6

Conservative
Systems

6–1
Chapter 6: CONSERVATIVE SYSTEMS 6–2

§6.1 INTRODUCTION
It was noted in previous Chapters that if a structure and the forces acting upon it collectively form
a conservative mechanical system, the residual force vector r may be expressed as the gradient of
the total potential energy  with respect to the state vector:

∂
r= . (6.1)
∂u

Furthermore, the decompositions  = U − P and r = p − f are related in the sense that

∂U ∂P
p= , f= . (6.2)
∂u ∂u
where p and f are the internal and external forces, respectively, U is the internal energy — which
reduces to the strain energy in the problems considered in this course — and P is the potential of
the applied loads, the negative of which is called the external work function W .
The force equilibrium equations r = 0 or f = p express the fact that the total potential energy is
stationary with respect to variations of the state vector when the structure is in static equilibrium.
Mathematically:
 
∂ T
δ = r δu =
T
δu = 0. (6.3)
∂u
where δu denotes a virtual displacement, δ being the variation symbol. Since δu is arbitrary, (6.3)
implies that r = 0.
If the structural system is conservative there are substantial advantages in taking advantage of that
property:
(1) If discrete force equilibrium equations are worked out by hand (either for complete structures
or finite elements) derivation from a potential is usually simpler than direct use of equilibrium,
because differentiation is a straightforward and less error prone operation, especially as regards
signs. Exercise 6.3 gives an example of this.
(2) The transformation of residual equations to different coordinate systems is simplified because
of the invariance properties of energy functions.
(3) The conventional finite element discretization method relies on the availability of an internal
energy functional.
(4) The tangent stiffness matrix is symmetric. Consequently equation solvers (and eigensolvers)
can take advantage of this property.
(5) Loss of stability can be assessed by the singular stiffness criterion, which is static in nature. If
the system is nonconservative, loss of stability may have to be tested by a dynamic criterion,
which is always more difficult and computationally expensive.
This Chapter introduces the concepts of internal and external potential for systems with finite
degrees of freedom. The presentation is not general in nature but relies on a few simple examples
complemented with exercises. The material is intended to serve as a “bridge” to the formulation of
geometrically nonlinear finite elements, which starts in the next Chapter.

6–2
6–3 §6.2 THE LOAD POTENTIAL

Deformed

;
u

;;; ;
F

Undeformed
;;; No change in load
magnitude or direction

Figure 6.1. Structure under concentrated dead load F

§6.2 THE LOAD POTENTIAL

The concept of load potential is the easiest to understand. This function, called P, is the potential
of the work done by the applied or prescribed forces working on the displacements of the points
on which those forces act. The negative of this potential W = −P is called the work function, but
this function will not be used in the present course.
§6.2.1 Concentrated Dead Loads
For a concrete example, consider a structure loaded by a single concentrated force F that does not
change in magnitude or direction as the structure displaces; see Figure 6.1). A force with these
properties is called a dead load.
If u is the deflection of the point of application of F in the direction of the force, then the work
performed is obviously Fu. Consequently,
P = Fu. (6.4)
If the structure is subjected to n loads Fk (k = 1, . . . n) and the corresponding deflections in the
direction of the forces are called u k , then

n
P= Fk u k . (6.5)
i=1

In general these forces will be defined by their three components along the axes x, y, z and are more
properly represented by vectors fk . For example, if at location k = 3 we have a force F3 acting in
the y-direction,  
0
f3 = F3 . (6.6)
0
Likewise, the displacement of points of application of fk is denoted by vector uk . The vector
generalization of (6.5) is then the sum of n inner products:

n
P= fkT uk . (6.7)
k=1

6–3
Chapter 6: CONSERVATIVE SYSTEMS 6–4

f y (x)
Deformed A B
u y (x)

y
;
;;; ;
x

Undeformed
;;;
Figure 6.2. Structure under under line load f y (x)
(directed upward) over segment AB.

Finally, if all applied force components are collected in the external force vector f (augmented with
zero entries as necessary to be in one-to-one correspondence with the state vector u) then we have
the compact inner-product expression
P = fT u. (6.8)

§6.2.2 Distributed Dead Loads

For distributed forces invariant in magnitude and direction, a spatial integration process is necessary
to obtain P. These forces may include line loads, surface loads or volume loads (body forces).
For example, consider the structure of Figure 6.2, on which a dead line load f y (x) acts in the y
direction along segment AB of the x axis. Then
 xB
P= f y (x) u y (x) d x, (6.9)
xA

where u y (x) is the y-displacement component of points on segment (A,B). A similar technique
can be used for volume (body) forces as illustrated in Exercise 6.1.

6–4
6–5 §6.4 THE INTERNAL ENERGY: HOW GEOMETRIC NONLINEARITIES ARISE

;
; 1
k

2
δ
x

Figure 6.3. Linear spring of stiffness k deforming along its axis.

REMARK 6.1
Substantial mathematical complications arise if some forces are functions of the displacements. For example,
in slender structures under aerodynamic pressure loads the change of direction of the forces as the structure
deflects may have to be considered in the stability analysis. These so-called “follower” forces, which introduce
force B.C. nonlinearities, are considered later in the course. Suffices to say here that no loads potential P
generally exist in such cases and the system is nonconservative.

§6.3 THE INTERNAL ENERGY: A LINEAR SPRING

The internal energy, called U , is the recoverable mechanical work “stored” in the material of the
structure by virtue of its elastic deformation. When this work is expressed in terms of strains and
stresses, as in following Chapters, it is called the strain energy. Note that only flexible bodies can
store strain energy; a rigid body cannot.
We shall illustrate the internal energy concept here by considering the simplest of all structural
elements already encountered in linear finite element analysis: a linear spring of stiffness k,
illustrated in Figure 6.3. Generalization to more complicated structures and structural components
will be made in subsequent Chapters.
If the spring is undeformed, its internal energy U can be conventionally taken as zero (because an
energy function can be adjusted by an arbitrary constant without changing its gradients). Now let
the spring deform slowly (to avoid inertial effects) such that its two ends separate by a distance
δ called the elongation. The internal spring force f¯ for an intermediate elongation 0 ≤ δ̄ ≤ δ is
f¯ = k δ̄. An elementary result of mechanics is that the strain energy taken up by the spring in its
deformed state is
 δ  δ
U= spring-force × d(elongation) = (k δ̄) d δ̄ = 12 kδ 2 . (6.10)
0 0

Suppose that the spring is fixed at end 1 and that end 2 can move only along the x axis, as in
Figure 6.3. Call u the x displacement of end 2. Then δ = u − 0 = u and the strain energy is
U = 12 ku 2 . According to (6.2) the internal force, which in this case is just the spring axial force p,
is the derivative of U with respect to u:
∂U
p= = ku. (6.11)
∂u
This is linear in the displacement u so nothing has changed so far with respect to linear finite
element analysis.

6–5
Chapter 6: CONSERVATIVE SYSTEMS 6–6

Deformed

1(x1 ,y1) ux1 uy1

uy2

y k

x
ux2
Undeformed
2(x2 ,y2)

Figure 6.4. Linear spring of stiffness k displacing on the x, y plane.

§6.4 THE INTERNAL ENERGY: HOW GEOMETRIC NONLINEARITIES ARISE

Now suppose that the spring can move arbitrarily on the plane x, y, as depicted in Figure 6.4. The
position of the deformed spring is completely defined by the four displacement components u x1 ,
u y1 , u x2 and u y2 , which we collect in the state vector


u x1
u 
u =  y1  . (6.12)
u x2
u y2

Let and d denote the spring lengths in the undeformed and deformed configurations, respectively.
The elongation δ is given by
 
δ= d − = ( x + x )2 +( y + y )2 − 2
x + 2,
y (6.13)

where x = u x2 − u x1 , y = u y2 − u y1 , x = x2 − x1 , y = y2 − y1 , in which x1 , y1 , x2 and y2

denote the x, y coordinates of the end nodes of the undeformed spring. Consequently

U = 12 kδ 2 = 12 k( 2
+ 2
d −2 d)
 (6.14)
= 12 k (2 2
+2 x x + 2
x +2 y y + 2
y −2 ( x + x) + (
2
y + y) .
2

6–6
6–7 §6.5 INTERNAL ENERGY: ADDITIVITY PROPERTY

The components of the internal forces are

 ∂U 
∂u
 x1 
 ∂U 
∂U  ∂u 
 y1 
p= = . (6.15)
∂u  ∂U 
 ∂u x2 
 
∂U
∂u y2

The actual expressions of the components in (6.15), which are nonlinear functions of the displace-
ments, are worked out in Exercise 6.2.
The important points that emerge from this example are:

1. The internal forces are nonlinear functions of the displacements, al-

though the spring itself remains constitutively linear. This nonlinearity
comes in as a result of geometric effects, and is thus properly called
geometric nonlinearity.
2. The effect of geometric nonlinearities can be traced to the change in
direction of the spring. Because if the spring stretches along its original
axis the internal force remains linear in the displacements. This change
of direction is measured by rotations.

Even for this simple case the exact nonlinear equations are quite nasty, involving irrational functions
of the displacements. The second property, however, shows that approximations to the exact
nonlinear equations may be made when the change in direction is “small” in some sense. This
feature is illustrated in Exercise 6.3.

§6.5 INTERNAL ENERGY: ADDITIVITY PROPERTY

If the structure consists of m linear springs, each of which absorbs an internal energy Uk , the total
internal energy is the sum of the individual spring energies:

U = U1 + U2 + . . . + Um . (6.16)

This additivity property is of course general because energies are scalar quantities. It applies to
arbitrary structures decomposed into structural components such as finite elements. Furthermore,
(6.16) is not affected by whether the structure is linear or nonlinear.
The last property explains why finite element equations should be derived from energy functions if
such functions exist. That is not, however, always possible.

6–7
Chapter 6: CONSERVATIVE SYSTEMS 6–8

Homework Exercises for Chapter 6

Conservative Systems
Note: the use of a symbolic algebra package, such as Mathematica or MathCad, is recommended for Exercises
6.3 and 6.4 to avoid tedious algebra and generate plots quickly. (There could be a gain from hours to minutes).

EXERCISE 6.1
[A:15] A body of volume V and density ρ is in an uniform gravity field g acting along the −z axis. The body
displaces to another position defined by the small-displacement field u(x, y, z). Find the expression of the
load potential P as an integral over the body if the change in shape of the body is negligible.

EXERCISE 6.2
[A:20] Work out the expression of the internal forces for (6.15). Then extend this relation to the three-
dimensional case in which the ends of the spring move by u x1 , u y1 , u z1 , u x2 , u y2 , u z2 in the x, y, z space.

EXERCISE 6.3
[A+N/C:30] Consider the shallow arch model shown in Figure E6.1. This consists of two identical linear
springs of axial stiffness k pinned to each other and to unmoving pinned supports as shown. The springs are
assumed able to resist both tensile and compressive forces. The distance between the supports is 2L. The
undeformed springs form an angle α with the horizontal axis.
The central pin in loaded by a dead vertical force of magnitude f , positive downwards, which is parametrized
as f = λk L. Only symmetrical deformations of the arch are to be considered for this Exercise. Consequently
the system has just one degree of freedom which we take to be the displacement u under the load, also positive
downwards. The response of this system exhibits the snap-through behavior sketched in Figure E6.2.
(a) Show that the internal energy U and load potential P of the two-spring system are given by
 2
1 1
U = kL 2
− , P = f u, (E6.1)
cos α cos θ

where θ is the angle shown in Figure E6.1, which is linked to u by the relation tan θ + u/L = tan α.

(b) Derive the exact equilibrium equation

∂
r (u, λ) = = 0, (E6.2)
∂u
in which  = U − P is the total potential energy, and λ = f /(k L) is the dimensionless state parameter.
For convenience rewrite this as
r (µ, λ) = 0, (E6.3)
in terms of the dimensionless state parameter
u
µ= . (E6.4)
L tan α

(c) Derive the exact equation for the limit load parameters

∂λ(µ) 
= 0. (E6.5)
∂µ µ=µ
L ,λ=λ L

6–8
6–9 Exercises

u k
k

;; ;;
θ
α

L L

Figure E6.1. Two-spring model of shallow arch.

L1

u or µ

L2

Figure E6.2. Snap-through response of shallow arch (sketch).

(Hint: the exact equation in terms of the angular coordinate θ is cos3 θ L = cos α). Solve this trigonometric
equation1 for the limit-load parameters λ L1 and λ L2 and the dimensionless displacements µ L1 and µ L2
at those points assuming that α = 30◦ .

1 Equation (E6.5) is equivalent to det K = 0 because for a one-DOF system det K = K = ∂λ/∂µ.

6–9
Chapter 6: CONSERVATIVE SYSTEMS 6–10

(d) If the arch initially is and remains sufficiently “shallow” throughout its snap-through behavior, we may
make the small-angle approximations,

cos α ≈ 1 − 12 α 2 , cos θ ≈ 1 − 12 θ 2 , sin α ≈ tan α ≈ α, sin θ ≈ tan θ ≈ θ. (E6.6)

Recast the energy, equilibrium equations, and limit load equations in terms of these approximations,
obtaining U as a quartic polynomial in θ , r as a cubic polynomial in θ, etc, then replace in terms of µ.
As a check, the residual equation in terms of λ and µ should be given by (4.16). Calculate the limit load
parameters λ L1 and λ L2 , and the dimensionless displacements
√ µ L1 and µ L2 at those loads. Verify that
these displacements correspond to the angles θ L = ±α/ 3.
(e) Draw the control-state response curves r (µ, λ) = 0, derived using the exact nonlinear equations and
those from the small-angle approximations on the λ, µ plane (as in the sketch of Figure E6.2, going up
to µ ≈ 2.5) for α = 30◦ .

EXERCISE 6.4
[A+N:15] Derive the current stiffness parameter κ defined in Equation (5.8) for the approximate (small-angle)
model of the two-spring arch of Exercise 6.3. Plot the variation of κ(µ) as µ varies from 0 to µ L1 at the first
limit point, with µ along the horizontal axis. Does κ vanish at the limit point?

6–10