Job shop scheduling with unit processing times

Size: px
Start display at page:

Download "Job shop scheduling with unit processing times"

Transcription

1 Job shop schduling with unit procssing tims Nikhil Bansal Tracy Kimbrl Maxim Sviridnko Abstract W considr randomizd algorithms for th prmptiv job shop problm, or quivalntly, th cas in which all oprations hav unit lngth. W giv an α- approximation for th cas of two machins whr α < 1.45, an improvd approximation ratio of O( log m log log m ) for an arbitrary numbr m of machins, and th first (2+ε)- approximation for a constant numbr of machins. Th first rsult is via an approximation algorithm for a string matching problm which is of indpndnt intrst. 1 Introduction Job shop schduling is a widly studid and difficult combinatorial optimization problm [11]. In this problm, w ar givn a collction of jobs and a st of machins. Each job consists of a squnc of oprations, which must b prformd in ordr. Each opration has a particular siz and must b prformd on a spcific machin. Th goal is to minimiz th makspan, dfind as th compltion tim of th last job to finish. Th problm is dfind formally in Sction 2. W considr th prmptiv cas with th objctiv to minimiz makspan. This problm is strongly NPhard vn for two machins [9]. If th numbr of machins is part of th input, thn th rsult of Williamson t al. [20] implis that thr is no polynomial-tim approximation algorithm with prformanc guarant bttr than 5/4 unlss P = NP. For th gnral nonprmptiv job shop schduling problm, th bst known approximation algorithm has prformanc guarant of O(log 2 mµ/ log 2 log mµ) whr m is th numbr of machins and µ is th maximum numbr of oprations in a job [19, 7]. If for vry job thr is at most on opration on ach machin, this bound can b improvd to O(log 1+ε m) for vry ɛ > 0 [5, 6]. This variant is calld th acyclic job shop schduling problm. In th cas of th acyclic job shop with unit procssing tims for vry opration, th famous paprs of Lighton t al. [12, 13] giv constant factor approximation algorithms. In th cas of th prmptiv job shop an approximation with ratio O(log mµ/ log log mµ) is known [12, IBM T.J. Watson Rsarch Cntr, P.O. Box 218, Yorktown Hights, NY mail:{nikhil,kimbrl,sviri}@us.ibm.com 19, 7]. A polynomial-tim approximation schm (PTAS) is known for th spcial cas of a constant numbr of machins and a constant numbr of oprations pr job [10] both for th prmptiv and nonprmptiv problms. In this papr w giv th following rsults for th prmptiv job shop schduling problm: 1. For m = 2: It is asy to s that any rasonabl schdul has an approximation ratio of 2. Svastianov and Wogingr [15] gav th first non-trivial 1.5-approximation algorithm for m = 2. An algorithm with a tightr approximation guarant in trms of th maximum machin load L and th maximum job lngth l, but still 1.5 in th worst cas, was givn by Andrson t al. [3]. All known approximation rsults for shop schduling (othr than approximation schms) us as a lowr bound th maximum of L and l. Svastianov and Wogingr [15] not than any approximation algorithm with ratio bttr than 1.5 for th two machin cas would rquir a nw, non-trivial lowr bound on th optimal makspan. In this papr, w giv such a rsult basd on th rlationship btwn th prmptiv job shop schduling problm and a string matching problm ovr th binary alphabt. Our algorithm has an approximation ratio of lss than For arbitrary m: W giv an algorithm with an approximation ratio of O(log m/ log log m) for m machins. This liminats th dpndnc on µ in th rsults of [12, 19, 7]. W giv anothr vry simpl algorithm that constructs a schdul of lngth (1 + ɛ)l + O(µ log m)p max in th non-prmptiv cas, and hnc a schdul of lngth (1 + ɛ)l + O(l log m) in th prmptiv cas (s Sction 2 for rlationship btwn prmptiv job shop and nonprmptiv job shop with unit procssing tims), whr p max is th maximum lngth of an opration. In som cass this givs a substantially strongr guarant than Svastianov s guarant [16, 17] of L + O(mµ 3 )p max. In particular, if th instanc is such that µ ɛl/ log m, thn our algorithm givs a (1 + ɛ) approximation. 1

2 3. For constant m: W show how our additiv approximation of (1+ɛ)L+O(l log m) can b usd to giv a polynomial tim (2 + ε)-approximation algorithm for any constant numbr of machins. (Not that w allow numbr of oprations pr job to b part of th input). Prviously, no algorithm with an approximation ratio indpndnt of m was known for th problm. 2 Modl and notation In th job shop schduling problm thr is a st J = {J 1,..., J n } of n jobs that must b procssd on a givn st M = {M 1,..., M m } of m machins. Each job J j consists of a squnc of µ j oprations O 1j,..., O µjj that must b procssd in ordr. Opration O kj must b procssd on machin M πkj, during p kj tim units. A machin can procss at most on opration at a tim, and ach job may b procssd by at most on machin at any tim. For a givn schdul, lt C kj b th compltion tim of opration O kj. Th objctiv is to find a schdul that minimizs th maximum compltion tim, C max = max kj C kj. Th valu of C max is also calld th makspan or th lngth of th schdul. For a givn instanc of th job shop schduling problm, th valu of th optimum makspan will b dnotd Cmax. For ach job j and machin i, l ij is th total amount of work in job j dsignatd for machin i. Lt l j = i M l ij dnot th total lngth of job j, and lt l = max j J l j. Lt L i = j J l ij dnot th load on machin i, and lt L dnot max i M L i. Clarly, max{l, l} Cmax. Lt µ = max j µ j b th maximum numbr of oprations in any job. In this papr w considr th prmptiv variant of th job shop schduling problm, in which vry opration can b prmptd during its xcution and rsumd latr without any pnalty. Of cours, w must still oby prcdnc constraints btwn oprations, and for vry opration O kj, th total tim allocatd on machin π kj must b qual to its procssing tim p kj. It is wll-known that thr xists an optimal schdul for th prmptiv job shop problm whr prmptions occur at intgral tims (s [4] for mor gnral rsults of this sort). W may assum that all opration lngths ar polynomially boundd sinc this assumption can b rmovd with only ε loss for any ε > 0 by standard scaling and rounding tchniqus. Thus, w will considr th prmptiv job shop schduling problm to b quivalnt to th nonprmptiv job shop schduling problm with unit procssing tims; i.., w split vry opration O kj with procssing tim p kj into p kj unit lngth oprations and assum that all inputs p kj, L i, l j ar polynomially boundd. W us ε throughout th papr to dnot a constant that can b mad arbitrarily small. W assum without loss of gnrality that for all i M, L i = L; this is asily achivd as follows. W add a dummy job for ach machin i if ndd, comprising L L i unit lngth oprations on machin i only. It is asy to s that this dos not chang th optimal makspan, sinc any schdul has lngth at last L and at last L L i idl tim stps in which th dummy job can b insrtd. 3 Th two-machin cas In this sction w considr th prmptiv two-machin job shop problm J2 pmtn C max. Th prviously bst known algorithms for this problm hav a worst cas approximation ratio of 1.5 [15, 3]. As mntiond prviously, Svastianov and Wogingr obsrvd that it is impossibl to achiv a bttr ratio if w us th trivial lowr bound of max{l, l} only. To s this, considr th instanc with two idntical jobs J 1 and J 2. Each of thm consists of L/2 oprations that rquir machin 1 followd by L/2 oprations that rquir machin 2. Clarly, th trivial lowr bound on th makspan is L. Howvr, it is asy to s that any fasibl schdul has makspan at last 1.5L. In ordr to gt a ratio strictly bttr than 1.5, w adopt th following approach: W first not that in th cass whn l < 0.88L or whn l > 1.16L th trivial lowr bound is nough to giv a ratio bttr than In th hard cas with a big job B of lngth about L (i.. l L), w attmpt to maximiz th numbr of oprations of othr jobs that ar prformd concurrntly with B. This is most clarly statd in trms of a string matching matching problm which w dscrib in Sctions 3.1 and 3.3 blow. W show how to solv this string matching problm so that w can prform concurrntly with B at last a (1 1/) fraction of th most possibl. W thn show that combining ths diffrnt cass givs a worst cas approximation ratio of about in th gnral cas. 3.1 Maximizing disjoint matchs btwn a st of binary strings and on long on. Considr th following problm. Lt S b a binary string and C = {S 1, S 2,..., S n } b a collction of binary strings. Lt l dnot th lngth of S, and lt l i dnot th lngth of S i. For a string X, lt X(i) dnot th i th charactr of X. W say that S i has a matching E i of valu k in S if thr xist indics a 1 < a 2 <... < a k and b 1 < b 2 <... < b k such that S i (a j ) = S(b j ) for all 1 j k. In othr words, S i has a matching E i of valu k in S if som subsqunc of k charactrs of S i can b matchd with a subsqunc in S. W dnot th valu of E i by E i. W also associat E i with th st 2

3 of indics b 1,..., b k. Matchings E i and E j of S i and S j, rspctivly, in S ar said to b disjoint if E i E j =. Problm: Find a collction of matchings E 1,..., E n of S 1,..., S n in S such that th E i ar pairwis disjoint, i.. E i E j = for all 1 i < j n. Th goal is to maximiz th cardinality of n i=1 E i. In othr words, w sk to match as much of S 1,...,S n as possibl, but with th rstriction that th matchings for any two S i and S j ar disjoint. W show th following: Thorm 3.1. Thr is a randomizd approximation algorithm for our string matching problm with xpctd ratio at last (1 1/). To prov Thorm 3.1, w considr an LP rlaxation of a natural intgr program formulation and show that it can b roundd to giv th dsird approximation ratio. Th proof is dfrrd until sction 3.3. W now show how Thorm 3.1 implis a 1.44 approximation for th job shop problm with two machins. 3.2 Solving job shop using th string matching algorithm Thorm 3.2. For vry ε > 0, thr is a randomizd (α + ε)-approximation algorithm for J2 pmtn C max, whr α = , with running tim polynomial in th numbr n of jobs, th maximum numbr µ of oprations in a job, and 1/ε. corrspondnc is invrtd: oprations on th first machin corrspond to 0 s and oprations on th scond machin corrspond to 1 s. W can obtain a schdul for a job shop instanc from a solution to th matching problm such that a match in th string problm corrsponds to an opration of a big job bing prformd in paralll with an opration of anothr job. Thus, th numbr of matchs is qual th numbr of tim units in which two oprations ar xcutd in paralll. Lt V dnot th optimal valu of th string matching instanc. V is th maximum possibl ovrlap btwn job B and th rmaining jobs, i.., th maximum numbr of unit-lngth oprations in B that can b xcutd concurrntly with oprations in othr jobs. Not this is not ncssarily th amount of ovrlap btwn B and th othr jobs in any optimal schdul, but is an uppr bound. This allows us to lowr bound th optimum makspan as follows: Assum that th maximum ovrlap btwn B and th othr jobs is achivd, furthr w allow th maximum ovrlap btwn th rmaining oprations in othr jobs. In particular, th rmaining oprations in othr jobs ar j B l j V, and w assum that optimum can schdul ths in paralll. Thus, th optimal makspan Cmax is at last l ( j B l j V ) = L + l/2 V/2 Proof. Lt 0 δ 1 b a constant to b dtrmind latr. W considr sparatly th following cass: 1. l 2δL. In this cas, w us th algorithm of Andrson t al. [3] which finds a schdul of lngth at most L + l/2, so th trivial lowr bound of L in this cas is nough to giv an approximation ratio of 1 + δ. 2. l 2 1L. Again w us th algorithm of Andrson t al. [3], which givs a schdul of lngth at most L + l/2. In this cas th trivial lowr bound of l implis a ratio of at most L/l + 1/2 / L > l > 2δL. In this cas, w us Thorm 3.1. W rduc J2 pmtn C max to th string matching problm as follows. String S corrsponds to a big job B job of lngth l (An arbitrary choic can b mad if thr ar two or mor such jobs). All othr jobs corrspond to strings in th st C. W crat a binary strings corrsponding to th jobs. For th job B, w crat a string of lngth l, whr oprations of B procssd on th first machin corrspond to 1 s in S and oprations on th scond machin corrspond to 0 s. For th othr jobs, th 3 Th quality follows by our assumption that ach machin has load xactly L and hnc that = 2L l. j B Using Thorm 3.1 w can obtain a schdul that has at last (1 1/)V matchs. In th worst cas, w assum that no othr oprations ar xcutd in paralll. Thus, in th worst cas, th xpctd lngth of th schdul at most 2L (1 1/)V. Thus th approximation ratio is at most f(v ) = 2L (1 1/)V L + l/2 V/2 It can b vrifid that for l 2 1L, f(v ) in monotonically incrasing in V. Thus sinc V l th approximation ratio has maximum valu at most 2 2(1 1/)δ with l = V = 2δL. Choosing δ = , w s that 2 2(1 1/)δ = 1 + δ and hnc in ithr of th thr cass abov, w obtain a schdul of lngth at most min{2 2(1 1/)δ, 1 + δ, /2}Cmax. This implis th dsird rsult.

4 3.3 Proof of Thorm 3.1W first considr a linar programming formulation for a rlaxation of th problm. W viw th rlaxd problm as follows: Considr a charactr S i (j) and imagin travrsing th string S on charactr at a tim starting from S(1) and going all th way to S(l). At stp k w match a fraction x i,j,k of S i (j) to S(k), and thn w discard (i.., lav unmatchd) a fraction y i,j,k of S i (j) bfor moving on to stag k + 1. W also assum that for ach i, j, thr is a stp 0, whr w can discard a fraction y i,j,0 of S i (j). Considr th following linar program: (3.1) k 1 (3.2) (3.3) (3.4) (3.5) (3.6) Maximiz k 1 (x i,j 1,t + y i,j 1,t ) t=0 n l i l i=1 j=1 k=1 x i,j,k (x i,j,t + y i,j,t ) x i,j,k 0 {i, j, k} t=0 l= S x i,j,k 1 k i,j x i,j,k = 0, S i (j) S(k) y i,j,k 0 x i,j,k 0 (x i,j,t + y i,j,t ) = 1 {i, j} t=0 Th first st of constraints modls th prcdnc constraints implid by th ordring of th charactrs in th strings. Thy say that th amount of S i (j) matchd up to S(k) and discardd up to S(k 1) is no mor than total amount of S i (j 1) matchd and discardd up to S(k 1). Th scond st of constraints nsurs that vry charactr is S is matchd to a total amount of at most 1. Th third st of constraint nsurs that 0s ar not matchd to 1s and vic vrsa. Th fourth and fifth sts of constraints nsur that x i,j,k and y i,j,k ar non-ngativ. Th final st of constraints say that ach charactr S i (j) is ithr matchd or discardd. W now show that th LP optimum is an uppr bound on th intgral optimum solution. Considr any intgral solution (I), and suppos that S i (j a ) is matchd to S(k a ), for a = 1,..., b. Thn for a = 1,..., b w st x i,ja,k a = 1 and x i,ja,t = 0 for all t k a. W also st y i,ja,t = 0 for all t. For j such that j a < j < j a+1, w st y i,j,ka = 1 and x i,j,t = 0 for all t. If j < j 1, w st y i,j,0 = 1. Clarly th valu of th objctiv function for this assignmnt of variabls is xactly th numbr of matchs in I. W nd only to show that this assignmnt satisfis all th constraints. Sinc x i,j,k and y i,j,k ar {0, 1} variabls in this assignmnt and vry S i (j) is ithr matchd or discardd, th last thr sts of constraints ar trivially satisfid. Th third st of constraints is satisfid sinc I dos not match a 0 to a 1 or vic vrsa. Th scond st of constraints holds bcaus ach S(k) is matchd to at most on charactr among th S i. For th first st of constraints, sinc x i,j,k and y i,j,k ar {0, 1}, such a constraint is violatd only if S i (j 1) has not bn matchd or discardd by stp k 1 and ithr 4 1. S i (j) is matchd at stp k. 2. S i (j) is matchd or discardd by stp k 1 or soonr. Howvr, it is asy to s that if S i (j) is matchd at to S(k), thn ithr j 1 was matchd to S(h) for h < k or discardd at stp g, whr g is th last stp bfor k whr som charactr from S i was matchd. If S i (j) is matchd or discardd by stp k 1 or soonr, it is trivial to s by construction that S i (j 1) is matchd or discardd bfor stp k 1 as wll. W now giv a randomizd procdur to round this LP solution. W will show that it producs a fasibl intgral solution which has a numbr of matchs qual to at last 1 1/ tims th optimum LP valu in xpctation. It is usful to viw th LP solution in th following quivalnt way. Lt 0 s i,j,k 1 dnot th xtnt to which S i (j) has bn matchd to S(1),..., S(k 1) or discardd during th first k 1 stps, i.. s i,j,k = k 1 t=1 (x i,j,t + y i,j,t ). Lt v i,j,k = s i,j,k + x i,j,k ; thus v i,j,k s i,j,k is xactly th xtnt to which S i (j) is matchd to S(k). Also not that y i,j,k = s i,j,k+1 v i,j,k. Not that th first st of constraints implis that s i,j 1,k v i,j,k. Rounding Procdur: 1. For ach string S i choos u i [0, 1] uniformly at random. 2. For ach i, j assign S i (j) to S(k) if and only if s i,j,k u i < v i,j,k. 3. Lt N(k) dnot th numbr of charactrs assignd to S(k) at th nd of th prvious stp. N(k) is a random variabl. If N(k) = 0, S(k) is not matchd to any charactr. If N(k) = 1, w match S(k) to th uniqu S i (j) assignd to it. If N(k) 2, w arbitrarily match S(k) to on of th charactrs, and discard th rmaining N(k) 1 charactrs (nvr to b matchd again).

5 W now study th proprtis of th obtaind solution. Givn two opn intrvals of numbrs I 1 = (l 1, u 1 ) and I 2 = (l 2, u 2 ), w say that I 1 > I 2 if l 1 u 2, i.. ach lmnt in I 1 is gratr than vry lmnt than I 2. Fix i and j. Sinc x i,j,k 0 and y i,j,k 0 it trivially follows that v i,j,k s i,j,k+1 and hnc th intrvals I k = (s i,j,k, v i,j,k ) ar pairwis disjoint. Thus, no S i (j) can b assignd to two or mor charactrs in S. At th nd of th third rounding stp, no S(k) is matchd to two or mor charactrs. Thus, to show th validity of th solution producd aftr rounding, w nd only to show that no prcdnc constraints ar violatd for any S i. Lmma 3.1. For a fixd i and k, lt I j dnot th (possibly mpty) intrval (s i,j,k, v i,j,k ). Thn, I j1 > I j2 for all j 1 < j 2 or quivalntly, v i,j2,k s i,j1,k. In particular this implis that I j1 I j2 = for all 1 j 1 < j 2 l i Proof. It suffics to show that I j 1 > I j for all j. By th first st of constraints w know that v i,j,k s i,j 1,k for all i, j, k. This implis th lmma. W now show that no prcdnc constraints will b violatd for any S i. Suppos S i (j 1 ) is assignd to S(k 1 ) and S i (j 2 ) to S(k 2 ) for j 1 < j 2 and k 1 k 2. Sinc w us th sam random u i for all charactrs in th string S i, it must b th cas that u i (s i,j1,k 1, v i,j1,k 1 ) and u i (s i,j2,k 2, v i,j2,k 2 ), which implis that v i,j2,k 2 > s i,j1,k 1. Sinc v i,j,k is monotonically incrasing in k, and k 2 k 1, this implis that v i,j2,k 1 > s i,j1,k 1. But this violats lmma 3.1. Thus w hav shown that th rounding producs a fasibl solution. W now analyz th quality of th solution. Lmma 3.2. Lt V dnot th optimum LP valu. Th rounding procdur matchs at last V (1 1/) charactrs in S in xpctation. Proof. Lt us considr th solution obtaind aftr th first two stps of rounding. Lt N(i, k) b a random variabl that dnots th numbr of charactrs from S i that ar assignd to th charactr S(k). Lt p i,k dnot j x i,j,k = j (v i,j,k s i,j,k ). By Lmma 3.1, w know that th intrvals I j = (s i,j,k, v i,j,k ) ar disjoint. Hnc th probability that som charactr of S i is matchd to S(k) is xactly qual to th probability that u i j I j, which is xactly p i,k. Thus N(i, k) is a Brnoulli random variabl with paramtr p i,k, that is, it is 1 with probability p i,k and is 0 othrwis. Thus, N(k) is th sum of m Brnoulli random variabls with paramtrs p 1,k,..., p n,k. Lt I 0 dnot th 5 random variabl that dnot th numbr of charactrs in S that hav zro matchs. Thus, I 0 = {k : N(k) = 0}. Clarly, th numbr of matchs in th LP is i,j,k x i,j,k = i,k p i,k. Th numbr of matchs in th intgral solution is th lngth of S minus th numbr of non-matchs. Thus, th numbr of matchs is l I 0. Now P r [N(k) = 0] = Π n i=1(1 p i,k ) n i=1 p i,k. Thus, E[I 0 ] l numbr of matchs is k=1 n l E[I 0 ] l (1 n k=1 (1 1/) i=1 p i,k Thus th xpctd l k=1 i=1 i=1 p i,k ) n p i,k Th last stp holds sinc 1 x x(1 1/) for 0 x 1. Not that n i=1 p i,k 1 follows from constraint st 2. This complts th proof of thorm An improvd bound for an arbitrary numbr of machins 4.1 An O(log m/ log log m)-approximation algorithm. In this sction w considr a simpl randomizd algorithm prviously considrd [12, 19, 7]. Namly, th algorithm chooss an intgr numbr t j btwn 0 and L 1 indpndntly at random for ach job J j J. It thn constructs an infasibl schdul procssing ach job J j in no-wait fashion in th tim intrval [t j, l j + t j 1]. Th problm with this schdul is that in som tim stps, som machins must procss mor than on job ach. Lt τ it b th numbr of oprations assignd to a tim stp t on machin M i by th abov randomizd procdur. To obtain a fasibl schdul from th infasibl on constructd by th randomizd shifting procdur, w xpand vry tim stp t into max i=1,...,m τ it stps. Sinc all oprations assignd to th sam stp blong to diffrnt jobs, w can fasibly schdul all such oprations using this intrval of lngth max i=1,...,m τ it. Finally, w concatnat th schduls so obtaind for all tim stps. Th total lngth of th final fasibl schdul is at most L+l t=1 max i=1,...,m τ it (whr th uppr limit of summation corrsponds to th uppr bound of L + l on th lngth of th infasibl schdul). By Chrnoff bounds it can b shown that with high probability max i,t τ it = O(log ml/ log log ml) [12,

6 19, 7]. Thrfor, with high probability th lngth of th final schdul is O(log ml/ log log ml) max{l, l}. Instad of using this high probability rsult, w will try to stimat th xpctd schdul lngth which is E( L+l t=1 max i=1,...,m τ it ) = L+l t=1 E(max i=1,...,m τ it ). W will do this using th following lmma. Th tchniqus ar standard and similar rsults can b found in [8, 14]. W dfr th proof of this lmma to Sction 6. Lmma 4.1. (Non-uniform Balls and Bins) Suppos w hav m bins and n balls. Evry ball j chooss a bin i at random with probability λ ij, i.. m i=1 λ ij 1. Th xpctd numbr of balls in vry n bin is at most on, i.. j=1 λ ij 1. Thn th xpctd maximum numbr of balls in any of th m bins is O(log m/ log log m). Lmma 4.1 immdiatly implis an uppr bound for L+l t=1 E(max i=1,...,m τ it ). In vry tim intrval of unit lngth w hav an instanc of th balls and bins problm with m bins and n balls. Ball j landing in bin i corrsponds to an opration of job J j bing schduld on machin M i at tim stp t by th randomizd shifting procdur. Lt p ijt b th probability of that vnt. Sinc our algorithm procsss at most on opration of vry job in any tim unit w hav that m i=1 p ijt 1. On th othr sid p ijt l ij /L and thrfor n j=1 p ijt 1. Thrfor, E(max i=1,...,m τ it ) = O(log m/ log log m) and w obtain th following. Thorm 4.1. Th xpctd makspan of th fasibl schdul obtaind by th abov randomizd algorithm is O(log m/ log log m) max{l, l}. 4.2 An approximation algorithm with an additiv prformanc guarant.our scond randomizd algorithm also chooss random intgr shifts t j in th intrval [0, L 1] for vry job J j. Th diffrnc is that instad of procssing vry job in th tim intrval [t j, t j + l j ], w procss job J j in th tim intrval [t j, t j + l j K ε log m 1] with qual dlays of K ε log m btwn conscutiv oprations of th sam job whr K ε = 4/ε 2 and ε > 0 is an arbitrary prcision paramtr. This schdul has lngth at most L + (K ε log m)l. This schdul also may b infasibl, but unlik th nowait schdul from th prvious sction w will show it can b transformd into a fasibl schdul of lngth (1 + ε)l + (1 + ε)(k ε log m)l. W now show how to transform an infasibl schdul into a fasibl on with (1 + ε) factor incras in th xpctd lngth of th schdul. Considr an infasibl schdul obtaind by th modifid randomizd shifting procdur. This schdul has lngth at most L + (K ε log m)l. W split th tim intrval 6 [0, L + (K ε log m)l] into conscutiv intrvals of lngth K ε log m; th last intrval may hav smallr lngth. Th main ida is that sinc ach intrval can contain at most on opration from ach job, th ordr of oprations within an intrval dos not mattr. In particular, any opration within an intrval can b schduld in any ordr in that intrval. Thus, it suffics to show that in any intrval, th xpctd numbr of oprations on th machin with maximum oprations is (1+ɛ)K ε log m. This will imply that all th oprations in an intrval can b fasibly schduld in xpctd in O(1 + ɛ)k ε log m tim stps. To bound th xpctd numbr of oprations in an intrval on th maximally loadd machin, w us th following variant of th balls and bins lmma, th proof of which can b found in Sction 6. Lmma 4.2. Lt 0 < ε < 1 and K ε = 4/ε 2. Suppos w hav m bins and n balls. Evry ball j chooss a bin m i at random with probability λ ij, i.. i=1 λ ij 1. Th xpctd numbr of balls in vry bin is at most n j=1 λ ij K ε log m. Thn th xpctd maximum numbr of balls in any of th m bins is (1+2ε)K ε log m. Thus w obtain th following. Thorm 4.2. Th xpctd makspan of th fasibl schdul obtaind by th abov randomizd algorithm is (1 + ε)l + O(log m)l. Rmark. Applying th sam algorithm for th nonprmptiv job shop schduling problm with gnral procssing tims and using dlays of (K ε log m)p max btwn conscutiv oprations of th sam job, whr p max is maximum procssing tim in th instanc, w can gt a fasibl schdul of lngth at most (1+2ε)L+ O(µ log m)p max, whr µ is maximum numbr of oprations pr job and a hiddn constant dpnds on ε. 5 A (2 + ε)-approximation for any constant numbr of machins W split th st of jobs into two sts: L = {J j l j εl/(k ε log m) is th st of big jobs and S = J \ L is th st of small jobs. Obsrv that sinc th numbr of machins is constant th numbr of big jobs is O(log m) which is a constant. Th st of small jobs is schduld by using th randomizd algorithm from th prvious sction. Thorm 4.2 guarants that th schdul lngth is at most (1 + O(ε)) max{l, l}. On th othr hand, th st of big jobs can b schduld optimally using a straightforward dynamic program, that for ach tim stp and for ach job stors how many oprations hav bn schduld thus far in th partial schdul [1]. In gnral, this givs a psudopolynomial tim algorithm as th procssing tims could

7 b xponntially larg. Howvr, for th prmptiv job shop problm, th lngth of th job is qual to th numbr of oprations and hnc polynomial in th siz of th input. Evn if w assum a tightr ncoding whr th numbr of conscutiv unit lngth of oprations on a machin is ncodd in binary, th job lngth can b mad polynomial using standard rounding tchniqus. Thrfor, w can schdul th big jobs with makspan at most Cmax (or (1 + ε)cmax if w includ th factor w los whn w apply rounding and scaling to dcras numbr of unit lngth oprations). Concatnating th two schduls, w obtain a schdul of lngth at most (2 + ε)cmax. 6 Proofs of Lmmas 4.1 and 4.2 W us th following vrsion of Chrnoff bounds as givn on pag 267, Corollary A.1.10, [2]. Lmma 6.1. Suppos X 1,..., X n, ar 0-1 random variabls, such that P r[x i = 1] = p i. Lt X = n i=1 p i and X = n i=1 X i. Thn P r[x X a] a (a+ X) ln(1+a/ X) W will also nd th following corollary. Lmma 6.2. P r[x X a min(1/5,a/4 X) a] Proof. Lt x = X/a. Thn th right-hand sid of th inquality in Lmma 6.1 can b writtn as a((x+1) ln(1+1/x) 1). Now, (x + 1) ln(1 + 1/x) 1 is dcrasing in x. At x = 2, its valu is 3 ln 5/3 1 1/5. For x > 2, it is at last (x + 1)(1/x 1/2x 2 ) 1 = 1/2x 1/2x 2 1/4x. Hnc, by th union bound, a ln ln m/2 P r[b > 1 + a] mp r[b i 1 + a] = m Now w hav E[B] = P r[b x] x=1 3 ln m ln ln m + 3 ln m ln ln m + m a 3 ln m ln ln m a 3 ln m ln ln m 3 ln m ln ln m + 2 m 1/2 ln ln m = O Proof. (of Lmma 4.2) By Lmma 6.2 P r[b 1 + a] a ln ln m/2 ( ) ln m ln ln m P r[b i K ε ln m + δ] P r[b i E[B i ] + δ] δ min(1/5,δ/4 X) W will only b intrstd in δ εk ε ln m and small ε > 0. Thus, P r[b i K ε ln m+δ] δ ε/4. and hnc by th union bound P r[b K ε ln m + δ] m δ ε/4. Thus, w hav that E[B] (1 + ε)k ε ln m + m δ ε/4 δ>εk ε ln m (1 + ε)k ε ln m + m 4 ε ε2 K ε ln m/4 Rcalling that K ε = 4/ε 2, w hav that E[B] (1 + ε)k ε ln m + 4/ɛ (1 + ɛ)k ε ln m + ɛk ε (1 + 2ε)K ε ln m Proof. (of Lmma 4.1) Lt B ij b a 0-1 random variabl that is 1 iff ball j gos to bin i. Lt B i dnot th numbr of balls in bin i. Finally, lt B = max i B i. As E[B i ] 1, by Lmma 6.1 w hav that P r[b i 1 + a] P r[b i E[B i ] + a] For a 3 ln m ln ln m, w hav that a (a+e[bi]) ln(1+a/e[bi]) a a ln(1+a) ln(1 + a) 1 ln(ln m/ ln ln m) (ln ln m)/2 which implis that, a ln ln m/2 P r[b i 1 + a] 7 7 Opn Problms Two outstanding opn qustions that rmain for th gnral prmptiv job shop problm ar: 1. Is thr an O(1)-approximation for th gnral prmptiv job shop problm with an arbitrary numbr of machins? As all lowr bounds on th makspan ar ssntially O(max{L, l}), a vry intrsting rsult would b to show an instanc such that th optimal makspan is not within a constant factor of max{l, l}. It is known that for nonprmptiv job shop schduling thr is no O(1)-approximation algorithm with rspct to max{l, l} vn for acyclic instancs [6]. W bliv that rsolving this qustion for prmptiv schduls would rquir significant nw insights into thir structur.

8 2. Is thr a polynomial tim approximation schm for th cas of a constant numbr of machins? Our algorithm in this papr givs a PTAS if l ɛl/ log m. On th othr hand, if w rstrict th problm to instancs with only a fw long jobs that compris all but an ε fraction of th load, thn w can gt a PTAS via dynamic programming. Making progrss on th gnral cas, in which both long jobs and short jobs may mak up significant portions of th load, would rquir an undrstanding of how small jobs and long jobs ar mixd togthr in an optimal schdul. A first stp in this dirction would b to undrstand th approximability of th string matching problm from Sction 3. Rfrncs [1] S. B. Akrs, A graphical approach to production schduling problms, Oprations Rsarch 4 (1956), [2] N. Alon, and J. Spncr, Th Probabilistic Mthod. John Wily & Sons, [3] E.J. Andrson, T.S. Jayram, and T. Kimbrl. Tightr Bounds on Prmptiv Job Shop Schduling with Two Machins. Computing 67 (2001), pp [4] P. Baptist, J. Carlir, A. Kononov, M. Quyrann, S. Svastianov and M. Sviridnko, Structural Proprtis of Prmptiv Schduls, submittd for publication. [5] A. Czumaj and C. Schidlr, A Nw Algorithmic Approach to th Gnral Lovasz Local Lmma with Applications to Schduling and Satisfiability Problms, Proc. 32 ACM Symposium on Thory of Computing (STOC), [6] U. Fig and C. Schidlr, Improvd bounds for acyclic job shop schduling. Combinatorica 22 (2002), no. 3, [7] L.A. Goldbrg, M. Patrson, A. Srinivasan, and E. Swdyk. Bttr approximation guarants for jobshop schduling. SIAM J. Discrt Math. 14 (2001), [8] G. Gonnt, Expctd lngth of th longst prob squnc in hash cod sarching. J. Assoc. Comput. Mach. 28 (1981), no. 2, [9] T. Gonzalz and S. Sahni. Flowshop and jobshop schduls: complxity and approximation. Oprations Rsarch 26, pp , [10] K. Jansn, R. Solis-Oba, and M. Sviridnko. Makspan minimization in job shops: a linar tim approximation schm. SIAM J. Discrt Math. 16 (2003), [11] E.L. Lawlr, J.K. Lnstra, A.H.G. Rinnooy Kan, and D.B. Shmoys. Squncing and Schduling: Algorithms and Complxity. In S.C. Gravs, A.H.G. Rinnooy Kan, and P.H. Zipkin (ds.), Logistics of Production and Invntory, Handbooks in Oprations Rsarch and Managmnt Scinc 4, North-Holland, Amstrdam, 1993, [12] F.T. Lighton, B. Maggs, and S. Rao. Packt routing and jobshop schduling in O(congstion + dilation) stps. Combinatorica 14, pp , [13] F.T. Lighton, B.M. Maggs, and A.W. Richa, Fast algorithms for finding O(congstion + dilation) packt routing schduls. Combinatorica 19 (1999) [14] M. Raab and A. Stgr, Balls into bins a simpl and tight analysis. Randomization and approximation tchniqus in computr scinc (Barclona, 1998), , Lctur Nots in Comput. Sci., 1518, Springr, Brlin, [15] S.V. Svastianov and G.J. Wogingr. Makspan minimization in prmptiv two machin job shops. Computing 60 (1998), [16] S. Svast janov, On som gomtric mthods in schduling thory: a survy. Discrt Appl. Math. 55 (1994), no. 1, [17] S. Svastianov, Bounding algorithm for th routing problm with arbitrary paths and altrnativ srvrs, Cybrntics 22 (1986), pp [18] Y. Sotskov and N. Shakhlvich, NP-hardnss of shopschduling problms with thr jobs. Discrt Appl. Math. 59 (1995), no. 3, [19] D. Shmoys, C. Stin, and J. Win. Improvd Approximation Algorithms for Shop Schduling Problms. SIAM Journal on Computing 23:3, , [20] D. P. Williamson, L. A. Hall, J. A. Hoogvn, C. A. J. Hurkns, J. K. Lnstra, S. V. Svast janov and D. B. Shmoys, Short shop schduls, Opration Rsarch 45 (1997), pp

CPS 220 Theory of Computation REGULAR LANGUAGES. Regular expressions

CPS 220 Theory of Computation REGULAR LANGUAGES. Regular expressions CPS 22 Thory of Computation REGULAR LANGUAGES Rgular xprssions Lik mathmatical xprssion (5+3) * 4. Rgular xprssion ar built using rgular oprations. (By th way, rgular xprssions show up in various languags:

More information

A Note on Approximating. the Normal Distribution Function

A Note on Approximating. the Normal Distribution Function Applid Mathmatical Scincs, Vol, 00, no 9, 45-49 A Not on Approimating th Normal Distribution Function K M Aludaat and M T Alodat Dpartmnt of Statistics Yarmouk Univrsity, Jordan Aludaatkm@hotmailcom and

More information

QUANTITATIVE METHODS CLASSES WEEK SEVEN

QUANTITATIVE METHODS CLASSES WEEK SEVEN QUANTITATIVE METHODS CLASSES WEEK SEVEN Th rgrssion modls studid in prvious classs assum that th rspons variabl is quantitativ. Oftn, howvr, w wish to study social procsss that lad to two diffrnt outcoms.

More information

Question 3: How do you find the relative extrema of a function?

Question 3: How do you find the relative extrema of a function? ustion 3: How do you find th rlativ trma of a function? Th stratgy for tracking th sign of th drivativ is usful for mor than dtrmining whr a function is incrasing or dcrasing. It is also usful for locating

More information

Job shop scheduling with unit processing times

Job shop scheduling with unit processing times Job shop scheduling with unit processing times Nikhil Bansal IBM T.J. Watson Research Center,P.O. Box 218, Yorktown Heights, NY 10598 email: nikhil@us.ibm.com Tracy Kimbrel IBM T.J. Watson Research Center,P.O.

More information

by John Donald, Lecturer, School of Accounting, Economics and Finance, Deakin University, Australia

by John Donald, Lecturer, School of Accounting, Economics and Finance, Deakin University, Australia Studnt Nots Cost Volum Profit Analysis by John Donald, Lcturr, School of Accounting, Economics and Financ, Dakin Univrsity, Australia As mntiond in th last st of Studnt Nots, th ability to catgoris costs

More information

Lecture 20: Emitter Follower and Differential Amplifiers

Lecture 20: Emitter Follower and Differential Amplifiers Whits, EE 3 Lctur 0 Pag of 8 Lctur 0: Emittr Followr and Diffrntial Amplifirs Th nxt two amplifir circuits w will discuss ar ry important to lctrical nginring in gnral, and to th NorCal 40A spcifically.

More information

Lecture 3: Diffusion: Fick s first law

Lecture 3: Diffusion: Fick s first law Lctur 3: Diffusion: Fick s first law Today s topics What is diffusion? What drivs diffusion to occur? Undrstand why diffusion can surprisingly occur against th concntration gradint? Larn how to dduc th

More information

(Analytic Formula for the European Normal Black Scholes Formula)

(Analytic Formula for the European Normal Black Scholes Formula) (Analytic Formula for th Europan Normal Black Schols Formula) by Kazuhiro Iwasawa Dcmbr 2, 2001 In this short summary papr, a brif summary of Black Schols typ formula for Normal modl will b givn. Usually

More information

New Basis Functions. Section 8. Complex Fourier Series

New Basis Functions. Section 8. Complex Fourier Series Nw Basis Functions Sction 8 Complx Fourir Sris Th complx Fourir sris is prsntd first with priod 2, thn with gnral priod. Th connction with th ral-valud Fourir sris is xplaind and formula ar givn for convrting

More information

Version 1.0. General Certificate of Education (A-level) January 2012. Mathematics MPC3. (Specification 6360) Pure Core 3. Final.

Version 1.0. General Certificate of Education (A-level) January 2012. Mathematics MPC3. (Specification 6360) Pure Core 3. Final. Vrsion.0 Gnral Crtificat of Education (A-lvl) January 0 Mathmatics MPC (Spcification 660) Pur Cor Final Mark Schm Mark schms ar prpard by th Principal Eaminr and considrd, togthr with th rlvant qustions,

More information

Upper Bounding the Price of Anarchy in Atomic Splittable Selfish Routing

Upper Bounding the Price of Anarchy in Atomic Splittable Selfish Routing Uppr Bounding th Pric of Anarchy in Atomic Splittabl Slfish Routing Kamyar Khodamoradi 1, Mhrdad Mahdavi, and Mohammad Ghodsi 3 1 Sharif Univrsity of Tchnology, Thran, Iran, khodamoradi@c.sharif.du Sharif

More information

EFFECT OF GEOMETRICAL PARAMETERS ON HEAT TRANSFER PERFORMACE OF RECTANGULAR CIRCUMFERENTIAL FINS

EFFECT OF GEOMETRICAL PARAMETERS ON HEAT TRANSFER PERFORMACE OF RECTANGULAR CIRCUMFERENTIAL FINS 25 Vol. 3 () January-March, pp.37-5/tripathi EFFECT OF GEOMETRICAL PARAMETERS ON HEAT TRANSFER PERFORMACE OF RECTANGULAR CIRCUMFERENTIAL FINS *Shilpa Tripathi Dpartmnt of Chmical Enginring, Indor Institut

More information

5 2 index. e e. Prime numbers. Prime factors and factor trees. Powers. worked example 10. base. power

5 2 index. e e. Prime numbers. Prime factors and factor trees. Powers. worked example 10. base. power Prim numbrs W giv spcial nams to numbrs dpnding on how many factors thy hav. A prim numbr has xactly two factors: itslf and 1. A composit numbr has mor than two factors. 1 is a spcial numbr nithr prim

More information

Rent, Lease or Buy: Randomized Algorithms for Multislope Ski Rental

Rent, Lease or Buy: Randomized Algorithms for Multislope Ski Rental Rnt, Las or Buy: Randomizd Algorithms for Multislop Ski Rntal Zvi Lotkr zvilo@cs.bgu.ac.il Dpt. of Comm. Systms Enginring Bn Gurion Univrsity Br Shva Isral Boaz Patt-Shamir Dror Rawitz {boaz, rawitz}@ng.tau.ac.il

More information

Adverse Selection and Moral Hazard in a Model With 2 States of the World

Adverse Selection and Moral Hazard in a Model With 2 States of the World Advrs Slction and Moral Hazard in a Modl With 2 Stats of th World A modl of a risky situation with two discrt stats of th world has th advantag that it can b natly rprsntd using indiffrnc curv diagrams,

More information

Parallel and Distributed Programming. Performance Metrics

Parallel and Distributed Programming. Performance Metrics Paralll and Distributd Programming Prformanc! wo main goals to b achivd with th dsign of aralll alications ar:! Prformanc: th caacity to rduc th tim to solv th roblm whn th comuting rsourcs incras;! Scalability:

More information

The example is taken from Sect. 1.2 of Vol. 1 of the CPN book.

The example is taken from Sect. 1.2 of Vol. 1 of the CPN book. Rsourc Allocation Abstract This is a small toy xampl which is wll-suitd as a first introduction to Cnts. Th CN modl is dscribd in grat dtail, xplaining th basic concpts of C-nts. Hnc, it can b rad by popl

More information

Econ 371: Answer Key for Problem Set 1 (Chapter 12-13)

Econ 371: Answer Key for Problem Set 1 (Chapter 12-13) con 37: Answr Ky for Problm St (Chaptr 2-3) Instructor: Kanda Naknoi Sptmbr 4, 2005. (2 points) Is it possibl for a country to hav a currnt account dficit at th sam tim and has a surplus in its balanc

More information

Foreign Exchange Markets and Exchange Rates

Foreign Exchange Markets and Exchange Rates Microconomics Topic 1: Explain why xchang rats indicat th pric of intrnational currncis and how xchang rats ar dtrmind by supply and dmand for currncis in intrnational markts. Rfrnc: Grgory Mankiw s Principls

More information

http://www.wwnorton.com/chemistry/tutorials/ch14.htm Repulsive Force

http://www.wwnorton.com/chemistry/tutorials/ch14.htm Repulsive Force ctivation nrgis http://www.wwnorton.com/chmistry/tutorials/ch14.htm (back to collision thory...) Potntial and Kintic nrgy during a collision + + ngativly chargd lctron cloud Rpulsiv Forc ngativly chargd

More information

Sharp bounds for Sándor mean in terms of arithmetic, geometric and harmonic means

Sharp bounds for Sándor mean in terms of arithmetic, geometric and harmonic means Qian t al. Journal of Inqualitis and Applications (015) 015:1 DOI 10.1186/s1660-015-0741-1 R E S E A R C H Opn Accss Sharp bounds for Sándor man in trms of arithmtic, gomtric and harmonic mans Wi-Mao Qian

More information

Mathematics. Mathematics 3. hsn.uk.net. Higher HSN23000

Mathematics. Mathematics 3. hsn.uk.net. Higher HSN23000 hsn uknt Highr Mathmatics UNIT Mathmatics HSN000 This documnt was producd spcially for th HSNuknt wbsit, and w rquir that any copis or drivativ works attribut th work to Highr Still Nots For mor dtails

More information

Factorials! Stirling s formula

Factorials! Stirling s formula Author s not: This articl may us idas you havn t larnd yt, and might sm ovrly complicatd. It is not. Undrstanding Stirling s formula is not for th faint of hart, and rquirs concntrating on a sustaind mathmatical

More information

Constraint-Based Analysis of Gene Deletion in a Metabolic Network

Constraint-Based Analysis of Gene Deletion in a Metabolic Network Constraint-Basd Analysis of Gn Dltion in a Mtabolic Ntwork Abdlhalim Larhlimi and Alxandr Bockmayr DFG-Rsarch Cntr Mathon, FB Mathmatik und Informatik, Fri Univrsität Brlin, Arnimall, 3, 14195 Brlin, Grmany

More information

Traffic Flow Analysis (2)

Traffic Flow Analysis (2) Traffic Flow Analysis () Statistical Proprtis. Flow rat distributions. Hadway distributions. Spd distributions by Dr. Gang-Ln Chang, Profssor Dirctor of Traffic safty and Oprations Lab. Univrsity of Maryland,

More information

In the previous two chapters, we clarified what it means for a problem to be decidable or undecidable.

In the previous two chapters, we clarified what it means for a problem to be decidable or undecidable. Chaptr 7 Computational Complxity 7.1 Th Class P In th prvious two chaptrs, w clarifid what it mans for a problm to b dcidabl or undcidabl. In principl, if a problm is dcidabl, thn thr is an algorithm (i..,

More information

Performance Evaluation

Performance Evaluation Prformanc Evaluation ( ) Contnts lists availabl at ScincDirct Prformanc Evaluation journal hompag: www.lsvir.com/locat/pva Modling Bay-lik rputation systms: Analysis, charactrization and insuranc mchanism

More information

The Constrained Ski-Rental Problem and its Application to Online Cloud Cost Optimization

The Constrained Ski-Rental Problem and its Application to Online Cloud Cost Optimization 3 Procdings IEEE INFOCOM Th Constraind Ski-Rntal Problm and its Application to Onlin Cloud Cost Optimization Ali Khanafr, Murali Kodialam, and Krishna P. N. Puttaswam Coordinatd Scinc Laborator, Univrsit

More information

MAXIMAL CHAINS IN THE TURING DEGREES

MAXIMAL CHAINS IN THE TURING DEGREES MAXIMAL CHAINS IN THE TURING DEGREES C. T. CHONG AND LIANG YU Abstract. W study th problm of xistnc of maximal chains in th Turing dgrs. W show that:. ZF + DC+ Thr xists no maximal chain in th Turing dgrs

More information

ME 612 Metal Forming and Theory of Plasticity. 6. Strain

ME 612 Metal Forming and Theory of Plasticity. 6. Strain Mtal Forming and Thory of Plasticity -mail: azsnalp@gyt.du.tr Makin Mühndisliği Bölümü Gbz Yüksk Tknoloji Enstitüsü 6.1. Uniaxial Strain Figur 6.1 Dfinition of th uniaxial strain (a) Tnsil and (b) Comprssiv.

More information

SPECIAL VOWEL SOUNDS

SPECIAL VOWEL SOUNDS SPECIAL VOWEL SOUNDS Plas consult th appropriat supplmnt for th corrsponding computr softwar lsson. Rfr to th 42 Sounds Postr for ach of th Spcial Vowl Sounds. TEACHER INFORMATION: Spcial Vowl Sounds (SVS)

More information

Budget Optimization in Search-Based Advertising Auctions

Budget Optimization in Search-Based Advertising Auctions Budgt Optimization in Sarch-Basd Advrtising Auctions ABSTRACT Jon Fldman Googl, Inc. Nw York, NY jonfld@googl.com Martin Pál Googl, Inc. Nw York, NY mpal@googl.com Intrnt sarch companis sll advrtismnt

More information

5.4 Exponential Functions: Differentiation and Integration TOOTLIFTST:

5.4 Exponential Functions: Differentiation and Integration TOOTLIFTST: .4 Eponntial Functions: Diffrntiation an Intgration TOOTLIFTST: Eponntial functions ar of th form f ( ) Ab. W will, in this sction, look at a spcific typ of ponntial function whr th bas, b, is.78.... This

More information

Expert-Mediated Search

Expert-Mediated Search Exprt-Mdiatd Sarch Mnal Chhabra Rnsslar Polytchnic Inst. Dpt. of Computr Scinc Troy, NY, USA chhabm@cs.rpi.du Sanmay Das Rnsslar Polytchnic Inst. Dpt. of Computr Scinc Troy, NY, USA sanmay@cs.rpi.du David

More information

Production Costing (Chapter 8 of W&W)

Production Costing (Chapter 8 of W&W) Production Costing (Chaptr 8 of W&W).0 Introduction Production costs rfr to th oprational costs associatd with producing lctric nrgy. Th most significant componnt of production costs ar th ful costs ncssary

More information

Policies for Simultaneous Estimation and Optimization

Policies for Simultaneous Estimation and Optimization Policis for Simultanous Estimation and Optimization Migul Sousa Lobo Stphn Boyd Abstract Policis for th joint idntification and control of uncrtain systms ar prsntd h discussion focuss on th cas of a multipl

More information

Fetch. Decode. Execute. Memory. PC update

Fetch. Decode. Execute. Memory. PC update nwpc PC Nw PC valm Mmory Mm. control rad writ Data mmory data out rmmovl ra, D(rB) Excut Bch CC ALU A vale ALU Addr ALU B Data vala ALU fun. valb dste dstm srca srcb dste dstm srca srcb Ftch Dcod Excut

More information

Section 7.4: Exponential Growth and Decay

Section 7.4: Exponential Growth and Decay 1 Sction 7.4: Exponntial Growth and Dcay Practic HW from Stwart Txtbook (not to hand in) p. 532 # 1-17 odd In th nxt two ction, w xamin how population growth can b modld uing diffrntial quation. W tart

More information

Establishing Wireless Conference Calls Under Delay Constraints

Establishing Wireless Conference Calls Under Delay Constraints Establishing Wirlss Confrnc Calls Undr Dlay Constraints Aotz Bar-Noy aotz@sci.brooklyn.cuny.du Grzgorz Malwicz grg@cs.ua.du Novbr 17, 2003 Abstract A prvailing fatur of obil tlphony systs is that th cll

More information

Use a high-level conceptual data model (ER Model). Identify objects of interest (entities) and relationships between these objects

Use a high-level conceptual data model (ER Model). Identify objects of interest (entities) and relationships between these objects Chaptr 3: Entity Rlationship Modl Databas Dsign Procss Us a high-lvl concptual data modl (ER Modl). Idntify objcts of intrst (ntitis) and rlationships btwn ths objcts Idntify constraints (conditions) End

More information

A Multi-Heuristic GA for Schedule Repair in Precast Plant Production

A Multi-Heuristic GA for Schedule Repair in Precast Plant Production From: ICAPS-03 Procdings. Copyright 2003, AAAI (www.aaai.org). All rights rsrvd. A Multi-Huristic GA for Schdul Rpair in Prcast Plant Production Wng-Tat Chan* and Tan Hng W** *Associat Profssor, Dpartmnt

More information

Analyzing the Economic Efficiency of ebaylike Online Reputation Reporting Mechanisms

Analyzing the Economic Efficiency of ebaylike Online Reputation Reporting Mechanisms A rsarch and ducation initiativ at th MIT Sloan School of Managmnt Analyzing th Economic Efficincy of Baylik Onlin Rputation Rporting Mchanisms Papr Chrysanthos Dllarocas July For mor information, plas

More information

SPREAD OPTION VALUATION AND THE FAST FOURIER TRANSFORM

SPREAD OPTION VALUATION AND THE FAST FOURIER TRANSFORM RESEARCH PAPERS IN MANAGEMENT STUDIES SPREAD OPTION VALUATION AND THE FAST FOURIER TRANSFORM M.A.H. Dmpstr & S.S.G. Hong WP 26/2000 Th Judg Institut of Managmnt Trumpington Strt Cambridg CB2 1AG Ths paprs

More information

Projections - 3D Viewing. Overview Lecture 4. Projection - 3D viewing. Projections. Projections Parallel Perspective

Projections - 3D Viewing. Overview Lecture 4. Projection - 3D viewing. Projections. Projections Parallel Perspective Ovrviw Lctur 4 Projctions - 3D Viwing Projctions Paralll Prspctiv 3D Viw Volum 3D Viwing Transformation Camra Modl - Assignmnt 2 OFF fils 3D mor compl than 2D On mor dimnsion Displa dvic still 2D Analog

More information

Combinatorial Prediction Markets for Event Hierarchies

Combinatorial Prediction Markets for Event Hierarchies Combinatorial rdiction Markts for Evnt Hirarchis Mingyu Guo Duk Univrsity Dpartmnt of Computr Scinc Durham, NC, USA mingyu@cs.duk.du David M. nnock Yahoo! Rsarch 111 W. 40th St. 17th Floor Nw York, NY

More information

CPU. Rasterization. Per Vertex Operations & Primitive Assembly. Polynomial Evaluator. Frame Buffer. Per Fragment. Display List.

CPU. Rasterization. Per Vertex Operations & Primitive Assembly. Polynomial Evaluator. Frame Buffer. Per Fragment. Display List. Elmntary Rndring Elmntary rastr algorithms for fast rndring Gomtric Primitivs Lin procssing Polygon procssing Managing OpnGL Stat OpnGL uffrs OpnGL Gomtric Primitivs ll gomtric primitivs ar spcifid by

More information

WORKERS' COMPENSATION ANALYST, 1774 SENIOR WORKERS' COMPENSATION ANALYST, 1769

WORKERS' COMPENSATION ANALYST, 1774 SENIOR WORKERS' COMPENSATION ANALYST, 1769 08-16-85 WORKERS' COMPENSATION ANALYST, 1774 SENIOR WORKERS' COMPENSATION ANALYST, 1769 Summary of Dutis : Dtrmins City accptanc of workrs' compnsation cass for injurd mploys; authorizs appropriat tratmnt

More information

CHAPTER 4c. ROOTS OF EQUATIONS

CHAPTER 4c. ROOTS OF EQUATIONS CHAPTER c. ROOTS OF EQUATIONS A. J. Clark School o Enginring Dpartmnt o Civil and Environmntal Enginring by Dr. Ibrahim A. Aakka Spring 00 ENCE 03 - Computation Mthod in Civil Enginring II Dpartmnt o Civil

More information

Simple and Effective Dynamic Provisioning for Power-Proportional Data Centers

Simple and Effective Dynamic Provisioning for Power-Proportional Data Centers Simpl and Effctiv Dynamic Provisioning for Powr-Proportional Data Cntrs Tan Lu, Minghua Chn, and Lachlan L. H. Andrw Abstract Enrgy consumption rprsnts a significant cost in data cntr opration. A larg

More information

Incomplete 2-Port Vector Network Analyzer Calibration Methods

Incomplete 2-Port Vector Network Analyzer Calibration Methods Incomplt -Port Vctor Ntwork nalyzr Calibration Mthods. Hnz, N. Tmpon, G. Monastrios, H. ilva 4 RF Mtrology Laboratory Instituto Nacional d Tcnología Industrial (INTI) Bunos irs, rgntina ahnz@inti.gov.ar

More information

AP Calculus Multiple-Choice Question Collection 1969 1998. connect to college success www.collegeboard.com

AP Calculus Multiple-Choice Question Collection 1969 1998. connect to college success www.collegeboard.com AP Calculus Multipl-Choic Qustion Collction 969 998 connct to collg succss www.collgboard.com Th Collg Board: Conncting Studnts to Collg Succss Th Collg Board is a not-for-profit mmbrship association whos

More information

Vibrational Spectroscopy

Vibrational Spectroscopy Vibrational Spctroscopy armonic scillator Potntial Enrgy Slction Ruls V( ) = k = R R whr R quilibrium bond lngth Th dipol momnt of a molcul can b pandd as a function of = R R. µ ( ) =µ ( ) + + + + 6 3

More information

Entity-Relationship Model

Entity-Relationship Model Entity-Rlationship Modl Kuang-hua Chn Dpartmnt of Library and Information Scinc National Taiwan Univrsity A Company Databas Kps track of a company s mploys, dpartmnts and projcts Aftr th rquirmnts collction

More information

Financial Mathematics

Financial Mathematics Financial Mathatics A ractical Guid for Actuaris and othr Businss rofssionals B Chris Ruckan, FSA & Jo Francis, FSA, CFA ublishd b B rofssional Education Solutions to practic qustions Chaptr 7 Solution

More information

Remember you can apply online. It s quick and easy. Go to www.gov.uk/advancedlearningloans. Title. Forename(s) Surname. Sex. Male Date of birth D

Remember you can apply online. It s quick and easy. Go to www.gov.uk/advancedlearningloans. Title. Forename(s) Surname. Sex. Male Date of birth D 24+ Advancd Larning Loan Application form Rmmbr you can apply onlin. It s quick and asy. Go to www.gov.uk/advancdlarningloans About this form Complt this form if: you r studying an ligibl cours at an approvd

More information

C H A P T E R 1 Writing Reports with SAS

C H A P T E R 1 Writing Reports with SAS C H A P T E R 1 Writing Rports with SAS Prsnting information in a way that s undrstood by th audinc is fundamntally important to anyon s job. Onc you collct your data and undrstand its structur, you nd

More information

Architecture of the proposed standard

Architecture of the proposed standard Architctur of th proposd standard Introduction Th goal of th nw standardisation projct is th dvlopmnt of a standard dscribing building srvics (.g.hvac) product catalogus basd on th xprincs mad with th

More information

Chapter 10 Function of a Matrix

Chapter 10 Function of a Matrix EE448/58 Vrsion. John Stnsby Chatr Function of a atrix t f(z) b a comlx-valud function of a comlx variabl z. t A b an n n comlxvalud matrix. In this chatr, w giv a dfinition for th n n matrix f(a). Also,

More information

Meerkats: A Power-Aware, Self-Managing Wireless Camera Network for Wide Area Monitoring

Meerkats: A Power-Aware, Self-Managing Wireless Camera Network for Wide Area Monitoring Mrkats: A Powr-Awar, Slf-Managing Wirlss Camra Ntwork for Wid Ara Monitoring C. B. Margi 1, X. Lu 1, G. Zhang 1, G. Stank 2, R. Manduchi 1, K. Obraczka 1 1 Dpartmnt of Computr Enginring, Univrsity of California,

More information

AP Calculus AB 2008 Scoring Guidelines

AP Calculus AB 2008 Scoring Guidelines AP Calculus AB 8 Scoring Guidlins Th Collg Board: Conncting Studnts to Collg Succss Th Collg Board is a not-for-profit mmbrship association whos mission is to connct studnts to collg succss and opportunity.

More information

Keywords Cloud Computing, Service level agreement, cloud provider, business level policies, performance objectives.

Keywords Cloud Computing, Service level agreement, cloud provider, business level policies, performance objectives. Volum 3, Issu 6, Jun 2013 ISSN: 2277 128X Intrnational Journal of Advancd Rsarch in Computr Scinc and Softwar Enginring Rsarch Papr Availabl onlin at: wwwijarcsscom Dynamic Ranking and Slction of Cloud

More information

Basis risk. When speaking about forward or futures contracts, basis risk is the market

Basis risk. When speaking about forward or futures contracts, basis risk is the market Basis risk Whn spaking about forward or futurs contracts, basis risk is th markt risk mismatch btwn a position in th spot asst and th corrsponding futurs contract. Mor broadly spaking, basis risk (also

More information

Abstract. Introduction. Statistical Approach for Analyzing Cell Phone Handoff Behavior. Volume 3, Issue 1, 2009

Abstract. Introduction. Statistical Approach for Analyzing Cell Phone Handoff Behavior. Volume 3, Issue 1, 2009 Volum 3, Issu 1, 29 Statistical Approach for Analyzing Cll Phon Handoff Bhavior Shalini Saxna, Florida Atlantic Univrsity, Boca Raton, FL, shalinisaxna1@gmail.com Sad A. Rajput, Farquhar Collg of Arts

More information

FACULTY SALARIES FALL 2004. NKU CUPA Data Compared To Published National Data

FACULTY SALARIES FALL 2004. NKU CUPA Data Compared To Published National Data FACULTY SALARIES FALL 2004 NKU CUPA Data Compard To Publishd National Data May 2005 Fall 2004 NKU Faculty Salaris Compard To Fall 2004 Publishd CUPA Data In th fall 2004 Northrn Kntucky Univrsity was among

More information

Introduction to Finite Element Modeling

Introduction to Finite Element Modeling Introduction to Finit Elmnt Modling Enginring analysis of mchanical systms hav bn addrssd by driving diffrntial quations rlating th variabls of through basic physical principls such as quilibrium, consrvation

More information

the so-called KOBOS system. 1 with the exception of a very small group of the most active stocks which also trade continuously through

the so-called KOBOS system. 1 with the exception of a very small group of the most active stocks which also trade continuously through Liquidity and Information-Basd Trading on th Ordr Drivn Capital Markt: Th Cas of th Pragu tock Exchang Libor 1ÀPH³HN Cntr for Economic Rsarch and Graduat Education, Charls Univrsity and Th Economic Institut

More information

Precise Memory Leak Detection for Java Software Using Container Profiling

Precise Memory Leak Detection for Java Software Using Container Profiling Distinguishd Papr Prcis Mmory Lak Dtction for Java Softwar Using Containr Profiling Guoqing Xu Atanas Rountv Dpartmnt of Computr Scinc and Enginring Ohio Stat Univrsity {xug,rountv}@cs.ohio-stat.du ABSTRACT

More information

Cloud and Big Data Summer School, Stockholm, Aug., 2015 Jeffrey D. Ullman

Cloud and Big Data Summer School, Stockholm, Aug., 2015 Jeffrey D. Ullman Cloud and Big Data Summr Scool, Stockolm, Aug., 2015 Jffry D. Ullman Givn a st of points, wit a notion of distanc btwn points, group t points into som numbr of clustrs, so tat mmbrs of a clustr ar clos

More information

Category 7: Employee Commuting

Category 7: Employee Commuting 7 Catgory 7: Employ Commuting Catgory dscription This catgory includs missions from th transportation of mploys 4 btwn thir homs and thir worksits. Emissions from mploy commuting may aris from: Automobil

More information

Rural and Remote Broadband Access: Issues and Solutions in Australia

Rural and Remote Broadband Access: Issues and Solutions in Australia Rural and Rmot Broadband Accss: Issus and Solutions in Australia Dr Tony Warrn Group Managr Rgulatory Stratgy Tlstra Corp Pag 1 Tlstra in confidnc Ovrviw Australia s gographical siz and population dnsity

More information

Long run: Law of one price Purchasing Power Parity. Short run: Market for foreign exchange Factors affecting the market for foreign exchange

Long run: Law of one price Purchasing Power Parity. Short run: Market for foreign exchange Factors affecting the market for foreign exchange Lctur 6: Th Forign xchang Markt xchang Rats in th long run CON 34 Mony and Banking Profssor Yamin Ahmad xchang Rats in th Short Run Intrst Parity Big Concpts Long run: Law of on pric Purchasing Powr Parity

More information

On the moments of the aggregate discounted claims with dependence introduced by a FGM copula

On the moments of the aggregate discounted claims with dependence introduced by a FGM copula On th momnts of th aggrgat discountd claims with dpndnc introducd by a FGM copula - Mathiu BARGES Univrsité Lyon, Laboratoir SAF, Univrsité Laval - Hélèn COSSETTE Ecol Actuariat, Univrsité Laval, Québc,

More information

On Resilience of Multicommodity Dynamical Flow Networks

On Resilience of Multicommodity Dynamical Flow Networks On Rsilinc of Multicommodity Dynamical Flow Ntworks Gusta Nilsson, Giacomo Como, and Enrico Loisari bstract Dynamical flow ntworks with htrognous routing ar analyzd in trms of stability and rsilinc to

More information

Free ACA SOLUTION (IRS 1094&1095 Reporting)

Free ACA SOLUTION (IRS 1094&1095 Reporting) Fr ACA SOLUTION (IRS 1094&1095 Rporting) Th Insuranc Exchang (301) 279-1062 ACA Srvics Transmit IRS Form 1094 -C for mployrs Print & mail IRS Form 1095-C to mploys HR Assist 360 will gnrat th 1095 s for

More information

An Broad outline of Redundant Array of Inexpensive Disks Shaifali Shrivastava 1 Department of Computer Science and Engineering AITR, Indore

An Broad outline of Redundant Array of Inexpensive Disks Shaifali Shrivastava 1 Department of Computer Science and Engineering AITR, Indore Intrnational Journal of mrging Tchnology and dvancd nginring Wbsit: www.ijta.com (ISSN 2250-2459, Volum 2, Issu 4, pril 2012) n road outlin of Rdundant rray of Inxpnsiv isks Shaifali Shrivastava 1 partmnt

More information

METHODS FOR HANDLING TIED EVENTS IN THE COX PROPORTIONAL HAZARD MODEL

METHODS FOR HANDLING TIED EVENTS IN THE COX PROPORTIONAL HAZARD MODEL STUDIA OECONOMICA POSNANIENSIA 204, vol. 2, no. 2 (263 Jadwiga Borucka Warsaw School of Economics, Institut of Statistics and Dmography, Evnt History and Multilvl Analysis Unit jadwiga.borucka@gmail.com

More information

A Theoretical Model of Public Response to the Homeland Security Advisory System

A Theoretical Model of Public Response to the Homeland Security Advisory System A Thortical Modl of Public Rspons to th Homland Scurity Advisory Systm Amy (Wnxuan) Ding Dpartmnt of Information and Dcision Scincs Univrsity of Illinois Chicago, IL 60607 wxding@uicdu Using a diffrntial

More information

Continuity Cloud Virtual Firewall Guide

Continuity Cloud Virtual Firewall Guide Cloud Virtual Firwall Guid uh6 Vrsion 1.0 Octobr 2015 Foldr BDR Guid for Vam Pag 1 of 36 Cloud Virtual Firwall Guid CONTENTS INTRODUCTION... 3 ACCESSING THE VIRTUAL FIREWALL... 4 HYPER-V/VIRTUALBOX CONTINUITY

More information

Gold versus stock investment: An econometric analysis

Gold versus stock investment: An econometric analysis Intrnational Journal of Dvlopmnt and Sustainability Onlin ISSN: 268-8662 www.isdsnt.com/ijds Volum Numbr, Jun 202, Pag -7 ISDS Articl ID: IJDS20300 Gold vrsus stock invstmnt: An conomtric analysis Martin

More information

ICES REPORT 15-01. January 2015. The Institute for Computational Engineering and Sciences The University of Texas at Austin Austin, Texas 78712

ICES REPORT 15-01. January 2015. The Institute for Computational Engineering and Sciences The University of Texas at Austin Austin, Texas 78712 ICES REPORT 15-01 January 2015 A locking-fr modl for Rissnr-Mindlin plats: Analysis and isogomtric implmntation via NURBS and triangular NURPS by L. Birao da Viga, T.J.R. Hughs, J. Kindl, C. Lovadina,

More information

Electronic Commerce. and. Competitive First-Degree Price Discrimination

Electronic Commerce. and. Competitive First-Degree Price Discrimination Elctronic Commrc and Comptitiv First-Dgr Pric Discrimination David Ulph* and Nir Vulkan ** Fbruary 000 * ESRC Cntr for Economic arning and Social Evolution (ESE), Dpartmnt of Economics, Univrsity Collg

More information

An International Journal of the Polish Statistical Association

An International Journal of the Polish Statistical Association STATISTICS IN TRANSITION nw sris An Intrnational Journal of th Polish Statistical Association CONTENTS From th Editor... Submission information for authors... 5 Sampling mthods and stimation CIEPIELA P.,

More information

Development of Financial Management Reporting in MPLS

Development of Financial Management Reporting in MPLS 1 Dvlopmnt of Financial Managmnt Rporting in MPLS 1. Aim Our currnt financial rports ar structurd to dlivr an ovrall financial pictur of th dpartmnt in it s ntirty, and thr is no attmpt to provid ithr

More information

New Concepts and Methods in Information Aggregation

New Concepts and Methods in Information Aggregation Nw Concpts and Mthods in Information Aggrgation János Fodor 1, Imr J. Rudas John von Numann Faculty of Informatics, Budapst Tch Bécsi út 96/B, H-1034 Budapst, Hungary E-mail: {Fodor, Rudas}@bmf.hu Abstract:

More information

CALCULATING MARGINAL PROBABILITIES IN PROC PROBIT Guy Pascale, Memorial Health Alliance

CALCULATING MARGINAL PROBABILITIES IN PROC PROBIT Guy Pascale, Memorial Health Alliance CALCULATING MARGINAL PROBABILITIES IN PROC PROBIT Guy Pascal, Mmorial Halth Allianc Introduction Th PROBIT procdur within th SAS systm provids a simpl mthod for stimating discrt choic variabls (i.. dichotomous

More information

Van der Waals Forces Between Atoms

Van der Waals Forces Between Atoms Van dr Waals Forcs twn tos Michal Fowlr /8/7 Introduction Th prfct gas quation of stat PV = NkT is anifstly incapabl of dscribing actual gass at low tpraturs, sinc thy undrgo a discontinuous chang of volu

More information

Topology Information Condensation in Hierarchical Networks.

Topology Information Condensation in Hierarchical Networks. Topology Information Condnsation in Hirarchical Ntworks. Pit Van Mighm Dlft Univrsity of Tchnology a ABSTRACT Inspird by th PNNI protocol of th ATM Forum, this work focuss on th problm of nod aggrgation

More information

Enforcing Fine-grained Authorization Policies for Java Mobile Agents

Enforcing Fine-grained Authorization Policies for Java Mobile Agents Enforcing Fin-graind Authorization Policis for Java Mobil Agnts Giovanni Russllo Changyu Dong Narankr Dulay Dpartmnt of Computing Imprial Collg London South Knsington London, SW7 2AZ, UK {g.russllo, changyu.dong,

More information

An Analysis of Synergy Degree of Primary-Tertiary Industry System in Dujiangyan City

An Analysis of Synergy Degree of Primary-Tertiary Industry System in Dujiangyan City www.ccsnt.org/ijbm Intrnational Journal of Businss and Managmnt Vol. 6, No. 8; August An Analysis of Synrgy Dgr of Primary-Trtiary Industry Systm in Dujiangyan City Qizhi Yang School of Tourism, Sichuan

More information

A Loadable Task Execution Recorder for Hierarchical Scheduling in Linux

A Loadable Task Execution Recorder for Hierarchical Scheduling in Linux A Loadabl Task Excution Rcordr for Hirarchical Schduling in Linux Mikal Åsbrg and Thomas Nolt MRTC/Mälardaln Univrsity PO Box 883, SE-721 23, Västrås, Swdn {mikalasbrg,thomasnolt@mdhs Shinpi Kato Carngi

More information

Fundamentals: NATURE OF HEAT, TEMPERATURE, AND ENERGY

Fundamentals: NATURE OF HEAT, TEMPERATURE, AND ENERGY Fundamntals: NATURE OF HEAT, TEMPERATURE, AND ENERGY DEFINITIONS: Quantum Mchanics study of individual intractions within atoms and molculs of particl associatd with occupid quantum stat of a singl particl

More information

TIME MANAGEMENT. 1 The Process for Effective Time Management 2 Barriers to Time Management 3 SMART Goals 4 The POWER Model e. Section 1.

TIME MANAGEMENT. 1 The Process for Effective Time Management 2 Barriers to Time Management 3 SMART Goals 4 The POWER Model e. Section 1. Prsonal Dvlopmnt Track Sction 1 TIME MANAGEMENT Ky Points 1 Th Procss for Effctiv Tim Managmnt 2 Barrirs to Tim Managmnt 3 SMART Goals 4 Th POWER Modl In th Army, w spak of rsourcs in trms of th thr M

More information

The price of liquidity in constant leverage strategies. Marcos Escobar, Andreas Kiechle, Luis Seco and Rudi Zagst

The price of liquidity in constant leverage strategies. Marcos Escobar, Andreas Kiechle, Luis Seco and Rudi Zagst RACSAM Rv. R. Acad. Cin. Sri A. Mat. VO. 103 2, 2009, pp. 373 385 Matmática Aplicada / Applid Mathmatics Th pric of liquidity in constant lvrag stratgis Marcos Escobar, Andras Kichl, uis Sco and Rudi Zagst

More information

Business rules FATCA V. 02/11/2015

Business rules FATCA V. 02/11/2015 Elmnt Attribut Siz InputTyp Rquirmnt BUSINESS RULES TYPE ERROR ACK Xpath I.Mssag Hadr FATCA_OECD Vrsion xsd: string = Validation WrongVrsion ftc:fatca_oecd/vrsion SndingCompanyIN Unlimit d xsd: string

More information

Cost-Volume-Profit Analysis

Cost-Volume-Profit Analysis ch03.qxd 9/7/04 4:06 PM Pag 86 CHAPTER Cost-Volum-Profit Analysis In Brif Managrs nd to stimat futur rvnus, costs, and profits to hlp thm plan and monitor oprations. Thy us cost-volum-profit (CVP) analysis

More information

Fraud, Investments and Liability Regimes in Payment. Platforms

Fraud, Investments and Liability Regimes in Payment. Platforms Fraud, Invstmnts and Liability Rgims in Paymnt Platforms Anna Crti and Mariann Vrdir y ptmbr 25, 2011 Abstract In this papr, w discuss how fraud liability rgims impact th pric structur that is chosn by

More information

Scheduling and Rostering. Marco Kuhlmann & Guido Tack Lecture 7

Scheduling and Rostering. Marco Kuhlmann & Guido Tack Lecture 7 Schduling and Rostring Marco Kuhlann & Guido Tack Lctur Th story so ar Modlling in Gcod/J Foral rawork or constraint prograing Propagation, global constraints Sarch Th story so ar Modlling in Gcod/J Foral

More information