Matrix Multiplication I

Size: px
Start display at page:

Download "Matrix Multiplication I"

Transcription

1 Matrx Multplcaton I Yuval Flmus February 2, 2012 These notes are based on a lecture gven at the Toronto Student Semnar on February 2, The materal s taen mostly from the boo Algebrac Complexty Theory [ACT] and the lecture notes by Bläser and Bendun [Blä]. Starred sectons are the ones I ddn t have tme to cover. 1 Problem statement Ths lecture dscusses the problem of multplyng two square matrces. We wll be worng n the algebrac complexty model. For us, an algorthm for multplyng two n n matrces wll mean a sequence of steps, where step l s a statement of the form t l r for any r R t l a or t l b for, {1,..., n} t l t p t q, where p, q < l and {+,,, /} c t p for p < l We wll say that such an algorthm computes the product C = AB f at the end of the program, c = a b. The runnng-tme or complexty of the algorthm s the total number of steps, dsregardng nput and output steps. Our model s non-unform. As an example, consder Strassen s algorthm for multplyng two 2 2 matrces, as coped from Wpeda: m 1 = (a 11 + a 22 )(b 11 + b 22 ) m 2 = (a 21 + a 22 )b 11 m 3 = a 11 (b 12 b 22 ) m 4 = a 22 (b 21 b 11 ) m 5 = (a 11 + a 12 )b 22 m 6 = (a 21 a 11 )(b 11 + b 12 ) m 7 = (a 12 a 22 )(b 21 + b 22 ) c 11 = m 1 + m 4 m 5 + m 7 c 12 = m 3 + m 5 c 21 = m 2 + m 4 c 22 = m 1 m 2 + m 3 + m 6 1

2 In our model (usng some bengn shortcuts), t wll loo le ths: t 1 a 11 + a 22 t 2 b 11 + b 22 t 3 a 21 + a 22 t 4 b 12 b 22 t 5 b 21 b 11 t 6 a 11 + a 12 t 7 a 21 a 11 t 8 b 11 + b 12 t 9 a 12 a 22 t 10 b 21 + b 22 t 11 t 1 t 2 t 12 t 3 b 11 t 13 a 11 t 4 t 14 a 22 t 5 t 16 t 6 b 22 t 17 t 7 t 8 t 18 t 9 t 10 t 19 t 11 + t 14 t 20 t 19 t 15 c 11 t 20 + t 17 c 12 t 13 + t 15 c 21 t 12 + t 14 t 21 t 11 t 12 t 22 t 21 + t 13 c 22 t 22 + t 16 The total complexty of ths algorthm s 26 operatons. Later we wll wrte such algorthms n a much better way. Strassen s algorthm can be used to multply two 2 n 2 n matrces. The ey s to wrte the matrces n bloc form: [ ] [ ] [ ] A11 A 12 B11 B 12 C11 C = 12 A 21 A 22 B 21 B 22 C 21 C 22 Each of the blocs s a 2 n 1 2 n 1 matrx. Applyng Strassen s algorthm to the bg matrces, there are some addtons and subtractons that we already now how to do, and 7 multplcatons of two 2 n 1 2 n 1 matrces. For the latter, we apply the algorthm recursvely. Eventually, everythng wll reduce to 7 n scalar multplcatons, and countless 1 addtons and subtractons. Ths constructon, called tensorng, s the basc tool used n matrx multplcaton algorthms. 1 Not really countless, as we show n the sequel. 2

3 Once we show that the runnng-tme s domnated by the number of scalar multplcatons, we can conclude that two matrces can be multpled n tme O(n log 2 7 ). Ths shows that ω log Here ω, also nown as the exponent of matrx multplcaton, s defned as the nfmum of all α such that two n n matrces can be multpled n tme O(n α ) (the constant can depend on α). 2 Normal form We spent much of the prevous secton on defnng the algebrac complexty model. Now t s tme to show that we can forget about t and concentrate on algorthms of a very specfc form. Ths s the result of the followng two theorems. Theorem 1. Suppose one can multply two n n matrces n runnng-tme T. Then there s an algorthm n the followng normal form, for M = 2T : For 1 M, compute α, a lnear combnaton of entres of A. For 1 M, compute β, a lnear combnaton of entres of B. For 1 M, compute p = α β. For 1, n, compute c as a lnear combnaton of the p. All these lnear combnatons are fxed (don t depend on A, B). Our orgnal presentaton of Strassen s algorthm s n ths form. You mght as what s the pont of usng ths transformaton. After all, for each orgnal operaton we now have two multplcatons, and there are also lots of lnear combnatons to compute, each requrng up to n 2 multplcatons and addtons. Ths s explaned by the followng theorem. Theorem 2. Suppose there s an algorthm n normal form that multples two n n matrces wth M = n α. Then ω α. So asymptotcally, all the extra addtons and multplcatons don t really count. Coppersmth and Wnograd [CW] showed that n fact ω < α (we wll comment on ths later on). Together, the two theorems show that we can forget about the algebrac complexty model and concentrate on normal-form algorthms, tryng to reduce M. 2.1 Proofs* The dea of the proof of Theorem 1 s that t s enough to eep trac of the lnear and quadratc parts of all computatons nvolved, snce hgher degree parts don t matter. Proof of Theorem Dvsons are confusng, so let s gnore them for now. We wll prove somethng very smlar to the statement of the theorem, wth M = T and α, β both lnear combnatons of both nput matrces. If there are no dvsons, then each quantty t computed durng the algorthm s a multvarate polynomal n the nput varables. We can group t nto ts homogeneous parts. The dth homogeneous part P (d) of a polynomal P conssts of all those monomals wth total degree exactly d. For 3

4 example, f P = 1 + 2a + b + a 2 + 3ab then P (0) = 1 P (1) = 2a + b P (2) = a 2 + 3ab Every polynomal s the sum of ts homogeneous parts. The outputs c are all quadratc,.e. homogeneous of degree 2. So we re not really nterested n t (d) for d > 2. In order to compute for all, t s enough to compute the constant, lnear and quadratc parts of everythng (.e.,, ). We wll convert an arbtrary program nto a new program computng these parts for all of the orgnal varables. Replace a statement t r wth Replace a statement t a wth Replace a statement t t ± t wth Replace a statement t t t wth r a 0 ± ± ± These statements are not basc, but that s not gong to matter. Fnally, replace an output statement c t wth c. It s easy to prove by nducton that the new program also computes matrx multplcaton. Another nducton shows that: 4

5 All constant parts are constant (don t depend on the nputs). All lnear parts are lnear combnatons of nputs. All quadratc parts are lnear combnatons of products the orgnal program. appearng up to step n From ths t s easy to extract the (modfed) normal form. 2. We ve almost reached our normal form. The only problem s that α and β may each depend on both parts of the nput. We can always wrte α = α A + α B, β = β A + β B, separatng the two parts. Snce α β = α A β A + α A β B + α B β A + α B β B, we can separate the two nputs at the cost of quadruplng M. Nothng bad would happen f we drop α AβA and α BβB, snce at the end we don t need these terms whch are quadratc n the entres of one of the matrces. So t s really enough only to double M. 3. Now t s tme to handle dvsons. Every ratonal functon can be extended to a formal power seres (gven that the denomnator s not zero at the orgn 2 ). So t s natural to do the same thng as before, addng the followng rule for statements t t /t : / ( )/t(0) ( + )/t(0). We got these statements from reversng what we got for multplcaton above. The rest of the proof goes through. To prove Theorem 2, we follow the same technque we used to show that Strassen s algorthm appled recursvely to 2 n 2 n matrces results n 7 n scalar multplcatons, only ths tme we also account for the rest of the operatons. We get a recurrence whose soluton s O(n α ). Proof of Theorem 2. Le we dd for Strassen s algorthm, we can extend the gven algorthm to an algorthm for multplyng two n n matrces. Denote ts runnng tme by T (), so T (1) = n α. What s T ( + 1)? The frst thng we have to do s compute M lnear combnatons n the entres of A, namely the α. Each lnear combnaton taes roughly 3n 2 operatons, for a total of 3Mn 2. The same number of operatons s needed to compute the M lnear combnatons β. Next, we multply these usng MT () operatons. Fnally, we compute n 2 lnear combnatons, one for each entry n the target matrx. So T ( + 1) n α T () + (6M + n 2 )n 2. One can show that necessarly α > 2 (n other words, you cannot multply two n n matrces wth M = n 2 ), and so the soluton to the mplct recurrence s T () = O(n α ). In terms of the matrx sze N = n, ths s T () = O(N α ). 2 That happens n our case snce the algorthm should correctly compute the product of two zero matrces. 5

6 3 Tensor notaton There s a neat way to wrte algorthms n normal form. Start by wrtng α t as a sum of a and β t as a sum of b. For Strassen s algorthm, for example, we d have α 1 = a 11 + a 22, β 1 = b 11 + b 22, and so on. Next, each c s a lnear combnaton c = M t=1 C t α tβ t. Defne γ t = t C t c, where we thn of c as formal varables. For example, n Strassen s algorthm we have γ 1 = c 11 + c 22. Let s loo at the expresson M t=1 α tβ t γ t. What s the coeffcent of c? It s the value that c gets at the end of the algorthm,.e. a b. So n a formal sense, M α t β t γ t = t=1 Here s Strassen s algorthm n ths form: 2,,=1 a b c = n,,=1 a b c. (a 11 + a 22 )(b 11 + b 22 )(c 11 + c 22 )+ (a 21 + a 22 )b 11 (c 21 c 22 )+ a 11 (b 12 b 22 )(c 12 + c 22 )+ a 22 (b 21 b 11 )(c 11 + c 21 )+ (a 11 + a 12 )b 22 ( c 11 + c 12 )+ (a 21 a 11 )(b 11 + b 12 )c 22 + (a 12 a 22 )(b 21 + b 22 )c 11 We denote the left-hand sde by 2, 2, 2. In general, n, m, p = n m p a b c =1 =1 =1 corresponds to multplyng an n m matrx by an m p matrx, obtanng an n p matrx as a result. We call these new obects tensors. They are generalzatons of matrces (see below). The mnmal M such that a tensor T can be represented as a sum M =1 α β γ, where α, β, γ are lnear combnatons of entres of A, B, C (respectvely) s nown as the ran of T, denoted R(T ). Strassen s algorthm shows that R( 2, 2, 2 ) 7 (we actually have equalty n ths case). In earler lterature, the ran s also called the number of essental multplcatons. How does tensor ran generalze matrx ran? Encode a matrx A of dmenson n m as n =1 =1 m a x y. 6

7 The ran of ths tensor s the mnmal R such that ( n m R n a x y = b ) m x c y. =1 =1 =1 In terms of matrces, ths corresponds to a representaton A = =1 R b c, =1 where b s a column vector and c s a row vector (so these are all outer products). Ths expresses each row of A as a lnear combnaton of the row vectors c, and so R s the row ran of A. Swtchng the roles, we see that R s also the column ran of A. The symmetrc form of ths representaton thus mmedately mples that the row ran equals the column ran. We can rephrase Theorem 2 usng our new termnology: =1 ω log n R( n, n, n ). (1) As we ve already commented, the nequalty s actually strct. Why dd we bother descrbng ths tensor notaton? The new notaton shows the symmetry among the varables a, b, c. Startng wth the tensor for n, m, p, f we swtch a wth b then we get the tensor p, m, n (we also need to swtch some of the ndces). Gong over all permutatons, we fnd out that R( n, m, p ) = R( n, p, m ) = R( m, n, p ) = R( m, p, n ) = R( p, n, m ) = R( p, m, n ). (2) Suppose we now that R( n 1, m 1, p 1 ) R 1 and R( n 2, m 2, p 2 ) R 2. What can we say about R( n 1 n 2, m 1 m 2, p 1 p 2 )? We can thn of the n 1 n 2 m 1 m 2 matrx A as an n 1 n 2 matrx whose entres are n 2 m 2 matrces. The matrces B, C can be decomposed analogously. We then apply the algorthm showng that R( n 1, m 1, p 1 ) R 1. Each tme we have to multply two elements, whch are now matrces, we use the algorthm showng R( n 2, m 2, p 2 ) R 2. Ths wll reduce everythng to computatons of R 1 R 2 matrx products. So R( n 1 n 2, m 1 m 2, p 1 p 2 ) R( n 1, m 1, p 1 )R( n 2, m 2, p 2 ). (3) Let s apply ths together wth the symmetry relatons: Ths has the mplcaton that R( nmp, nmp, nmp ) R( n, m, p ) 3. Ths generalzes (1). We ll see (and use) even more general forms below. ω 3 log nmp R( n, m, p ). (4) 7

8 4 Border ran Bn came up wth the followng dentty: ɛ(a 11 b 11 c 11 + a 11 b 12 c 21 + a 12 b 21 c 11 + a 12 b 22 c 21 + a 21 b 11 c 12 + a 21 b 12 c 22 )+ ɛ 2 (a 11 b 22 c 21 + a 11 b 11 c 12 + a 12 b 21 c 22 + a 21 b 21 c 22 ) = (a 12 + ɛa 11 )(b 12 + ɛb 22 )c 21 + (a 21 + ɛa 11 )b 11 (c 11 + ɛc 12 ) a 12 b 12 (c 11 + c 21 + ɛc 22 ) a 21 (b 11 + b 12 + ɛb 21 )c 11 + (a 12 + a 21 )(b 12 + ɛb 21 )(c 11 + ɛc 22 ). On the left-hand sde, we have ɛ tmes the tensor correspondng to the followng partal matrx multplcaton: [ ] [ ] [ ] a11 a 12 b11 b 12 c11 c = 12. a 21 0 b 21 b 22 c 21 c 22 From a computatonal pont of vew, here s how you d use ths dentty to compute ths partal matrx multplcaton approxmately (f you could calculate wth nfnte precson). Pc ɛ > 0 very small, and use the dentty. Dvde the result by ɛ. We obtan roughly the result. By pcng ɛ nfntesmal, we can compute the matrx multplcaton exactly. But there s a way to do t even wthout nonstandard analyss, as we ll see below. Frst, let s see how you d use the dentty for multplyng large square matrces. Puttng together two copes of the dentty, we get an dentty equatng ɛ 3, 2, 2 + O(ɛ 2 ) wth a sum of 10 terms (the bg-o notaton hdes terms whch grow le ɛ 3 or smaller). Symmetrzng, we get an dentty equatng ɛ 3 12, 12, 12 + O(ɛ 4 ) wth a sum of 10 3 terms. Tang the Nth tensor power, we get an dentty equatng ɛ 3N 12 N, 12 N, 12 N + O(ɛ 3N+1 ) wth a sum of 10 3N terms. What do we do wth the new dentty? We re only really nterested n all terms wth coeffcent ɛ 3N. How do we solate them? Consder the product (a 12 + ɛa 11 )(b 12 + ɛb 22 )c 21. If we wanted to compute only the coeffcent of ɛ, we would need to compute a 11 b 12 c 21 + a 12 b 22 c 21. The general stuaton s smlar. For each gven term α β γ, n order to compute the coeffcent of ɛ 3N, we wll tae terms correspondng to coeffcents d α, d β, d γ from α, β, γ (respectvely) such that d α + d β + d γ = 6N, and sum all of these. There are at most ( ) 3N+2 2 = O(N 2 ) such terms. So ω log 12 N O(N N ) 3 log The polynomal factor O(N 2 ) doesn t really mae any dfference n the lmt. In other words, the fact that the dentty was only approxmate maes no dfference. Bn s dentty (after tang two copes of t) shows that R( 3, 2, 2 ) 10. The border ran R(T ) of a tensor T s exactly the mnmum M such that ɛ d T + O(ɛ d+1 ) = α β γ for some d. Our argument shows n general that ω 3 log nmp R( n, m, p ). (5) 8

9 Border ran satsfes many of the propertes of ran. For example, t s symmetrc 2 and submultplcatve 3. We can thn of t as a generalzaton of ran whch s good enough to obtan bounds on ω. Bn s partcular dentty can be leveraged even more, showng that ω 3 log , usng the methods of 6. In earler lterature, these denttes are nown as λ-computatons (so ɛ s replaced wth λ). Also, negatve powers of λ are used so that the left-hand sde has an error term of magntude O(λ). 5 Schönhage s τ theorem Schönhage dscovered the followng surprsng dentty: ɛ 2 A B C + X 3+ Y 3+ Z + O(ɛ 3 ) = 3 =1 =1 =1 =1 =1 =1 3 (A + ɛx 3+ )(B + ɛy 3+ )(ɛ 2 C + Z)+ 3 A (B 4 ɛ =1 3 Y 3+ )(ɛ 2 C 4 + Z) + =1 A 4 B 4 (ɛ 2 C 44 + Z) Ths dentty shows that ( 4 =1 A ) 4 =1 B 3 (A 4 ɛ =1 Z. 3 X 3+ )B (ɛ 2 C 4 + Z)+ =1 R( 4, 1, 4 1, 9, 1 ) 17. (In fact there s equalty.) The dentty shows how to compute the outer product 4, 1, 4 of two vectors of length 4 along wth the nner product 1, 9, 1 of two other vectors of length 9 (the symbol emphaszes the fact that the two products concern dfferent varables) wth only multplcatons. Ths s surprsng, snce easy arguments show that R( 4, 1, 4 ) = 16, and wth one more multplcaton t s suddenly possble to compute along an extra (long) nner product. How do we use ths dentty? Schönhage nvented hs τ-theorem (also: asymptotc sum nequalty) to answer ths queston. We wll approach hs soluton through a seres of smple steps. For smplcty, we wll only explctly consder ran, but everythng we do also wors wth border ran, whch s how we state our results. 1. Suppose we had an dentty showng that R( n, n, n ) M (here n, n, n s the drect sum of copes of n, n, n wth dsont varables). How would we use t to multply square matrces? Suppose we wanted to multply two n T n T matrces, where T s very large. We apply our new algorthm recursvely. In the frst level we can t really tae advantage of the full abltes of our algorthm, snce we only have one matrx product to compute. In the second level, we already have M of these, so we need to apply our algorthm recursvely only M/ tmes. As we go along, we wll be able to group the products we need to compute at each level to groups of sze, and leverage our algorthm almost perfectly. So asymptotcally, our algorthm behaves as f we had a way to multply two n n matrces usng M/ products (note that M/ need not be ntegral). 9

10 The detals wor out, showng R( n, n, n ) ω log n. (6) 2. A smple symmetrzaton argument shows that 3. Consder now Schönhage s dentty, showng R( n, m, p ) ω 3 log nmp. (7) R( 4, 1, 4 1, 9, 1 ) 17. Compute the Nth tensor power: ( N ) R c n, m, p 17 N, where c = =0 ( ) N, n = p = 4, m = 9 N. Our goal s to obtan a bound usng (7). If we were amng at a bound ω 3τ, then we d need 17 N c (n m p ) τ. It s natural to defne accordngly the volume of a tensor n, m, p by V τ ( n, m, p ) = (nmp) τ, and extend the defnton addtvely to drect sums. Ths noton of volume s multplcatve,.e. V τ (T 1 T 2 ) = V τ (T 1 )V τ (T 2 ), so ( N ) V τ c n, m, p = V τ ( 4, 1, 4 1, 9, 1 ) N = (16 τ + 9 τ ) N. =0 Snce there are only N + 1 terms on the left, there must be some term satsfyng c (n m p ) τ (16τ + 9 τ ) N N + 1 We want ths to be roughly equal to 17 N, so we choose τ so that Wth ths value of τ, formula (7) mples that 16 τ + 9 τ = 17. ω 3 log n m p 17 N c 3τ. In the lmt, we actually get ω 3τ. In our case, τ 0.85 and 3τ Ths proof generalzes to gve Schönhage s τ-theorem: ( ) (n m p ) τ = R n, m, p mples ω 3τ. (8). 10

11 An alternatve form shows why t s worthy of ts other name, the asymptotc sum nequalty: ( ) (n m p ) ω/3 R n, m, p. (9) Coppersmth and Wnograd [CW, Corollary 3.2] showed that R( n, m, p 1, R( n, m, p ) m(n + p 1), 1 ) R( n, m, p ) + m. Ths shows that startng wth an algorthm for n, m, p, we can always obtan a better algorthm (yeldng a better ω through the asymptotc sum nequalty). In other words, no sngle algorthm for n, m, p can yeld the optmal ω. A smlar result of thers from the same paper (Corollary 3.5) mples that the nequalty n (8) s always strct. 6 Fast multplcaton of rectangular matrces* For ths secton, we change our focus and concentrate on multplcaton problems of the type n, n, n α. We want to fnd a value of α such that R( n, n, n α ) = Õ(n2 ). Followng Coppersmth s footsteps [Cop2], we consder another dentty due to Schönhage: ɛ 2 (a 11 b 11 c 11 + a 11 b 12 c 21 + a 12 b 21 c 11 + a 13 b 31 c 11 + a 22 b 21 c 12 + a 23 b 31 c 12 ) + O(ɛ 3 ) = (a 11 + ɛ 2 a 12 )(b 21 + ɛ 2 b 11 )c 11 + (a 11 + ɛ 2 a 13 )b 31 (c 11 ɛc 21 )+ (a 11 + ɛ 2 a 22 )(b 21 ɛb 12 )c 12 + (a 11 + ɛ 2 a 23 )(b 31 + ɛb 12 )(c 12 + ɛc 21 ) a 11 (b 21 + b 31 )(c 11 + c 12 ). Ths dentty descrbes a fast way to approxmately multply a partal 2 3 matrx wth a partal 3 2 matrx usng only fve multplcatons: ( ) a11 a 12 a 11 b 12 ( ) 13 b b 0 a 22 a 21 0 c11 c = c b c 22 It s mportant here that ths dentty s tght: there s a trval matchng lower bound for the border ran, comng from the number of ndetermnates n the A matrx. Tae ths dentty and tensor t N tmes to get a new dentty nvolvng 5 N multplcatons, showng how to multply a partal 2 N 3 N matrx wth a partal 3 N 2 N matrx. These matrces have a very complcated pattern of zeroes, but we wll only be nterested n the number of non-zero entres n each column of the frst matrx, and the number of non-zero entres n each row of the second matrx. There are N + 1 dfferent types of such ndces (does that sound famlar?). Type nvolves m = ( ) N 2 columns of the frst matrx wth n = 2 non-zero entres, and matchng rows of the second matrx wth p = 2 N non-zero entres. Suppose we group together all ndces of type, and zero all other entres n the two matrces. We get an dentty for approxmately computng AB, where A conssts of m columns, each havng 11

12 n non-zero entres, and B conssts of m rows, each havng p non-zero entres. So t s morally a tensor of type n, m, p. In fact, we can actually encode n, m, p nsde the product AB (see below). Ths shows that R( n, m, p ) 5 N. Symmetrze once to get R( n m, n m, p 2 ) 5 2N. We would le n m 5 N. Snce n m = ( ) N 4, n m s maxmzed for = 4N/5, at whch pont n m = Θ(5 N / N). For ths we have p 2 = 4N/5 = 5 αn for α = (log 5 4)/ Concludng, ( ) 5 N R, 5N, 5 αn 5 2N. N N Tang n = 5 N / N, Ths mples that R( n, n, n α ) = O(n 2 log n). R( n, n, n α ) = O(n 2 log 2 n). The usual tensorng trc shows that one can multply an n n α matrx by an n α n matrx usng O(n 2 log 2 n) arthmetc operatons. Moreover, as observed by Ryan Wllams [Wl], we can descrbe all these transformatons unformly. Ryan Wllams used fast matrx multplcaton to solve SAT on certan symmetrc crcuts on all nputs very qucly. Later, Andreas Börlund found a more elementary dynamc programmng accomplshng the same tas, so that fast matrx multplcaton s no longer needed. 6.1 Proof It remans to prove that n, m, p matrx multplcaton can be embedded n AB matrx multplcaton f A conssts of m columns, each havng n non-zero entres, and B conssts of m rows, each havng p non-zero entres. Let A be an n m matrx. We wll construct a (constant) matrx T translatng between A and A: gven A, we wll be able to fnd A so that A = T A. The same constructon wll also gve a matrx S translatng between an arbtrary m p matrx B and B, n the sense that B = BS. We can then reduce A B to AB usng A B = T (AB)S. How do we fnd the matrx T? We show one way to do ths for small numbers, the general case beng smlar. Suppose that n = 2 and that A has three rows, so the matrx T has dmensons 2 3. Our matrx T wll be [ ] T = The matrx T corresponds to the lnear transformaton T (x, y, z) = (x + y, x + z). Ths lnear transformaton has the property that, gven the constrant that only two of x, y, z are non-zero, we can stll encode any par (a, b): T (a, b a, 0) = (a, b) T (a, 0, b a) = (a, b) T (0, a, b) = (a, b) Now tae each column of the matrx A, and use the approprate formula to encode t nto a column of A. Ths encodng ensures that A = T A. More generally, the property we need from the matrx T s the all square mnors are non-zero. Ths s certanly the case for a random matrx T. 12

13 6.2 Alternatve method In the same paper descrbng the precedng constructon, Coppersmth also descrbed a slghtly nferor constructon gvng only α However, the same method, combned wth the dentty and methods of [CW], gves a much better result, α Ths has recently been mproved by Le Gall [Gal] to α These constructons apparently cannot be used for Wllams result snce they only gve O(n 2+ɛ ) algorthms for any ɛ > 0. The dea s to start wth an dentty by Schönhage whch we have already consdered (wth dfferent parameters): ɛ 2 A B C + X 2+ Y 2+ Z + O(ɛ 3 ) = 2 =1 =1 =1 =1 =1 =1 2 (A + ɛx 2+ )(B + ɛy 2+ )(ɛ 2 C + Z)+ 2 A (B 3 ɛ =1 2 Y 2+ )(ɛ 2 C 3 + Z) + =1 A 3 B 3 (ɛ 2 C 33 + Z) Ths dentty shows that ( 3 =1 A ) 3 =1 B 2 (A 3 ɛ =1 Z. 2 X 2+ )B (ɛ 2 C 3 + Z)+ =1 R( 3, 1, 3 1, 4, 1 ) 10. Agan, ths dentty s tght. The dea now s to tae a hgh tensor power wthout symmetrzng: N ( ) R N 3, 4 N, 3 10 N. =0 The dea now s to choose an approprate and zero everythng else. Whch should we pc? The same method used for provng the asymptotc sum nequalty shows that for large M, R( 3 M, 4 (N )M, 3 M ) 10NM ( N ) M. In order to get a tght bound, we would le the rght-hand sde to be approxmately (3 M ) 2. In other words, tang the Mth root, we want ( N ) 9 10 N. When provng the asymptotc sum nequalty, t was enough to comment that such a can be found by tang the maxmum over ( N ) 9, snce the sum of these N + 1 terms s 10 N. In ths case we also need to now the value of, whch s roughly 0.9N. Ths shows that roughly speang, R( 3 0.9NM, 4 0.1NM, 3 0.9NM ) 9 0.9NM. Puttng n = 3 0.9NM, the other ndex s 4 0.1NM = n α for α = 0.1 log 4/0.9 log When unrollng the constructon, for techncal reasons we only get a bound of the form O(n 2+ɛ ) rather than Õ(n2 ) as before. 13

14 References [ACT] Peter Bürgsser, Mchael Clausen and M. Amn Shorollah, Algebrac Complexty Theory, Sprnger, [Blä] Marus Bläser, Complexty of blnear problems (lecture notes scrbed by Faban Bendun), scrpt.pdf, [Cop] Dan Coppersmth, Rapd multplcaton of rectangular matrces, SIAM J. Comput. 11: , [Cop2] Dan Coppersmth, Rectangular matrx multplcaton revsted, J. Comp. 13:42 49, [CW] Dan Coppersmth, Shmuel Wnograd, On the asymptotc complexty of matrx multplcaton, SIAM J. Comput. 5: , [Gal] Fran cos Le Gall, Faster algorthms for rectangular matrx multplcaton, ArXv, [Wl] Ryan Wllams, Non-unform ACC crcut lower bounds, CCC

v a 1 b 1 i, a 2 b 2 i,..., a n b n i.

v a 1 b 1 i, a 2 b 2 i,..., a n b n i. SECTION 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS 455 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS All the vector spaces we have studed thus far n the text are real vector spaces snce the scalars are

More information

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by 6 CHAPTER 8 COMPLEX VECTOR SPACES 5. Fnd the kernel of the lnear transformaton gven n Exercse 5. In Exercses 55 and 56, fnd the mage of v, for the ndcated composton, where and are gven by the followng

More information

where the coordinates are related to those in the old frame as follows.

where the coordinates are related to those in the old frame as follows. Chapter 2 - Cartesan Vectors and Tensors: Ther Algebra Defnton of a vector Examples of vectors Scalar multplcaton Addton of vectors coplanar vectors Unt vectors A bass of non-coplanar vectors Scalar product

More information

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Luby s Alg. for Maximal Independent Sets using Pairwise Independence Lecture Notes for Randomzed Algorthms Luby s Alg. for Maxmal Independent Sets usng Parwse Independence Last Updated by Erc Vgoda on February, 006 8. Maxmal Independent Sets For a graph G = (V, E), an ndependent

More information

Recurrence. 1 Definitions and main statements

Recurrence. 1 Definitions and main statements Recurrence 1 Defntons and man statements Let X n, n = 0, 1, 2,... be a MC wth the state space S = (1, 2,...), transton probabltes p j = P {X n+1 = j X n = }, and the transton matrx P = (p j ),j S def.

More information

1 Example 1: Axis-aligned rectangles

1 Example 1: Axis-aligned rectangles COS 511: Theoretcal Machne Learnng Lecturer: Rob Schapre Lecture # 6 Scrbe: Aaron Schld February 21, 2013 Last class, we dscussed an analogue for Occam s Razor for nfnte hypothess spaces that, n conjuncton

More information

We are now ready to answer the question: What are the possible cardinalities for finite fields?

We are now ready to answer the question: What are the possible cardinalities for finite fields? Chapter 3 Fnte felds We have seen, n the prevous chapters, some examples of fnte felds. For example, the resdue class rng Z/pZ (when p s a prme) forms a feld wth p elements whch may be dentfed wth the

More information

BERNSTEIN POLYNOMIALS

BERNSTEIN POLYNOMIALS On-Lne Geometrc Modelng Notes BERNSTEIN POLYNOMIALS Kenneth I. Joy Vsualzaton and Graphcs Research Group Department of Computer Scence Unversty of Calforna, Davs Overvew Polynomals are ncredbly useful

More information

PERRON FROBENIUS THEOREM

PERRON FROBENIUS THEOREM PERRON FROBENIUS THEOREM R. CLARK ROBINSON Defnton. A n n matrx M wth real entres m, s called a stochastc matrx provded () all the entres m satsfy 0 m, () each of the columns sum to one, m = for all, ()

More information

The OC Curve of Attribute Acceptance Plans

The OC Curve of Attribute Acceptance Plans The OC Curve of Attrbute Acceptance Plans The Operatng Characterstc (OC) curve descrbes the probablty of acceptng a lot as a functon of the lot s qualty. Fgure 1 shows a typcal OC Curve. 10 8 6 4 1 3 4

More information

What is Candidate Sampling

What is Candidate Sampling What s Canddate Samplng Say we have a multclass or mult label problem where each tranng example ( x, T ) conssts of a context x a small (mult)set of target classes T out of a large unverse L of possble

More information

Loop Parallelization

Loop Parallelization - - Loop Parallelzaton C-52 Complaton steps: nested loops operatng on arrays, sequentell executon of teraton space DECLARE B[..,..+] FOR I :=.. FOR J :=.. I B[I,J] := B[I-,J]+B[I-,J-] ED FOR ED FOR analyze

More information

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ). REVIEW OF RISK MANAGEMENT CONCEPTS LOSS DISTRIBUTIONS AND INSURANCE Loss and nsurance: When someone s subject to the rsk of ncurrng a fnancal loss, the loss s generally modeled usng a random varable or

More information

Ring structure of splines on triangulations

Ring structure of splines on triangulations www.oeaw.ac.at Rng structure of splnes on trangulatons N. Vllamzar RICAM-Report 2014-48 www.rcam.oeaw.ac.at RING STRUCTURE OF SPLINES ON TRIANGULATIONS NELLY VILLAMIZAR Introducton For a trangulated regon

More information

n + d + q = 24 and.05n +.1d +.25q = 2 { n + d + q = 24 (3) n + 2d + 5q = 40 (2)

n + d + q = 24 and.05n +.1d +.25q = 2 { n + d + q = 24 (3) n + 2d + 5q = 40 (2) MATH 16T Exam 1 : Part I (In-Class) Solutons 1. (0 pts) A pggy bank contans 4 cons, all of whch are nckels (5 ), dmes (10 ) or quarters (5 ). The pggy bank also contans a con of each denomnaton. The total

More information

Implementation of Deutsch's Algorithm Using Mathcad

Implementation of Deutsch's Algorithm Using Mathcad Implementaton of Deutsch's Algorthm Usng Mathcad Frank Roux The followng s a Mathcad mplementaton of Davd Deutsch's quantum computer prototype as presented on pages - n "Machnes, Logc and Quantum Physcs"

More information

21 Vectors: The Cross Product & Torque

21 Vectors: The Cross Product & Torque 21 Vectors: The Cross Product & Torque Do not use our left hand when applng ether the rght-hand rule for the cross product of two vectors dscussed n ths chapter or the rght-hand rule for somethng curl

More information

Forecasting the Direction and Strength of Stock Market Movement

Forecasting the Direction and Strength of Stock Market Movement Forecastng the Drecton and Strength of Stock Market Movement Jngwe Chen Mng Chen Nan Ye cjngwe@stanford.edu mchen5@stanford.edu nanye@stanford.edu Abstract - Stock market s one of the most complcated systems

More information

Production. 2. Y is closed A set is closed if it contains its boundary. We need this for the solution existence in the profit maximization problem.

Production. 2. Y is closed A set is closed if it contains its boundary. We need this for the solution existence in the profit maximization problem. Producer Theory Producton ASSUMPTION 2.1 Propertes of the Producton Set The producton set Y satsfes the followng propertes 1. Y s non-empty If Y s empty, we have nothng to talk about 2. Y s closed A set

More information

8 Algorithm for Binary Searching in Trees

8 Algorithm for Binary Searching in Trees 8 Algorthm for Bnary Searchng n Trees In ths secton we present our algorthm for bnary searchng n trees. A crucal observaton employed by the algorthm s that ths problem can be effcently solved when the

More information

Linear Circuits Analysis. Superposition, Thevenin /Norton Equivalent circuits

Linear Circuits Analysis. Superposition, Thevenin /Norton Equivalent circuits Lnear Crcuts Analyss. Superposton, Theenn /Norton Equalent crcuts So far we hae explored tmendependent (resste) elements that are also lnear. A tmendependent elements s one for whch we can plot an / cure.

More information

1. Measuring association using correlation and regression

1. Measuring association using correlation and regression How to measure assocaton I: Correlaton. 1. Measurng assocaton usng correlaton and regresson We often would lke to know how one varable, such as a mother's weght, s related to another varable, such as a

More information

Quantization Effects in Digital Filters

Quantization Effects in Digital Filters Quantzaton Effects n Dgtal Flters Dstrbuton of Truncaton Errors In two's complement representaton an exact number would have nfntely many bts (n general). When we lmt the number of bts to some fnte value

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Max Wellng Department of Computer Scence Unversty of Toronto 10 Kng s College Road Toronto, M5S 3G5 Canada wellng@cs.toronto.edu Abstract Ths s a note to explan support vector machnes.

More information

The Greedy Method. Introduction. 0/1 Knapsack Problem

The Greedy Method. Introduction. 0/1 Knapsack Problem The Greedy Method Introducton We have completed data structures. We now are gong to look at algorthm desgn methods. Often we are lookng at optmzaton problems whose performance s exponental. For an optmzaton

More information

Series Solutions of ODEs 2 the Frobenius method. The basic idea of the Frobenius method is to look for solutions of the form 3

Series Solutions of ODEs 2 the Frobenius method. The basic idea of the Frobenius method is to look for solutions of the form 3 Royal Holloway Unversty of London Department of Physs Seres Solutons of ODEs the Frobenus method Introduton to the Methodology The smple seres expanson method works for dfferental equatons whose solutons

More information

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module LOSSLESS IMAGE COMPRESSION SYSTEMS Lesson 3 Lossless Compresson: Huffman Codng Instructonal Objectves At the end of ths lesson, the students should be able to:. Defne and measure source entropy..

More information

Project Networks With Mixed-Time Constraints

Project Networks With Mixed-Time Constraints Project Networs Wth Mxed-Tme Constrants L Caccetta and B Wattananon Western Australan Centre of Excellence n Industral Optmsaton (WACEIO) Curtn Unversty of Technology GPO Box U1987 Perth Western Australa

More information

THE METHOD OF LEAST SQUARES THE METHOD OF LEAST SQUARES

THE METHOD OF LEAST SQUARES THE METHOD OF LEAST SQUARES The goal: to measure (determne) an unknown quantty x (the value of a RV X) Realsaton: n results: y 1, y 2,..., y j,..., y n, (the measured values of Y 1, Y 2,..., Y j,..., Y n ) every result s encumbered

More information

Extending Probabilistic Dynamic Epistemic Logic

Extending Probabilistic Dynamic Epistemic Logic Extendng Probablstc Dynamc Epstemc Logc Joshua Sack May 29, 2008 Probablty Space Defnton A probablty space s a tuple (S, A, µ), where 1 S s a set called the sample space. 2 A P(S) s a σ-algebra: a set

More information

HÜCKEL MOLECULAR ORBITAL THEORY

HÜCKEL MOLECULAR ORBITAL THEORY 1 HÜCKEL MOLECULAR ORBITAL THEORY In general, the vast maorty polyatomc molecules can be thought of as consstng of a collecton of two electron bonds between pars of atoms. So the qualtatve pcture of σ

More information

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12 14 The Ch-squared dstrbuton PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 1 If a normal varable X, havng mean µ and varance σ, s standardsed, the new varable Z has a mean 0 and varance 1. When ths standardsed

More information

Texas Instruments 30X IIS Calculator

Texas Instruments 30X IIS Calculator Texas Instruments 30X IIS Calculator Keystrokes for the TI-30X IIS are shown for a few topcs n whch keystrokes are unque. Start by readng the Quk Start secton. Then, before begnnng a specfc unt of the

More information

Joe Pimbley, unpublished, 2005. Yield Curve Calculations

Joe Pimbley, unpublished, 2005. Yield Curve Calculations Joe Pmbley, unpublshed, 005. Yeld Curve Calculatons Background: Everythng s dscount factors Yeld curve calculatons nclude valuaton of forward rate agreements (FRAs), swaps, nterest rate optons, and forward

More information

An Alternative Way to Measure Private Equity Performance

An Alternative Way to Measure Private Equity Performance An Alternatve Way to Measure Prvate Equty Performance Peter Todd Parlux Investment Technology LLC Summary Internal Rate of Return (IRR) s probably the most common way to measure the performance of prvate

More information

How To Calculate The Accountng Perod Of Nequalty

How To Calculate The Accountng Perod Of Nequalty Inequalty and The Accountng Perod Quentn Wodon and Shlomo Ytzha World Ban and Hebrew Unversty September Abstract Income nequalty typcally declnes wth the length of tme taen nto account for measurement.

More information

NMT EE 589 & UNM ME 482/582 ROBOT ENGINEERING. Dr. Stephen Bruder NMT EE 589 & UNM ME 482/582

NMT EE 589 & UNM ME 482/582 ROBOT ENGINEERING. Dr. Stephen Bruder NMT EE 589 & UNM ME 482/582 NMT EE 589 & UNM ME 482/582 ROBOT ENGINEERING Dr. Stephen Bruder NMT EE 589 & UNM ME 482/582 7. Root Dynamcs 7.2 Intro to Root Dynamcs We now look at the forces requred to cause moton of the root.e. dynamcs!!

More information

INSTITUT FÜR INFORMATIK

INSTITUT FÜR INFORMATIK INSTITUT FÜR INFORMATIK Schedulng jobs on unform processors revsted Klaus Jansen Chrstna Robene Bercht Nr. 1109 November 2011 ISSN 2192-6247 CHRISTIAN-ALBRECHTS-UNIVERSITÄT ZU KIEL Insttut für Informat

More information

Lecture 3: Force of Interest, Real Interest Rate, Annuity

Lecture 3: Force of Interest, Real Interest Rate, Annuity Lecture 3: Force of Interest, Real Interest Rate, Annuty Goals: Study contnuous compoundng and force of nterest Dscuss real nterest rate Learn annuty-mmedate, and ts present value Study annuty-due, and

More information

+ + + - - This circuit than can be reduced to a planar circuit

+ + + - - This circuit than can be reduced to a planar circuit MeshCurrent Method The meshcurrent s analog of the nodeoltage method. We sole for a new set of arables, mesh currents, that automatcally satsfy KCLs. As such, meshcurrent method reduces crcut soluton to

More information

Nonbinary Quantum Error-Correcting Codes from Algebraic Curves

Nonbinary Quantum Error-Correcting Codes from Algebraic Curves Nonbnary Quantum Error-Correctng Codes from Algebrac Curves Jon-Lark Km and Judy Walker Department of Mathematcs Unversty of Nebraska-Lncoln, Lncoln, NE 68588-0130 USA e-mal: {jlkm, jwalker}@math.unl.edu

More information

Time Value of Money Module

Time Value of Money Module Tme Value of Money Module O BJECTIVES After readng ths Module, you wll be able to: Understand smple nterest and compound nterest. 2 Compute and use the future value of a sngle sum. 3 Compute and use the

More information

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification Lecture 4: More classfers and classes C4B Machne Learnng Hlary 20 A. Zsserman Logstc regresson Loss functons revsted Adaboost Loss functons revsted Optmzaton Multple class classfcaton Logstc Regresson

More information

Calculation of Sampling Weights

Calculation of Sampling Weights Perre Foy Statstcs Canada 4 Calculaton of Samplng Weghts 4.1 OVERVIEW The basc sample desgn used n TIMSS Populatons 1 and 2 was a two-stage stratfed cluster desgn. 1 The frst stage conssted of a sample

More information

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence 1 st Internatonal Symposum on Imprecse Probabltes and Ther Applcatons, Ghent, Belgum, 29 June 2 July 1999 How Sets of Coherent Probabltes May Serve as Models for Degrees of Incoherence Mar J. Schervsh

More information

Compiling for Parallelism & Locality. Dependence Testing in General. Algorithms for Solving the Dependence Problem. Dependence Testing

Compiling for Parallelism & Locality. Dependence Testing in General. Algorithms for Solving the Dependence Problem. Dependence Testing Complng for Parallelsm & Localty Dependence Testng n General Assgnments Deadlne for proect 4 extended to Dec 1 Last tme Data dependences and loops Today Fnsh data dependence analyss for loops General code

More information

L10: Linear discriminants analysis

L10: Linear discriminants analysis L0: Lnear dscrmnants analyss Lnear dscrmnant analyss, two classes Lnear dscrmnant analyss, C classes LDA vs. PCA Lmtatons of LDA Varants of LDA Other dmensonalty reducton methods CSCE 666 Pattern Analyss

More information

Section 5.4 Annuities, Present Value, and Amortization

Section 5.4 Annuities, Present Value, and Amortization Secton 5.4 Annutes, Present Value, and Amortzaton Present Value In Secton 5.2, we saw that the present value of A dollars at nterest rate per perod for n perods s the amount that must be deposted today

More information

Hedging Interest-Rate Risk with Duration

Hedging Interest-Rate Risk with Duration FIXED-INCOME SECURITIES Chapter 5 Hedgng Interest-Rate Rsk wth Duraton Outlne Prcng and Hedgng Prcng certan cash-flows Interest rate rsk Hedgng prncples Duraton-Based Hedgng Technques Defnton of duraton

More information

Simple Interest Loans (Section 5.1) :

Simple Interest Loans (Section 5.1) : Chapter 5 Fnance The frst part of ths revew wll explan the dfferent nterest and nvestment equatons you learned n secton 5.1 through 5.4 of your textbook and go through several examples. The second part

More information

Product-Form Stationary Distributions for Deficiency Zero Chemical Reaction Networks

Product-Form Stationary Distributions for Deficiency Zero Chemical Reaction Networks Bulletn of Mathematcal Bology (21 DOI 1.17/s11538-1-9517-4 ORIGINAL ARTICLE Product-Form Statonary Dstrbutons for Defcency Zero Chemcal Reacton Networks Davd F. Anderson, Gheorghe Cracun, Thomas G. Kurtz

More information

Rotation Kinematics, Moment of Inertia, and Torque

Rotation Kinematics, Moment of Inertia, and Torque Rotaton Knematcs, Moment of Inerta, and Torque Mathematcally, rotaton of a rgd body about a fxed axs s analogous to a lnear moton n one dmenson. Although the physcal quanttes nvolved n rotaton are qute

More information

SPEE Recommended Evaluation Practice #6 Definition of Decline Curve Parameters Background:

SPEE Recommended Evaluation Practice #6 Definition of Decline Curve Parameters Background: SPEE Recommended Evaluaton Practce #6 efnton of eclne Curve Parameters Background: The producton hstores of ol and gas wells can be analyzed to estmate reserves and future ol and gas producton rates and

More information

Question 2: What is the variance and standard deviation of a dataset?

Question 2: What is the variance and standard deviation of a dataset? Queston 2: What s the varance and standard devaton of a dataset? The varance of the data uses all of the data to compute a measure of the spread n the data. The varance may be computed for a sample of

More information

Optimal outpatient appointment scheduling

Optimal outpatient appointment scheduling Health Care Manage Sc (27) 1:217 229 DOI 1.17/s1729-7-915- Optmal outpatent appontment schedulng Gudo C. Kaandorp Ger Koole Receved: 15 March 26 / Accepted: 28 February 27 / Publshed onlne: 23 May 27 Sprnger

More information

Logical Development Of Vogel s Approximation Method (LD-VAM): An Approach To Find Basic Feasible Solution Of Transportation Problem

Logical Development Of Vogel s Approximation Method (LD-VAM): An Approach To Find Basic Feasible Solution Of Transportation Problem INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME, ISSUE, FEBRUARY ISSN 77-866 Logcal Development Of Vogel s Approxmaton Method (LD- An Approach To Fnd Basc Feasble Soluton Of Transportaton

More information

Using Series to Analyze Financial Situations: Present Value

Using Series to Analyze Financial Situations: Present Value 2.8 Usng Seres to Analyze Fnancal Stuatons: Present Value In the prevous secton, you learned how to calculate the amount, or future value, of an ordnary smple annuty. The amount s the sum of the accumulated

More information

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College Feature selecton for ntruson detecton Slobodan Petrovć NISlab, Gjøvk Unversty College Contents The feature selecton problem Intruson detecton Traffc features relevant for IDS The CFS measure The mrmr measure

More information

NON-CONSTANT SUM RED-AND-BLACK GAMES WITH BET-DEPENDENT WIN PROBABILITY FUNCTION LAURA PONTIGGIA, University of the Sciences in Philadelphia

NON-CONSTANT SUM RED-AND-BLACK GAMES WITH BET-DEPENDENT WIN PROBABILITY FUNCTION LAURA PONTIGGIA, University of the Sciences in Philadelphia To appear n Journal o Appled Probablty June 2007 O-COSTAT SUM RED-AD-BLACK GAMES WITH BET-DEPEDET WI PROBABILITY FUCTIO LAURA POTIGGIA, Unversty o the Scences n Phladelpha Abstract In ths paper we nvestgate

More information

Formulating & Solving Integer Problems Chapter 11 289

Formulating & Solving Integer Problems Chapter 11 289 Formulatng & Solvng Integer Problems Chapter 11 289 The Optonal Stop TSP If we drop the requrement that every stop must be vsted, we then get the optonal stop TSP. Ths mght correspond to a ob sequencng

More information

LECTURE 1: MOTIVATION

LECTURE 1: MOTIVATION LECTURE 1: MOTIVATION STEVEN SAM AND PETER TINGLEY 1. Towards quantum groups Let us begn by dscussng what quantum groups are, and why we mght want to study them. We wll start wth the related classcal objects.

More information

Section 5.3 Annuities, Future Value, and Sinking Funds

Section 5.3 Annuities, Future Value, and Sinking Funds Secton 5.3 Annutes, Future Value, and Snkng Funds Ordnary Annutes A sequence of equal payments made at equal perods of tme s called an annuty. The tme between payments s the payment perod, and the tme

More information

On some special nonlevel annuities and yield rates for annuities

On some special nonlevel annuities and yield rates for annuities On some specal nonlevel annutes and yeld rates for annutes 1 Annutes wth payments n geometrc progresson 2 Annutes wth payments n Arthmetc Progresson 1 Annutes wth payments n geometrc progresson 2 Annutes

More information

The Noether Theorems: from Noether to Ševera

The Noether Theorems: from Noether to Ševera 14th Internatonal Summer School n Global Analyss and Mathematcal Physcs Satellte Meetng of the XVI Internatonal Congress on Mathematcal Physcs *** Lectures of Yvette Kosmann-Schwarzbach Centre de Mathématques

More information

Chapter 4 ECONOMIC DISPATCH AND UNIT COMMITMENT

Chapter 4 ECONOMIC DISPATCH AND UNIT COMMITMENT Chapter 4 ECOOMIC DISATCH AD UIT COMMITMET ITRODUCTIO A power system has several power plants. Each power plant has several generatng unts. At any pont of tme, the total load n the system s met by the

More information

A Probabilistic Theory of Coherence

A Probabilistic Theory of Coherence A Probablstc Theory of Coherence BRANDEN FITELSON. The Coherence Measure C Let E be a set of n propostons E,..., E n. We seek a probablstc measure C(E) of the degree of coherence of E. Intutvely, we want

More information

Multiplication Algorithms for Radix-2 RN-Codings and Two s Complement Numbers

Multiplication Algorithms for Radix-2 RN-Codings and Two s Complement Numbers Multplcaton Algorthms for Radx- RN-Codngs and Two s Complement Numbers Jean-Luc Beuchat Projet Arénare, LIP, ENS Lyon 46, Allée d Itale F 69364 Lyon Cedex 07 jean-luc.beuchat@ens-lyon.fr Jean-Mchel Muller

More information

Lecture 2: Single Layer Perceptrons Kevin Swingler

Lecture 2: Single Layer Perceptrons Kevin Swingler Lecture 2: Sngle Layer Perceptrons Kevn Sngler kms@cs.str.ac.uk Recap: McCulloch-Ptts Neuron Ths vastly smplfed model of real neurons s also knon as a Threshold Logc Unt: W 2 A Y 3 n W n. A set of synapses

More information

Generalizing the degree sequence problem

Generalizing the degree sequence problem Mddlebury College March 2009 Arzona State Unversty Dscrete Mathematcs Semnar The degree sequence problem Problem: Gven an nteger sequence d = (d 1,...,d n ) determne f there exsts a graph G wth d as ts

More information

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm Avalable onlne www.ocpr.com Journal of Chemcal and Pharmaceutcal Research, 2014, 6(7):1884-1889 Research Artcle ISSN : 0975-7384 CODEN(USA) : JCPRC5 A hybrd global optmzaton algorthm based on parallel

More information

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP)

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP) 6.3 / -- Communcaton Networks II (Görg) SS20 -- www.comnets.un-bremen.de Communcaton Networks II Contents. Fundamentals of probablty theory 2. Emergence of communcaton traffc 3. Stochastc & Markovan Processes

More information

POLYSA: A Polynomial Algorithm for Non-binary Constraint Satisfaction Problems with and

POLYSA: A Polynomial Algorithm for Non-binary Constraint Satisfaction Problems with and POLYSA: A Polynomal Algorthm for Non-bnary Constrant Satsfacton Problems wth and Mguel A. Saldo, Federco Barber Dpto. Sstemas Informátcos y Computacón Unversdad Poltécnca de Valenca, Camno de Vera s/n

More information

Logistic Regression. Steve Kroon

Logistic Regression. Steve Kroon Logstc Regresson Steve Kroon Course notes sectons: 24.3-24.4 Dsclamer: these notes do not explctly ndcate whether values are vectors or scalars, but expects the reader to dscern ths from the context. Scenaro

More information

AN EFFECTIVE MATRIX GEOMETRIC MEAN SATISFYING THE ANDO LI MATHIAS PROPERTIES

AN EFFECTIVE MATRIX GEOMETRIC MEAN SATISFYING THE ANDO LI MATHIAS PROPERTIES MATHEMATICS OF COMPUTATION Volume, Number, Pages S 5-578(XX)- AN EFFECTIVE MATRIX GEOMETRIC MEAN SATISFYING THE ANDO LI MATHIAS PROPERTIES DARIO A. BINI, BEATRICE MEINI AND FEDERICO POLONI Abstract. We

More information

Fisher Markets and Convex Programs

Fisher Markets and Convex Programs Fsher Markets and Convex Programs Nkhl R. Devanur 1 Introducton Convex programmng dualty s usually stated n ts most general form, wth convex objectve functons and convex constrants. (The book by Boyd and

More information

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic Lagrange Multplers as Quanttatve Indcators n Economcs Ivan Mezník Insttute of Informatcs, Faculty of Busness and Management, Brno Unversty of TechnologCzech Republc Abstract The quanttatve role of Lagrange

More information

Multiple stage amplifiers

Multiple stage amplifiers Multple stage amplfers Ams: Examne a few common 2-transstor amplfers: -- Dfferental amplfers -- Cascode amplfers -- Darlngton pars -- current mrrors Introduce formal methods for exactly analysng multple

More information

REGULAR MULTILINEAR OPERATORS ON C(K) SPACES

REGULAR MULTILINEAR OPERATORS ON C(K) SPACES REGULAR MULTILINEAR OPERATORS ON C(K) SPACES FERNANDO BOMBAL AND IGNACIO VILLANUEVA Abstract. The purpose of ths paper s to characterze the class of regular contnuous multlnear operators on a product of

More information

On Robust Network Planning

On Robust Network Planning On Robust Network Plannng Al Tzghadam School of Electrcal and Computer Engneerng Unversty of Toronto, Toronto, Canada Emal: al.tzghadam@utoronto.ca Alberto Leon-Garca School of Electrcal and Computer Engneerng

More information

J. Parallel Distrib. Comput.

J. Parallel Distrib. Comput. J. Parallel Dstrb. Comput. 71 (2011) 62 76 Contents lsts avalable at ScenceDrect J. Parallel Dstrb. Comput. journal homepage: www.elsever.com/locate/jpdc Optmzng server placement n dstrbuted systems n

More information

7.5. Present Value of an Annuity. Investigate

7.5. Present Value of an Annuity. Investigate 7.5 Present Value of an Annuty Owen and Anna are approachng retrement and are puttng ther fnances n order. They have worked hard and nvested ther earnngs so that they now have a large amount of money on

More information

Energies of Network Nastsemble

Energies of Network Nastsemble Supplementary materal: Assessng the relevance of node features for network structure Gnestra Bancon, 1 Paolo Pn,, 3 and Matteo Marsl 1 1 The Abdus Salam Internatonal Center for Theoretcal Physcs, Strada

More information

Level Annuities with Payments Less Frequent than Each Interest Period

Level Annuities with Payments Less Frequent than Each Interest Period Level Annutes wth Payments Less Frequent than Each Interest Perod 1 Annuty-mmedate 2 Annuty-due Level Annutes wth Payments Less Frequent than Each Interest Perod 1 Annuty-mmedate 2 Annuty-due Symoblc approach

More information

GRAVITY DATA VALIDATION AND OUTLIER DETECTION USING L 1 -NORM

GRAVITY DATA VALIDATION AND OUTLIER DETECTION USING L 1 -NORM GRAVITY DATA VALIDATION AND OUTLIER DETECTION USING L 1 -NORM BARRIOT Jean-Perre, SARRAILH Mchel BGI/CNES 18.av.E.Beln 31401 TOULOUSE Cedex 4 (France) Emal: jean-perre.barrot@cnes.fr 1/Introducton The

More information

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis The Development of Web Log Mnng Based on Improve-K-Means Clusterng Analyss TngZhong Wang * College of Informaton Technology, Luoyang Normal Unversty, Luoyang, 471022, Chna wangtngzhong2@sna.cn Abstract.

More information

An efficient constraint handling methodology for multi-objective evolutionary algorithms

An efficient constraint handling methodology for multi-objective evolutionary algorithms Rev. Fac. Ing. Unv. Antoqua N. 49. pp. 141-150. Septembre, 009 An effcent constrant handlng methodology for mult-objectve evolutonary algorthms Una metodología efcente para manejo de restrccones en algortmos

More information

On Leonid Gurvits s proof for permanents

On Leonid Gurvits s proof for permanents On Leond Gurvts s proof for permanents Monque Laurent and Alexander Schrver Abstract We gve a concse exposton of the elegant proof gven recently by Leond Gurvts for several lower bounds on permanents.

More information

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK Sample Stablty Protocol Background The Cholesterol Reference Method Laboratory Network (CRMLN) developed certfcaton protocols for total cholesterol, HDL

More information

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement An Enhanced Super-Resoluton System wth Improved Image Regstraton, Automatc Image Selecton, and Image Enhancement Yu-Chuan Kuo ( ), Chen-Yu Chen ( ), and Chou-Shann Fuh ( ) Department of Computer Scence

More information

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy 4.02 Quz Solutons Fall 2004 Multple-Choce Questons (30/00 ponts) Please, crcle the correct answer for each of the followng 0 multple-choce questons. For each queston, only one of the answers s correct.

More information

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S S C H E D A E I N F O R M A T I C A E VOLUME 0 0 On Mean Squared Error of Herarchcal Estmator Stans law Brodowsk Faculty of Physcs, Astronomy, and Appled Computer Scence, Jagellonan Unversty, Reymonta

More information

NPAR TESTS. One-Sample Chi-Square Test. Cell Specification. Observed Frequencies 1O i 6. Expected Frequencies 1EXP i 6

NPAR TESTS. One-Sample Chi-Square Test. Cell Specification. Observed Frequencies 1O i 6. Expected Frequencies 1EXP i 6 PAR TESTS If a WEIGHT varable s specfed, t s used to replcate a case as many tmes as ndcated by the weght value rounded to the nearest nteger. If the workspace requrements are exceeded and samplng has

More information

Power-of-Two Policies for Single- Warehouse Multi-Retailer Inventory Systems with Order Frequency Discounts

Power-of-Two Policies for Single- Warehouse Multi-Retailer Inventory Systems with Order Frequency Discounts Power-of-wo Polces for Sngle- Warehouse Mult-Retaler Inventory Systems wth Order Frequency Dscounts José A. Ventura Pennsylvana State Unversty (USA) Yale. Herer echnon Israel Insttute of echnology (Israel)

More information

The Mathematical Derivation of Least Squares

The Mathematical Derivation of Least Squares Pscholog 885 Prof. Federco The Mathematcal Dervaton of Least Squares Back when the powers that e forced ou to learn matr algera and calculus, I et ou all asked ourself the age-old queston: When the hell

More information

Fast degree elevation and knot insertion for B-spline curves

Fast degree elevation and knot insertion for B-spline curves Computer Aded Geometrc Desgn 22 (2005) 183 197 www.elsever.com/locate/cagd Fast degree elevaton and knot nserton for B-splne curves Q-Xng Huang a,sh-mnhu a,, Ralph R. Martn b a Department of Computer Scence

More information

The circuit shown on Figure 1 is called the common emitter amplifier circuit. The important subsystems of this circuit are:

The circuit shown on Figure 1 is called the common emitter amplifier circuit. The important subsystems of this circuit are: polar Juncton Transstor rcuts Voltage and Power Amplfer rcuts ommon mtter Amplfer The crcut shown on Fgure 1 s called the common emtter amplfer crcut. The mportant subsystems of ths crcut are: 1. The basng

More information

Lecture 3: Annuity. Study annuities whose payments form a geometric progression or a arithmetic progression.

Lecture 3: Annuity. Study annuities whose payments form a geometric progression or a arithmetic progression. Lecture 3: Annuty Goals: Learn contnuous annuty and perpetuty. Study annutes whose payments form a geometrc progresson or a arthmetc progresson. Dscuss yeld rates. Introduce Amortzaton Suggested Textbook

More information

A Performance Analysis of View Maintenance Techniques for Data Warehouses

A Performance Analysis of View Maintenance Techniques for Data Warehouses A Performance Analyss of Vew Mantenance Technques for Data Warehouses Xng Wang Dell Computer Corporaton Round Roc, Texas Le Gruenwald The nversty of Olahoma School of Computer Scence orman, OK 739 Guangtao

More information

An Overview of Financial Mathematics

An Overview of Financial Mathematics An Overvew of Fnancal Mathematcs Wllam Benedct McCartney July 2012 Abstract Ths document s meant to be a quck ntroducton to nterest theory. It s wrtten specfcally for actuaral students preparng to take

More information

An Empirical Study of Search Engine Advertising Effectiveness

An Empirical Study of Search Engine Advertising Effectiveness An Emprcal Study of Search Engne Advertsng Effectveness Sanjog Msra, Smon School of Busness Unversty of Rochester Edeal Pnker, Smon School of Busness Unversty of Rochester Alan Rmm-Kaufman, Rmm-Kaufman

More information