7.

BOUNDARY CONDITIONS
AND
GENERAL CONSTRAINTS

TheSpecificationofKnownJointDisplacements
ReducestheNumberofEquationstobeSolved

7.1 INTRODUCTION

Thefundamentalsofstructuralanalysisandmechanicsasappliedtothelinear
staticanalysishavebeensummarizedinthefirstseveralchaptersofthisbook.
However, additional computational and modeling techniques used to solve
specialproblemsremaintobepresented.

It has been established that the displacement method, where the joint
displacements and rotations are the unknowns, generates a system of joint
equilibrium equations. Bothstatically determinate andstatically indeterminate
structuresaresolvedbythedisplacementmethod.Theglobalstiffnessmatrixis
the sum of element stiffness matrices and can be formed with respect to all
possible joint displacement degrees of freedom. The minimum number of
supports required for a stable system is that which will prevent rigid body
movementofthestructure.
7-2 STATIC AND DYNAMIC ANALYSIS

Thereareseveralreasonsthatthegeneraldisplacementmethodisnotusedfor
noncomputercalculations.Formostproblems,thesolutionofalargenumberof
equations is required. Also, to avoid numerical problems, a large number of
significantfiguresisrequiredifbothbendingandaxialdeformationsareincluded
in the analysis of frame structures. One notes that the two traditional
displacement analysis methods, moment distribution and slopedeflection,
involve only moments and rotations. When those traditional displacement
methodsareextendedtomoregeneralframetypestructures,itisnecessarytoset
theaxialdeformationstozero;which,inmodernterminology,istheapplication
ofadisplacementconstraint.

Ithasbeenshownthatforthedevelopmentoffiniteelementstiffnessmatricesit
isnecessarytointroduceapproximatedisplacementshapefunctions.Basedon
thesameshapefunctions,itispossibletodevelopconstraintsbetweendifferent
coarseandfinefiniteelementmeshesintwoandthreedimensions.

7.2 DISPLACEMENT BOUNDARY CONDITIONS

One of the significant advantages of the displacement method is the ease in

specifyingdisplacementboundaryconditions.Considerthefollowingsetof N
equilibrium equations formedincluding the displacements associatedwith the
supports:
N

Ku R Or,insubscriptnotation K ij u j Ri i 1,...N (7.1)

j 1

If a particular displacement un is known and is specified, the corresponding

load,orreaction Rn ,isunknown.Hence,the N1 equilibriumequationsare
writtenas:
n 1

K
j 1
ij u j Ri K in u n i 1,...n 1

N
or, K u R (7.2)
K
j n 1
ij u j Ri K in u n i n 1,...N
CONSTRAINTS 7-3

Thissimplemodificationtothestiffnessandloadmatricesisappliedtoeach
specifieddisplacementandthe nth rowandcolumnarediscarded.Forafixed
support,wherethedisplacementiszero,theloadvectorsarenotmodified.Those
modifications, resulting from applied displacements, can be applied at the
element level, before formation of the global stiffness matrix. After all
displacements have been calculated, the load associated with the specified
displacementscanbecalculatedfromthediscardedequilibriumequation.This
samebasicapproachcanbe used where the displacements are specifiedas a
functionoftime.

Itshouldbeapparentthatitisnotpossibletospecifyboth un and Rn atthe

same degree of freedom. One can design a structure so that a specified
displacement will result from a specifiedload; therefore, this is a structural
designproblemandnotaprobleminstructuralanalysis.

7.3 NUMERICAL PROBLEMS IN STRUCTURAL ANALYSIS

Manyengineers uselarge values for element properties when modelingrigid

parts of structures. This can cause large errors in the results for static and
dynamicanalysisproblems.Inthecaseofnonlinearanalysisthepracticeofusing
unrealistically large numbers can cause slow convergence and result in long
computerexecutiontimes.Therefore,thepurposeofthissectionistoexplainthe
physical reasons for those problems and to present some guidelines for the
selectionofpropertiesforstiffelements.

Elementswithinfinitestiffnessandrigidsupportsdonotexistinrealstructures.
Wecanonlysaythatanelement,orasupport,isstiffrelativetootherpartsofthe
structure.Inmanycases,therelativestiffnessofwhatwecallarigidelementis
10to1,000timesthestiffnessoftheadjacentflexibleelements.Theuseofthese
realisticvalueswillnotnormallycausenumericalproblemsintheanalysisofthe
computermodelofastructure.However,ifarelativevalueof10 20 isused,a
solutionmaynotbepossible,becauseofwhatisknownastruncationerrors.

Toillustratetruncationerrors,considerthesimplethreeelementmodelshownin
Figure7.1.
7-4 STATIC AND DYNAMIC ANALYSIS

k K k

u1 , F1 u2 , F2

Figure7.1ExampletoIllustrateNumericalProblems

Theequilibriumequationsforthissimplestructure,writteninmatrixform,are
thefollowing:

K k K u1 F1
K K k u F (7.3)
2 2

Most structural analysis programs are written in double precision, and the
stiffnesstermshaveapproximately15significantfiguresandcanbeintherange
of10308to10+308.Therefore,ifthestiffelementhasastiffnessofK=1020k,the
termK+kistruncatedtoKandtheequilibriumequationsaresingularandcannot
be solved. If K=1012 k, approximately 12 significant figures are lost and the
solution is accurate to approximately three significant figures. The equation
solvers used in the wellwritten structural analysis programs can sometimes
detectthistypeoferrorandwarntheuser.However,forlargesystems,thistype
oferrorcanbecumulativeandisnotalwaysdetectedbythecomputerprogram.

This problem can be avoided by using realistic stiffness values, or by using

constraintsintheplaceofverystiffelements.Thisisonereasontherigidfloor
diaphragm constraint is often used in the solution of multistory buildings,
because the inplane stiffness of the floor system is often several ordersof
magnitudegreaterthanthebendingstiffnessofthecolumnsthatconnectthestiff
floorslabs.

Innonlineardynamicanalysis,iterationisoftenusedtosatisfyequilibriumatthe
endofeachtimestep.Ifelementshavealargestiffnesschangeduringthetime
step, the solution can oscillate about the converged solution for alternate
iterations.Toavoidthisconvergenceproblem,itisnecessarytoselectrealistic
stiffness values; or displacement constraints can be activated and deactivated
duringtheincrementalsolution.
CONSTRAINTS 7-5

7.4 GENERAL THEORY ASSOCIATED WITH CONSTRAINTS

Structuralengineershaveuseddisplacementconstraintsinstructuralanalysisfor
overacentury.Forexample,thetwodimensionalportalframeshowninFigure
7.2hassixdisplacementdegreesoffreedom(DOF).Therefore,sixindependent
jointloadsarepossible.

u x 1 , u y 1 , u 1 u x 2 , u y 2 , u 2
u x 1 , u 1 , u 2
R x1 , R y1 , R 1 R x 2 , R y 2 , R 2
R x1 , R 1 , R 2

1 2 1 2
A P P L IC A T IO N O F
3 C O N T R A IN T S
6 DOF 3 DOF

Figure7.2UtilizationofDisplacementConstraintsinPortalFrameAnalysis

Usinghandcalculationsandtheslopedeflectionmethod,itiscommonpractice
toneglectaxialdeformationswithinthethreemembersoftheportalframe.In
mathematicalnotation,thosethreeconstraintequationscanbewrittenas:

u y1 0
u y2 0 (7.4)
u x 2 u x1

Asaresultoftheseconstraints,thefollowingloadassumptionsmustbemade:

R y1 0
Ry 2 0 (7.5)
Rx1 R x1 R x 2
7-6 STATIC AND DYNAMIC ANALYSIS

Note the similarities between the displacement compatibility conditions,

Equation(7.4),andtheforceequilibriumrequirements,Equation(7.5).

Fromthissimpleexample,thefollowinggeneralcommentscanbemade:

1. Theapplicationofaconstraintequationmustbejustifiedbyaphysical
understandingofstructuralbehavior.Inthiscase,wecansaythattheaxial
deformationsaresmallcomparedtolateraldeformation u x1 .Also,theaxial
deformationsinthecolumnsdonotcausesignificantbendingforceswithin
the other members of the structure. In addition, vertical loads cannot be
appliedthatcancausehorizontaldisplacementsintherealstructure.

2. Ingeneral,foreachapplicationofaconstraintequation,oneglobaljoint
displacementdegreeoffreedomiseliminated.

3. Theforceassociationwitheachaxialdeformation,whichhasbeensetto
zero,cannotbecalculateddirectly.Becausetheaxialdeformationhasbeen
set to zero, a computer program based on a displacement method will
produce a zero axial force. This approximation can have serious
consequences if automatic code design checks are conducted by the
computerprogram.

4. Theconstraintequationsshouldbeappliedattheelementstiffnesslevel
beforeadditionofelementstiffnessmatricestotheglobaljointequilibrium
equations.

7.5 FLOOR DIAPHRAGM CONSTRAINTS

Many automated structural analysis computer programs use masterslave

constraintoptions.However,inmanycasestheusersmanualdoesnotclearly
definethemathematicalconstraintequationsthatareusedwithintheprogram.To
illustratethevariousformsthatthisconstraintoptioncantake,letusconsiderthe
floordiaphragmsystemshowninFigure7.3.

Thediaphragm,orthephysicalfloorsystemintherealstructure,canhaveany
numberofcolumnsandbeamsconnectedtoit.Attheendofeachmember,atthe
CONSTRAINTS 7-7

diaphragmlevel,sixdegreesoffreedomexistforathreedimensionalstructure
beforeintroductionofconstraints.
7-8 STATIC AND DYNAMIC ANALYSIS

u(iz) ?
u(iz)
u(iy) (i)
uz(i) u y
uz(i) u(iy)
u(m
y
)
y(i) u (i) (i)
u u(mz) u(ix)
x x
x(i)
m ux(m)

A. Typical Joint I on Floor Systemin x-y Plane B. Master-Slave Constraints

Figure7.3RigidDiaphragmApproximation

Fieldmeasurementshaveverifiedforalargenumberofbuildingtypestructures
thattheinplanedeformationsinthefloorsystemsaresmallcomparedtothe
interstoryhorizontaldisplacements.Hence,ithasbecomecommonpracticeto
assumethattheinplanemotionofallpointsonthefloordiaphragmmoveasa
rigid body. Therefore, the inplane displacements of the diaphragm can be
expressedintermsoftwodisplacements, u (m )
u (m ) ,andarotationabout
x and y
thezaxis, u (m )
z .

Inthecaseofstaticloading,thelocationofthemasternode(m)canbeatany
locationonthediaphragm.However,forthecaseofdynamicearthquakeloading,
themasternodemustbelocatedatthecenterofmassofeachfloorifadiagonal
massmatrixistobeused.TheSAP2000programautomaticallycalculatesthe
locationofthemasternodebasedonthecenterofmassoftheconstraintnodes.

Asaresultofthisrigiddiaphragmapproximation,thefollowingcompatibility
equationsmustbesatisfiedforjointsattachedtothediaphragm:

u(xi ) u(xm) y ( i ) u( mz)

(7.6)
u(yi ) u(ym) x ( i ) u( mz)
CONSTRAINTS 7-9

(i )
Therotation u z mayormaynotbeconstrainedtotherigidbodyrotationofthe
diaphragm. This decisionmust be based on howthe beams andcolumns are
physicallyconnectedtothefloor system.Inthecaseofa steel structure, the
structuraldesignermayspecifythatthefloorslabisreleasedinthevicinityofthe
joint,whichwouldallowthejointtorotateindependentlyofthediaphragm.On
the other hand, in the case of a pouredinplace concrete structure, where
columns and beams are an intricate part of the floor system, the following
additionalconstraintmustbesatisfied:

u( iz) u( mz) (7.7)

Orinmatrixform,thedisplacementtransformationis:

u(xi ) 1 0 y ( i ) u(xm)
(i) ( i ) ( m)
u y 0 1 x u y or, u
(i )
T ( i ) u ( m) (7.8)
u( iz) 0 0 u(iz) u( mz)

Ifdisplacementsareeliminatedbytheapplicationofconstraintequations,the
loadsassociatedwiththosedisplacementsmustalsobetransformedtothemaster
node.Fromsimplestaticstheloadsappliedatjointicanbemovedtothe
masternodembythefollowingequilibriumequations:

Rx( mi) Rx( i )

R (ymi) R (yi ) (7.9)
R( mi
z
)
R( iz) y (i )
R (xi ) x (i)
R (yi )

Orinmatrixform.theloadtransformationis:

Rx( mi) 1 0 0 R(xi )

( mi)
Ry 0 1 0 R(yi ) Or, R (mi) T ( i )T R ( i ) (7.10)
R( mi )
( i ) x ( i ) 1 R( iz)
z y

Again, one notes that the force transformation matrix is the transpose of the
displacementtransformationmatrix.
7-10 STATIC AND DYNAMIC ANALYSIS

Thetotalloadappliedatthemasterpointwillbethesumofthecontributions
fromallslavenodes.Or:

R (m) R
i
( mi )
T ( i )T
R (i) (7.11)

Now,consideraverticalcolumnconnectedbetweenjointiatlevelmandjointj
atlevelm+1,asshowninFigure7.4.Notethatthelocationofthemasternode
canbedifferentforeachlevel.

uz
j
y(j) uz u y

x(j) ux
m+1 DOFat i andj
uz
uy (i)
y(i) i
x
ux m
DOFat mandm+1

Figure7.4ColumnConnectedBetweenHorizontalDiaphragms

FromEquation(7.6)itisapparentthatthedisplacementtransformationmatrix
forthecolumnisgivenby
CONSTRAINTS 7-11

u (xm)

u (xi ) 1 0 0 0 0 0 y (i ) 0 0 0 0 0 0 0 u (ym)
(i )
u y 0 1 0 0 0 0 x (i) 0 0 0 0 0 0 0 u (zi )

u (zi ) 0 0 1 0 0 0 0 0 0 0 0 0 0 0 u(ix)
(i )
u x 0 0 0 1 0 0 0 0 0 0 0 0 0 0 u(iy)
(i )
u ( i ) 0 0 0 0 1 0 0 0 0 0 0 0 0
0 uz
(iy)
u z 0 0 0 0 0 1 0 0 0 0 0 0 0 0 u( mz)
( j)
u x 0 0 0 0 0 0 0 1 0 0 0 0 0 y ( j ) u (xm1)

u (yj ) 0 0 0 0 0 0 0 0 1 0 0 0 0 x ( j ) u (ym1)
( j)
u z 0 0 0 0 0 0 0 0 0 1 0 0 0 0 u (zi )

u ( j ) 0 0 0 0 0 0 0 0 0 0 1 0 0 0 u(ix)
( jx)
uy 0 0 0 0 0 0 0 0 0 0 0 1 0 0 u(iy)

u ( j ) 0 0 0 0 0 0 0 0 0 0 0 0 1 0 u(iz)
z
( m1)
uz
(7.12)

Orinsymbolicform:

d Bu (7.13)

Thedisplacementtransformationmatrixis12by14ifthezrotationsareretained
asindependentdisplacements.Thenew14by14stiffnessmatrix,withrespectto
themasterandslavereferencesystemsatbothlevels,isgivenby:

K B T kB (7.14)

where k istheinitial12by12globalstiffnessmatrixforthecolumn.Itshould
be pointed out that the formal matrix multiplication, suggested by Equation
(7.14), need not be conducted within a computer program. Sparse matrix
operationsreducethenumericaleffortsignificantly.
7-12 STATIC AND DYNAMIC ANALYSIS

Inthecaseofabeamatadiaphragmlevel,theaxialdeformationwillbesetto
zero by the constraints, and the resulting 8 by 8 stiffness matrix will be in
referencetosixrotationsandtwoverticaldisplacements.Therefore,theforcein
thebeamelementwillbezero.

7.6 RIGID CONSTRAINTS

Thereareseveraldifferenttypesofconstraintsthatrequiredisplacementsatone
pointtoberelatedtodisplacementsatanotherpoint.Themostgeneralformofa
threedimensionalrigidconstraintisillustratedinFigure7.5.

u( jz)
u( jy)
uz( j) u(yj)

z j ux( j) u( jx)

u(iz)
y u(iy) m
(i) (i)
u z u y

i ux(i) u(ix)
x
Figure7.5RigidBodyConstraints

Thepointsi,jandmareallpointsonabodythatcanbeconsideredtomove
withsixrigidbodydisplacements.Anypointinspacecanbeconsideredasthe
masternodeforstaticloading;however,fordynamicanalysis,themasternode
mustbeatthecenterofthemassifwewishtorestrictourformulationtoa
diagonalmassmatrix.

It is apparent from the fundamental equations of geometry that all points

connectedtotherigidbodyarerelatedtothedisplacementsofthemasternodeby
thefollowingequations:
CONSTRAINTS 7-13

u(xi ) u (xm) ( z ( i ) z ( m) ) u( my) ( y ( i ) y ( m) ) u( mz)

u(yi ) u (ym) ( z ( i ) z ( m) ) u( mx) ( x ( i ) x ( m) ) u( mz)
u(zi ) u (zm) ( y ( i ) y ( m) ) u( mx) ( x ( i ) x ( m) ) u( my)
(7.15)
u( ix) u( mx)
u( iy) u( my)
u( iz) u( mz)

TheconstraintequationsforpointjareidenticaltomatrixEquation(7.15)withi
replacedwithj.

7.7 USE OF CONSTRAINTS IN BEAM-SHELL ANALYSIS

An example that illustrates the practical use of a threedimensional rigid

constraintisthebeamslabsystemshowninFigure7.6.

Figure7.6ConnectionofBeamtoSlabbyConstraints

Itisrealistictousefournodeshellelementstomodeltheslabandtwonode
beam elements to model the beam. Both elements have six DOF per node.
However, there are no common nodes in space to directly connect the two
elementtypes.Therefore,itislogicaltoconnectnodei,atthemidsurfaceofthe
slab,withpointjattheneutralaxisofthebeamwitharigidconstraint.Ifthese
constraintsareenforcedattheshellnodesalongtheaxisofthebeam,itwill
allowthenaturalinteractionofthetwoelementtypes.Inadditiontoreducingthe
numberofunknowns,itavoidstheproblemofselectinganeffectivewidthofthe
7-14 STATIC AND DYNAMIC ANALYSIS

slab.Also,itallowsnonprismaticbeams,wheretheneutralaxisinnotona
straightline,toberealisticallymodeled.Tomaintaincompatibilitybetweenthe
beamandslab,itmaybenecessarytoapplytherigidbodyconstraintatseveral
sectionsalongtheaxisofthebeam.

7.8 USE OF CONSTRAINTS IN SHEAR WALL ANALYSIS

Anotherareainwhichtheuseofconstraintshasprovenusefulisintheanalysis
of perforated concrete shear walls. Consider the twodimensional shear wall
showninFigure7.7a.

Figure7.7BeamColumnModelofShearWall

A. SHEAR WALL WITH LINE LOADS B. FINITE ELEMENT MODEL

COLUMNS

BEAMS
3 DOF PER
RIGID ZONES RIGID ZONE

C. DEFINE BEAMS & COLUMNS D.BEAMCOLUMNMODEL

Many engineers believethat the creation of a twodimensional finite element

mesh,asshowninFigure7.7b,isthebestapproachtoevaluatethedisplacements
andstresseswithintheshearwall.Intheauthorsopinion,thisapproachmaynot
bethebestforthefollowingreasons:

1. As previously illustrated, the use of fournode plane elements for frame

analysisdoesnotaccuratelymodellinearbending.Theapproximationof
CONSTRAINTS 7-15

constantshearstresswithineachelementmakesitverydifficulttocapture
theparabolicsheardistributionthatexistsintheclassicalframeelement.

2. Ifaveryfinemeshisused,thelinearfiniteelementsolutionwillproduce
near infinite stresses at the corners of the openings. Because the basic
philosophyofreinforcedconcretedesignisbasedoncrackedsections,itis
notpossibletousethefiniteelementresultsdirectlyfordesign.

3. Using common sense and a physical insight into the behavior of the
structure,itispossibletouseframeelementstocreateaverysimplemodel
thataccuratelycapturesthebehaviorofthestructureanddirectlyproduces
resultsthatcanbeusedtodesigntheconcreteelements.

Figure7.7cillustrateshowtheshearwallisreducedtoaframeelementmodel
interconnected withrigidzones. The columns are first defined by identifying
regionsofthestructurethathavetwostressfreeverticalsides.Thebeamsare
thendefinedbyidentifyingareasthathavetwostressfreehorizontalsides.The
length of each beam and column should be increased by approximately 20
percentofthedepthoftheelementtoallowfordeformationsneartheendsofthe
elements.Theremainingareasofthestructureareassumedtoberigidinplane.

Basedonthesephysicalapproximations,thesimplemodel,showninFigure7.7d,
is produced. Each rigid area will have three DOF, two translations and two
rotations.Theendoftheframeelementsmustbeconstrainedtomovewiththese
rigidareas.Therefore,thismodelhasonly12DOF.Additionalnodeswithinthe
frameelementsmayberequiredtoaccuratelymodelthelateralloading.

7.9 USE OF CONSTRAINTS FOR MESH TRANSITIONS

Itisafactthatrectangularelementsaremoreaccuratethanarbitraryquadrilateral
elements. Also, regular eightnode prisms are more accurate than hexahedral
elementsofarbitraryshape.Therefore,thereisamotivationtouseconstraintsto
connectafinemeshwithcoarsemesh.
7-16 STATIC AND DYNAMIC ANALYSIS

Figure7.8UseofConstraintstoMergeDifferentFiniteElementMeshes

Toillustratetheuseofconstraintstomergedifferentsizedelements,considerthe
threedimensionalfiniteelementshowninFigure7.8.

TheeasiestmethodtogeneratethemeshshowninFigure7.8istousecompletely
differentnumberingsystemstogeneratethecoarseandfinemeshareasofthe
finiteelementmodel.Thetwosectionscanthenbeconnectedbydisplacement
constraints. To satisfy compatibility, it is necessary that the fine mesh be
constrainedtothecoarsemesh.Therefore,theshapefunctionsofthesurfaceof
thecoarsemeshmustbeusedtoevaluatethedisplacementsatthenodesofthe
finemesh.Inthiscase,the36DOFofthe12finemeshnodes,numbers21to32,
are related to the displacements at nodes 13 to 16 by 36 equations of the
followingform:

u c N 13 u13 N 14 u14 N 15 u15 N 16 u16 (7.16)

Theequationisappliedtothex,yandzdisplacementsatthe12points.The
bilinearshapefunctions, Ni,areevaluatedatthenaturalcoordinatesofthe12
points.Forexample,thenaturalcoordinatesfornode25arer=0ands=1/3.It
isapparentthatthesedisplacementtransformationscanautomaticallybeformed
CONSTRAINTS 7-17

and applied within a computer program. This approach has been used in
computerprogramsthatuseadaptivemeshrefinement.

7.10 LAGRANGE MULTIPLIERS AND PENALTY FUNCTIONS

In rigidbody mechanics the classical approach to specify displacement

constraintsisbyusingLagrangemultipliers.Amorerecentapproachusedin
computational mechanics is to use penalty functions, within the variational
formulationoftheproblem,toenforceconstraintconditions.

Thepenaltymethodcanbeexplainedusingasimplephysicalapproachinwhich
theconstraintisenforcedusingasemirigidelement.Toillustratethisapproach
Equation(7.17)canbewrittenas:

N 13 u13 N 14 u14 N 15u15 N 16 u16 u c 0 eor,e B c u (7.17)

An equation of this form can be written for all degrees of freedom at the
constraintnode.Thedisplacementtransformationmatrix B c isa1by5matrix
foreachconstraintdisplacement.Fortheconstraintequationtobesatisfied,the
error e must be zero, or a very small number compared to the other
displacementsintheequation.Thiscanbeaccomplishedbyassigningalarge
stiffness k c ,orpenaltyterm,totheerrorintheconstraintequation.Hence,the
force associated with the constraint is f c k c e and the 5 by 5 constraint
elementstiffnessmatrixcanbewrittenas:

k c BTc k c B c (7.18)

Asthevalueof k c isincreased,theerrorisreducedandthestrainenergywithin
theconstraintelementwillapproachzero.Therefore,theenergyassociatedwith
theconstraintelementcanbeaddeddirectlytothepotentialenergyofthesystem
beforeapplicationoftheprincipleofminimumpotentialenergy.

It should be pointed out that the penalty term should not be too large, or
numericalproblemsmaybeintroduced,asillustratedinFigure7.1.Thiscanbe
avoidedifthepenaltytermisthreetofourordersofmagnitudegreaterthanthe
stiffnessoftheadjacentelements.
7-18 STATIC AND DYNAMIC ANALYSIS

TheLagrangemultiplierapproachaddstheconstraintequationstothepotential
energy.Or:
J
1 T

2
u Ku u T R B u
j 1
j j (7.19)

where j isdefinedastheLagrangemultiplierfortheconstraint j.Afterthe

potential energy is minimized with respect to each displacement and each
Lagrangemultiplier,thefollowingsetofequationsisproduced:

K B u R
BT (7.20)
0 0

ThenumberofequationstobesolvedisincreasedbyJadditionalequations.
Equation(7.20)hasbothequilibriumequationsandequationsofgeometry.Also,
the symmetric matrix is not positivedefinite. Therefore, pivoting may be
requiredduringthesolutionprocess.Hence,thepenaltymethodisthepreferable
approach.

7.11 SUMMARY

Traditionally,constraintswereused toreducethenumberofequationstobe
solved.Atthepresenttime,however,thehighspeedofthecurrentgenerationof
inexpensive personal computers allows for the doubleprecision solution of
severalthousandequationswithinafewminutes.Hence,constraintsshouldbe
usedtoavoidnumericalproblemsandtocreatearealisticmodelthataccurately
predictsthebehavioroftherealstructure.

Constraintequationsarenecessarytoconnectdifferentelementtypestogether.In
addition,theycanbeveryusefulinareasofmeshtransitionsandadaptivemesh
refinement.

Caremustbeexercisedtoavoidnumericalproblemsifpenaltyfunctionsareused
to enforce constraints. The use of Lagrange multipliers avoids numerical
problems;however,additionalnumericaleffortisrequiredtosolvethemixedset
ofequations.
CONSTRAINTS 7-19