• 検索結果がありません。

An Efficient Algorithen for the Minimum-Range Ideal Problem

N/A
N/A
Protected

Academic year: 2021

シェア "An Efficient Algorithen for the Minimum-Range Ideal Problem"

Copied!
2
0
0

読み込み中.... (全文を見る)

全文

(1)

1−F−5

1995年度日本オペレーションズ。リサーチ学会 秋季研究発表会

Åm芯侃e五eⅡ止Å孔g①『弛肋m馳〉『も馳eM五m五mⅧmロ凪amg母Ⅰ鮎a凰p『①他凰em

O2401130 中根本俊男(筑波大学社会工学研究科)

且。 Ⅱm屯灯OdⅧCも五om

StlppOSeWeare given aposct(r>artial1y or(lcredsct) P=(E,ゴ),arealwciglltW(e)associatedwitlleaCll elemente∈EandapositivcilltCgerk.AsubsetofE

iscal1edanideatIofPifcゴC′∈IallVaySiml)1ies

C∈I.Weconsidertllel)rOblemaBfollows: Pr8岬:Millilnizel欝−U(e卜讐デー申) Subjectto J∈J(p),けl=た (1・1) whereT(ア)istlleSetOfa11tlleidealsofP,auldlXlis thecardillalityofa負nitesetX.Wecal1thisproblcm tlleminimllm−rallgeidealprol)lcln(P,ange)・Wbl10Pe thatthi$prOblemscrvesaLS・aSllbproblemforlcvcli11g rcsollrCeSforlargcscalcscllC(1uliIlgplal‖llllg・

The oI)tinlizatiollprOblell10Ilthci(1ealis valual)1c

becausemallyaPPlicatioll$illrea‖ifbarcfbrl11alizcdas theidealproblem.TllCreforc,VariollStyl)eSOfthisclass

OfproblelmShavcbee11WCllrcsearclled.ItisilltereSting

tonotethaもthecar(1inality−rCStricte(lidcall)rOblelnis JVP−llardiftheobjectivefullCtionislillCar,butitll astrollglypolynomialalgoritllmiftheobjcctivcis to minimizet・herange・Totllea11tl10r’skl10WIcdgc,110011C hasevcrconsidcrcdPran6e・

The min−rallgC PrOblcmisalsoanintcrestingcom−

binatorialoptimizatiollprOblem.ScveralrcscarcllCrS havestudiedan11mbcrofmin−rallgCl)rOblcl11S,incl11(ト ingtllea$SigllmelltprOblenlt3],tlleSpal111ingtreeprob− 1em【1】,alldthecutproblelnf2】・Forthcsel}rOblel11S,a gencralalgorithmllaSbccllprOpOSedill【3】・Siml)ly叩− plyingtllegenCrala]gorithmin(3】,P,a,,6CCanbcsoIved il10(7Tt71・)tilne,WllerC7l・=lElalldmistlleSmallcst n11111bero一礼rcstorepl・eSent・ア. However,WCl)rCSe11tal10(nlogn+m・)algoritllln fortllisprol)1cln.ThisalgoritlllllisdiffcrclttfromtllC algorithlnSprOpOSedfortlleabovemin十「孔11geCOmbilla− torialproblcms.‡tisalsoprovcdtlla†・tllispl・Oblcll111aβ al10(nJogn+m・)lowcrl)Oulld・TllismcallStllatも11e algoritlllnl)rCSelltCdillthisl)apCrisoptilnal・ 2. W払e M五m五maxIdea且p町Ob且em

Weconsider the minimaxidealproblemall(1tllC ma3;・

iminidealproblem whicl11)lay animportallt rOle to

SOIvePran6C・Tlleformeris(1c餌cdasfouows: Prnini−n礼X:11−ill(l項)l(1・1))・ Thcfo1lowi11galgorithmsoIvesPminimax・ Å1goriもhm MINIMAX(ア,た) Stepl:PlltJ:=臥 S七ep2:Repeatthefollowing(a)and(b)ktimes・ (a)C:=(eIeisaminimalelementofア(E−])). (b)IfC=0・tllenStOp(thereiミnOidealofsizek OfP)・Othcrwise餌damln−Weightelement ∂inC,alldJ:=JU(a)・ Theorem2。乱:MINIMÅⅩ(ア,た)∽叫印加釘=血涙m肌 ide几∼扉ア豆乃0(几logγl+†花)・ ロ Similarly,Wede丘ncthelatter(Pmaximin)andconstruCt all乱1goritllm,Whiclliscal1cdMAXIMIN(ア,k). 3。A Na五veA皿gor兢払m ForrealvaluesαalldPwithα<β,letL(α)=(ele∈ β,t〃(e)≦α)alld∬(β)=伺e∈β,ひ(e)>β),弧d br弧elemellte∈β,1etダ(e)=(占Ieゴ占,占∈β). TllCn,aSetE−Ue∈L(e,)F(e)−Ue∈H(P)F(e)isanideal Of7).Forsimplicity,WeuSeア(α,β1asallabbreviation fortllCSubposetillducedbytheaboveideal.Similarly, Weabbrcviateア(E−Ue∈L(α)F(e))to P(α,∞),and ア(β−∪。∈JJ(β)ダ(e))topト∞,β】・

First,WeShow thatthere exists the subposetof7)

Whose mill−rangeidealis found ea$ily.Given areal

Val11e勘,COmpute the optimalvalue of PminilmX On P(α‘,∞),dcnotedbyβi+1・if7,(α‘!OO)haBanidealof

sizek.TllCn,idelltibT7)(cti,βi+l】.

memma3・孔:釣rαgJ壷deαりげβizeたげア(叫,β汁1】,

Weん仰靴=llaX(叩(e)le∈J)=βり小 口

Colnl)illing tllislemmaand thc fact tllat PT8nge be−

COmCSeqllivalellttOPmaximinifeveryidealhasthesame

maxim11111WCight,Wehavethelcznmabelow.

memma3・2:エef J亘1 むe α

mαエ富m名和

盲deαg扉γ(αi,βけl】・rんe符丁汀l豆βαm古かrα叩e盲deαg

げγ(αi,β汀1ト ロ

Next,We Sllal1expa.nd the discussion aboveinto a

ll孔ivealgorithm.Letαo be asu伍cicntlysmal1value.

Tllen We havc7)(α0,∞)=P and we can丘nd the nrstappropriatevalueβ1byapplyingM‡NIMAX(P,k)・ FurtllCrmOre,by applying MÅXIMIN(7)(α0,βl],k)a lllill−rangCi(lcalIlOf7)(α0,β1]is obtained・De丘ne α1=1nin(w(e)lc∈Il)・Aftcrtllat,repeattlleabove proccsscsfbrαi(i≧1)obtainedinthepreviousitera− tiolllllltiltllereisnoi(1ealofsizekof7)(αi,∞).We OToshio NEMOTO:^ r)octora]Candidate studying at

Doctoralprogramin Socio−Economic Ptann川g,University or T811kuba・,Tsukllha,30うJAPAN.

e−mail:nemOtO◎sllako.sk・tSltkllba・.aC.JP

一且36−

(2)

(a)Ifci∈SaJldlu(ei)≦α,thellD3(ei)・ Ifci∈Salldα<w(ei)≦β,thellS:= ぶ−(ei)alldJ:=JU(ei)・ Ifei∈∫alldβ<ひ(eり,tl■lenβ:=u(ef), ∫‥=5−(ei)all(1J:=JU(ei). l)rOVCtllefactt・11at†・11Creisalllill−rallgCi(lcalofPill (印=1,…,ヴ‡ol)t孔ille(1孔l)0、▼e.Tll(汀CrOl・e,it ta・kes O(rzl(nlog77・+m))til11Ct・0負11(1amill−rallgCi(lcal・Tllis tillleis110tfastcrtllalltllCgelleralalgoritl1111【3]・

4. Improved Implementation

れ)improvc thc til11C Of t・11Cllaive algoritllln tO

O(nlogn+77.・),WCilltrO(1ucctl−reellCWidca5・ Fil・St・1y,WCCXamillCal−1CtllO(1toi(1clltih,P(−∞,P)・

LetJll)C t・11C SCtJol)taillCdl)y rCl)CatillgStep20f MINIMAX(P,k)wllile mill(w(c)lc∈C)≦β1101(1s

a・ftel・MINIMAX(P,^:)is(lollC,、VllCrC Pis tllC Ol,ti−

1・nalval11C Of Pminil,.aX.Tllell,We havc tllatJ′ =

β−∪。∈〃用)F(e),i・e・,ア(J′)=ア(−∞,外

SccolldlyiWe PrOl)OSC how toi(1clltif* P(α,∞)・ Illgelleral,l10111Ctl10(1cxccl)t for t・11C basic scal・Ch−

Inethodis fo1111d.Howcver,1)y uSlllg t・hcillk)rJlla−

tiol10f7)(−∞,β)01)taillC(1illtllCpl・eViousstel},E− ∪。叫α)ダ(e)canl)ei(1ellti鮎(11乃rt・11erOll(川・illgt・111・ee Operati(ⅢS. Dt(e):Deletiol10fダ(e)k)re∈(ガーリe∈冊)F(e)トf・ D2(e)‥Deletiol10ffl(e)hre孔Clle∈∫(α)・ D3(e):pelctiol10汀(e)roreacllrelll礼iIli】1ge∈エ((一っ・ Here,Jis 孔I11孔Xi川illi(1e礼lorア(−∞,β】,α = min(w(e)le∈f)孔−1(lf(可=(eIe∈f,−U(e)=α)・Ill hct,itisl10tllCCCSSarytOident・ibTP(α,∞)coll−1)1ct・Cly tofindamillimaxi(1ealofit.TllCal)OVCOI)CratiollSarC

carried out as tllC nCe(1arises.

Tllirdly,WCilltrO(1ucc allCW PrCl)rOCCSSOr SCllCmC,

whicll11SCS t.1VOlleWOr(1crsollEdc611C(lbelow.IllCar− ryingoutMINIMAX(P,1,r),CVCryCIcll−ClltillEl,Clo11gS toJillt11川.Tlleminima9”rderqfPisdcGllC(1astllC Or(1cr11WhichMINIMAX(7,,”・)collSi(lcrstllCelcll−Cllt・・ Inasllllila.rra5llioll,(1efilletlle m丑止肌わーOderげアl)y MAXIMIN(7),r,・). Theorem4.1:爪)r几冊ideαり扉ア付和d冊i押よ叩erん Ⅷ軌0≦た≦けし班eβllむβe£J⊂・Jc㈹壷蘭叩可用 e∫ee血れmわ血−Order(m血れorder川アi… 血血禦減h丹Ⅷ血血品叫函叩(J)・ ロ

HellCe)glVelltllCl11illill−fLX−Ordcr and tllC maXilllill− or(1er ofア〉as al)reprOCeSSOr,itisl10tlleCCSSary t・O arra)lgetllCSCtOfalltllClllillilllalcIcl11CntSOftllCull− derlyingslll)pOSCttOSOIvcPl11i11inⅦXalldPnlaXiIlli]1・

Now,1et us(1cscribc a11e代cic11tilnl)lelllelltatiol10f

tllel血vealgo山I111l・

Algorithm RANGE+(P,k)

Stepl;Dchle thcmilliln礼X−Or(1crand tllCl−1aXil−1ill− OrderofP,alldgivecacllCIcll−elltinEtlleil一(lcxill thcminilllaX−Ordcr.PutJ:=仇S:=E,ra・n・9e:= ∝),α:=−∝〉,β:=−()〇・alldよ:=1. Step2:ⅥりIile5−≠仇(lotllefo】lo、Villg,(2−1)t・0(2−5)・ (2−1):RcpcattllCfollowillgltll†・il[Jl=k・ 0 0 g g n n e e L t t Lん た <ニ ▼J ▼J d d 血弧 乃 几 > > (b)よ▲‥=よ+1.If toStel)3.If to(b′). (2−2):Wllilcw(ei)≦β,dotllefollowillg. (al)Ifci∈Salldw(ei)≦α,thenD3(ei)・ Ifei∈ぶalldα<l〃(ei),tllen∫:=∫− †ei),]:=JU(ei),R11(lthebiggest占∈J illtllCmaXimi1l−OrderalldDl(占). (b′)i:=i+1・Ifi>n,thencal1(2−3)and (2−4),弧dgotoStep3. (2−3):Filld the.mill−Weightclelllent占∈Jand α:=ひ(けIfrα↑lge>β−α,tllenrαれ・タe:= β−αalldβ◆:=β. (2−4):Wllilew(占)=α,doD2(占)and Rndtlle nlin−WCiglltCIc)nent占∈J. Step3:R・etricvcamill−rangeidealbyusillgP’. Theorem4・2:鱒ANGE+(7),k)correctLyco77岬ules a mれ−r(打叩だideα上げアi几0(几log†い卜m)fime. □

5. Lower Bound

Wcsllal1collSi(lcralowerboulldforPranse・

Theclosestknumbersproblem(Cl(k)):Givenn

real11ul11l)CrSalldapositiveintegcrk,findknum− bcl・SWl10SCrallgCissmal1est. Lemma5・1:Cl(k)hasann(nlogn)towerbound・ロ

We prove thc fact that Cl(k)is n−tranSformable to

Prangc・Ti1Crefore,WChavetllefollowlnglemma・・

もemma5・2:P.a.−g。ん那 αm n(m′log几+m)わびer

むoTl†ld. 口

Theorem5.3:The aLgorilhm RANGE+(7),k)re一

匹ireβ町=ogTけ7れう£まme,ひ九icん由0〆imαJ・ □

Refbrences

tl】Z・GalilalldB・Scllicl)er:011RndillgmOSt1111iform Sl)allll111g trCe・Discrele^ppLiedMalhemalics20 (1988)173−175・ 【2】N・IくatollalldIく・Iwal10:EfRcientalgoritllm$for millilnumrallgCC11tprOblems.Nelworks24(1994) 395−407. 【3】S・Martc1lo,W・R・Pu11eyblallk,P・Totha11d

Wcrra:BalallCCd optimization problems.Opera−

!よ0耶虎eβe即℃んエe〃e得3(1984)275−278・

ー137−

参照

関連したドキュメント

Hence, the degree theory constructed in [1] is an extension of the classical Leray–Schauder degree in Hilbert space.. It is unique, single-valued and has the usual properties of

These allow us to con- struct, in this paper, a Randers, Kropina and Matsumoto space of second order and also to give the L-dual of these special Finsler spaces of order two,

Kirchheim in [14] pointed out that using a classical result in function theory (Theorem 17) then the proof of Dacorogna–Marcellini was still valid without the extra hypothesis on E..

After that, applying the well-known results for elliptic boundary-value problems (without parameter) in the considered domains, we receive the asymptotic formu- las of the solutions

For a given complex square matrix A, we develop, implement and test a fast geometric algorithm to find a unit vector that generates a given point in the complex plane if this point

Secondly, once we have established the solvability of SPDEs within the stochastic parabolic weighted Sobolev spaces H γ,q p,θ (O, T ) , we have to exploit the L q (L p ) –regularity

Pi˜nar gave an unified approach to the orthogonality of the generalized Laguerre polynomials {L (α) n } n≥0 , for any real value of the parameter α, by proving their orthogonality

In section 4, we consider another boundary controllability problem for the higher order linear Schr¨ odinger equation, in which only the value of the first spatial derivative (at x =