Professional Documents
Culture Documents
Inmanyproblemsthedecisionvariablesmusthave
integervalues.
Example: assignpeople,machines,andvehiclesto
activitiesinintegerquantities.
INTEGERPROGRAMMING
Ifthisistheonlydeviationfromlinearprogramming,it
Ifthisistheonlydeviationfromlinearprogramming
it
iscalledanintegerprogramming(IP) problem.
Ifonlysome variablesarerequiredtobeinteger,the
modeliscalledamixedintegerprogramming(MIP)
SanFranciscoPoliceDep.problemisanIPproblem.
WyndorGlassCo.problemcouldbeanIPproblem;how?
JooMigueldaCostaSousa /AlexandraMoutinho
IntegerProgramming
Prototypeexample
CaliforniaManufacturingCompanyisconsideringexpansion,
buildingafactoryinLosAngeles,SanFranciscoorinbothcities.
Onenewwarehousecanalsobeconsideredinacitywhereanew
factoryisbeingbuilt.Maximum$10milliontoinvest.
Objective:findfeasiblecombinationofalternativesthatmaximizes
thetotalnetpresentvalue.
thetotalnetpresentvalue
Inintegerprogrammingthedivisibilityassumption
mustbedropped.
Anotherareaofapplicationrelatestoproblems
involvingyesornodecisions,whichhavebinary
g
variables.
TheseIPproblemsarecalledbinaryinteger
programming(BIP) problems.
AsmallexampleofatypicalBIPproblemisgivenin
thefollowing.
JooMigueldaCostaSousa /AlexandraMoutinho
Decision
number
262
Capital
required
BuildfactoryinLosAngeles?
x1
$9million
$6million
BuildfactoryinSanFrancisco?
x2
$5million
$3million
BuildwarehouseinLosAngeles?
x3
$6million
$5million
BuildwarehouseinSanFrancisco?
x4
$4million
$2million
JooMigueldaCostaSousa /AlexandraMoutinho
263
Groupsofyesornodecisionsoftenconstitutegroups
ofmutuallyexclusivealternatives:onlyone decision
inthegroupcanbeyes.
Occasionally,decisionsoftheyesornotypeare
contingentdecisions:decisionthatdependsupon
previousones.
SoftwareoptionsforsolvingBIP,IPorMIPmodels:
1 ifdecisionj isyes,
xj =
j = 1,2,3,4
0 ifdecisionj isno,
Z =totalnetpresentvalueofthesedecisions.
MaximizeZ =9x1 +5x2 +6x3 +4x4.
Constraints:
6x1 +3x2 +5x3 +2x4 10
Mutuallyexclusivealternatives
Contingentdecisions
xj isbinary,forj =1,2,3,4.
JooMigueldaCostaSousa /AlexandraMoutinho
Decision Netpresent
variable
value
BIPmodels
Alldecisionvariableshavethebinaryform:
x3 x1 and x4 x2
Yesornoquestion
BIPmodel
x3 +x4 1
261
264
Excel
MatLab
LINGO/LINDO
MPL/CPLEX
JooMigueldaCostaSousa /AlexandraMoutinho
265
BIPapplications
Formulationexamples
Investmentanalysis,suchastheCaliforniaMan.Co.
Siteselection,offactories,warehouses,etc.
Designingaproductionanddistributionnetwork,or
moregenerallytheentireglobalsupplychain.
Dispatchingshipments,schedulingroutes,vehicles
andtimeperiodfordepartureandarrivals.
Airlineapplications,ase.g.fleetassignmentandcrew
scheduling.
Schedulinginterrelatedactivities,assetdivestures,etc.
JooMigueldaCostaSousa /AlexandraMoutinho
266
Example1
Example1:makingchoiceswhendecisionvariables
arecontinuous.R&DDivisionofGoodProducts Co.
hasdevelopedthreepossiblenewproducts.
Requirement1: fromthethree,atmost twocanbechosen
tobeproduced.
Eachproductcanbeproducedineitheroftwoplants.
However,managementhasimposedanother
restriction:
Requirement2: justoneofthetwoplantscanbechosenas
theproducerofthenewproducts.
JooMigueldaCostaSousa /AlexandraMoutinho
267
Formulationoftheproblem
Productiontimeused
foreachunitproduced
Production
timeavailable
perweek
Product1
Product2
Product3
Plant1
3hours
4hours
2hours
30hours
Plant2
4hours
6hours
2hours
40hours
Unitprofit
(103 $)
Sales
potential
(unitsper
week)
Similartoastandardproductmixproblem,suchastheWyndor
GlassCo.ifwedropthetworestrictionsand requireeach
producttouseproductionhoursinbothplants.
Letx1,x2,x3 betheproductionratesoftherespectiveproducts:
Maximize Z = 5 x 1 + 7 x2 + 3 x 3
subjectto 3 x 1 + 4 x2 + 2 x 3 30
4 x 1 + 6 x2 + 2 x 3 40
x1 7
x2 5
x3 9
Objectives: choosetheproducts,theplantandthe
productionratesofthechosenproductstomaximize
totalprofit.
JooMigueldaCostaSousa /AlexandraMoutinho
268
Formulationoftheproblem
269
Auxiliarybinaryvariables
Forrequirement1,threeauxiliarybinaryvariables(y1,y2,y3) are
introduced:
Forrealproblem,restriction1 addstheconstraint:
Numberofstrictlypositivevariables(x1,x2,x3)mustbe 2
ThismustbeconvertedtoanIPproblem.Itneedsthe
introductionofauxiliarybinaryvariables.
Thisisintroducedinthemodelwiththehelpofanextremely
largepositivenumberM,addingtheconstraints:
Restriction2 requiresreplacingthefirsttwo
functionalconstraintsby:
x 1 My 1
x2 My2
x 3 My 3
Either3 x 1 + 4 x2 + 2 x 3 30
or4 x 1 + 6 x2 + 2 x 3 40musthold.
y1 + y2 + y 3 2
Thisagainrequiresanauxiliarybinaryvariable.
JooMigueldaCostaSousa /AlexandraMoutinho
x 1 , x2 , x 3 0
JooMigueldaCostaSousa /AlexandraMoutinho
y j isbinary,forj = 1,2,3.
270
JooMigueldaCostaSousa /AlexandraMoutinho
271
Auxiliarybinaryvariables
Completemodel(MIP)
Maximize Z = 5 x 1 + 7 x2 + 3 x 3
subjectto
x1 7
Forrequirement2,anotherauxiliarybinaryvariabley4 is
introduced:
1
y4 =
0
x2 5
x3 9
if4x 1 + 6 x 2 + 2 x 3 40musthold(choosePlant2)
if3x 1 + 4 x 2 + 2 x 3 30musthold(choosePlant1)
x 1 My 1 0
x2 My2 0
x 3 My 3 0
Thisaddstheconstraints:
3 x 1 + 4 x2 + 2 x 3 30 + My 4
y 1 + y2 + y 3 2
4 x 1 + 6 x2 + 2 x 3 40 + M (1 y 4 )
3 x 1 + 4 x2 + 2 x 3 My 4 30
y 4 isbinary
4 x 1 + 6 x2 + 2 x 3 + My 4 40 + M
and
JooMigueldaCostaSousa /AlexandraMoutinho
272
Solution
JooMigueldaCostaSousa /AlexandraMoutinho
273
SouthwesternAirwaysneedstoassignthreecrewsto
coveralltheupcomingflights.
Table showstheflightsinthefirstcolumn.
Other12columnsshowthe12feasiblesequencesofflights
foracrew.
foracrew
Numbersineachcolumnindicatetheorderoftheflights.
Exactlythreesequencesmustbechosen(onepercrew).
Morethanonecrewcanbeassignedtoaflight,butitmust
bepaidasifitwasworking.
Lastrowshowsthecostofassigningacrewtoaparticular
sequenceofflights.
274
DataforSouthwesternAirways
1
1.SanFranciscotoLosAngeles
2.SanFranciscotoDenver
1
1
3.SanFranciscotoSeattle
1
1
4.LosAngelestoChicago
9.DenvertoChicago
1
3
2
3
JooMigueldaCostaSousa /AlexandraMoutinho
1
2
11.SeattletoLosAngeles
275
Objective: minimizethetotalcostforthethreecrew
assignmentsthatcoverallflights.
12feasiblesequencesofflights:12yesorno
decisions:
Shouldsequencej beassignedtoacrew?
The12binaryvariablestorepresentthedecisionsare:
10.SeattletoSanFrancisco
12
7.ChicagotoSeattle
2
11
3
3
10
1
2
6.ChicagotoDenver
1
2
8.DenvertoSanFrancisco
JooMigueldaCostaSousa /AlexandraMoutinho
Formulationoftheproblem
Feasiblesequenceofflights
Flight
Cost(1000)
JooMigueldaCostaSousa /AlexandraMoutinho
Example:SouthwesternAirways
MIPproblemwith3continuousandfourbinary
variables.
Optimalsolution:
y1 =1,y
1,y2 =0,y
0,y3 =1,y
1,y4 =1,x
1,x1 =5.5,x
5.5,x2 =0,x
0,x3 =9.
9.
Thatis,produceproducts1and3withproduction
rates5.5unitsperweekand9unitsperweek
respectively,andchoosePlant2forproduction.
Resultingtotalprofitis$54,500perweek.
5.LosAngelestoSanFrancisco
x i 0, fori = 1,2,3
y j isbinary, forj = 1,2,3,4
2
5
2
9
276
1
xj =
0
ifsequencej isassignedtoacrew
otherwise
JooMigueldaCostaSousa /AlexandraMoutinho
277
Formulationoftheproblem
Solution
MinimizeZ = 2 x 1 + 3 x2 + 4 x 3 + 6 x4 + 7 x 5 + 5 x 6 + 7 x 7 + 8 x 8 + 9 x 9
+9 x 10 + 8 x11 + 9 x 12
subjectto
x 1 + x 4 + x 7 + x10 1 (SFtoLA)
x2 + x 5 + x 8 + x 11 1
x 3 + x 6 + x 9 + x 12 1
x 4 + x 7 + x 9 + x 100 + x 122 1
x 1 + x 6 + x 10 + x 11 1
x4 + x5 + x9 1
x 7 + x 8 + x 10 + x 11 + x 12 1
x2 + x 4 + x 5 + x9 1
x 5 + x 8 + x 11 1
x 3 + x 7 + x 8 + x 12 1
x 6 + x 9 + x 10 + x 11 + x 12 1
12
x
j =1
JooMigueldaCostaSousa /AlexandraMoutinho
andx j isbinary,
forj = 1,2, ,12
Oneoptimalsolutionis:
x3 =1(assignsequence3toacrew)
x4 =1(assignsequence4toacrew)
x11 =1(assignsequence11toacrew)
Andallotherxj =0.
Totalcostis$18,000.
Anotheroptimalsolutionis:x1 =x5 =x12 =1.
= 3 (assignthreecrews)
278
Discussion
JooMigueldaCostaSousa /AlexandraMoutinho
279
SolvingIPproblems
Thisexamplebelongstoaclasscalledsetcoveringproblems,
withanumberofpotentialactivities (e.g.flightsequences)and
characteristics (e.g.flights).
Objective:determinetheleastcostlycombinationofactivities
thatcollectivelypossesseachcharacteristicatleastonce.
Si isthesetofallactivitiesthatpossesscharacteristici.
Aconstraintisincludedforeachcharacteristici:
x
jSi
Insetpartitioningproblems theconstraintis
x
jS j
=1
JooMigueldaCostaSousa /AlexandraMoutinho
280
SolvingIPproblems
DifferencetoLPisthatIPhavefarfewersolutions.
IPproblemshaveafinite numberoffeasiblesolutions.
However:
Finitenumberscanbeastronomicallylarge!Withn
Fi it
b b t
i ll l
!With
variablesaBIPproblemhas2n solutions,having
exponentialgrowth.
LPassures thataCPFsolutioncanbeoptimal,
guaranteeingtheremarkableefficiencyofthesimplex
method.LPproblemsaremucheasiertosolvethanIP
LPproblemsaremucheasiertosolvethanIP
problems!!
problems
JooMigueldaCostaSousa /AlexandraMoutinho
281
SolvingIPproblems
Consequently,mostIPalgorithmsincorporatethe
simplexmethod.ThisiscalledtheLPrelaxation.
Sometimes,thesolutionoftheLPproblemis the
solutionoftheIPproblem,suchas:
Primarydeterminantsofcomputationalcomplexity:
1. numberofintegervariables,
2. thesevariablesarebinary orgeneral integervariables,
3. anyspecialstructure intheproblem.
ThisisincontrasttoLP,wherenumberof
constraintsismuchmoreimportantthanthe
numberofvariables.
AsIPproblemsaremuchmoredifficultthanLP,we
couldapplyLPandround theobtainedsolution...
Yes?
Minimumcostflowproblem,includingtransportation
problem,assignmentproblem,shortestpathproblem and
maximumflowproblem.
Specialstructures(seeexamples2and3):mutually
exclusivealternatives,contingentdecisions orset
coveringconstraints canalsosimplifytheproblem.
JooMigueldaCostaSousa /AlexandraMoutinho
Areintegerproblemseasytosolve?
282
JooMigueldaCostaSousa /AlexandraMoutinho
283
Example1
Example2
Minimize
Z = x2
subjectto x 1 + x2 0.5
x 1 + x2 3.5
x 1 , x2 0,
and
MinimizeZ = x 1 + 5 x2 subjecttox 1 + 10 x2 20
x 1 2
andx 1 , x2 0,integers.
x 1 , x2 integers.
JooMigueldaCostaSousa /AlexandraMoutinho
284
SolvingIPproblems
Thesealgorithmswillbediscussedlater.
PureIPproblemscanconsidersometypeof
enumerationprocedure.
Thisshouldbedoneinacleverwaysuchthatonlya
tinyfractionofthefeasiblesolutionsisexamined.
Branchandbound withadividetoconquer
techniquecanbeused.
dividing(branching)theproblemintosmallerandsmaller
subproblems untilitcanbeconquered
conquering(fathoming)bybounding howgoodthebest
solutioncanbe.Ifnooptimalsolutioninsubset:discardit.
MostpopulartraditionalmethodforsolvingIP
problemsisthebranchandboundtechnique.
286
Example:CaliforniaManuf,Co.
JooMigueldaCostaSousa /AlexandraMoutinho
287
Branching
Moststraightforwardwaytodividetheproblem:
fixthevalueofavariable:
Recallprototypeexample:
MaximizeZ =9x1 +5x2 +6x3 +4x4
subjectto
and
(5)xj isbinary,forj =1,2,3,4.
JooMigueldaCostaSousa /AlexandraMoutinho
285
BranchandboundappliedtoBIP
Thus,abetterapproachtodealwithIPproblemsthat
aretoolargetobesolvedexactlyareheuristic
algorithms.
Heuristicsandmetaheuristics areextremelyefficient
forverylargeproblems,butdonotguarantee tofind
anoptimalsolution.
JooMigueldaCostaSousa /AlexandraMoutinho
JooMigueldaCostaSousa /AlexandraMoutinho
288
JooMigueldaCostaSousa /AlexandraMoutinho
Branching
Bounding
Dividing(branching)intosuproblemscreatesatree
withbranches (arcs)fortheAll node.
Thisisthesolutiontree orenumerationtree.
Branchingvariable istheoneusedforbranching.
Thebranchingcontinuesornotafterevaluatingthe
subproblem.
OtherIPproblemsusuallycreatesasmanybranches
asneeded.
Abound isneededforthebestfeasiblesolutionof
eachofthesubproblems.
Standardwayistoperformarelaxation ofthe
g
g onesetofconstraintsthat
problem,e.g.bydeleting
makestheproblemdifficulttosolve.
Mostcommonistorequireintegervariables,soLP
relaxation isthemostwidelyused.
JooMigueldaCostaSousa /AlexandraMoutinho
JooMigueldaCostaSousa /AlexandraMoutinho
290
Boundinginexample
Fathoming
Asubproblem canbeconquered(fathomed,i.e.
searchtreeispruned)inthreeways:
1. WhentheoptimalsolutionfortheLPrelaxationofa
subproblem isinteger,itmustbeoptimal.
Example
E
l :for
f x1=0,(
(x1, x2, x3, x4)=(0,1,0,1),isinteger.
) ( ) i i
Example:forthewholeproblem,(5)is
replacedbyxj 1andxj 0forj=1,2,3,4. Using
simplex:
(x1, x2, x3, x4)=(5/6,1,0,1),withZ=16.5
Thus,Z
,
16.5forallfeasiblesolutionsforBIP
5
problem.CanberoundedtoZ 16(why?)
LPrelaxationforsubproblem 1 (x1=0):
(x1, x2, x3, x4)=(0,1,0,1),withZ=9
LPrelaxationforsubproblem 2 (x1=1):
(x1, x2, x3, x4)=(1,4/5,0,4/5),withZ=16.5
JooMigueldaCostaSousa /AlexandraMoutinho
Itmustbestoredasfirstincumbent (bestfeasible
solutionfoundsofar)forthewholeproblem,alongwith
valueofZ:
Z* =valueofZ forfirstincumbent
IntheexampleZ* =9.
Subproblem 1issolved,soitisfathomed (dismissed).
292
JooMigueldaCostaSousa /AlexandraMoutinho
Fathoming
InSubproblem
p
2thatdoesnotoccur,theboundof16is
,
largerthan9.However,itcanoccurfordescendants.
AsnewincumbentswithlargervaluesofZ* arefound,it
becomeseasiertofathominthisway.
3. Ifthesimplexmethodfindsthatasubproblems LP
relaxationhasnofeasiblesolution,thesubproblem
hasnofeasiblesolution andcanbedismissed.
JooMigueldaCostaSousa /AlexandraMoutinho
293
Summaryoffathomingtests
291
294
or
or
Asubproblemisfathomed(dismissed)if
Test1: Itsbound Z*
Test2: ItsLPrelaxationhasnofeasiblesolutions
Test3: OptimalsolutionforitsLPrelaxationis
integer.
Ifbetter,thissolutionbecomesnewincumbent,andTest
1isreappliedforallunfathomedsubproblems.
JooMigueldaCostaSousa /AlexandraMoutinho
295
Fathominginexample
BIPbranchandboundalgorithm
Initialization: SetZ* = .Applybounding,fathoming
andoptimizationstepsdescribedbelowtothe
wholeproblem.Ifnotfathomed,performiteration.
Stepsforeachiteration:
1. Branching: Amongtheremainingsubproblems,
selecttheonecreatedmostrecently.Branchfrom
thisnodebyfixingthenextvariableaseither0or1.
2. Bounding: Foreachnewsubproblem,obtainits
bound byapplyingitsLPrelaxation.
Resultofapplyingthethreetestsisinfigurebelow.
Subproblem1 isfathomedbytest3.
JooMigueldaCostaSousa /AlexandraMoutinho
296
BIPbranchandboundalgorithm
RounddownZ forresultingoptimalsolution.
JooMigueldaCostaSousa /AlexandraMoutinho
Completingexample
3. Fathoming: Foreachnewsubproblem,applythe
threefathomingtests,anddiscardsubproblems
thatarefathomedbythetests.
Optimalitytest:
y
Stopwhentherearenoremaining
g
subproblems.Thecurrentincumbentisoptimal.
Otherwise,performanotheriteration.
Iteration2. Remaining
subproblems areforx1 =1.
Subproblem 3
((fixx1 =1,x
, 2 =0):
)
MaximizeZ =9+6x3 +4x4
subjectto
(1)5x3 +2x4 4
(2)x3 +x4 1
1
(3)x3
(4)x4 0
(5)xj isbinary,forj =3,4.
JooMigueldaCostaSousa /AlexandraMoutinho
297
298
Example
JooMigueldaCostaSousa /AlexandraMoutinho
Subproblem 4
((fixx1 =1,x
, 2 =1):
)
MaximizeZ =14+6x3 +4x4
subjectto
(1)5x3 +2x4 1
(2)x3 +x4 1
1
(3)x3
(4)x4 1
(5)xj isbinary,forj =3,4.
299
Example
LPrelaxationisobtainedbyreplacing(5)by0 xj 1
j=3,4. Optimalsolutionsare:
LPrelaxationforsubproblem 3:
(x1, x2, x3, x4)
)=(1,0,0.8,0),withZ
(1,0,0.8,0),withZ =13.8
13.8
LPrelaxationforsubproblem 4:
(x1, x2, x3, x4)=(1,1,0,0.5),withZ =16
Resultingbounds:
Allthreefathomingtestsfail,sobothareunfathomed.
Boundforsubproblem 3:Z 13
Boundforsubproblem 4:Z 16
JooMigueldaCostaSousa /AlexandraMoutinho
300
JooMigueldaCostaSousa /AlexandraMoutinho
301
Iteration3
Iteration3(cont.)
LPrelaxation:replace(5)by0 x4 1. Optimal
solutionsare:
LPrelaxationforsubproblem 5:(x1, x2, x3, x4)=(1,1,0,
Subproblem 4hasthelarger
bound,sonextbranchingis
donefrom(x1,x2)=(1,1).
Subproblem 5
(fixx1 =1,x2 =1,x3 =0):
MaximizeZ =14+4x4
subjectto
(1)5x3 +2x4 1
(2),(4)x4 1
(5)x4 isbinary
JooMigueldaCostaSousa /AlexandraMoutinho
0.5),Z =16
LPrelaxationforsubproblem 6:Nofeasiblesolutions.
6 Nofeasiblesolutions
Subproblem 6
(fixx1 =1,x2 =1,x3 =1):
MaximizeZ =20+4x4
subjectto
(1)2x4 4
(2)x4 0
(4)x4 1
(5)x4 isbinary
Boundforsubproblem 5:Z 16
Subproblem 6isfathomedbytest2,butnot
subproblem 5.
302
Iteration3(concl.)
JooMigueldaCostaSousa /AlexandraMoutinho
303
Iteration4
Nodecreatedmostrecentlyisselectedforbranching:
x4 =0:(x1, x2, x3, x4)=(1,1,0,0)isfeasible,withZ =14,
x4 =1:(x1, x2, x3, x4)=(1,1,0,1)isinfeasible.
Firstsolutionpassestest3(integersolution)and
secondpassestest2(infeasible)forfathoming.
Firstsolutionisbetterthanincumbent,soitbecomes
newincumbent,withZ* =14
Reapplyingfathomingtest1(bound)toremaining
branchofSubproblem 3:
Bound=13 Z* =14(fathomed).
JooMigueldaCostaSousa /AlexandraMoutinho
304
SolutiontreeafterIteration4
JooMigueldaCostaSousa /AlexandraMoutinho
305
OtheroptionsinBranchandBound
Branching canbedonee.g.fromthebestbound
ratherthanfromthemostrecentlycreated
subproblem.
Bounding isdonebysolvingarelaxation.Another
possibleoneise.g.theLagrangian
p
g
g g relaxation.
Fathoming criteriacanbegenerallystatedas:
Crit.1: feasiblesolutionsofsubproblem musthaveZ Z*,
Crit.2: thesubproblem hasnofeasiblesolutions,or
Crit.3: anoptimalsolutionofsubproblem hasbeenfound.
SomeadjustmentsnecessaryforBranchandbound
tofindmultipleoptimalsolutions.
JooMigueldaCostaSousa /AlexandraMoutinho
306
JooMigueldaCostaSousa /AlexandraMoutinho
307
BranchandboundforMIP
BranchandboundforMIP
Generalformoftheproblem:
SimilartoBIPalgorithm.SolvingLPrelaxations are
thebasisforbounding andfathoming.
Maximize
Z = c j x j
j =1
subjectto
a x
j =1
and
ij
4changesareneeded:
1. Choiceofbranchingvariable.Onlyintegervariables
thathaveanoninteger value intheoptimalsolution
fortheLPrelaxationcanbechosen.
bi , fori = 1,2, , m,
x j 0, forj = 1,2, , n
x j isinteger, forj = 1,2, , I ; I n.
JooMigueldaCostaSousa /AlexandraMoutinho
308
JooMigueldaCostaSousa /AlexandraMoutinho
BranchandboundforMIP
309
Recurringbranchingvariable
2. Asintegervariablescanhavealargenumberof
possiblevalues,createjusttwo newsubproblems:
xj*:noninteger valueofoptimalsolutionforLPrelaxation.
[xj*]=greatestinteger xj*.
Rangeofvariablesfortwonewsubproblems:
f
bl f
b bl
xj* [xj*]andxj* [xj*]+1.
Eachinequalitybecomesanadditionalconstraint.
Example:xj* =3.5,then:xj* 3andxj* 4.
310
JooMigueldaCostaSousa /AlexandraMoutinho
BranchandboundforMIP
MIPbranchandboundalgorithm
Changesneeded:
3. Boundingstep:valueofZ wasroundeddown inBIP
algorithm.Nowsomevariablesarenotinteger
restrictedsoboundisvalueofZ without rounding.
g
4. Fathomingtest3:optimalsolutionforthe
subproblemsLPrelaxationmustonlybeinteger for
integerrestricted variables.
JooMigueldaCostaSousa /AlexandraMoutinho
311
312
JooMigueldaCostaSousa /AlexandraMoutinho
313
MIPbranchandboundalgorithm
MIPbranchandboundalgorithm
3. Fathoming(cont.):
2. Bounding: Foreachnewsubproblem,obtainits
bound byapplyingitsLPrelaxation.
Test3: OptimalsolutionforitsLPrelaxationhasintegervalues
forintegerrestricted variables.(Ifthissolutionisbetterit
becomesnewincumbent,andtest1isreappliedforall
p
)
unfathomedsubproblems).
33. Fathoming:
g Foreachnewsubproblem,applythe
p
pp y
threefathomingtests,anddiscardsubproblems
thatarefathomedbythetests.
Optimalitytest: Stopwhentherearenoremaining
subproblems.Thecurrentincumbentisoptimal.
Otherwise,performanotheriteration.
JooMigueldaCostaSousa /AlexandraMoutinho
SeeMIPexamplesinPL#7andinpage518of
Hilliersbook.
314
JooMigueldaCostaSousa /AlexandraMoutinho
BranchandcutapproachtoBIP
Automaticproblemprocessing forBIP
ComputerinspectionofIPformulationtospot
reformulations thatmaketheproblemquickerto
solve:
Branchandboundwasdevelopandrefinedinthe
60sandearly70s.
Cansolveproblemsupto100variables.
Branchandcutapproach
pp
wasintroducedinthemid
80s,andcansolveproblemswiththousandsof
variables.
Onlysolvelargeproblemsiftheyaresparse (lessthan5or
even1%ofnonzero valuesinfunctionalconstraints).
Usesacombinationofautomaticproblemprocessing,
generationofcuttingplanes andB&Btechniques.
JooMigueldaCostaSousa /AlexandraMoutinho
316
Fixingvariables: identifyvariablesthatcanbefixedat0or
,
p
1,becauseothervaluecannotleadtofeasibleandoptimal
solution.
Eliminatingredundantconstraints: identifyandeliminate
constraintsthatareautomaticallysatisfiedbysolutions
thatsatisfyallotherconstraints.
Tighteningconstraints: tightenconstraintsinawaythat
reducesfeasibleregionofLPrelaxationwithout
eliminatinganyfeasiblesolutionsfortheBIPproblem.
JooMigueldaCostaSousa /AlexandraMoutinho
Tighteningconstraints
LPrelaxationincluding
feasibleregion.
315
317
GeneratingcuttingplanesforBIP
LPrelaxationaftertightening
constraint.
Cuttingplane (orcut)isanewfunctionalconstraintthat
reducesfeasibleregionforLPrelaxationwithout
eliminatinganyfeasiblesolutionsofIPproblem.
Procedureforgeneratingcuttingplanes:
1.
Considerfunctionalconstraintinformwithonly
nonnegativecoefficients.
2. FindagroupofN variablessuchthat
a) Constraintisviolatedifeveryvariableingroup=1andall
othervariables=0.
b) Itissatisfiedifvalueofany variableschangesfrom1to0.
3.
Resultingcuttingplane:
sumofvariablesingroupN 1.
JooMigueldaCostaSousa /AlexandraMoutinho
318
JooMigueldaCostaSousa /AlexandraMoutinho
319
10
ConstraintProgramming
Statingconstraints
Combinationofartificialintelligencewithcomputer
programminglanguagesinthemid80s.
Flexibility instating(nonlinear)constraints:
1 Mathematicalconstraints,e.g.,x
1.
Mathematicalconstraints e g x +y <z.
<z
2. Disjunctiveconstraints,e.g.,timesofcertaintasks
cannotoverlap.
3. Relationalconstraints,e.g.,atleastthreetasks
shouldbeassignedtoacertainmachine.
JooMigueldaCostaSousa /AlexandraMoutinho
320
Example
JooMigueldaCostaSousa /AlexandraMoutinho
321
ConstraintProgramming
StepsinConstraintProgramming:
1. Formulateacompactmodelfortheproblemby
usingavarietyofconstrainttypes(mostnotofIP
type).
2. Efficientlyfindfeasiblesolutionsthatsatisfyall
ff
f f
f
theseconstraints.
3. Searchamongfeasiblesolutionsforanoptimalone.
Strengthofconstraintprogrammingisinfirsttwo
steps,whereasthemainstrengthofIPisinstep3.
Currentresearch:integrateCPandIP!
Consider:
x1{1,2},x2{1,2}, x3{1,2,3},x4{1,2,3,4,5}
Constraints:
1. Allvariablesmusthavedifferentvalues;
2. x1+x3=4
Applydomainreductionandconstraintpropagationto
obtainfeasiblesolutions:
x1{1},x2{2}, x3{3},x4{4,5}.
JooMigueldaCostaSousa/AlexandraMoutinho
322
JooMigueldaCostaSousa /AlexandraMoutinho
323
11