1

Advanced Finite Element Analysis

Prof. R. KrishnaKumar
Department of Mechanical Engineering
Indian Institute of Technology, Madras

Lecture - 29

Yeah we had, in the last class we had looked at what is called as total Lagrangian
formulation and we had, we had said that two types of formulation that we can follow.
One is an updated Lagrangian and another is total Lagrangian. We had had a discussion
about this.

(Refer Slide Time: 1:12)


We had looked at the total Lagrangian formulation; we were half way through and we
realized that there is lot of algebra that is involved in the total Lagrangian formulation.
So, let us just quickly look at what these are before we go further and look at the finite
element formulation for the total Lagrangian. We write down the functional. All of you
are familiar with it. The functional in terms of the original configuration or reference
configuration and of course we have two terms. You, you know this. Essentially what we
did was to now find out the first variation of the terms that are involved in this W.
2


In order to do that for example, for this it was straight forward. That is the variation of the
external loads or the potential loss due to external loads. Here, we found that if we have
to do this, we have to have a small, you know jugglery or algebraic manipulations that
have to be done or in other words, this term now has dow W by dow C
IJ
. See, of course
this can also be expressed in terms of E
IJ
; we will see that in a minute. In order to do that,
we may have to have some formula in place for this manipulation. So, say for example, if
I have half delta C
IJ
, say, sorry, delta C
IJ
S
IJ
which can be written as delta E
IJ
S
IJ
, so we
can, we need to have what or we need to know what this is.

In order to do that, we will substitute or we will first find out what is the first variation of
F.

(Refer Slide Time: 3:02)


This first variation of F comes from the fact that F can be written in terms of the
displacement u or in other words, if you want further going down this you can write x is
equal to u or x minus X is equal to u, sorry, yeah x is equal to u plus X, sorry. From here
you can write down dow u or dow x by dow capital X; you can write that down in terms
of say, from here you can write dow x by dow capital X, because this is a tensor, is equal
3

to dow u that is here it comes u i comma capital I and so on plus I. That is the term which
comes here for F, so from this you get here and you use and you see that the variation of
this is equal to zero. So, you will get delta u, delta F
i
is equal to delta u i comma capital I.
Of course you know that small i is in the current and the capital I is in the reference
configuration.

With this, now with this, what we do is to look at the first variation.

(Refer Slide Time: 4:24)


This is an important thing - first variation of E which is the, our Green strain. Then you
take each of these terms. Of course we know that E
IJ
is equal to half of F transpose F
minus I. So, you apply the variation for F transpose and F. So, you will get two terms and
of course the variation of I goes to zero. These two terms are now taken or this is taken
along with S
IJ
and then, or in other words, this is multiplied by S
IJ
. So, you will get, using
the symmetry of S
IJ
you will get a term like this. Then, now using the derivation which
we have done now on the first variation of F along with this and then using the Greens
theorem, ultimately you get delta pi to be written in this fashion, from which we said that,
what is that we arrived at? We said that, yes, exactly; so, we can get the equilibrium
equation.
4

(Refer Slide Time: 5:29)


Of course, this is the whole thing. We will get the equilibrium equation, this and we will
get the boundary conditions from this. This is what we saw in the last class and we also
said that delta E for example, if it is, if it is to be written in terms of 11 22 33 12 and so
on, you can write down delta E to be like this.

(Refer Slide Time: 5:45)


5

So, this will be the starting point of our finite element formulation. Is this clear? Now, if
you closely look at it, if we look at the concepts, forget about the terms that are involved.
They are no different from what you did in your earlier class; they are the same. The only
difference now is that the type of strain measures and the stress measures that you are
using, the stress and strain measures that you are using, are different. So, that involves lot
more algebraic jugglery. The concept is exactly the same, there is no difference with it;
but because of these large number of terms, the expressions becomes extremely complex.
Is it clear?

We said that we can go and do the finite element formulation straight away from this.

(Refer Slide Time: 7:07)


The finite element formulation starts with writing down X in terms of say, the X
I
in terms
of the sigma, what is it? Shape function times the X nodal, very good, in terms of say, I
am putting alpha here, where alpha is summed up to the total number of nodes and so this
is alpha I and so on. This is our regular formulation for the iso-parametrical formulation
which we had used again in our earlier classes, so that now u is again written in the same
fashion. If you, if you look at u
i
, then it is again written in the same fashion N alpha u i
alpha and so on, where again alpha indicates the nodal quantities. Is there any question?
6

Essentially what we are doing now is, note this carefully; forget about for the time being
the algebraic manipulation. We are exactly following again the same procedure. Only
thing is now we are going to get all the terms that are involved here from the discretized
quantities. All the terms there we are going to obtain from the discretized quantities. For
example, if I, if I have to have u i comma of course I, what do you do? The same thing
what you did in your earlier classes and that is what resulted in, for example the B matrix.
Now, B matrix is slightly going to be different, but the concept of B matrix is very
similar, so that this can be written as again N alpha comma I u i alpha. Is it clear? Okay.
Now, what we do is to substitute this into this expression, in order to get a B matrix.
Actually, let us see, let us just do that. Let us see what we get? Is there any question so
far, clear?

Let us write down for each of the node, let us take a three dimensional problem, because
this is a three dimensional element is what we are looking at. Delta E we have written,
written in a column, as a column vector with six entries. So, what we are looking at is a
three dimensional formulation.

(Refer Slide Time: 10:05)


7

For the three dimensional formulation let us say that we have, each node has of course
three degrees of freedom. Let us say that, let us call that as say, delta u 1 alpha, delta u 2
alpha, delta u 3 alpha, where alpha is the nodal degree of freedom. Let us see what we
get. Let us say, look at the first expression here. What I do is look at the first expression.
This is where some doubts were there. From here, from here I substitute it into this
expression. Note that, from here I substitute it into that expression. Note that I is
repeated, so that you will get actually there, what is that you will get? Substitute that say,
F. Let me just take that first term what we had written F
i1
delta u i comma 1.

How do I now get the first term in my B matrix, what we had written in the last class? I
am giving you a clue, u i comma I we had written this as sigma N alpha comma I u i
alpha. So, substituting it, substituting this into this expression, we get actually a big
expression for sigma. For sigma we will get a big expression. Let us, let us just substitute
that, so this here. What are the terms you have? You will have F.

(Refer Slide Time: 12:03)


It is going to be really big; F
11
into delta u
11
delta u
11
. So, that I have to obtain from this.
So, delta u
11
, which will involve alpha 1 u 1 alpha; sigma of N1 u1 alpha. Since it is delta
here, I will get delta u
1
alpha. Is this clear, plus what is it that I will get? Second term,
8

plus F
21
sigma N alpha comma 2 delta u 2 alpha plus F
31
sigma N alpha comma 3 delta u
3 alpha. No, sorry alpha comma 1, 2 1 yes alpha sorry 1; 1 1, because that 1 is constant
here. So, this sigma were summed up by from alpha. So, this will be like N 1 comma 1
delta u 1 plus N 2 comma 1 and so on. It will, it will be a big expression.

In order to, in order to not to write the whole expression we split this B into B alphas; we
split this into B alphas, that say for example, this we split it into a number of alphas
properly aligned where each of them belong to one node. Then comes my first expression
or first delta E term.

(Refer Slide Time: 14:54)


So, that will be, that is first that is B delta E in terms of B alpha, of course sum of them,
can be written as, from here first term F
11
N alpha comma 1, F
11
N alpha comma 1; let me
multiply that by delta u
1
alpha delta u
2
alpha delta u
3
alpha and of course, sigma of alpha,
I have just taken it outside. Then, the second term will be F
21
N alpha comma 1 and the
third term will be F
31
N alpha comma 1. That is the first entry here. Is this clear? That is
the first entry. So, you can, you can imagine how big it is. Now, the fourth entry which is
there is going to be still more bigger; I would like you to complete it, this fourth entry
here, okay. Can I, can I give you two minutes? Can you write down and tell me what it is.
9

Though these are not important from concept wise, but I just want you to understand how
it is written. So, F u i comma 2, so F
11
first term will be F
11
delta u i comma 2, you have
to put N alpha comma 2 that will be the first term plus, plus the second term plus i 2 F
i2
;
note this here, note this here. So, that will be plus F
12
delta u, now again alpha; alpha
comma 1, very good, alpha comma 1, sorry, delta N, not delta u, N alpha comma 1. This
whole thing will be multiplied by delta u 1 alpha.

This will be the first term. Then, there will be a second term, there will be a third term.
The second and the third term will be terms which will be multiplied by the term delta u 2
alpha delta u 3 alpha. This whole thing can be, forget about this; you, you can fill this up,
it is not a problem or I mean you can get this matrix also from very, all standard text
books give this. Look at Zinkevichs for this. But, ultimately what we get is that this can
be written as say B alpha, this alone, one term alone, into delta u alpha. Is this clear?

Now, if you, if you look at this, actually this can be split again. I am not going to do that,
split again into two terms. Actually this has L terms and NL terms or B matrix can be
split into a linear part plus non-linear part. Linear part is the same as what you did in the
previous course and there will be a non-linear part. My next step is exactly the same as
what we did in our earlier classes. What is that we did? We substituted this back into my
expression for the variation, the first variation in order to determine K. So, I substitute it
back into my first expression what we wrote and then let us see what happens for delta Pi.
Substitute that into that expression and take delta u alpha out; remember this was, this
was very similar.








10

(Refer Slide Time: 19:30)


Now, only thing is this alpha I am writing there just to state that, since it is big I cannot
write one B; next step I am just going to write that down, but just stating there, because
sum up, you have to sum up with respect to each of the alphas. S dv minus f
alpha
is equal
to zero. This was exactly the way we had written before and this gives rise, where f
alpha
is
of course, if you want a matrix and f
alpha
is of course the ones which consists of the body
forces and the surface forces. In other words, this is my starting point for what? For the
Newton-Raphson scheme; remember that this is the starting point for my Newton-
Raphson scheme. If you remember these terms, where B transpose sigma, delta u was
outside, B transpose sigma minus f is equal to zero. If you remember this is how we
started Newton-Raphson. Let me write it in the same way as we did in the last, you know,
some time back, where we had B transpose sigma; remember that we had put this as B
transpose dow sigma by dow epsilon and so on.

But right now, I cannot write that next step in the same simple fashion as we did before,
because this B, in fact B if you had noticed in my last session, when I wrote the Newton-
Raphson scheme, B was well behaved, nice guy, it was independent of fs, there was no f,
independent of u. So, I was, I was able to write this as how did we write? B transpose
dow sigma by dow epsilon into dow epsilon by dow u into delta u, so that we had K
T
. If
11

you remember we had a lovely K
T
term without much problems which was B transpose
D
T
B. This is for small strain non-linear problem. This is the K
T
we had. But now, this B
has or is much more complex than what it was in my small strain problem.

(Refer Slide Time: 22:29)


So, let me, let me write down, now that you know what this Bs contains, B contains you
know, B alphas we can write this down as say for example, B transpose S all the sums of
it; I hope you will do the correct this one dV minus f is equal to zero. This is my starting
point for my Newton-Raphson scheme. Now, how do I write this down, what are the
terms that are involved? Let us see what, what these terms are? In other words, what do
you think would be the difficulty now or what are the other terms, other terms which I
have to add than what I had done in my previous course? Let us see whether you are,
whether you are able to say that. Since B involves u and f may also involve u, actually
when I want to calculate K
T
, I will have three terms. One term involving, first term is
very similar to say for example, similar to what we did in the last class that is B transpose
dow S by dow u. That can be written as B transpose dow S by dow E dow E by dow u, so
transpose dow S by dow E dow E by dow u dV. Of course, delta u is there but I am
writing only K
T
plus what is the other fellow? Dow B transpose.

12

Now this, this has u in it. B has u in it. This is how we had written down, because it has
deformation in it, F in it. So, that term also has to be differentiated with respect to u, so
that term will be dow B transpose by dow u S term dV. That is the next term which I will
have and f is not going to be or f may or may not depend upon u. f may depend upon u, I
will come to that in a minute, what it is. So, that term also will be there. We will
investigate each one of them and what they mean. Is it clear, what is the difference
between this and small strain? Yeah, we will come to that in a minute. How does this
term come into picture? We will name each one of them.

(Refer Slide Time: 25:48)


This guy is due to the material contribution and is called as the, if I put K
T
, it is called as
the K
M
matrix, the first guy here. That is the material contribution. The second guy here
is due to geometric stiffness. That is the beauty of this whole formulation. You can, you
can understand each one of these terms. First one is there is a material contribution to
stiffness. The second term, note this, this is due to B which has f in it. This is what is
called as the geometric contributor for stiffness and the last term is due to the load and
that is called as the load contributor of the stiffness. So, you have, because we have split
this P
L
plus P
NL
or in other words, there are non-linear terms associated with B that
13

means that the stiffness matrix has contribution from geometry or geometrical changes as
well. These are what are called as the stiffness terms due to load.

Some times people call this as follower forces, follower forces. How does it change or
why is there, it is a very good question; why does it contribute suddenly to the stiffness.
Let us see that.

(Refer Slide Time: 27:15)


Let us just understand this term first and let us come back to this question in a minute.
Let us first understand this term, what it means. Suppose I have a cantilever beam and I
apply a load. If it is a very small deformation, you would not have even bothered to ask a
question whether the load is going to act in the same vertical direction or is it that this
load will act say, normal to it. If the, if the deformations are small you would not even
bother to do that say, pressure is acting on to a surface. Now, if the deformations are very
small, you would not bother much to look at the pressure, if you had defined for example,
pressure in terms of x direction, you would not define or you would not bother if this x
direction there is a change. But, on the other hand when the deformations are large, the
pressure which say for example, was acting like this, where say for example, this is x
direction or this load which was acting in the downward direction, if this deformations
14

are large here in this cantilever or on this pressure, then there may be a large variation in
the way the force acts. In other words, there was no horizontal component to this when it
was like this, but when the deformations are large, there may be a horizontal component
to it and so on. In other words, the load follows the deformation which means that f terms
varies with u. So, that is the contribution for the stiffness matrix and that is what we call
as follower forces. Yeah, any questions?

Student: f term is outside that integral or it is inside that integral?

Of course, it is outside. B transpose sigma minus f is the, is the side term. But, when I
take dow u dow psi by dow u into delta u for my Taylor series expansion, then dow f by
dow u into delta u will come here. In previous case it was not there.

Student: When we are doing this one, then this term will be there.

Yeah, that is why it is, this is not inside the integral. It is outside dow f by dow u term.
That is what is the contribution for my K
L
term.

Student: Here but, the force always reduces the stiffness of the .. Is there any reason?

Force reduces I am not, I did not comment, yes. Yeah, yeah, but I do not know how dow f
by dow u is going to vary, whether that is going to be negative or positive. So, you cannot
make generalized statements whether the deformation reduces the stiffness or not; very,
very interesting example. Yes, but, but the, but you cannot make very general statement
whether deformations, you cannot make a general statement that deformations reduces
stiffness; it may not.

I will, I will, yeah, I will tell you a physical example; that will make things clear. In other
words, why you cannot make a statement, with deformation, will a system stiffness
decrease? Suppose I have a, I have a sheet of paper. Let us say that we take this cloth.
The cloth is say I am just holding it just like this. No load, no loads are applied. You
15

apply the load on top of it like this. There is going to be some deformation. Suppose, now
I deform this cloth like this. I apply load in this direction and even in the biaxial way,
both directions, I apply a load. Now, this cloth is now deformed. If you apply the load
perpendicular to the cloth what will happen to this cloth? The deformations are going to
be less. In other words, this has been stiffened by stress. It is called as stress stiffening.
The deformations in the other directions had actually in fact enhanced the stiffness of this
cloth in a perpendicular direction. So you cannot, that is why you cannot make a general
statement that deformations reduce stiffness. Is this clear? These are the three terms that
are involved and they are called as K
M
, K
G
and K
L
terms.

Now, let us go to the next step of defining it for updated Lagrangian. How do you define
this whole problem of updated Lagrangian? No, different, very straight forward. The only
thing what I am going to do is transform all the quantities which I had written to the
current co-ordinates. What is the current co-ordinates? Current co-ordinates, yeah, are
small i, but current co-ordinates in this case, is actually not Eulerian, but the converged
state at say t
n
, at say t
n
, time step n. This is the, actually what is called as the current
configuration. In other words, you update, you update all the co-ordinates of the body to
the current or t
n
by adding to the coordinates the displacements, the total displacements.
Let us see how we are going to transfer this. How are we going to transfer all the terms?
Not these terms, I mean you have to go back to my original statement where we had say,
delta E
IJ
S
IJ
. Let us see how we are going to transfer.

Let us take that term. So, what we do? Term by term I am going to transfer. I am going to
carefully transfer this to the current co-ordinate system. So, let us take, because of that let
us take delta E
IJ
, the first variation of E
IJ
. Let us see how you can write that. Of course,
what is it that is going to be useful to us when we transfer it? Deformation, f; very good.
Where are, where are these kinds of things that we did? Note, we did this when we did
Nansons formula. We did the similar thing when we calculated dv to capital DV ratios; d
small v to d capital V ratios, where we did, you know, similar pull back operations and
push forward operations. What essentially we did was to play with the current co-
ordinates and the deformed co-ordinates. Let us see, we will do that. Let us see how we
16

write this for upgraded Lagrangian. Unfortunately, I do not have them on the board. So,
the old ones, so, you may have to write it down from your notes. Let us see how you are
going to write delta E
IJ
.

(Refer Slide Time: 35:03)


Let us do a small exercise. You just try it out, so, delta E
IJ
, we had written this some time
back. How did I write? Half of say, half of delta u i comma j into or I will, I will add this
and multiplication I will do it ultimately; delta u j comma i into F
iI
F
jJ
. Look at this term
what is inside, what is inside? That is nothing but delta epsilon
ij
. So delta epsilon
ij
F
iI
F
jJ

look how neatly whole thing falls in place when we move from the reference to the
current configuration.

Now, let me E
IJ
S
IJ
, what happens to this term? Let us see how many of you are able to
recognize and say something; S
IJ
, the second Piola-Kirchhoff stress, S. So, what is the
relationship between the second Piola-Kirchhoff stress and Cauchy stress? Sigma is equal
to FS F transpose 1 by j or J inverse or in other words, F SF transpose j, F SF transpose is
equal to J sigma. Let us see if you are able to recognize any of them here. Look at what
these three terms in indicial notations means? Fantastic, so that is it. So, this is equal to
dow epsilon
ij
into J sigma
ij
. Is that clear? Now, what is that you do? You go back and
17

substitute for my delta E
IJ
S
IJ
in my previous expression for delta pi. Let us see what you
get.

Why? How come, how is that, how is that possible? See, what we have essentially,
essentially what we have done is to, absolutely, absolutely; please note, yes, very
interesting question. Why, why did I get J ? If I look at the total energy term, yes, but
when I look at the total energy term delta E
IJ
S
IJ
d capital V is what I have there, when I
integrate it. I will have now small dv. If I now substitute it back into my expression, so
that is why I said you have to be careful and they are not blind transformations. Go back
and put this. This is very interesting thing. So, you put this back into my first expression
there, delta E
IJ
S
IJ
term.

(Refer Slide Time: 38:52)


So, you will get delta epsilon
ij
sigma
ij
J d capital V. So, that J d capital V will go now to
small dv. What was omega
0
now, when I now convert this into small dv will go to the
current configuration. Of course you will have, this is my delta pi you will have minus
delta pi external. This is my first variation. The first variation should be equal to zero. Is
that clear? So, you have to be careful in converting every term.

18

(Refer Slide Time: 40:09)


In the same fashion, if I write it in terms of say, dt delta E
IJ
, so, integral delta E
IJ
when I,
when I differentiate it and write it as say, D
T
into dE, in other words, it is, D
T
is the
relationship between delta S and delta E and write this as IJ KL into dE
kl
dV. Let us see,
if you want, this is my first term for my K expression. Say, let us see how you write this.
Substitute this, one by one term and see how you write this. This has to be d delta
epsilon
ij
d
IJ KL
. This is, this is a small catch in what we are doing into delta epsilon
kl
. But
note that, we always define d capital IJ KL or D
T
to be delta S by delta E. That is how,
that is the theory. So, I have to calculate an equivalent d, equivalent d in the current
configuration.

In other words, in other words, note this carefully. In other words, we know how to
transform S, stress, we know how to transform strains; we just now saw it. How do you
now transform the D that is capital D? If I, if I can call it as tangent modulus, tangent
elastic modulus, how do you now transform this, because usually you are given only D
capital. If you look at most of the strain energy forms, all the strain energy forms that you
write say, W is a function of E only. Obviously you cannot write that straight away here.
You have to transform it to the current configuration. Please do that, let us see.

19

You need not do push back pull forward, you know, right now. Actually that is what it
means, but let us see how you arrive at this. J ust substitute it or in other words, how am I
going to write delta epsilon
ij
, if I write say, with respect to small d as the one which
belongs to the, as the elastic modulus for the current configuration ijkl d epsilon
kl
, then
what is this relationship with respect to this? That is the question. Not very difficult; this
is delta epsilon
ij
F
iI
F
jJ
and this guy here is d epsilon
kl
F
kK
F
lL
, substitute that. So, I define,
note this carefully, I define d
ijkl
to be F
iI
F
jJ
F
kK
F
lL
into D
IJ KL
. The only other thing that is
there is dV is equal to J dV and usually that is also included here, so that j, sorry, not here
in the next one here, so that J into J into d
ijkl
is equal to F
iI
F
jJ
F
kK
F
lL
into D
IJ KL
. That is
how you find out; this is very, very important expression. Let me write that. That is how
we find out the elastic modulus in the current configuration.

(Refer Slide Time: 45:07)


J dijkl is equal to F
iI
F
jJ
F
kK
F
lL
D
IJ KL
. No, but we are in the indicial notation. No, we are in
the indicial notation; we are not in matrix notation. Matrix notation, of course, will have
an effect. These are indicial notations that we are doing. To make things clear, so that you
know where the summing up and so on happens, we are doing the indicial notations.
Now, of course, we will have a look at the KG terms slightly more closely in the next
class. So, that also has to be transformed. That term is usually called as G
alpha beta
term.
20

We will, we will look at that thing in the next class, but before that what I want to
comment is, very important comment, that by this time you would have realized, by this
term you would have realized that both the updated and the total Lagrangian formulations
are equivalent, there is no difference between them. They are the same, basically because
mathematically in a very correct fashion we are trying to get from the reference to the
current, the current to the reference and so on. So, no way we miss out anything in this.
The results should be the same. So it is, it is almost meaningless to talk about whether
updated Lagrangian is better or total Lagrangian is better from an accuracy point of view,
from an accuracy point of view. But, at the same time, but at the same time please note
that these guys here reduced to such a nice extent and so close, so close to the linear
formulation, not only linear formulation even non-linear small deformation formulations
that the extension of the code from or implementation of the code from linear to non-
linear is very simple with updated Lagrangian.

Though we know that it is not just that if you put delta epsilon
ij
I will get the current
configurations small deformation code, because there is one more term which we have to
look at that. I told you that we will have a look at it in the next class, but the whole point
is that the only difference or only major advantage for this formulations are that they are
easy to implement. Is there any question? The second term we will expand it and we will
have a closer look in the next class and we will finish the finite element formulation. But
before we close, there is one more comment just coming up in the next class that this is
only a displacement based formulation, but unfortunately, very unfortunately, many of
these formulations may not be valid, may not be valid if you do large strain large
deformation problems.

Very good; there are problems where the problem can be geometrically non-linear, but
the strains can be small, so that we can still use an elastic type of constitutive equation,
but add the geometric part of the stiffness matrix. In other words, this formulation is also
valid for linear elastic case when the strains are small, when the displacements and hence
the geometrical variations are large. So, that kind of problems, there are quite a few
problems that exist. For that kind of problems we can still use this formulation. But, when
21

the strains are large and the displacements are also large, then we have certain problems
which we have to take care of. We will see that in the next class. We will stop here and
continue in the next class.