12

The Corotational
Description:
Concepts

12–1
Chapter 12: THE COROTATIONAL DESCRIPTION: CONCEPTS 12–2

§12.1 INTRODUCTION
The corotational (CR) kinematic description of geometrically nonlinear structural analysis is the
most recent of the three kinematic descriptions introduced in Chapter 7 (TL, UL and CR) to be
developed for nonlinear structural analysis. It was developed in the 1980s and has not yet penetrated
the major commercial FE codes.
The key concept of the CR description is the “splitting” or decomposition of the reference config-
uration into two:
1. The initial or base configuration, which is kept fixed throughout the analysis, and serves as an
immovable reference configuration C0 . The corresponding coordinate system is usually the
same for the entire structure.
2. The corotated configuration C R varies from element to element (and also from node to node
in some CR variants). For each individual element, its CR configuration is obtained as a rigid
body motion of the element base configuration. The associated coordinate system is Cartesian
and moves together with the element like a “shadow”. Element deformations are measured
with respect to the corotated configuration. See Figure 12.1.

REMARK 12.1
The idea of separating rigid body and purely deformational motions is an old one in solid and continuum
mechanics. It originally arose in theories of small deformations coupled with finite rotations. Important
applications of these are aircraft and spacecraft dynamics (see next Remark). Because of currently rapid
developments in the subject it is difficult to trace precisely how these concepts came to be adopted in finite
element analysis. A definitive historical account may have to wait until the subject matures and survey articles
are written.

REMARK 12.2
The rigid-plus-deformational decomposition idea for an entire structure was originally pursued in the context
of the dynamics and control of orbiting space structures and aircraft structures, for which it obviously makes
sense. The corotated system follows the vehicle. This approach was systematized by Fraeijs de Veubeke
in an important paper [1]. This article essentially closed the subject as regards the complete structure. The
CR approach takes down the separation idea to individual elements. In that regard it may be viewed as a
combination of the kinematics of de Veubeke’s approach with the idea of subdividing the structure into finite
elements.

REMARK 12.3
Unlike the core-congruential TL formulation studied in Chapters 10-11, a finite element discretization has
to be decided upon before any meaningful equations can be written out for large motions. This is because
the definition of corotated systems depends on the finite element definition. Whether a pure “core CR”
formulation exists is presently an open research problem that remains to be explored in the future. An
important breakthrough in this direction is the “element-independent” CR description discussed later, which
achieves some of the attributes of a core formulation.

REMARK 12.4
The CR kinematics description is occasionally confused with the convected-coordinate (CC) description of
motion, which is extensively used in fluid mechanics. Both may be encompassed within the class of moving

12–2
12–3 §12.2 ELEMENT INDEPENDENT CR DESCRIPTION

Corotated
Current C
Reference configuration
splits into C0 and CR

Deformational motion

Base or initial configuration C0

Rigid body motion

Figure 12.1. Corotational kinematic description concept for a beam element

coordinate kinematic descriptions. The CR description, however, maintains orthogonality of the moving
system(s) thus achieving an exact decomposition of rigid-body and deformational motions. On the other hand,
convected coordinates form a curvilinear system that “fits” the change of metric as the body deforms. Of course
the difference tends to disappear as the discretization becomes progressively finer, but the fact remains that the
convected metric necessarily embodies some kind of deformations. Such deformations are more important in
solid than in fluid mechanics (because Newtonian fluids “forget” displacements). This explains the general
lack of success of the CC approach in structural mechanics.

§12.2 ELEMENT INDEPENDENT CR DESCRIPTION

As noted previously, the CR description is the youngest and is still in an unsettled state of develop-
ment. A consequence of this state of flux is that research groups are pursuing different angles. No
universally accepted version exists. In the present course we shall stress the element-independent
corotational (EICR) description developed by Rankin and coworkers at Lockheed Palo Alto Re-
search Labs [2-5], and perfected by Haugen in his recent dissertation [6].
The term “element independent” is not meant to say that the corotated equations are independent
of the finite element discretization. Rather it emphasizes that the corotational operations of adding
and removing rigid body motions can be modularly implemented as a “front end filter” between the
assembler/solver and the element subroutine library, as sketched in Figure 12.2. This possibility
has great practical importance because it implies that the element library of an existing (linear or
nonlinear) finite element program being converted to the CR description need not be drastically
modified.

REMARK 12.5
There are other element-based CR implementations that are more closely embedded with the element library.

12–3
Chapter 12: THE COROTATIONAL DESCRIPTION: CONCEPTS 12–4

Add Form
Assembler rigid body element
motions stiffness & force

Extract Evaluate
Solver deformational element
motions stresses

CR "Filters" Element library

Figure 12.2. The EICR as a modular ‘element front-end filter’

Few have been implemented, however, within general-purpose nonlinear finite element codes. Exceptions to
this statements are: (1) the “ghost CR” formulation developed by Bergan, Nygård, Mathisen and coworkers at
Norway [7] and implemented in the program FENRIS, and (2) the EICR itself, which has been implemented
in the program STAGS.

§12.3 COROTATIONAL VERSUS LAGRANGIAN DESCRIPTIONS

It was noted in Chapter 7 that the Corotational description is gaining ground primarily at the expense
of the Updated Lagrangian (UL) description. If this trend continues by the year 2000 the TL and
CR descriptions will dominate the spectrum of general purpose nonlinear finite element programs.
From present perspective, it appears that the main advantages of CR are:

(A1) Effective treatment of problems involving large-rotation but small strains. This covers
many technologically important problems in mechanics and engineering because most
structural materials can experience only fairly small strains in service.

(A2) Takes advantage of existing small-strain (linear) finite element libraries, especially if the
EICR variant is used.

(A3) Decouples small-strain material nonlinearities from geometric nonlinearities.

(A4) Is well adapted to the treatment of structural elements with rotational degrees of freedom
(beams, plates, shells) for arbitrarily large rotations. Such elements are notoriously
difficult to treat with the TL description.

(A5) Interfaces naturally with multibody dynamics (MBD) programs.

12–4
12–5 §12.4 INFINITESIMAL RIGID MOTION MODES

Some of the disadvantages of CR, as compared to TL, that may be mentioned are:
(D1) CR is not is particularly suited to large-strain elastic problems because the additional
complexity brought by the dual configurations is not compensated by other advantages.
Whether this statement can be extented to cover large-strain material nonlinearities (as
in forming processes) is not clear at this point.
(D2) May lead to an unsymmetric tangent stiffness matrices for elements with 3D rotational
degrees of freedom if spin variables are used. However, as shown by Haugen [6] post-
symmetrization seems to cause little harm in static analysis.
(D3) Involves difficult mathematics if rotational degrees of freedom are to be accurately treated.
(D4) It seems effective primarily for elements whose initial geometry is simple: 2-node bars and
beams, 3-node and 4-node plates and shells. For elements with more complex geometry,
such as a 3-node beam, difficulties arise [6]. Fortunately such simple elements are the
most commonly used in geometrically nonlinear analysis.
REMARK 12.6
Because of the ongoing developments in CR this list cannot be considered immutable or complete.

§12.4 INFINITESIMAL RIGID MOTION MODES

The central operation of the CR description is the decomposition of the element motion into rigid
body motions (RBMs) and deformational motions. In the EICR version, projector matrices play a
crucial role in the decomposition. To motivate and illustrate the key concepts we begin first with
the consideration of the infinitesimal motion case treated in linear finite element analysis.
Thus, consider an individual element1 with n v nodal degrees of freedom collected in state vector u.
We restrict attention to infinitesimal motions under dead loading about an unstressed configuration.
The potential energy of the element in terms of u is

5(u) = U (u) − P(u) = 12 uT Ku − fT u, (12.1)

where the (linear) stiffness matrix K and the node force vector f are constant.
The element has n r independent infinitesimal rigid body modes or IRBMs. The nodal values of the
m th IRBM is denoted by Rm . For most elements, n r = 1, 3, 6 in one, two and three dimensional
motions, respectively. These modes are collected as columns of a n v × n r matrix R:

R = [ u R1 u R2 . . . u Rnr ] (12.2)

This will be called the infinitesimal rigid body matrix. Now decompose the motion characterized
by u into RBM and deformational:
X
nr
u= Rm αm + ū = Rα + ū, (12.3)
m=1

1 To simplify notation, elements will not be identified by element tags in the sequel unless necessary.

12–5
Chapter 12: THE COROTATIONAL DESCRIPTION: CONCEPTS 12–6

Subspace of Purely
Deformational Motions

u Rα

Total Motion
90 o
Subspace of Rigid
Body Motions
_
u=Pu

Figure 12.3. Sketch of motion decomposition (12.3)

where vector α collects the IRBM amplitudes and ū is the pure deformational motion to be extracted
from u. The latter must then be orthogonal (see sketch in Figure 12.3) to the IRBMs collected as
columns of R, i.e.
RT ū = 0. (12.4)
If the IRBMs are linearly independent, matrix R has full rank. Consequently the square symmetric
matrix RT R has full rank and can be inverted.

§12.4.1 The Linear Projector

Premultiplying (12.3) by RT and taking account of (12.4) we obtain

RT u = RT Rα, α = (RT R)−1 RT u, (12.5)

which substituted into (12.3) yields

u = ū + R(RT R)−1 RT u (12.6)

where as noted above RT R must be nonsingular if R has full rank. Therefore

¡ ¢
ū = I − R(RT R)−1 RT u = Pu, (12.7)

where
P = I − R(RT R)−1 RT , (12.8)

12–6
12–7 §12.4 INFINITESIMAL RIGID MOTION MODES

I being the n v × n v identity matrix. The symmetric matrix P has the property P2 = P and is
therefore a projector. The particular form (12.8) is called the linear projector because it pertains
to infinitesimal motions only. As (12.7) shows, the linear projector acts as a filter that extracts the
deformational motion ū from u. Note also that PR = 0.

REMARK 12.7
The property P2 = P can be confirmed by direct matrix multiplication:

P2 = PP = I − 2R(RT R)−1 RT + R(RT R)−1 RT R(RT R)−1 RT = I − R(RT R)−1 RT = P. (12.9)

The physical significance of P2 u = P(Pu) = Pu is that a projector reapplied to the projection has no effect.
More generally, Pn u = Pu for any integer n ≥ 1.
A important spectral property of a projector matrix is that its eigenvalues are either zero or one. More precisely,
P has n r zero eigenvalues and n v − n r unit eigenvalues. The eigenvectors of the former (“null eigenvectors”
in the usual terminology) span the space of IRBMs.

REMARK 12.8
If the IRBMs chosen as columns of R are orthonormalized, RT R = I, where the identity matrix I is of
dimension n r × n r , and the inverse of RT R drops out. The expression for P simplifies accordingly. Such
normalization is convenient in numerical computations.

REMARK 12.9
The solution (12.5) for α is precisely the least-squares solution of the overdetermined system Rα = u.

REMARK 12.10
Although the projector (12.8) is symmetric, this is not necessarily the case for finite motions. Observe also that
P is not generally invertible because its eigenvalues are zero or one, and its rank is only n v − n r . Physically
this means that the total motion u cannot be recovered from the deformational part ū.

§12.4.2 Example 1: Bar in 1D and 2D

Consider a 2-node bar element of length L 0 moving in one-, two- or three-dimensional space. For
the one dimensional case the element freedom vector is uT = (u 1 u 2 ). The only infinitesimal rigid
motion is u = const = 1, which is defined by u R1 = (1 1). Thus
· ¸ ·1 ¸ · ¸
1 −1
1 1
− 12
R= , R(R R) R =
T T 2 2
, P= 2
. (12.10)
1 1
2
1
2
− 12 1
2

Applying this projector to u gives

·1 ¸
(u
2 1
− u2)
ū = Pu = , (12.11)
1
(u
2 2
− u1)

which is physically obvious.

12–7
Chapter 12: THE COROTATIONAL DESCRIPTION: CONCEPTS 12–8

For the two-dimensional case in which uT = (u X 1 u Y 1 u X 2 u Y 2 ), we may chose u X = 1, u Y = 1

and u X = −Y, u Y = X as three independent IRBMs from which R is constructed as:
1
0 −Y1 
0 1 X1 
R= . (12.12)
1 0 −Y2
0 1 X2

If the origin of coordinates is placed at the element midpoint then −X 1 = X 2 = 12 L 0 , Y1 = Y2 = 0

and
1 0 0 
 0 1 − 12 L 0 
R= , (12.13)
1 0 0
1
0 1 L
2 0

which has the advantage of having mutually orthogonal columns. Consequently RT R is diagonal
(although not the identity) and its inversion is trivial.
The construction of the projector matrix and the treatment of the three-dimensional case are given
as Exercises.

§12.4.3 Example 2: Plane Beam

Consider a two-node plane linear beam of length L 0 with six degrees of freedom:

uT = [ u X 1 uY 1 θ1 u X2 uY 2 θ2 ] . (12.14)

We may again chose u X = 1, u Y = 1 and u X = −Y, u Y = X as independent IRBMs, and again it

is convenient to place the (X, Y ) coordinates at the center of the element. Then
 
1 0 0
0 1 − 12 L 0 
 
0 0 1 
R=
1
. (12.15)
 0 0  
0 1 1
L 
2 0
0 0 1

The columns of this R are mutually orthogonal but not orthonormal.

§12.5 ENERGY INVARIANCE

If one restricts consideration to infinitesimal motions about an equilibrium state, the potential energy
5(u) must be invariant under infinitesimal rigid body motions. In particular the constraint

5(u) = 5(u − Rα) = 5(ū) = 5(Pu), (12.16)

12–8
12–9 §12.6 FINITE MOTIONS

must hold for arbitrary u. Applying this invariance condition to the quadratic form (12.9) we
obtain
5(ū) = 12 uT Ku − uT f
(12.17)
= 12 ūT PT KPū − ūT PT f = 12 ūT Kū − ūT f̄
in which
K = PT KP, f̄ = PT f. (12.18)
These are called the projected stiffness matrix K and the projected force vector f̄, respectively.

REMARK 12.11
If the stiffness matrix K does not satisfy rigid body motions exactly, that is, KR 6= 0, application of the projector
P produces a K that does, because KR = PT KPR = 0. A similar observation holds for the force vector.
This property has been systematically exploited by Rankin and Nour-Omid [4,5] to “sanitize” contaminated
stiffness equations at the element level.

§12.6 FINITE MOTIONS

The study of the previous two sections deal with infinitesimal motions about a configuration of an
arbitrary body. In particular a finite element or assembly of elements. It is therefore relevant to
geometrically linear analysis.
In geometrically nonlinear analysis one deals with finite motions. The problem of separation of
rigid and deformational motions is conceptually straightforward if one is familiar with nonlinear
continuum mechanics. But serious difficulties arise when trying to translate those concepts into
finite element computational mechanics.

§12.6.1 The Polar Decomposition

To see the source of difficulties consider the finite motion of a body in a Total Lagrangian description:

x = x(X), (12.19)

This motion from the reference configuration C0 to a current one C is assumed given. The defor-
mation gradient F associated with (12.19) is easily computed. The polar decomposition theorem
of continuum mechanics says that F has the multiplicative decomposition

F = QU = VQ (12.20)

Here Q is an orthogonal matrix, which characterizes the rotation tensor of each particle. U and V
are symmetric positive definite matrices called the right stretch and left stretch tensors, respectively.
Either characterizes the deformation of each particle.
On first impression the decomposition (12.20) appears as a sensible and rigorous way to formulate
a CR description. The matrix Q describes the rigid rotations, which can be combined with transla-
tional motions, while one of the stretch tensors may be used as deformation measure. And indeed
this idea is used in the Convected Coordinate (CC) description of fluid mechanics alluded to in

12–9
Chapter 12: THE COROTATIONAL DESCRIPTION: CONCEPTS 12–10

Remark 12.4. The key difficulty in solid mechanics is that F generally depends on position X , and
so does the rotation matrix Q. Physically: different particles rotate by different amounts. Because
in nonlinear FE computations one cannot carry along an infinite number of such matrices, some
compromises have to be made.
The second difficulty relates to the computability of Q. In two dimensions the decomposition
(12.20) involves extracting the eigenvalues of the 2 × 2 symmetric matrix FT F, which requires
solving quadratic equations. This is easy numerically and moderately difficult symbolically. In
three dimensions the corresponding problem involves solving cubic equations. This is not difficult
numerically but becomes formidable symbolically. Unfortunately the formulation of CR elements
requires symbolic computations.
The computability problem has steered FE researchers into the use of geometric methods to describe
finite rotations.

§12.6.2 Describing Finite Rigid Rotations

In considering formulation of CR finite element models, a key decision responds to the question:
how should the rigid rotations be computationally represented? Or equivalently: where do we
place the moving (corotated) system(s), and by which rules do those system(s) move following the
motion of the model? Several answers have been proposed and investigated:
(1) Chose one moving system for the entire structure. This is often called a body-attached system
in the literature.
(2) Place one moving system at each node. The choice is called a nodal-based CR formulation.
(3) Place one moving system at each element. The choice is called an element-based CR formu-
lation. This is the only one studied here.
(4) Place one moving system at each element Gauss point. This is obviously a refinement of the
previous choice, but it has not been extensively explored.
Choice (1) has been used extensively in structural analysis, independently of finite elements, because
it is very natural for certain structures. For example, a highly rigid orbiting satellite with some
flexible appendages, or a fairly rigid flying aircraft. The moving system is normally attached to the
center of mass. The determination of the axes directions was systematically studied by Fraeijs de
Veubeke [1]. He showed that the problem is substantially difficult and, worse, may have no unique
solution, if the deformational motions are not infinitesimal.
Choices (2) through (4) are related to the presence of a finite element discretization. The rationale
is that the “area of influence” of each moving system is restricted to one element in cases (3) and
(4) or elements attached to a node in case (2). This localization makes it is easier to meet the
assumption of small deformational motions by refining the FE mesh if necessary.
These general ideas are made more precise in the following Chapter, which studies the CR formu-
lation of a simple element.

12–10
12–11 §12.7 REFERENCES

§12.7 REFERENCES

[1] B. M. Fraeijs de Veubeke, The dynamics of flexible bodies, Int. J. Engrg. Sci., 1976, pp. 895–913.
[2] C. C. Rankin and F.A. Brogan, “An element-independent corotational procedure for the treatment of
large rotations,” ASME J. Pressure Vessel Technology, 108, 1986, pp. 165–174
[3] C. C. Rankin, “Consistent linearization of the element-independent corotational formulation for the
structural analysis of general shells,” NASA Contractor Report 278428, Lockheed Palo Alto Research
Laboratory, Palo Alto, CA, 1988.
[4] C. C. Rankin and B. Nour-Omid, “The use of projectors to improve finite element performance,” Com-
puters & Structures, 30, pp. 257–267, 1988.
[5] B. Nour-Omid and C. C. Rankin, “Finite rotation analysis and consistent linearization using projectors,”
Comp. Meth. Applied. Mech. Engrg., 93, (1991), pp. 353–384.
[6] B. Haugen, Buckling and stability problems for thin shell structures using high-performance finite ele-
ments, Ph. D. Dissertation, Dept. of Aerospace Engineering Sciences, University of Colorado, Boulder,
CO, 1994.
[7] P. G. Bergan and M. K. Nygård, Nonlinear shell analysis using Free Formulation finite elements, in Finite
Element Methods for Nonlinear Problems, Springer Verlag, Berlin, 317-338, 1989.

12–11
Chapter 12: THE COROTATIONAL DESCRIPTION: CONCEPTS 12–12

Homework Exercises for Chapter 12

EXERCISE 12.1
Suppose the linear stiffness matrix of a one-dimensional bar is computed (because of some error) as
h i
1+δ −1
K=k . (E12.1)
−1 1+δ

where k = E A/L is the axial stiffness of the equivalent spring. This matrix does not satisfy the rigid body
motion test KR = 0 if δ 6= 0. By constructing the projected stiffness matrix K = PT KP explicitly, verify
that on application of the projector (12.10), the projected stiffness passes that test and is therefore free of that
defect.2

EXERCISE 12.2
Complete the derivation of the projector P for the two-dimensional bar, starting from the expression (12.13)
of R.

EXERCISE 12.3
Complete the derivation of the projector P for the two-dimensional beam, starting from the expression (12.15)
of R.

EXERCISE 12.4
Construct the 6 × 5 infinitesimal rigid-body matrix R for a three-dimensional bar of length L 0 . For simplicity,
align the bar along the X axis so that the end nodes are located at (− 12 L 0 , 0, 0) and ( 12 L 0 , 0, 0). Do not work
out the projector.
Note: This assignment is somewhat tricky. For, if there are six rigid-body motions in three dimensional space,
how come R has only five columns? Several finite element “experts” have flunked this question.3 Hint:
think about the fact that columns of R must be nonzero because of the requirement of full rank, and of the
infinitesimal “torsional” rotation about X .

2 This is the “RBM filtering” technique rediscovered every few years or so by several FEM authors over the past three
decades.
3 In their defense, most of them were applied mathematicians who know little about mechanics.

12–12
12–13 Solutions to Exercises

Homework Exercises for Chapter 12

Solutions

EXERCISE 12.1

Using (E12.1) and P given by (12.10), the projected stiffness matrix K is given by
· ¸ h i
2+δ −(2 + δ) 1 −1
K = P KP =
T 1
k = k(1 + 12 δ) . (E12.2)
2
−(2 + δ) 2+δ −1 1

For a one dimensional bar, the rigid body motion test, using R = [1 1]T , is
h ih i h i
2+δ −(2 + δ) 1 0
KR = 12 k = .
−(2 + δ) 2+δ 1 0

It is seen that the test is now identically satisfied.

EXERCISE 12.2

 
1 0 −1 0
 0 0 0 0
P=  −1 (E12.3)
0
1
2 0 1
0 0 0 0

EXERCISE 12.3
 1
0 0 − 12 0 0 
2
 0 2 L0 − 2 L0 
 4 + L 20 4 + L 20
0
4 + L 20 4 + L 20

 
 
 0 L0 1− 2 − L0 2 − 2 
 4 + L 20 4 + L 20
0
4 + L0 4 + L 20 
 
P=  (E12.4)
 − 12 0 0 1
0 0 
 2 
 
 0 − 2 − L0 0 2 − L0 
 4 + L 20 4 + L 20 4 + L 20 4 + L 20 
 
0 L0 − 2 0 − L0 2 1− 2
4 + L 20 4 + L 20 4 + L0 4 + L 20

12–13
Chapter 12: THE COROTATIONAL DESCRIPTION: CONCEPTS 12–14

EXERCISE 12.4

The three translational rigid body modes can be formed by inspection:

1 0 0
0 1 0
0 0 1
R1 = Rt X =  
1; R2 = RtY =  
0; R3 = Rt Z =  
0. (E12.5)
     
0 1 0
0 0 1

The infinitesimal rigid body rotations can be expressed by

µ ¶
∂u Z ∂u Y
θX = 1
− = C,
2
∂Y ∂Z
µ ¶
∂u X ∂u Z
θY = 1
− = C, (E12.6)
2
∂Z ∂X
µ ¶
∂u Y ∂u X
θZ = 1
− = C.
2
∂X ∂Y

where the constants C may be taken to be unity for conveniency. Since the bar is aligned along the X -axis, θ X
cannot be formed using the given degrees of freedom. Consequently the “torsional” rotation θ X is invisible,
that is, cannot be defined by the element degrees of freedom.4 The corresponding rigid body matrix column
would be null, which violates the requirement that such columns be linearly independent. Thus the remaining
rigid body motion vectors are those corresponding to θ Z and θY , which are defined by columns R4 and R5
respectively:
 −Y   Z 
1 1
 X1   0 
 0   
R4 = Rθ Z =   , R5 = RθY =  −X 1  , (E12.7)
 −Y2   Z2 
   
X2 0
0 −X 2
Using the given coordinates of the bar nodes, i.e. Z 1 = Z 2 = Y1 = Y2 = 0 and −X 1 = X 2 = L 0 /2, the rigid
body matrix is
1 0 0 0 0 
0 1 0 − 2 L0
1
0 
0 0 1
L 
R= 2 0 .
1 0
1 0 (E12.8)
0 0 0 
 1 
0 1 0 L
2 0
0
0 0 1 0 − 12 L 0
Note: the signs of any column of R may be reversed and its columns may be listed in any order; the resulting
matrix is still correct.

4 Physical interpretation: a pinned bar in a 3D truss can be arbitrarily twisted about its axis without any nodal motions.

12–14