is
prescribed, e.g. prescribed settlements or displacements of points of support. Matrix
notation
0 0 0 0 0 0 0
1 2 3
[ , , ] [ , , ]
T T
x y z
p p p p p p p , tensor notation
0
( 1, 2, 3)
i
p i .
3.3 Variational principles of mechanical problems of FEM
71
The following fields are derived from the four basic fields 1.4.:
12. Tensor field d of components of deformation derived from vector field u:
d
T
u v w v w w u u v
x y z z y x z y x
1
+ + +
1
]
(3.3.23)
A priori there is no relation between the derived field d and field , neither any
relation between other two derived fields d ,
0
,
0
.
13. Tensor field of stress components derived from the tensor field of deformation
components through physical, in general nonlinear, relations
x y
x y
W( ) W( ) W( )
etc. (3.3.24)
where W is the density (intensity) of the potential energy of the elastic deformation of
the body in point ( , , ) x y z . This scalar function of , , x y z can be in general of form
( , ) W , but for the explicit expression of stress components it is necessary to obtain
form ( ) W . The function must be determined through physical experiments. In
physically linear theory W is defined by known quadratic expressions [5,31].
Consequently, formulas (3.3.24) are linear and represent general Hookes law
C = (3.3.25)
where C is the matrix of elastic constants (art. 3.3.2.)
14. Tensor field of components of strain
0
, derived from tensor field of dislocation
stresses
0
through similar physical relations:
x y
x y
() () ()
.
etc
(3.3.26)
where , is substituted by symbols
0
,
0
in order to induce field
0
. is the
density (intensity) of the complementary energy in point ( , , ) x y z of body . It is
once again a scalar function of , , x y z and is related to function W through the
Legendre transformation:
( )
T T
W + = (3.3.27)
which explains its name, as the product
T
is the full virtual work of finite stress
components over finite strain components in point ( , , ) x y z . The total energies in
body are then
,* int int
Wd d
(3.3.28)
Formula (3.3.26) is linear in a physically linear theory and follows directly from
Hookes law (3.3.25)
1
C (3.3.29)
The following simple quadratic formula holds for functions and W
T
x x y y xy xy
1 1
( )
2 2
W + + + K (3.3.30)
3.3 Variational principles of mechanical problems of FEM
72
Using (3.3.29) or (3.3.25), formula (3.3.30) can be expressed just by means of quadratic
expressions in or in .
Certain boundary values are assigned to the derived fields if the point ( , , ) x y z of body
gets closer to boundary . In general, no relation between these values and given
0
p ,
0
u is
required in advance. Similarly, the derived fields do not have to be necessarily in equilibrium,
compatible, etc.
Let us assign a number functional to body with boundary
p u
+ and to
3 3 6 6 18 + + + functions of , , x y z defined as vector fields u,
0
p and tensor fields ,
0
:
int ext
D
+ + (3.3.31)
by means of formulas:
( )
int
W d
(3.3.32)
p
ext T T
d d
b u p u (3.3.33)
( ) ( )
0 0 0
u
T T
D
d d
+
d p u u (3.3.34)
int
is the total potential energy of the elastic deformation of the body,
ext
is the total
potential energy of primary external forces of the body, i.e. specified loads b, p in body
and on its boundary or, more precisely, on part
p
of this boundary. The primary state of
the body is always used as the comparison level.
D
is the total virtual work of components
of dislocation stress
0
in dislocations d , including the virtual work of dislocation load
0
p on boundary dislocations
0
u u. The dislocation stress can be considered an external
tensor field of force character that does virtual work over the difference between d and
called dislocation. It is clear that if we assume that is an integrable and compact field
derived from field u, then we get d , as following from (3.3.23) d is us just derived from
u. Then d 0 = and 0
D
, i.e.
int ext
+ , art. 2.2., (2.2.9).
D
can be briefly called
dislocation potential energy.
The extension of the definition of virtual work for continuum must be accepted axiomatically
as proved by experience, as the transition from a finite or countable set of mass points to a
continuum is too radical (in terms of both physics and mathematics). In this context we can
say that functional according to (3.3.31) to (3.3.34) is the total virtual work of the system
of forces
0 0
, , , , ) b p p ( acting on body and its boundary over virtual displacement
0
, , , ) u d u u ( .
In general, any system of functions
0 0
, , , , ] b p p [ can be taken as a system of forces and
any system of functions u [, ] can be taken as a virtual displacement. In addition, d and
0
u u are derived from the latter in order to determine the value of functional . After such
a choice, however, no physical law holds for functional . Let us assume that + is a
continuous physical body in the state of statical equilibrium and that all four original fields
3.3 Variational principles of mechanical problems of FEM
73
0 0
, , , u p have been changed to new fields
0 0 0 0
in on + + + + u u p p (3.3.35)
At the same, the given load b in , p on also supporting conditions
0
u on
u
remain
unchanged. This change (variation of fields
0 0
, , , u p ) results in the change of the
functional value to the value of + .
This change is
( )
( ) ( ) ( ) ( )
0 0 0 0 0 0
( )
p
u
int ext
D
T T
T T T T
W d d d
d d
+ +
+
1 1 + + + +
] ]
b u p u
d d p u u p u u
(3.3.36)
If we extend the principle of virtual work to a generalised virtual displacement
0 0
, , , u p , then the general variational principle reads
( ) 0
int ext
D
+ + (3.3.37)
It is in fact an extension of the Lagrangean variational principle by the dislocation potential,
which can be viewed as an amendment that expresses the impact of secondary conditions
compatibility conditions in and on . These conditions or, more precisely the lefthand
sides of nullified equations that express these conditions, are in (3.3.34) multiplied by
Lagrangean coefficients
0
and
0
p . The classical Lagrangean principle assumes a priori that
these conditions are met and thus expression (3.3.34) equals to zero.
3.3.4 Consequences for some variants of FEM procedures
First of all, it can be proved that all the equilibrium conditions in body and on
boundary
p
, geometrical relations between u and in , geometrical boundary conditions
on u and also physical relations follow from the principle (3.3.37). It is sufficient to select in
turns such variations in which the only nonzero variation is
0
, then
0
p , u and [48].
These four simple variations do not consider all possible variations of four independent fields
0 0
, , , u p and represent only the prove that the variational principle (3.3.37) is superior to
all basic relations in the theory of physically generally nonlinear elasticity. In linear elasticity
the general formula (3.3.24) is replaced by Hookes law (3.3.25) and formula (3.3.30) holds
for W . In geometrical terms, the whole reasoning is linearised by formulas (3.3.23) that hold
only for small u and d, i.e. . For large displacements u but small deformations, it is
enough to apply formulas (3.3.23) extended by quadratic terms, which practically exploits all
technical applications except materials with a fractional modulus of elasticity, such as rubber,
etc. Such problem is considerably complicated and those who are interested in more
information are referred to [32].
3.3 Variational principles of mechanical problems of FEM
74
The general variational principle in the form (3.3.37) was derived by Hu Hai  Chang
already in 1955 as what is termed the principle of generalised potential energy, as it is a direct
extension of the LagrangeDirichlet principle of dislocation potential
D
. Equation (3.3.36)
is a generalisation of Lagrangean variational equation [5, 31].
We integrate the first term in (3.3.34), i.e. the product
0
T
d per partes according to
the GaussOstrogradsky formula over area with boundary , we divide the boundary
integral over into two parts (
p
and
u
) and merge the boundary integral in (3.3.33) and
(3.3.34):
( ) ( )
* *
0 0 0
+
p
u
T T
ext T T
T T T
D
d d
d d
b b u p p u
p u
(3.3.38)
where in brief notation
* * * * * * * *
*
* *
*
* * *
*
*
* *
*
* * * *
* * * *
* * * *
, , , ,
1
1
1
T T
x y z x y z
yx
x zx
x
xy y zy
y
yz
xz z
z
x x yx zx
y xy y zy
z xz yz z
b b b p p p
b
x y z
b
x y z
b
x y z
p m n
p m n
p m n
1 1
] ]
+ +
+ +
+ +
b p
(3.3.39)
, , l m n are direction cosines of the outer normal to .
The meaning of the functionals is thus modified. Hu Hai  Chang called this principle
(3.3.37) the principle of generalised complementary energy. If we follow the Legendre
transformation and make substitution in functional (3.3.32)
T
W , then if
0
the
element
T
is deducted by the element 
T
. If we add the volume integrals in (3.3.32) and
(3.3.38) in the substitution into (3.3.32), the density of potential energy W in (3.3.31) is
replaced by the formula for the density of complementary energy .
The practical consequence of both mentioned forms of the variational principles in FEM is:
p) What causes certain problems in the application of the Lagrangean variational
principle is the selection of the approximation of the deformation state of the body, as
the principle assumes that all conditions of continuity (compatibility) are met and
expects that only one field is under variation. Practically, it is directly vector field u,
because if we select the vector field u as a system of three functions ( , , ) u x y z ,
3.3 Variational principles of mechanical problems of FEM
75
( , , ) v x y z , ( , , ) w x y z that are continuous throughout the whole body , the
unobjectionability of the application of the principle is granted.
In classical variational methods it is necessary to choose the shape of functions u for
the whole body with parameters that are used throughout the whole body . In
simplified terms, only one formulation (formula) for substitute functions u holds for
the whole body . This fact has often been criticised as it is sometimes almost
impossible to find a suitable formula for function u that holds throughout the whole
, and the task is even more complicated by the necessity of satisfying also all the
geometrical boundary conditions. If, however, such a formulation for u is found, then
the fact that all the geometrical requirements of the Lagrangean variational principle
are satisfied is a big advantage.
In the finite element method, this advantage does not apply to the mere selection of the
formula for function u over elements
k
. This factor was initially ignored due to the
fact that in the years 19561964 the method was pursued mainly by engineers
technicians who intuitively assumed that the introduction of nodal parameters of
deformation common to all elements with a contact in the node can for a refined mesh
give a true picture of the nonnodal continuity. The convergence, however, does not
have to be monotonous. The practical consequence is that even quite a fine mesh does
not guarantee generally accurate results. We may express an intuitive theorem that the
results obtained with incompatible elements are the poorer the greater is the proportion
of dislocation potential
D
. Therefore, it is recommended to use compatible elements.
q) A similar reasoning holds for the application of Castiglianos variational principle in
which we have to start only with the equilibrium states and in which we once again
have just one field subjected to variation.
Usually, it is tensor field for which we ensure in advance that it meets the
requirements for any parameters of Cauchys equilibrium equations and statical
boundary conditions. In classical variational methods we select the field using one
formula that holds throughout the whole body . On the other hand, in the finite
element method an arbitrary selection of over elements (without further analysis)
does not guarantee that the statical conditions of equilibrium on boundaries between
elements are satisfied if we limit ourselves only to the continuity in mesh nodes. Non
correct element may be created. An intuitive concept that refinement of the mesh
increases the number of nodes in which the continuity of the stress is guaranteed in the
parameters is not justified as, at the same time, the area of these discontinuities
increases and the size of the border between elements grows. Consequently, the force
methods required the same studious analysis as the deformation variant.
r) The force variant is less frequently used due to clear practical advantages of the
deformation variant. It is, however, impossible to overlook the considerable
importance of the possibility to obtain solution of both types (a) and (b) for the same
problem. It is a wellknown fact that the compatible solution approaches:
a) the exact solution from below in terms of energy, i.e. it gives the lower limit of the
energetic functional. On the other hand, the equilibrium solution approaches
b) the exact solution from the top in the same sense. The intuitive prove follows
3.3 Variational principles of mechanical problems of FEM
76
already from the fact that in the deformation method we define just a finite number of
parameters and thus introduce into body many geometrical constraints that do not
exits in reality, while in the force method we with the same number of the
parameters release the constraints. Naturally, also exact proofs could be presented.
Direct utilisation of this fact is possible if the body is subjected to just one load. Then,
according to the Betti theorem, it follows that the solution a (b) gives the lower
(upper) limit of the ordinates of influence surfaces, or functions. Using this we can
for common load configurations (such as e.g. full distributed load or periodic groups
of forces) make estimates about the lower or upper limit of the result, which is the
fundamental issue of each approximation.
s) The mixed variant of FEM was introduced into practice already around 1970 through
what is termed hybrid elements. These elements define two independent fields of
displacement (symbolically marked u) and stresses ( o ). One of the oldest elements is
Herrmans plate element with independent deflections w and moments P. Both fields
are linear over the triangular element and introduced are three parameters w in
vertices and three parameters
n
M in the centres of sides with normals n . The
popularity of hybrid elements in practice gradually faded as they a priori satisfy
neither equilibrium conditions as in paragraph (a), nor the continuity conditions as in
(b). The engineer can quite simply assess to what extent the solution satisfies the
equilibrium conditions. Deviations can be clearly seen e.g. in stepchanges in the
distribution of internal forces at the boundaries between elements, in differences
between given external nodal forces and those forces that correspond to the calculated
parameters are printed by the programs like what is called nodal loads we mean
the real (IST in German) as against wanted (SOLL in German). The solution may
produce certain unwanted forces (residua) on free unloaded edges. If these deviations
are negligible in comparison with the quantities that are decisive for the design (e.g.
2% of the absolute extreme of similar values in the whole structure), technical
standards accept even such calculation. On the other hand, it is hopeless in common
practice to try to evaluate the admissibility of deviations in the continuity between
elements or between the structure and its supports. Even a subtle dislocation may be a
cause of fatal errors in the stressstate, which can be proved when we input such a
dislocation as a deformation load especially very stiff structures are sensitive.
Therefore, purely forcevariant Castiglianos (b) elements and classical hybrid
elements can be nowadays only rarely seen.
In following years, finite elements based on more general HellingerReissners principle with
pure deformation parameters emerged. They guaranteed, when refined, a good convergence
towards what is called weak solution of the given problem. They are used in the presentday
practice, especially after deformation rotational parameters of type were introduced. See
also the end of art. 3.3.5.
3.3.5 Special configurations used in FEM
Hu Hai  Chang and K. Washizu formulated the principle in a general form similar to
(3.3.37) in 1955 for both elastic and plastic zones. Our text deals only with elastic
applications. It is possible to proceed from the classical variational principle to the general
3.3 Variational principles of mechanical problems of FEM
77
principle in such a way that the condition of continuity (LagrangeDirichlet) or equilibrium
(Castigliano  Menabrea) is not set a priori, but as a secondary condition that must be met by
the functions minimising functional to the greatest possible extent. For example, when the
Lagrangean variational principle is extended, the dislocation stresses and loads
0
,
0
p have
the meaning of the Lagrange coefficients by which the lefthand sides of the nullified
secondary conditions are multiplied in the element that includes the total value of
functional .
The other way round, both the classical principles (Lagrange and Castigliano) can be
derived from the general principle, as well as other more uptodate, or what we may call non
classical, principles (Hellinger  Reissner, Veubeke, etc.). The derivation of these more
specific principles only requires that we assume in the general principle that fields
0 0
, , , u p
are subject to variations that meet some of the conditions which otherwise follow from the
general principle, such as Eulers differential equations of the variational problem (3.3.37). As
already said, there are in total five conditions:
A) in body
15. The physical relation between tensor fields and , which is generally in the form
(3.3.24) or (3.3.26); for the material with a linear relation between and in the
form (3.3.25) or (3.3.29); i.e. Hookes law.
16. The geometrical relation between fields u and . Tensor field can be derived from
vector field u by means of wellknown formulas for d . This relation can be also
expressed through the SaintVenant conditions of compatibility, on condition that we
first calculate only with field . If they are met, then there also exists a vector field u
from which the tensor field can be derived.
17. Cauchys equations of equilibrium, i.e. the wellknown statical relation between tensor
field and vector field of loads b.
B) on boundary
p u
+
:
18. Geometrical boundary conditions for field u on
u
.
19. Statical boundary conditions for field ; see detailed notation (3.3.39), as the
principle leads to the identity of fields
0
.
If we sort these conditions in the given order, then the general variational principle can be
classified by symbol EEEEE, as all the conditions are Eulers equations of the appropriate
variational problem. For other less general principles we assume that one or more from the
five presented conditions is satisfied a priori, which is marked by symbol a . These special
situations sorted in the order following from [50] are summarised in table 3.1.
It can be mathematically proved that what is termed correct finite elements that make it
possible to satisfy the conditions (a priori) have the following physically favourable property:
when the mesh is gradually refined, the solution converges to such a state of the structure that
embodies the exact value of the appropriate energetic functional briefly said: it converges
in terms of energy. This is related to the basic fact that the classical solution using the
differential equations of equilibrium (Cauchy, Lam) and continuity (SaintVenant, Beltrami)
very demanding on the continuity of functions (up to the third derivative of the
3.3 Variational principles of mechanical problems of FEM
78
displacement components, inclusive!) called strong solution, is in FEM replaced by a
problem of seeking an extreme of some energy significantly less demanding on the continuity
of functions over the analysed domain. It is a different solution (in the sense solution = result)
and is called weak solution see the end of art. 3.3.4. The functional space, in which the
solution is sought, is thus dramatically limited. It is a certain small subspace from the space
belonging to the strong solution. Let us symbolically mark a technically interesting function,
e.g. deflection or stress, by symbol F (strong solution) and f (weak solution). The principal
question for engineering applications is whether f differs from F and what the
consequences on the design may be. Those who are interested in these convergence issues are
kindly referred to the literature cited e.g. in [5]. This complicated issue can be briefly
summarised as follows:
t) The energetic convergence of FEM does not generally imply the convergence in all
functional norms, i.e. in all defined remainders between the distribution of functions
F , f , e.g. the integrals from the squares of remainders ( ) F f . The same applies
also to the derivative of functions F and f and naturally also to individual values
i
F ,
i
f in a certain point i . Only in exceptional situations it is possible to deduce from the
energetic convergence the convergence in one point, on condition that such a value
itself can be decisive for the energy. This typically involves what is termed influence
lines with one concentrated load, e.g. a plate subjected to concentrated load
i
P in point
i has in linear mechanics according to the Clapeyron theorem the total potential
energy of internal and external forces related to the primary state 2
i i
Pw .
u) In FEM with correct Lagrangean elements energy
int
converges to the exact
value, i.e. deflection
i
w under load P converges to the exact deflection too. But,
already for two loads ,
i k
P P we have ( ) 2
i i k k
Pw P w + . The refinement of the
mesh in a good program results in correct , but there exist infinitely large number of
deformation pairs ,
i k
w w that can give it. Of course, the engineers intuition tells that
no wild pair is possible, but the computer has no intuition. It only works with
formal, fully transmittable terms see art. 5.1.1.1. Therefore, pure mathematical
analyses of base functions of elements are welcome, as well as proofs that the solution
converges even in functions f (even though it is still a weak solution) where it is not
possible to make any statements about higher derivatives as in the case of strong
solution F .
v)
108. Conditions
109. in body 110. on boundary
106. 107. Variational principles
111. P
hys
112. G
eom
113. S
tat
114. G
eom
115. S
tat
116.
.
117. General (Hu HaiChang, K.
Washizu)
118. E 119. E 120. E 121. E 122. E
125. vers
ion a)
126. a 127. E 128. E 129. E 130. E
123.
I.
124. Hellinger
Reissner:
131. vers
ion b)
132. E 133. a 134. E 135. E 136. E
3.3 Variational principles of mechanical problems of FEM
79
137.
II.
138. LagrangeDirichlet: 139. a 140. a 141. E 142. a 143. E
146. vers
ion a)
147. a 148. E 149. a 150. E 151. a
144.
V.
145. Castigliano
Menabrea:
152. vers
ion b)
153. E 154. a 155. a 156. E 157. a
158.
.
159. Boundary (Trefftz) 160. a 161. a 162. a 163. E 164. E
167. vers
ion a)
168. a 169. E 170. a 171. E 172. E
173. vers
ion b)
174. E 175. a 176. a 177. E 178. E
179. vers
ion c)
180. E 181. E 182. a 183. E 184. a
185. vers
ion d)
186. E 187. E 188. a 189. E 190. a
165.
I.
166. Special [50]
191. vers
ion e)
192. a 193. a 194. E 195. E 196. E
Table 3.1:
The overview of a priori satisfied conditions (symbol a) in various variational principles. Other conditions are
Euler equations of the problem (symbol E).
w) A common user quite often meets the consequences of the introduction of the weak
solution in elements with rotational parameters of deformation that cannot generally
satisfy the compatibility on common borders (Novozhilovs effect, see later in the
text), i.e. in general in elements resulting in dislocations that are minimised by means
of the general HellingerReissner principle. The obtained solution thus satisfies neither
the conditions of equilibrium nor the conditions of continuity. Both improve with the
refinement of the mesh, but in terms of energy, e.g. the dislocation potential, work
done by internal forces on the dislocations between elements, reduces. In general, this
does not imply the convergence in functions, respectively in norms of remainders of
two functions, not mentioning the values assigned to a certain point. This usually
escapes the attention of undemanding users due to the standarddefined 2% tolerance
for numerical accuracy, because the nature of the problem does not lead to any
substantial differences in comparison with the exact solution F obtained analytically
in a simple test.
x) In practice, no one knows the exact solution of their problem since absolutely no one,
including analysts, knows it for more complex shapes, loads and supports of the
analysed structure. The results of FEM are accepted in good faith (bona fide), i.e. the
reputation of a software company is trusted, which is verified through independent
tests. Or alternatively one relies on the fact that structures designed by means of FEM
programs serve reliably their purpose for years and (even) were not too expensive, i.e.
they are safe and economical! This is an intuitive factor in the trust in the FEM
program that cannot be underestimated by the engineering science.
3.3 Variational principles of mechanical problems of FEM
80
3.3.6 Example of evaluation of elements by means of the
variational principle
After the initial period between 19561965 when elements based on principle III from
table 3.1 (Lagrange, deformation) were exclusively used, elements of type IV from table 3.1
were developed (Castigliano, force) and also elements of type II from table 1 (Reissner [51],
mixed). In classical matrix analysis of structures composed of beam they correspond to
deformation, force and mixed method, which means that the unknown parameters are of
geometric or static nature or both. Close attention was paid to mixed elements and a
considerable numerical effect was expected from them due to the minimum requirements
concerning the grade of the base functions. Even today, users are still interested in offers of
companies that contain what is termed Herrmans plate element: triangular with parameters
1 2 3
, , w w w
(deflections) in nodes and
1 2 3
, , m m m (bending moments in midsides opposite to
the vertices 1,2,3), acting through compression or tension on faces in the direction of the
normal of the side. Linear base functions are sufficient. Many similar mixed elements based
on the Reissner principle have been developed, in particular version IIb, in which the physical
relation between fields and is not fixed in advance (symbol "a" = a priori). Authors usually
claimed that the classical deformation variant III was improved through the introduction of
another free field added to the free (unknown) field u, which cause that the equilibrium
conditions were better met (the geometrical conditions are satisfied a priori). However, the
analysis [52] revealed the following:
The best tensor field within the meaning of the Reissner principle is such that is
derived from field u (and the corresponding field ) through physical formulas (3.3.24), for
linear configuration through (3.3.25). This field can be obtained directly by means of
unimproved Lagrangean variational principle!
Other authors used version IIa, i.e. they did not meet a priori the geometrical
constraints in body , which represents the wellknown SaintVenant equations of
compatibility. The common force variant IV was reportedly improved through the
introduction of another free field u in addition to free field , the aim of which was to better
meet the conditions of continuity in body (statical conditions of equilibrium are satisfied a
priori). With no particular difficulty it was however shown ([52]) that the following statement
is true:
The best tensor field within the meaning on the Reissner principle is identical with
the best within the meaning of the Castigliano principle!
No improvement of FEM with force elements is thus possible using the Reissner
principle. With regard to the quality of field u, corresponding to the force solution, there is
nothing definite to be said about it, as field corresponding to the sought field according to
(3.3.26), in a linear configuration according to (3.3.29), does not have to be integrable into
field u, because the SaintVenant conditions of compatibility were not guaranteed. We have
six differential relations between components of tensor that follow in fact from the
requirement of continuity of components , , u v w of field u up to the third derivative inclusive
(classical smoothness of displacement components).
3.3 Variational principles of mechanical problems of FEM
81
3.3.7 Buflers variational principles
The generalisation of the reasoning from art. 3.3.3.3.3.6. was made already in 1969
by H. Bufler [53]. In particular, he assumed that all fields can vary over time t , which means
that we deal with a problem of four variables , , , x y z t . The principle of virtual displacements
then uses the variations of components of displacement u
i
in the analysed state in time t ,
which is emphasised by the name dynamic principle of virtual displacements. Similarly, the
principle of complementary work (Castigliano) is transferred into the dynamic principle of
virtual forces. This extends the reasoning from art. 3.3.3. to 3.3.6. (table 3.1) to problems of
dynamics. What is however important for further (also the presentday) development of FEM
is the introduction of a significantly more general dislocation potential on the boundaries
between elements and the possibility that the impulse theorem can be met only approximately
(Second Newtons law, Ft mv ) in the surroundings of each point of body . Art. 3.3.3.
established four independent fields
0 0
, , , u p , in tensor notation
0 0
, , ,
i ij ij i
u p and (from
them) derived fields (3.3.23) to (3.3.29). Now we add the following:
Vector field
i i
v u t where is the density of the matter of the body and
i
u t is
the speed of the motion of the point, or, more precisely, its components into axes , , x y z
( 1, 2, 3) i . The field is called briefly specific impulse.
Dislocation border
d
, called also the internal boundary, which is a set of all boundaries
between finite elements that are located within body , but not on its boundary . This
boundary is as in art. 3.3.3. divided into two (not inevitably continuous) parts
u p
+
on which the geometrical ( u ) or static ( p ) conditions are prescribed. These, however, now
depend also on time t and thus we have kinematic ( u ) or dynamic ( p ) conditions.
Discontinuity in the components of displacement is now allowed on the dislocation boundary
(in the way that may really happen in FEM according to the nature of the FEM variant).
Boundary
d
divides in each point body by a section on one bank of which one finite
element is located and on the second bank of which another element exists, on which the
displacement is marked as prime (e.g. u = u prime). In general, certain stepchanges appear
in the displacement components that are called:
Vector field of dislocations, defined only on
d
:
( 1, 2, 3)
ui i i
d u u i
In FEM practice, this field has zero values in nodes of the mesh if the deformation FEM
variant (art. 2.2.) is used for which components u
i
are in common parameters of all elements
adjacent to one node. Dislocations commonly happen for incompatible base functions, e.g. the
wellknown rectangular plate element with twelvemember biharmonic polynomial ( , ) w x y ,
which was so far used in older program systems. This is nowadays known even in FEM
practice. What is less known is the fact that dislocations may happen in the force variant just
for the reason that this variant provides no guarantees in advance concerning the continuity of
the body.
Stepchanges of stress may appear on the dislocation boundary
d
. These changes are called:
3.3 Variational principles of mechanical problems of FEM
82
Tensor field of stepchanges in stress, defined only on
d
:
ij ij ij j ij ij j
d m m
where two adjacent elements
e
and
e
have their external normal
j
m and
j
m defined by
means of the components of its unit vector
1 2 3
, , m m m and
1 1
m m ,
2 2
m m ,
3 3
m m , on
condition that we establish that the local positive direction of the normal to the element
boundary points from the element outwards.
Now, we could generalise art. 3.3.3 without formidable obstacles, however, the
auxiliary formulas and the final notation are formally complex. For the purpose of this
overview of FEM principles let us mention the general variational equation that follows
through independent variations from the formulation common to both options (variation of
displacement or forces):
( ) ( ) ( )
( ) ( )
( ) ( )
,
, ,
1
1 1
( )
2 2
1
0
2
u
d u
u d
ij j i i i i i i ij j i i
ij i i ij i j j i ij i i i
ij j i i i ij ij j
X v u d v u v d n p u d
d u u d u u d u u p d
n p u d du m d
+ + + +
1
+ + + + +
1
]
+ +
(3.3.40)
Equation (3.3.40) has 8 members and with independent variations , , ,
i i ij i
u v p
in
integration domains , , ,
u p d
can be clearly seen that it can be satisfied identically only
if the coefficients for these variations are equal to zero. Thus we get gradually 8 Eulers
equations of the analysed variational problem:
20. Dynamic equations of equilibrium in body (Cauchys equations).
21. Definition of the specific impulse, see vector field
i
v .
22. Dynamic boundary conditions on boundary part
p
, where vector of stress
i
p is
specified.
23. Condition of continuity for tensor of stresses
ij
along the whole internal boundary
d
, i.e. stepchanges in stress must be zero.
24. Kinematic relation between vector field u
i
and tensor field
ij
in body .
25. Kinematic boundary conditions on boundary part
u
, where vector of displacements
i
u
is specified.
26. Dynamic relation for vector of reactive stresses
i
p
on boundary part
u
, where
displacements are prescribed.
27. Condition of continuity for displacement u
i
along the whole internal boundary
d
.
3.3 Variational principles of mechanical problems of FEM
83
The Buflers variational equation (3.3.40) holds generally and does not assume
existence of the potential of external or internal forces. It can be thus used for the solution of
problems with nonconservative external forces, with nonelastic material, etc. What remains
unresolved is the question whether the Lagrangean multipliers of secondary conditions can be
applied to variational equations as the existence of the corresponding functional is not clear.
The equation can, however, be viewed also from the point of view of physics as an
axiomatic extension of the principle of virtual work, which already several times proved
correct during the development of mechanics. For the first time with the extension from rigid
bodies to flexible ones and then with the invention of Galerkin method [10,11] and so on.
Equation (3.3.40) does not look appealing to practical FEM users, but it presents in a
concise way an explanation of the behaviour of a great many of new FEM elements offered
today, with an explanation of what can be expected from these elements. The overview of
todayalreadyclassical elements (Lagrange, Castigliano, Reissner) in statical problems to
which the simpler variational equation (3.3.36), (3.3.37) applies was provided in table 3.1,
art. 3.3.4. The table contains in fact 12 element families I, IIa, IIb, III, ... VIe. Elements III
have best stood the test of time, which will be further justified in the following passages. A
similar overview of element families for the general variational equation would be too
extensive. For the time being, most of them have not been developed at all and many of them
would show that they improve nothing. As the presentday explosion of new FEM systems
sometimes contains completely new features, we will present here at least an abridged
amendment to table 3.1 in the following table 3.2 that focuses only on statical problems,
which means that the specific impulse is omitted. From dozens of options we selected only
two from table 3.1 and (in the last row) what is termed rigid elements that are offered by some
companies as well.
197. Order
198.
.
199. 2
.
200. 3
.
201.
.
202.
.
203. 6. 204. 7.
205.
.
206. Domain
207.
208.
209.
p
210.
211.
212.
u
213.
u
214.
215. Meanin
g
216.
YN.
217.
Q.
218. S
PEC.
219. I
MP.
220. D
YN.
221. B
OUND
222.
TEP
223.
224.
IN.
225.
EL.
226. K
IN.
227. B
OUND
228. D
YN.
229. B
OUND
230.
TEP
231.
232. General
ly
233. 234. E 235. E 236. 237. 238. E 239. E 240.
241. Tab.
1, I.
242. 243.  244. E 245. 246. 247. E 248. a 249.
250. III. 251. 252.  253. E 254. 255. 256. a 257. a 258.
259. Rigid
elements
260. 261.  262. E 263. 264. 265. E 266. E 267.
Table 3.2:
3.3 Variational principles of mechanical problems of FEM
84
Overview of a priori satisfied conditions (symbol a=a priori) for some special situations of application of the
general variational equation. Symbol E (Eulers equation) denotes which condition should be approximately met
by the solution.
The symbol E warns that the conditions stated in the same order as in the numbered
list in the text of art. 3.3.7 by numbers 1 to 8 are satisfied only approximately. In the
classical methods these would be Eulers equations of the variational problem. The chosen
example of rigid elements is interesting as it cannot be assigned to table 3.1 it thus belongs
among typical extensions within the meaning of Buflers variational equation (3.3.40). A
finite element is considered a completely rigid, absolutely nondeformable body. Its actual
elastic or plastic properties are concentrated in its edges, e.g. in the case of a triangular shell
elements into their perimeters. A single variant thus offers an incredibly simple element with
three parameters
1 2 3
, , w w w in vertices (for the analysis of spatial shells), w is the deflection
in the direction of the normal to the element.
All the possible combinations of letters E and a in table 3.2 have not been considered so far,
but most of them will not probably be a big asset. The only factor that deserves the attention
of the users of FEM programs is the fact that various new elements can be reliably assigned to
a certain category and that it is possible to evaluate whether their weak point (symbol E) is
not in fact desirable for specific problems solved by them and whether fulfilment of a certain
condition a priori (letter a) is not for that particular group of problems detrimental. The
common conditions of continuity of fields and u on internal boundaries
d
are strongly
violated in the relatively new category of rigid elements suitable e.g. for the analysis of global
behaviour of rockfil dams. The contact is not in fact continuous but it is realised by means of
substantially smaller contact faces on which the contact stress reaches values that are larger
(normally by a factor 100) that the values in continuum. This is related to the possibility to
define what is termed transitional or contact elements of dimension ( 1)D n for domain D n
that can be used for a successful modelling of various technical effects.
3.3.8 Inverse variational principles
The name is not entirely accurate but it took hold thanks to pioneering publications
[54], [55]. The simplest definition of an inverse problem is: all fields that define the stress
state of a domain are known. We are supposed to immerse body whose shape (boundary)
is unknown into this domain. We seek such a body that is in a certain sense optimal. It is
thus an opposite task to common problems in which we know body and try to find its
stressstate. This task and also a much wider range of problems can be solved if we generalise
Buflers variational equation (3.3.40), or directly the original dynamic principle of virtual
displacements ([5], first edition from 1972, pp. 303, formula (48)) for variable body and
boundary . We thus accept the existence of the variation of and , which was for
fields that are continuous throughout the whole body introduced already by V. Horak
[54], [55] in the secondary conditions of constant volume V of body or surface U of
boundary (or both the conditions can be demanded simultaneously).
More general formulation can be obtained if we:
y) either (i) admit that also noncontinuous fields can appear in Horaks inverse
4.1 Geometric properties of elements
85
variational principle and (ii) introduce dislocation boundary
d
that can be possibly
also subject to variations,
z) or admit that Buflers variational principle may contains also variations of body or
boundary or both ( and ). Moreover, if we even allow for variations of
dislocation boundary
d
, it may result in the definition of a principle that optimises
the division of the domain for noncontinuous fields [56].
This more general principle will be stated without proof using tensor notation.
*
is the
variable body,
*
,
*
d
the variable boundary. Variation
*
applies also to integration domains
* * *
, ,
d
.
The inverse dynamic principle of virtual displacements:
( ) ( )
( ) ( ) ( )
* * *
*
* *
*
* * * *
* * * *
1
1 1
0
2 2
p
u d
ij ij i i i i i i i i
ij ij ij ji i i i ij ij j i
d X v u d v u v d p u d
u u d p u u d m du d
+
1
+ + + +
' ;
1
]
(3.3.41)
The inverse principle of virtual forces can be modified the same way. In addition to
adaptations leading to the variational equation (3.3.40), also all elements in the fixed limits of
integrals , ,
d
must be expressed. This can be achieved using the Horaks formulas for
the variation of functionals caused by the variation of integration domains [54].
The meaning of variational principles was for a long time underestimated in FEM practice.
One of the reasons was the inappropriately selected application example in [54]. The
theoretically challenging text of the authors thesis was followed by a single example:
determination of an optimal height ( h ) to width ( b ) ratio of a rectangular crosssection of a
beam subjected to bending on condition that the beam is made (cut) from a cylinder of a
specified diameter. That problem had been solved in the distant past using a substantially
simpler approach called golden cut. Later works by V. Horak and other authors presented a
much wider range of problems related to the optimisation of the shape of a body. In the future
we can expect further progress in the application of FEM to such problems.
4 Finite elements
4.1 Geometric properties of elements
4.1 Geometric properties of elements
86
4.1.1 Differential elements and finite elements
Classical analysis employs differential elements, e.g. dx dy dz for bodies in right
handed global coordinates , , X Y Z , or dr rd dz in semipolar coordinates , , r z , or in
the simplest configuration ds for 1D beams with the coordinates s as the position of a
crosssection measured from the chosen origin of the beam. The characteristic feature is that
these are limit shapes, i.e. it is assumed that differentials , dx dy etc. approach zero without
any limits, which is often expressed inaccurately in a popular way that the elements are
infinitely small. What is in fact substantial for these elements is the variability in the vicinity
of zero. No equation from the classical analysis could be derived if we imagine an element
0 0 0 or with constant . Even the fundamental term of the classical analysis,
derivative, is defined by a limit transition, e.g.
0
0
2
2
0
( ) ( ) ( )
lim
( , ) ( , ) ( , )
lim
( , ) 1 ( , ) ( , )
lim
df x f x f x
dx
f x y f x y f x y
x
f x y f x y f x y
x x x
+ _
,
(4.1.1)
The abovementioned formulas express the values of derivatives through limits of
differences of values that are valid beyond the point x , i.e. for x + , and values that are valid
in the point x . Therefore, these are forward derivatives. The formulas can be written also for
backward derivatives then we have to use differences ( ) ( ) f x f x , or as middle
derivative using a symmetrical form ( 2) ( 2) f x f x + etc. Engineering faculties devote
quite a considerable part of mathematics curriculum to derivatives and therefore, here we only
refer to the appropriate textbooks, books or guides and popular monographs [57], [58]. For
the needs of the following text, it is useful to remind especially the term continuity of
functions and derivatives. Function f is continuous in point B of domain 1D, 2D or 3D
with coordinates x , ( , ) x y or ( , , ) x y z if:
aa) it is in point B defined by certain value
0
f ,
bb) values f in the vicinity of point B approach this single common value
0
f , if the
points of the vicinity approach point B from any side or direction.
A similar definition applies to an arbitrary derivative f . If we use an illustrative concept of
derivative ( ) f x as a tangent (tan) of the angle between (i) the tangent to the graph of f and
(ii) the x axis, then the conception of a common tangent on the left or righthand side
(alternatively limit middle secant) leads to the popular definition of continuity of the first
derivative: the graph of the function is not allowed to have a peak in point B. This applies
also to two variables if we consider sections const y or const x across the graph ( , ) f x y .
If we admit the conception of n dimensioned graphs embedded into ( 1) n + dimensional
space, it is possible to handle the term peak even in these graphs that cannot be displayed by
means of any technical measures in a real 3D model of the space (Euclid, Newton). Similarly
for the second derivative, it is possible under certain assumptions approach to the conception
4.1 Geometric properties of elements
87
of graph curvature (1 variable) or three components of the curvature of a planar graph
(2 variables) that must have in the case of satisfied continuity the same limit from all
sides, etc.
4.1.1.1 How many differential elements are there?
This provocative question can be perceived in different ways. If it is to mean how
many different shapes of differential elements can be introduced in continuum, then the
answer is: the same number as the number of all possible coordinate systems. In 1D
continuum it is just one possible element ds , which is generally curved, but straight in a
straight continuum ( dx ). In 2D and 3D continuum we can theoretically have an unlimited
number of coordinate systems, even though only some are used in practice (rightangled,
semipolar, curved, etc.). Consequently, we get an unlimited number of shapes of differential
elements. However, the question can be understood in different way: We have a given 1D,
2D, or 3D body, we define in it just one fixed system of coordinates and we divide the body
into differential elements all of the same shape. How many of these elements are there? This
formulation of the question has a closer relation to the following text dealing with FEM and it
is useful to think about it a bit more. Every FEM user, including students, would give the
correct answer that their number is unlimited, as their dimension approaches, in limit, zero,
or as students sometimes put it: they are infinitely small and therefore there are infinitely
many of them. The basis for this concept is that the whole body is composed of these
e
d :
e
e
(4.1.2)
If all the elements are identical and their number is n , then:
e
n (4.1.3)
In limit:
0
lim lim" "*"0"
e
e
n
n
(4.1.4)
The popular idea of n as a sequence of natural number 1, 2, 3, n K could lead to
a misconception that there are countable infinity of these elements, which is briefly marked
0
. The subscript 0 indicates the lowest possible cardinality of the set of the infinite number
of elements, which is termed alephnull. If we admit, within the same conception, that
differential element
e
has a zero measure, then the meaning of expression (4.1.4) is just a
sum of countable (even though infinite) number of nulls (zeros), i.e. 0 ! Apparently,
something is wrong. And painstaking students who had good teachers of mathematics report
the same problem.
A continuum of an arbitrary dimension (1D, 2D, etc.) and size (from the smallest
possible domains to metagalactic dimensions) is defined by unlimited divisibility. Each of
its parts has again properties of a continuum. More illustrative example of 1D continuum: the
interval between arbitrarily close points x , x h + is again a continuum, it thus have the same
4.1 Geometric properties of elements
88
number of points as the continuum between and +. The same must be understood in
the meaning of the possibility to assign the elements of one set to the elements of a second
set. In the case of infinite sets we cannot speak about a concrete number since the number is
unlimited, but about cardinality which is identical for the two sets, if such assignment can be
made and no elements remain unassigned in either set. What is not correct in the result 0
is apparently the wrong conception of a continuum that sticks to the common sense of a
practical engineer who does not accept some strange higher infinity. It is sufficient to get over
this useless backwardness and simply admit that the number of points in an arbitrary
continuum is greater that countable infinity, i.e. that they cannot be assigned to natural
numbers 1,2,3,...
If we had an unlimited amount of time and started to assign natural numbers 1,2,3,... to
the points of any howsoever small (e.g. ( , ) x x h + ) or large (e.g. (0, ) ) continuum, which
would represent a neverending process, so that some infinitely longlived creatures would
have to carry on with it after the extinction of the universe, they would see with some
astonishment that the continuum has not decreased at all, that it still has the same cardinality.
After billions of years of assigning, e.g. with the speed of 1 point / second, still almost no
point of the continuum would be marked, which is the mathematical formulation of the
continual measure of a countable, even though infinite, set of points.
We can imagine a differential element around each point of a continuum, so that in
limit 0 dx , 0 dy , etc., we get a set of differential elements which is not only infinite, but
which has much larger cardinality than countable infinity. Thus the number of differential
elements can be marked e.g.
1
, with the subscript 1 indicating a greater cardinality aleph
one. The extremely interesting mathematical considerations, disputes and proofs about
whether there is something between
0
and
1
, the alternative uptodate definition of the
measure, etc., are not interesting for FEM users. On the other hand, it is very useful for the
needs of modelling of inputs and interpretation of outputs of FEM programs to consider the
principal capabilities of FEM in general, with no regard to advertising slogans printed on
the glossy paper of worldrenowned companies colourful marketing materials. These
capabilities are ultimately limited by a simple factor:
The number of finite elements in FEM is always finite, as it follows from the name
of the method and the nature of the matter: division of body (4.1.2) to elements of the same
dimension (FEM) and similarly for the boundary of the body (BEM) or partially discretised
problems (FLM and strip method). The number of parameters d, i.e. geometric quantities
in the deformation variant, or statical quantities in the force variant, or both in the mixed
variant, bound to the nodes of the mesh is finite as well. Therefore, always a double
degradation of the problem takes place:
cc) Reduction of the cardinality of the continua aleph one to a countable set alephnull,
reduction of
1
to
0
.
dd) Reduction of countable infinity
0
to finite number of N free (independent)
parameters d.
Formally it is manifested by:
ee) Replacement of differential equations by the variational principles in which we handle
countable sets of free fields (art. 3.3.2.).
4.1 Geometric properties of elements
89
ff) Reduction of the system of
0
equations to systems of N equations, which can be
processed by the current computers.
4.1.2 Advantages and disadvantages of finite elements
The advantages prevail for both rational users of FEM, mathematicians and
programmers. The biggest advantage can be explained through the oldest and nowadays most
frequently used deformation variant of FEM in statical problems, art. 2.3. The selection of the
finite element is only under numbers 1 to 5 of the overview. It covers the shape of the
element, its base functions and deformation parameters. The whole procedure under numbers
6 to 19 is standard and applicable to all elements. It just has to be programmed with general
number of base functions and deformation parameters as what is termed main program. It is
extended by a continually updated library of finite elements, i.e. subprograms or routines
according to points 1 to 5. This, to a certain extent, turns the situation against classical
methods, in which it was necessary to deal with each shape of body with boundary
separately and in which the differential element remained the same. The following principle
applies to FEM:
The algorithm of FEM solution is independent on the shape of body and its
boundary , including possible openings, notches, unusual loads and supports etc. The
essential part of the FEM algorithm is just the main program that realises the addition
theorem following from the additivity of energetic or other (orthogonalising) functional, no.
13 and 17 of the overview in art. 2.3., together with a library of subprograms for various
finite elements.
It is difficult to explain the advantages of this concept today, when there are only a
few engineers who had to calculate structures by hand using the classical methods and thus
there is nothing to compare with. Hardly anybody would believe that only 60 years ago
(1945) a unique theory was used for each shape of a shell and that all of them failed if one or
more airconditioning openings had to be designed. And it sounds like a fairy tale, even
though it is a proven fact, that in 1955 an engineer was given 14day vacation to analyse a
frame with 16 unknowns and that after the return to work with the solved system of 16
unknowns for 6 righthand sides he was celebrated as a superhero of statics. These historical
notes were included here for the reason that it is the most popular way to explain the
connection of FEM and stateoftheart computers.
The stateoftheart computer including the presentday top PCs have similar
approach to the double degradation of the problem as FEM (art. 4.1.1.):
a) They operate only with a countable set of numbers.
b) They can take into account only a finite number of numbers from a countable set, e.g.
in double precision of N numbers 10
17
, 2.10
7
, 3.10
17
, ..., 10
17
 2.10
17
, 10
17
 10
17
,
10
17
, in arbitrary precision N < .
Analytical operations (derivative, integration) are carried out by computers usually
numerically as simple algebraic operations. Best suited for it are polynomial functions,
analytically the simplest expressions of the type of rational functions (termed correction
functions), etc. This leads us to another advantage of finite elements:
4.1 Geometric properties of elements
90
Finite elements are an ideal means to exploit the capacity of the presentday computers
that are not capable of working with a continuum, but that can handle exceptionally well
(speed, accuracy) a finite set of numbers.
Now we get the weak points of FEM that, if correctly accepted and known, cannot overweight
the advantages in common engineering practice. They represent in fact the dark side of the
advantages:
c) FEM can express neither any singularities, which are typical for mechanics of a
continuum, nor various limit terms of the exact analysis. FEM does not work with any
differential formulas, equations or terms. It is based on variational principles of
mechanics or on other formal principles of a general nature (orthogonalisation with
weight functions = weighted residual principles, art. 1.3 and 3.3.3) and it obtains what
is called weak solution through their approximate analysis using a finite number of
base functions. This solution features lower (by a factor of ten or more) demands on
continuity of sought functions in the analysed domain (structure). For this reason,
FEM cannot give a true picture of higher derivatives of sought functions along internal
boundaries
d
between elements, as they are not defined there at all. If the users
manage to live with the fact that it is not possible to extract from FEM and if they
learn the correct technical interpretation of the outputs (taking into account the
discussed fact), they will experience no disadvantage. However, if they try to obtain
something that does not exist, without paying the attention to the core of the matter,
then even the friendliest ever hotline support cannot help them. All explanation given
over the phone sound to them more like excuses of foxy authors and distributors of the
software who are stealing the precious time of the engineers just when they need to (i)
perform the design of a plate above a line support for the known bearing strength, (ii)
design the reinforcement of a flat slab above a column modelled by means of a single
simple fixed support, or when they need to (iii) design a similar plate subjected to a
concentrated load for which, with gradually refined mesh, the output value
approaches the absolutely useless Pucher singularity
2
ln 8 w r r with an infinitely
large moment, etc. In order to provide accurate information, we have to add that,
today, there exist dozens of what is termed singular finite elements that can be
inserted into required parts of the model in similar situations (acuteangle corners of
openings or boundaries, concentrated force actions, etc.) and that allow for exact
modelling of such singularities. On one hand, they are a rather inorganic element in
common programs, but, on the other hand, they can be included into the addition
theorem in a normal way. After a period of massive use of these improvements
(around 1980), their application became rare, mainly due to marketrelated reasons.
The users called and asked how to size the structure for these infinities. They learnt
that the infinite values represent local point infinities with no meaning for the
strength, as the decisive factors are what is termed integral strength factors, which,
for the simplest configuration, are integrals from the distribution of quantities along a
certain curve, the values that numerous programs printed in their output. What
followed was a disillusion from the improved accuracy and the return to good old
elements that themselves smoothed such singularities, with best results in the case that
loads and supporting conditions were modelled in such a way that corresponded to
technical practice where, after all, no point singularities cannot be created (and why, if
4.1 Geometric properties of elements
91
they are so inconvenient). We should realise that FEM is not an academic issue but a
useful tool for practice. The development of FEM rarely followed exact scientific
demands, even though whole teams of mathematiciansanalysts, who did not know
what the engineers would model by those elements, were engaged in this process. This
was a good thing, since otherwise we would have today seemingly purposeless
hierarchies of elements, convergence criteria, error estimates and many other useful,
but only formal pieces of knowledge without any technical content.
d) The advantage stated above in (a) (i.e. the finite number of parameters and application
of computers that can operate just with the finite number of digits) is linked to a well
known disadvantage that is briefly called numerical instability. The simplest
explanation is rather dramatic. With regard to the measure of the continuum of
numbers, almost no number is stored in any numerical computer. Mathematics uses
this diction to express the fact that the continual measure of displayed numbers equals
to zero. It is a kind of a sieve where almost all numbers pass through. Double
precision only mitigates this problem to some extent but it does not resolve it. It could
be solved only through the introduction of interval arithmetic: either classical, i.e.
number x is defined as interval
1 2
x x x < < , or spherical, i.e. number x includes
dispersion r , which for 1D problems equals interval ( 2) ( 2) x r x x r < < + . In 2D
and 3D we think about points in circular or, as the case may be, surroundings of point
x , or generally in space D n about the appropriate supersphere hence the term
spherical arithmetic. The problem is pursued by a special commission named GAMM
(Gesellschaft fr angewandte Mathematik und Mechanik), led by Professor K. Nickel,
who regularly reports at annual GAMM conferences. In 1996 this conference took
place for the second time (since the establishment in 1920) in Prague. The first time it
was in 1968. We can extract from the published materials two facts that are interesting
for FEM users: unfavourable fact that the calculation takes considerably longer time
(e.g. 300% of time in common arithmetic) and favourable fact that all fuzzy tests and
all numerical instabilities are not relevant any more.
4.1.3 How not to get lost in the collection of finite elements
Let us begin with the same question as in art. 4.1.1 where we asked: how many finite
elements are there? Once again, there are several conceptions available. If we assume that the
question covers elements of different shapes and sizes, it has no practical meaning since it
represents an infinite set of type
1
. Users, usually, raise the question that takes no account of
particular lengths and angles, i.e. for example all triangles with three vertex nodes represent
one element type, on condition that the sought quantities are distributed through the same
functions. Thus the question How many types of elements are there? is meaningful.
Within this book we limit to mechanical problems and we will formulate two question
variants of different sophistication:
e) How many types of elements are theoretically known today including the analysis
of their mechanical properties and problems related to their application in different
problems?
There is a simple answer even to this question: once again, it is an infinite set that (not
4.1 Geometric properties of elements
92
taking into account the particular lengths and angles) is of type
0
, which means a countable
set. For various particular types of elements we today know countable infinite hierarchies
(sequences, families) of elements the elements of which have gradually increasing numbers of
parameters, i.e. degrees of freedom for their behaviour, which makes it possible for them to
satisfy the ever growing demands on the smoothness of the results along the boundaries
between elements.
The term smoothness means the continuity in function values and in derivatives up to
a certain degree d inclusive, which, for 1D, 2D and 3D problems, means formulas:
( )
( )
( )
0, 1, 2, ,
d f x
d
dx
K (4.1.5)
( )
( , )
, 0, 1, 2, , max ( )
f x y
d
x y
+
+
K (4.1.6)
( )
( , , )
, , 0, 1, 2, , max ( )
f x y z
d
x y z
+ +
+ +
K (4.1.7)
The function itself, in this notation, is considered the zeroth derivative, i.e. for example.
(0) (0)
(0) 0 0
( ) ( , )
( ) ( , )
d f x d f x y
f x f x y
dx x y
(4.1.8)
Technical problems always require that certain minimal demands on smoothness are
met. For example, we cannot allow any crack, hollows or cuts on the boundaries between
elements. It is necessary to require that components of displacement , , u v w are continuous
everywhere. Such continuity in the function value is the lowest demand on smoothness. The
class of functions that meets this requirement inside body (analysed structure) is called
class
0
C . The letter C stands for continuity and subscript 0 denotes zeroth derivative, the
function itself. There are problems in which we have stricter demands on the smoothness. A
typical example is the Kirchhoff theory of 2D plates where we introduce just one component
of displacement in the midplane ( , ) w x y and the other two are derived from the assumption
of keeping the normal perpendicular.
( , ) ( , )
( , , ) ( , , )
w x y w x y
u x y z z v x y z z
x y
(4.1.9)
The continuity of , u v results in the requirement that the first derivative of function
( , ) w x y is continuous. Therefore, we have to work with the class of functions
1
C in 2D area
of the plate. Technically speaking: no breaks in deflection surface ( , ) w x y are allowed, which
would mean stepchanges in the first derivatives that manifest themselves as peaks in the
graphs. The demands can be made stricter, e.g. we could require in addition to continuity
(4.1.9), that bending moments (depend on the second derivative) or shear forces (depend on
the third derivative) are continuous as well. Then we could accept only function classes
2
C or
3
C . Other examples will be presented later. For the time being, let us emphasize the
characteristic of the hierarchy of elements that is most important from the users point
of view: higher elements generate function classes
d
C of higher smoothness d .
4.1 Geometric properties of elements
93
f) How many types of elements are included in the presentday assortment of FEM
programs?
If we focus on the prevailing deformation FEM variant for standard statical 1D, 2D
and 3D problems, then the answer is not too satisfactory for curious users we have
thousands of element types the brief overview of which would fill in several volumes. Larger
program systems include more than 100 types of elements, often in numerous variants
depending on various keys reflecting the demands of the users.
It is difficult for common users to find the identity of elements offered by various
companies if they study just the users guides. It is inevitable to read also theoretical manuals
that are published by reputable companies. However, some details are not included there
either. Normal elements are extended by (i) transition elements that make it possible to
connect two elements featuring different valence (see later in the text) and (ii) contact
elements, featuring measure equal to zero in the analysed domain, that give a true picture of
only special configurations of connections between two elements of connections of elements
to the surroundings. The number of all today offered and commercially successful elements
for the deformation variant of FEM can be roughly estimated to be about 1,000. The overview
given in the following articles can help the engineers not to get lost.
4.1.4 1D elements
4.1.4.1 Hermite 1D polynomials in FEM
All the statements in the following overview will be stated without proofs, which can
be found in [5]. The same reference contains also formulations that are stricter in terms of
mathematics. Here, we will use the formulation that is closer to engineers who use FEM
programs (without causing unacceptable inaccuracies).
Each 1D element (fibre, truss, part of a frame or grid beam, column, pile, etc.) has a
very simple geometry. It is a line segment IJ or an arc IJ with endpoints I ( 0) x and J
( ) x L , for which we can introduce coordinates x with the origin in point I, so that an
arbitrary point B has coordinate
B
x equal to the length of line segment or arc IB (fig. 4.1). 1D
element forms a 1D continuum of points in the interval [0, ] L . In practice, we need to know
various functions in this interval, e.g. components of displacement or rotation of beam
sections, internal forces , , , , ,
y z x y z
N Q Q M M M , components of external force or moment
loads, generally variable geometrical characteristics, etc.
4.1 Geometric properties of elements
94
Figure 4.1:
a) Straight 1D element
b) Curved 1D element
c) Hermite interpolation polynomial, general configuration
In FEM we have to accept the fact that the exact results can only rarely be obtained for these
functions (only for what is termed elementary situations). Practically, the sought function will
be approximated by a polynomial with variable x , which is in fact a linear combination of
( 1) n + monomials in x :
0 1 2 3
1, , , , ...,
n
x x x x x x (4.1.10)
with coefficients
0 1 2 3
, , , , ...,
n
c c c c c (4.1.11)
i.e. common type of formula
4.1 Geometric properties of elements
95
2
0 1 2
( ) ...
n
n
p x c c x c x c x + + + + (4.1.12)
briefly
0
( )
n
k
k
k
p x c x
(4.1.13)
There are many ways to create such a polynomial if it is supposed to satisfy given conditions
(Lagrange, Stirling, Bessel polynomials, etc., see [57, 58]. The most suitable for FEM
proved to be Hermite polynomials, which follows from this favourable characteristic (fig.
4.1c):
In every point 0, 1, 2, K , with coordinates
0 1 2
...
s
x x x x < < < < it is possible to define
0 1 2
, , ,...,
s
conditions for ( ) p x , with the sum of these conditions equal to the sum of
coefficients (4.1.11), i.e.:
0 1 2
... 1
s
n + + + + + (4.1.14)
In one point it is possible to define one (with the corresponding 1 ) or more ( 1) >
conditions for ( ) p x and derivatives ( ) p x with respect to x that are denoted by the order of
the derivative in the brackets:
( )
( )
( )
( ) 1, 2,
a
a
a
d p
p x a
dx
K (4.1.15)
or by a number of primes at lower derivatives, e.g.
(1) (I) (2) (I I)
( ) ( ), ( ) ( ), ... p x p x p x p x (4.1.16)
The conditions must be systematic, i.e. they must relate to value p (always) and all
subsequent derivatives
I I I
, , p p K (for greater number of ). No derivatives can be skipped,
including the 0
th
derivative and the function itself. A set of such conditions is called
Hermitean set. If we mark the given numbers by letter y , the Hermitean conditions are
always in the form:
0 0
1 1
( 1) ( 1) (0) (1)
0 0 0 0 0 0
( 1) ( 1) (0) (1)
1 1 1 1 1 1
( 1) ( 1) (0) (1)
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
s s
s s s s s s
p x y p x y p x y
p x y p x y p x y
p x y p x y p x y
K
K
M
M
K
(4.1.17)
These conditions are satisfied by the Hermitean interpolation polynomial, the general form of
which can be explicitly written in the form of a rather unintelligible sum (see [5]). Its meaning
is in the applicability to arbitrary conditions in the form (4.1.17). The formula is significantly
simplified in the following practical applications:
28. All numbers in (4.1.14) are equal to one, i.e. only the function values and no
derivatives are prescribed in all the points. Apparently, 1 n + values must be
prescribed for a polynomial of  n th degree. This is the wellknown (both from school
and from practice) (oldest) Lagrangean interpolation polynomial. Unfortunately, it
4.1 Geometric properties of elements
96
is not suitable for the purposes of FEM, as it operates with many intermediate points.
It is clear from the principle of FEM (art. 1.  2.) that it is essential to try to
concentrate the determining values to the ends of the interval. This is best satisfied by
the second extreme configuration.
29. We select only two endpoints
0
x (for 1D element
0
0 x ) and
s
x ( )
s
x L as
defining. We prescribe the same number of conditions (4.1.17) in each point, i.e.
0
( 1) 2
s
n + . As the number of conditions must be integer, we have to limit
ourselves to odd degrees n . Consequently, for 1 n we obtain a linear Hermitean
polynomial defined only by functional values at ends
0
,
s
y y , which is a standard linear
interpolation. For 3 n we get a cubical Hermitean polynomial, defined at each end
by the function value and its derivative. In FEM, this situation is most common
for 1D elements. It is technically interpreted in various ways and is decomposed into
what is termed unit parametrical states, see fig.4.2. For 5 n we obtain a quintic
polynomial degree 5. Three conditions for , , p p p , etc. are prescribed at each end.
This special configuration can be characterised by table
Chyba! Nenalezen zdroj odkaz.:
4.1 Geometric properties of elements
97
Figure 4.2:
Decomposition of function w(x) of the deflection of a prismatic beam into base functions V
i
(x). The picture
shows the part between nodes 1 and 3, in which six parameters of deformation
i
fully defines the distribution
of deflection w(x)
4.1 Geometric properties of elements
98
Figure 4.3:
Most often used 1Delements beam type for: a) planar truss girders, b) spatial truss girders, c) planar frames,
d) planar grids, e) spatial frames. Shown are their parameters of deformation in global axes.
4.1 Geometric properties of elements
99
Figure 4.4:
a) Example of a 1Delement with three nodes IJK compatible with a 2Delement of a Mindlin shell with 15
parameters of deformation. B) Example of the definition of the position I
C
J
C
of a physical 1D element in space
by means of an auxiliary axial line segment I
A
J
A
and eccentricities e
I
,e
J
.
4.1 Geometric properties of elements
100
(4.1.18)
The literature dealing with FEM almost always use the term Hermitean polynomial only for
this special situation, which is a kind of an established slang term justified by the extreme
endeavour to repel all the determining quantities to the end of the interval.
4.1.4.2 The most known 1D elements
The most often used is a straight beam element IJ as in fig. 4.1a. FEM took this
element from a classical matrix analysis of beam structures, in which the terms beam and
element coincided. Fig. 4.3 presents five basic situations a), b), c), d), e) having 4, 6, 6, 6, 12
parameters of deformation d with the components , , , , ,
x y z
u v w (no. 2 of the overview in
art. 2.3). Components in element coordinates are used to work with the element for 2D
elements it is the centroidal axis
C
x and the principal centroidal axes of the crosssection
C C
, y z . The stiffness matrix and vector of load parameters are then defined in these
coordinates no. 11 and 14 of the overview in art. 2.3. To operate with the whole structure,
all this is transformed into the global coordinates that are common to the whole structure no.
12,13 and 16,17 in art. 2.3. Less frequent is the element with three nodes as in fig. 4.4a, which
can have in general spatial configuration 2 6 18 parameters in nodes I, J, K (K is the centre
of the element). Usually however, what is termed Kirchhoffs principle is utilised, which is
based on the expectation of a certain permanence of geometrical relations even for the
Mindlin model of a beam (see art. 4.2), where the rotations of mass sections are independent
of deflections. The article 4.2 indicates that it represents two independent components of
rotation
z
(instead of dv dx ) and
y
(instead of dw dx ). If we do not expect large
distortion of the right angles between directions , x y and , x z and if we require that the
Mindlin model with the increasing slenderness converges to the Kirchhoff one, the
distribution of
z
and
y
should be expressed by means of a polynomial one degree lower
than v and w. The simplest elements in fig. 4.4a assume that the distribution of , , u v w
follows polynomials of degree 2 and , ,
x y z
polynomials of degree 1. The centre K then
introduces only parameters , , u v w, and, as a result, the whole element has 15 parameters of
deformation. This element is often used as an element compatible with 2D elements (see e.g.
Special Hermite polynomials with two defining points at interval ends:
Polynomial
degree
Number of
conditions at
one end
Type of conditions
Total number of
conditions and
polynomial
members
1 1
p
2
3 2 , p p 4
5 3 , , p p p 6
n ( 1) 2 n +
( 1) 2
, , , ,
n
p p p p
K
1 n +
4.1 Geometric properties of elements
101
fig. 4.6b, 4.7b and 4.8d in the following art. 4.1.5.1) with nodes in the middle of edges. Such
elements, however, make the numerical processes more complicated and prolong the
computation time. Therefore, modern FEM abandons these midnodes and, in general, all
nonvertex nodes. It can be made on the level of one element through interpolation or
preferably preelimination (termed static condensation). Even better is to eliminate such nodes
from the element through the definition of other vertex parameters such as rotation , see art.
3.3.2, formulas (3.3.6), (3.3.7) and the corresponding text in the following art. 4.1.5.
Unfortunately, this has been discovered only after nearly 20 years of fumbling influenced by
the personality of B. M. Irons, who in 1980 in the famous title The Finite Element
Technique still called such effort as vasting money. He was apparently strongly influenced
by Lagrangian variational principle and who was annoyed by zero displacement modes see
art. 3.3.2.
The introduction of type rotational parameters in 2D and 3D elements allowed for a
unified conception of these parameters for all 1D, 2D and 3D elements. These parameters are
commonly used at the ends (in the vertices of line segments) of 1D elements since the
oldest beam conceptions. Their physical meaning is rotation , ,
x y z
of end crosssections
around central axes , , x y z .
The geometry of a 1Delement can be defined directly by its ends I, J. The orientation of
principal centroidal axes
C C
y z is already a physical property, art. 4.2. For many structures
it may be, however, suitable to define first an auxiliary grid of lines I
A
J
A
, which can be called
axial configuration of the structure. The physical (real technical) elements are then attached
with (in general) eccentricities
I J
, e e (Fig. 4.4b) to the nodes of this configuration with respect
to the actual design of joints. The eccentricities are position vectors I
A
I
C
and J
A
J
C
that can be
defined through their global components in coordinates ( , , )
G
x y z common for the whole
structure:
G G G G G G
xI yI zI xJ yJ zJ
e e e e e e (4.1.19)
Or alternatively by the axial components in the auxiliary coordinates ( , , )
A
x y z , introduced
separately for each line segment J
A
J
A
according to the given fixed rules:
A A A A A A
xI yI zI xJ yJ zJ
e e e e e e (4.1.20)
1D elements defined in this way behave in FEM like polygonal tribeams I
A
I
C
J
C
J
A
the end
parts I
A
I
C
, J
C
J
A
of which are absolutely rigid (they do accumulate no deformation energy).
Therefore, it represents a comfortable modelling of rigid joints. Elastic joints can be modelled
by means of elastic joint beams
I J
, e e , see chapter 5. Special situations when the elastic
connection of the beamend (or several beamends) must be taken into account can be
handled with just a dimensionless contact element with a very simple stiffness matrix. On
condition that the elastic connections do not depend on each other, all four main stiffness sub
matrices are diagonal. The elimination of the deformation parameters at the ends of the beam
(similar to a general statical condensation of internal parameters of a substructure or
superelement) can give a transformed stiffness matrix with respect to the elastic connections
to the joints. They then enter the addition that is used to create the global stiffness matrix of
the whole structure. The set of physical elements defined with eccentricities (4.1.19) or
(4.1.20) from the axial configuration is called central configuration of the structure or also
physical configuration, as it is formed by the axes of physical (real) beams. This terminology
4.1 Geometric properties of elements
102
is useful for nonlinear tasks too.
4.1.4.3 Thinwalled beams of open cross section
Beams, whose thickness of individual partial cross section part
i
t is small comparing
to beam cross section dimensions ( , ) b h of we call thinwalled. Literature specifies rate app.
1/10.
There are differentiated open and closed cross sections, according to fact whether
centre line constitutes closed curve. Thereinafter we will deal only with open cross sections
according to theory of Prof. Vlasov. There belong e.g. I, C, T, U sections.
Theory is based on 2 assumptions:
30. Deformation of cross section outline in its plane does not exist. It follows that cross
sections twist in their plane as rigid aggregates.
31. Elements of centre line surface originally rectangular, remain after deformation
rectangular, as well, i.e. their angular deformation is nought.
If the element is stressed only by torque, in the element could originate:
gg) aaaa) only tangential stresses. This phenomenon originates at free torsion (Saint
Venant), diamonding (deplanation) of cross section is not restrained, it could freely
wage.
hh) bbb) tangential stresses and normal stresses. Most frequent case free torsion is
restrained by boundary condition, loading, change of torque value along length of the
beam, or change of cross section along length of the beam. This is a bounded torsion.
We neglect it at massive cross sections, at thinwalled it play significant role on
overall elasticity of the beam.
Bounded torsion could arise also due to other types of loading, it need not be just a torque. It
could be also shear force that does not pass through centre of shear, or axial force acting apart
from centre of gravity of the cross section.
Vector of deformation:
, , , , , ,
T
x y z
u v w 1
]
u
Differential equations of equivalence:
2
2
3
3
x
sv T
x
w
x
w w
d
M GI
dx
d
B EI
dx
d
M EI
dx
4.1 Geometric properties of elements
103
where
sv
M is a free torsion (St.Venant),
w
M is a bounded torsion (Vlasov, warping),
B is bending torsion bimoment.
Total torque is:
x sv w
M M M + . After substitution from antecedent equations we receive
differential equations of torsion:
x T w
M GI EI
For solid crosssections we assume that the shear centre coincides with the centroid of the
crosssection. This means that:
 the calculation of static quantities is simplified,
 in the evaluation of internal forces it is not necessary to distinguish to which point the forces
are related.
For thinwalled crosssections, the shear centre is usually not identical with the
centroid. The ideal situation when both points coincide can happen only for crosssections
that are symmetrical around both principal axes (Isection, Xsection). For crosssections that
are symmetrical around one axis, both the centroid and shear centre are located on the axis of
symmetry, but at a certain distance from each other.
The calculation requires that the sectional characteristics be specified in the following
way:
sectional area
x
A , moments of inertia ,
y z
I I are related to the centroid,
shear areas ,
y z
A A , torsional moment of inertia
x
I and sectorial moment of inertia I
are
related to the shear centre.
A joint in a real structure may connect two beams with a different position of the shear centre
(e.g. two different thinwalled crosssections or a solid crosssection with a thinwalled cross
section). The assembly of stiffness matrices of individual beams can be performed in a usual
way, but the different positions of the shear centres must be matched together they must be
related to a specific point of the crosssection. This point may be:
the centroid of the crosssection (this point is unambiguous and is common to all beams
connected to the joint),
the shear centre of one specific beam with all the remaining beams connected to that beam in
the joint being transformed into the LCS of that specific beam.
The most suitable is the first option when we transform nodal parameters related to the shear
centre, which is based on:
4.1 Geometric properties of elements
104
addition of the torsional moment due to the eccentricity of shear forces and
addition of bimoment B due to bending moments.
INITIAL STRESS
Let us assume that a beam is initially subjected to forces
0 0 0 0
( , , , )
y z
N M M B .
Using the formula in [23] we have:
0
0 0 0
0
y
z
z y
M
P M B
A I I I
+
The geometric matrix (the effect of the stressstate on the stiffness) can be written as a linear
function of the initial stressstate:
0 1 0 2 0 3 0 4 g y z
N M M B + K K K K K
The final matrix
g
K is symmetrical and it dimension is (14 14) .
Thinwalled finite element
Number of nodes:
2
Shape functions:
1 2
2 3
1 2 3 4
2 3
1 2 3 4
2 3
1 2 3 4
u a a x
v b b x b x b x
w c c x c x c x
d d x d x d x
+
+ + +
+ + +
+ + +
For thick crosssection is used only linear term:
1 2
d d x +
Example no. 1
Critical force of cantilever of unequal angle. Length 2[m] L , 210 [GPa] E , 84[GPa] G .
Cross section rectangular, free of slopes and radiuses, thickness of arms 10 [mm], length
centreline arms 250 [mm] and 150 [mm].
Numeric solution and solution according to Vlasov approaches with increasing length of
beam the Euler critical force.
4.1 Geometric properties of elements
105
analytical Euler analytical Vlasov FEM 1D model FEM 2D model
[kN]
crit
P 2250 650 632 658
Results.
4.1.5 2D elements
4.1.5.1 Triangular elements
The geometrically simplest 2D element is a triangle. It is what is termed a simplex in
2D space which is similar to a line segment in 1D, tetrahedron in 3D, super pentahedron
in 4D, etc. The theory of simplex is wellelaborated (simplex simple, the simplest possible
shape). With regard to FEM the interested readers are referred to texts in [5], where proofs
can be found together with further references. As a brief introduction we will state some
useful terms and formulas relating to common 2D triangular elements.
In particular, it is clear that a continuous distribution of an arbitrary quantity over a triangle
can be described by means of a polynomial of degree n with two variables:
2 2
1 2 3 4 5 6
3 2 1
7 8 1
( , )
...
n n
N N
p x y a a x a y a x a xy a y
a x a x y a xy a y
+ + + + + +
+ + + + +
(4.1.21)
where
( ) ( ) 1 2 2 N n n + + (4.1.22)
If we want to write the complete polynomial of degree n in a more compact form, let us
rewrite coefficients
1
, ,
n
a a K in (4.1.21) in such a way that the coefficient at product
i i
x y ,
where i j k + , is written
( ) k
ij
a . Then, instead of (4.1.21) we can write
( )
0
( , )
n
k i j
ij
k i j k
p x y a x y
+
(4.1.23)
where the symbol
i j k +
means that the sum is over all ordered pairs ( , ) i j of nonnegative integers the sum of which
equals to k . It can be derived from (4.1.23) that the number of terms of a complete
polynomial of degree n with two variables is given by formula (4.1.22).
We try to use in FEM applications interpolation polynomials with the highest possible
number of conditions determining the polynomial concentrated in the vertices of the triangle.
The reason is simple: a condition prescribed for an edge is common at most to two triangles,
but a condition prescribed in a vertex is common to all triangles that share this particular
4.1 Geometric properties of elements
106
vertex. This is the reason why the total number of conditions in the same triangular mesh is
lower if more conditions are concentrated to vertices.
Let us explain it by the following example: A polynomial of third degree can be defined on a
triangle unambiguously in two ways:
32. By functionvalues in vertices, centroid and points that divide sides into thirds (in total
10 values).
33. By functionvalues in vertices, centroid and by first derivatives in vertices (in total 10
values).
If we divide a square by two diagonals to four triangles, then the first variant (1) gives the
total number of prescribed values equal to 25, while the second option (2) results only in 19
values. The more triangles are in the mesh the greater the differences in the total number of
values are.
The problems with triangular elements in FEM are of analytical and topological nature. The
topological problem is that the triangulation of domain (wall, plate, section across a body
of revolution, section across a prismatic volume) generates a vast number of elements and
even their numbering does not help to improve orientation. Modern FEM program get over
this weak point through the implementation of quadrilateral elements (art. 4.3) composed of
four triangles with some inconvenient deformation parameters being eliminated or condensed
(art 4.3.23). However, this does not settle the analytical problem which was after years of
groping eventually solved by Czech mathematicians A. Zenisek and M. Zlamal [3, 5],
which is still cited in the world literature. In essence, it is a seemingly simple problem:
A function ( , ) f x y in 2Ddomain e.g. plate deflection ( , ) w x y , stress component in a
planar problem ( , )
x
x y , bending moment in a plate ( , )
x
M x y , etc. is to be substituted by
a function that is prescribed separately for each element into which the domain is
divided by (fixedly selected) triangulation. This prescription should be a complete
polynomial of degree n (4.1.21). Naturally, its coefficient may be different for every
element. What is identical is just the degree n the substitution should be so smooth to fit
into class
d
C with d being the required level of continuity (see art. 4.1.3, formulas (4.1.5)
(4.1.8) and the related text).
For
0
C i.e. the lowest level of continuity solely in functional values (with allowed breaks
in boundaries between elements) there is in fact nothing to solve, as the requirement is met
already by the ageold Courant plated surface from 1943. A plane a bx cy + + over every
triangle defined by three ordinates f in its vertices is sufficient. Linear polynomials were in
FEM actually the first applied base functions (fig. 4.5) and many FEM programs still have
them in their element libraries for the analysis of planar problems. The question arises
concerning the level of continuity that can be guaranteed by polynomials of degree 2, 3, etc.,
at the boundary between elements. The answer can be general for an arbitrary degree n and it
will show that only the degrees 5, 9, 13, etc., can guarantee
1 2 3
, , C C C , etc., continuity, i.e. up
to the first, second, third, etc., derivative inclusive. Therefore, it is convenient to group the
polynomial into foursomes:
4 0,1, 2,... 1, 2, 3, 4 n m m + (4.1.24)
According to (4.1.22) every polynomial has ( ) ( ) 1 2 2 N n n + + terms and, therefore, the
4.1 Geometric properties of elements
107
same number N of parameters is required for its determination. They cannot be chosen
arbitrarily if we want to satisfy the main principle of FEM: to concentrate the highest possible
number of parameters into the vertex nodes of an element. We mark the points as in fig. 4.6.
In order to obtain the maximum brevity and full generality, we write functions ( , ) p x y in
point P or Q with coordinates ( , ) x y using the symbol (P) p or (Q) p and we indicate the
point by an index.
In mechanical FEM problems the function p is usually one of the components of a
displacement or small rotation vector. The approximations of vector field over elements are
created this way. Function p can be of course also e.g. a stress component or deformation
component, torque function (Prandtl), Airy function, capillary pressure of liquids in two
phase environments, etc.
Partial derivatives of this function will be symbolically marked as follows:
1 2
1 2 1 2
( , )
p
D p
x y
+
(4.1.25)
4.1 Geometric properties of elements
108
Figure 4.5:
The simplest triangular FEM element for planar problem or deformation in walls and prismatic volumes.
Displacement components u(x,y) are distributed linearly over the element.
a) Six deformation parameters d
1
, d
2
, , d
6
has the meaning of displacement
components u
i
, v
i
, , v
k
in element vertices.
b) Six parameters of external forces (load).
c) Unit element e
1
d) Distribution of six base functions V in one element is linear too.
4.1 Geometric properties of elements
109
Figure 4.6:
Depiction of points in the triangle in which the parameters of deformation are defined. Vertices P
1
, P
2
, P
3
,
centroid P
0
, points on edges Q with determining index: lower (subscript) starting with 1, 2, 3 in the middle of
edges 12, 23, 31. If there are more points on one edge, the notation is the same and additional indices 1,2; 3,4;
5,6 are attached to the newly added points. Upper (superscript) in brackets (k) mean that the points divide the
edge into k+1 intervals. Fig. shows only a few first configurations of FEM elements with regard to the following
fig. 4.7. Deformation parameters are most often the displacement components, sometimes their derivatives and
in vertices also the components of small rotation.
4.1 Geometric properties of elements
110
Figure 4.7:
The first nine terms in the hierarchy of triangular FEM elements with the polynomial of degree n=1 to 9 having
(n+1)(n+2)/2 coefficients. This corresponds to the same number of parameters that unequivocally define these
polynomials, i.e. 3, 6, 10, 55 parameters for a), b), c), etc., respectively. The small full circle in the figure
indicates points where the parameter is the value of function p, the little arrow means the derivative dp/dn in the
direction of the arrow n, the double arrow represents the second derivative d
2
p/dn
2
, the circle with number no
(no.=1, 2, 3 or 4 in the figure) means value p and all partial derivatives up to the order no.
4.1 Geometric properties of elements
111
Figure 4.8:
a) Rectangular or oblique coordinates in a rectangle or rhomboid. b) Continuity between rectangular or planar
coordinates L
1
, L
2
, L
3
in a triangle. c), d) e) Decomposition of function p (L
1
, L
2
, L
3
) into base functions of
degree 1, 2 and 3, the parameters are just function values p in 3, 6 and 10 nodes according to fig. f) Assumed
distribution of deformation of one side of the element following what is termed beam deflection in beams with
rotation parameters. g) Element deformed only by three rotation parameters.
4.1 Geometric properties of elements
112
For example the statement that values (P) D p
K
(4.1.26)
where p v denotes the derivative with respect to the normal and superscripts , , , 1 k are
defined as follows:
2 2 1 1,...,
2 1 1 1, 1, , 1
2 1 1 1,...,
2 1 1 1 1; 1, , 1
m m k m
m m k k m
m m k m
m m k k m
+
+
+ + +
K
K
(4.1.27)
This statement can be used to create an infinite number ( 1, 2, ) n K of elements whose
geometrical and continuity properties are known and can be directly used for the solution of
FEM problems with specified technical requirements. A classical example is the element for
bending of the Kirchhoff plate. Convergence to a weak solution (art. 3.3.) is guaranteed
already by continuity
1
C (first derivative), i.e. elimination of breaks between elements, which
is satisfied by a polynomial of fifth degree. If we want also the
2
C continuity (second
derivatives, bending moments) a polynomial of degree 9 is required. For higher demands, e.g.
3
C (third derivative, shear forces), we have to work with a polynomial of degree 13. Further,
we will see that a Mindlin plate has lower demands on the degree of the polynomial at the
cost of increase in the number of degrees of a point of a 2D model from 1 ( w) to 3
( , ,
x y
w ). For better understanding, the first nine terms of a general hierarchy (4.1.26),
(4.1.27) are drawn in fig. 4.7. In more uptodate modifications of FEM we try to achieve
even higher concentration of parameters into the vertices than permitted by the polynomial
analysis. This can be achieved by eliminating or condensating nonvertex parameters (art.
4.3). Consequently, the element in fig 4.7e, which originally had 3 6 3 21 + parameters,
becomes an element with 3 6 18 vertex parameters of type , , , , ,
x y xx yy xy
w w w w w w (in
plates).
Also interesting is the evolution of opinions on the useful polynomial degree n . Historically
first was 1 n (1943 R. Courant, 1956 CloughOhio), 2 n (Veubecke, 1965) in planar
problem up to 5 n (1968) in plates, among others M. Zlamal. A complete hierarchy was
created by A. Zenisek (1971). The application of higher polynomials collided with the
capacity of computers. They became used only after 1985 thanks to what is termed
4.1 Geometric properties of elements
113
p refinement of results as an alternative to h refinement, or the combination termed h p
or p h process. The stateoftheart FEM programs allow for both an automatic increase of
the polynomial degree ( p process) and refinement of the mesh, i.e. decrease of the hnorm of
their size ( h process) according to comfortably input demands of the user. It was
contributed, among others, by excellent research performed by I. Babuska (living in the U.S.
since 1962) concerning the adaptive form of FEM carried out between 19801995. The
opinion on the required degree of the polynomial was also changed by the fact that the Irons
taboo was overcome after 1980, i.e. components of small rotation, and possibly even the
rotation fields independent on the displacement fields, were permitted as deformation
parameters. See art. 3.3.2. and notes relating to this issue in art. 4.1.4.2.
4.1.5.2 Triangular elements with polynomials in L
1
, L
2
, L
3
Let us notice the main difference between the rectangular (or oblique) coordinates , x y on a
rectangle (or rhomboid) 0 x a , 0 y b and on a triangle. For a rectangle (or rhomboid)
we can introduce dimensionless coordinates according to fig. 4.8a:
1
1 1 1
1 1
y
x
A
A x y
a A b A
(4.1.28)
where A ab is the area of the whole rectangle (or rhomboid) and
1x
A and possibly
1y
A
,
are
parts of this area in the interval
1
0 or, as the case may be,
1
0 . Coordinates
(4.1.28) are natural in a certain sense, apparently the best and most illustrative of all possible
ones. Such a level of naturalness and illustrativeness cannot be achieved through any
rectangular or oblique coordinates for a triangle (fig. 4.5). In FEM it was quite early
discovered (already about 1960) that just the ratio of areas in (4.1.28) is typical and what is
termed natural coordinates started to be used, called planar coordinates by numerous authors,
which are applied in many presentday FEM programs. The definition is quite simple
(fig. 4.8b). Each point P has three dimensionless coordinates
1 2 3
, , L L L . These are ratios of
triangle areas (P23), (P31), (P12) to the area of the whole element (123), symbolically:
1 2 3
( 23) ( 31) ( 12)
(123) (123) (123)
P P P
L L L (4.1.29)
Even this definition makes it clear that
1 2 3
1 L L L + + (4.1.30)
as the three partial triangles always exactly cover the whole element. In addition, a very
simple expression, or equation of sides of the element, is clear. Each point on side 12, 23 or
31 of the element results in a zero triangle (P12), (P23) or (P31), therefore:
3 2 1
0 (side12), 0 (side 31), 0 (side 23) L L L (4.1.31)
Moreover, vertex coordinates are obvious as in them always one partial triangle occupies the
whole element. It is that triangle which is formed by a vertex with the opposite side. Thus, for
a general marking of the point
4.1 Geometric properties of elements
114
1 2 3
( , , ) ( , ) P L L L P x y (4.1.32)
the vertices P
1
, P
2
, P
3
are:
1 1 1 1
2 2 2 2
3 3 3 3
(1, 0, 0) ( , )
(0,1, 0) ( , )
(0, 0,1) ( , )
P P x y
P P x y
P P x y
(4.1.33)
There is a simple linear relation between planar and rectangular coordinates. If the planar
coordinates
1 2 3
, , L L L of point P are known, its rectangular coordinates , x y can be calculated
from the coordinates of its vertices in the same (otherwise arbitrary) coordinates
1 1
( , ) x y ,
2 2
( , ) x y ,
3 3
( , ) x y :
1 1 2 2 3 3
1 1 2 2 3 3
1 2 3
1 (verification formula)
x L x L x L x
y L y L y L y
L L L
+ +
+ +
+ +
(4.1.34)
4.1 Geometric properties of elements
115
The form of this formula is useful also for other FEM elements, see further in the text
(quadrilateral elements). Planar coordinates
1 2 3
, , L L L can be viewed as influence lines of
point P in terms of coordinates of the vertices of the element. For example,
1
L shows the
influence of coordinate
1
x and
1
y of vertex P
1
on coordinates x and y of point P. Let us go
back to e.g. the unit element in fig. 4.5c) with vertex coordinates (0,0), (1,0), (0,1). Formulas
(4.1.33) have the form following also directly from the ratio between triangle heights:
2 3
1, 1 x L y L
The multiplication by one (1m) is not omitted in order to keep the dimension of lengths , x y .
The inversion of (4.1.34) can give us formulas to calculate planar coordinates
1 2 3
, , L L L from
known coordinates , x y of point P. The most lucid notation is:
( )
( )
( )
1 1 1 1
2 2 2 2
3 3 3 3
2
2
2
L a b x c y A
L a b x c y A
L a b x c y A
+ +
+ +
+ +
(4.1.35)
where 2A means the double of the area of element 123. It can be found fast from the
determinant:
1 1
2 2
3 3
1
2 1
1
x y
A x y
x y
(4.1.36)
Coefficients in formulas (4.1.35) have a common cyclic formula, where ijm 123, 231 or
312:
i j m m j i j m i m j
a x y x y b y y c x x (4.1.37)
As planar coordinates
1 2 3
, , L L L are just linear transformation of rectangular ones , x y ,
their application does not change the polynomial degree on the triangle. Therefore, a
similar hierarchy holds as in art. 4.1.5.1., with nodal polynomials of degree 1, 5, 9, 13, etc.,
for the generation of function class
0 1 2 3
, , , C C C C , etc., in the analysed 2D structure, i.e. with
the requirement on continuity both in the function (
0
C ) or up to partial derivatives of degree
1, 2, 3, etc., inclusive (
1 2 3
, , C C C , etc.).
This fact was unknown to many authors for long time. Consequently, there was a certain
period in the history of FEM when even reputable experts attempted to solve the known
discrepancy between the number of coefficients of a polynomial of third degree (4.1.21)
(which is according to (4.1.22) 4 5 2 10 N ) and the number of vertices of a triangle
multiplied by three (parameters , ,
x y
w w w ), i.e. 3 3 9 . To add the centroidal value w as the
tenth parameter according to fig. 4.7c) was numerically inconvenient as the elimination and
condensation were not yet normally used and mathematicians warned about it. The reason
was that it lowered the convergence by a factor of ten or hundred. A vast number of triangular
FEM elements based on
1 2 3
, , L L L coordinates were derived at those times, in particular of
degree 3. A very useful integration formula still used in FEM literature found its
4.1 Geometric properties of elements
116
application in the formation of stiffness matrices:
( )
1 2 3
! ! !
2
2 !
a b c
A
a b c
L L L dA A
a b c
+ + +
(4.1.38)
The continuity of derivatives could not be guaranteed by the polynomials of third degree. The
class of functions was still
0
C . Various sophisticated modifications of what is termed
correction functions, which were not polynomials but rational functions, shifted them
towards class
1
C . Some of them were published (B. M. Irons, O.C. Zienkiewicz, [1,2]), others
were kept secret and even today some companies treat them as confidential information as
they reportedly represent the core of the success of specific elements. The correction
functions lose their importance in presentday elements. On the other hand, the elements with
rotational parameters now newly require new types of functions relating to the possibility of
strain without energy (zero energy modes) or even zero displacements at a certain
configuration of rotational parameters (zero displacement modes). The technical literature
uses abbreviation ESF (extra shape functions) for functions that enlarge the class of allowable
element deformation and RDOF (rotational degrees of freedom) through the rotational
parameters. We mentioned the necessity to use a special kind of reinforcement in the elements
(penalty energy, penalty stiffness) already in art. 3.3.2.
As an example let us use the first three members in the hierarchy of triangular elements with
polynomials in planar coordinates. Contrary to the hierarchy in rectangular coordinates (art.
4.1.5.1.), where following from fig. 4.7 also partial derivatives are used, we can do just
with function values
i
p in points 1, 2, 3, i K of the triangle, mainly in its vertices and on its
sides. Generally, the polynomial has then the form of a linear combination
1 2 3 1 2 3
1
( , , ) ( , , )
N
i i
i
p L L L p v L L L
(4.1.39)
where 3, 6, 10 N , etc. for degree 1, 2, 3 n , etc. Base, or influence, functions analogous to
fig. 4.5d in linear configuration are very simple (fig. 4.8c):
1 1 2 2 3 3
V L V L V L (4.1.40)
1 2 3 1 1 2 2 3 3
( , , ) p L L L p L p L p L + +
Quadratic polynomials (fig. 4.8d) require introduction of different base functions for vertices
and for midpoints of sides::
( ) ( ) ( )
1 1 1 2 2 2 3 3 3
4 1 2 5 2 3 6 3 1
2 1 2 1 2 1
4 4 4
V L L V L L V L L
V L L V L L V L L
(4.1.41)
( )
1 2 3 1 1 1 6 3 1
( , , ) 2 1 4 p L L L p L L p L L + + K
It is similar for cubic polynomials (fig. 4.8e), where centroid 10 is added and each side has
two nodes in the thirds of sides:
( ) ( )
1
2
3 1 3 2 1, 2, 3
i i i i
V L L L i
( )
9
4 1 2 1 5 9 4
3 1 etc. for and V L L L V V (4.1.42)
4.1 Geometric properties of elements
117
10 1 2 3
27 V L L L
( )( )
1
1 2 3 1 1 1 1 10 1 2 3 2
( , , ) 3 1 3 2 ... 27 p L L L p L L L p L L L + +
To conclude this article, let us mention at least one example of correction function for the
distribution of deflection w of a Kirchhoff triangular plate element:
2 2
1 2 3
1 2 3
1 2 2 3
( , , )
( )( )
L L L
w L L L
L L L L
+ +
(4.1.43)
Following from (4.1.31) it is clear that w reaches on all three sides of the element values
5 2
0 0 0 and a proof can be provided that also the derivatives w n are along two sides
(12 and 13) equal to zero. w n is nonzero only along side 23, i.e. the mass normals h of
the plate rotate. If we use cyclic substitution in (4.1.43), we can obtain two similar functions
with nonzero w n only on side 12 or only on side 13. Linear combination
123
w of these
three functions with suitable selected parameters of type w n in the middle of the sides
thus has two very useful features:
ii) it does not break continuity of deflection w between elements, i.e. class
0
C ,
jj) it can establish continuity w n , i.e. smooth the breaks in the deflection surface of
the plate between elements (class
1
C ), on condition that we add
123
w to a common
polynomial of degree 3, which in general does not guarantee this continuity as we
already know from art. 4.1.5.1., fig. 4.7. Without
123
w it would have to be a
polynomial of degree 5.
It is possible to create an infinite number of correction functions of type (4.1.43) and really
about 100 of them were made between 1960 and 1968. Universities in the U.S.A. in 1968
explicitly forbade submitting of new PhD theses dealing with this theme, as there was a
danger of real explosion of complicated formulas. The presentday FEM programs usually
have such elements in their libraries, as already stated. The users may find it useful to read
the following note: correction functions are not polynomials and thus their integration and
sometimes even derivation are performed numerically. Singularitysurprises are not
eliminated for higher derivatives used e.g. to obtain moments in plates (second derivative)
and shear forces (third derivative). Formulas are more complex see the rule about the
derivation of rational functions
2
( ) ( ) f g f g fg g , which was applied several times! If
the element is not properly treated in terms of programming, its internal forces do not have
to be reliable. Fortunately enough, we are now witnessing an overall tendency of swapping
from Kirchhoff elements to Mindlin ones, for which function class
0
C is sufficient.
4.1.5.3 Quadrilateral elements with polynomials in x,y
This is the oldest known FEM element used already in 1956. The very oldest element
is a rectangular one shown in fig. 4.9. The original authors based their approach on a naive
idea that unfortunately thrived for long in fields isolated from technical information: it is
just enough to generalise polynomials from beams, i.e. 1D elements, to 2D space. This really
gives an applicable element for planar elastic problems that are a kind of extension of tension
4.1 Geometric properties of elements
118
and compression in both directions amended by planar shear to provide for full membrane
(wall) stressstate. Tension or compression in 1D elements requires just a linear base
displacement function ( ) u x . For a 2D element in fig. 4.9a) two bilinear displacement
functions are sufficient
1 2 3 4
5 6 7 8
( , )
( , )
u x y a a x a y a xy
v x y a a x a y a xy
+ + +
+ + +
(4.1.44)
together with eight deformation parameters in four vertices 1,2,3,4 in the following order
[ ] [ ]
1 2 3 4 5 6 7 8 1 1 2 2 3 3 4 4
, , , , , , , , , , , , , ,
T T
d d d d d d d d u v u v u v u v d (4.1.45)
Base functions are of simple form (similar to fig. 4.5 for a triangle). At the same time, it is
guaranteed that a set of these elements generates in the analysed 2D domain function class
0
C , with continuous function values u,v in all points including the boundary between the
elements. This is sufficient for a weak solution of the problem.
The solution of a Kirchhoff plate requires that the deflection surface ( , ) w x y is
continuous in derivatives w x , w y , i.e. function class
1
C . The generalisation from a
beam used originally (for almost 9 years) a biharmonic polynomial with twelve terms
2 2
1 2 3 4 5 6
2 2 3 3 3 3
7 8 9 10 11 12
( , ) w x y a a x a y a xy a x a y
a xy a x y a x a y a xy a x y
+ + + + + +
+ + + + + +
(4.1.46)
which satisfies biharmonic equation (the subscript denotes partial derivative)
2 0
xxxx xxyy yyyy
w w w + + (4.1.47)
which means that plane load on the element ( , ) 0 p x y . This was originally considered a
positive feature (everything is concentrated into nodes). Only on the occasion of the first
declassification of FEM which was in the meantime used for the analysis of many strategic
structures including the preparation of the APOLLO module for landing on the Moon at the
legendary first FEM conference (1
st
Conference on Matrix Methods, WPAFB, Ohio, 1965), a
warning was published that (4.1.46) does not generate function class
1
C . The continuity of
first derivatives is guaranteed only in nodes of rectangular mesh, not along their common
sides where the plate is broken. The proof is selfevident. After a partial derivation with
respect to x , only degree 3 is left in y , and thus w x goes along sides 0 x and x a
(fig. 4.9a) following a cubical parabola in y . This inevitably requires four defining
parameters, e.g. values w x and derivative with respect to y , i.e. ( )
2
y w x w x y
in every side end, i.e. in nodes of the element. This means that the original number of 12
parameters three in each node ( , , ) w w x w y increases to 16 (derivatives are marked
briefly by a subscript):
[ ]
1 2 3 4 5 16 1 1 1 1 2 4
, , , , ,..., , , , , ,...,
T
T
x y xy xy
d d d d d d w w w w w w 1
]
d (4.1.48)
The polynomial with twelve terms (4.1.46) must be extended by additional four terms, in
order to get a regular relation d Sa (no. 2, art. 2.3.). These are
2 2 2 3 3 2 3 3
13 14 15 16
a x y a x y a x y a x y + + + (4.1.49)
4.1 Geometric properties of elements
119
amending (4.1.46) to what is termed Ahlins bicubic polynomial with 16 terms, which
exploits all possibilities of degree in variable x or y up to the degree 3. It is an incomplete
polynomial of sixth degree, as the last term is of degree 6, but many terms of degree 4, 5 and
6 are missing (e.g.
4 6
, x y y , etc.).
Polynomial (4.1.46) with amendment (4.1.49) can be simply written as an Ahlin bicubic
polynomial, if we use two subscripts for the notation of coefficients a meaning the power of
x (1
st
subscript) and y (2
nd
subscript):
3 3
0 0
( , )
i j
ij
i j
w x y a x y
(4.1.50)
What can be successfully applied to FEM are Ahlins bi  n th polynomials with odd
1, 3, 5, 7 n , etc., with
2
( 1) n + coefficients including the absolute term
00
a :
0 0
( , )
n n
i j
ij
i j
w x y a x y
(4.1.51)
It can be proved [5] that they ensure the continuity between elements up to (n1)th
derivative inclusive, i.e. following from fig. 4.9b to 4.9d:
In our country, they were commercially used for the first time already in 19691973 in
the oldest versions of NE01 and NE03 programs by I. Nemec. This revealed that the
continuity of the first derivative can be sometimes unwelcome, e.g. in multicell slabs with
linear hinges in the longitudinal direction it is necessary to admit the break in the transverse
direction. This can be solved through doubling deformation parameters ,
y xy
w w in all nodes
(fig. 4.9e).
For n = 1 3 5 7
Continuity: function + 1
st
derivative
1
st
and 2
nd
derivative
up to 3
rd
derivative
For Kirchhoff
plate:
of deflection
rotation without
breaks
curvature
(moments)
shear forces
In planar elastic
problem:
of displacement
deformation
components
deformation
components
without breaks
Saint
Venant
equations
4.1 Geometric properties of elements
120
Figure 4.9:
a) The oldest known rectangular element with four nodes in vertices.
b) Generalisation to a rhomboid in oblique coordinates x,y.
c) Four deformation parameters in a bilinear element.
d) 16 deformation parameters in a bicubic element.
e) Doubling of parameters to ensure discontinuity (break) in the ydirection.
f) Character of the boundary of a region divided to rectangles, the region may contain openings (hatched).
g) Character of the boundary of a region divided to rhomboids.
h) Quadrilateral element composed of four triangular subelements.
i) Effective element of NE14, NE15 programs. When two neighbouring nodes coincide, it transforms itself
without any problems into a triangular element composed of three triangles.
4.1 Geometric properties of elements
121
Ahlins elements offer a big numerical advantage in the concentration of all parameters into
nodes = vertices. Polynomial (4.1.51) with
2
( 1) n + coefficients, if we admit only odd degrees
2 1 n m + , require that just
2
( 1) m+ parameters are introduced in the nodes. For 0 m , 1 n
it is just one parameter (fig. 4.9c), for 1 m , 3 n there are four parameters (fig. 4.9d), etc.
A practical disadvantage of Ahlins elements in the original form is that they are capable of
covering only certain areas (fig. 4.9f and 4.9g). Their boundary must be composed of line
segments that are parallel with axes , x y . This disadvantage can be overcome through the
transformation of the shape into a general quadrilateral, see the following article. Another
possibility is to compose the quadrilateral of four triangles as in fig. 4.9h. The internal node is
located in the intersection of two diagonals defined by the midpoints of the sides. This
guarantees geometrical regularity in the case that the element degenerates into a triangle due
to the coincidence of two adjacent nodes. This, together with the introduction of rotation
parameters of deformation for the planar stressstate, leads to a very effective modern node in
fig. 4.9i. All deformation parameters, in total 4 6 24 , are concentrated into its four
vertices.
4.1.5.4 Iso, hypo and hyperparametric elements
The basic operation for the definition of these elements is the projection of a square
onto an arbitrarily curvilinear quadrilateral. An additional operation consists in the selection
of base functions most often components of displacement or rotation on this quadrilateral.
Usually, both operations use the same functions that are called in the first operation shape
functions and in the second one base functions, which gives rise to isoparametric elements.
Nodal coordinates ( , )
v v
of the arbitrarily curvilinear quadrilateral can be considered to be
the element shape parameters. The values of the unknown function in these nodes ( , )
v v v
p
are the parameters of deformation related to the base functions, most often components of
displacement or rotation of these nodes (fig. 4.10). The oldest isoparametric element is in fact
the triangle from art. 2. with six shape parameters ( , )
v v
and six deformation parameters
( , )
v v
u v , 1, 2, 3 v , linear shape and base functions, known already in 1956. Also its extension
by nodes in the centres of sides ( 4, 5, 6) v , with twelve parameters of both shape and
deformation (Veubecke, 1965) and quadratic functions, is an isoparametric element
(fig. 4.7b).
It is however possible to define a triangular element with straight sides ( 1, 2, 3) v
over which the displacement components follow quadratic functions and in which other nodes
in the centres of sides are utilised ( 1 v to 6). Such an element is in terms of base functions
hyperparametric and in terms of shape functions hypoparametric. A similar example of a
quadrilateral can be shown in fig. 4.10a (selection of the shape) with a higher base function
with higher number of nodes, e.g. 4.10b,c, being selected for the distribution of the unknown
function.
If we lower base functions in the selected element, e.g. only linear functions with
nodes 1, 3, 5, 7 for the element from fig. 4.10b) and the values in nodes 2, 4, 6, 8 are
determined only by these linear functions (interpolated) and are not among the unknown
parameters of deformation we obtain an element that is hypoparametric in terms of base
4.1 Geometric properties of elements
122
functions and hyperparametric in terms of shape functions.
In practice mainly isoparametric elements are used. Their analysis and properties are
based on special functions called parametric functions ( , ) p , analysed in detail in Czech
publication [5], pp. 148 to 196, accompanied with a great number of examples. The same
publication also removes errors that commonly appeared in these functions between 1965 and
1978 even in contributions by prominent authors. The error usually resulted from an incorrect
generalisation of simple bilinear functions to what is termed biquadratic, bicubic, etc.,
functions without verification of values of the coefficients and absolute terms. The procedure
was strongly influenced by a very popular class of polynomials called by O.C. Zienkiewicz
[2] serendipity. For more precise definition see [5]. Users of parametric functions should
know just the main piece of knowledge that it is better to define these functions separately for
every degree 1, 2, 3 n , etc., as general expressions are almost always incorrect.
4.1 Geometric properties of elements
123
Figure 4.10:
Projection of a unit square with vertices (t1, t1) onto a general curvilinear quadrilateral, nodal points for degree:
a) n=1 (only straightline sides), c) n=3 (cubic parabolas).
4.1 Geometric properties of elements
124
The nature of parametric functions can be illustrated on three examples from fig. 4.10. Let us
apply the procedure by A. Zenisek from [5], which holds for an arbitrary degree n .
Let us have a square with vertices ( 1, 1) , (1, 1) , ( 1, 1) , ( 1, 1) in the Cartesian coordinate
system , . Let us divide its sides to n identical intervals. There are in total 4( 1) n
dividing points and together with square vertices we get 4n nodal points. Depiction of these
points for n 1 to 3 is clear from fig. 4.10. Square vertices are now marked with symbols
*
1
( 1, 1) A ,
*
1
( 1, 1)
n
A
+
,
*
2 1
( 1, 1)
n
A
+
,
*
3 1
( 1, 1)
n
A
+
.
Let us assume polynomials in the form
( , )
( , )
a b
ab
a b
p
(4.1.52)
where the summation is carried out over all integer pairs ( , ) a b having the following
properties:
1. 0 a n , 0 b n ,
2. only one of the numbers a,b can be greater than one. Set [ ( , ) a b ] of all these pairs is thus of
the following form
( ) [ ( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ]
, 0, 0 , 0,1 , ..., 0, ,
1, 0 , 1,1 , ..., 1, ,
2, 0 , 2,1 , 3, 0 , 3,1 ,..., , 0 , ,1
a b n
n
n n
1
]
(4.1.53)
and has 4n members.
The set of all polynomials (4.1.52) forms 4n dimensional linear function space. For our
needs we seek such a base of this subspace (i.e. 4n base functions
1
( , ) p ,
2
( , ) p , ...,
4
( , )
n
p ) which has an advantageous feature for FEM: all base functions ( )
i
p i k have
nonzero value in nodes
*
k
A , only function
k
p has the value equal to one in the circumstance.
This can be written by means of Kroneckers symbol
ik
, which is equal to zero for i k and
equal to one for i k , in a commonly used form:
( )
*
( ) , 1, 2,..., 4
i k ik
p A i k n (4.1.54)
The literature cited in [5] correctly derives the most common situations according to fig. 4.10,
which will be mentioned here explicitly for their significant importance in FEM:
1 n (fig. 4.10a)
( )( ) ( )
1
4
( , ) 1 1 1,..., 4
v v v
p v + + (4.1.55)
where ,
v v
are coordinates of vertices
*
k
A , i.e. ( 1, 1) , ( 1, 1) , ( 1, 1) , (1, 1) .
2 n (fig. 4.10b)
4.1 Geometric properties of elements
125
( )( )( ) ( )
( ) ( ) ( )
( )( ) ( )
1
4
2
1
2
2
1
2
( , ) 1 1 1 1, 3, 5, 7
( , ) 1 1 2, 6
( , ) 1 1 4, 8
v v v v v
k k
k k
p v
p k
p k
+ + +
+
+
(4.1.56)
3 n (fig. 4.10c)
( )( ) ( ) ( )
( )( )( ) ( )
( ) ( ) ( ) ( )
2 2
9 10
32 9
2
27 1
32 3
2
27 1
32 3
( , ) 1 1 1, 4, 7,10
( , ) 1 1 3 2, 3,8, 9
( , ) 1 1 3 5, 6,11,12
v v v
k k k
k k k
p v
p k
p k
+ + +
+ +
+ +
(4.1.57)
For 2 n the nodal coordinates are, respectively, according to fig. 4.10b ( 1, 1) , (0, 1) etc.,
up to ( 1, 0) , for 3 n (fig. 4.10c) the twelve node coordinates are ( 1, 1) ,
1
3
( , 1) ,
1
3
( , 1) , (1, 1) ,
1
3
(1, ) , etc., up to
1
3
( 1, ) . Property (4.1.54) can be easily verified through
direct substitution. Experienced readers probably anticipate the influence principle that will be
now explained.
The most illustrative is the situation for 1 n (fig. 4.10a). In terms of element shape,
formula (4.1.55) can be applied as what is termed shape function, i.e. for the transformation
of unit square (its sides are equal to 2) with vertices according to the text following (4.1.55)
and fig. 4.10a) left to a general quadrilateral. Each point ( , ) can be assigned
unequivocally a point ( , ) x y according to a simple rule of transformation of a rectangular
mesh (net) with identical meshes (elements) (in fig. we selected
1
3
) to a
more general mesh specified by uniform division of sides of the quadrilateral to identical
intervals (there are six them in fig. 4.10). It can be easily verified that coordinates of point
( , ) x y that is an image of point ( , ) are
4 4
1 1
( , ) ( , )
v v v v
x x p y y p
(4.1.58)
where p are functions (4.1.55), ( , )
v v
x y coordinates of four vertices 1, 2, 3, 4 v . In
particular for the vertices we get correct relations:
( ) ( ) ( ) ( )( ) ( )
( )( ) ( ) ( ) ( ) ( )
( )( ) ( ) ( )( ) ( )
( ) ( ) ( ) ( )( ) ( )
1
1 1 1 4
1
2 2 2 4
1
3 3 3 4
1
4 4 4 4
1 1 1 1 1 1 0 0 0
0 1 1 1 1 1 1 0 0
0 0 1 1 1 1 1 1 0
0 0 0 1 1 1 1 1 1
x x x
x x x
x x x
x x x
+ + + + +
+ + + + +
+ + + + +
+ + + + +
(4.1.59)
and similarly for
1
y to
4
y . For centre 0, 0 we obtain
[ ]
[ ]
4
1 2 3 4
1
4
1 2 3 4
1
1 1
(1 0)(1 0) 1 1 1 1 1 1
4 4
1 1
(1 0)(1 0) 1 1 1 1 1 1
4 4
c v
c v
x x x x x x
y y y y y y
+ + + + +
+ + + + +
(4.1.60)
4.1 Geometric properties of elements
126
which are the coordinates of the centroid of the quadrilateral. Notice please that each vertex
1 v to 4 or more precisely its coordinates ,
v v
x y contribute to coordinate , x y just by value
( , )
v
p , which can be perceived as an influence or source function for coordinate x or y .
This is an analogy to the sum of effects of four loads P on a certain static quantity if we
know its influence surface.
What is important is the fact that formulas (4.1.58) and (4.1.55) assign (again) a straight line
segment to the points of a certain straight line segment (that is a part of a rectangular mesh
fig. 4.10a left, e.g.
0
const . ). The reason is that only changes linearly and thus
(4.1.58) becomes an equation of a straight line segment with parameter . Consequently, also
sides 1 , 1 , 1 , 1 become straight line segments sides of a quadrilateral.
Similarly, it can be proved that formulas (4.1.56) and (4.1.57) transform a rectangular mesh
, to a curvilinear mesh composed of parabolas of second and third order, which holds
even for the sides of the element. This way we can obtain elements with curved sides. They
are useful in applications where we need rather smooth substitution of the boundary of the
analysed domain one curved side on the boundary is sufficient, others may be straight. In
general, the influence principle is still valid for an arbitrary n :
4 4
1 1
( , ) ( , )
n n
i i i i
i i
x x p y y p
(4.1.61)
For 1 n formulas (4.1.58) are with notation i v (nodes are only in vertices v ). For
2, 3 n , etc., the nodes are also located on sides of elements outside of the vertices, are
marked k and together with v are numbered 1 i to 8 (fig. 4.10b), 1 i to 12 (fig. 4.10c),
etc. A general summation expression proved in [5] is useful for the verification of the
formulas:
4
1
( , ) 1
n
i
i
p
(4.1.62)
In terms of the distribution of displacement components ( , ) u x y , ( , ) v x y within the extent
of one element, we can use polynomials ( , )
i
p as base functions. We use a formula similar
to (4.1.61):
4 4
1 1
( , ) ( , ) ( , ) ( , )
n n
i i i i
i i
u x y u p v x y v p
(4.1.63)
However, instead of coordinates of nodes 1 i to 4n , we now handle their displacement
components ,
i i
u v , which are in FEM used as deformation parameters. They are shared by
elements that have a common node, regardless of whether a vertex v or sidepoint k . This
safely ensures the required continuity of functions , u v over the whole domain. It is not
necessary to select in (4.1.63) the same n as in (4.1.62). If we do so, we get an isoparametric
element with exactly the same number of shape parameters (coordinates) as deformation
parameters (displacement components). It is the most frequent element of this type. Otherwise
we get a hypoparametric or hyperparametric element, as already stated at the beginning
of this article, in terms of shape or distribution of base functions, which must be for the
needs of users explicitly stated in the manuals.
4.1 Geometric properties of elements
127
4.1.5.5 Surface elements recommended by the authors
Planar elasticity and deformation
Geometry is defined in plane. It is possible to solve both cases of planar elasticity:
 planar elasticity loading, as well as reactions act in wall plane. Vector of stress:
, , 0, , 0, 0
T
x y xy
1
]
= , after substitution to physical equations:
( )
( )
( )
1 1
1 1
1 1
x x y z xy xy
y y x z yz yz
z z x y zx zx
E G
E G
E G
1
+
]
1 +
]
1
+
]
we will reach vector of strain: , , , , 0, 0
T
x y z xy
1
]
= . Deformation ( , , ) w x y z in direction
upright to the plane of plate exists, but it is irrelevant. Wall in this direction freely change
thickness, nothing prevent this.
 planar deformation  layer with thickness 1 h cut from the points that does not allow
deformations in direction upright to the wall plane. Vector of strain: , , 0, , 0, 0
T
x y xy
1
]
= .
We will reach stress vector through inversion of physical equations: , , , , 0, 0
T
x y z xy
1
]
= .
Wall try to deform also upright to its plane, what is prevented by adjacent layers.
In next we work with reduced vectors of stresses and strains:
x
y
xy yx
1
1
1
1
]
Missing components is possible to computed of physical equations. Thickness h within one
element is constant.
Vector of strain and corresponding geometric equations that comprise only linear part:
x
y
xy
u
x
v
y
u v
y x
1
1
1 1
1
1
1
1
]
1
+
1
]
In solution of geometrically nonlinear tasks by the NetonRaphson method is used
4.1 Geometric properties of elements
128
quadratically Green vector of strain
II
that comprises nonlinear terms that are in theory of
small deformation neglected:
2 2
2 2
II
1
2
1
2
u v
x x
u v
y y
u u v v
x y x y
1 1
_ _
+ 1 1
, ,
1 1
]
1
1
1 _ _
+ 1
1
1 , ,
1 ]
1
1
+
1
]
Element could be physically isotropic, or orthotropic. Matrix of physical constant
M
C :
11 12 13
22 23
33
.
M
C C C
C C
sym C
1
1
1
1
]
C
In case of isotropy is valid:
13 23
0 C C
planar elasticity: planar deformation:
( )
11 22 2
33
12 2
1
2. 1
.
1
E
C C
E
C
E
C
( )
( ) ( )
( )
( ) ( )
( )
( ) ( )
11 22
33
12
E. 1
C C
1 . 1 2.
E. 1
C
2. 1 . 1 2.
.E. 1
C
1 . 1 2.
+
Material orthotropy is given by angle of rotation  angle of orthotropy to planar coordinate
element system.
Vector of deformation in case thermal loading:
0
[ , , 0]
T
T T , where T is change of
temperature from stadium of preparation (zero elasticity) after actual status 1  fig. 4.11.
Figure 4.11:
Uniform change of temperature.
4.1 Geometric properties of elements
129
Result of solution are at:
 planar elasticity of internal forces n (normal and shear):,
x
y
xy yx
n
n
q q
1
1
1
1
]
n
 planar deformation of stresses (normal and shear):.
x
y
xy yx
1
1
1
1
]
Membrane isoparametric elements with rotational degrees of freedom (Serendipity family)
Each vertex of element has 3 degrees of freedom  2 translations in wall plane walls and
rotation: [ , , ]
T
z
u v u .
Numeric integration of matrices
There is utilised Gauss numeric integration that converts integral to sum of products of
weighted coefficients and functional values of integrated function. Number of integration
points and their position is for each element different; their exact description is presented at
description of each element. At composition of matrix of elements by numeric integration is
not allowed to change its parameters, numeric integration is firmly bounded with each type of
element and there is selected such number of integration points and their position that
provides best results comparing to amount of machine instructions. Integration on unit square
is:
( ) ( )
1 1
1 1
1 1
, ,
m m
i j i j
i j
f d d ww f
where
m is a number of integration points,
w are weight coefficients,
( ) , f are functional of value of shape function in corresponding point.
For 3node element is used analogical procedure, boundary of integral are 0,1 .
Elimination of nodes in centre of sides
Field of deformation parameters of 3nodal and 4nodal element is approximated by quadratic
multinomial, original geometry is with node in centre of each side, fig. 4.12. On each side of
the element is anticipated that is straight. This enables to eliminate central node, what is
4.1 Geometric properties of elements
130
moreover favourable from aspect of width of the system of equations and thus consumption of
machine time. Resulting geometry of the element is described freely of centre node it is
linear. Resulting element is than subparametrical.
Figure 4.12:
Edge of element with centre node.
Original elements have nodes at the centres of sides [ , ]
k k
u v , these nodes are condensed by
transformation of deformation parameters to vertices by interpolation:
( ) ( )
1
2 8
j i
k i j zj zi
y y
u u u
+ +
( ) ( )
1
2 8
j i
k i j zj zi
x x
v v v
+ +
There will arise element that is compatible with other finite elements 1D and 3D that have
corresponding degrees freedom in rotation.
Spurious mode control
For isoparametric elements is necessary to numerically stabilize elements, to avoid zero
energy modes. In case of 3nodal element there appears only equal rotation of spurious mode,
in case of 4nodal element there is also hourglass mode. Stabilisation consists in adding of
small contribution to the potential energy to element. Size of the parameter
R
and
H
is
not possible to be changed.
1/ Hourglass mode
It occurs only at 4nodal beams if there is used reduced integration. This mode vanishes, if
there is used disordered net, or more elements, fig. 4.13.
There is valid:
4.1 Geometric properties of elements
131
1 2 3 4 z z z z
Element has added energy
H
E :
H H H xy H
E V G
H
parameter of energy (
3
1 10
)
V element volume
( )
1
1 2 3 4 4 H z z z z
+
xy
G modulus of elasticity in shear
Figure 4.13:
Hourglass spurious mode.
2/ Equal rotation mode
It occurs at 3nodal and 4nodal elements, fig. 4.14.
There is valid:
1 2 3 4 z z z z
Element have added energy
R
E :
R R R xy R
E V G
R
parameter of energy (
6
1 10
)
V element volume
R
relative rotation, computed in the centre of the C element:
4.1 Geometric properties of elements
132
1
1
.
n
R C zi
i
n
,
where
1
.
2
C
C C
v u
x y
_
,
Figure 4.14:
Equal rotation spurious mode.
4nodal element:
Figure 4.15:
8nodal origin 16 DOF membrane without RDOF.
4nodal new 12 DOF membrane with RDOF.
Number of nodes:
8
Unit coordinates of nodes:
{ 1, 1}, {1, 1}, {1, 1}, { 1, 1}
Shape functions:
4.1 Geometric properties of elements
133
( ) ( )( )
( )( )( )
( )( )( )
( ) ( )( )
( )( )
( )( )
( )( )
( )( )
1
1 4
1
2 4
1
3 4
1
4 4
2
1
5 2
2
1
6 2
2
1
7 2
2
1
8 2
1 1 1
1 1 1
1 1 1
1 1 1
1 1
1 1
1 1
1 1
N
N
N
N
N
N
N
N
+
+ + +
+ +
+
+
Numeric integration:
Reduced (2 2)
integration points and weight multiplier:
[ ]
1 3, 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3
1, 1, 1, 1 w
1
]
1
]
Evaluation:
Function of surface on extrapolation of qualities of integration points to peaks:
1 2 3 4
F a a a a + + +
Loading:
Element could be loaded:
of node force (moment) in the node,
uniform loading on edge,
volume loading,
relative deformation of surface caused by uniform warming up, eventual uniform
shrinkage.
3nodal element:
4.1 Geometric properties of elements
134
Figure 4.16:
6nodal origin 12 DOF membrane without RDOF.
3nodal new 9 DOF membrane with RDOF.
Unit coordinates of nodes:
{0, 0}, {1, 0}, {0, 1}
Shape functions:
( )
( )
2 2
1
2
2
2
3
2
4
5
2
6
1 3 3 2 4 2
2
2
4
4
4
N
N
N
N
N
N
+ + +
+
+
Numeric integration:
Selective reduced (4point), 3 for ,
x y
, 1 for
xy
integration points and weight multiplier:
[ ]
[ ]
[ ]
1 6, 2 3, 1 6, 1 3
1 3, 1 3, 2 3, 1 3
1 3, 1 3, 1 3, 1 w
Evaluation:
1 2 3
F a a a + +
4.1 Geometric properties of elements
135
Normal stresses are extrapolated of integration points to vertices according to the function F .
Shear stress is constant, there is evaluation centre of gravity of the element.
Matrices of finite elements
Sttiffness matrix
Dependency among vector field of deformation and unknown nodes by deformation
parameters u is expressed by:
L
Nu B u . Linear deformation matrix
L
B is obtained by
derivation of shape functions N .
Matrix of differential operators:
0
0
T
x y
y x
1
1
1
1
1
]
Matrix of shape functions:
1 2
1 2
0 0
0 0
N N
N N
1
1
]
N
K
K
Of potential energy of internal forces
i
:
1
2
T
i
d
1
]
%
Than, geometric matrix of the element
G
K is:
T
G NL NL
d
K B B %
Plate elements
Planar constructions, whose all points lay in one common plane. Thickness h is small,
comparing to dimensions of the plate. At determination of characteristic dimension L of the
plate (what is in case of rectangular plate a shorter dimension, in case of round plate diameter)
it is possible to set up solutions of plate according to corresponding theory to:
34. Thin plate ratio 1 50 1 10 h L < <  technical theory of bend of thin plates, based on
Kirchhoffs assumptions. There belong almost all ceiling plates.
35. Thick plate  ratio 1 10 1 5 h L < < ReissnerMindlins theory that respects shear of
normal line slope due to shear deformation ,
xz yz
. There could be used also thin
plates, while it is necessary to eliminate shear of solving of cross section at small
thicknesses.
While the ratio is 1 5 h L >  construction should be already evaluate as 3D task, at
ratio 1 50 h L < the construction is necessarily to be solved a membrane, what is possible in
the shell model, or membrane force are accentuated and in overall energetic potential could
not be neglected.
Force loading acts upright to the plane of the plate, moment around axes of the plane
plate.
Normal line to the centreline plane of the element remains after deformation straight, it
depends on theory of calculation, whether it remain perpendicular to the centreline plane
(Kirchhoffs theory), or not (ReissnerMindlins theory).
Each vertex of element has 3 degrees of freedom: deflexions and rotation: [ , , ]
T
x y
w u .
Vector of deformation in case of thermal loading: [ ]
0
, , 0, 0, 0
T
T h T h where T
is difference of temperature on upper and bottom surface, fig. 4.17.
4.1 Geometric properties of elements
137
Figure 4.17:
Nonuniform change of temperature.
Result of solutions are internal moments and force:
x
y
xy yx
x
y
m
m
m m
q
q
1
1
1
1
1
1
1
]
m
Thickness of element is constant.
Loading:
Element could be loaded:
by node force (moment) in vertex,
by uniform force loading on edge,
by volume loading,
by curvature of surface caused by nonuniform warming up, eventually nonuniform
shrinkage.
Application of affect of flexible subsoil.
Flexible subsoil with shearing spreading of the KolarNemec type could be easily added to the
stiffness matrices. Stiffness matrix of subsoil is added to the stiffness matrix of the plate
element. Potential energy of subsoil is given by:
( )
2 2 2
1 2 2
1
2
S S S
p x xx y yy
C w C w C w d
+ +
Thin plates
Element is based on Kirchhoff theory of thin plate, where is neglected shear
4.1 Geometric properties of elements
138
deformation  slope. Deformations of plate are described by 1 function of 2 variables ( , ) w x y .
Rotations are derivations of the deflexions. Out of assumptions of this theory there is valid
, , 0, , 0, 0]
T
x y xy
= [ .
Normal line to nondeformed surface of the plate remains a normal line also after its
deformation, and under assumption of zero displacement , u v points of centreline surface
there is valid:
x
y
w
u z z
x
w
v z z
y
After substitution to the geometric equations:
x xy
y yz
z zx
u u v
x y x
v v w
y z y
w w u
z x z
+
+
+
we will reach vector of deformation that vitiates linearly by high of cross section h . We
describe this vector for short by the vector of curvature of deflexions surface , what is for
plate more convenient inscription:
2
2
2
2
2
2
.
x
y
xy
u w
z
x x
v w
z
y y
u v w
z
y x x x
1
1
1
1
1
1
1
1
1
]
1
+ 1
1
]
2
2
2
2
2
2
.
x
y
xy
w
x
w
y
w
x y
1
1
1
1
1
1
1
1
]
1
1
1
]
Between both vectors there is valid relation: z
Normal stress
z
is comparing to stresses
x
and
y
negligible small, they are not
considered. Vector of stress is [ , , 0, , 0, 0]
T
x y xy
. Out of condition of zero slope there
are zero also shear of component
yz
and
zx
. Discrepancy of assumptions with reality of this
theory is in:
1/
( )
1
z x y
E
1
+
]
physical equations, but 0
z
.
4.1 Geometric properties of elements
139
2/
1
0
1
0
yz yz
zx zx
G
G
,
but out of conditions of equilibrium there is valid that course of shearing stresses by high of
cross section is parabolic.
Number of nodes:
3
Coordinates of nodes:
{1, 0, 0}, {0,1, 0}, {0, 0,1} in planar coordinates
Deflexions function of the plate:
2 2 2 2 2 2
1 1 2 2 3 3 4 1 2 5 2 3 6 3 1 7 1 3 8 2 1 9 3 2 10 1 2 3
2 w a L a L a L a L L a L L a L L a L L a L L a L L a L L L + + + + + + + + +
Meaning of parameters
1 9
a
: deflexions w and rotations ,
x y
in each node.
Parameter
10
a is linear combination of parameters:
10 4 5 6 7 8 9
( ) 4 a a a a a a a + + + + + . To this
parameter appertain deflexions w in the centre of gravity of element, we work in calculation
free of internal nodes, and that is why we eliminate this parameter.
1 2 3
, , L L L are planar coordinates. At 3nodal element it is more convenient to work in
the system of planar coordinates, as in the Cartesian system, due to calculation of integrals.
Each planar coordinate
i
L of the item P expresses ration of surfaces:
( )
( )
, ,
, ,
i
A P j k
L
A i j k
Point P defined inside the element with planar coordinates
1 2 3
[ , , ] L L L is transformed to the
Cartesian system with coordinates [ , ] x y according to:
1 1 2 2 3 3
1 1 2 2 3 3
1 2 3
1
P
P
x x L x L x L
y y L y L y L
L L L
+ +
+ +
+ +
It follows that point 1 of the element has planar coordinates of [1, 0, 0], point 2 [0, 1, 0], and
point 3 [0, 0, 1], fig. 4.18.
4.1 Geometric properties of elements
140
Figure 4.18:
Area coordinates on the triangle
Inverse relating to antecedent 3 equations represent equations for calculation of planar co
ordinates of the point P, defined in the Cartesian system:
( ) ( ) 2
i i i i
L a b x c y A + + , accordingly for , j k .
Area of the element A could be defined e.g. according to:
1 1
2 2
3 3
1
1
det 1
2
1
x y
A x y
x y
i j k k j
i j k
i k j
a x y x y
b y y
c x x
we will obtain expressions for , , ,
j j k
a b c K by cyclic exchange of indexes.
We derive each function defined in the system of planar coordinates
1 2 3
( , , ) f L L L for needs
of stiffness matrix, vector of volume forces and internal forces as composite function
according to:
1 2 3 3 1 2
1 2 3
1 2 3 1 2 3
( , , ) 1
2
f L L L L L L f f f f f f
b b b
x L x L x L x L L L
_
+ + + +
,
1 2 3 3 1 2
1 2 3
1 2 3 1 2 3
( , , ) 1
2.
f L L L L L L f f f f f f
c c c
y L y L y L y L L L
_
+ + + +
,
4.1 Geometric properties of elements
141
Integration:
Implicitly analytically.
Calculation of integral is in this case simple, we advance according to Fellippe equation:
( )
1 2 3
! ! !
2
2 !
p q r
A
p q r
L L L dA A
p q r
+ + +
Thick plates
Element is based on assumption of ReissnerMindlins theory:
Normal lines to the centreline plane remain straight also after deformation, however, they
are not upright to the centreline plane of the plate, but there is neglected deplanation of cross
section,
Normal line of stress
z
is comparing to stresses
x
and
y
negligible small, such as in the
Kirchhoff theory. Of this condition their is also in this theory contradiction with 0
z
. This
phenomenon originates by reduction of task dimension.
Vector of deformation , , 0, , , ]
T
x y xy yz zx
= [ is composed of deflexion deformation of
surface , ,
x y xy
that are linearly changed by high of the cross section and constant slope by
high of the cross section ,
yz zx
.
y
x
x
y
xy
y
x
x
y
y x
1
1
1
1
1
1
1
1
]
1
1
1
]
y
yz
zx
x
w
x
w
y
1
+
1
1
1
1
1
]
1
]
While ,
x
w and
y
are independent variable, contrary of the the Kirchoff's plate theory.
The element could bet physically isotropic, as well as orthotropic, shape of physical matrix
B
C is:
11 12 13
22 23
33
44 45
55
0 0
0 0
0 0
.
B
C C C
C C
C
C C
sym C
1
1
1
1
1
1
1
]
C
In case of isotropy there is valid:
4.1 Geometric properties of elements
142
( )
( )
( )
( )
13 23 45
3
11 22
2
3
33
2
3
12
2
2
44 55
0
12 1
1
2 12 1
12 1
12 2 1
C C C
Eh
C C
Eh
C
Eh
C
Eh
C C
+
Material orthotropy is represented by angle of rotation of angle orthotropy to the planar co
ordinate system of the element.
LYNN  DHILLON
Element defined in the Cartesian coordinates with linear approximation functions with
adding of quadratic beams for deflexion for improvement of convergence. Problem with shear
locking at decreasing thickness is solved by introduction of numericstabilisation test:
( ) ( )
44 55 11 22
, 500 , C C C C A
where A is planar area of the element.
Number of nodes:
3
Coordinates of nodes:
2 2 3 3
{0, 0}, { , }, { , } x y x y
Shape functions:
( )
2 2
1 1 1
1 2 3 8 5 8 6 2 2 2
4 5 6
7 8 9
x
y
w a a x a y a x a a xy a y
a a x a y
a a x a y
+ + + +
+ +
+ +
Integration:
Implicitly analytically
4.1 Geometric properties of elements
143
MITC4
Isoparametric linear element according to [Bathe], elimination of shear locking at decreasing
thickness is done by mixed interpolation of deflexion, rotation and slope.
Qualities of the element:
element matrix is obtained by full Gausss numeric integration ,
element has no zero mode of energy
Number of nodes:
4
Unit coordinates of nodes:
{ 1, 1}, {1, 1}, {1, 1}, { 1, 1}
Shape functions:
( ) ( )
( ) ( )
( )( )
( )( )
1
1 4
1
2 4
1
3 4
1
4 4
1 1
1 1
1 1
1 1
N
N
N
N
+
+ +
+
Numeric integration:
Full (2 2)
Integration of points and weight multiplier:
[ ]
1 3, 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3
1, 1, 1, 1 w
1
]
1
]
Shell structures
Facetshell structure is an operating term for planar (2dimensional) structures situated in 3
dimensional space and loaded so that there is not possible in them to separate the plate impact
from the wall impact. There is used an element from planar tension. Bending element depends
on selected plate theory. The final stress vector: [ , , 0, , , ]
T
x y xy yz zx
, strain vector:
[ , , , , , ]
T
x y z xy yz zx
.
Internal forces are composed of a vector of membrane stresses and vector of m bending
moments and shear forces.
4.1 Geometric properties of elements
144
x
y
xy yx
1
1
1
1
]
x
y
xy yx
x
y
m
m
m m
q
q
1
1
1
1
1
1
1
]
m
The matrix of physical constants of the element C is composed of a bending part
B
C ,
membrane part
M
C . Material orthotropy is defined by the angle of the rotation of the angle
of orthotropy to planar coordinate system of the element, into which are because of so
defined element transformed physical quantities.
In case of physical nonlinearity, (for example by origination of cracks along the height of a
section) will come to moving of the position of bearing of centreline plane in comparing to
the original one, which divided the width of the element into 2 equal parts. This effect
respects a sub matrix
BM
C , it describes cohesion of wall and plate effects:
.
B BM
M
sym
1
1
]
C C
C
C
Large deformations
In solving of geometrically nonlinear tasks by the NetonRaphson method is used quadratic
element of Green vector in deformation
II
, which includes nonlinear terms that are
neglected in the theory of small deformations:
2
2
II
1 w
2 x
1 w
2 y
w w
x y
1 1
_
1 1
,
1 1
]
1
1
1 _
1
1
1 ,
1 ]
1
1
1
]
Subsoil
Every finite element can have along the entire surface continuous contact with effective
model of a subsoil of the KolarNemec type, which is defined by five constants in planar
coordinates [ , , ]
p p p
x y z :
1 1 1 2 2
S S S S S
x y z x y
C C C C C
Relevant forces for deformations are membrane
,
( )
u v
r r like and bending ( , , )
w x y
r t t like.
4.1 Geometric properties of elements
145
1 2
1 2
1
p
p S p p S
u x x x p
p
p S p p S
v y y y p
p S p
w z
w
r C u t C
x
w
r C v t C
y
r C w
The constants
1 1
,
S S
x y
C C express the resistance against planar movements of centreline plane of
the element (friction). The constants
1 2 2
, ,
S S S
z x y
C C C are coefficients of relations expressing
resistance of surrounding against the movement and angular rotation. The constant
1
S
z
C
responds to the Winkler model,
2 2
,
S S
x y
C C Pasternak model. In most cases is
1 1
S S
x y
C C and
2 2
S S
x y
C C .
Sandwich elements
The element is along the height divided into a few (at least 2) isotropic, or orthotropic layers,
ideally resistant linked together, so that there does not come to shearping. Every layer has its
physical parameters, which are possible to solve so far only by physical linear elasticity.
According to NewtonRaphson is possible to solve such elements by geometrical non
linearity. The number of layers is unlimited. The thickness of every layer is constant. Load of
an element is analogous, as well as with facetshell elements.
Alternate homogeneous crosssection with ideal static quantities, which are dealt with in
calculation will be created from a composite crosssection. For specification we alternate
resistance come out of equality of work of internal and external forces. Evaluation of internal
forces comes off in optional point along the height of a crosssection, where on the basis of
relevant equations of elasticity is finished calculation of final quantities.
Example:
Bimetallic strip  console, constant change of the temperature.
7
3 10 E [Pa], 0 ,
1 2
0.05 h h [m], 100 T [K],
5
1
1 10
,
5
2
2 10
. The length of a beam
10 L [m], the width 1 [m].
Figure E4.1:
Geometry, crosssection of an element and results process of stresses along the height of a crosssection.
4.1 Geometric properties of elements
146
Deformations and stresses
x
u [m]
z
u [m]
1I
[Pa]
2I
[Pa]
1II
[Pa]
2I I
[Pa]
numerically 0,150 0,750 7500 15000 15000 7500
analytically 0,150 0,750 7500 15000 15000 7500
Results.
4.1 Geometric properties of elements
147
4.1.6 3D elements
4.1.6.1 Tetrahedron
The simplest and oldest 3D finite element is a tetrahedron as a natural extension of a
sequence of what is termed simplex in D n space. A simplex is the simplest shape, a line
segment in 1D, triangle in 2D, tetrahedron in 3D, superpentahedron in 4D, etc. It is
intercepted by the lowest possible number of shapes of dimension ( 1)D n , i.e. for 1 n by
two points (dimension 0D), for 2 n by three line segments (dimension 1D), for 3 n by
four triangles (dimension 2D), for 4 n by five tetrahedrons (dimension 3D), etc. Spaces D n
or ( 1)D n can be curved, i.e. they can be immersed into spaces ( 1)D n or D n , which gives
rise to curved elements.
We will describe the most frequently used 3D tetrahedron intercepted by four planar
triangles. It is rare for presentday programs to offer this element separately. It usually forms
a subelement of a brickshaped element, which is a prism, or more generally a block or an
arbitrary hexahedron, etc. see art. 4.1.6.2. A whole hierarchy of polynomials in , , x y z of
degree 1, 2, 3, n K on a tetrahedron is known, with m coefficients, where
( ) ( )( )
1
6
1 2 3 M n n n + + + (4.1.64)
which for 1, 2, 3, n K is in turns 4, 10, 20, ... coefficients with monomials x y z
,
( ) 0 1, 2, 3 < + + , etc. This corresponds for each of the three displacement
components , , u v w to the same number of m deformation parameters in m nodal points
which must be selected in compliance with certain rules. The first three situations are shown
in fig. 4.19ac. The tetrahedron has in total 3m parameters of deformation. These include
components of displacement , , u v w and possibly also their derivatives in nodes. For 1 n
and 2 only displacement parameteres are udes, in fig. 4.19a)b) marked by full circle. It
represents 3 4 12 ( 1) n and 3 10 30 ( 2) n parameters. For 3 n we have
3 20 60 parameters. The components and their first derivatives (marked by index) must be
introduced in all vertices, i.e. in total 3 4 4 48 quantities ( , , , , , , )
x y z z
u u u u v w K . In
addition, also introduced must be 3 4 12 displacement components ( , , ) u v w in centroids
of the sides (fig. 4.19c). A complete hierarchy was elaborated by A. Zenisek [5]. Up to 8 n ,
these polynomials guarantee the continuity of , , u v w only in function values, i.e. function
class
0
C . Only the polynomial of degree 9 with 220 coefficients and 660 deformation
parameters can generate function class
1
C , and continuity also in the first derivatives, i.e. in
components of strain . This element was normally unworkable in technical applications due
to the performance capacity of presentday computers.
It is now convenient to make a summary of features of polynomials in simplexes in
space 1D, 2D, 3D, i.e. in a line segment, triangle and tetrahedron. If the dimension of the
space is d (1,2,3), the continuity in function values
0
( ) C can be achieved already by a
polynomial of first degree ( , , ) a bx a bx cy a bx cy dz + + + + + + with 2,3,4 coefficients.
Continuity in the first derivatives (for 1 d > partial ones) requires a polynomial of degree 3, 5,
4.1 Geometric properties of elements
148
9 with 4, 21 and 220 coefficients. The continuity in second derivatives needs polynomials of
degree 5, 9, 17 with 6, 55 and 1,140 coefficients, etc. The number of coefficients relates to
just one approximation function, which means that the number of parameters of deformation
is either the same (if we work with just one unknown function, e.g. deflection ( ) w x in a 1D
problem) or higher, e.g. three times if we have a 3D problem with three unknown
displacement components , , u v w. As the demand for what is termed pversion of FEM,
which improves the solution for a rather coarse mesh and large elements through increasing
the polynomial degree in elements, has increased in recent years, we will address this issue
again in art. 4.3.3.
4.1 Geometric properties of elements
149
Figure 4.19:
3Delements FEM: ac) Tetrahedrons with polynomials of degree 1  3. df) Bricks with trilinear, triquadratic
and tricubic polynomials. Tetrahedron b) with quadratic polynomials and 3 x 10 = 30 parameters of deformation
of type u, v, w can be modified to a significantly more effective element with 4 x 6 = 24 parameters of type u, v,
w,
x
,
y
,
z
only in vertices with preserved favourable properties of quadratic approximation as explained in
art. 4.1.6. Then there are no parameters in the centres of sides. Similar improvement can be applied to the brick
in fig. e), again through parameters of rotation , which means that it has only 8 x 6 = 48 parameters, all in
vertices, see art. 4.1.6.1 and 4.1.6.2 and cited literature. Merging of all upper nodes 3, 11, 4, 20, 8, 15, 7, 19 into
a single node can give a useful element in the shape of a foursided pyramid, which does not cause any
mathematical difficulties.
4.1 Geometric properties of elements
150
4.1.6.2 Bricks
The most often used 3D elements are elements termed bricks. This originally slang
term has been widely used in FEM since 1960. It has been adopted in most languages and
normally is not translated. Originally, it was a body with the shape of a cube, prism or block
with six planar or curved sides, i.e. a hexahedron, or depending on the number of sides also
dodecahedron with eight vertices. Later on, other shapes were developed too, e.g.
pentahedrons (fig. 4.20). The main advantage against tetrahedrons from art. 4.1.6.1. is better
orientation in the division of the domain and the possibility to introduce numerically
convenient base functions. During the years of FEM development, several bricks were
constructed as pure superelements composed of five tetrahedrons, which, apart from better
orientation, brought no other advantages. Progress was made only when what is termed
isoparametric 3D elements was introduced.
It is in fact a consistent extension of the idea of 2D isoparametric elements (art.
4.1.5.4.) to a 3D space. This extension was initially (19621968) made through a formally not
well thought out addition of z terms to , x y terms. Defect elements were created even in top
ranking FEM centres, e.g. B. M. Irons in Swansea, Wales, in 1975 honestly admitted errors in
his base functions following the correspondence with authors of publication [5], where A.
Zenisek published the first correct formulas for what is termed triquadratic and tricubic
polynomials. The rule from 2Ddomains that each polynomial degree must be defined
separately and general formulas fail holds for 3D bricks as well. In order to illustrate the
stated, let us present the correct formulas for the first three polynomial degrees called briefly
tri n th, in the domain of a unit cube with the side equal to 2 and vertices as in fig. 4.19df,
the centre of the cube is in the coordinate origin (0,0,0). Generally, the tri n th polynomials
have the following form:
( , , )
( , , )
a b c
abc
a b c
p a
(4.1.65)
where the summation is carried out over all integer triplets ( , , ) a b c having the following
properties:
1. 0 , 0 , 0 a n b n c n ,
2. only one of the numbers , , a b c can be greater than one.
It is clear that for 1 n we have a polynomial with the highest term of degree 3 , for
2 n the highest terms are of degree 4
2 2 2
, , , for 3 n the highest terms have
degree 5
3 3 3
, , etc. Based on the highest power of one coordinate, a rather vague
name tri n th polynomial is used, 1, 2, 3 n , etc. The simplest ones are trilinear
polynomials ( 1) n , assigned in turns to individual vertices of the cube in fig. 4.19d with
coordinates ( , , )
v v v
, 1 v to 8, which reach only unit values +1 or 1:
4.1 Geometric properties of elements
151
( )( ) ( ) ( ) ( , , ) 1 8 1 1 1 1, 2,..., 8
v v v v
p v + + + (4.1.66)
Features of polynomial (4.1.66) in 3D are identical to the features of polynomial (4.1.55) in a
2D problem, see art. 4.1.5.4.
If we use it as the shape function in the meaning of formula (4.1.61) which is in 3D
extended by coordinate z :
12 4
1
12 4
1
12 4
1
( , , )
( , , ) 1 , , 1
( , , )
n
i i
i
n
i i
i
n
i i
i
x x p
y y p
z z p
(4.1.67)
then for 1 n the sum deals with 12 4 8 terms assigned to 8 vertices 1 i v to 8.
Rectangular mesh , , is transformed by (4.1.67) to the mesh of straight line
segments, and also edges of the cube are transformed to straight line segments and sides of
the cube becomes a warped line surface with the shape of a hyperbolic paraboloid, fig. 4.19d.
We get a brick of degree 1. If we use (4.1.66) also as base functions in the meaning of
formula (4.1.63), extended in 3D by displacement component w:
12 4
1
12 4
1
12 4
1
( , , )
( , , ) 1 , , 1
( , , )
n
i i
i
n
i i
i
n
i i
i
u u p
v v p
w w p
(4.1.68)
then the displacement components , , u v w follow inside the 3D brick trilinear polynomials
( 1 n ). This creates the simplest isoparametric 3D brick. Its shape is defined by coordinates
of eight vertices, its displacement components by eight triplets ( , , ) u v w in these vertices, in
total 24 parameters ( , , )
v
x y z and deformation ( , , )
i
u v w , 1 i to 8. In a standard FEM
terminology we call it brick 24, which is today included almost in all FEM programs.
The next closest higher polynomial (4.1.65) for 2 n , triquadratic, cannot be written
using a collective formula for all 12 2 4 20 nodes (fig. 29e), which was in vain attempted
until circa 1970. We will present here the correct form according to [5]. Eight polynomials
relating to eight vertices
( , , )
v v v
are of the following form:
( )( )( ) ( ) ( , , ) 1 8 1 1 1 2
v v v v v v v
p + + + + + (4.1.69)
four polynomials relating to nodal points (0, , )
k k
in centres of sides :
( )( )( )
2
( , , ) 1 4 1 1 1
k k k
p + + (4.1.70)
four polynomials relating to nodal points ( , 0, )
k k
in centres of sides :
4.1 Geometric properties of elements
152
( )( )( )
2
( , , ) 1 4 1 1 1
k k k
p + + (4.1.71)
four polynomials relating to nodal points ( , , 0)
k k
in centres of sides :
( ) ( ) ( )
2
( , , ) 1 4 1 1 1
k k k
p + + (4.1.72)
in total 8 3 4 20 + polynomials, in domain 1 , , 1 .
In the present times, the highest degree of polynomial (4.1.65) implemented in FEM
programs is 3 n . The correct form of these polynomials is determined by their relation to
nodes of the element in fig. 4.19f. This represents 8 vertices and 12 2 27 nodes in the
thirds of the sides, in total 32 nodes. Eight polynomials relating to eight vertices ( , , )
v v v
are of the following form:
( )( ) ( ) ( )
2 2 2
( , , ) 1 64 1 1 1 9 19
v v v v
p
1
+ + + + +
]
(4.1.73)
four polynomials relating to nodal points ( 1 3, , )
k k
in one third of sides :
( )( )( )( )
2
( , , ) 9 64 1 3 1 1 1
k k k
p + + (4.1.74)
four polynomials relating to nodal points ( 1 3, , )
k k
+ in two thirds of sides :
( ) ( ) ( ) ( )
2
( , , ) 9 64 1 3 1 1 1
k k k
p + + + (4.1.75)
similarly eight polynomials relating to nodal points ( , 1 3, )
k k
t in the thirds of sides :
( )( )( )( )
2
( , , ) 9 64 1 1 3 1 1
k k k
p + t + (4.1.76)
and eight polynomials relating to nodal points ( , , 1 3)
k k
t in the thirds of sides :
( )( )( )( )
2
( , , ) 9 64 1 1 1 3 1
k k k
p + + t (4.1.77)
in total 8 3 8 32 + polynomials in domain 1 , , 1 .
Similarly to (4.1.62) in 2D problems, the summation statement must hold for all correctly
defined polynomials (4.1.65):
12 4
1
( , , ) 1
n
i
i
p
(4.1.78)
FEM practice, in addition to the already mentioned brick 24, commonly uses brick 60,
an isoparametric element as in fig. 4.19e with the same triquadratic polynomials (4.1.69) 
(4.1.72) for element shape (4.1.67) and also for the distribution of displacement components
(4.1.68). The edges of the element are generally formed by a parabola of second order passing
through three nodes, i.e. the vertices and the midpoint of the edge. The sides of the element
are generally curved. Program packages may also contain the isoparametric brick 96 as in
fig. 4.19f, with the edges formed by parabola of third order with one inflection allowed. These
elements are isoparametric, because they have the same number of shape and deformation
parameters. The shape of brick 60 is defined by means of 3 20 60 coordinates of nodes
( , , ) x y z and the deformation is prescribed by 3 20 displacement components , , u v w in
these nodes. For brick 96 we have 3 32 96 parameters of shape and deformation.
4.1 Geometric properties of elements
153
Analogously to 2D problems discussed in art. 4.1.5.4. it is possible to define hyper
or hypoparametric elements in terms of shape or deformation. It is applied to situations
when we need the best representation of the boundary (smooth without breaks) and, therefore,
we use higher polynomials for the shape functions. At the same time we limit ourselves to a
coarse approximation of displacement components, i.e. to a lower number of unknown
parameters of deformation, e.g. to a trilinear distribution specified just by nodal values
( , , u v w). For other nodes we content ourselves with the interpolation, in which we substitute
the coordinates of nonvertex nodes into the trilinear polynomials. This creates an element
that is hyperelastic in terms of shape and hypoelastic in terms of base. Nowadays, however,
this is not a typical approach as the computational capacity of presentday computers does not
represent such a strong limiting factor for the allowable number of unknowns, as it was in the
past.
The presentday FEM programs offer tetrahedrons and bricks that concentrate all the
parameters of deformation into the vertices. This is possible thanks to the rotation parameters
of deformation see the previous art. 3.3.2, form. (3.3.6), (3.3.7) and notes in subsequent
paragraphs. The algorithms of such elements are not simple, they require a large number of
transformations, additional functions (ESP = extra shape functions) and a special treatment of
singular states (penalty energy, penalty stiffness) in order to eliminate zero displacement
fields under certain configurations of vertex rotational parameters, which influences the
stiffness matrix in a way similar to excessive release of the body. The basic principles were
published in [93] and applied to [65, 73]. They can be briefly summarised as follows:
They are based on quadratic fields of displacement components, i.e. for a tetrahedron
on the element in fig. 4.19b) with 30 deformation parameters of displacement type and for a
brick from the element in fig. 4.19e) with 60 parameters of deformation of the same type, i.e.
, , u v w. All parameters relating to the midpoints of edges are eliminated through special
interpolation according to fig. 4.8f), which can be popularised as a beamlike state, as linear
interpolation is used for the effect of endpoint displacements and cubic interpolation for the
effect of rotational parameters that were added into all of the vertex nodes. Through standard
modifications we then obtain a tetrahedron stiffness matrix of dimension (24, 24) and a brick
stiffness matrix (48, 48). There are six parameters in every vertex (six degrees of freedom
, , , , ,
x y z
u v w ), i.e. for a tetrahedron in total 4 6 24 and for a brick 8 6 48 . The
matrix created in this way is, however, singular, as certain constellations of rotational
parameters lead to no (zero) displacement field. This can be avoided, similarly to normal
singularity, if zero values are prescribed for some rotational parameters in the analysed
structure, which is analogous to the necessity to define certain support conditions in order to
prevent the rigid body motion. This approach was, however, abandoned (in around 1980) for
many good reasons (subjectivity and the fact that the elements were still illconditioned).
What proved useful was ordinary stiffening of elements with regard to these effects, which
was applied from 1960 to eliminate equations of type 0 0 in complementary connections of
two planar systems in space. In terms of physics, this represents an insertion of certain
mechanisms into an element in which deformation energy forms even under the mentioned
circumstances (penalty energy, mentioned earlier in the text). Moreover, the overall
displacement field is amended (enriched) by special states ESP (extra shape functions) that
are quantified by, usually three, additional formal parameters that are related to the whole
element and can be imagined e.g. in the centroid. Publication [93] uses a purely Lagrangean
conception of the element. There exist even other elements, e.g. with mutually independent
displacement and rotation fields, the incompatibility of which decreases with the refinement
4.1 Geometric properties of elements
154
of the mesh thanks to the application of more general variational principles [27, 49, 51, 53,
62]. A list of literature contributing to the problem of rotational parameters contains over 100
items. It is neither feasible nor purposeful to present all of them to ordinary FEM users.
Therefore, those who are interested in this topical FEM issue are referred to only a few
selected articles [89] to [107], listed chronologically, that are friendly to the engineering
mentality and that make it possible to follow the interesting development from 1973 ([89],
naive and partially incorrect approach) to 1991 ([107], very effective procedure, verified in
ANSYS [65]).
Bricks can be modified in various ways. For example, it is not necessary that the
division of their edges is uniform, brick 96 does not have to have the nodes exactly in the
thirds, just one (boundary) side can be curved and the other (internal) planar, etc. It is
definitely useless to introduce curved sides of bricks inside of a 3D domain. Useful bricks are
those with the shape of a nonagon (fig. 4.20), which are suitable to fill some corner
subdomains of analysed volumes (blocks, cast volumes, concrete dams). Another name of this
type of brick is pentahedron. It is possible to create a similar hierarchy of polynomials like
bricks, each polynomial being of shape
0
( , , ) ( , )
n
k
k
k
p f
(4.1.79)
where
0
( , ) f and
1
( , ) f are polynomials of  n th degree and
2
( , ), , ( , )
n
f f K are
polynomials of degree 1. Polynomial (4.1.79) has generally d members, where
( ) ( ) ( ) 1 2 3 1 d n n n + + + (4.1.80)
Let us consider a nonagon in fig. 4.20 with vertices (0,0,1), (1,0,1), (0,1,1), (0,0,1),
(1,0,1), (0,1,1). Nodal points of this nonagon are for 1 n the vertices, for 2 n the vertices
and middle points on edges and fro 3 n the vertices, points dividing the edges to thirds and
the centroid of the triangular sides. The total number of nodes is for all configurations
specified by expression (4.1.80). The nodal points are sorted according to fig. 4.20 and
numbered 1, 2, , d K . The detailed analysis of the properties of polynomials (4.1.79) was
made by A. Zenisek in [5]. For the sake of clarity, let us present at least the first two cases of
degree n :
1 n , six polynomials assigned to vertices 1 to 6, variables , within the scope of the unit
triangle (0,0), (1,0), (0,1), third variable within the interval 1 1 , fig. 4.20a):
( )( ) ( )
1 1
1 2 2 2
( , , ) 1 1 ( , , ) 1 p p
( ) ( ) ( )
1 1
3 4 2 2
( , , ) 1 ( , , ) 1 1 p p + (4.1.81)
( ) ( )
1 1
5 6 2 2
( , , ) 1 ( , , ) 1 p p + +
2 n , 3 4 3 1 15 d + polynomials assigned to nodes 1 to 15 according to fig. 4.20b), the
same scope of variables , , :
( )( ) ( ) ( ) ( )
1
1 2 2
1 1 2 2 2 1 1 p p
( ) ( ) ( )
1
3 4 2
2 2 1 2 1 p p
( ) ( ) ( ) ( )
1
5 6 2
1 2 2 2 1 1 p p
4.1 Geometric properties of elements
155
( )( ) ( )
2 2
7 8
1 1 1 p p (4.1.82)
( ) ( )( )( )
2
1
9 10 2
1 1 1 2 2 p p + +
( ) ( ) ( ) ( )
1
11 12 2
2 1 1 1 2 2 p p + + +
( ) ( ) ( )
1
13 14 2
2 1 1 2 2 p p + + +
( )( )
15
2 1 1 p +
Nonagons can be continuously connected to bricks simple example can be seen in fig.
4.20d). At the same time, it is possible to eliminate, in advance, the unknown deformation
parameters ( , , ) u v w in the nodes of the nonagon that are not needed for the connection (art.
4.3.). Moreover, the effectiveness of the element in fig. 4.20 b1) can be increased similarly to
the tetrahedron (fig. 4.19b) and brick (fig. 4.19e). Instead of the parameters in the centres of
the edges, rotations in the vertices are introduced. They represent small rotations, i.e. a vector
with three components. Each vertex has thus all the six degrees of freedom and the element
has in total 6 6 36 deformation parameters.
4.1 Geometric properties of elements
156
Figure 4.20:
Nonagon: a) n=1, b) n=2, c) n=3, d) connection of a nonagon to brick 60. Other name of this element is
pentahedron. Case n=2 can be improved with the help of rotational parameters so that all parameters of
deformation are just in vertices, in total 66=36, as explained in art. 4.1.6.2.
4.1 Geometric properties of elements
157
4.1.6.3 Toroid
3D elements include also elements created by rotation of a 2D element around a
certain axis o that is common to the whole problem. If the 2D element were circular, we
would get a torus. The original element is usually a quadrilateral or triangle and, therefore, we
call the element annular (analogy to anulus) or simply toroidal. With reference to geometry
and base functions, everything told about 2D elements in art. 4.1.5 holds. The difference is
significant in terms of physical nature. Contrary to plane stress or plain strain described in fact
by three components of tensor or , in a toroid we get a general stressstate with all six
components of these tensors. In a special configuration of an axially symmetrical problem in
axes , , r z , where all planes const . form the plane of symmetry, the shear stress
components ,
r z
are eliminated and only four stress components , , ,
r z rz
remain
together with similar strain components , , ,
r z rz
. This differs from 2D elements of
planar problems with , ,
x y xy
and , ,
x y xy
only in the fourth component
and
.
Component
is analogous to component ( )
z x y
v E + in a plane stress
problem with 0
z
. The difference is essential in the physical relation between tensor
and where the impact of incurvation of element fibres into circles of various radii r takes
hold.
4.1.6.4 Special 3D elements
It is natural that various special technical problems initiated development of different
finite elements that are more suitable for the particular task than standard versatile elements.
About 100 of such problemoriented elements were published just in the period between 1965
and 1985, and it is estimated that approximately 1,000 useful modifications of shape and base
functions were not made public. These were made partially according to the procedure given
in art. 4.3 and sometimes following the engineering intuition. The attempts to document all
FEM elements miserably failed already around 1980. Famous Japanese centres admitted that
they could trace about 8% of innovations at most, even when focusing only on literature
searches of topranking publications and company prints. Manuals describe such elements
only from the users perspective and if any theoretical manuals exist, they are usually not
detailed enough to analyse the element in question (and possibly integrate into ones own
system). The user must rely on what is called numerical tests that are already about 20 years
centrally gathered and analysed e.g. in NAFEMS publications. Authors tests are not
accepted. The tests are made by independent professionals and the comparison of results is
made for similar conditions (number of unknowns, computation time). The results are
commonly discussed at specialised conferences, which proved useful and which represents a
noeffect advertisement. The conclusions may be surprising and procedures and elements
promoted for log time sometimes lose their glory. Around 1980 1985 this happened to
hybrid elements when examples were presented in which the welltried elements for thick
shells (semiloof, Reissner), etc., failed. On the other hand, some procedures for the selection
4.1 Geometric properties of elements
158
of base functions not based on any mathematical derivation but following utterly from an
engineering intuition or a lucky idea proved their worth This includes in particular the well
known serendipity family [2].
The name has its origins in the late 18
th
century. It is an artificial word created by
Horace Walpole (17171797), 4
th
Earl of Orford, Member of the United Kingdom Parliament
from 1741 to 1768. In his fairy tale he endowed the three princes of Serendip with the ability
(serendipity) to make discoveries by accident and sagacity. Nowadays, this is considered a
significant attribute of creative thinking and is supported with e.g. the discoveries of Xrays,
radioactivity, radioastronomy (quasars, pulsars), etc. With regard to FEM, many elements and
procedures developed from seemingly unimportant and secondary reasoning, just thanks to
the fact that someone noticed that it could be utilised usefully in a different way, more
generally, that something could be generalised from 1D and 2D and 3D, etc. This human
ability is nowadays distinctly limited by powerful computers that induce a feeling that such
typically human, almost emotional, processes of development of knowledge are not
appropriate for the end of 2
nd
millennium, that they cannot be correct, that they cannot be
proved, etc. This is, however, a false belief. It is a very desirable part of modelling in
technical practice. If we carried out a survey into the history of FEM development starting
from 1956, we could reveal that a vast majority of such suddenly appearing lucky ideas were
later proved by mathematicians to be entirely correct.
With regard to what is termed serendipity family, it represents an attempt to extend the
polynomials used in 2D problems in art. 4.1.5.4. and bricks in art. 4.1.6.2. to general degree n
through an analytical transformation from 1D problems (beams) to 2D problems (plates,
walls, shells) and to 3D problems (volumes in Euclidian space). Only after a few years of
practical application, the mathematical analysis was provided by A. Zenisek in [5]. He proved
that in 3D problems the serendipity family is defined only for 4 n . This means the final stop
for inventing of new members of this family of elements.
4.1.6.5 Solid elements recommended by the authors
3D isoparametric elements with rotational degrees of freedom (Serendipity family)
Every vertex of an element has 6 degrees of freedom: [ , , , , , ]
x y z
u v w . Elements do
not have nodal point in the centre of sides, edges can be straight only. Elements are defined
by geometry of nodes and isotropic material. In case of 6nodal point element and 8nodal 
point is possible to use material orthotropy, that is used in contact elements.
4.1 Geometric properties of elements
159
Figure 4.21:
20nodal point original 60 DOF brick without rotational degrees of freedom.
8 nodal point new 48 DOF brick with rotational degrees of freedom.
Is used numeric gauss integration. Every element has relevant number of integral
points, while because of the fastness of calculation is used reduced integration. This results in
origination of zero energy states, which are treated by additional resistances. We distinguish 2
types: equal rotations and hourglass. For every element s area is used analogical technique of
additional resistance introduction into resistant matrix of an element so as for 2D membrane
elements.
For all isoparametrical elements is used equal principle of transformation of nodal 
point deformations in centres of sides into vertices so as it is in the case of 2dimensional
elements, but it is completed by the third dimension. Movements in centres of the sides are
eliminated by mediation:
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
1
2 8 8
1
2 8 8
1
2 8 8
j i j i
k i j zj zi yj yi
j i j i
k i j xj xi zj zi
j i j i
k i j yj yi xj xi
y y z z
u u u
z z x x
v v v
x x y y
w w w
+ + +
+ + +
+ + +
Space elements with smaller number of vertices than 8 are solved in 2 ways:
36. by degeneration of 8nodal point element, fig. 4.22,
37. by defining of own shape functions and by position of integral points.
Both techniques are implemented, for a user is available only the technique No.2. Hereinafter
is in detail described only this technique.
8nodalpoint element  brick is derived from quadratic izoparametric element with 20 nodal
points. Such an element allows using curved edges, its disadvantage is large width of a half 
belt and consequently large consumption of machine time. By above mentioned elimination
of nodalpoints in centres of sides of 20nodalpoint brick was achieved equations system
4.1 Geometric properties of elements
160
with tighter halfbend, what lead to acceleration of calculation. From the original stiffness
matrix of the 6060 element will arise a matrix of the 4848 size.
Figure 4.22:
Scheme of brick degeneration.
Stiffness matrix
Hereinafter presented shape functions N create together with matrix of differential operators
of 6 geometrical equations:
L
N u B u =
while:
0 0 0
0 0 0
0 0 0
T
x y z
y x z
z y x
1
1
1
1
1
1
1
1
]
Transformation of derivations of nondimensional coordinates to Cartesian coordinates is:
4.1 Geometric properties of elements
161
x y z
x
x y z
y
x y z
z
1
1
1
1
' ; ' ;
1
1
1
1
]
Inverse relation of solution of equations:
y z z y z y y z y z z y
x
1 z x x z x z z x z x x z
y J
x y y x y x x y x y y x
z
1
1
1
1
' ;
1
1
1
]
' ;
Where J is determinant of Jacobi matrix of transformation:
( , , )
x y z x z y y z x
J J
y x z z x y z y x
+
+
Evaluation of Jacobean in any point:
1 1 1
M M M
j j j
i k i k k i
i j k
i j k
j j j
i k k i k i
N N N
N N N N N N
J x y z
N N N
N N N N N N
+
_
+
,
Loading vector
Vector of volume forces:
v
V
dV
f Nb
Vector of initial strain of temperature has in case of 3D shape:
0
0 0 0
T
x y z
T 1
]
0
T
e L
V
dV
f B C
4.1 Geometric properties of elements
162
Geometric matrix
Linear part of the deformation matrix
L
B , is assemble on base of matrix of differential
operators , introduced at previous text. Nonlinear part
NL
B , necessary for solving of
stability (or for calculation of tangent matrix) is of shape:
1 2
1 2
1 2
0 0
0 0
0 0
NL
N N
x x
N N
y y
N N
z z
1
1
1
1
1
1
1
1
]
B
K
%
K
K
Cauchys matrix of stresses:
x xy xz
xy y yz
xz yz z
1
1
1
1
]
%
On base of presented assumptions we are able to assemble geometric matrix
G
K , with which
we could approach to solving of eigennumbers:
det
L G
K K 0
Mass matrix
This matrix is not diagonal in most of cases. However, it is symmetric, positively defined. We
convert it to diagonal one to mass vector by thereinafter described algorithm. We adjust mass
matrix in general:
T
M
V
dV
K N N
to shape of vector according to algorithm:
1/ we calculate
( , )
1
for 1, ,
N
i M i j
j
M K i N
K
2/ set up
( , )
0 for
M i j
K i j
( , )
for 1, ,
M i i i
K M i N K
4.1 Geometric properties of elements
163
Stress
Results  stresses we will obtain directly in integrating Gauss points. For each element we
will reach field of stresses:
1 1 1 1 1 1
[ , , , , , , , , , , , , , , , , , ]
g x xN y yN z zN xy xyN yz yzN zx zxN
K K K K K K
N is number of integration points. Each element has specific number of integration points
and their position in dimensionless coordination system , , .
Taking into account evaluation of results in 3D, it is more convienent to provide user
results directly in vetices
v
, what we solve with their extrapolation to vertices of brick
tranship of virtual hyperbolic function F that is different for each type of element. We will
reach transformation matrix
Numeric integration:
Selective, reduced (6point), 5 for , ,
x y z
, 1 for , ,
xy yz zx
integration points and weight multiplier:
[ ]
[ ]
[ ]
[ ]
1 6 1 2 1 6 1 6 1 4 1 4
1 6 1 6 1 2 1 6 1 4 1 4
1 6 1 6 1 6 1 2 1 4 1 4
9 20 9 20 9 20 9 20 4 5 1 w
Evaluation:
Function of surface for extrapolation of qualities of integration points to vertices:
1 2 3 4
F a a a a + + +
4.1 Geometric properties of elements
165
Pyramid element
Number of nodes
10
Unit coordinates of nodes:
{ 1, 1, 0}, {1, 1, 0}, {1,1, 0}, { 1,1, 0}, {0, 0,1}
Shape of the function
( ) ( )( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
( )
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
( )( )( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
1
1 4
1
2 4
1
3 4
1
4 4
5
6
7
8
9
1
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
2 1
1 1 1 2 1
1 1 1 2 1
1 1 1 2 1
1 1 1 2 1
N
N
N
N
N
N
N
N
N
N
+
+
+ + + +
+ +
+
+ +
+ +
+
( ) ( ) ( )
( ) ( ) ( )
( )( ) ( )
( ) ( ) ( )
0
11
12
13
1 1 1
1 1 1
1 1 1
1 1 1
N
N
N
+
+ +
+
Numeric integration
reduced 8  point 2 2 2
integration of points and weight multiplier :
1 2
3 4
5 6
0, 50661630334978769 0, 26318405556971380
0,12251482265544100 0, 54415184401122496
0, 23254745125350801 0,10078588207982500
[ ]
[ ]
[ ]
[ ]
1 1 1 1 2 2 2 2
1 1 1 1 2 2 2 2
3 3 3 3 4 4 4 4
5 5 5 5 6 6 6 6
, , , , , , ,
, , , , , , ,
, , , , , , ,
, , , , , , , w
Evaluation:
Function of surface for extrapolation of qualities of integration points to vertices:
4.1 Geometric properties of elements
166
1 2 3 4 5 6 7 8
F a a a a a a a a + + + + + + +
Triangular prism (wedge)
Number of nodes
15
Unit coordinates of nodes:
{0, 0, 1}, {1, 0, 1}, {0,1, 1}, {0, 0,1}, {1, 0,1}, {0,1,1}
Shape of the function
( )( ) ( ) ( )( )
( ) ( ) ( )( )
( ) ( ) ( )
( )( )( ) ( )( )
( )( ) ( )( )
( )( ) ( )
( ) ( ) ( )
1
1 9 2
1
2 10 2
1
3 11 2
1
4 12 2
2
1
5 13 2
2
1
6 14 2
2
7 15
8
1 1 2 2 2 1 1
1 2 2 2 1 1
1 2 2 2 1
1 1 2 2 2 1 1
1 2 2 1 1
1 2 2 1
2 1 1 1
N N
N N
N N
N N
N N
N N
N N
N
+
+
+ + +
+ +
+ +
( ) 2 1
Numeric integration
reduced 6  point 3 2
integration points and weight multiplier :
[ ]
[ ]
[ ]
1 6, 2 3, 1 6, 1 6, 2 3, 1 6
1 6, 1 6, 2 3, 1 6, 1 6, 2 3
1 3, 1 3, 1 3, 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3, 1 3, 1 3 w
1
]
Evaluation:
Function of surface for extrapolation of qualities of integration points to vertices:
Function of extrapolation is obtained by combination of triangle and line element:
( )
1 2 3
b b b + + , ( )
4 5
b b + , their multiplying and adjusting to shape:
1 2 3 4 5 6
F a a a a a a + + + + +
4.1 Geometric properties of elements
167
8nodal brick
Number of nodes
20
Unit coordinates of nodes:
{ 1, 1, 1}, {1, 1, 1}, {1,1, 1}, { 1,1, 1}, { 1, 1,1}, {1, 1,1}, {1,1,1}, { 1,1,1}
Shape of the function
( )( )( ) ( )
( )( )( ) ( )
( ) ( )( ) ( )
( ) ( )( ) ( )
( )( )( )( )
( ) ( ) ( ) ( )
( ) ( )( )( )
( )( )( )( )
1
1 8
1
2 8
1
3 8
1
4 8
1
5 8
1
6 8
1
7 8
1
8 8
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
1 1 1 2
N
N
N
N
N
N
N
N
+
+ + +
+ +
+ +
+ + +
+ + + + +
+ + + +
( )( )( )
( )( )( )
( )( ) ( )
( ) ( ) ( )
( )( ) ( )
( )( )( )
2
1
9 4
2
1
10 4
2
1
11 4
2
1
12 4
2
1
13 4
2
1
14 4
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
N
N
N
N
N
N
+
+
+
+ +
( )( )( )
( ) ( ) ( )
( ) ( ) ( )
( )( )( )
( )( ) ( )
( )( )( )
2
1
15 4
2
1
16 4
2
1
17 4
2
1
18 4
2
1
19 4
2
1
20 4
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
N
N
N
N
N
N
+ +
+
+
+ +
+
Numeric integration
reduced 8  point reduced (2 2 2)
integration points and weight multiplier:
[ ]
1 3, 1 3, 1 3, 1 3, 1 3 , 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3, 1 3 , 1 3, 1 3, 1 3
1 3, 1 3, 1 3, 1 3, 1 3, 1 3, 1 3, 1 3
1,1,1,1,1,1,1,1 w
1
]
1
]
1
]
4.1 Geometric properties of elements
168
Evaluation:
Function of surface for extrapolation of qualities from integration points to vertices:
1 2 3 4 5 6 7 8
F a a a a a a a a + + + + + + +
Examples
For comparison we have selected following examples.
Example No. 1
Plate Fig. E4.2, 6 4 m a b , thickness 0, 2 m h , 20 GPa E , 0,15 ,
3
2300 kg m , supported along entire circumference fixing. We have solved own
frequency and own shapes.
Figure E4.2:
Plate fixed along circumference.
dividing
1
f [Hz]
2
f [Hz]
3
f [Hz]
6 4 1 54.10 90.68 162.7
12 8 1 46.98 72.53 116.9
60 40 2 45.61 70.05 110.1
analytically 45.974 71.145 112.58
Results.
4.1 Geometric properties of elements
169
Figure E4.3:
Dynamics  1
st
own shape of oscillation.
Figure E4.4:
Dynamics  2
nd
own shape of oscillation.
Figure E4.5:
Dynamics  3
rd
own shape of oscillation.
4.1 Geometric properties of elements
170
Example No. 2
Plate Fig. E4.6, 9 3 m a b , thickness 0,1m h , 32, 5 GPa E , 0, 0 ,
supported along entire circumference articulated. Uniform membrane loading
2
3, 33 MNm
n
q , resultant 1MN N . We have solved critical loading and own shapes.
Wall buckling in square valves  this shape represents minimal resistance against buckling.
Figure E4.6:
Plate along circumference fulcrumed, axially loaded.
dividing
cr
N [MN]
18 6 2 42.87
45 15 2 35.50
90 30 2 34.71
analytically 33.01
Results.
Figure E4.7:
Stability  1. own shape of buckling.
4.1 Geometric properties of elements
171
Example No. 3
Plate of previous example  fig. E4.8. Loading uniform shear
2
100 kNm
t
q at
circumference. We have solved critical loading and own shapes, fig. E4.9.
Figure E4.8:
Plate by circumference fulcrumed, loaded by shear flow.
dividing
cr
Q [MN.m
1
]
18 6 2 24.26
45 15 2 17.76
analytically 17.46
Results.
Figure E4.9:
Stability 1
st
own shape of buckling.
4.1 Geometric properties of elements
172
Example No.4
Rotation of the element as rigid body, fig. E4.10. Theory of large rotations. Brick
supported by hinge in 1 node, loaded with force in oposite node, solved by geometric
nonlinearity according to NewtonRaphson.
Figure E4.10:
Geometry and deformed shape.
CONTACTS
A connection of two rigid plates (2D elements) between which the relevant physical relations
hold true is called a contact. A contact allows for a small mutual displacement of the two
elements.
Basic assumption is, that during increases of load will not come to change of contact space of
2 elements, which have a contact among themselves. From that follows necessity of
knowledge of a contact region before calculation. By this assumption is limited number of
tasks, which are possible to solve. In the next stage of development can come to further
development in this issue on the basis of user response.
The limitation of the scope of the problem follows from the introduction of an element that is
geometrically bound to two contact surfaces. The displacement (translation) in the plane of
the contact must be small. No such restriction is necessary for the normal direction.
The stressstrain diagram in the normal direction can be:
 full action in both tension and compression,
4.1 Geometric properties of elements
173
 elimination of tension,
 elimination of compression.
The most frequent is the second situation when two rigid elements are in mutual contact and
can be separated from each other. The stiffness in the normal direction is taken from stiffness
of the basic material of the contact plates.
The stressstrain diagram in the tangential direction can be:
 full action in shear,
 elastic Coulomb friction,
 elastic friction with a limited maximum shear
s
,
s n
 max. allowed stress in friction
n
normal stress on the contact plane
frictional coefficient
 rigid friction.
The most common is the second situation when the shearing failure in the contact surfaces
occurs only after the ultimate shear force, which depends on the pressing force and friction
coefficient, has been exceeded.
Process of defining the contact is necessary except of geometry of contact areas, to choose
adequate physical law of contact. This one is defined especially for normal treatment on
contact plane and tangentially in its plane (friction). Normal influence does not cause bigger
problems in calculation. On the contrary, by introduction of friction conditions in plane of the
contact and mainly, which is dependent on normal force to the surface (Coloumb) mostly in
the course of calculation causes problems with convergence, whole calculation can be
markedly slowed down..
Elimination of tension Elastic Coulomb friction
4.2 Physical properties of elementse
174
4.2 Physical properties of elementse
4.2.1 Physical models of materials of elements
An article dealing with physical properties of elements is not, strictly speaking,
relevant to the FEM issue, as FEM does not apply any special physical laws, i.e. relations
between static quantities (internal forces, components of stress tensor) and geometric
quantities (components of displacement or rotation and their derivatives, components of strain
tensor). These laws must be observed in any method: analytical and numerical, classical and
modern. With regard to the variety of materials used in analysed buildings, structures and
their environment or subsoil, there exists a separate and wideranging scientific and
technological field based on an experimental basis. If a physical law is to be used in
technical applications, it is not enough just to philosophise them out. It is essential to obtain
for them reliable parameters as input data for calculations. Otherwise, any prognosis of
behaviour of an object has no chance to find a purposeful application. FEM assumes that a
reliable research of characteristics of all materials present in the analysed problem has been
carried out and that stressstrain diagrams are available and valid for all configurations
(states)) predictable in the structure (assembly, service, limit). These states are more analysed
by means of constitutive laws, which can contain also the process of gradual loading and
unloading, i.e. the path of the load or components of vector u a tensors , at time t . It is
suitable to divide the problems analysed by means of FEM into two fundamentally different
groups:
a) Finite problems, in which it is not necessary to take into account the way how the
analysed state of stress or deformation originated, i.e. the history of the analysed state. The
most typical example is the analysis of an elastic system. In general, such a system is defined
by uniquely corresponding (i) tensor stress field and (i) tensor strain field , without the
necessity to know what was leading to the current state. For example, it is not important
whether some components of were greater or lower before, etc. This general definition
includes, as a special case, also the educational theorem saying that after unloading an elastic
body returns to the original state. Naturally, it returns to this original state after loading too,
on condition that it was subjected to lower load before, and if we consider the original state
without stress and strain ) 0 ( . Then, if all external loads disappear, the body returns to
the original state as well. If there was some initial stress
0
, the body returns to this stress. It
may seem superfluous to speak of returns, as the history is completely irrelevant and it is
sufficient to analyse just a single particular state of the elastic body.
In simpler situations the relation between components of and components of is linear,
which can be written according to no. 8 and 9 tab., art. 2.3 and art. 3.3.2, formula. (3.3.15) 
(3.3.17) preferably in matrix form to be as concise as possible:
C = (4.2.1)
with square matrix of physical constants C of dimension ( , ) s s , if s is the number of
components of tensors and . According to Glossary, the dimension in ordinary elastic
4.2 Physical properties of elementse
175
problems is 3 s to 6. The same article contains examples where the matrix vector is
composed of what is termed internal forces of 1D or 3D element. They represent in fact
integrals of stress components and their moment over the crosssection of a beam or over the
thickness of a slab, wall or shell, see the following art. 4.2.2. to 4. Most programs assume that
for one element C is a matrix of constants, which means that nonhomogenous shapes must
be divided to so small elements so that the change of C could be modelled by a steplike
diagram. On the other hand, there are programs that allow for modelling of changes of C
inside the element, usually by means of a polynomial of low degree that is defined
unequivocally by nodal values of C, similarly to shape and base functions (briefly physical
functions). For example, the very useful element SHELL43ANSYS of a Mindlin plate is a
general quadrilateral, in the vertices of which different thickness
1
h to
4
h can be specified,
which represents a physical term (not geometrical) for 2D space, as there is no geometrical
dimension in the direction of the normal. Function ( , ) h x y is inside the element a bilinear
polynomial of type (4.1.55). Generally, instead of (4.2.1) we can write:
( ) ( ) ( ) x C x x = (4.2.2)
where x x for 1D elements, ( , ) x y x for 2D elements and ( , , ) x y z x for 3D elements,
alternatively with dimensionless coordinates ; , ; , , or cylindrical ones , , r z etc. It
is typical for FEM that the variability of C does not lead to any significant numerical
difficulties. According to the table in art. 2.3., no. 10, stress matrix is simply varying as
well, and if the integration in no. 11 at stiffness matrix K is carried out exclusively
numerically using the Gauss formula, only the values of C in the integration nodes apply.
The case of a nonlinear law, combining components of with components , is
more complex. This includes in particular laws (3.3.24) and (3.3.26) that can be used to
differentiate tensor and tensor out of the density of potential energy W or
complementary energy , which leads to a linear relation (4.2.1) or (4.2.2), with the later
happening only occasionally if we deal with special quadratic functions W and . Generally
nonlinear relations with a vector function are obtained experimentally from tests with a
controlled strain:
( ) f (4.2.3)
or with a controlled stress:
( ) g (4.2.4)
or the implicit dependence
( , ) F 0 (4.2.5)
The feature of the finite problem i.e. the independence on the history preceding the analysed
state is preserved. It represents a nonlinear elasticity and formula (4.2.5) can be applied to
some elasticplastic substances only for the special configuration of initial monotonous
loading, without the possibility to monitor the behaviour during subsequent unloading.
b) Problems depending on the path include all problems excluding those stated under
(a). A typical example is a viscous substance in which the magnitude of the deformation is
determined among others by the speed of stress alteration over time t , over which the stress
acts. The range of such problems is significantly broader than for (a). The finite problems are
in fact a mere abstraction that is useful for the first (and in practice often the last as well)
4.2 Physical properties of elementse
176
prognosis of the behaviour of a real object. Even a brief classification of problems under (b)
goes beyond the scope of this text. Practically speaking, it always represents a nonlinear
problem. In order to clearly demonstrate how the dependence on the path of the analysed
process over time arises, let us combine what is termed elementary rheological models (fig.
4.23) into systems, in which two or more models are arranged one next to another (parallel
arrangement) or one after another (serial arrangement). Three simple examples are shown in
fig. 4.24 with the scheme of corresponding stressstrain diagrams for force ( ) P t
monotonously increasing over time t attached at the bottom. If ( ) P t has other than
monotonous distribution, i.e. if the system is alternately loaded and unloaded, the behaviour
of rheological models can be very complex. Only systems with up to 3 4 parts have been
analysed in detail so far. An illustrative example of the Kelvin twopart model is in fig. 4.25.
In this text, let us emphasize just the fact following from the graph in fig. 4.25c):
If we know the magnitude of force P that acts on the system at time t , we can
determine neither the deformation, nor the percentage of force P transferred by model H
(spring) and by N (damper). If the problem is to be defined at all, we must know the state of
the system at time 0 t , e.g. the initial state without any stress and deformation, and the
complete history of the magnitude of P up to the analysed time instant, i.e. function ( ) P t . In
the case of such a definition we talk about loading by controlled force, as we assume that a
certain mechanism introduces the force ( ) P t into the system in this way. If the deformation is
defined (elongation or shortening in the cited examples), we talk about controlled
deformation and force P acting on the system is unknown. This force could be read from a
dynamometer connected to jaws of a press used for the test carried out with the model. It
could be simply demonstrated that force P depends on the complete history of the introduced
displacement ( ) u t prior to the instant of measurement.
In particular, fig. 4.25c shows that even such a simple twopart model (spring and
damper) completely fails to comply with the fundamental requirements of finite problems
the independence of the state on the path of load over time. It behaves fundamentally
differently from an elastic body. In order to determine (in our example just two internal
forces ,
el vis
P P ), it is insufficient to input just (in our example just translation u ), similarly
as the pure value u does not determine force P . If we study the model at a certain time
instant t , in which there is a certain displacement u , we cannot determine the force in the
spring and in the damper at all, until we know what has been happening since the initial state.
Such behaviour can be seen quite often e.g. in foundation engineering where the settlement of
a building strongly depends on the history that the subsoil went through not only in geological
terms (history of geostatic stressstate and structural strength) but also in terms of the
construction process (excavations, temporary unloading of the vertical geostatic component)
and operation (alternate loading of adjacent objects, e.g. in a system of tanks for bulk
material). More details can be found in [8, 9, 22].
4.2 Physical properties of elementse
177
Figure 4.23:
Six basic rheological models marked by initials of authors (the last without author) and corresponding stress
strain diagrams for monotonous loading by force P causing elongation u of the model.
4.2 Physical properties of elementse
178
Figure 4.24:
Three examples of complex rheological models and corresponding stressstrain diagrams with a brief wording of
the constitutional law for the behaviour at time t under constant force P.
4.2 Physical properties of elementse
179
Figure 4.25:
The Kelvin model according to fig. 4.24 K for cyclic loading by a constant force P alternated with complete
unloading to the value P = 0. The dependence of the model state (i.e. its deformation and percentage of
transferred force P in parts H and N) on the full history of loading starting from the initial state of the model.
4.2 Physical properties of elementse
180
4.2.2 What is the effect of physical properties in FEM
algorithms
4.2.2.1 3D constitutive laws
To find out the physical properties of materials is the problem for experiments and no
FEM algorithm can substitute this kind of information. On the other hand, it is convenient if
the author of the FEM program and the experimenter were not completely isolated from each
other, as only a certain form of information is useful for a particular FEM methodology and
any other result would have to be transformed in a complicated way. The complex
constitutive laws in the implicit form (4.2.5) are usually implemented only in problem
oriented not versatile programs, e.g. for the prognosis of behaviour of large soil structures
(earth dams, caverns, tunnels). An ordinary user most often comes across the explicit form
(4.2.3) that takes form (4.2.1) for linear problems.
This form can be directly and without problems applied to FEM algorithms for 3D problems,
in which the dimension of matrix and is 6 and symmetrical matrix (6, 6) C has 21 of
elements that may be generally different, see art. Glossary, 2.3. and 3.3.3. The tensor notation
clearly indicates the stress components ( , 1, 2, 3)
ij
i j , strain components ( , 1, 2, 3)
kl
k l
and physical constants
ijkl
c . The summation is made over indices , k l and the signs of these
sums are omitted for the sake of brevity:
ij ijkl kl
c (4.2.6)
Thus, a fourthorder tensor of physical constants c is introduced. It is transformed as a tensor
during the rotation of the coordinate trihedral , , x y z . The matrix notation takes advantage
of the symmetry of tensors and , i.e.
ij ji
,
kl lk
. Six generally different
components are numbered using one index and are sorted into column matrices written, for
the sake of brevity, as transposed rows:
[ ] [ ]
T T
1 2 3 4 5 6
, , , , , , , , , ,
11 22 33 23 31 12
= (4.2.7)
[ ] [ ]
T T
1 2 3 4 5 6
, , , , , , , , , ,
11 22 33 23 31 12
=
If we keep this cyclic order, tensor is in a geometrically linear mechanics derived from
vector field u according to Glossary by means of the rule Gu (the first matrix G in
Glossary). The same rule applies to equilibrium conditions in the transposed form:
T
+ G x 0 . Matrix of physical constants C in the relation C is a square matrix (6,6)
with the elements marked with two subscripts according Glossary:
[ ] , 1, 2, , 6
ik
C i k C K (4.2.8)
Due to the symmetry
ik ki
C C , the maximum number of different constants is 21 for general
material anisotropy. Various types of orthotropy reduce this number and in the case of
isotropy in which the physical properties do not depend on the analysed direction, all the
4.2 Physical properties of elementse
181
constants can be derived from two fundamental characteristics. Usually, modulus of elasticity
E and Poissons coefficient of transverse contraction are used. These two are related to
shear modulus G through the formula
( ) 2 1
E
G
+
(4.2.9)
Optionally, another relation can be applied see e.g. [31, 4246].
Nonlinear problems, firstly, make the relation between strain tensor and displacement
tensor u more complex (i.e. matrix of operators G), and, secondly, the definition of stress
must be done more accurately. There exist voluminous both theoretical and experimental
studies the outcomes of which have been implemented into FEM practically always only in
the simplified form (4.2.1) with constants (4.2.8), e.g. in incremental form in matrix
notation:
d d C = (4.2.10)
or in index notation
ij ijkl kl
d c d (4.2.11)
Constants
ijkl
c have in this for the nature of tangent modulus. They are part of input data for
different levels of stress and deformation, in the form of a function, table or graph. The reason
why the presentday exceptionally developed nonlinear constitutional relations cannot get into
FEM algorithms in a different way is simple. Contemporary computers and their peripherals
can operate with continuous inputs and outputs, the cooperation with analogous devices is not
eliminated, etc., but the only tried and standing core or atom of every algorithm is the linear
matrix algebra, and in FEM in particular the solution of a system of linear algebraic
equations. Without exaggeration, we can say that all FEM algorithms conform to this fact.
Several concessions must be made in nonlinear problems just with regard to the physical
laws, so that we get linear systems, as even the simplest assumptions without these
concessions result in systems of equations of degree 3 to 4, which are totally unusable for
many reasons.
Consequently, an ordinary FEM user comes into contact with the physical properties of
materials in fact only in the form of a certain number of physical constants: at least two
( , ) E and at most the whole set of 21 constants for different levels of stress and deformation
of an anisotropic matter. They can neither input them randomly nor assume that the FEM
program derives them from nothing. They represent irreplaceable input data. Users obtain
them from standards, from their own or ordered experiment, from accepted published sources,
from databanks (e.g. geofund), etc.
In addition to these constants that represent deformation properties of a substance, the
physical constants include also strength characteristics, which are in fact certain limit values
or relations between them. If these limit values are reached, one of the following happens:
failure, plasticizing, change of structure of a multiphase environment, bulk material, soil, etc.
They represent technical limits of elasticity, plasticity, strength, structural stability and, in
general, quantitative transition towards another physical model. In FEM algorithms they often
appear in the solution of limit states, e.g. propagation of cracks in reinforced concrete
structures initiated when the tensile strength of concrete is locally exceeded, which with
increasing load results in continuous areas that all the internal forces are transferred solely by
4.2 Physical properties of elementse
182
the steel reinforcement. After its plasticization, a certain collapse mechanism is created and a
part or the whole structure collapses. Strength characteristics of soil are a typical example
already for the simplest Coulomb law on limit shear stress
nt
, acting in the direction of t on
the surface with normal n :
tg
nt n
c + (4.2.12)
If there is no normal stress
n
, the limit shear stress equals to cohesion c , which for bulk
materials equals to their cohesion, for continuous model it is the tensile strength that
originates under pure tension
nt
in the direction that halves the right angle ( , ) n t . For non
cohesive soil (sand, gravel) 0 c , and what remains is just the friction between individual
grains with the coefficient tg f , with representing the friction angle. This parameter,
however, comes in play even in cohesive soil and, consequently, a FEM program for a
strength problem of this type has two constants: , c . They may have different value for
various construction and service stages. For example, at the beginning we have values ,
u u
c
(subscript u means undrained), when consolidation of subsoil ended we get values ,
ef ef
c
(subscript ef = effective, directly between soil grains).
Deformation and strength constants defined in a real 3D material are theoretically clearly
and unambiguously specified for the needs of FEM algorithms. They can be found
experimentally and are constitutive in nature, i.e. hold the information on the real behaviour
of material in physical processes and this information is not distorted by additional
speculations about the reduction of the dimension of the problem. Only 3D objects exist in
reality and their 3D physical models fully respect this fact.
4.2.2.2 Reduction of the dimension of a problem
Problems that are prevailing in common practice model given objects by means of 1D
and 2D. Therefore, the problem that emerges is what the physical properties of such elements
can be if we know that only 3D bodies are real. A wideranging series of studies have been
focusing on this topic since the very beginning of FEM. In fact, every users manual must
tackle this issue and give instructions concerning the representation of physical properties of
the material of the structure. Consequently, we limit ourselves to the main principles that
usually remain unnoticed in the torrent of otherwise useful information.
In the first place, it is convenient to empathise with the dimension of the element and
consider what can be defined geometrically in this dimension. For 2D elements (plates, walls,
shells) nothing extending into the third dimension can be defined this way, i.e. thickness h of
elements. According to art. 4.1.5. an element is defined on a surface let us call it planar
(surface) with coordinates ,
P P
x y (fig. 4.26a). A spatially curved 2D element is defined in a
similar way on a 2D surface, which adds neither complexity nor simplicity to our problem
the 2D space is just curved (inserted into a 3D space), but is still a 2D space (fig. 4.26b). A
real physical element is a 3D body, filled continuously with mass normals to the middleplane
( , )
P P
x y of the element in such a way that we measure out 2 h in the direction
P
z + and
4.2 Physical properties of elementse
183
P
z , which creates the positive face of the element 2
P
z h + and negative face 2
P
z h .
The physical volume of the element is located in between them. They form the physical
boundary of the element in
P
z direction. Boundaries in
P
x and
P
y directions are defined by
outer normals. If h is constant, we obtain an element of a constant thickness. In general, the
thickness may be variable. This however changes nothing on the core of the following text. A
simple fact is true:
The thickness in a 2D element has a character of a physical constant. It is never used in
any algorithm on its own without being multiplied by some physical constant, e.g.
deformational constant E , gravitational constant g , etc. Only 3D stressstate and strain ,
can exist in a real physical element. In a 2D element we have to introduce some
representatives of it. It follows from the nature of the problem that it must represent the
distribution of , just inside the interval 2 2
P
h z h + , which is not geometrically
present in the 2D space, i.e. it must substitute functions ( )
P
z , ( )
P
z in this interval. As far
as the functions of stress are concerned, the simplest approach in terms of statics is to
work with their resultants or with the resultants of their moments. It is at the same time
necessary in order to satisfy the elementary conditions of equilibrium. This is the way we
define in planar coordinates what is termed membrane (wall) and bending (plate) internal
forces in walls, plates and shells, as stated in Glossary:
x x y y xy xy yx
n dz n dz q dz q
x x y y xy xy yx
m z dz m z dz m z dz m
(4.2.13)
x xz y yz
q dz q dz
4.2 Physical properties of elementse
184
Figure 4.26:
a) Planar 2Delement, b) Curved 2Delement, generally curvilinear planar coordinates, c) 2D internal forces and
their positive direction, d) 1Delement, centroidal coordinates x
C
y
C
z
C
, e) 1D internal forces and their positive
direction in a signconsistent convention, f) end vectors f.
4.2 Physical properties of elementse
185
The integration is carried out from 2 h to 2 h . If no negative sign is introduced, we deal
with a signconsistent definition of internal forces, which features certain advantages in
connection with practical outputs and their processing, determination of the location of the
tensile face under bending, etc. The dimension of quantities n and q is [kN/m], that of m is
[kNm/m=kN]. Technical names: n axial forces (tension +, compression ), q shear forces,
m moments ( ,
x y
m m bending moments,
xy
m torsional moment). The positive direction is
clear from fig. 4.26c. Stress components
z
are neglected, as under common load they have
no technical meaning in comparison with other components.
A more dramatic reduction is made in the case of 1D beam elements. We can operate just
with a straight line segment within the interval 0
C
x L or with a part of an arc 0 s L
for curved elements, which is not a significant difference. The missing two dimensions in
C
y
and
C
z directions cannot be in 1D space considered to be geometrical quantities and they do
not appear in algorithms separately without some physical constant. We use the centroidal
(central) coordinate system
C
x (beam axis), ,
C C
y z (principal centroidal axes) as in fig.
4.26d. All classical (Navier, Bernoulli) and also newer (Mindlin) theories of beams neglect
stress components , ,
y z yz
in comparison with , ,
x xy xz
. For the sake of brevity we omit
in , , x y z the index C of the central axes. The considered components must be represented
through their resultants over what is called a crosssection of the beam, related to its centroid:
the term crosssection
x
A means a 2D area of a 3D beam volume, filled with points that are
located in the normal plane ( )
C C
y z , perpendicular to the
C
x axis. The whole volume of the
beam is an infinitely rigid of its crosssections with boundaries represented by its faces (end
sections) and surface. Similarly, we could construe the physical volume corresponding to a
2D element as an infinitely rigid set of its mass normals, also known as the pin model.
In 1D elements we have to do with a single point, centroid C of the crosssection, into which
the resultants of three considered stress components in this crosssection are concentrated. In
practice, we call them internal beam forces:
( )
x x y xy x z xz x
x xz xy x y x x z x x
N dA Q dA Q dA
M y z dA M zdA M ydA
(4.2.14)
The integration is performed over the whole area
x
A of the crosssection. The formulas
contain just one minus sign at the torsional moment
x
M , which cannot be eliminated
regardless of which coordinate system we have selected. There is no minus sign before the
integral. This signconsistent convention has its practical advantages in the design of beams
subjected to bending (fig. 4.26d). Positive moments ,
y z
M M result in tension at the positive
side of the beam, which is the side with positive ,
C C
y z
coordinates. In the theory of beams it
may be convenient to use the minus sign before the last integral (4.2.14), which results in
vectorconsistent convention. In this convention, the components of vectors
2 2
, P M
acting at
the righthand end of the beam (generally at x l ) have the same sign as the internal forces at
this end. At the other end, components
1 1
, P M have the opposite sign. Besides, also other
conventions, usually traditional in nature, are used in technical practice. An ordinary user can
learn about the convention applied in the purchased program from manuals and quite often it
4.2 Physical properties of elementse
186
is not necessary to know it at all, if the program offers convenient graphical postprocessors
that draw the diagram on the tensile side where the reinforcement must be inserted, etc. It is
however useful to be familiar with the convention if one compares the output of reactions
(vector components acting at bound ends of beams) with internal forces that are not simple
vectors, but doublevectors.
Each section of the beam at point
1
x has two banks, one limits the part
1
x x < before
the section and the other one the part
1
x x > after the section. The doublevector of internal
forces is composed of the vector acting on the part before the section and the vector with the
opposite direction of the components acting on the part after the section. This feature is
caused by the fact that internal forces (4.2.14) are the integral representation of stress
components that are not vectors, but tensors. After a part of the body is released in the section,
the action of the released part manifests itself only through interrupted stress components that
follow the principle of action and reaction on both parts of the body, in the opposite direction
when expressed in the vector form.
For the sake of clarity, let us state in detail the relation between endvectors acting on a 1D
element (fig. 4.26e)
T T
T T
1 2 1 1 1 1 1 1 2 2 2 2 2 2
,
x y z x y z x y z x y z
P P P M M M P P P M M M 1 1
] ]
f f f (4.2.15)
with internal forces at its ends
T T
T T
1 2 1 1 1 1 1 1 2 2 2 2 2 2
,
k x y z xd yd zd x y z xd yd zd
N Q Q M M M N Q Q M M M 1 1
] ]
S S S (4.2.16)
For all conventions we have only changes of the sign, which can be briefly written by using
diagonal matrix (12,12) DIA , which have zeros in all elements except the main (descending)
diagonal. The elements of this diagonal are equal to plus or minus one. Consequently, the
following relation is always true:
k k
f DIA S S DIA f (4.2.17)
In the vectorconsistent convention we have
[ ] DIAG 1, 1, 1, 1, 1, 1; 1, 1, 1, 1, 1, 1 DIA (4.2.18)
Using the convenient signconsistent convention (fig. 4.26e):
[ ] DIAG 1, 1, 1, 1, 1, 1; 1, 1, 1, 1, 1, 1 DIA (4.2.19)
Program ESA uses convention that can be defined by following signs in DIAG:
[ ] DIAG 1, 1, 1, 1, 1, 1; 1, 1, 1, 1, 1, 1 DIA (4.2.20)
4.2.2.3 Description of deformation in a reduced problem
In reality, only 3D volumes exist. Their overall deformation is completely described
by means of three functions , , u v w of three variables , , x y z . From this we can derive the
deformation in details (detailed description of geometrical changes), i.e. elongation,
compression, skewing, volume and shape changes in an arbitrarily small vicinity of every
4.2 Physical properties of elementse
187
point ( , , ) x y z . To be more precise, each vector field
[ ]
T
( , , ) ( , , ) ( , , ) u x y z v x y z w x y z u (4.2.21)
can be assigned a tensor field of strain through some geometrical operation G, which in
mathematical terms represents mainly differentiation
Gu = (4.2.22)
The simplest operation G is in what is termed geometrically linear mechanics, where we
have at most the first partial derivative with respect to , , x y z , in the first power (examples
can be found in Glossary at symbol G where six components of classical deformation tensor
are listed under symbol ). They represent three relative changes in length in x , y ,
z direction and three slopes which represent the change of the originally right angles
between directions ( , ) y z , ( , ) z x and ( , ) x y . Formulas (4.2.24) can be simply derived through
the summation of the contributions of individual displacement components to the total slope.
Formulas (4.2.25) that are very similar apply to the components of small rotation defined
already in art. 3.3.2, in which there are differences of these contributions. In the right handed
system , , x y z , the sign of the contribution is defined by the unified convention of the
positive rotation about the x + axis (direction from y + towards z + ) and similarly around the
y + axis (from z + towards x + ) and about the z + axis (from x + towards y + ). The
components are the average from the rotation of all line segments passing the given point and
they can be roughly interpreted as the rotation of a small vicinity of the point as a rigid unit.
Consequently, in linear mechanics we get nine components of tensor and operation G has
a simple form that can be explicitly written:
x y z
u x v y w z (4.2.23)
1 1
( / / )
2 2
1 1
( / / )
2 2
1 1
( / / )
2 2
yz yz
zx zx
xy xy
w y v z
u z w x
v x u y
+
+
+
(4.2.24)
1
( / / )
2
1
( / / )
2
1
( / / )
2
x yz
y zx
z xy
w y v z
u z w x
v x u y
(4.2.25)
The components of slope are written in theoretical texts with subscripts indicating the plane of
the slope. The quantity can be then transformed as a tensor if the position of the coordinate
trihedral , , x y z changes. Most FEM manuals, however, use technical components of that
are equal to the whole change of the right angle between the original directions, i.e. full slope.
The components of rotation (4.2.25) appear in FEM only from 1975 1980 and are marked
by one index indicating the rotation axis or by two indices indicating the plane of rotation.
4.2 Physical properties of elementse
188
There is no stressstate directly linked to these components as they do not appear in the
physical relation C . A statement by B. M. Irons wellknown coauthor with O. C.
Zienkiewicz saying that any attempt to introduce the rotation into the deformation
parameter is a waste of company money caused that really effective elements with rotations
were developed only 20 years after FEM was born. After 2D elements also excellent 3D
elements were created see art. 4.1.6.2. Now we can exploit a whole set of 1D, 2D and 3D
elements, including their problemfree connection as all the shared nodes have the same
valence. We have six deformation parameters with a clear meaning of displacements
(translations) and rotations, which fact is wholeheartedly welcome in engineering practice.
It is not significant for the following reasoning that in geometrically nonlinear mechanics on
condition of large displacements or deformations formula (4.2.22) can be more complex than
(4.2.23)  (4.2.25), as it contains in commonly used approximation the derivatives in the
quadratic terms and we strictly meet also irrational expressions under square roots. The main
point is that relation (4.2.22) is fully sufficient to find out the tensor of deformation in a 3D
problem with 3D elements.
In the case of 2D elements, it is clear that it is not possible to define vector field u (4.2.21)
directly in the 2D area of an element (e.g. plane ( , )
P P
x y ) also for the reason that the variable
P
z is not positioned in this area. Therefore, it is necessary to start with values ( , )
P P
x y u and
assume a certain analytical extension towards the
P
z direction. This is not typical only of
FEM, but also of classical methods for the solution of 2D systems, walls, plates and shells.
The general starting point is what is termed Reissners sequences [62], i.e. in fact the
expansion into Taylors series in variable z , which can be written separately for each
component , , u v w of vector u and we can even use different number of members in the
series. Using an economical common notation omitting the superscript P for planar
coordinates , , x y z of a 2D element:
0
1
0
( , , )
( , , ) ( , , 0)
1!
( , , )
!
z
m m
m m
z
x u x y z
u x y z u x y
z
z u x y z
Z
m z
_
+ +
,
_
+ +
,
(4.2.26)
The higher the number m of terms, the lower the absolute value of the remainder
1 m
Z
+
on
condition that certain requirements mentioned in the cited texts dealing with the mathematical
analysis are met. Standard FEM algorithms now use only two examples of expansion (4.2.26)
that respect just the first two terms and that differ just in the assumption with regard to the
second term. Manuals usually call them the Kirchhoff and Mindlin 2D elements.
Historically younger and more topical is the Mindlin element, which is used practically in all
presentday packages of FEM programs for both plates and shells. Its theory is based on a
quite inconspicuous article [63] from 1951 that explained the difference between the
behaviour of thin and thick plates subjected to bending vibration through the introduction of
two rotations ,
x y
of the mass normal of the plate which were independent on its deflection
surface w. At the same time, it kept Kirchhoffs assumption that the normal remains straight,
its length ( ) h does not change and on condition of small slopes of the deflection surface it has
everywhere the same displacement component w, so that just the identity remains at this
component in series (4.2.26)
4.2 Physical properties of elementse
189
0
( , , ) ( , , 0) ( , ) w x y z w x y w x y (4.2.27)
In addition, let us neglect the membrane (wall) stressstate and deformation of the element
that is in linear mechanics commonly superimposed to the bending ones and is analysed
conveniently with components , u v in the plane of the 2D element using [43], or with all
details using [31]. Under bending, in the middle plane of a physical element
( , , 0) ( , , 0) 0 u x y v x y (4.2.28)
which eliminates the first term of (4.2.26). The value of the second term is given by the first
derivative u z or v z at point 0 z , which for small slopes represents the inclination
angle of the mass normal around
C
y or
C
x , i.e.
y
or
x
. According to fig. 4.27 we introduce
vector signs for the angles and thus the second terms of (4.2.26) get the following form in a
right handed system:
( , , ) ( , )
( , , ) ( , )
x
u x y z z x y
v x y z z x y
(4.2.29)
4.2 Physical properties of elementse
190
Figure 4.27:
a) Geometric assumption of Mindlin plate, b) Kirchhoff plate, c) Mindlin beam, d) Navier beam.
4.2 Physical properties of elementse
191
This, together with (4.2.27), determines the complete vector field u of the physical element
of a Mindlin plate by means of three functions defined only in its centroidal plane 0 z :
0
( , ) ( , ) ( , )
x y
w x y x y x y (4.2.30)
Notice, please, that the originally right angle between axes ( , )
P P
x z and ( , )
P P
y z changes
after deformation due to sloping
0 0 xy y yz x
w x w y + (4.2.31)
Kirchhoff element implemented in older FEM programs and used as an alternative for plates
and shells of very small thickness even in the presentday programs is based on a radical
assumption that no shear deformations (4.2.31) and thus:
0
0
0
0
xz y
yz x
w x
w y
(4.2.32)
As a result, instead of three functions (4.2.30) just a single function
0
( , ) w x y is sufficient to
determine the complete vector field u in the physical element:
0
0
0
( , , ) ( , )
( , , ) ( , )
( , , ) ( , )
w x y z w x y
u x y z z w x y
v x y z z w x y
(4.2.33)
Modern FEM programs usually employ Mindlin elements and use a special numerical
procedure called briefly shear locking to ensure that the reduction of the element
thickness h does not lead to numerical instability and that the results are close to the
Kirchhoff assumption, which is a theoretically and experimentally verified trend. Individual
programs differ only in the magnitude of the parameter used for the shear locking and serious
manuals state this value. A user who would obtain slightly different results from two well
tried programs, can thus find the explanation and does not lose time by searching for non
existing bugs.
The situation is similar with 1D elements where we today use mainly Mindlin elements
offering the slope and (as an alternative) less classical elements based on the Bernoulli
Navier assumption that are called Navier elements. A brief explanation is in fig. 4.27c,d.
Instead of mass normals h we have in beams (nonwarping) crosssections
x
A . Instead of the
middle plane 0
P
z we have a centroidal axis
C
x of the beam where 0
C C
y z . In terms
of geometry we can consider the Mindlin beam a onedimensional Cosserat continuum in
which each point has all six degrees of freedom both in translation and rotation with regard to
centroidal axes , ,
C C C
x y z . For the sake of brevity, let us now omit superscript C and write
the distribution of the degree of freedom along the x axis in the form of six functions of one
variable x :
( ), ( ), ( ), ( ), ( ), ( )
x y z
u x v x w x x x x (4.2.34)
If vector field (4.2.21) defined through three functions ( , , ) u x y z , ( , , ) v x y z is to be
completely defined by these six functions, we have to make some assumption concerning the
distribution along y and z coordinates, i.e. across crosssection
x
A . Theoretically, it is
4.2 Physical properties of elementse
192
possible to start with the expansions similar to (4.2.26) and define the whole hierarchy of
beam models. Practically, this is necessarily awkward, as technically accurate models can be
obtained through simple assumptions concerning the behaviour of crosssection
x
A . In
principle, there are two possibilities:
kk) Points ( , , ) x y z filling planar crosssection
x
A in the initial state fill after deformation
some general area
*
x
A
, i.e. the crosssection warps, it is no longer planar. The mode
of warping can be prescribed on the basis of theoretical analyses of the behaviour of
real beams. The best known is the prescription of displacement component ( , )
A
u y z
perpendicular to the, generally inclined, plane of
x
A according to Vlasovs rule for
sector areas of what is termed thinwalled beams. This introduces, in addition to
(4.2.34), also another geometrical quantity warping parameter ( ) x .
ll) Crosssection
x
A moves during the deformation of a beam as a rigid planar unit.
Therefore, it has just six degrees of freedom (4.2.34) defined in its centroid C . If there
is no bond between them, it represents a more general instance of a beam let us call
it briefly the Mindlin beam (fig. 4.27c) following the analogy with the Mindlin model
of a plate discussed earlier in the text. In order to get a better overview of the
distribution of , , u v w in the 3D volume of such a beam, it proved useful in technical
practice to introduce two groups of load cases and deformations called axial and
transverse effects. If we select the , , x y z axes in the principal centroidal axes of the
beam and its crosssection , ,
C C C
x y z (and omit the superscript C for the sake of
brevity), these effects are independent on each other, can be analysed separately and
superimposed. The axial effects can be divided into translation (tension or
compression), when each crosssection simply moves in the x direction as a rigid
unit, i.e.
( , , ) ( ) ( , , ) ( , , ) 0 u x y z u x v x y z w x y z (4.2.35)
and rotational (torsion) when the crosssection rotates about the xaxis and,
therefore, for small rotations
x
we have:
( , , ) 0 ( , , ) ( ) ( , , ) ( )
x x
u x y z v x y z z x w x y z y x (4.2.36)
The transverse effects can be divided into effects in the plane ( ) xy when component v
applies as what is termed deflection and component
z
as the rotation of crosssection
x
A . In
geometrically linear reasoning we neglect the effect of the rotation and we consider v to be
relating to all points of the crosssection. We get the following displacement field:
( , , ) ( ) ( , , ) ( ) ( , , ) 0
z
u x y z y x v x y z v x w x y z (4.2.37)
The effects in the plane ( ) xz are characterised by deflection w and rotation of the cross
section
y
with this displacement field:
( , , ) ( ) ( , , ) 0 ( , , ) ( )
y
u x y z z x v x y z w x y z w x (4.2.38)
This defines the whole displacement field using six functions of one variable (4.2.34), i.e. the
reduction of a 3D problem to a 1D problem is carried out. The classical theory of beams
4.2 Physical properties of elementse
193
uses another simplification and defines a beam without the slope (Bernouilli, Navier) that can
be briefly called according to the analogy with plates a Navier beam. The added requirement
is that the plane of crosssection
x
A remains even after deformation the normal
(perpendicular) plane to the bent beam axis. The last two functions (4.2.34) can be then
derived from the first function through differentiation with the sign convention for angles as
vectors (fig. 4.27d) respected:
( ) ( ) ( ) ( )
y z
x dw x dx x dv x dx (4.2.39)
The deformation of such a classical beam is described just by four functions of one variable
, , ,
x
u v w . The assumption of the perpendicularity of
x
A to the beam axis can also be
understood in the way that the change of the initially right angles between directions , x y and
, x z is not permitted, i.e. there is no slope
0 0
xy xz
(4.2.40)
This physical assumption is disputable in reality, as it implies zero shear transverse stresses
after multiplication by shear modulus G
0 0
xy xy xz xz
G G (4.2.41)
After the integration over crosssection
x
A this results in zero shear forces:
0 0
y z
Q Q (4.2.42)
which is possible only for load by pure bending without any shear, i.e. constant moments
zd
M or
yd
M with a constant curvature of the beam. This is a very rare situation. Every other
type of load leads to nonzero , , Q . This changes the right angles between the initial
directions , x y and , x z . Formulas (4.2.39) no longer apply and must be extended by the
effect of slope. It is convenient to write it with derivatives of w on the lefthand side and
extend thus the idea about the behaviour of the Mindlin beam:
( ) ( ) ( )
( ) ( ) ( )
y xz
z xy
dw x dx x x
dv x dx x x
+
+
(4.2.43)
A detailed analysis of the corollaries with regard to FEM is contained in [42], chapter 3. For
the purpose of this text, formulas (4.2.34)  (4.2.43) are sufficient for further physical
reasoning.
4.2.2.4 Components of deformation in a reduced problem
If we know the complete displacement field u in a 3D body modelled by means of 2D or 1D
elements, we can derive tensor field of strain without any reduction problems through the
ordinary 3D rule in the simplest linear form Gu (art. 2.3., no. 6 and 7), or using more
complex operations in the case of a nonlinear problem. Then we can determine the tensor
stress field from the constitutional physical relations in the form C (art. 2.3., no. 8) or
using more complex relations (art. 3.3.2), again, without any technical speculations. If we
reduce the dimension of the problem to 2D or 1D, we do not work directly with tensor , but
4.2 Physical properties of elementse
194
with its integral representatives, 2D internal forces (4.2.13) or 1D internal forces (4.2.14), fig.
4.26, which can be summarised in matrix vectors:
[ , ] [ , , ] [ , , , , ]
T T T T
m b m x y xy b x y xy x y
n n q m m m q q s s s s s (4.2.44)
[ ] , , ,
, ,
T T
T
T T T T
a b a xd b bxy bxz
T T
bxy y zd bxz z yd
N M
Q M Q M
1 1
] ]
1 1
] ]
S S S S S S S
S S
(4.2.45)
2D internal forces (4.2.44) were divided to membrane ones (wall) marked with index m and
bending (plate) with index b . 1D internal forces (4.2.45), which are usually written in the
following order
, , , , ,
T
y z xd yd zd
N Q Q M M M 1
]
S (4.2.46)
were divided into axial effects
a
S and transverse effects, which are further divided to bending
and shear in plane ( ) XY and ( ) XZ , i.e.
bxy
S and
bxz
S . This is convenient for physical
relations into which we need to introduce instead of tensor components on which
internal forces s or S do the virtual work. At the same time, the virtual works done by
components of on components of corresponding to the applied reduction of the
dimension from 3D to 2D or 1D equal, which can be formally written by the following
equations:
2D
3D 2D
T T
dxdydz dxdy
s (4.2.47)
1D
3D 1D
T T
dxdydz dx
S (4.2.48)
The derivation of strain components
2D
and
1D
of reduced problems is one of the main
themes in the textbooks of theoretical mechanics. It is often done using various popularising
forms, in which there is a risk of factual mistakes. Taking into account only more recent texts,
we can recommend [2, 4, 31, 42, 43, 61]. The consistent conception of reduced problems in a
way that it is always necessary to know the exact situation in the 3D body whose 2D or 1D
model is analysed was introduced already in classical mechanics by G. Hamel, A. Ndai and
in particular A. and L. Fppl [47]. The importance of this principle has not diminished with
the development of numerical methods and is respected by serious authors. This was often
underestimated in common practice and internal forces s , S and their strain virtual
equivalents
2D
,
1D
were unthinkingly accepted as separately existing and everything
representing quantities without any regard to the fact through which reduction they originated.
No attention is paid to the fact that all real bodies of any dimensions or proportions of these
dimensions are only 3D bodies and that there is no other stressstate than 3D that is defined
by tensors and . On the other hand, this does not prevent us from using resultants,
whenever it is useful, e.g. if we need to balance the tension in a concrete structure with
reinforcement. Users of FEM programs, vast majority of which employs 2D and 1D elements
(i.e. the real 3D dimension is reduced), should be informed in the manual which type of
reduction has been applied and what analysis the given model can be used for, see chapter 2.
Any attempts to use such elements for the solution of typical 3D problems of stress
4.2 Physical properties of elementse
195
concentration, etc., are pointless, as the contemporary FEM already offers both the required,
even though demanding, software with 3D elements and advanced computers.
Let us state here explicitly (according to manuals [42, 43]) the components of deformation
assigned to internal forces s and S for the most frequent FEM problems. For 2D
problems analysed by means of shell elements with internal forces (4.2.44) we can similarly
divide the strain components to membrane ones (wall)
2Dm
and bending ones (plate)
2Db
:
( )
2D 2D 2D 2D 2D 2D 2D
* *
2D
, , ,
, , , ,
T
T T
m b m x y xy
T
b y x y x xz yz
1 1
] ]
1
]
(4.2.49)
Partial derivatives with respect to x and y are marked by a prime and a dot. Let us notice
that the membrane stressstate can use standard components of deformation known from
plane stress problems. The stressstate is constant along thickness h of the element. The
bending stressstate is introduced for each of five internal forces by such a quantity that is
present in the formula for virtual work (4.2.47) on a differential element dxdy . For bending
moments
x
m and
y
m with what we call direction or differentiation or reinforcement
subscripts , x y (the direction of tension or reinforcement at the face of the plate), it represents
mutual inclination of mass normals to the plate in sections , x x dx + and , y y dy + about the
y and x axis. Similarly, for torsional moment
xy
m it is about the x  and y axis, which is
added together and both vector index i and signs of angles
x
and
y
are respected. For
shear forces ,
x y
q q it represents slope that can be expressed by means of three functions
(4.2.30) according to formulas (4.2.31) which means that we have no additional unknown
functions. The described more general Mindlin model of a plate can be for plates of a very
small thickness replaced by the Kirchhoff model according to (4.2.32) (4.2.33) with just
one unknown function
0
( , ) w x y . This, however, results in a physical paradox that shear forces
,
x y
q q are nullified due to zero slopes (4.2.32). Kirchhoff got around this through the
derivation of these forces from the conditions of moment equilibrium of the dxdy element
with regard to the x  and y axis, which caused that they became independent on bending
moments , ,
x y xy
m m m and they cannot be physically derived from the slope. What remains is
the relation between moments and angle increments that are, according to (4.2.32), replaced
by second derivatives of a single function
0
( , ) w x y deflection surface of the plate:
2 2 * 2 2
0 0
* 2 2 2
0 0 0
2
y x
y x
w x w y
w x y w x y w x y
(4.2.50)
Let us notice that rotation components (art. 3.3.2) do not apply, which is in accordance
with physical experience that stress arises only due to pure deformation and is not affected
by the rotation of a small vicinity of the point as a rigid unit, or more precisely by the
average rotation of all directions passing through this point.
In 1D beam problem, the six internal forces (4.2.45) are virtually assigned simple
deformation quantities. Following the order of common notation and output (4.2.46), this is
the following set:
4.2 Physical properties of elementse
196
1D
, , , , ,
T
x xy xz x y z
k k 1
]
(4.2.51)
Internal forces , ,
y z
N Q Q work in the element dx on its relative elongation
x
and slope
,
xy xz
. Internal moments
,
,
xd yd zd
M M M work on relative slope
x
and on curvatures
caused by the differences in the angles of adjacent crosssections ,
y z
. We can employ six
degrees of freedom (4.2.34), which can be, according to art. 4.2.2.3, used to express values
(4.2.51) in the following way:
x xy Q xz Q
x x y y z z
u dx v dx w dx
dx k dx k dx
(4.2.52)
We introduced the negative sign for curvature
z
k , which applies to a technically
convenient definition of bending moments, see art. 4.2.2.2. and the text following formula
(4.2.14). This minus sign disappears in the vectorconsistent convention. Users of FEM
programs do not have to know the convention, on condition that they use exclusively the
graphical output of the bending moments that are drawn at the tensile face of the beam. On
the other hand, in numerical outputs the sign plays an important role, is mentioned in manuals
and numerous users find it on their own through a simple comparison with moment diagrams.
The total (final) diagram of transverse displacement , v w can be put together from the part
resulting from bending ,
M M
v w and the part resulting from shear ,
Q Q
v w (detailed explanation
can be found in [42], chapter 3):
M Q M Q
v v v w w w + + (4.2.53)
At the same time, relation (4.2.43) is valid, which enables us eliminate slope from
components (4.2.52) and deal just with the introduced six degrees of freedom (4.2.34) without
decomposition (4.2.53):
xy z xz y
dv dx dw dx + (4.2.54)
This gives the system of strain components (4.2.51) the form suitable for FEM:
( ) ( )
1D
, , , , ,
T
z y x y z
du dx dv dx dw dx d dx d dx d dx
1
+
]
(4.2.55)
An extreme example is Navier slender beams, in which zero slope can be assumed, see
(4.2.39) (4.2.40). This situation is similar to the Kirchhoff 2D elements, i.e. shear forces
,
y z
Q Q cannot be physically derived from zero , but only from the conditions of moment
equilibrium of element dx . Instead of angles ,
y z
, we can use the derivatives of w, because
y
dw dx ,
z
dv dx . Formulas (4.2.55) get simplified to the following form:
2 2 2 2
1D
, 0, 0, , ,
T
x
du dx d dx d w dx d v dx 1
]
(4.2.56)
used now only in the classical theory of beams, with the effect of transverse shear not taken
into account.
For 1D elements we do not encounter any more the problems related to the components of
4.2 Physical properties of elementse
197
small rotation in 2D and 3D areas. Each point of a 1D element is the model of one cross
section of the beam with six degrees of freedom: three translations , , u v w and three rotations
, ,
x y z
. The question how to deal with them effectively comes to the fore only in
geometrically nonlinear problems (art. 5.1.4) with large rotations that can no longer be treated
as vectors. Some programs apply the conception according to which rotation is treated as a
quaternion (four data: the direction of rotation axis , , and the magnitude of rotation ),
see e.g. Argyriss algorithms in [61, 65]. The incremental approach is more feasible if load
increments are selected so small that they produce only small increments of rotation, with
reference to the previous (reference) configuration of the structure, that may be considered
vectors. However, it is always necessary to properly separate the pure strain of the beam from
its rigid body motion, including its rotation around its centroidal axis.
4.2.2.5 Physical constants of 2D FEM elements
In the previous paragraphs, static and geometric quantities of a reduced 2D problem were
defined. These are: internal forces s (4.2.44) and strain components
2D
(4.2.49), related to
each other through the definition of virtual work (4.2.47), the density of which (the value
related to a unit area) is specified by coefficient
2D
T
s . When a particular problem is to be
analysed, it is necessary to know the physical relation between s and
2D
. For the simplest
configuration this relation can be linear and have the form of generalised Hookes law
C (4.2.1), (4.2.2), (4.2.6), for more complex situations incremental form is suitable
(4.2.10), (4.2.11), etc. In 3D problems these relations are not burdened with any speculation,
they express the physical essence or constitution of the matter hence the name constitutional
in art. 4.2.2.1. The reduction 3D2D brings to the physical relations one geometric quantity
element thickness h existing in the third dimension, that is not contained in the 2D area. It
gets there only during the integration of , which gives rise to internal forces s (4.2.44).
Furthermore, it comes into play when
2D
is created from 3D tensor and 3D vector u. The
aim is always to have the relation that is either (i) linear in all components:
2 2D
s C (4.2.57)
with the matrix of physical constants
2
C (8,8), or (ii) more complex, incremental, but always
if possible explicit, e.g. (4.2.3) etc. Matrix
2
C is marked with subscript 2 corresponding to
the dimension of the problem and is substantially different from matrix C of the nonreduced
3D problem, whose subscript 3 is not used, as it represents the fundamental constitutive
quantities. The difference is that
2
C is not determined just by the physical properties of the
matter, but also by statical and geometrical assumptions that were made during the reduction
of the problem to the 2D dimension. In FEM practice matrix
2
C includes also what is termed
shape orthotropy (models of ribbed or corrugated slabs), change of constitutive matrices C
over thickness h (effect of steel bars in concrete, stratification) and similar effects. There
exist a vast number of published (hundreds) and unpublished matrices
2
C , starting with
resources from 1900 1950 before FEM was developed, as this reasoning is independent on
the method of solution of the structure itself. An older overview can be found in Czech
publication [31], a newer FEMoriented overview is given in [64]. Despite this, it sometimes
4.2 Physical properties of elementse
198
happens that a FEM user who is to design a 2D structure cannot find what they would expect
on the basis of their engineering feeling even in the new manuals [43], [65], etc. Should this
happen, they can create the input data for
2
C themselves.
The starting point is the definitions of internal forces (4.2.13) in which stress components
are substituted by strain components according to undisputed 3D constitutional relations
that must be in explicit form, e.g. (4.2.1), (4.2.2) or (4.2.3). This gives rise to no speculation
as to the reduction of the dimension. This only comes after the physically existing
components are replaced by
2D
of the 2D model. These are generally defined by energetic
(virtual) equivalence (4.2.47). A FEM user practically always opts for the Mindlin model
(4.2.49), occasionally with the Kirchhoff assumption (4.2.50), i.e. they get satisfied with a
rigid mass normal h . The membrane components of the 2D model are thus transferred
unchanged into the whole 3D body as its strain components.
2D
2D
2D
( , , ) ( , )
( , , ) ( , )
( , , ) ( , )
x x
y y
xy xy
x y z x y
x y z x y
x y z x y
(4.2.58)
The consequences of bending components of the 2D model (4.2.49) are more complex. They
can be found from general formulas (4.2.23)  (4.2.25) in which the displacement components
, , u v w are replaced by expressions (4.2.27), (4.2.29). We get:
( )
0
x y xy x y y x
y x xz y
z yz x
u x z v x u y z z z
v y z w x
w y
+ +
+
(4.2.59)
4.2 Physical properties of elementse
199
For the sake of brevity, we omit the marking of variables , , x y z and index 0 at w, which is
the shared deflection of the whole normal h (4.2.27). As a result, we can substitute into
internal forces (4.2.13). If the constitutive 3D relation between and has the form of a
general anisotropy, matrix C (6,6) in the formula C = is full and each stress component
depends on all six strain components, e.g.
11 12 13 14 15 16 x x y z yz zx xy
c c c c c c + + + + + (4.2.60)
After integration of the first formula (4.2.13) we get the following relation between internal
force n
x
and components
2D
(4.2.49) of the strain of the 2D model:
11 2D 12 2D 13
14 15 16 2D
( ) ( ) 0
( ( ))
x y y x
x
yz zx xy y x
c z c z c
n dz
c c c z
1 + + + +
1
+ + + +
1
]
(
(
(4.2.61)
The integration is carried out over the thickness of the physical body of the element
2 2 h z h . After the integration we get the required formula that forms the first row in
matrix (4.2.57):
2
2,11 2D 2,12 2D 2,13 2D 2,14
* *
2,15 2,16 2,17 2,18
( ) ( )
Dx
x x y xy y
x y x xz yz
n C C C C
C C C C
+ + + +
+ + + +
(4.2.62)
where:
2,11 11 2,12 12 2,13 16
2,14 11 2,15 12 2,16 16
2,17 15 2,18 14
C c dz C c dz C c dz
C c z dz C c z dz C c z dz
C c dz C c dz
(4.2.63)
Similarly, we get the 2
nd
to 8
th
line of matrix (4.2.57) from the 2
nd
to 8
th
formula (4.2.13). If
we sort the internal forces (4.2.13) into a matrix vector
, , ,
, , , ,
T T
T T
m b m x y xy
T
b x y xy x y
n n q
m m m q q
1 1
] ]
1
]
s s s s
s
(4.2.64)
in a similar way we sorted the strain components (4.2.49), we can write the whole obtained
physical relation (4.2.57). In general, matrix
2
C is full, but always symmetrical, which means
that we do not have 8 8 64 constants
2ij
C , but only 36 different constants. The material of
the structure can be generally anisotropic, which results in the full integration of membrane
and bending conditions. It is clear already from the dependence of membrane internal force
x
n on all, i.e. also bending, components of strain. Such complex modelling is rare in practice,
even though the functioning of the FEM program itself is not affected. An ordinary user
cannot, due to financial and time constraints, afford to order all the values that form
constitutive matrix C (21 items), almost no data can be traced in data banks, and it can be
considered a success if reliable C are obtained for a much simpler orthotropic configuration.
In the case of orthotropy, matrix C splits into two separate submatrices (3, 3)
n
C for axial
4.2 Physical properties of elementse
200
components and (3, 3)
s
C for shear components. On condition that the axes of orthotropy
coincide with the planar axes of the element,
s
C is diagonal, which reduces the number of
independent physical constants of the 3D matter to 6 3 9 + :
11 12 13 44
12 22 23 55
13 23 33 66
(3,3) 0
(6,6)
0 (3,3)
c c c c 0 0
(3,3) c c c (3,3) 0 c 0
c c c 0 0 c
n
s
n s
1
1
]
1 1
1 1
1 1
1 1
] ]
C
C
C
C C
(4.2.65)
Axial components ,
z z
(reasons and consequences can be found in [31], [43]) are neglected
in walls, plates and shells, as they are considerably lower than , , ,
x y x y
. It is then
sufficient to know 3 physical constants
11 12 22
, , c c c for
n
C and 3 for
s
C , in total 6 constants. It
can be easily proved that this significantly simplifies matrix
2
C too. For example, only the
first two members with
11 12
, c c apply in formulas (4.2.60) and (4.2.61), which means that four
of eight constants (4.2.63) are not employed and what remains are two constants linked to
membrane components
2D 2D
,
x y
and two for bending components ,
y x
. It is now
convenient to add an important note that often escapes the attention of most users of FEM
programs:
Constitutive 3D constants
ij
c after the integrals remain in formulas (4.2.63). Therefore, it is
possible to model even structures with varying physical properties over the thickness, e.g.
members with greater stiffness at one face, layered members (called sandwiches), etc. The
behaviour of corresponding 2D members is then, however, more general than commonly
known. If we relate all internal forces to their geometrically middle plane 0 z , it, in general,
results in the interaction of membrane and bending stressstate. This is not a hindrance in
general shell (platewall) elements, on condition that the FEM program allows for such an
input. However, the situation simplifies if this variation of physical properties is
symmetrical with reference to plane 0 z . This happens for most sandwich plates with both
soft core (more rigid surface layers) and rigid core (symmetrical insulation cover layers). As a
result, both
11
( ) c z and
12
( ) c z in formulas (4.2.63) are symmetrical functions of z , product
11
( ) zc z and
12
( ) zc z are asymmetrical functions and their integral over the interval 2, 2 h h
equals zero. Consequently, another two constants
2,14 2,15
, C C are no longer employed and
membrane forces ,
x y
n n depend only on membrane components of strain
2D 2D
,
x y
. If the
axes of material orthotropy were different from the planar  x and  y axes of the 2D element
(which is a frequent situation, as the axes for automatic mesh generation are assigned to
individual elements by the program according to unified rules independently on the physical
properties), then just the third member of (4.2.62) with membrane slope
xy
would be added,
which does not affect the independence of the membrane conditions with regard to
bending. A similar formula can be actually easily proved even for the next two components
,
y xy
n q and, moreover, it can be shown that also the bending conditions are independent
not affected by the membrane components.
4.2 Physical properties of elementse
201
The just derived independence is implemented into most (almost all) FEM programs in such a
way that the user is required to provide (for the needs of the input) only very simplified
matrix of physical constants of relation (4.2.57) that splits into two independent matrices:
2 2D 2 2D
(3,1) (3,3) (3,1) (5,1) (5,5) (5,1)
m m m b b b
s C s C (4.2.66)
Both matrices are symmetrical and, therefore,
2m
C has just six independent elements that
must be input by the user of the FEM program in a given order:
(1,1), (2, 2), (3, 3), (1, 2), (1, 3), (2, 3) (4.2.67)
In case of isotropic plate made of a material which is defined just by its modulus of elasticity,
Poissons coefficient and thickness:
, , E h (4.2.68)
Then the following constitutive equations are valid for the plane stress:
( ) ( )
( ) ( )
2
2
1
1
x x y
y y y
xy xy
E
E
G
+
+
(4.2.69)
The coefficients of the membrane material stiffness matrix are then as follows:
( )
2
2,11 2,22 2,33
2,12 2,11 2,13 2,23
1
0
C C Eh C Gh
C C C C
(4.2.70)
Matrix
2b
C for bending of the element is a little more complicated. On the grounds of
symmetry, this does not involve 5 5 25 different numbers, but only 15 for a general
anisotropy and 6 3 9 + for the most frequent configuration, in which case we can assume
mutual independence of the elementary bending and shear states. Actually, in that case,
matrix
2
(5, 5)
b
C further splits into two separate symmetrical matrices
2
(3, 3)
bo
C ,
2
(2, 2)
bs
C ,
on condition that we divide the corresponding internal forces
b
s (4.2.64) and strains
2Db
(4.2.49) to bending and shear:
[ , , , , , ]
T
y z xd yd zd
N Q Q M M M S (4.2.71)
2 2D 2 2D bo bo bo bs bs bs
s C s C (4.2.72)
Nine elements of matrices of physical constants must be specified:
11 22 33 44 55 12 13 23 45
, , , , , , , , C C C C C C C C C (4.2.73)
The constitutive equations can be then written in the following expanded form:
11 12 13
*
12 22 23
*
13 23 33
44 45
45 55
0 0
0 0
( ) 0 0
0 0 0
0 0 0
x y
y x
xy y x
x xz
y yz
m C C C
m C C C
m C C C
q C C
q C C
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
] ] ]
(4.2.74)
4.2 Physical properties of elementse
202
The simplest configuration is an isotropic substance that was already analysed in the section
dealing with the membrane stressstate, see (4.2.68) (4.2.70). For this situation, only three
constants are input and they are stored in the first three positions. The program automatically
detects from the zeros stored in the remaining positions that an isotropic matter is dealt with
and calculates all nine constants. Formulas can be obtained through an elementary integration
over the interval 2 2 h z h , similarly as it was shown for the membrane forces. As the
integral definition contains at moments in (4.2.13) multiplication with the z coordinate, the
substitution of stress components that are linear in z produces integrals from the square of z ,
i.e. the order is
3
h . The order for shear forces (constants in z ) is just h . We get:
( )
3 2 3
11 22 33
44 55 12 11
13 23 45
12 12 12
0
C C Eh C Gh
C C Gh C C
C C C
(4.2.75)
For a uniform distribution of shear stress ,
xz yz
over thickness h we have 1 . Using the
analogy with a beam of rectangular crosssection, it is possible to apply parabolic distribution
and shear coefficient 1, 2 . Depending on the circumstances, also different values of
can be used for 2D models of 3D physical walls, plates and shell of various shapes.
None of the physical relations contains components of rotation . This complies with
perceiving a classical continuum (Boltzmann) as a thick set of nonoriented, i.e. non
polarised, points in which it is not possible to detect if they rotate". Each point has only three
relevant degrees of freedom displacements. In a 2D model, each of its points substitutes the
whole mass normal of length h (the thickness of the modelled wall, plate or shell). As it
follows from both the Kirchhoff (no slope) and Mindlin (with slope) hypothesis, the normal
remains straight and its rotation models the rotation of a point of the 2D model. In terms of
geometry we assign one point with additional three degrees of freedom rotational. This is
the case of the polarised (Cosserat) continuum, where it is possible to discover that the point
rotates, as if we equip it with some suitable marks, e.g. like on a globe. Therefore, some
FEM manuals contain information that the Cosserat model is used. This, however, does not
apply in terms of physics. The Cosserat continuum has nine different components of stress
and strain (nonsymmetrical tensors, theorem of reciprocity does not apply) and in addition to
common components , it contains another stressstate due to the mutual rotation of points
termed moment stressstate. In 2D models, for example, moments (4.2.74) are also linked to
the difference of rotations of adjacent points, but physical constants C of this relation are
obtained after the reduction of the 3D dimension to 2D by means of constants of a classical
continuum.
4.2.2.6 Physical constants of 1D FEM elements
The reduction of a 3D problem to a 1D one is much more dramatic and only six
constants on condition that we deal with an ordinary prismatic straight 1D element without
warping crosssection are left to express the physical properties of a 3D body that is
4.2 Physical properties of elementse
203
considered to be a 1D beam. It features six internal forces (4.2.14) and six degrees of freedom
(4.2.34) of point x that models crosssection
x
A and six components of 1D strain (4.2.51),
virtually assigned to the internal forces in such a way that each of them does virtual work only
on its own component. The advantage is that, contrary to formula (4.2.57) for 2D problems,
we now deal with a simpler, diagonal relation in the form
1 1D
S C (4.2.76)
where according to (4.2.14) a (4.2.51) there is:
[ , , , , , ]
T
y z xd yd zd
N Q Q M M M S
(4.2.77)
1D
[ , , , , , ]
T
x xy xz x y z
e k k (4.2.78)
1 1,11 1,22 1,33 1,44 1,55 1,66
DIAG , , , , , C C C C C C 1
]
C (4.2.79)
For 1D elements, we pay attention to only three stress components , ,
x xy xz
and, for these
components, we employ just three rows from the constitutive physical relation C = . An
isotropic material is practically always assumed and we can apply the simplest formulas:
x x xy xy xz xz
E G G (4.2.80)
With regard to axial effects, tension and compression cause no problems and the
integration of the first formula gets
x x x x
N E dA E dA
thus
1,11 x
C EdA
(4.2.81)
which, for the simplest configuration of a constant E over crosssection
x
A
,
is a well known
stiffness quantity
1,11 x
C EA (4.2.82)
Quite often we have two moduli, for example steel
a
E and concrete
b
E , in crosssection part
a
A (not necessarily continuous, e.g. steel reinforcement bars) and
b
A . The integration is
carried out separately for
a
A and
b
A and we get
1,11 a a b b
C E A E A + (4.2.83)
Apparently, even more general configurations, e.g. laminated beams, are not complicated. It is
usually convenient to factor out some constant modulus
0
E and deal with the dimensionless
ratio after the integral:
1,11 0
0
( , )
x
E y z
C E dA
E
((
(4.2.84)
This form is suitable also for physically nonlinear substances with a tangent modulus
4.2 Physical properties of elementse
204
E of the diagram ( )
x x
f .
Torsion as the second axial effect requires the integration of the fourth formula in (4.2.14), in
which the substitution (4.2.80) introduces slopes that originate from relative twisting
x x
d dx . Generally, such a strain is accompanied with warping of the crosssection, i.e.
with formation of displacement components ( , ) u y z , which breaks the assumption that the
crosssection behaves like a rigid unit (under all other five types of loading). If we insisted
that no warping of the crosssection occurs, we would have to introduce residual axial stresses
( , )
x
y z , which is in fact really done in what is termed restrained warping. Consequently,
the seventh internal force bimoment B must be defined, whose resultant is zero, i.e. it
results into a system stress
x
that is in equilibrium in the crosssection. This theory is
applied in what is termed thinwalled beams, which means, in particular, various steel rolled
crosssections, etc. This configuration will be skipped here and the readers are referred to
standard texts dealing with the theory of V. Z. Vlasov, A. Umansky, etc., contained in a
directly applicable form in the textbooks of steel structures and corresponding standards. We
limit ourselves just to what is termed thickwalled (solid) beams of ordinary rectangular,
circular, Tshaped, etc. crosssection. For them we can admit, without any negative impact on
technical accuracy, that free warping occurs under torsion, which means the state when
( , ) 0
x
y z , i.e. no additional axial stresses constrain the beam. Even this simplified problem
cannot be solved elementary, but by means of either the strain method with the unknown
function of warping ( , ) u y z , or force method with what is termed the Prandtl stress function
( , ) y z . It is related to shear components of the stress in the fourth formula in (4.2.14)
through derivative formulas
xy xz
z y (4.2.85)
Therefore, function is a certain potential, similarly to the classical Airy function known in
planar elasticity problems or the Maxwell or Galerkin functions used in spatial problems.
Many years of practice lead to the compilation of tables with results for the most frequent
crosssection. These tables contain both torsional stiffness
1,44
C from formula (4.2.76):
1,44 xd x
M C (4.2.86)
and formulas for extreme shear stress, usually in the form that is similar to the traditional
formula for bending:
max
xd
M W
The selection of values for ordinary crosssections can be found in tabular form in [43], pp.
6365 for isotropic shear modulus G where the following applies
1,44 k
C GI (4.2.87)
where
k
I [m
4
] is purely geometric crosssectional characteristic with the same physical
dimension [m
4
] as moments of inertia ,
x y
I I or polar moment of inertia
p
I .
This is a good place to mention one common mistake committed by users of FEM programs
due to the fact that quantities , , ,
x y p k
I I I I have the same dimension [m
4
]. This mistake results
from the fact that for a circular crosssection (symmetrical hollow crosssection as well) with
4.2 Physical properties of elementse
205
diameter 2 D R we get the following equity:
2 4 4
2 32
k p x
I I r dA R D
(4.2.88)
This equity does not exist for any other crosssection, which means that
k
I is not a polar
moment of inertia and, moreover, is not related to inertia (such as ,
x y
I I ) at all. Values
p
I
contained in numerous tables serve a completely different purpose and must be substituted for
k
I ! This can be actually clearly seen from the oldest approximate formula derived already by
SaintVenant for crosssections close in shape to circle, e.g. polygons:
4
40
k x p
I A I (4.2.89)
For a precise circle this formula produces
( )
4
2
3 4
4
40 2 20
k
R
R
I
R
,
which, applying the notation
3 2
and approximation
2
10 , corresponds with
formula (4.2.88) variation defined by means of exact
2
9,87 , i.e. 1.3%.
Of all the formulas for individual crosssection shapes listed in [42], pp. 6365, let us
mentions here a rectangle with sides h b :
h b 1 1.2 1.5 2 3 Multiplier:
k
I 0.1404 0.1661 0.1957 0.2286 0.2633 0.3333
3
b h
The assumption of SaintVenant torsion is strongly negatively affected for ratio / h b greater
than 3. Often published approximate formula
3
3
k
I b h for slender rectangles is not
recommended. It is better to model such beams using 2D shell elements. The cited publication
[42], pp. 5960, contains the analysis of the accuracy of the GrashofZhuravsky assumption
for transverse shear (see the following text). For slender, flat crosssections, e.g. a shallow
ellipse, the extreme shear stress is burdened with an error that is largest for 0 , when it
even increases towards infinity with the growing shallowness of the ellipse. The error
vanishes for 0.5 . It is therefore recommended to abandon the 1D model for common
crosssections with 0.2 to 0.3 and employ 2D elements.
Bending and shear in ( ) XY plane are in a 1D beam model expressed through the relation
between two internal forces (4.2.45) and two strain components (4.2.51)
1,22 1,66 y xy zd z
Q C M C k (4.2.90)
Both slope and curvature can be expressed by means of formulas (4.2.52), (4.2.54), i.e. the
form (4.2.55) is used instead of (4.2.51) for the strain components, in order to have only six
degrees of freedom (4.2.34) in the FEM algorithm:
( )
1,22 1,66 y z zd z
Q C dv dx M C d dx (4.2.91)
4.2 Physical properties of elementse
206
The procedure to determine physical constants
1,22
C and
1,66
C is the same as for axial effects,
i.e. it is based on the integral definition (4.2.14), where
xy
and
x
are substituted by strain
components
xy
and
x
in their undisputed 3D meaning. Then we can proceed to 2D
components of strain according to (4.2.52), (4.2.54) and calculate the integrals over the area
of crosssection
x
A . A problem is encountered in the case of
xy
what distribution of
( , )
xy
y z should be assumed, as the reduction principle 3D 1D in art. 4.2.2.4. leaves this
issue completely open. It could be e.g. possible to take ( , ) const .
xy xy
y z , which
corresponds in physical terms to ( , )
xy xy xy
y z G . The second formula in (4.2.14) then
gives a very simple result:
1,22
y xy x xy x x xy
x
Q dA G dA GA
C GA
(4.2.92)
Unfortunately, this would break not only the concordance with experimentally proven
distribution of ( , )
xy
y z , but also the fundamental principle such as the reciprocity theorem for
shear stress components
xy yx
at the face of the beam that is obviously without any shear
stress when subjected to transverse axial load. Fortunately enough, already the classical
theory of elasticity found a simple hypothesis (GrashofZhuravsky) that provides sufficient
compliance with all requirements for common solid crosssections, on condition that these are
symmetrical around the bending XY plane, i.e. around the vertical principal centroidal axis
C
Y . The resultant of the stress components ,
xy xz
is
2 2
xn xy xz
+ (4.2.93)
for which we assume that its ray intersects the
C
Y axis in point O that is for certain level of y
shared by all coordinates. The point O is defined by the intersection of the tangent to the
profil contour in the level y . Due to symmetry, both tangents in points ( , ) y z and( , ) y z
intersect in the same point O. This overcomes the indeterminacy of the 1D model and the
subsequent procedure is already straightforward. It is presented in detail in all elementary
textbooks dealing with the theory of elasticity. FEM users should be aware of one
dissimilarity in the recommended formulas:
Less technical texts, university textbooks in particular, simplify the problem even more and in
the energetic equivalence (4.2.47) completely neglect the impact of components
xz
in
comparison with
xy
under bending and shear in the XY plane. Similarly, they neglect the
influence of
xy
in comparison with
xz
for a problem in the XZ plane they simply take
into account only the components that are parallel with the plane of the problem. More
elaborate texts respect both components. In order to provide for a practical application in
FEM inputs, the result is always modified to the form analogous to (4.2.92). Most often, what
is termed crosssectional area
y
A effective in shear is introduced in plane XY , and similarly
z
A in plane XZ :
y x y z x z
A A A A (4.2.94)
4.2 Physical properties of elementse
207
This introduces coefficients
y
and
z
expressing the influence of the distribution of shear
stresses over the crosssection.
Roughly, the rule applies that 1 > , i.e.
y x
A A < ,
z x
A A < , which means that the assumption
that
x y z
A A A stiffens the beam with regard to transverse shear and may be thus
dangerous. For common crosssections we can consider values 1.2 (rectangle), 1.18
(full circle), 1.11 (full hexagon), 2.0 (thinwalled circles  pipes), 2.1 to 2.8
(rolled crosssections no. 8 to 45 under normal bending and shear in the plane of their web).
The traditional form of formula (4.2.92) is modified according to (4.2.94) to the following
form:
1,22 1,33 y z
C GA C GA (4.2.95)
The second physical constant of formula (4.2.91) for bending due to moment
zd
M is simpler
with regard to both objectivity and calculation. The sixth formula of (4.2.14) is used, where
( )
x x z z
E E du dx E d y dx Ey d dx
It directly gives the sought after relation:
( )
2
zd x x x z
M ydA Ey dA d dx
with physical constant
2
1,66 x
C Ey dA
(4.2.96)
If the modulus of elasticity is constant over crosssection
x
A , E can be factored and the
integral contains just the wellknown expression for the moment of inertia of a crosssection
about its principal centroidal z axis:
1,66 z
C EI (4.2.97)
Similarly, for bending in plane XZ , we can obtain the relation between moment
yd
M and
strain component
y
d dx of the 1D model of a beam according to the fifth formula in
(4.2.14):
( )
2
yd x x x y
M zdA Ez dA d dx
with physical constant
2
1,55 x
C Ez dA
(4.2.98)
which is, for constant E over the crosssection, equal to an ordinary product:
1,55 y
C EI (4.2.99)
with the moment of inertia of the crosssection about its principal centroidal y axis.
For a homogenous cross section the pertinent cross section stiffness can be introduced in the
following formula (4.2.100):
4.2 Physical properties of elementse
208
1,11
C
1,22
C
1,33
C
1,44
C
1,55
C
1,66
C
Homogenous
crosssection:
x
EA
y
GA
z
GA
x
GI
y
EI
z
EI
(4.2.100)
For nonhomogenous cross sectio these stiffnesses must be calculated by pertinent numerical
quadrature.
For homogenous cross section the following values must be input:
, , , , , , ,
x y z x y z
A A A I I I E (4.2.101)
and the program calculates the required (2 2 ) G E + . The calculation of , ,
y z x
A A I is an
easy task for users only in the case of very simple crosssections, which are overviewed in
[43]. A manual calculation is not possible for complex configurations. Therefore, it was
completely automated, e.g. by program [94], in which the user defines just the shape of the
crosssection. Coefficient for the effective crosssectional area in shear
y y
A A ,
z z
A A is determined by the program from the exact theoretical formula by means of a
numerical integration. Torsional stiffness
k
I can be calculated by FEM that exploits a
triangular element of T. Moan (1973, Trondheim, Norway) with three stress parameters in a
node. It is a function of torsion F and its partial derivative with respect to y and z , which is
equal to components of shear stress ,
xy xz
, occurring under SaintVenant torsion of a beam
with free warping (unconstrained torsion of solid (thickwalled) beams).
4.2.2.7 Physical constants of toroids
A special status in FEM belongs to rotationally symmetrical shapes divided to toroidal
elements according to art. 4.1.6.3. Even if in the inputs we deal with a 2D area, in which the
given 3D body is generated through the rotation about an axis, we do not use 2D physical
relations of art. 4.2.2.5, but actually the nonreduced 3D relations C = , or more complex
3D relations between and , yet in cylindrical coordinates ( , , ) r z . It is of no advantage
to introduce here internal forces. FEM program can be divided to two groups. Most of them
can be applied only to problems with a rotationally symmetrical load. It is actually the first
term in the expansion of a general load with respect to variable , i.e. into a series with terms
cos n , 0, 1, 2, n K , as cos(0) 1 const . Advanced programs allow for the input of an
arbitrary load whose harmonic analysis with respect to cos n provides a set of loading
components, the effect of which is analysed separately, because it features different stiffness
matrices, and therefore the lefthand sides of the equations Kd f are also different. Most
often, only the first term ( cos ) is used in practice to approximately express the effect of
wind (pressure and suction).
4.2 Physical properties of elementse
209
4.2.2.8 Gas elements
Special type of element that is possible to be used in case of validity of state equation
of ideal gases. They are constructions that are closed, filling if constituted by gas that at the
same time assures synergism. It could to be insulation glass, where among several solid layers
of glass and case of glued foil is present of given production temperature and pressure.
The potential energy of external forces causes a change in the potential energy of the gas that
depends on the change of volume and pressure.
i
p V (4.2.102)
The dependence of the work on these quantities is shown in the diagram.
Figure 4.28
Diagram of isotermic action of an ideal gas.
Let us assume that there is no change of the temperature in the gas let us consider an
isometric action according to Boyle Mariotte law which says that for a fixed mass of ideal gas
at fixed temperature, the product of pressure and volume is a constant.
const . pV (4.2.103)
4.2 Physical properties of elementse
210
Figure E4.11:
Air hall solidly supported along bottom edge, asymmetrically loaded by uniform force loading, geometry,
deformed shape.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
211
5 Modelling of Structures for FEM
Analysis
5.1 Introduction to the Theory and Practice of
Creation of FEM Models
5.1.1 Presentday Approach to Modelling of Structures and
Soil Environment
5.1.1.1 Objects and Terms
When New Generation Computing, An International Journal on Fifth Generation
Computers (University of Tokyo, editor T. Motooka) magazine was established in 1983
with the participation of reputable experts from all developed countries, there was a prognosis
made in the editorial articles that the 5
th
generation of computers would be on commercial
sale in about 19891990. Their expected parameters, based on the experienced theoretical
studies of informatics, arose a considerable optimism as far as the possibilities of modelling
of engineering tasks were concerned, in particular the full automation of this process
liberating the user from demanding considerations. As we know today, the predictions were
only partly met. Moreover, in about 1990 the theory of modelling was further advanced with
active participation of engineers, mathematicians and physiologists and it became clear that
leaving the engineering problems related to the modelling to an uncompromising algorithm
was not desirable because it could potentially put an end to the whole technical progress. This
is actually conditioned by the human ability of invention, which cannot be transformed into
an algorithm. It is even possible to say that mere deduction does not lead to any truly new
piece of knowledge. The outstanding physicist E. Mach [24] was right when he already in
1912 defended the statement that all pieces of theoretical knowledge are acquired only as
useful concentrates of massed experience (empiriocriticism) with the specific human ability to
create abstract terms playing its role.
A user of a FEM program is, first of all, interested in a certain object that exists and
should be evaluated or that does not exist yet and is to be designed. The object we deal with
can be a real object that can be, or will be able to be, perceived, or it may be an abstract
object of some internal spiritual activity, a schedule of ideas, etc. In any case, the object
features four characteristics:
38. It is the subject of the users interest.
39. It may be defined, differentiated from other objects.
40. It represents a whole at a given level of consideration.
41. It is formulated as a concept, either in the existing standards, manuals, literature or it
is created according to the symbolic rule [36] just for the needs of the problem in
question:
5.1 Introduction to the Theory and Practice of Creation of FEM Models
212
( , , , ) P F S I p (5.1.1)
F is the form, i.e. the name of the term, symbol, sentence, etc, S is its contents,
communicable by communication means or by a language, verbally or graphically, I
is the intuitive, noncommunicable contents and p is the assignment to a particular
object O. The identification term assigns the object to a broader class of objects { } O .
The descriptive term expresses the properties of a real object:
mm) c
lassification (qualitative)
nn) topological (typical in a system of FEM elements, comparative, it places the element
into the right position within the system),
oo) metrical (quantitative, the magnitude of lengths and angles of an element, structure,
loading, etc).
What is considered the least dangerous for users are formal terms in which the
contents is 100% communicable, i.e. I O . These are all purely mathematical terms, which
by their contents remain in mathematics and have no physical contents. Computers can work
with such terms only. All engineering considerations must be mathematically modelled
through these terms, otherwise a computer cannot be used. On the other end of the spectrum,
there are intuitive terms, whose contents is 100% noncommunicable ( 0 S ). And
completely out of the spectrum there are vague terms, i.e. an unclear, fashionable, lax
connection of words uttered without any idea and expressing nothing sound even to the
author, 0 S , 0 I . If they occur in manuals or even in textbooks, they are of little use and
represent a waste of time for the user or company. On the other hand, they are invaluable and
indispensable in art where the explanation of the work may be just intuitive, and in politics as
the art of the possible with the necessity to avoid any formulations that are hard to change
after the election and that make coalition agreements difficult to conclude. No belleslettres,
music, picture, statue, etc. or successful politics could be made of just formal terms. If we
rightly eliminate the vague terms from FEM, we cannot do it to the full extent with the
intuitive terms for a simple reason:
We do not understand FEM just as a purely mathematical matter, but we take into
account also the engineering component. If this component used just formal terms (100%
S ), all technical progress would stop and everything would remain at the level communicated
in the presentday resources through the mathematical form. The engineering component
applies in the modelling of structures with the aim to treat them using the FEM
methodology. Already the decision about the dimension of the model 3D, 2D, 1D (art.
5.1.1.2) cannot be unambiguously prescribed by any manual. Depending on the immediate
need in the given stage of a project, the user must use the professionalism and decide
intuitively whether a 1D beam model, a 2D shell model or even, for a detail near a bearing, a
3D model will be used. Similarly, the user must decide about the model for external forces,
i.e. loading. They may feel e.g. that it is not enough to concentrate all the load in nodes using
a primitive rule of a simple beam, although no formal exact mathematical analysis is at hand..
They select one support type of the structure from the types offered by the FEM program:
point or linear, rigid or flexible without exactly analyzing the effect of the flexibility on
stress distribution. They feel that finding the real flexibility for the input data could be
5.1 Introduction to the Theory and Practice of Creation of FEM Models
213
dramatically more complicated task than the analysed problem itself, although they make no
formal analysis of this feeling. FEM program manuals must provide enough room for all
these intuitive considerations. Simply said, in addition to formal terms that are necessary for
the communication with PCs, manuals must use normal terms as well.
Normal terms in technical literature have different ratio : S I , on average about 80% of
S and 20% of I , but e.g. in soil mechanics, in foundation engineering, etc. there is
sometimes even 60% of S and 40% of I , which is typical for piles. Without intuitive
imagination cultivated in expert contacts (reading, conferences, seminars), many resources are
not applicable because they silently assume this imagination. The user must simply know
what is a pile, although e.g. a gravelsand drainage pile is something completely different
(local change of subsoil, stiffening, acceleration of primary consolidation). The best score of
: S I is that of steel structures. Reinforced concrete is considerably much worse off due to the
intuitive imagination about the way the steel reinforcement may take over the tensile stress of
a 3D region if this is already affected by cracks in which there is no tensile stress at all,
whether and how shear can arise there, etc. Obviously, theoretical and experimental studies
have been elaborated dealing with these phenomena, with some simplified conclusions
accessible to a common user mostly in standards, but they often come up with nothing
applicable to the very problem in question. The situation in soil mechanics is the worst due to
its subject matter. Nearly every introduction to a textbook contains a warning that only an
experienced expert with about thirtyyear practice can make a good design of a building
foundation, which is in fact just before their retirement. Literally, it means that sufficient
intuition can be obtained only over such a period of time, because no satisfactory abstraction
and evolution of generalizing experiences into a communicable contents S have taken place
yet.
What is important is the differentiability of terms, i.e. the ability of the FEM user to
recognize at a given level of consideration whether the term relates to the object. The level
may be variably demanding with regard to the differentiability. It gradually grows from a
generally technical area to a design, production and calculation area (absolute in the contact
with a PC). It may be generally factual, more complex and arearelated, objective in the given
year 1997 and ultimately absolute in the mathematics. Roughly speaking, the differentiability
has lower limit d and upper limit h with possible density of occurrence v in between
ultimately a continuum in the interval ( , ) f d h , e.g. in mathematics all real numbers between
0 and 1. Normally, the assignment may be much weaker, e.g. only a few instances may be
included.. For example, the term typified prefabricated beam K67 and I73 for bridge
structures included only four possible spans 21, 24, 27 and 30 m l , which specified all its
dimensions, reinforcement etc. according to typified drawings. The term large diameter pile
according to CSN 73 1002 (1982) included all cases of diameter d larger than 0.6 m and
lengths D greater than 4d , without any limitation from the top. The differentiability may be
naturally different individual for each person. In a group of experts it increases to team
differentiability and if the team is extended by other experts it may grow to details with sharp
boundaries. There exists a certain effective differentiability that corresponds economically to
the most advantageous division of work. The active age of one man is approximately 15 000
md (= mandays) with 30 000 working seconds per one manday. The amount of information
received in each second is about 2 b (bits) with the exception of at most 6 bits over continuous
30 minutes, which has no influence on the total amount of the received information:
5.1 Introduction to the Theory and Practice of Creation of FEM Models
214
15000 2 30000 900000000 b 900 Mb 125 MB
Up to 90% of this amount is sooner or later forgotten, which means that even the most
experienced expert has not too much information in his operating memory (brain) compare
it with a disk of a common PC. In addition to a team work it is obviously necessary to engage
a kind of an external memory, databank, Internet, etc. Only the most concentrated
knowledge, i.e. the theory, should be left to the brain. An expert that is able to quote by
heart the contents of one standard, but who knows nothing about the purpose of the principle
of virtual works (which is extraordinarily useful in both linear and nonlinear mechanics)
waists his internal memory for useless things that can be easily obtained from external
memories (standards). FEM programs relieved us of the drudgery of calculations, but they are
simply not capable of performing typically human considerations. Also the following
characterization of term P created for object O is related to human qualities:
It is the objectivity of the term that can be evaluated by measure
0
M
0
0
n
M
m
(5.1.2)
where m in the denominator is equal to the number of all subjects (persons, teams) that have
created term P (5.1.1) for the given object O. The numerator
0
n is the number of subjects
used for the creation of terms P with identical communicable parts S . To be exact, formal
mathematical terms have
0
1 M , vague terms have the tendency towards
0
0 M , common
technical terms at various levels have
0
M in the interval (0,1) . Let us take an example from
soil mechanics: participants of advanced training for structural engineers created for a specific
foundation a term with measure
0
0.6 M up to 0.8. The lower ratio appears traditionally in a
pile foundation, the higher one in a raft foundation. The measure
0
M was lower in a group of
newly graduated participants who studied the subject, approximately 0.5. Generally, it was
proved that the objectivity of terms is greater for steel structures and is high in the field of
modelling for FEM calculations without the interaction with subsoil.
The given example of the creation of a term to a particular object is a deductive
process. The opposite is the inductive process, when we are looking for the object for a term
created by abstraction. The engineering activity belongs to constructive processes, when we
create artificially an object for our terms. It is connected with the rate of completeness
c
M
of the delimitation of the term
c
c
n
M
m
(5.1.3)
The amount of all subjects creating object O on the basis of term P is marked m.
The amount of all identically created objects is
c
n . It denotes the identity on the given
differentiability ( , , ) d h , explained above. We take into account only the communicable
part S of term P that was transferred to all m subjects by the primary subject that created
term P for object O. Insubstantial details may differ in the intuitive part. In technical
practice, the rate of completeness
c
M varies within quite broad limits according to the
character of the subject. The average value is about 0.8, which means that the number of
substantially different objects is roughly 0.2 (20%). The effort of the authors of standards is to
5.1 Introduction to the Theory and Practice of Creation of FEM Models
215
increase the rate
c
M . It is traditionally low in the field of foundation engineering which has
been already mentioned (raft foundations, deep foundations, pile foundations, micropiles,
etc.). It is not a Czech speciality, it also occurs on an international scale. E.g. EUROCODE 7
does not contain any chapter about common largesize foundation slabs, it is necessary to
check them according to the principles laid in the chapter Spread Foundation, which covers
rather pad foundations, strip foundations or grillage foundations. Users of FEM programs who
want to analyse foundation slabs, grillage foundations or piles [8, 9, 22], etc. must involve
their intuition in the process of selection of the most suitable model at the given level of
structural design.
Design of a technical object O always assumes a sufficient understanding in the
sense that the created object O satisfies all actual requirements during the construction and
service life and, possibly, all limit states prescribed in standards. In addition to the limit
bearing capacity (safety against collapse due to any possible reason), the limit deformation
is critical in todays practice more and more often, which means in general terms the
applicability in operation, the tolerance of deflections, settlement, etc. The progressing
development of FEM programs resulted in gradual disappearance of primitivism, which is
practicism based on rough estimates. The development rather takes the trend to precisionism,
which is the currently maximally reachable technical level of the project. It is clear from fig.
5.1 that both extremes are harmful, because they may result in increase of total costs on the
realisation of the technical object. These must be increased by the costs of repairs of defects,
adaptations, renovations, etc. that always make a threat if a primitive design was applied.
What must be also included are the costs of a higherquality investigation of the properties of
materials and subsoil, the time and financial factor related to longer and higherquality
calculations, etc. Obviously for various levels of designreadiness, i.e. designs for
competitions, detail designs of objects of various importance and investment costs under
various conditions, etc. a certain economical level of definition of the term and design of the
technical object exists in which the total costs are minimal (Fig. 5.1). The main principle for
the creation of a model is that neither misunderstanding and misapprehension of the term
and object nor excessive costs from both extremes may occur. Also this principle should be
born in mind when users chose a suitable FEM program from todays rich offer. It is certainly
pointless to primitively estimate the tension in a certain structural detail if it can be easily
calculated by FEM. On the other hand, it is also pointless to resolve the issue of a safe
foundation of the building in the complex buildingfoundationsubsoil system using 3D brick
elements for the soilmassive, which may lead to a system with
5 6
10 10
deformation
parameters. Moreover, an advanced research of physical input data which further increases
the price of projectpreparation is adequate for such a complicated calculation.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
216
Figure 5.1
Economical level of the definition of the term and and design of a technical object, according to [36].
5.1 Introduction to the Theory and Practice of Creation of FEM Models
217
Figure 5.2
Application of different models of structures and their subsoil for FEM analysis. Selection of an effective FEM
model in practice.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
218
Figure 5.3
A typical evolution of the creation of calculation models in three phases, according to [75].
The primitive phase I is characterised by a considerable unfamiliarity with real processes and
estimates made by practitioners. The experience with the costs for repair of defects lead
gradually to phase II that is exploited by real users structural engineers and scientific and
research institutions oriented to the needs of the structural engineers and sponged on by a
considerable amount of theoreticians. Their work is often sterile, it falls into a cemetery of
nonuseful ideas and is forgotten forever. The real benefit is brought by the transition to phase
III where the complexity of the model is adequate to the necessity to guarantee both safety
and economy of the product.
5.1.1.2 The Selection of an Effective FEM Model in Practice
5.1 Introduction to the Theory and Practice of Creation of FEM Models
219
The term effective FEM model (hereinafter
efm
P ) comprises the modelling of physical
phenomena undergoing in the structure, the choice of the dimension of the calculation model
n D, the selection of the type of finite elements including the degree of polynomial p and the
mesh density h , the formulation of static and geometrical conditions (loads and supports), the
interaction with neighbouring buildings, subsoil and soil environment, the consideration of
how the outputs will be interpreted in the design office, in particular which limit states are
decisive and which secondary, whether exact stresses are needed for the assessment of steel
structures or whether just the resultants are adequate for the design of reinforcement of
reinforced concrete structures, and many problemoriented details. It is clear even from this
that the communicable part of this term (5.1.1) (that may be expressed by a language of
any sort verbally or graphically) does not characterise its contents to the full extent of 100%.
A part of contents I is necessarily left intuitive. The objectivity
0
M according to (5.1.1) and
also the rate of completeness
c
M according to (5.1.2) are lower than one. A pessimistic
approach to
efm
P considers it even to be vague, fashionable, changeable depending on the
development and accessibility of PCs in practical engineering, variable according to prices of
software and peripheries, etc., without any particular communicable S and intuitive I
contents, as the presentday explosion of computer technologies and FEM programs gives no
time to develop any desirable intuition. It is difficult to argue with this opinion, as there is
much truth in it. It is, however, not useful for structural engineers at all. It gives no guidance
as far as the choice is concerned. Let us try to perform a purely economic evaluation that can
be understood in the market environment. Let us neglect the subjective approach to the
analysis of structures, which comprises (i) the curiosity of the user who is interested to know
more about things than is necessary to put the design into practice and (ii) the interest of the
researcher who wants to collect a set of results at various  h p level and who uses FEM to
play interesting computer games or to obtain annexes for thesis, etc.
The economic evaluation must start from the total costs of all the FEM calculations
relating to the given project in the phase of preparation and service of the analysed structure
(Fig. 5.2). The user must realise that there is no sense in selecting a more complex model
unless they order for it the corresponding physical input data or their investigation. This
applies mainly to the subsoil of buildings, as each structure must be laid on a foundation (with
the exception of space satellites). The high precision of the superstructure model with the
foundation comes completely in vain in a thick liquid, which the Winklers subsoil in fact is.
The reliability of a whole is decided by the weakest link of the chain. If the savings achieved
in the calculation are 100%, i.e. we apply no FEM model and limit ourselves to an expert
estimate, it represents a correct allowable procedure. It is sufficient if the project is properly
signed, as calculations are required by Building Control departments only for more complex
projects in order to prove that there is no danger of considerable public damages, threat to
human lives or life interests of neighbours, etc. Consequently, both the structural engineer and
investor take the risk of defects in the structure. These defects may appear already during the
construction (which usually has just economic, not juridical consequences) or later during the
service life of the building. Once a defect is discovered, it is usually advisable to perform the
FEM calculation that determines both the origin of the fault and the responsible party,
suggests the appropriate structural changes, etc. These calculations require the input data
obtained through measurements performed on the structure, which is usually an expensive
matter: the magnitude of absolute and relative displacements, the strength of specimens of
materials taken, the photo documentation of the defect, the nondestructive defectoscopic
5.1 Introduction to the Theory and Practice of Creation of FEM Models
220
monitoring (ultrasonic and radiation devices, verification of the position of reinforcement),
etc.
The threat of defects in the structure and considerable damage leads to the opposite
extreme, analogous precisionism (Fig. 5.1) in the overall conception of the project, now in
the field of FEM calculations (Fig. 5.2). What is typical for this attitude is the increase in the
dimension of finite elements n D (art. 5.1.2). The users do not realise that the division of a flat
slab laid on columns to 3D elements (art. 4.1.6) could make some details in stresses more
precise only on condition that they input adequate input data about physical properties of
concrete and reinforcement, including the detailed location of the reinforcement, which is not
yet known. The modulus of elasticity of concrete
b
E and its Poissons coefficient
b
have
distinctly cumulative character and are not applicable to tensile regions with hairlike cracks.
If the users manage to get more exact physical relations between and , it is just an
additional expense of time or money for invoices paid to experts and laboratories. The result
may be an unpleasant surprise: practically nothing has changed in the resultants that are
needed for the design of reinforcement in comparison with the cheaper calculation using 1D +
2D elements. Analogously, also the application of an expensive expert program of the most
uptodate hpversion may result in disillusion, as the results practically do not differ from
those obtained by a cheap program using professionally correct segmentation and choice of
mesh size.
It is necessary to realise that no one is interested in the FEM calculation itself, unless it
is not the final goal, e.g. the research of the properties of an element, convergence issues, etc.
which is the problem of the authors of the program. FEM is just an intermediate link that is
more or less useful in a certain phase of the design, and it may happen that the zero variant is
the best option, i.e. there is no calculation at all. Mostly, however, there exists some effective
model for which the total costs of all calculations made within the project (including possible
additional corrections) are minimal (Fig. 5.2). It follows from the previous explanation that it
is not possible to present any 100%valid rule for this selection and many considerations
must necessarily remain intuitive. This irritates inexperienced users of FEM programs and
they complain either that the less complex model has not produced the expected results
(usually in a certain point of stress concentration) or that the calculation with the most exact
model took many hours and that as they knew nothing about the physics of the materials but
only the modulus of elasticity E the result was a triviality that could have been estimated
for free in advance. The most frequent complaints arise from the misunderstanding of the
reduction of the dimension of the model theoretically explained in art. 4.2.2.2., Fig. 4.26 
4.27, formulas (4.2.1)  (4.2.100). Therefore, we will dedicate a few practical remarks to this
question in the next article.
To conclude the considerations about effective FEM models, it is suitable to point out
that it is not a formal, intuitive or vague term (art. 5.1.1.1.), but a normal term with
communicable ( S ) and noncommunicable intuitive ( I ) contents that develops both over real
time (history of technology), and over physiological time of the life of an individual with
growing experience or existence and changes of a working team. This process was suggested
with a kind of exaggeration as early as in 1976 by H. Duddeck [75] and is schematically
indicated in Fig. 5.3. The phase II is nowadays typical for FEM models. Everyone has already
understood that mere estimates are not sufficient to achieve success in competition, in tenders
and price quotes. Therefore, FEM is used generally, but similarly to any demand for
technology a great many people whose immediate interest is not an engineering work but
5.1 Introduction to the Theory and Practice of Creation of FEM Models
221
FEM itself parasitize on it. This interest may be considerably useful. Without professional
mathematicians we would hardly find our way in the finite elements  see the pioneering
works of M. Zlamal and A. Zenisek. [3] to [5]. Also thousands and thousands of theses and
conference presentations, the purpose of which was just to represent an item in the list of
produced publications (required by the school or academy), belong to this group. Most of
them brought nothing understandable for people in practice or, in the worst case, they
frightened the readers by unproven statements about disadvantages of e.g. commonly used
Lagrange elements, which were temporarily abandoned in favour of a vast number of
elements based on other variational principles (art. 3), but which are once again used widely.
Fig. 5.3 indicates also a completely sterile branch of theoreticians who are far away from
practice and who doggedly follow more and more complicated models, which are sometimes
not understood even by similarly obsessive theoreticians, which brings immense losses of
spiritual potential due to duplicities and errors flooding hundreds of international conferences,
bulletins and monographs. The approach of theoreticians who are working directly for
practice and who are directly commercially interested in practical application of their theories
is indicated in the figure by a thick line and follows the trend towards the engineering
mastership characterised by a maximal possible simplicity for the required level of
accurateness of the model. Due to the infinite diversity of possible applications, it is
impossible to derive explicitly the what is not substantial and what can be neglected. The
intuition, which can be trained only through (i) a continuous contact with practice, (ii)
successes and failures, (iii) lessons taken from mistakes made by others and from ones own
slips, (iv) a continuous dealing with topical requests of hotline users and (v) the principle to
use exclusively true statements, plays an important role here, even though it is unpleasant to
everyone including the author of the theory and model and even though it may be sometimes
accompanied with temporary financial losses.
5.1.2 Dimensions of the Model for FEM analysis
The selection of the dimension of the model for FEM calculation is extremely
important both objectively and economically. Therefore, finite elements of various 1D to 3D
dimensions were discussed in detail in theoretical chapter 4, art. 4.1.4 to 4.1.6, fig. 4.1 to 4.20.
It was stressed in art. 4.2 that only 3D objectvolumes exist in physical terms and that objects
of lower dimensions 2D and 1D are abstract (art. 5.1.1.1) and the terms related to them
must be understood in that way. The reduction of the dimension is of course very useful. It is
only necessary to understand correctly the meaning of 2D internal forces and deformations of
walls, slabs and shells and 1D internal forces and deformations of beam elements of beams,
trusses, frames and grids, ribs and other stiffeners of planar structures. The information in art.
4.2.2.2 to 4.2.2.6, fig. 4.26  4.27 dealt with this issue. These pieces of knowledge will be
illustrated here through a few practical examples of modelling for FEM calculations. In
order to understand which considerations are in question, it is necessary to emphasise the
following fact at the very beginning as it is unjustly omitted by many:
Most users of FEM programs have no feeling of subjective happiness connected with
the use of those programs. Briefly said, they do not enjoy themselves and FEM rather
represents a required task (similar to performing some arithmetical operations on a calculator)
that is necessary for the main goal: the prognosis of the behaviour of the structure. If one
5.1 Introduction to the Theory and Practice of Creation of FEM Models
222
needs to find out e.g. the result of the mathematical operation 15.73 / sin 10.5, they type the
numbers and functions on a keyboard and read the result or store it in the memory for
subsequent use. Only very few individuals are interested in the number of decimal digits of
and the representation of sine function, whether the calculator uses in a subsequent operation
just the digits displayed on the display or whether it employs the internally stored 16digit
value. Perhaps no one is interested at all in (i) which Taylors series was used to derive the
sine, (ii) which test is used to check the number of elements in this series to reach 16 valid
digits, (iii) how the decimal inputs are converted to binary, octal or hexadecimal forms
accessible to the arithmetic unit of the calculator, etc. All this was solved by specialists in the
prehistory of electronics. It may be found in various older resources and encyclopaedias.
Nobody today asks mathematicians or company dealers about it. Using a hyperbole we may
say that the corresponding experts have built for themselves an eternal monument of glory
and, at the same time, they dug the grave for the interest in consulting service in this field, as
it works without any mistake. A similar situation is today with the systems of programs based
on FEM that reached such perfection and usersfriendliness that the questions relating to FEM
itself disappeared completely, even though they were quite frequent in our country in the
years 1970 1980. The interest in training courses dealing with FEM principles dropped,
while the interest in the creation of models for FEM increased however, for purely
utilitarian reasons as an assistance to practice. This is given by the position of the user of
FEM program for whom it represents just a part, even through an important part, of the whole
work on a project according to Fig. 5.4, blocks 6 to 11. Even the presentday level of input
and output preprocessors, graphical environment and automation of many tasks does not
allow for removing the human factor from block 6 (preparation) and 11 (checking by
professional feeling before releasing the results for the next project phase). In block 11 we
may discover crucial factual mistakes in inputs (formal mistakes will be reported by the
program in time and the calculation will not be performed), e.g. in loading, stiffnesses of
elements and connections, units, etc. Such inputs define a formally correct problem, the
program can process it, but it has nothing in common with the given structure it represents a
completely different problem. It is necessary to (i) come back to the node E or as far as to D
of the flow chart in Fig. 5.4, (ii) correct the mistakes and (iii) repeat the calculation. Not even
the most stateoftheart graphical environments, e.g. WINDOWS 2000, can protect against
factual mistakes, handling of inputs is just more comfortable and faster. If the transition to
subsequent project phase reveals that the design is unrealistic (e.g. the bending moments
cannot be transferred by means of any reasonable reinforcement), it is sometimes necessary to
go back to the initial concept of the statical model.
Repeating of the FEM calculation is sometimes enforced by factual mistakes of the
user some sources of which are given in Fig. 5.5. Many manuals produced by different
companies are written for promotional reasons in such a way as if all users were ideal
unmistakable beings, while the truth is just the opposite. There exists no FEM program,
including the newest expert hpversion, which could prevent factual mistakes. Formally
correct input data are processed by the program even if they represent a technical nonsense
e.g. a foundation slab made of thin rubber (error in E modulus, invalid dimension of
thickness).
5.1 Introduction to the Theory and Practice of Creation of FEM Models
223
Figure 5.4
An overview of the activities of a structural engineer designer, from the initial concept of the statical model
and model for FEM program (block 1) to the supervision during the realisation of the project (block 13).
5.1 Introduction to the Theory and Practice of Creation of FEM Models
224
Figure 5.5
Main sources of mistakes made by the structural engineer designer that result in the necessity to repeat the
FEM solution using the same model but different input data, or using a changed model or applying more detailed
model upgraded to a higher dimension from 1D to 2D, etc.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
225
Figure 5.6
Two basic types of 1D beam models used in common FEM programs. a) The newer Mindlin model taking into
account the influence of transverse shear (slope), geometrically understood as Cosserats 1D continuum with six
independent degrees of freedom in each point. b) A classical beam according to BernoulliNavier hypothesis
with zero shear deformation. c) Explanation of the difference between (a) and (b) on a truss girder. d) An
ordinary solid beam
5.1 Introduction to the Theory and Practice of Creation of FEM Models
226
5.1.2.1 The 1D Models
Practically all newer FEM programs use the model of a beam according to Fig. 5.6a.
Exceptionally, classical beams neglecting the influence of deformation due to shear forces
(Fig. 5.6b) may occur. The theory of reduction of 3D and 1D was explained in art. 4.1.4 and
4.2.2. The users should be aware of what they gain by the reduction and what they lose. If
only six values the resultants of three stress components , ,
x xy xz
(another three are
neglected in the whole body of the beam) are available in one section of a beam, then an
infinite number of stressstates correspond to this group of six numbers and they cannot be
differentiated by the 1D model at all. Let us emphasize now just that for short beams the
deformation may be caused mostly by shear and thus the calculation using a classical model
gives a dangerously small deflection, even just 10% of the actual one, which is thus ten times
larger. Also bending moments M and shear forces Q are commonly affected by the error
which even relatively grows in limit situations (load located near supports) above all limits,
yet for quantities that are small in absolute value. For quantities that are decisive for sizing the
errors reach even tens of percent. Only very slim beams are not sensitive to the influence of
shear deformation (slope) as most of deformation in them is due to bending. The influence of
transverse shear can be illustratively demonstrated on a truss girder (Fig. 5.6c) where the
upper and lower chords transfer bending moments (stress
x
is concentrated into the chords)
and the diagonals and verticals (generally the beams between chords) transfer shear forces
(shear stress
xy
concentrated into their axial forces). The model of a classical beam is a truss
with absolutely stiff diagonals and verticals that do not allow for the change of the right angle
of the diagonals. If we omit the diagonals and verticals, we get a limit configuration that
models a beam with no shear resistance in which the total deformation is due to shear due to
the change of the shape of the diagonals and verticals which ultimately results in a
kinematically indeterminate structure. Real solid beams (Fig. 5.6d) behave rather in a
classical way, but their shear stiffness (art. 4.2.2.6) is finite and corresponds to an
intermediate configuration Fig. 5.6c.
Fig. 5.7 is a good example to clearly demonstrate what we lose by the reduction 3D
2D 1D. Each physical body, even a beam, is threedimensional and no other stressstate
then the 3Dstressstate in a continuum exists. If we use a 2D model, we lose the possibility to
follow the details in the interval 2 2 h z h < < . An infinite number of stress distributions
may be assigned to the same internal forces of a 2D model in the above mentioned interval
it is sufficient if they have identical resultants, more exactly identical integrals (4.2.13),
art. 4.2.2.2, Fig. 4.26c or the same intensities in point ( , ) x y of the 2D model.
The reduction to a 1D model is more dramatic, as the overall stressstate of a section
of the beam i.e. three functions of two variables ( , )
x
y z , ( , )
xy
y z , ( , )
xz
y z are
characterised by six numbers , , , , ,
y z xd yd zd
N Q Q M M M defined by integrals (4.2.14), art.
4.2.2.2., Fig. 4.26e. In a symbolic way, we deal with the degradation of information whose
cardinality is
2
3 into six numbers, which is true even if we admit just the countable
infinite number of points ( , ) y z of the section even though we deal with the infinity of a
larger cardinality of a continuum. Practically said: the given six numbers that are called the
internal forces in the section of a beam cannot characterise any differences in the stressstate
5.1 Introduction to the Theory and Practice of Creation of FEM Models
227
, ,
x xy xz
if the resultants (4.2.14) are identical.
It follows from the above that any system of forces with a zero resultant can be
added to the load, because it does not change the total resultant (4.2.14). The example is in
Fig. 5.7b (it is not possible to differentiate the load on the upper and lower flange), Fig. 5.7c
(supports of the beam) and Fig. 5.7d (various stressstates resulting in the same moment
yd
M ). Following from the Saint Venant principle, the static effect of an equilibrium system
of forces acting on a small domain
0
with the characteristic dimension h manifests itself
just in this domain and its close vicinity up to the distance kh , where 2 k (ordinary beams)
and 4 (more complex configurations). This is in fact the main reason why we operate with the
resultants (4.2.14). In a Saint Venant domain
0
and its close vicinity we cannot analyse the
detailed stressstate by means of beam internal forces. Those who are interested in the stress
in the vicinity of the beam end must define its support conditions more accurately than by
mere definition of deflection w and rotation of the end section.
On condition that we require even more detailed analysis that cannot be covered by
such approach (e.g. the concentration of stress in the corners of flanges welded to the end
plate or the stressstate of an anchoring concrete block) a 3D model may be necessary, as the
general spatial stressstate cannot be simplified by any credible assumption.
Common practice often applies 1D elements whose centroidal axes do not meet in the
common point of intersection, which results in eccentric joints (Fig. 5.8). Older FEM
programs modelled these
5.1 Introduction to the Theory and Practice of Creation of FEM Models
228
Figure 5.7
a) The reduction of the dimension of the model. b) If we add a state with a zero resultant we may transfer the
load from one flange to the other one, which cannot be covered by a 1D model. c) A 1Dmodel cannot
differentiate whether the beam is supported on the upper or lower flange. d) Quite different distributions of stress
x
along the section give the same bending moment M
yd
for the model with a 1D beam.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
229
Figure 5.8
Various structures composed of beams, modelled by means of 1D elements. a) Continuous beams with
differently strengthened flanges. b) Concrete frames with different effective widths of slabs. c) Eccentric
connections of beams. d) Eccentricity of 1Delement in relation to the basic 2Dmodel of a slab or wall. e)
Eccentricity of the joints of beams in a truss girder. f) Eccentricity of the rib in relation to the middle plane of a
shell. g) Eccentric joints in masts.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
230
Figure 5.9
Modelling of eccentric connections of beams to nodes of an analysed system composed in general of 1D and
2D elements in the configuration when the centroidal axis of the beam remains parallel with the line
connecting the nodes and the eccentricity is defined just by two components in axial axes x
A
, y
A
.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
231
Figure 5.10
General eccentricity in which centres of gravity I
C
, J
C
of the endsections of the beam are arbitrarily shifted from
nodes I
A
, J
A
of the mesh of the analysed structure that consists generally of 1D and 2D elements. Vectors of
eccentricity e
I
, e
J
are generally different, therefore the centroidal axis x
C
= I
C
J
C
is not parallel with the axial line
segment x
A
= I
A
J
A
.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
232
Figure 5.11
Main centroidal axes x
C
, y
C
, z
C
of each 1D beam element determine its position in space through the line
connecting the centres of gravity x
C
= I
C
A
C
that is defined in eccentric joints according to Fig. 5.10 and through
the rotation of the coordinate axes y
C
z
C
from the basic position (defined by the program) by a certain angle
0
.
To be more userfriendly, uptodate FEM programs enable the user to define this position simply and in an
illustrative way by an arbitrary point B
y
located in plane y
C
x
C
or B
z
in plane z
C
x
C
. We select a distinctive
point e.g. on the surface of the beam, which enables the user to work in the interface with AutoCAD, see [80].
5.1 Introduction to the Theory and Practice of Creation of FEM Models
233
eccentricities by means of what was termed jointbeams whose lengths were equal to the
eccentricities and whose stiffness was either (i) infinitely large, practically e.g. 10
6
times
larger than that of structural beams or (ii) finite which made it possible to express the
elasticity of the joint. This lead to an increased number of nodes of the calculation model and
increased number of unknown Ns , notwithstanding the unpleasant inputs relating to the
definition of the fictive beams. Newer programs treat the eccentric nodes in a smarter way
without increasing the number of unknown Ns . As early as in 1980 the program system
NEXX enabled the user to define a parallel connection of a beam, e.g. a rib of a slab, to the
nodes of the system in a way shown in Fig. 5.9, the legend of which gives the explanation of
the procedure. The possibility to define a general eccentricity according to Fig. 5.10 was
implemented in the same system in 1993. This creates a new possibility to define a system
composed of beams. It is sufficient to define the axial configuration as a system of lines
connecting axial nodes ,
A A
I J that (in general) do not have to be structural nodes. Moreover,
even the whole axial configuration may be just a geometrical auxiliary system. In the next
step, the central configuration consisting of centroidal axes ,
C C
I J of structural, physical,
beams is defined. This can be achieved through the definition of three components of the
vector of eccentricity ,
I J
e e at each endpoint of the beam, which can be done depending on
the view of the user either in global axes , ,
G G G
X Y Z that are common for the whole
system, or in axial axes defined separately for each line segment ,
A A
I J . This defines the
position of a physical beam in space with the exception of one parameter that represents the
position of the coordinate axes
C C
y z . To do that, the structural engineer may define any
point
y
B in the main centroidal plane
C C
x y or
z
B in the main centroidal plane
C C
x z . Such a
point may be easily chosen for example on the surface of the beam (e.g. in beams made of
rolled profiles in the intersection of the web axis and the face of the flange and it is similar in
concrete T sections etc). All ambiguities are thus resolved. The input is always unambiguous,
has no exceptions and may be easily connected to AutoCAD.
To conclude, let us mention that the position of 1D element in space must be
unambiguously defined including the position of central axes
C C
y z , i.e. angle
x
or point
y
B or
z
B also in the case of elements of solid circular and hollow circular profile, i.e. bars
and tubes. Their axes
C C
y z are not physically unambiguously determined their position
is arbitrary. If the user provides no input (default), the program assigns them a standard
(termed basic) position with one of the axes situated in the basic projection plane. The
calculation is performed without any problems and unless the user is interested in the position
on the beam in which the extreme stress occurs, there are no difficulties. If, however, a
reinforced concrete thickwalled tube with a distinct bending in one plane is to be analysed
which should happen even in a complex irregular distribution of longitudinal tensile
reinforcement then the user would have to be aware of the direction of moments ,
yd zd
M M ,
i.e. they would have to know the position of axes
C C
y z . The situation is similar with steel
tubes if the position of the stress extreme is to be found. For such situations it is definitely
sensible to select the position of axes
C C
y z in advance and in such a way (i) that e.g. the
axis
C
z
is situated in the direction of the expected extreme or in the direction that is important
for other structural reasons and (ii) that it makes the calculation model clear.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
234
5.1.2.2 2D Models
The reduction of the dimension of a problem to 2D was theoretically explained in art.
4.2.2.25, fig. 4.26  4.27 where we introduced 2D internal forces (4.2.13) and deformations
(4.2.49) related with each other through physical formula (4.2.57) with the matrix of physical
constants
2
D . The elements of this matrix may be determined from constitutive
(corresponding to 3D reality) constants
ik
d by means of integrals (4.2.63) within the interval
of the thickness of the 2D model ( 2, 2) h h . The decomposition of internal forces into three
membrane (wall) forces , ,
x y xy
n n q and five bending (slab) forces , , , ,
x y xy x y
m m m q q
according to (4.2.64) is useful in technical terms. Using common assumptions of a linear
mechanics and orthotropy of materials we may obtain two separate physical relations for the
membrane and bending states (4.2.66). The membrane state is simpler in fact the plane
stress with matrix of physical constants
2
(3, 3)
m
D , for example see (4.2.72). What is more
complicated is the bending state with matrix
2
(5, 5)
b
D . Its decomposition into the bending
part
2
(3, 3)
bo
D and shear part
2
(2, 2)
bs
D according to (4.2.75) is almost always sufficient in
common practice. The decomposition of the required nine constants (4.2.73) is described in
detail in (4.2.74) with an example of isotropy (4.2.75). Therefore, the reduction of a 3D body
of a slab, wall or shell to a 2D model is theoretically clear. In this paper we will try to point
out some useful remarks to FEM users concerning the technical meaning of 2D models whose
finite elements were described in art. 4.1.5
There are no technical problems with models of walls with plane stress, because all
FEM programs assume silently (without mentioning it in manuals) the existence of what is
termed generalised plane stress. This is the state very close to reality with no normal stress
z
acting on faces of walls. The detailed theoretical analysis, differences against the
theoretical plane stress and plane strain and the overall analysis of what is termed plane
problem of elasticity can be found in Czech title [31]. Structural engineers can do with the
following remark: Individual 2D models differ from each other just by their physical
constants and, therefore, we may obtain every solution by means of a FEM program on
condition that we input the corresponding constants. Only three of them must be input for the
most frequent isotropic case: modulus of elasticity E , Poissons ratio and thickness h that
(as we know from art. 4.2.2) belongs in 2D models to physical data, because the geometrical
dimension in the direction of the normal does not exist. The algorithms also contain the
thickness h only in the product Eh . Only in postprocessors, which determine the 3D stress
state, thickness h may occur independently, e.g. in section modulus
2
6 W h related to
width 1 b of a section of the wall. Similarly, there are no troubles with the membrane stress
state of shells, especially if the FEM program uses planar shell finite elements, which is in
fact applied in majority of FEM programs. If the finite element is a planar one, its membrane
parameters define the same stressstate regardless whether it is a part of a wall or a shell.
Certain difficulties arise of course in 2D models of plates or in the bending state of
shells, which we explain on an example of a planar slab (Fig. 5.12). Many users of FEM
5.1 Introduction to the Theory and Practice of Creation of FEM Models
235
programs have the tendency to require from the model more than it is able to offer and to
absolutise its results up to absurd conclusions contradicting both the correct technical
interpretation and the simple engineering intuition. A slab is in fact a very frequent model and
may be of various shapes with corners, openings and cuts, may be exposed to continuous and
concentrated loads, may have linear, column and elastic flexible supports, etc (Fig. 5.12a).
Misunderstandings resulting from unfamiliarity with the theory sometimes occur in practical
applications. For example, an ordinary rectangular bridge slab subjected to a uniformly
distributed load should according to the opinion of a naive structural engineer behave as a
simple beam (Fig. 5.12b), which really occurs for zero transverse contraction 0 . If we
input for concrete 0.2 , the program draws and prints in the corners of the slab large
concentrations of reactions ( ) r y . In the effort to remove them, the user applies finer mesh of
finite elements, but the concentration increases and, naturally, its extent is reduced. Technical
support of the program producer confirms it as a correct result which would be even more
distinctive for larger values of , e.g. roughly 500% for 0.5 . What can reassure the user is
the print of verification results obtained by the prestige program ANSYS as well as the exact
theoretical solution that gives in the corner an infinitely large q the well known corner
singularity. Moreover, it is possible to explain this effect in a popular way by cutting the slab
into beams according to Fig. 5.12c. For 0.2 and the slab subjected to bending the lower
faces of the beams get narrower and the upper ones get broader, which is in a compact slab
possible only through the effect of transverse bending in the y direction accompanied by
moments 0.2
y x
M M which in the y direction act everywhere in an infinitely wide slab. If
a finite bridge slab has a free edge, the condition of 0
y
M can be met by the removal of
0.2
y x
M M , which means by the application of moments 0.2
x
M along this edge. This will
necessarily produce in linear supports of the slab additional reactions ( ) r y that violate the
constant r expected by the naive structural engineer. The system of removing moments
0.2
x
M and corresponding reactions ( ) r y is in equilibrium and if the bridge slab is
considerably wide, i.e. B L ? , it has the Saint Venant character. It affects just the parts near
the free edges. In the middle of the slab the distribution of reactions ( ) r y is practically
uniform, as an educated structural engineer familiar with beams can intuitively expect. The
whole mistake is thus grounded on the incomprehension of the differences between a slab and
a series of beams caused by the Poissons coefficient of transverse contraction.
The 2D model is in practice often used to substitute the typical thinwalled box
bridge structures, which is usually accurate enough, if vertical webs guarantee proper shear
transfer between the horizontal slabs (Fig. 5.13a). The user must solve the problem of
longitudinal overhanging strips bearing usually just the pavements but considerably
interacting with the whole structure. Two extremes are possible: (i) to extend the neutral
plane or axis
1
o
of the section as far as to the edge (Fig. 5.13d), or (ii) to put it into the middle
plane
2
o of the overhanging strip (Fig. 5.13b). The correct solution would be obviously
represented by an intermediate position
3
o (Fig. 5.13c). The problem was analysed in detail
for both coarse (Fig. 5.13e) and fine (Fig. 5.13f) mesh including extensive photoelastic tests
performed on models. Also the transversal arrangement of the structure proved to be decisive.
If densely distributed diaphragms are used, they provide for the rigidity of the shape of the
crosssection. Consequently, axis
3
o
may be closer to
1
o even for large projection of the
overhanging part. These studies were of a considerable importance at the time when the
5.1 Introduction to the Theory and Practice of Creation of FEM Models
236
capacity of common PCs made it possible to solve just systems of equations with 1000 to
10000 unknowns. Today, it is possible to include 2D shell elements into 3D space and
model the box structure in 3D. Equations with 10 000 to 100 000 unknowns can be solved by
the presentday PCs without any problems.
FEM programs automatically assign certain planar coordinates x
P
, y
P
to 2D elements, e.g. applies the principle
that x
P
is parallel with the intersection line of plane of the element and the global coordinate plane x
G
y
G
and its
positive direction is given by the requirement that the angle of axes (x
P
, y
G
) must be acute (smaller than 90)
while 0 is also allowed (Fig. 5.14). The z
P
axis is always the normal to the plane of the element and its
positive direction is defined by numbering of IJK or IJKL vertices in a triangular and quadrilateral element
respectively. When viewed in the direction of the z
P
axis the numbering goes clockwise. The y
P
axis completes
the x
P
and z
P
axes to create a right handed system x
P
, y
P
, z
P
. This rule causes a certain discomfort, because the
internal forces are calculated, printed and drawn in these coordinates (fig. 5.14) and also all deformations are
related to them (fig. 5.15). It may happen that for a general situation when a plated surface is used to model a
complicated shell, e.g. a bucket of a turbine, a bucket of an excavator etc., each element can have different
direction of x
P
, y
P
towards the global x
G
, y
G
, z
G
axes. Effective postprocessors may provide the directions of
principal moments and there also exist postprocessors designing optimum reinforcement of reinforced concrete
shells in a complex configuration of up to 2 x 3 layers in selected directions. Theoretically, it is necessary to
satisfy all exceptional cases when the definition of the position of planar x
P
axis fails, because the above
mentioned intersection line of (x
G
y
G
) does not exist. This is handled by the program and it will explained,
for better clarity, on the most frequent example of a box structure with (i) horizontal walls parallel with (x
G
y
G
), (ii) frontal wall parallel with (y
G
z
G
) and (iii) side walls parallel with (x
G
z
G
), see Fig. 5.16 and 5.17. At the
same time, it is obvious from these figures that the user may influence the positive direction of planar normals z
P
of all 2D elements in such a way, that e.g. the positive faces z
P
= h/2 lie completely inside the box, or on lower,
back and right faces etc. Positive bending moments m
x
, m
y
then produce tension on these faces, i.e. tensile
reinforcement must be designed in reinforced concrete structures. Consequently, a certain systematic approach
is very desirable.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
237
Figure 5.12
Some consequences of the 2D dimension of the model of the slab for the users of FEM programs: a) Shape,
supporting and loading singularities. b) A bridge deck with free edges, concentration of reactions in corners with
nonzero transverse contraction. c) Explanation of the origin of the corner effect. d) Fixing of the slab into a rigid
column. e) An estimate of the distribution of moments in the slab above the real column. f) The state after the
column is removed. g) The detail around the column head must be modelled by means of 3D elements.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
238
Figure 5.13
A box section of a bridge deck solved by a 2D model. a) Uncertainty of the position of the neutral area or axis o
of the section of the overhanging part supporting usually the pavement. b) A rather pessimistic estimate of o
2
in
the centroidal plane of the overhanging part. c) The state as analysed by a more complex model with 2D shell
elements inserted into 3D space. d) Technical estimate for a dense system of transverse diaphragms that
guarantee the stability of the shape of the whole section of the bridge deck including the overhanging parts. e)
Too coarse finite element mesh. f) Minimal fineness of the mesh for the estimate of real distribution of the
quantities in the transverse direction.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
239
Figure 5.14
a) The principle by which the FEM program ESA PT assigns the planar coordinate system x
P
, y
P
, z
P
to finite
elements. If the triangles are subelements of quadrilateral elements, the directions of the axes are the same for
the whole quadrilateral, as far as it is a planar one. b) All internal forces are calculated in the planar coordinates.
Then they may be further processed by a postprocessor that can determine the principal values, directions, etc.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
240
Figure 5.15
Internal forces according to Fig. 5.14 cause the deformations of the differential element dx
P
dy
P
of the finite
element: a) membrane (wall), b) bending, c) torsional, d) transverse shear.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
241
5.1.2.3 Systems Consisting of 1D and 2D Elements
Many FEM programs allow for modelling of structure by means of 2D shell and 1D
beam elements in one model. This covers in particular planar structures (walls, slabs, shells)
with ribs, bracing, ties, side beams, etc. The theoretical issues relating to the mutual
compatibility were indicated in art. 4.1. The user is most interested in the technical
interpretation of the output data, for which we give here a few remarks illustrated by
figures 5.18 to 5.22. They deal mainly with reinforced concrete structures. The presentday
standards for the design of reinforcement are still affected by the beam era and provide
instructions and formulas (verified by experience) for the reinforcement of Tsections in
which the web is formed by a beam of a rectangular crosssection and the flange by what is
called effective slab width. This originally exactly defined term (R. Chwall, 1944) for the
special configuration was gradually generalised and became used in standards for both
concrete and steel structures in connection with two conceptions of analysis of ribbed shells.
The first one was the method of substituting framework which reduces the system to 1D beam
elements equipped with the effective widths of the slab structure, i.e. it concentrates the
physical body into a set of linear 1D elements. The other one was the method of substituting
continuum which calculates just with 2D slab structure and which models the influence of
beam ribs just through physical constants of this structure. This includes in particular shape
orthotropy used not only in ribbed, but also otherwise shaped walls, slabs or shells
(corrugated walls, boxes).
The presentday level of FEM software and PCs gives an optimistic idea about the
uselessness of the two presented conceptions, as the stresses in 1D and 2D elements are
handled separately. For steel structures we come across the issue of continuity of stress in the
connection of 1D and 2D elements, as the correct base functions of Lagrangean elements
guarantee just the continuity of components of displacement, which is sufficient for the
convergence of what is called weak solution. The steplike changes in the stress may be
considered a consequence of not satisfying the equilibrium conditions, they appear also
between the elements of the same 2D dimension. In the field of reinforced concrete structures
the users come across an traditional problem of steel reinforcement that was already
commented earlier. Outputs of FEM programs contain 1D internal forces of beam elements
and 2D internal forces of shell elements. Nothing else is available for the design of the
reinforcement using either a suitable postprocessor or manual calculation. However, these
forces are in central coordinates , ,
C C C
x y z of 1D elements and planar coordinates , ,
P P P
x y z
of 2D elements. But these coordinates are not generally parallel as can be seen in Fig. 5.18.
There exist simple orthogonal systems where (as required) the components are parallel and
can be easily summed (Fig. 5.19). In that case there is only one substantial discrepancy in the
impossibility of a versatile coincidence of the signs of torsional moments in 2D and 1D
elements.
It may be formally removed. When bending moments (Fig. 5.20) are summed we have
to respect the fact that for 2D elements we deal with the intensity of physical dimension
(kNm/m), art. 4.2.2.2., formula (4.2.13). Under the simplest precondition that this intensity is
constant in a certain effective width b , the values M mb + and analogously the normal forces
N nb + are added together. It is similar with the summation the shear forces Q qb + and
5.1 Introduction to the Theory and Practice of Creation of FEM Models
242
torsional moments
k k
M m b + , always with the subscripts of the corresponding axes (Fig.
5.21). In a nonorthogonal system (Fig. 5.22) we must first transform the internal forces into
identical axes, in order to be able to sum the components. If the ribs are thinly distributed it is
more suitable not to use the effective width b , but to perform the integration of internal
forces of 2D elements over the whole width between the ribs. The actual distribution of the
reinforcement depends on a set of other details and engineering invention, which is typical for
reinforced concrete structures.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
243
Figure 5.16
a) An example of planes in a box structure. In all three planes the positive direction +n was selected to coincide
with the positive direction of the global axes. In visible planes, +n goes from inside to outside of the box. Note:
In quadrilateral elements, please, substitute the numbering of vertices IJK by IJKL. b) The orientation of the
planar axes of the elements in face plane r of the box (more generally, in the planes parallel with global
coordinate plane X
G
Z
G
). Only the direction of the +X
P
axis can be determined unambiguously it is parallel
with the intersection of plane r and plane X
G
Y
G
and forms with the +X
G
axis angle in the interval 0 < /2,
i.e. 0. It is a general rule, not an exceptional plane. The positive direction of normal +n = +Z
P
may be selected
independently for the front and back plane into or out of the box and vertices IJK should be marked
accordingly.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
244
Figure 5.17
a) The orientation of the planar axes of the elements in horizontal planes of the box (more generally, the planes
parallel with the global coordinate plane X
G
Y
G
). Only the direction of the +X
P
axis +X
L
// +X
G
(exception
from b) can be determined unambiguously. The positive direction of the normal +n = +Z
P
may be chosen as
going into or out of the box and the local numbering of vertices IJK must be made accordingly. The +n direction
can be chosen independently for the upper and lower plane, e.g. everywhere going out of the box. b) The
orientation of the planar axes of the elements in the side planes of the box (more generally, the planes parallel
with the global coordinate plane Y
G
Z
G
). Only the direction of the +X
P
axis +Y
L
// +Y
G
(exception) can be
determined unambiguously. The positive direction of normal +n = +Z
P
may be chosen independently for the left
hand and righthand side out of or into the box and the local numbering of vertices IJK must be made
accordingly.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
245
Figure 5.18
a) The planar coordinates of a 2D element of a FEM program. b) The central coordinates of 1D elements
adjacent to some sides of 2D elements for different order of coding of their ends which determines the positive
direction of the x
C
axis.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
246
Figure 5.19
a) The simplest case of the interaction of 2D and 1D elements in FEM programs: perpendicular ribs in the X
P
and Y
P
directions. b), c) Coinciding directions of axes X
C
// X
P
, X
C
// Y
P
and coinciding position of relative sides
z > 0. d), e), f), g) The effect of the coding of 1D elements on the direction of the positive X
C
axis. h ), i), j)
Impossibility to get coincidence of the signs of torsional moments of 2D and 1D elements without taking their
coding into account.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
247
Figure 5.20
When bending moments of 2D and 1D elements are summed, it is necessary to respect the differences in their
physical character. For 2D elements it is the intensity in a point. The meaning of the values printed by the FEM
program is: if the vicinity of this point was subjected to bending f constantly along width b = 1m, the moment
acting over this width would be numerically equal to this intensity. If we use another, e.g. the effective, width b,
then moment m b acts over this width in the 2D element. On condition that m is the assumed a constant
intensity. Just this moment in (kNm) may be summed with the moment (kNm) in the 1D element.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
248
Figure 5.21
Shear forces q of 2D elements printed by a FEM program are point intensities. If they acted constantly over
width b = 1m of a section across a 2D element, then the area h 1 (m
2
) would be subjected to the force (kN)
numerically equal to the printed value q. If we use another, e.g. the effective, width b, the shear force on the area
hb (m
2
) is equal to q b (kN). Just this force can be physically summed with the shear force in the 1D element
(kN). Similarly, the torsional moment in the 2D element m
xy
is also the point intensity and only the
multiplication by length b, along which its constant intensity is assumed, produces the moment (kNm) that can
be physically summed with the torsional moment M
x
in the 1D element only if its positive direction is taken
into account correctly. Algebraic sums cannot be generally performed.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
249
Figure 5.22
Note relating to the coaction of a 1D element with two adjacent 2D elements in FEM programs. Generally, the
internal forces of all three elements are printed in different coordinates: in 2D planar and 1D central. Before the
conversion of 2D quantities to physical dimension that can be summed with 1D quantities (i.e. force in (kN) and
moments in (kNm)), it is necessary to transform the 2D quantities into the directions defined by the central axis
of the 1D element. In addition, the distribution of the intensities of the internal forces in the 2D elements must be
taken into account and it must be decided whether we sum their integrals (in the case of significant variation) or
whether we simply multiply the intensities obtained from the transformation by some effective width b taken
from standards or regulations. These standards extrapolate the term b considerably out of the scope of the theory
of 2D+1D systems. It may happen that no formula for b is applicable for a general 3D configuration, in
particular for ribs located in two directions, complex loading conditions and complex geometry of the whole
system.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
250
5.1.3 Numerical Stability of the Calculation of FEM Models
5.1.3.1 Defective FEM Results Due to Arithmetics
In practice, we may usually easily find out who is to blame for intuitively defective
results of FEM analysis: whether the user (invalid inputs) or the author of the program
(mishandled critical places of the algorithms). However, we may come across a situation that
no one is to blame, because the input data are perfectly correct and the program for many
years worked fine for similar problem. It is no surprise for mathematicians and there are
specialists who deal just with such problems. The German Society of Applied Mathematics
and Mechanics (GAMM) established as early as after 1980 a department headed by K. Nickel
[76] the research of which is very useful even for engineers. Recently, a special magazine for
mathematical modelling of systems [77] has been started where we may find interesting
topics for all those who learnt in their practice that the difficulties of modelling of reality does
not have to end with the mathematical formulation and creation of programming algorithms
even when FEM is applied. We will give here a few remarks for common users that may
explain some failures and give advice about what to do in such situations.
Let us start from the elementary example in Fig. 5.23a that is even statically
determinate, which means that there will be no doubts concerning the exact solution.
A vertical suspension 012 consists of two different ties 01, 12 with lengths
1
L and
2
L and
tensile stiffnesses [kN/m]
1 1 1 1
2 2 2 2
K E A L
K E A L
(5.1.4)
It is subjected to vertical force P . Therefore, the same axial force
1
N P ,
2
N P is in both
ties. The first tie elongates by length
1
w , the other one by
2
w , the node 1 drops by length
1
and node 2 by
2
, while the formulas of technical elasticity evidently apply:
1 1 1 1 1
2 1 2 1 1 1 2 2 2
w PL E A
w w PL E A PL E A
+ +
(5.1.5)
When written using stiffnesses we get
1 1 1
2 1 2 1 2 1 2
( )
P K
P K P K P K K K K
+ +
(5.1.6)
Practically all presentday FEM programs solve this problem by the deformation method with
two unknowns
1
and
2
, for which they assemble two equations of vertical equilibrium as a
very primitive special case of the procedure given in art. 2.3. The detailed form of the system
of two equations K f is:
5.1 Introduction to the Theory and Practice of Creation of FEM Models
251
1 2 1 2 2
2 1 2 2
(1) ( ) 0
(2)
K K K
K K P
+
+
(5.1.7)
From the second equation we get:
2 2 1 2 2 1
( ) P K K P K + + (5.1.8)
which, when substituted into the first equation, gives
1
from the equation with one
unknown:
1 2 1 2 2 1
( ) ( ) 0 K K K P K + + (5.1.9)
which, for exact arithmetics, is:
[ ]
1 2 2 1
( ) K K K P + (5.1.10)
which for
2 2
0 K K fully corresponds to the elementary solution (5.1.6).
The real arithmetics of PCs with the finite number of valid digits applied in the solution of the
system of equations (5.1.7) using the elimination does not have to lead to the exact equality
2 2
0 K K . Even the beginners in programming know that what is called zero tests must be
extended to a small interval , otherwise they would not work. Therefore, the parentheses
(5.1.10) may contain the result
2 2
K K and we have an inaccurate relation:
[ ]
1 1
1 1
( )
K P
P K
+
+
(5.1.11)
This produces the inaccurate reaction of the whole tie:
1 1 1 1 1
( ) (1 ) R K K P K P K + + (5.1.12)
In common practice, the differences in the size of the stiffnesses are not too big and thus the
numerical error is so small in comparison with the value of
1
K that nobody will spot it in
the output. However, let us imagine the situation that
2 1
K K ? e.g.
9
2 1
10 K K , which may
occur for
2 1
E E ? (steel 12, rubber 01) or
2 1
A A ? (dramatically different crosssections) or
2 1
L L ? (considerably different lengths). When real (single precision) numbers are used, we
may encounter error
9
2
10 K
, which leads to
9 9
1 2 2
10 10 1 K K K
(5.1.13)
Consequently, formula (5.1.12) produces not the correct reaction R P , but the incorrect one
2 R P , which even a beginner must notice in the output it is a 50% error in equilibrium.
We may provide even more drastic examples from real practice: In order to harmonise the
free vibration of a very rigid foundation block of a turbogenerator, a long adjustable steel tie
was used to connect it with the 60 m deep anchoring block. The stiffness of the tie was
negligible in comparison with the stiffness of the blocks and the first solution failed. Let us
give a possible correction: If we divide a very flexible structural element into a larger number
of finite elements, then their stiffness will increase in indirect proportion to lengths L and the
5.1 Introduction to the Theory and Practice of Creation of FEM Models
252
discussed effect will decrease or disappear at all. One of the oldest principles of FEM follows
from that:
The division of the structure into finite elements should never produce dramatic
differences in stiffnesses of adjacent elements.
The graphical representation of the sensitivity of the system of equations (5.1.7) can
be seen in Fig. 5.23b. We have in fact equations of two lines in coordinate axes
1 2
, . Both
equations can be transformed into a practical tangent form
[ ]
2 1 2 2 1 1
2 1 2 1
(1) ( )
(2)
K K K k
P K p
+
+ +
(5.1.14)
The tangent of angle between the first line and the
2
axis is k , the second line forms
with the
2
axis angle 45. The first line passes through the origin, the interval on the
2
axis defined by the second line is p . What is important for the stability of the solution is
obviously the value of the tangent of the line (1):
1 2 2
( ) k K K K + (5.1.15)
If, for example,
1 2
0.732 K K , then 1.732 k , 60 and the intersection of lines (1) and
(2) in Fig. 5.23 is distinct. Its coordinates
1 2
, represent the numerically stable solution of
the system of equations (5.1.7) or (5.1.14). If
1 2
0.073 K K , it leads to 1.073 k , 47 ,
the lines intersect each other under a small angle 2. The point of intersection is not distinct,
the graphical solution would already be unreliable, but the PCs arithmetics would master it
without any problems. The problems occur only with
1 2
K K = the lines are almost parallel.
In the limit, 1 k and the problem has no finite solution the parallel lines intersect each
other in the vanishing point
1 2
.
The situation is similar for three degrees of freedom (symbolically shown in Fig.
5.23c). We obtain three equations with three unknowns
1 2 3
, , analogous to (5.1.7) or
(5.1.14), each of them being the equation of a plane in 3D space
1 2 3
, , . The solution is
represented by the coordinates of the common point, i.e. the point of intersection of these
three planes. If the planes are collinear, they have always one point of intersection see e.g.
the coordinate planes which have a common origin (0, 0, 0) . If a pair of planes or even three
planes are
5.1 Introduction to the Theory and Practice of Creation of FEM Models
253
Figure 5.23
a) A suspension consisting of two different ties. b) Roots D
1
, D
2
of the system of equations (5.1.7) are the
coordinates of the intersection point of two lines (1), (2) from formula (5.1.14). For accuracy e we have strips
the width of which is 2e. The point of intersection may lie in the intersection of these strips. The more the two
lines get parallel, the bigger the intersection of the strips is. c) The example with 3 degrees of freedom, roots D
1
,
D
2
, D
3
are the coordinates of the point of intersection of the three planes. For accuracy e the possible positions
of the point of intersection form a 3D body whose size grows with the decreasing angle of the two planes. d) The
symbolic representation of Ndimensional point D and corresponding vector of righthand sides f. For N=1 it is a
common diagram f
1
= f(D
1
), for a linear problem f
1
= K
1
D
1
.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
254
nearly parallel, the point of intersection is not distinct. And for planes that are exactly
parallel in he limit there exist no finite set
1 2 3
, , which would satisfy the system of three
equations.
Now we will discuss particular examples of FEM from practice where the system
of equations K f has
4
10 to
6
10 unknown parameters
1 2
, , ,
N
K . For 2 or 3 N we
were looking for the intersection of two straight lines or three planes. In general, we deal with
N dimensional linear bodies, inserted into ( 1) N + dimensional space that may be
illustratively called N superplanes starting with 4 N . The solver of the equations
integrated into a FEM program has to find the common point of intersection of
N

superplanes, i.e. N dimensional arithmetical point
1 2
, , ,
N
K . We do not need a big
imagination to realise how many nearly parallel couples may appear in todays common case
of 100 000 superplanes and which factors may influence the distinctness of the position of the
point of intersection of such a quantity of linear bodies (of N dimensional variants).
Mathematics has payed a great attention to this and related problems for many decades.
Important factors are (i) the method that is used to find the point of intersection or the roots of
the system of equations, (ii) the possibilities for adaptations or transformations into what is
called well conditioned systems and effective solution methods. As the solution of
equations in nonlinear problems is performed in one problem commonly 10 to 100 times,
the solver of equations is literally the engine of the program and its effectiveness either
increases or decreases the applicability of the program. That is the reason why we will return
to this issue again in art. 5.1.4. In this chapter, let us give just the following useful
consideration:
Let us imagine that we have obtained some solution of the equation system (5.1.7)
about which we know that its error is , i.e. we do not have exact roots
1 2
, , but
approximate values
* *
1 2
, . We know neither the algorithm nor the precision of the
arithmetics and, therefore, we can analyse neither the causes of the error nor its magnitude.
Nevertheless, we want to evaluate whether the solution
* *
1 2
, is useful. Let us substitute the
approximate values into equations (5.1.7) and we get some other righthand sides:
* *
1 2 1 2 2 1
* *
2 1 2 2 2
(1) ( ) 0
(2)
K K K P
K K P P
+
+
(5.1.16)
If
1
P is very small in comparison with P , it represents, physically, some negligible additional
loading of node 1 (Fig. 5.23a) and if
2
P is nearly identical with P , the node 2 is nearly
subjected to the required load and, as a result, the approximate solution may be accepted as
the basis for further design. The rate of approximation may be, for example, the traditional
difference of 2% in comparison with the required value P that means that we admit
2
P in
the interval 0.98P to 1.02P and in node 1 we will tolerate the unwanted additional load of
0.02P to 0.02P . For more complex problems we can use, for example, norm (272) or tests
(288), (289), etc.
Let us generalise this indirect solution in which we try to find the input of righthand
sides
i
f for roots
i
for an arbitrary number of unknowns N , that means for a system of N
equations K f . For each set of unknowns
obtained by an indirect method, i.e. the solutions corresponding to some righthand sides
*
f ,
we may be interested only in solution corresponding to an acceptable righthand side f .
The abovementioned tests cover obviously also the inaccuracies of the arithmetics resulting
from rounding of intermediate results to a given number of valid digits. Therefore, it may be
difficult to distinguish them from the inaccuracies resulting from other sources. There are,
however, situations when the source of the error is quite clear. If, for example, a program
code working with real (single precision) numbers produces error (5.1.13) and the program
prints the value of the reaction 2 R P with the 50% error even though we analyse a
statically determined structure with evident R P and the input value is correct we must
have come across a numerical instability, which may be overcome by transition to double
precision code working with numbers with precision
18
10
, and in
multiple arithmetics it may be even smaller, but there are always infinite amount of
numbers of the continuum between two displayable numbers. It is possible to
formulate a mathematical statement that almost no number is displayed from the set of
the continuum of real numbers, which means that the continual measure of displayed
numbers equals zero. No assignment can create enough numbers to fill continuously
any small or large interval of the continuum. The consequences can be illustrated on
elementary operations of subtraction and division. Expressions ( ) ( ) a b c d are
sensitive. If ( , ) a b and ( , ) c d are almost identical, small deviations in numbers may
cause even the change of the sign of the result. There exist even seemingly stable
operations with surprising failures.
One of the oldest algorithms was known already in the years long before Christ to Euclid who
5.1 Introduction to the Theory and Practice of Creation of FEM Models
256
probably derived it. It is the decomposition of an arbitrary real number a into the form of a
continued fraction containing only integers
0 1 2
, , , c c c K according to Fig. 5.24a, e.g.
3.1 3 1 10 + , 3.14 3 1 (7 1 7) + + , etc. The fraction is finite for rational numbers, otherwise
it is infinite. The set of numbers
0 1 2
, , , c c c K always defines some number a . Numbers
0 1 2
, , , c c c K for number a can be found comfortably through repeated division. The
corresponding algorithm e.g. in ALGOL is:
begin
real a;
integer c;
start:c=entier(a); (the whole part of a)
if a=c then goto stop;
print (c); (integers are printed here, one after another)
a:= 1/(ac); (core of the calculation)
go to start;
stop: print (a);
end
The computer thus prints a series of numbers c , which are in the Euclids fraction
marked
0 1 2
, , , ,
n
c c c c K and which can be called the decomposition of a real number into
integers. If a given number a is rational, the test a c applies always, even though
sometimes after a very long series of numbers, and the computer stops, otherwise n .
To perform the calculation, we do not inevitably have to use a PC, a usual calculator
may be sufficient. As an example we use the already mentioned number 3.14 with the finite
decomposition {3, 7, 7}. This is really the result obtained on acalculator. But for another
calculator (the precision of numbers in the register is
99
10
this:
{3, 7, 6, 1, 142857142, 1, 9, } K
and the process did not end either. And 3.14 is a rational number. For the transcendental
number the fraction would be correctly infinite, but for the rational value shortened to
15 decimal numbers it would be naturally finate. Let us give the obtained decompositions
according to Fig. 5.24a:
3.14159 26535 89793
Exactly: {3, 7, 15, 292, 1, } K
EL514: {3, 7, 15, 1, 292, 1, 1, } K
K.Nickel: {3, 7, 15, 1, 293, 10, } K
5.1 Introduction to the Theory and Practice of Creation of FEM Models
257
EL5020: {3, 7, 15, 1, 292, 1, 1, 1, 1, 1, 15, 2, 1, 3, } K
Figure 5.24
a) The decomposition of number a into a continued fraction. b) Babushkas paradox for Kirchhoff slab: the
solution of a regular polygon with n vertices converges for n to the deflection of the centre of the slab, which
is only 0.615multiple of the deflection of a circular slab. c) Zone c between the mesh density (number of
elements) n
1
and n
2
in which most the common applications of FEM usually fluctuate and where the
predominant error is caused by the decomposition of the exact function into base functions over elements, briefly
by interpolation. The finer the mesh, the larger the number of equations N and the larger the numerical error.
There exists division n
o
in which the total error is minimal. If a complex application of FEM is used to extensive
systems (N=10
5
to 10
6
), the numerical error may predominate.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
258
Large numbers that appeared in the number 3.14 at the 5
th
position of the
decomposition signal that something is probably wrong with the arithmetics. No excessively
large number occurs for number , but even there we cannot speak about unambiguity. We
cannot even say that the multiple character of the arithmetics or the increase in the accuracy
of displayed numbers has any generally positive effect.
Errors caused by the arithmetics, which can be analysed or removed purely
mathematically (art. 5.1.3.2) must be distinguished from the errors resulting from the physical
nature of the model of the structure. For example, a simply supported Kirchhoff slab in Fig.
5.24b has the shape of a regular polygon with n vertices and was analysed in 1968 by the
team of authors of [3] for growing 12, 36, 60 n , etc. by means of a fully compatible
triangular element and polynomials of 5
th
degree (art. 4.1.5.1) with the elimination of needle
like elements from the centre. The team expected the convergence towards deflection
0
w of
the centre of a solid circular slab. And really, for 60 n the printed
0n
w only slightly differed
from
0
w . However, with n growing further, it started to decrease and for a large n the result
was approaching the limit of
0
0.615w . At first, the coauthor mathematician considered it to
be a defect of the element. Nevertheless, it was a defect of the model and complex boundary
condition 0 w defined along he whole periphery. For such configuration the derivative
0 w n along both edges that meet each other in one node, which implies 0 w n in an
arbitrary direction n , i.e. the nodal mass normal of the slab is fully fixed. This decreases the
deflection of the slab, even though not up to the deflection of a fully fixed slab
0
0.200w . It
would require that a set of boundary normals with the cardinality of the continuum be fully
fixed, while for n we still have just a countable set. This is covered by the mathematical
theory and it represents another illustrative example of the difference in the cardinality of sets
that was already mentioned in the passage dealing with the display of numbers in a computer.
I. Babushka addressed this problem as early as in 1964 at the 1
st
Czechoslovak conference on
mechanics in Smolenice and it is also published in documents [3,5]. The paradox disappears if
we use the Mindlin model of a slab with independent rotation of mass normals with
components ,
x y
(art. 4.2.2). It disappears even in the Kirchhoff model, on condition that
we require 0 w only in the boundary nodes and not on the boundary sides. This eliminates
the cause of point fixation. FEM users can learn from this that a seemingly less exact input of
boundary conditions may improve the solution. It is an instruction that proper attention should
be always paid to the model that is used in the particular FEM program and that is described
in the manual. The input data should be adapted to correspond to the physical substance of the
model including the boundary conditions if the given structure should be reliably modelled.
5.1.3.2 Presentday Possibilities of Improving Arithmetics in FEM
Calculations
We will introduce just briefly three topics addressed by numerical mathematicians in
the corresponding section of GAMM [76] that relate to the solution of the system of
equations, which is the main arithmetical part of FEM. The first one is arithmetics of
fractions in which we limit ourselves to rational numbers. They (except zero) can be always
5.1 Introduction to the Theory and Practice of Creation of FEM Models
259
written as fraction m n, where m and n are integers. There are no rounding errors in this
arithmetics. The systems of equations can be solved exactly and the roots can be presented in
the form m/n as well. Various tests suggest that the solution takes about five times longer
time. The drawback is the impossibility to express exactly irrational numbers, especially roots
and transcendental numbers such as , e , functions log, sin, etc.
The solution was found at around 1970 in what is termed interval or spherical or
(generally said) set arithmetics. It is based either on the interval in which the number a must
lie, that is
1 2
( , ) a a a , or on the form
1 1
( , ) a a r a r + , which once again represents two
determining numbers this time the centre
1
a of the interval and the half of length r . For
complex number, this second form leads to a circle in a plane, or generally, for numbers
defined by more numbers to a sphere, or supersphere in the corresponding space hence the
name spherical arithmetics. The advantage is that the amount of numbers describing the
intervals is reduced, because r may be common for all. The term single number does not exist
in this arithmetics (Dedekinds section on the number axis), there is only a set of numbers
hence the name set arithmetics. Simple procedures are derived for numerical operation
( , , , ) + and the result is again a set of numbers, i.e. the interval in which the result lies.
The term of equality b c must be understood as a complete equivalence of interval
1 2
( , ) a a ,
or
1 1
( , ) a r a r + , etc. Moreover, a very useful term of inclusion substituting the equality is
introduced for situations when it is sufficient that two sets representing two numbers b and c
have a nonempty intersection. Roughly speaking, it is an approximate equality of the order
r it is possible to elaborate tests of programs for it, etc. If we take into account the fact that
input data in technical practice are also intervals or sets with tolerance r , it is obvious why
the interval arithmetics became established so quickly. In our country it was first reported by
J. Kratochvil and F. Leitner already in 1972. The contemporary literature represents an
endless series of several hundreds of titles. The advantages can be summarised into the fact
that it is an apparatus that is capable of displaying the numerical continuum by means of the
finite amount of numbers without discrepancies and unambiguously for all computers.
Even if the set arithmetics overcomes most of the numerical troubles, it will not solve some of
them on its own. As an example we may state a typical logical (decision, branching)
statement, e.g.
if a=b then c else d;
which for the equality a b invokes jump to c , otherwise to d . But what is the equality in a
digital computer? Can also inclusion be considered equality?
Let us demonstrate it on an example: Let the numbers , a b are produced during the
execution of the program in this way: 3 1 3 a ; 1 b . Let us have a usual n multiple
arithmetics, in which the result of the operation 1 3 0.333 33 K and the result
0.999 99 a K , that means that a b for any precision and that command d is performed
incorrectly instead of command c ! It may seem that this can be removed by the introduction
of the set arithmetics. On the other hand, in the set arithmetics not just two, but three
situations may occur, because the numbers , a b are described by sets (intervals)
, a A b B :
Case 1: A B identity of intervals,
Case 2: 0 A B zero intersection
5.1 Introduction to the Theory and Practice of Creation of FEM Models
260
Case 3: 0 A B nonzero
For case 1 the jump to statement c is definitely correct. For case 2 the jump to
statement d is correct. But for case 3 it is not decided what should be done. This case 3 must
be added as the 3
rd
possible value of the logic expression: true, false, unknown. If the latter
happens, it is necessary to start some additional decisionmaking procedure, so that the
algorithm might proceed to the correct target. For example, the equality a b in the test will
be in this case substituted by the inclusion A b and, knowing the numerical precision in the
computer, set A is created in such a way that A a , e.g. (0.999 99, 1.000 01) A K K . As
1 b , it must be true that b A and jump to statement c will be correctly performed. We
presented a very simple example that would be handled by an experienced programmer by an
adaptation of the test, supposing that they would come across any discrepancy during
debugging. But the cases may be significantly more complex and an unusual exceptional case
of a test can be met only after long years of flawless operation of the program. Consequently,
it pays off to prevent these incidents in advance also in the procedures of FEM programs.
5.1.4 Modelling of Nonlinear Behaviour of Structures by
means of FEM Algorithms
5.1.4.1 User Approach to Nonlinear FEM Problems
The commonly used linear mechanics is from the physical point of view only a
special limiting case of a general mechanics. The analysed structure does not change at all in
terms of geometry. The components of displacement vector u are negligible with regard to
the dimensions of elements. The components of the tensor of deformation and vector of
rotation are near zero and, therefore, they may be neglected with regard to 1, i.e. to the
dimensionless unit. Quantities u , and are considered to be only the source of stress
state that is attributed to the nondeformed body that is used to specify the conditions of
equilibrium. In terms of physics, we assume that and are related to each other by a linear
relation by a generalised Hookes Law. Also the geometrical relations between and u are
linear, they do not contain any expressions of 2
nd
and higher degree, roots, etc. Statically, we
work just with external forces which do not depend on the deformation of the structure. We
deal with a conservative system of primary (given loads) and secondary (unknown reactions)
external forces that do a common virtual work over virtual displacements defined by the sum
or by the integral of products of forces and components of displacement in their direction.
With regard to the primary state u 0 they have the potential energy of the position
(art. 3.3). The support (boundary) conditions do not change during the process of
deformation, they are scleronomous. The results of a linear problem do not depend on (i)
how the given external forces reached their final size, (ii) how they are distributed over the
structure, (iii) whether loading or unloading occurred briefly said they do not depend on the
loading path over time. Similarly, the Hookes Law does not contain any information about
the path of stress and deformation over time. For a given or we may unambiguously
5.1 Introduction to the Theory and Practice of Creation of FEM Models
261
find or without examining the history through which the material of the structure
passed before the investigated state.
The assumptions of the linear mechanics are so strong that in reality they are never
fully satisfied and, therefore, it is the engineer who must decide whether they accept all of
them and limit themselves to a linear version of FEM programs. In exceptional situations, e.g.
if only slightly stressed and deformed steel structures subjected to monotonous static load
with permanent connections in joints and bearings are analysed, there is no substantial doubt
concerning the justification of the full linearization. Usually, however, some or all the given
assumptions can be rightly criticised and their fulfilment is questionable. Slender steel
structures may show deflections that evidently influence the arms of the forces, i.e. their
moments related to the points of the deformed system. Partial plasticisation of certain parts in
the analysed limit state can be expected and allowed, etc. The material of concrete structures
is far from the assumptions of Hookes Law and prestressed systems are very sensitive to
even relatively small deflections (fig. 24d). Each structure must be laid on some foundation
and the subsoil (which has a significant impact on the behaviour of the structure) is always
strongly physically nonlinear. Moreover, its deformation depends on the history of the
loading process, on consolidation, changes of water regime, etc. Membrane and cable roof
structures are subjected to nonconservative forces, water pressure is a typical force that
follows the deformation and remains permanently perpendicular to the load bearing surface,
etc. Briefly summarized: There are substantially fewer structures for the analysis of which we
can apply the linear mechanics without any objections than those which evidently do not
meet the strict assumptions.
In the EU countries the civil engineering design practice was strongly influenced in the years 1991 to
1997 by seven EUROCODES numbered temporarily EC1 to EC7. Majority of them was already accepted by
Czech standards CSN P ENV 1991, etc. with corresponding symbols of the field including what is termed
national application documents (NAD). Nonlinear calculations are required if the expected error of the linear
theory exceeds 10%. It is up to the structural engineer to decide who and how should anticipate this error.
These facts contradict with the fact that the presentday design practice of ordinary
structures (estimated 95% of all static calculations) is fully linearised with the exception of
foundation engineering where the physical nonlinearity of subsoil is, at least roughly, taken
into account. Otherwise, we would obtain completely unreliable settlements (dramatically
larger than the real ones), noneconomical foundations and invalid prognoses of nonuniform
settlements that fail to meet technological tolerances. Besides, applicable standards EC7,
CSN, DIN, NORM, etc. would not be satisfied. As we deliberately skip this problem in this
chapter, as it will be described in chapter 6 in more details, we may, even in this paper, limit
ourselves to what is called superstructure. The weakest link in terms of the theory of
reliability is today the foundation engineering and thus the limitation to pure structures is just
a methodological issue. Such a limitation can be applied in practice only in exceptionally
favourable foundation conditions of less deformable subsoil or in some types of statically
determined foundations, e.g. bridges on three supports.
The reasons why structural engineers in real practice use in 95% of situations linear
FEM programs can be summarized into two groups:
rr) Traditional lectures at universities, textbooks, technical guides, numerous standards
and manuals are still usually based on the linear mechanics with an occasional
5.1 Introduction to the Theory and Practice of Creation of FEM Models
262
reference to a certain nonlinear effect that is sometimes considered unimportant. A
typical example is the presentday approach to the calculation of internal forces in
reinforced concrete structures where both EC2 and relating national standards admit
the linear calculation, even though the design takes into account the physical non
linearity and heterogeneity of material. Also the commonly offered FEM programs
available for an affordable price for smaller companies have contained, until recently,
only the linear statics.
ss) The time and financial factor slow down the application of nonlinear FEM
programs. Large mechanical engineering companies, universities and scientific
institutes can already utilise efficient systems, e.g. ANSYS [73], which are not
accessible to entrepreneurs with limited financial resources. They can perform the
required calculations, but on order and they have to pay for it. A calculation performed
by your own program on your own PC has the following disadvantages: (i) it takes
significantly longer time than the linear problem, (ii) it may require more complicated
input data, e.g. the nonlinear physical relation between bending moments and
curvatures in cracked reinforced concrete, detailed geological profile of the subsoil,
constants for conditions of plasticity, etc. The static calculation is just a small part of
the whole project, see art. 5.1.2., Fig. 5.4 and 5.5. To pay a great attention to the
project is reasonable only if the savings made during the realisation of the project
exceed the expenditures spent on obtaining more detailed input data and
performing the calculation, or if there is a chance to get another advantage, for
example in a tender for a contract, or if it is not possible to guarantee the safety of the
building in some ultimate or serviceability limit states, especially stability and
deformation.
The procedure applied in practice follows subconsciously the schedules given in art.
5.1.1., Fig. 5.1 to 5.3. Practice is not interested in the two extremes: primitivism and
perfectionism, i.e. no calculation on the one side or unnecessarily complex nonlinear analysis
on the other side. It intuitively minimises the total time and financial demands on the
construction and its project. Each case has its specific features. The abstract term it is
necessary to use a nonlinear mechanics has its communicable part S and non
communicable intuitive part I , see art. 5.1.1.1, formula (5.1.1). The ratio of objectivity of
this necessity
0
M (5.1.1) the ratio of completeness of the specification of the problem
c
M
(5.1.2) is definitely smaller than one. In the set of subjects, persons and teams only a certain
part
0
n shares the same communicable form that (in some particular situations) the nonlinear
calculation is unconditionally necessary. In chart (282) they hold at least one positive answer
YES to the questions about the expected extent of the change of the shape of the structure
and about the influence of internal forces on the stiffness of elements. If both answers are
NO, the linear calculation is justified on condition that there is no risk of shape instability that
would require the application of determinant equation (286).
Let us notice that none of the above mentioned positions is exactly quantified. That is
where the intuitive part of decision about the necessity of application of a nonlinear
mechanics is. The experts simply conclude on the grounds of his greater or smaller
experience, concern about the future of the construction, subconscious reactions based on
5.1 Introduction to the Theory and Practice of Creation of FEM Models
263
what they learnt from various seminars, conferences and other information sources that a
nonlinear program should be used. They cannot give any exact reason for this conclusion.
To be able to do so they would have to know the result. The intuition occasionally fails and a
sophisticated program consuming tentimes more computer time produces almost the same
result as the linear mechanics the quantities decisive for the design differ e.g. just by 2%
and large differences are only in nonsubstantial details. The engineering proficiency (Fig.
5.3) is, among others, related to a genius intuition and simplicity the goal is a highquality
construction and not the calculation itself. An extraordinarily efficient aid that helps to
achieve this target is FEM, the results of which either objectively confirm the prognosis of the
behaviour of the structure or correct it and provide a welcome guidance for the future. Unlike
the mathematicians who base their considerations just on exactly proved theorems and work
with formal terms P (100% S , 0% I ), (popularly said: they solve what they are capable of),
the engineers solves what must be solve and this solution is not the core or final aim of their
effort. From time to time we may even come across an opinion that the engineer must at least
roughly know in advance what the result will be. If we take into account the fact that they
must input the analysed structure in a way that is not too far from the reality in terms of
designed crosssections of beams, thicknesses of slabs, reinforcement, etc. which is the
decisive factor in the design of statically indeterminate structures (and practically nearly all
structures are statically indeterminate), this ability to foresee is really desirable, otherwise
there is a risk that the calculation would have to be repeated even several times.
Software companies offering FEM programs emphasise in the marketing materials
printed in colours the userfriendliness of their products, efficient graphical preprocessors and
outputs, quality implementation of all possible nonlinear and timedependent problems,
which is after all true information. However, for tactical reasons, it is not mentioned how
educated the user must be to exploit at least 5% from the corresponding program packages or
not to remain just at the level of linear modelling. Similarly, time and financial demands
relating to gathering of input data corresponding to the level of the program are not stated. All
thee factors decide in practice about the total effectiveness of the resources spent on the
preparation of the project, i.e. also about the fact to which extent the nonlinear mechanics is
employed.
5.1.4.2 Assembly of Equation Systems in Nonlinear FEM problems
FEM program systems are usually versatile and make it possible to adjust the required
mode according to (282) or stability calculation according to (286) already during the
declaration phase. In addition, it is possible to follow individual solution steps, to control their
accuracy and final accuracy of the level and distribution of loads, to follow the convergence,
etc. The solution can be performed also manually and can be compared with (i) the exact
calculation (irrational expressions with roots) or with (ii) commonly applied cubic calculation
that is not suitable for rubber materials. This approach clarifies all necessary terms of basic
equation (280) for one step of the general algorithm, which solves the transition of the
analysed structure from an arbitrarily loaded and stressed configuration 1 into a subsequent
configuration 2. Structural engineers with an average expertise can skip elementary textbooks.
For them, the following concise explanation that skips all the details that are interesting
5.1 Introduction to the Theory and Practice of Creation of FEM Models
264
primarily for mathematicians and programmers.
The presentday computers require the complete linearization of one step (of the
atom of the solution), because the only applicable means is the solution of the system of
linear algebraic equations. In algorithms of complex problems this may be implemented just
as a subprogram of some more complicated nonlinear procedure that is incorporated into a
package of mathematical programs with attractive names. It does not alter the fact that, after
all, our PCs are not capable of using any other means. Theoretical documents prefer the
procedure from the most general definitions of mechanical quantities (requiring irrational
geometrical relations with roots (Novozhilov)) to quadratic geometry (GreenLagrange tensor
of deformation and 2
nd
PiolaKirchhoff tensor of stress) which, however, when applied in
FEM without any adaptations, generates systems of equations of 4
th
degree. These can be
relatively easily adapted to equations of 3
rd
degree. The reduction to 2
nd
degree is already
bound by strict assumptions and in order to achieve complete linearization, it is necessary to
make a kind of intervention into physical relations. The procedure requires various
adaptations, symmetrisation, definition of quantities of tensor character, etc. The final effect
can be explained to users of FEM programs in a popular way from below without stating
anything from the whole theoretical base. It requires to limit ourselves to socalled reasonable
or well conditioned problems and to make a sincere confession of what we are able to solve.
We can analyse this problem: We have an arbitrary structure or body with
boundary in state 1 in which the shape is
1
and boundary
1
and in which it shows, in
comparison with the initial state, parameters of deformation
1
, stressstate
1
and in which
it is subjected to external load represented by virtual nodal equivalents
1S
f that are in
equilibrium with internal forces or stress
1
. We are interested in shape
2
with boundary
2
in state 2 in which it is subjected to external forces with virtual nodal equivalents
2
f . The
unknowns are the parameters of deformation
2
and stressstate
2
in state 2. The
deformation and force parameters are defined or measured in both states in the same
coordinate system
G
X (let us call it global) that does not change during the whole process of
solution. The stressstate is related to the same areas of sections across the physical body ,
which means that we may perform decomposition of state 2 to the sum of known state 1 and
unknown increment:
2 1
+ (5.1.17)
2 1
+ (5.1.18)
2 1S
f f f + (5.1.19)
We assume that only small rotations occur during the transition from state 1 to state 2,
and that they can be handled as vector. The question how to obtain the total rotations after
passes through several states will be solved later using a special consideration in which we
separate the pure deformation of elements from their motion as a rigid body.
We are able to solve increments from the system of equations. The analysed
structure or body may consist of 1D elements with internal forces
1
S , 2D elements with
internal forces
1
s and possibly also 3D elements with stress
1
in state 1. As both
1
S and
1
s
are just integral representatives of real physical stresses
1
according to art. 4.2.2.2., formulas
(4.2.13), (4.2.14), it is obviously enough to use the notation
1
in parenthesis, which points
5.1 Introduction to the Theory and Practice of Creation of FEM Models
265
out the fact that matrix
K
that depends not only on stressstate
1
but also on
1
, because the transformation matrix T
depends on
1
. This is marked by symbol
1 1
( , )
K 0 .
The only inaccuracy of the solution is the fact that we assume that each 1D element remains
straight after the deformation, even though its axis in this case is an exact circle. This
inaccuracy can be arbitrarily reduced through the refinement of the mesh of 1D elements in
the cantilever. The more the solved structure differs from this ideal configuration especially
the larger the axial forces are the bigger the error is. The axial force N has the biggest
impact of the six internal forces of beams. This may be compensated by what is termed e
correction, which in fact corresponds to nodal balancing
2 1S
f f with the remaining five
components of the deformation of the beam being neglected. Taking into account the current
state of the development of the theory and computer technology, programs of this type are
meaningful as a simpler approximate calculation of suitable configurations. The influence of
K :
5.1 Introduction to the Theory and Practice of Creation of FEM Models
268
[ ]
1 2 1
( )
L S
+ K K f f (5.1.27)
All transformations by means of matrices
1
( ) T do not apply any more and matrix
L
K is
constant during the whole solution. This may mean just one run through the block of the
solution of equation (5.1.27) that has been prepared from found nodal parameters
1S
f that
form with
1
an equilibrium system of state 1. It can be stored in the memory from one of
previous calculations. The program calculates parameters and to be safe it verifies
whether the selection of this variant by the user is justified. It performs all transformations
that depend on as if the user opted for a more complex variant 4 and continues with tests of
the given load.
Variant 4 does not differ from variant 3 in the first step, on condition that the full value of
load (full load) is input without any increments. However, the deformation is assumed (in
advance) so large that the dependence of matrices K on . cannot be neglected. As a result,
we use equation (5.1.24) in a physically linear form:
[ ]
1 1 1 2 1
( ) ( , )
L S
+ K K f f (5.1.28)
This variant enables the user to control the complexity of the calculation through tolerances
specified in tests. Experienced users can intuitively guess from the character of the problem
that there is no danger of divergence of in no phase of loading (i.e. that there is no risk of
loss of stability of the shape), they do not insist on the accuracy of levels and distribution of
load in partial incremental phases and divide the total load into n increments just to ensure
that the tolerance for rotations in one step is not exceeded. A relatively small n can ensure it,
e.g. 3 n . On the other hand, this iteration is always necessary after the last increment,
otherwise no one could guarantee what is the factual load of the structure producing the given
graphical and numerical output data.
The variants 5 to 8 differ considerably from the variants 1 to 4 by the fact that the physically
nonlinear behaviour of materials of elements of the structure is taken into account together
with their connections which will be omitted for the time being for the sake of simplicity (it
may be substituted by bound or contact nonlinear elements). The complexity of variants 5 to
8 changes according to the requirements of the user and according to the character of the
problem. Firstly, we explain the simplest case of what is called finite problems that do not
depend on the path over time. Consequently, both the real time t and the fictive development
parameter t vanish from all calculations. In practice, it is in particular the case of
monotonous incremental loading of the structure, if the stress and deformation increase in
all elements without any discontinuity in stressstrain diagrams, e.g. elimination of tension
after cracking and after similar steplike changes in diagrams. This does eliminate the
possibility of loss of elasticity and development of plastic, i.e. irreversible, deformations. As
the phase of unloading is not solved and thus there is no way to say whether it is the case of
nonlinear elasticity (a possible return to the initial state) or elasticplasticity (after the return a
part of deformations is permanent). For these problems the program requires that the stress
strain diagrams of the material of the elements be input in the usual form obtained from tests
that can be performed either with controlled deformation , (i.e. function in the form
( ) ) or with controlled stress ( ) . Implicit formulas of the type ( , ) f 0 are
5.1 Introduction to the Theory and Practice of Creation of FEM Models
269
not suitable. For the simplest configurations we use the relation for uniaxial stressstate
( )
x x
f , more precise Hookes Law with one cubic member
3
1 3 x x x
E E + , bilinear
law with two different moduli
1 2
, E E varying at a certain level of
x
while the other modulus
may be zero (plasticity without strengthening), similarly for G modulus in shear, etc. In each
state 1 the program can find both deformation
1
of the element and stressstate
1
depending on the achieved values of parameters
1
, because the history of the structure plays
no role here.
The variant 5 is then controlled by a considerably simplified equation (5.1.24) or by a
slightly extended equation (5.1.26):
1 2 1
( )
L S
K f f f (5.1.29)
In the simplest situation, state 1 is the initial state with zero
1 1
, ,
S
f
1
, the soughtafter state 2
is final (full load state). Contrary to the linear Hookes Law D (valid for any
1
and
2
,
i.e. permanently during the transition from state 1 to state 2), the moduli or other constants of
the physical relation between s and vary. The moduli usually decrease the material
softens, but they may also increase the material hardens. It may happen that we determine
some average values of the moduli between states 1 and 2 that have the following properties:
The calculation using variant 1 gives, with a satisfactory precision, the same result as variant
5. These are what is called secant moduli. In the stressstrain diagram of a uniaxial stressstate
( )
x x
f these are the tangents of angles between the secant of the graph and the
x
axis.
Ideally, it would be enough to solve equation (5.1.29) only once. Let us write it in the form
1 2
( , )
L
K f (5.1.30)
After solving and corresponding stressstate
2
we may determine the real load acting on
the structure i.e. to calculate
2
f from stress
2
. If the userspecified tolerance TOLF (288
289) is satisfied, the calculation ends. If the tolerance is not met, the calculation must be
repeated with a new estimate of the secant moduli. That, however, does not have to be a better
one, if we consider that a structure today usually has 1000 10 000 elements and all of them,
to a different extent, participate in the creation of factual load
2
f . Therefore, it is usually more
reliable to perform the calculation using the tangent moduli with the load divided into n
increments. In each of them the stressstrain diagram is substituted by the tangent in the point
corresponding to state 1 that was reached in the previous step. For more complex physical
laws we deal with tangential planes or superplanes. Matrix
1
( )
L
K is then a typical tangent
matrix of stiffness marked usually
T
K , see (280). This procedure solves equation (5.1.29)
n times, every time with a different lefthand side. It must be ensured that no systematic error
occurs and that the whole process does not produce results that are permanently above or
under the real ones and continuously move away from them. This can be ensured by at least
occasional equilibrium iteration that has been already mentioned in the text about the secant
modulus. It is sufficient to apply a simplified iteration reduced to just one step, in which the
factual load of the structure is determined. This state is then used as the starting point for the
next increment. There exist many practical adaptations of this procedure, see [17, 32, 61, 70,
73]. Also the averages of tangent moduli at the beginning and at the end of the increment
5.1 Introduction to the Theory and Practice of Creation of FEM Models
270
proved useful, i.e. a new repetition of the calculation of one step, etc.
Variants 6 to 8 have been derived from variants 2 to 4 through a similar adaptation that was
used to create variant 5 from variant 1. The difference is that it is reasonable to insist on a
certain reasonable accuracy of the level and distribution of the real load of the structure in
comparison with the required state after the considered load increment. Subsequently, the
tangential moduli of the stressstrain diagrams of the material of elements apply in this state.
Another difference is that it is necessary to specify the term stress. It is usually sufficient to
use an intelligible engineering conception and the stress is related to the same planes of
elements in their element coordinates. These move together with the elements as rigid bodies
and their pure deformation is separated. In 1D elements the internal forces simply relate
always to the same crosssection, regardless of the changes of its position in space. Similarly,
in 2D elements they relate to the same mass normal h . More details can be found in [3, 49,
61, 65, 69 to 74].
5.1.4.4 More Complicated Constitutive Relations and Projects Depending on
the Path
The current state of FEM programs and computer technology allows for the input of
physical properties of materials corresponding to their real behaviour that may be very
complex. What became a problem is how to obtain these properties effectively from
experiments and how to store this information for numerical calculations. This is the issue
addressed by specialists who publish corresponding publications, e.g. [78]. A special group is
represented by geomechanical data [79], which are not covered in this publication for the sake
of brevity and for which we refer to documents [8, 9]. The physical relations are elaborated in
more detail also in newer FEM texts, e.g. [80]. FEM also penetrated many nonmechanical
fields and it becomes commonly applied method in problems dealing with the protection of
the environment [81]. Nontraditional models [82] are newer. They abandon the classical
conceptions [83, 84]. The current trend is to increase the reliability and applicability of FEM
in modelling and optimisation in the design in all engineering fields, see the preparation of the
congress [85]. From the enormous number of innovations of physical character, the problems
depending on the path (this means either a formal history of the loading process or the real
time factor t , i.e. also the duration of individual phases preceding the analysed state of the
structure) are those that have the biggest chance to be applied in common civil engineering
and mechanical practice. An example of the dependence on t was given in article 4.2.1, Fig.
4.23  4.25. The typical feature of problems depending on time is obvious already from the
simple Kelvins rheological model (Fig. 4.24, 4.25): The input of the current load of the
structure does not determine its stressstate and deformation. In general, an arbitrary number
of stressstates and deformationstates correspond to such an instantaneous loadstate
depending on the previous history of loading and unloading. Also the speed of loading and
unloading and the time period during which the load is constant have an impact on the result.
This is in agreement with the examination of structures whose material or subsoil feature
strongly rheological properties, e.g. reinforced concrete foundation slabs with the effect of
creep, relaxation, cracking, consolidation of cohesive soils, change of water regime, etc. taken
5.1 Introduction to the Theory and Practice of Creation of FEM Models
271
into account.
Older FEM programs sometimes offered a special process to follow the relaxation,
i.e. the drop of stress over time t under constant deformation, or creep, i.e. the growth of
deformation under a constant stress. Newer programs have introduced the possibility to
follow the relaxation in a more general way for an arbitrarily controlled distribution of
deformation
1
( ) g t with given function
1
g (Fig. 5.25a,b), which in the special case of a
constant distribution
1
in time t leads to a traditional relaxation. Similarly, it is possible
to separately follow a generalised creep with the input of introduced stress
3
( ) g t , Fig.
5.25c. In real materials both phenomena can occur simultaneously according to more or less
complex constitutive laws that can be described (in a popular way) as a variable plane
( , , ) F t in Fig. 5.25d, even though the dependence is more complicated. In symbolic
coordinates , , t the plane
1
defines a section ( ) t representing a common creep.
Similarly, the dashed section ( ) t across the plane
1
represents a normal relaxation. The
points of dashdotted line
0
( ) f can be considered the representation of a physically non
linear dependence of on in time 0 t . Similarly,
1
( ) f in time
1
t t , where,
however, the symbolism of (Fig.5.25d) fails, because function
1
f does not generally depend
only on and
1
t , but also on the course and duration of individual phases of the development
of ( ) t over time
1
0 t t . Complex FEM programs require the cooperation of the
structural engineer with an expert on the given material. When dealing with consolidating
subsoil (where the tensor of stress is according to Terzaghi principle divided into the
spherical tensor of hydrostatic or hydrodynamic pressure of water in pores of soils and the
remaining general tensor of effective stress between individual grains of the soil), a
consultation with an engineergeologist is normally required. This twophase particular
environment can be accompanied by a third, gaseous, phase (in the pores not completely filled
with water). The constitutive law is adapted according to the degree of saturation (saturation
of pores with water).
An aid that proved useful for the explanation of the difference between the linear
and nonlinear mechanics is the example of a strut or stay in Fig. 5.26 that can be solved
scientifically or with a various rate of simplification and with various numerical
modifications even manually, which contributes to the confidence of users in FEM
programs that sometimes produce surprising results. Common structural engineers are, from
both the faculty and practice, so soaked with the linearity and resulting principles of linearity
(proportionality) and summation of effects, that they trust the results of a nonlinear
mechanics only when they can verify them themselves using their own effort and reasoning.
For clarity reason, we will present here just a few typical results and conclusions. All of them
relate to symmetrical stays or struts in Fig. 5.26 ad, whose solution can be reduced to the
analysis of behaviour of one beam subjected to tension or compression, on condition that we
limit ourselves just to a vertical load in nodes. One beam is subjected to vertical force P . The
condition of symmetry takes the effect in the right end of the beam with the vertical sliding
support (Fig. 5.26e), that prevents horizontal displacement u , which means that the problem
has just one unknown: sagging and deflection in the stay and strut, respectively. A classical
solution follows from the conditions of equilibrium in the nondeformed state and from the
Hookes Law (Fig. 5.26e):
5.1 Introduction to the Theory and Practice of Creation of FEM Models
272
2
( sin ) v P D
D EA L
(5.1.31)
with a common tensile or compression stiffness D. It is advantageous to introduce
dimensionless quantities according to the marks of length in Fig. 5.27e
sin v L b L p P EA
v L b L P EAp
(5.1.32)
Then, the classical solution has the form
2 2
p p (5.1.33)
and in coordinate axes , p it represents a straight line indefinitely long for both the stay
and strut, passing trough the origin ( , ) (0, 0) p in Fig. 5.27a.
For clarity reason, a particular example has been selected that can be for the stay easily
demonstrated by a small model of thin rubber:
2 2
1
2
2[ ], 0.0029 ,
0.05 , 0.05385165 ,
0.02 ,
37.13907 [ ],
arctg( ) 21.80141 ,
sin 0.371391,
0.137931.
EA N L m
a m L m
b m
D EA L Nm
b a
b L
The equation of the line is (5.1.34):
0,137931 p p P EA v L (5.1.34)
A thin rubber fibre can sustain even large elongations without failure (rupture), e.g. 5 .
In the limit we may imagine the state, in which both fibres of the stay are nearly vertical
( ) v and no geometrical change of the inclination of the fibre appears and the limit
relation holds:
0.628609 p (5.1.35)
marked by the dashed straight line in Fig. 5.27a. The behaviour of the stay is given by the
drawn curve with the tangent (5.1.34) and asymptote (5.1.35). Its equation can be easily
derived by means of the exact Pythagorean theorem:
( ) ( )
0.5
2
1 1 2 p
1
+ + +
1
]
(5.1.36)
The inclination of the tangent of the graph is in fact given by the derivative:
5.1 Introduction to the Theory and Practice of Creation of FEM Models
273
( ) ( ) ( )
0.5 1.5
2
2 2
1 1 2 1 2
T
dp
K
d
+ + + + + + (5.1.37)
For very small sagging 0 we get
2
0.137931 dp d , like according to (5.1.34). For
large sagging we may use strong inequality
2
2 1 + ? and substitute the 1 in
parenthesis by number
2
from the interval 0 to 1, without changing considerably the value
in the parenthesis. As a result, we can easily extract
2 2 0.5
( 2 ) ( ) + + + and we get
(in agreement with (5.1.35)):
1 1 p dp d + (5.1.38)
5.1 Introduction to the Theory and Practice of Creation of FEM Models
274
Figure 5.25
Special cases of sections across a general constitutive plane F(s, e, t). a) The distribution of stress s = h
1
(t) in
time t under controlled deformation e=g
1
(t) introduced into the element in the vicinity of the analysed point x, y,
generalization of relaxation. b) The same for another distribution e=g
2
(t) of the introduced deformation.
Generally different stresses s
1
, s
2
correspond to the same parameters e
1
, t
1
for different history e. c) The
distribution of deformation e=h
3
(t) for controlled stress s=g
3
(t), generalization of creep, dependence of value e
3
not only on parameters s
3
, t
1
, but on the whole history s. d) Illustration of relaxation and creep as parts of a
general rheological process.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
275
Figure 5.26
a) to d) Symmetrical stays, in which the solution of a single tie or beam 12 is sufficient. e) to g) Geometrical
relations for large sagging. h), i) The limit case of the initial horizontal position of the elements of the stay.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
276
Figure 5.27
a) The example of the stay used in the text. The solid line represents the exact solution, the dashed line shows the
asymptote for an infinitely large sagging, the dashdotted line means the linear technical solution for very small
sagging. b) The limit case of the stay whose beams are horizontal in the initial state. In a geometrically linear
mechanics, we deal with a kinematically indeterminate problem that cannot be solved. If we admit large sagging,
the solution exists similarly to case b).
5.1 Introduction to the Theory and Practice of Creation of FEM Models
277
Figure 5.28
The behaviour of the strut from Fig. 5.26 b for all possible values of load P. The change of the strut into the stay
(due to break through) happens in part a between deflections w
1
and w
2
. Only the stay can transfer another
increment of force P. This part can be experimentally examined only with controlled deformation w. If we attach
a vertical beam or a spring with stiffness k to the strut, three cases may occur: a) The phenomenon of the break
through remains preserved in another interval a. b) The breakthrough does not happen, only a little indication of
it. c) The system is permanently stable, which occurs for larger stiffnesses k .
5.1 Introduction to the Theory and Practice of Creation of FEM Models
278
This means that the exact solution (5.1.36) includes also both limit cases. In addition,
it includes also the case that cannot be solved by means of a technical linear static calculation
(Fig 5.27b). If 0 , the initial position of the beams of the stay is horizontal. Linear
programs for PCs report immediately after the input that the problem is kinematically
indeterminate. An attempt to solve one equation for one unknown sagging discloses the fact
that stiffness K is zero in comparison with . That means that the equation of type K P
gives an unusable result . In Fig. 5.27b this is marked by the horizontal tangent to the
graph in the origin. The exact formula (5.1.37) provides for 0 exactly such sagging
that is necessary to transfer force P in accord with the behaviour of the model! Large P or
p P EA is naturally accompanied with large or n L . In limit it holds that the
asymptote 1 p n see Fig. 5.27b where also the detail of behaviour with very small is
shown. It leads us to an important general conclusion for users of nonlinear FEM programs:
In geometrically nonlinear mechanics there exists no term of static or kinematic
determinateness or indeterminateness. All problems can be in principle solved. If the
degree of freedom is not limited in any way (by a claw or stop), the system deforms up to the
state in which the static equilibrium of loads and reactions is reached, i.e. the equilibrium of
external forces of the system. Only if such a case can be never reached, the Newtons laws of
motion are applied and the problem is transformed into a dynamic motion problem, e.g.
forced damped vibration, etc. This is rather a rare exception for common static loads which
then change their character according to the nature of the response of the structure, see chap. 5
art. 5.4.
A drastic example: When we attempt to release the fixation of a horizontal cantilever
with vertical load P at the free end, i.e. after the insertion of a hinge into the fixed end, the
linear theory fails. Consistently geometrically nonlinear theory finds (e.g. by the method of
controlled deformation) some states of the broken cantilever, it assigns zero bearing capacity
P to all of them until it reaches the vertical position of the cantilever when it correctly
assigns normal force N P . If a kind of stop was defined in the problem somewhere along
the path of the cantilever (e.g. at some deflection
K
of its loaded end), then the motion of the
cantilever would stop and force P would be transferred by the reaction R in the stop.
The derived exact solution (5.1.36), (5.1.37) valid also for the strut, or for the
reversed stay, in which the length 0 b < and, therefore, 0 b L < . The behaviour of the
strut in the whole extent p is displayed in Fig. 5.28. Several intervals of variability of p can
be noticed there. For all negative p , it is simply a strut subjected to load acting upwards
whose behaviour is identical to the behaviour of the stay it is stable in all phases. The
response of the strut to positive p acting downwards is at the beginning stable according to
(5.1.36), (5.1.37) with negative . Quite simple algebraic operations can be applied to
discover that for a certain deflection (better displacement)
K
the derivative (5.1.37) equals
zero and that for
K
it is negative. In technical terms, it means that load
K
p ,
corresponding to critical value
K
, is the maximum possible one. In an attempt to increase the
load, the strut would break through and converts itself into the shape of the stay, where such
increase is already possible. The process during the breakthrough is dynamic, as there exists
an excess of the force, i.e. acceleration occurs according to the Newtons law. The break
through may be easily followed on a model made of suitable bars. If we return to Fig. 5.26a
and 5.26b we can imagine a rotationally symmetrical system of beams of the strut that
5.1 Introduction to the Theory and Practice of Creation of FEM Models
279
approximately model a flat shell with membrane stressstate. The breakthrough is in every
day practice known as the behaviour of the bottom of an oilcan under required compression
and also as the behaviour of various switches, etc. it is thus a common and actual
phenomenon and the graph in Fig. 5.28 is no technical surprise. We will extend it further by
this consideration:
Let us place a vertical beam or spring with axial stiffness k (in the relation R kv )
under the loaded node of the strut. For the beam
K K K
k E A L , for the springs k is the
known spring constant. It also transfers a part of the load (according to the ratio of its own
stiffness k to the stiffness of the strut) that is in the geometrically nonlinear process variable
and is given by incremental form (5.1.37) by the formula
T
dp K dv .
The summation of both parts of the load produces the graph in Fig. 5.28, from which it
is obvious that three situations may occur:
tt) A small stiffness in comparison to
T
K , i.e.
K
k k < the phenomenon of the break
through occurs again, only the magnitude of the load under which another increase
in loading starts to be impossible changes and the strut becomes the stay strengthened
by the spring (Fig. 5.28a).
uu) Critical stiffness
K
k , equal to
T
K in a horizontal position that is passed during the
break through when the strut becomes the stay. When corresponding load is reached,
we can observe in the model a very easy increase of the deflection under a very small
(in the limit none) increase of the load (Fig.5.28b).
vv) Large stiffness
K
k k > . The graph in Fig. 5.28c is constantly growing. The system is
permanently stable. No breakthrough occurs.
The lesson learnt by the user of FEM programs: The systems solved in practice usually
have 10 000 100 000 elements. Elements stressed differently in different loading phases
meet in one node. If we limit ourselves just to their axial forces N (truss girders), we may
say, in quite a popular way, that some beams may act in the node as springs that stabilise
the buckling behaviour of other beams. The behaviour of a technical structure would be
characterised by an N dimensional graph analogous to Fig. 5.28 (where 1 N ) and with
1000 10000 N with a vast number of inflection, saddle, etc. points. Numerically it
demonstrates itself in coaction with other internal forces of 1D elements (similarly in 2D
elements) by the fact that the program may during the nonlinear calculation report a collapse
of the system (local or even global). It is quite complicated to distinguish whether it is a real
collapse for the given level and distribution of load (i.e. the bifurcation of equilibrium, shape
instability) or just a collapse during some equilibrium iteration due to the influence of an
auxiliary constellation of the system of forces. The user has practically just one option: to
increase the number of increments, i.e. to reduce one level of loading undergoing the
equilibrium iteration. If the system continues to collapse either locally or globally, it is very
probable that the system is really unsuitable for transfer of the given load and some structural
changes must be made. A mere change of crosssections and thicknesses of elements usually
has no or little effect. It is better to change the whole conception of the structure, i.e. to
modify the geometry, introduce stiffeners, etc.
5.1 Introduction to the Theory and Practice of Creation of FEM Models
280
5.1.4.5 The Selection of the Number of Increments and the Course of the
Equilibrium Iteration
Most FEM programs for nonlinear analysis require a kind of usercooperation in the
solution and expert versions of the programs provide some assistance. Only robust programs
have chance for success in practice.
Briefly said: they produce equilibrium results and the users are able to compare the
load with their input values. However, they have practically no chance to compare the
analysed shape of the deformed system (structure) with the real shape that can be determined
through measurement of the real system or at least its model, or alternatively by means of
very complex numerical analysis with such a large number of small elements that it surpasses
the capacity of presentday PCs. What is the core of the approximation of the results?
The complete linearization of one step of the is possible only under such drastic
simplifications that any additional preconditions would completely kill the nonlinearity of
the problem. Let us summarise at least the following main sources of errors in one step of the
solution from chapter 1:
ww) W
hen the influence of the increment, or the correcting load increment, on the reference
configuration 1 is calculated, every 1D or 2D element passes into configuration 2
generally deformed (elongated, compressed, bent, skewed, twisted). The rigid body
motion of the element is som