Chapter 6

Integer Programming: Branch-and-Bound Algorithm

FIRST APPROXLMATION
i\n Integer program is a linear program with the odd! requirement that ull be integers
Chnptor I). Thcrdor<, n fiw op,..axlmatlo to thC! 'olullon o( llJ\Y inl<J<r progrnm may be obwncd
by Ignoring tbe integer roquinunent and ;olvlng the resulting lin<llr program by one or dtc
ulready presented. J( the optimal so.lution to the linear program happens 111 be integcol. Uten thi> !olution
is also the optimal solution to the odglnal Integer program (see Problem 6J). Otl>erwi!C- and this is
the u>unl situation-one may round the components or the first approximauon to tb< nearest ICII5ible
onteger; and obtain n oppro:c:immioo. This procedure is oRen carried out, cspo<:ially whcn the first
r.approximation involves In rae numbers. but it-can be inaooum.te wheo Lhe numbers are smnll (see Problem
6.5)
llRANCHINC
I( the first llpproxtmution contuin u vurinble that not int.ogrnl. say x;. then 1
1
< xj < 1
1
, where
t
1
and i::. ate consecutive. nomJC"g.J tivc in-tegers:. Two nc w intcg.:r programs then crc.Llcd b)' IIUgmenting
the orlginlll intOJ!<r program Wtth clther the conotrnnnt t
1
1
1
or the
1
2 1
2
, Th11 proccs.<,
call! brmtd t{flg, Ita> the dTecl ol sltrinlung the leas ible region in way tb3t rltnunnt .. !'rom Curther
consideration the current nonintegrnl solutjon (or .<
1
but sull preserves pcnsibk integral solullons to
the origmal problem. (Sec Problem 6.8.)
6. 1 As a tim upproJtimation 10 the ioJeatl' proaram
max.smuc:o .: = 10-"
1
+ 'C:
tubjtcl 10 2.t
1
+ Sx, S II
Wltb: x
1
and .c.
1
nCtnnepuvc 1uld
(6./J
we mrt>ider tht llQallt<d lincor progrAm obtain! by deletlna the lntesral noqulremenL Uy the W>luli0<1
i rtldily lound to be Jl XI 0, with ; .. SS. Sil'IC<' S < .t; < 6. brnehtnl ""'"I$ the twg new integer
progronu1
maximi:z.c. : = 10.'('
1
+ x,J
.!IU.bjOJI lO; + SX.t s I I
., j
(6.1)
.11'
1
llnd t",a nonnranfh+e lind integra)
= l0.'C
1
+ .x,
A\lbjc..:l tO: l..t I + 5;t; .l II
x, 6
With '
1
and nnd 1Rt,Crtl
For the two Integer programs aeatcd by tbe bnt nchmg process. first approxomationo are obtained
by again ignoring t he mtegcr requirements and sol ving thr resulting linear proynms. If either 6m
124
CHAP. 6) tNHGER I'ROGRAMMlt-:G: BRANCHANO.BOUND ALGORITHM t25
approximation is slllll nonintegrnl, then the iillegtl" proyafil whicb rise to Lllat first a)lproxlmetion
lx't'Om<-s u candida1c for furlher branchins.
Exomple 6.2 li<rng grophiCO\l methods. we find 1ha1 t6.1) has the first approximoti"n x; : 5. xj: 0.2,
with : 50.2. while.,.-ogram (6J) has no fc'Jsible solution. ThU!O. program (6.1) is a Cll.ndtdaie for lunher hranchin&-
Sincc 0 < 1:t < I. '-""t &ug.mcnt (6 . .?) ,.ith either .t:; s 0 or x,. I. and obtain lhe two new programs
(in .,- 0 I> fonctd) and
m:uhnltt; : 10.T
1
+ "':
to: 2x
1
+ Sx, s I I
s s
... , s 0
with: x, and xJ nt..1nnC"gathe ::tnd inttgrul
11'lll.rimac:: ! = ( 0x 1 + ,"(;
!IU.bjtct lO' l.\'t + St"J s ' 1
,,
s 5

with; .\:
1
and x
1
and integral
(6.4)
<o.Jl
\Vtth tM unegtr rtqu1rcmenu rgnored. the solution lo proiram ii xt =- 5 . cl = 0. wifb : =.50. while tbl:
1(1lutil>n 10 pr0,8rJm (6.$) l, x: = 3, " ; I, ..vhh : .s 31. Sinoe both lhese first appro)lam.ations are integrat oo
fun her brnnC'bing is required
BOUNDrt'G
Assume that the obje<.tive functJon is to be ma:clmlzed. Branchmg continues uo-til an first
approximation (which is thus an integral solution) is obtained. The value of the objective for this first
imegral solution bec-omes a IQwer bound for tile pr<lblem. ond all progmms whose first approxinunions,
integral or not, }icldl values of 1he objective function smnUcr thnn tile lower bound are discarded.
Exomplo &.3 PrOMnm (6.4) pos...scs on int<g111l solution wilh : ' =SO: hen S(l be"'""''' a 00\ond lor the
probltm. Prog""m (6.Sl has solution wolb : c 31. Since 31 is less tluln 1he hound 50. profP"'Im (6.5) is
from funber "ml wauttlltant be.en JD t'lltn if itl first ap(Noxi,tatiQn htltl bttm

Branching continu<l' from tlto<e programs having oonimegral first approximatloM that give values
or tho objective function greater than the lower bound. If. in the process. n new integral solution is
uncovered having D value or the objective function thao t he current lower bound. then this v:due
or the objective funcdon becomes the new lower bound. The progmm yielded the old lower bound
i:s elimlnaltd, L\5 :s.rc: till programs whose first approximations g.ive values or the objecti'Yc3 function smaller
the MW lower bound. The branching process contJnucs untU there a.rc no progtallls with nonintcgral
first o.pproximations remaining unde-r AL this poiot. the current lowerbou.nd solution is
the opumal soluuoo to the orii,Zinal integer program.
1 the objective function is to b<: millimized, the procedure remains tho same, I hat upper bounds
arr used. Th"' the ' "Jiue of the fir$1 intcgml olution becomes on upper bound for the problem, and
programs are eliminated when heir first npprox.imtue tva1ues are grt'ater than the cunent upper bound.
126 INTEGER PROGRAMMING: BRANCH-AND-BOUND AI.GOIUTHM (Cit"P. 6
COMPUTATIONAL CONSIDERATIONS
Ont alway' branches from thot progrm whi<h appears moM ocarly oplimol Wlipn th<>re arc "
number of candidates lor lunher bnltJChing. ont chooses that having the :-value, Sf the
function is to be ma.<intited. or that havmg tht smallest :-value, if tbc fuqctloo is to be
minirniud.
Additional constraints are added one at a ume. If a fint involves more thao one
nomntegral vanabk the new comtraints are imposed on that variable which is furlhesr;,ltrom being an
integer: i.e . that variable whose fractional patt is closest to O.i In case of a tie. tbt !!Olver arbitrarily
chooses one or the urlablcs.
Fin..Jiy. it i& poS$1btc for on intc.ger program or an nsscx:i"tcd linear progam to h#vc more than
one optimnl solutiou. In both caOS, we odbcrc to tbe com-cn!lon adopted m Chaptcf I, orbunmly
designating one of the 50iutiom M tbe optimal one and disregarding the rest.
Solved Problems
6.1 Draw a ;chcmattc diagram (tree) depicting !he results of Examples 6.1 6ll.
Sec Fig, 6- t. The original inteaer program. here (6./ ). is dc"gcu.1t!d by a clr-:led I, and progrtlmt
ft:trmed th!oug.h br1nching art dc:lianaltod io the order of their r:ttation by circled s:ueceu-1:, integt",.. lhLS.
progrum (6.:') through (6 . .f) BI'C designated by Clfllled 2 through 5, r11pCXthly, fh ( .. t approximate
solution to each program 11 wntten by the circle dao.BJlltlng the p<ogn.m. Eeh cir<lc (_l!osrm) Is th<n
connected by ullne to that elrole(proJ11am) wht<b generted it vla the bmxbio& ptoedS. l '!oowcoru.trttint
1huc dd1otd the btanch 1$ wrinen abovt tht lu$t .. f'i.aatty, a luge cross b drllwll lhrou a orde- If rbe
program hu been eUmill3ted Crom funber ootuiderarion. ltenl%, branch dlmioar!d
bcc:au.:;e h not branch S ,..;u dtntin.atcd by boundJnt in. Example 6.l. Sl "" oo
nonintegm1 brancbcsleft to consKJtr. tbe scbematitt diayam india:ues lbltl progrtm I sul'fed w.t.h .l:f - '
.x; = 0. and .: .... SO.
maxim.ize: : - l x
1
+ 4.t
1
subject ur. 2x
1
<, S 6
Jx, + J,<, :> 9
with; x, and x, nonnegative and integral
CHAP6] INT0ER PROORAMMINO BRANCH ANO.BOUNO ALGORITHM 127
Nqlc-clrnalbc rniC'JOf :r<qUlmMnl. we obulll r 2.!$, J,j. 1lll : IUS. as lbc 10
lbc auocr11cd bncar J)O)IT2JD. Srnct: t lt furtbn from an "'''"' dun x:, c - u 10 l<llmllc lbt
btartcll and '::!: 2.


mluoml: =-h .. .. ,1 mutmru: :- lx
1
.,.
tubjt 10. lx, ..
6 \Ut>,cclto: 21<, .. x, 6
2.t, Jf, s 9 2.<
1

' S I
AJ 2
Wttb Y I t .\ J ftOn.Mp ltYC:
.. . ,h: x,. :r
1
nonncpuYC
and lnt<pl IUid
Tbc fuJI 10 Propwm 2 " c; 1S. 1<$- I. 1th : 11.5. lhe fint to
f'tosram l "' t ; I S. cj 2. .. uh : 12.5. TheM rcso;lt} arc thown 111 rtJ. 6-2. Stilet Ptopams 2
3 both ha>c nonrntqnllll:Sl "c COtlld bnii1Cb from atllc:r one"'" ehOC* ProJtllDI)
'' b4) the la!JH IIQII y Oplnml) >illuc ot the ot>,cclrve lunc11on Here I < xr < Z. oo tbt !IN Pf'OIIIliiiJ
IU'C
,.,,...,
h-z,_. j
marrw:r: : )x, + -4.r, lft.Uli!IJlC : lt,- . .. ,
iUb.IC'CI 10; b, + .... 6 .oubjcciiO; 2.1<, s 6
h, t l'<, s 9
x, l!: 2
-'t l!: l
... S l

:tl
,.,th
' ' nonncpti>"C
wnh.
'' .t l ftOftntJalJ "C
aftd lnlqral llDd rntqnl
Tbcl'< It no JOiutlon 10 I'Top111m S (it Is " hll" the 110lution 10 Propm V!llh the rnt<S<r
coMIBillu lano;..S " x:- I, ;a J 1/3, "'ith : 12..33 See Fl 6-l. llw! can continue fro;n
<Jtbtr Provam 2 or l>ropatn 4, we choooe 4 It lw lbc Jfdltcf : ""lue. Ucrc 2 < " f < 3. .a
, . II.S
128 INTFOF.R PROORAMMI '<l 8R<\N('IIMO.B0l Nl> AlGORITHM
Pr.xrem 6
r'IUUm&lC ! }\ 1 - ha
l<, ': s 6
!,, '':
"(.1
Sl
,, s l
' ": nonl>t"fih
nd lntccral
,...,.,.,.,. -
:- h
1
_.. h-
1
II> I \J S (>
:!,, ....

'
.Sl
\ >
tth ' \: MnM,JU\C
nd ont<JII
[("liM' 6
Tlluolutklft 10 Prosr= & "lll11bc m<J<rc..ntnoonn tw<-..1 .. 1. ; - 1. ub : 11 St...,. th"
It ID IIIU.,ai 0011111011. : II btcoma I ..,...., b.JIUIKI !.tr J'I"Obkm .. Ill) J'I"OJIIM I ; lwc
wallet tlwl II Ill bdcdonh be dumlutcd. The fim w Ptobkm Ito , 1 0. '; \ .,.otb
: 11 -dllusan wcp2holouon "uh 1 lmo<T 1\>oeoc
tk - ,.,,.., bound. 5nd the profl'llm th.lt 1\'nm>lal tht old ""'"'' b<.>omd. Pt>Jr<UII 6. " dtlnt""ud (r<>l!l
(11J1htt -<tdntoon. &> IS PrOCi&m l Fopn 1>-2 II"" <110'"> *' lutxbc< ldt 10 a>n\ldn C>lhtt than 111<
- CCtnC$pondlliJ 10 thc C'llm."OI lo-'<f bollnd C'o1l>al,_ll) , tlno the >01111- 10
: 0. <! l ... uh:' -1!
U Soh( Ptoblcm I .9
Dmpp1111111< IDI<jCf rcqUimnctlh (nlm J'l"l>lfhiM 1/) uf ProNcm 19. "'<,.\lit "'"'"'Uit\1 hMJt
J!fOJram Rnt, to find Problem S 4), '1 2. i 18. ' 0. \ l :!0. : 0. : nh : 4S.
Tllb <> Jbc lint app<O>II!Llllon SoMe ot " lntcl .,.,, .... , 11 h tlw tht o>pum.;l Sl'lUtl<nl to 111< C>rlJ1nal
ptOJfllll\.
SA Sohc Problem 16.
lJ,DOnll&lhe IDtC&<"f RqiUICID<JIIi Ill JlfDSrllm l-It ol Prot>lan I 6. ,.. -+t>on ' j 1 l 0 I 1-1>.
\ I "llh : - 55000. ti the liB! &I'PfOlll'IWtiOil St- t ,. 001 rntop;.lt lontxh JC t,.o -
and sol<e t.ICll ,.lib tht ant.,...- Cl'ft\lnunt> tlaori The tc.lllb on.IA.-.....S 111 P II' Ptop-am
J pen_,... tnkJfal .-.l11uon 1th 1 !''lhat rrcnn than the :ul..r ol Prottm ! ("o>ttwq11C11LI) l't
clrmtn.ttt Prop-am ! and 4ICCJClll the >OIIIuon to Prl)(lr= J the unc ' ; I. '; ll. , ; 161! 1.
( ! 4999. lt.Jth : - usooo.
1.5 t.hr cri"'f'\ tnohcd n rouodtn& 1he iint to the pRIJI.lm tn
Problcmt6.= and 6 4 to tnlcgcn and lhcn l.t\.1111 am""'"' .... the uptunal a.>nei
CHAP. 6) IKTEGER PROGRAMMI NG: JlRAt-ICHANDBOUND ALGORITHM 129
6.6
fi rs uppro,imatlorn in Probkm 6.2 was =-2.25 .xi - LS. We wi(b 10 round 10 the do,sest
pOint in tlrr.fNsiblt rl1)i01t. Now. or rhe: rour integer points surrounding tbe first approximation. only
one, Cl. I). 15 round the feasible region. Thus we lakt r - 2. " ) - I, with a : = 10,
the proposed C\palmzd sdu1roo The tnac optimal $01u1km was found ali ; = thUJ the rounded $0lu1ion
dC"vii!IU."J (rvm lhr true S\)hllit>n by more lhDTl 16 perct'Jll.
Th< fi"t npproxim1uiun in Prob1c:rn 6.4 w1u r f -=-- .-.: ; - 0, .xf - .'(: - S(k10. Rounding .\'f doy,n.
ltJ rem am we obtain xt '-' f 0. xj ... !666, t l SOOOa& the toorditunes altho npiimlll
:lOiulwn. nu: OOf'NJ\p<mding SS4996. de\illldl rrom lhc t rue f.<,I Uil Of't, : - S55000. by thAU
0008 pr<:GI11
minimize: :. =- x, + x,
subjc<t lo: 2.<
1
+ 2.<
1
5
12x
1
T 5.<, :S: 30
with: x
1
and x, nonnegative and imcgral
A first approxim:ttloft 10 this program ss ;( T = LS, = 0, whh s:: = 2.5. Rounding .x: up. thereby
we 3, x; = 0. wilh : =- 3. an cstimaae or the optamaJ soluuon a.o I he ori&inal
prc:'s,rrJm. Obstrvt. hov. cver. chat for inrtg.rat value,; or the \'"itriables. the (unction must iiS(I( be
ln1csrnl. The ror the apprtlxirnaaion .: .. ::o: 1.5, pro,ides a Jowl!'r bound for 1he optimal objo:til'e:
oonsequcotly. the Qptimal objectiv-e c:annoa be sma1Jer than 3. Since we an es1im.ate whiC'h tutains the
vA! ut. 3. atb: esuma1c muSt be optlmat. i.t .. l . co 0. wilh : - J.
6.7 Solve the ktltlpsuck prubiHIJ formulated in Problem 1.8.
The simple' me1hod rou!d b< US<d 10 find tb< [trSI pproxnnotlon for I)<Ogram (]) of Problem 1.8. A
more cfficicnr pro....--edurt' l! the
The crttiad (actor tn determining whether an rtcm IS taken ts nol its weight or v;lluc: per se but lhe
rrttio of the t.,.,:o, value pou.nd. We derM)u: this factor as de.Ji,obiliry, adJoin n to the data. and conS!roc1
1'3ble ft. r. where: the: hems ttrc listed m order of desirabiUty. Tu QbUno tbe solution to
the knars.ack. probkm 'w\'l th I he integer constralnls Jgnortd, we take .as much ar eadl itan as possible
1-..ithout <x<:din& the wcighl linHI), beginning wi1h the mosl deslrabk. It follows from I 111111
the aprrt>xim:uion COMi:'t't In aU of item 2 (I he mosr defirable <"nt-). aU Q( item S ( t ht- next most desi,rable
item). Ulld 30 lb ol item .l: xf = O t j = I. xt = 3();35. x! 0. "f I, "ith :- = 135
ltc:m lb Value O..imbiiG),
'11luc lb
2l 60 U l
$ 7 IS 2.14
J 3S 70 2.00
I 52 100 1.92
IS IS 1.00
thiJ: lirst appJo:dmatktn i$ n<'rUn\cg.ral, WC' braoch b.)' ausmenting rhe oripnal with
drher x, 0 or x, >. I BdClre doing so. howe\er, we note that s.iocc Y, is requiced ro be the
cons1min1 \:,t '=' 0 can be tl8htClled 10 f .t 0: a1ld Uth.'C a1 most one or on will be raken. the coru.truiot
I C:3P bt lightened ta 'C'- 1. This indicated in the trc.: diqgram, Fig. 6-4
Dropping the integer requircmem.lli. we f.k1crminc lbe op1im"l 1otudon' Ul bclth Prng.am' l ond l in
Fig. 6-4, \Ding l nble 6 1 10 find tb(' be:tt n'i ooruL'itenl 1he conntiiinu.. For Ptogtanl 1. w-t obtain
' ; .. J0/52. l. xj Q, ,.! 0. .:x; 1. with= - 112.69. and fur Progrnm 3, -0, \:! - l . Y1 l,
Y! ON Q, \'; W 2f7. wiab 134.28,.
130 INTEOER PROGRAMMINO BlANCH-AND-BOUND ALCORITliM (CHAP b
J' .- 119.61
Flc.li-4
Con1inulng 1he bn>noh.ond-bound Pl.,..... we oompl<1e Fia. The flnllnovaJ i01ullon I ob .. in<
an Prog.r:tnt S.. "'-;th 90.. A SC'COnd inteJna.llt)l.uti.on k obtained In Proaam 10. ith UO. Si* th.lt
tcoood :,value ft l:.rgcr than the fiat, we el.ttnift:.tc Progl'*:m fh u wdl Ptogntnu 9 11nd II Pto&ram s.
't& :vaJuc than the 1.!-tttrt:nl lowc:J bound. .co th:t.l we n'I UI.I tiUI hracp!!h from lL Tbe
n:>ui1111J! Prognm 12 h .. 1 : -volue <lnllller lh>n I J(), wbil Progrnm IJ il infdau'ble; h<ni:c they loo ""'
c:limlnatC\J We f'C(ftllln Y.'tlb only Program 10: thcrdoh:, iu .aolution-takc only itC1:111 2 j:jDtJ l . Cor a Ultll
value cr llO IJ the opuroal JOiutlon.
ttt uch of the bninet1-antlbdund olght hllve been 4't'01drd Wr: tmo'*' In advan" tha1 either
0 or x,, l in the apbtrutJ If .x, o. thef' Pro_grum 2 oonddr.ll with thr original propam.
ttnd lbc- : -v:tluc: 132..69 obttuned '&hell the requuernentil 1te dropped (lhettby ap:an41na the fc.uiblc
tt.gion} mUit be treata tb.tul, or aJ IQ.\1 cquill h\ the true: optimum. SlmUady, ll x
1
t, we see frOLn Prnsram
l thlat the true op11mum cllnnol exccod I 3428. Whkhcvcr tbc CIUIC, we .&rt wumi Uuu the IJ\Ie optimum
tS len than Ill. for lntegrlll vulua ol the vu.riublcs.,: i i.ntcgral; in (lb.'t, it b 1 multiple of S. inac the
values of tho tttmi art multiples or 5. Thudore. the true optunu.tn ,, A1 r'nMI 130. Now, rc,)bndin& 1M fln1
approxtmaw "'lution to Program 3 givaxr O..xt l,.x; l, s: 0,. xt 0, whh : " 130. Con!lt!qumdy,
thu. jOiution is optunal.
6.8 Discuss the geometrical S1&nificn"" o( making ll1e fil$1 in Problem 6.! .
The '"lion for Problem 6.2 with 11>c 1ntescr requirerncnu isnorod h tho Jblldcd n:gion in FiJ.
the ft;ulble region fur Problem 6.2 18 V''<n is the "'' of II in1egor poinu (markod whh cr.....,)
bdonging to 1bc. sh:adcd The fin1 pprOAtma.uon rs 1hc circled c:xtranc point.
AI " mull or bnu'K'hjng. I he ku.Jiblc rcg.iun ror Program 2, with the intqe:r ignQrod, if
R<aion I In Fig. hcrcu RCJion II in t11c wno flguro r<pre>cn1S 1ho f<Juiblc '"lPo" ror Prosnun 3
with the int*r requin:mcntJ r'K:J)eclcd. l hlll Region I IIIMI n lOJtlber contllln illl the fc&diblc
in1egor poinu of Fig. and only abose lnl<gor po1ntc. Hence. if the ontd.W in1<gor proaram hat
CHAP. 6] I NTEGER PROGRAMMI NG: BRANCH-AND-BOUND ALGORITHM Ill
,, .,
I) (h)
Oplimal solution {as h dOC$, in this caJC). chat solution wiJI be- optimal for ooc or the cwo new inteacr
progmmJ. Conversely. 1/ th4r two flfl'ft' i"tttl"' programs lu1ot solurUJru. ont of thtS# .fOlutioru (t!Jt Olk'
"''ith rht largrr f" 1/td Nl.k l( t1 NUI..dml:utlfm will M oplil"ffDI Otlglntd '"'e"flt't prQ{IrCilfL
The V:)J.idity Q( lht b(l\lndina technique folloWS rrom the parc:nthc:lict&l rat'liatk jU$1 tnldO.
6.10
6.11
Supplementary Problems
Solvo rottowing problems by usc or the brancll-:mdbound algontbm.
DU&.Jimuc: : .x-
1
+ 2$
1
+
JUbjcct 10: 2K
1
- lK: + lx, S I I
v.1th: 11!1 vorirs.blc:t nonr&e_gad\'C lUld
mnalmiu: ; = .x 1 +- 2.l'
1
+ Jx, + x ..
16. Jx, + 1.t, + x, + 4r, 10
3x, + Jx, + z., + Sx, s 5
whh. aU variables oonncguhc and integral
rruuimize: : lx, + + l'.J
sub)CCi to: S.<
1
+ b, + x, s 15
2x, + t
1
+ 7:c,s20
" + Jx2 + 2xl :S 2:S
all varillblq: nonncg.ath<c: nd intepul
IJ2 INTEGER PROGRAMMI NG BRANCH-AI'fO-BOUNO ALGORITHM
6.12
6.13 Problem I 20.
: IO:t-
1
+ lx
2
+ J 1.1:,
JubjCC1 tO' lx
1
+ 7;t
1
;.. rJ 4
... +ax, - lx> D 11
with: all wrillbles nonnogative and lntqral
[CHAP. 6
6.14 Solve Probl<m 6.7 by opplyl n tllo branth-and-bc>und algorithm dirootly to progrnm (J) o(Prol>ltm 1.8 and
thls proaduro with the upproach tooen ln Problem 6.7
Chapter 7
Integer Programming: Cut Algorithms
At each stage of brnnching in rhe branch-andbound algorirhm the current fcasiblc region (for the
current program with integ.r rcmictions ignored) is cur into two smaller regions (one of them may be
empty) by the imposition Of two OOW tODSiraints derived from the first approximatJo.n (O I be CUrrent
program This splitting is such that the optimal solution to the ourrent program must show up as the
optimal solution to one of tbe two now prognuru; ( Problem 6.8). The cut algorithms of the present
chapter op<ratc in like fashion. t.he only difference being t hat a single new constraint is added
at each stage, whereby t he feasible region is dimimshed v.itl10ut being split.
Ttl GOMORY ALGORITHM
The new constraints are determined by the following thrcctcp pra<:<ldure. (See Problem 7.5.)
STEf' h In the current final s;mplex tableau, select one (itoy one) of the nooint<gn>l variables and.
tdJimur assigning :ero 11afues tQ rht 110nha.fi( t-arinbles, consider the constraint equation
represented by tbc row of t he selected variable.
Eum,.7.1 The Slmplti< tablcu
x, x, . ,
x.
'
x, - 112 0 - 713 1/2 1112
.'{: 1/2 0 - I l/4
4
()
0 3/4 Z5/2
g.i\'ts th.e opri1nal (i.e., the C"Urrtnl firsr approlll imali.on) -as xf 11/2, xi 1. with earh or the nonbas.ic
,._\:: . .md set equal to u.ro. noninteger usignmtnt (or :<) came frotn be: 6nt mw of 1hc tahlca""
wbi<:h represents I he constraint
- !.<, + .<, - lx. t !x, '1 ( 7. /)
STEP 2: Rewri te each fr.tctiooal coefficient and cottStanl in the constrnint equation obtained from Step
I as the sum or an integer and fr.tction between 0 and l. Then rewrire the equation
so rhn t the lcfthnnd :rid< contains only terms with [ructional cocllicicnli (and a fractional
conscant). while the righthand SJdc contains only terms with integral coefficients (and an
integral constllllt}.
Example 7.2 Equation (1.1) bomc;,;
(- 1 + l>x, +.<, + ( - 3 +J).<. + (0 + i)x, = S + l


+ jx, + !x, -l =1 + .r, -x, + 3l<, (7.))
STEI' 3. Require the lefihand side or the rewritt<:n equation to be noonegati"e. The resulting inequality
Is new conslraint.
133
134 INTEGER PROGRAMMI NG: CUT ALGORITHMS [CHAP 7
Enmp .. 7.3 frt>lll (7..!1,
or
is the flCWCQDSIJlUDL
COMPUTATIONAL CONSIDERATIONS
Computing time l saved by appending the new constraint inequality obtained from Step 3 to the
constraint equations dCSC!ribed in the curreJJt finaJ simplex lablo.au r1ther lhan to the alaebraically
equivul<:.nt eonS! given in the original program. (See Problem 7.1.)
The Oomory cut algorithm may not converge; Jhnt ls. uo integral soluJion may n01 be obuuned
rcgrudlcss of the number or heruJlons. GeneruUy. howev..-, if the algorithm does converge, h converges
reasonably qulckly. F'or tbls reuson, an upper limh on Jbc number oriteratlons lo be anempted is olten
tstabllshed before lht compu13tion Is innla1ed. If I he inlegral soluJion is nol oblalncd wi1bin this bound.
the algorithm is abandoned.
There are no theoretical reosons for choa.iog between the Gomory and branchand-bound
nlgorilhm .. The bmnthnnd-bound algorithm i< the newer of the IWO prc><rdurr:s, and &ppell,.. tO be
fuvnred slighlly on1ong prnctitionors.
7.1
Solved Problems
mru:imi?.e: : Zx, + ;W".t
SUb)ll 10: 2x
1
+ h
1
:!> 17
Jx
1
+ 2xJ :S: 10
With: .x, and .<, nonnegative and loltgrol
(I)
lh.e hnegcr requlremcnli abd applysng LbC' simplex method to the ruul&lng lineAr progn&""
wo obuu.n Tublcau 1 u the oputnJtlt!*bleau aflc:r Olle hcrtuio.n.
...
'

,,
"
..
x,
.. , 0 ll/l I -2/l ] 1/l ., 0 () I -'12 11/6 17(2
'
2/) 0 1/l
U)/3 ,,
I 0 0 0 1{3 l

0 I 0 11'2 -1/2 1/2
0 0 2{l 1:0/J
0
Tobloeau I
0 0 1/2 6 1)/2
Tollkoa 2
TllC firs:r &pproMmalon 10 is .'(r 10/1. .xJ Jljl, xt xl 0. 8o1b x f l1nd xj
arr noniotegral Atbiuurilyselecting :cf. 'olo't consid-er the comuuint rc.pceSf:tHed by the second row
I, 1hc. row defitung .tf; nunel}.
' + .... , + i - .,.
Writin caC'h frltCti on u 1hr Q( an lntetctr al'ld a fracuan btlV.'R 0 ttnd 1, wt bavc
Or
jx, + I' - I C: 0
or
7] INTEGER PROGRAMMING: CUT ALGORITHMS 135
ta$ 1hc nev. Rewri1in1 the consH'IJint'f of the orig1nal program (I) in thr Ju8lrn1cd by
Tabk11u I and addil1$ 1he new c:cnJ;uaint. we ;enerate lhe- proenm
mallimize : = l.l'
1
+ Y
1
+ OKJ ..1. Ox.-
subJ<CI li>:
1
/.t
1
+ '> -J.<.- )'
(1)
1."t: 1-- 4'\Z I
wi1b; aJI '-ari.bles nt.mM-&athc: a nd
A '-anab1c. x,., tlnd an &fllfietal vMfable.. .c-6. are inlrodtteed into lhr! inequalit) connraint of (1).
and the two--phase method tS applied, with ;t
1
, .'C the in.idat set o(ba.sk vari:abk:s. The: opumaJ
Tableau 2 is obtaintd after onl}' ittratitJn The firj.t .apprCixbnation to program (1) is thus .x-f - 3,
'fj = If!. -"i = 17/2, x: = :c; = 0. Choo.<in,& :c! to Ult n.:w coostn.liot, we obtain from the t hied row
ufToblcau
or
This;. combmed wnh the constraints or program (J) m rorm$ tuggestai b)' Tabteau 2. gives th.e new
intcacr progmm
. ... ... u u
.\:l - 1' "t .xs-,
+ ix
1
a 3
H.<. - !-<, l
:<" + x,
whh: <tU nooJ.Cg:itthe ud mtegrlll
(J)
lg.nnring integer ond 11pplyins the rncchod 1P pwgrMl) with .x-,. x
2
, .T.J Qod
(Jtrtlfici .. l) in, uill bMac lCt. we obtJ.i.n lha optimal 3 .
<, . ,
'
... x, x.
,,
0 0 -IJ, J 0 11/6 2013
.... 0 0 - IJJ 0 I f) S/3
,,
0
I)
I 0 - Ill
'
u 0
I) -l
0
() ()
l (i Q 19/}
Tdlnu 3
A nev. itcrali\Jn of the it. (roc:n 8/3 in Tableau TbJa resu.ht in a progr:un
solu.liun l\ with xt - 3. ,\t = 0, ltnd : c. 16. this sotuuoo i1 then 1bc optimal soluuon to imescr
prosmm (I),
7:1 Discu<.< 1he gom<lricnl significncc of I he 6r:s1 c<Jns1rain1 in Problem 7.1.
lnilially. che region coru:ists uf Rll polnu in tht fir.t qu;adrnn: tu:avfna Integral tt\:u
'lliiL'tfY
+ s 17 nnd
""'the I'(! in" morkcd by in Fig. 1 1 (o).
The mnnraint 1tddcd 10 I he originJ I prns.mm (I) wu 2.(, + $ , 1: it led to prc.>,nun {2). S<JJ-;ing 1hc
136
' :
INTEGER PROGRAMMJ!-10: CUT Al.CORITHMS
fitst '*PrfOIUMtioa
oo P"''""' (I)
,.,
I
,,
'
?I
2s, + (10- 3;<, - 2x,) ;o 1 or
[CHAP. 7
(b)
The effect of impo!itng. s J tS indftlncd n 1 a.mall plcc:c tOOIIuning lbc curran f1n:1
tapproxi.nulion ... slic:td oJT lhc fea11blc No in1egn polnt., b lo.:L
1-1 Solve Problem 1.12.
The fir$1 approxunation tp th1$ mtt-p proaram (teoc Pro.blctn :5,10 witb lhe variab\ts relabeled) Is
700. tjJ 500. -<!
1
1000. x:,-x!
1
xtJ 0. v.ith : Since tht.. fima,ppromadon 11
it iJ aloo th< optimal liOiunon to the iotqer proJ111m. Under tbi optimal :oc:hcdulo. 100 bo>cs ill
he l>lppcd (rom f11<1e>ry I to rctOJ1CI" 1 500 boCI from f><tory I tu maikr 3, and 1000 bol<s from f:octory
l to Rtaller I Tbe IOtllllbippiDI 0051 is $216
7.4 Solve Problem 1.5.
Program ( of Probl= 1.5, brousbt inro s!Jlndlud form. b
mi1imize; ; = 2(b
1
+ l.U + ll"C., + b'. + 0."t
5
+ Ox + O.x, + Ox + Mx. + Mx,
0
subj(('l 10: 4x
1
-r b1 +- -x.. + t 5'

-x,


t "'
'
all varitblc$ nonncgativ.t and integral
1
0 7
1
(f)
the Otcacr ru1rit1ions and sohin.a. this pro-na:rn by 1.he 1wo-phasc rnechod. we obtain Tableau I
CHAP 7) INTF.GF.R PROGRAMMIN(): CU'T ALGOR.ITHMS 137
after lhrtenerations. The fiat approximation to (I) is lhus x: I.H. x; a 7,x; a 5, with -:. -:;a: 279.

x, . ,

x.

.
'
I 0 0 - 0.3 0.005 0 - 1.6 0 us
x, 0 0 I 0.2 - <U 0 0.4 0 s
x, 0 0 0 0.3 I 1.6 0 S.l5
.<, 0 I 0 0 0 0 I 0 7
x. 0 0 0 - 0.2 0.2 0 -0.4 I 2
0 0 0 2.4 2.6 0 2.8 0 - 279
Ta-.&eau I
No.-. tltis first approxrmation may be rounded to the Uue-gral sol.utk>n x
1
= 1. Xz = 7, .t
3
= S,
wi1h = = 284. h follows I hal the dcsirtd minimum ""nno1 excred 284. Ollthe other hAnd .... rerrins to 1hc
original progrm (4) or Probl<m LS, we see 1hot for integral valut< of the vriables = is .., .,..., integer:
in view of che Jo"'-:er bound : = 279 by I be flrst approximat ion. the minimal : anno1 be
tess rhan UIO. TberefO<e. rhe mini=l: can only he 280. 282, or 284. and we are guaranteed thai the error
(ommitt-:d in (2, 7. 5)1 cue the oplimal 50lution at wor.il
J84- 280
- -1 . 3/.
2$0
(Strtlng from Tubleau I. one lind.! fter si>< 11emtJot>J or the Gomory lgorithm thot (2, 7, S)' is ln faet
rhe optlmol solurl on.J
7.5 Devti<}P the G<lmory cui algorithm.
the: optimal tableau fhll rt:tulu rro.m applyrnt lht method co An inttacr ptOgntm
wi1h the iotegt:r liflorod. and :u:t\lmC one:- of the: buic variabln. x.,, is nonintegral. Tht.
co"straint equatio11 correspoodinglo the tableau row determined "t mui1 the rorm
(/ )
where the .rum itt O\'er all vrui.abk!L The ,... tennJ arc: the coetndenu and the coru1ant term hppearing
in the tllhh:.1u ro.,.,, detc:rminlng .x Sinc:e x-. IJ obt3i ned rrom CJ) by .st!UinJ the nonbiuiic vurinblct equal h>
ttrn, iJ folio"'" lhut J'o i:s il.I!Wt no-nhncgral.
Write eiH:h f"lctm in (J j u tht JUI'll or un i r11C!C'l and a J.Onne,ative fraction less lhlln 1:
and fo ... io + /o
of tbc be uro, but f u as @l,,.ll_f&fl l ced to be positfvo. Equation (I) becomes
.. r o, + hlx,- r. + r.
or
(2)
If ooeb .vriabk iJ rcqutrcd to he mteg..,l, then tbc lcfthaod side of (2) IS which forces the
alow lb be im::gral Sua. sinc.-e CllCb /j and .x
1
is JO too ts '/;xJ. Tbe righHunw:l
sfdt of (1) then I) an Integer wbtch is smaller ttwl a posith'e fraction less cban that is, a lnteser.
ru
Tbts is the new constrain-l in 1be Oomory aleoricbrn.
7.6 nnotht!r cut a.lgornhm.
O.msidcr (I} of Problem U nontwk arhtble :(
1
ii then .t, Yt is nonintegr11L 1f x
11
is
to become then .at lust 01'h: ,,f tb< :t
1
mu1U be made dHTc:R:ru from 'Lcru. 11U
13k IN1 CtoCR PRCKJRAMMtNO: CUT AI. GORI11fMS [CH.AI'. 1
v:uiil.ble> rcquirccJ lo be nounc:pthc and lntC!gfal il rouo,...& Uwt '.li lr;u;i. " 'lc- noubot.\tC tun.ahiC' U1UJol be
1n"de trcnuer thtth oc eqmal to L Thb ht tur.:o lmpl.es that the sum or aU lhc acmb.sic Vl.n!lbln u1w:t be
LWde gr-eater th:in or equal to t. I( thii condnioo rs usod as the nev. 10 be adjoin'*s to the
progrum. ha'e the llrst .,y l'>ao1.jg.
7.7 Use the cut aJgoriLhm developed In Problen\ 7.6 10 solve
-: - ),
1
+ 4.v.!
11bject1<>: h, + <, s; 6
+ :b; s 9
wit h -., and -., nonncyauvc and integral
Introducing stack vnnllbles .. and tc 3nd then the rC5ulting pml(l3fn. .,,b the intcw:t
rcqmrcmcnts Ignored. th" s:Lmpiex obtam Tablr:,u I
"'
.,
,\ J .. ,
I
'
()
07S 2.ZS
,,
0 - o.s I.S
0
()
0.:!5 1, 25 11.7S
Tolllelu I
i he firs1 >lppt(nlrrusll.gn lhctdt)n:, :c t - 1..25, xi - L.S. '4hich tS nol wtegt!.IL Tbr vit.tiablet lU'C
'l Knd .-..... sn tht new t:Onsu-:ai nc is )t! 1-- .\a I. Appcndmg tlus (OnsLr:rinl w T).bteao 1. :a.fie.r ht- inlroducuon
.lf surplu!C ' uriablc *<, aod ll!tlflc. .. l vruiablc .:lb-. and $Olt'1ng the rcsulung .. m by the tW phllsc method,
we gcrwrotc :!.

.,

'
...
I 0 0 - I
"'
0 0
"'
0 0
0 0 0
Tableau l
'
0.75
-O.S
- I

1.5
l
l.t2S
I
11 folJOYw""J from Ta.blcau ltbou xf-= _:t: j = ;:, = I. .,.Ul'l ..-. and l: .. oonbasic. SOIUUWl
is: we roL:e .l
4
..., ..t, I 8$ 1he ne,.. oonwarnt. Adjoining. this 10 T:1bk3u l. afta the
intTodUC:UOn o( 1UrpiUJ and artifJdJJ \'llri3bll: .1..,, QO.d 1Q)vi"i th.C' n::::sultfn-g f rOJ:r.tm by the
1wo-pbi&SCt fl'W'lb.od v; e generate T:J.blt-::.u 1
.<, x,
" '
... ,
"'"'
.,
l 0 0 -1.75 0 0.7S 0.7S
,,
0 I 0 t.S 0 - 0.5
.t, 0 0 .I 2 0 - I 2
, 0 0
f) I I - I I
0 0 Ill 0.75 0 015
I
12.25
TaWtau J
I
CHAI' 7) INTEGER PROGRAMMING: CUT AL<lO'RITHMS
From Tableau J. the tttrrent optunal lll nonintegral. with ntlflbasic variables x
4
and
now eonttr:l..nl is t.h lli '( .. ...._ .,:
11
I it to 3 and the program by tile
;\\ophasr "T obt:Un xj - 0, x1 3, with : 12 Sint"e cbis solution is intcaral. it ts tilt'
tOJu1ion ' "' the originAl program
Supplementary Problems
1-M Use: C.hc Gomory algorhhm to
subjM lo; :r, + 2.l:: + .lx,; s 9
lt, + b'J. +
Mtb: all variabks nonn<1<a<i.o and inte;ual
7.9 S<>lvc Problem 1.3 by tbc Gomory algorithm.
7.10 Solve Pr<lhiem 6.9 by the Gomory algorithm
7. 11 Solve Pmhlcm 10 by the Gol1!ol') l.gorithm
1.12 Sol-ve Problem 6 II by 1he Gumor) !tlgorithm..
7.1J Solve Problem 6.9 by t (u, CUI olgurithm of Problem 7.6,