Equivalence Checking of Dissimilar Circuits II
|
|
- Mark Terry
- 6 years ago
- Views:
Transcription
1 Equvalence Checkng of Dssmlar Crcuts II Cadence Berkeley Labs 1995 Unversty Ave.,Sute 460, Berkeley, Calforna,94704 phone: (510) , fax: (510) CDNL-TR August 2004 Eugene Goldberg (Cadence Berkeley Labs), Abstract In ths report we revst the theory ntroduced n [2]. We formulate t n terms of correlaton functons so showng that the ntroducton of flterng functons s not necessary. We also descrbe an algorthm of equvalence checkng for crcuts wth a known specfcaton that s based on computaton of correlaton functons only (no flterng functons are computed). 1. Introducton Ths report s smply a rewrte of paper [2] that presented some results on complexty of equvalence checkng of crcuts wth a common specfcaton. The results of [2] were formulated n terms of correlaton and flterng functons. In ths report, we show that the ntroducton of flterng functons s not necessary and the theory of [2] can be formulated n terms of correlaton functons only. Ths also apples to publcatons [3] and [4] n whch the ntroducton of flterng functons can be avoded ether. (Papers [2] and [4] are avalable onlne from the web page ) In the report we follow the structure of paper [2] wth the followng exceptons. To reduce the repetton of text we omtted materal presented n Sectons 1, 5 and 6 of [2]. The major changes are made n secton 4 whch presents the man result. In partcular, we added subsecton 4.4 n whch we explan why flterng functons can be dropped. Fnally we added a secton descrbng an algorthm for equvalence checkng of crcuts N 1, N 2 wth a known common specfcaton S (Secton 5). Ths algorthm s dentcal to that of [4] wth the followng two exceptons. We formulate ths algorthm n terms of correlaton functons only (flterng functons are not used). We emphasze the fact that for ths algorthm the specfcaton S does not have to be represented explctly. The algorthm only needs to know the parttonng of N 1 nto subcrcuts that are mplementatons of blocks of S. 2. Common Specfcaton of Boolean Crcuts In ths secton, we ntroduce the noton of a common specfcaton of Boolean crcuts. Let S be a combnatonal crcut of mult-valued blocks (further referred to as a specfcaton) specfed by a drected acyclc graph H. The sources and snks of H correspond to prmary nputs and outputs of S. Each non-source node of H corresponds to a mult-valued block computng a mult-valued functon of mult-valued arguments. Each node of n of H s assocated wth a mult-valued varable A. If n s a source of H, then the correspondng varable specfes values taken by the correspondng prmary nput of S. If n s a non-source node of S then the correspondng varable descrbes the values taken by the output of the block specfed by n. If n s a source (respectvely a snk), then the correspondng varable s called a prmary nput varable (respectvely prmary output varable). We wll use the notaton C=G(A,B) to ndcate that a) the output of a block G s assocated wth a varable C; b) the functon computed by the block G s G(A,B); c) only two nodes of H are connected to the node n n H and these nodes are assocated wth varables A and B. Denote by D(A) the doman of the varable A assocated wth a node of H. The value of D(A) s called the multplcty of A. If the multplcty of every varable A of S s equal to 2 then S s a Boolean crcut.
2 Now we descrbe how a Boolean crcut N can be produced from a specfcaton S by encodng the mult-valued varables. Let D(A)={a 1,,a t } be the doman of a varable A of S. Denote by q(a) a Boolean encodng of the values of D(A) that s a mappng q:d(a) {0,1} m. Denote by length(q(a)) the number of bts n q that s the value of m. The value of q(a ), a D(A) s called the code of a. Gven an encodng q of length m of a varable A assocated wth a block of S, denote by v(a) the set of m codng Boolean varables. In the followng exposton we make the assumptons below. Assumpton 1. Each gate of a Boolean crcut and each block of a specfcaton has two nputs and one output. Assumpton 2. The multplcty of each prmary nput (or output) varable of a specfcaton s a power of 2. Assumpton 3. If A s a prmary nput (or output) varable of a specfcaton, then length(q(a))=log 2 ( D(A) ) Assumpton 4. If a 1 and a 2 are values of a varable A of a specfcaton and a 1 a 2, then q(a 1 ) q(a 2 ). Assumpton 5. If A and B are two dfferent varables of a specfcaton, then v(a) v(b) =. Remark 1. From Assumpton 2, Assumpton 3, and Assumpton 4 t follows that f A s a prmary nput (or output) varable, a mappng q:d(a) {0,1} m s bjectve. In partcular, any assgnment to the varables of v(a) s a code of some value a D(A). Defnton 1. Gven a Boolean crcut I, denote by Inp(I) (respectvely Out(I)) the set of varables assocated wth prmary nputs (respectvely prmary outputs) of I. Defnton 2. Let X 1 and X 2 be sets of Boolean varables and X 2 X 1. Let y be an assgnment to the varables of X 1. Denote by proj(y,x 2 ) the projecton of y on X 2.e. the part of y that conssts of the assgnments to the varables of X 2. Defnton 3. Let C=G(A,B) be a block of specfcaton S. Let q(a),q(b),q(c) be encodngs of varables A,B, and C respectvely. A Boolean crcut I s sad to mplement the block G f the followng three condtons hold: 1) The set Inp(I) s a subset of v(a) v(b). 2) The set Out(I) s equal to v(c). 3) If the set of values assgned to v(a) and v(b) form codes q(a) and q(b) respectvely where a D(A), b D(B), then I(z )=q(c). Here z s the projecton of the assgnment z=(q(a),q(b)) on Inp(I), I(z ) s the value taken by I at z, and c=g(a,b). Remark 2. The reason why Inp(I) may not nclude all the varables of v(a) and/or v(b) s that the functon G(A,B) may not dstngush some values of A or B. (G(A,B) does not dstngush, say, values a 1,a 2 D(A), f for any b D(B), G(a 1,b)=G(a 2,b).) So to mplement G(A,B) the crcut I may need only a subset of varables of v(a) v(b). Ths sad, for the sake of smplcty, we wll wrte I(q(a),q(b)) meanng I(q (a),q (b)), q (a)= proj(q(a),inp(i)) and q (b)=proj(q(b),inp(i)). Defnton 4. Let S be a mult-valued crcut. A Boolean crcut N s sad to mplement the specfcaton S, f t s bult accordng to the followng two rules. 1) Each block G of S s replaced wth an mplementaton I of G. 2) Let the output of block G 1 (specfed by varable R) be connected to an nput of block G 2 (specfed by the same varable R) n S. Then the outputs of the crcut I 1 mplementng G 1 are properly connected to nputs of crcut I 2 mplementng G 2. Namely, the prmary output of I 1 specfed by a Boolean varable x v(r) s connected to the nput of I 2 specfed by the same varable of v(r) f x Inp(I 2 ). In Fg. 1a a specfcaton of three blocks s shown. The functonalty of two dfferent mplementatons of the block C=G 1 (A,B) (Fg. 1b) are shown n Fg. 1c and 1d. Here D(A)={a 0,a 1 }, D(B)={b 0,b 1,b 2,b 3 } and D(C)={c 0,c 1,c 2 }. Snce A and B are prmary nput varables, they are encoded wth a mnmum length encodng and q 1 (A)=q 2 (A) and q 1 (B)=q 2 (B) where q 1 (a 0 )=0, q 1 (a 1 )=1, q 1 (b 0 )=00, q 1 (b 1 )=01, q 1 (b 2 )=10, q 1 (b 3 )=11. Fnally, the encodngs q 1 (C) and q 2 (C) are q 1 (c 0 )=00, q 1 (c 1 )=10, q 1 (c 2 )=01 and q 2 (c 0 )=100, q 2 (c 1 )=010, q 2 (c 2 )=001. Remark 3. Let N be an mplementaton of a specfcaton S. Let p be the largest number of gates used n an mplementaton of a mult-valued block of S n N. We wll say that S s a specfcaton of granularty p for N. C G 3 A B F E K G 1 G 2 (a) I 1 (q 1 (A),q 1 (B)) q 1(A) q 1(B) q 1(C) (c) C=G 1 (A,B) A B C a 0 b 0 c 0 a 0 b 1 c 1 a 0 b 2 c 1 a 0 b 3 c 0 a 1 b 0 c 1 a 1 b 1 c 2 a 1 b 2 c 2 a 1 b 3 c 0 q 2(A) q 2(B) q 2(C) Fgure 1. A specfcaton and the functonalty of two mplementatons of a block Defnton 5. The topologcal level of a block G n a specfcaton S s the length of the longest path from a prmary nput of S to G. (The length of a path s measured n the number of blocks on t. The topologcal level of a prmary nput s assumed to be 0.) Denote by level(g) the topologcal level of G n S. (d) (b) I 2 (q 2 (A),q 2 (B))
3 Let N be an mplementaton of a specfcaton S. From Assumpton 4 t follows that for any value assgnment h to the nput varables of N there s a unque set of values (x 1,,x k ), where x D(X ) such that h=(q(x 1 ),,q(x k )). That s there s oneto-one correspondence between assgnments to prmary nputs of S and N. The same apples to prmary outputs of S and N. Defnton 6. Let N be an mplementaton of S. Gven a Boolean vector y of assgnments to the prmary nputs of N, the correspondng vector Y=(x 1,..,x k ), x D(X ) such that y=(q(x 1 ),,q(x k )) s called the pre-mage of y. Proposton 1. Let N be a crcut mplementng specfcaton S. Let I(G) be the mplementaton of a block C=G(A,B) of S n N. Let y be a value assgnment to the prmary nput varables of N and Y be the pre-mage of y. Then the values of prmary outputs of I(G) form the code q(c) where c s the value taken by the output of G when the nputs of S take the values specfed by Y. Proof. The proposton can be proven by nducton n topologcal levels of varables of the specfcaton S. Accordng to Remark 1, the proposton holds for the varables of topologcal level 0 (prmary nput varables of S). Let C=G(A,B) be a block of S and level(g)=n, n>0. Let I(G) be the mplementaton of G n N. By the nducton hypothess, values taken by the varables of v(a) and v(b) n N under the nput assgnment y should be q(a) and q(b) respectvely. Here a and b are values of varables A and B under the nput assgnment Y. Then from Defnton 3 t follows that the outputs of I(G) take the values of q(c) where c=g(a,b). Proposton 2. Let N 1, N 2 be crcuts mplementng a specfcaton S. Let each prmary nput (or output) varable X of S have the same encodng n N 1. Then Boolean crcuts N 1 are functonally equvalent. Proof. Let y be an arbtrary assgnment to nput varables of N 1. Snce the encodngs of prmary nput varables of S n N 1 are the same, then the pre-mage Y of y for N 1 s the same. Let C be a prmary output varable of S assocated wth a block G. From Proposton 1 t follows that the values taken by the mplementatons I 1 (G) and I 2 (G) of G n N 1 are equal to q 1 (c) and q 2 (c) respectvely. Here c s the value taken by the output of G under nput assgnment Y and q 1 and q 2 are encodngs of the prmary output varable C n N 1. Snce C has the same encodng n N 1, then q 1 (c) = q 2 (c). Defnton 7. Let N 1, N 2 be two functonally equvalent Boolean crcuts. Let N 1, N 2 mplement a specfcaton S so that for every prmary nput (output) varable X encodngs q 1 (X) and q 2 (X) (used when producng N 1 respectvely) are dentcal. Then N s called a common specfcaton (CS) of N 1. Defnton 8. Let S be a CS of N 1. Let p 1 (respectvely p 2 ) be the granularty of S wth respect to N 1 (respectvely N 2 ). Then we wll say that S s a CS of N 1 of granularty p = max(p 1,p 2 ). Defnton 9. Gven two functonally equvalent Boolean crcuts N 1, N 2, S s called the fnest common specfcaton f t has the smallest granularty p among all the CSs of N Equvalence Checkng as SAT Snce n ths report we formulate the complexty of equvalence checkng n terms of resoluton proofs, we recall a common way of reducng equvalence checkng to the satsfablty problem. Defnton 10. A dsjuncton of lterals of Boolean varables not contanng two lterals of the same varable s called a clause. A conjuncton of clauses s called a conjunctve normal form (CNF). Defnton 11. Gven a CNF F, the satsfablty problem (SAT) s to fnd a value assgnment to the varables of F for whch F evaluates to 1 (also called a satsfyng assgnment) or to prove that such an assgnment does not exst. A clause K of F s sad to be satsfed by a value assgnment y f K(y)=1. The standard converson of an equvalence checkng problem nto an nstance of SAT s performed n two steps. Let N 1 be Boolean crcuts to be checked for equvalence. At the frst step of ths converson, a crcut M called a mter [1] s formed from N 1. The mter M s obtaned by 1) dentfyng the correspondng prmary nputs of N 1 ; 2) XORng each par of correspondng prmary outputs of N 1 ; 3) ORng the outputs of the added XOR gates. So the mter of N 1 evaluates to 1 f and only f for some nput assgnment a prmary output of N 1 and the correspondng output of N 2 evaluate to dfferent values. Therefore, the problem of checkng the equvalence of N 1 s equvalent to testng the satsfablty of the mter of N 1. At the second step of converson, the satsfablty of the mter s reduced to that of a CNF formula F. Ths formula s a conjuncton of CNF formulas F 1,..,F n specfyng the functonalty of the gates of M and a one-lteral clause that s satsfed only f the output of M s set to 1. The CNF F specfes the -th gate g of M. Any assgnment to the varables of F that s nconsstent wth the functonalty of g falsfes a clause of F (and vce versa, a consstent assgnment satsfes all the clauses of F.) For nstance, the AND gate y=x 1 x 2 s specfed by the followng three clauses ~x 1 ~x 2 y, x 1 ~y, x 2 ~y. 4. Equvalence Checkng n General Resoluton In ths secton, we prove some results about the complexty of equvalence checkng of crcuts wth a CS of granularty p. The man dea of the proof s that f S s a CS of N 1, then ther equvalence checkng reduces to computng the correlaton functon for encodngs q 1 (C),q 2 (C) of each varable C of S. The two man propertes of correlaton functons are that They can be bult based only on the nformaton about the topology of S and about assgnment of gates of N 1 to blocks of S. The correlaton functon for encodngs q 1 (C),q 2 (C) of the varable C specfyng the output of a block G(A,B) can be computed locally from the correlaton functons of encodngs of varables A and B and CNFs specfyng mplementatons I 1 (G) and I 2 (G). So these functons can be computed n topologcal order startng wth nputs and proceedng to outputs.
4 4.1 Class M(p) and general resoluton Defnton 12. Gven a constant p, a CNF formula F s a member of the class M(p) f and only f t satsfes the followng two condtons. F s the CNF formula (obtaned by the procedure descrbed n Secton 3) specfyng the mter of a par of functonally equvalent crcuts N 1. N 1 has a CS of granularty p. Defnton 13. Let K and K be clauses havng opposte lterals of a varable (say varable x) and there s only one such varable. The resolvent of K, K n varable x s the clause that contans all the lterals of K and K but the postve (.e. lteral x) and negatve (.e. lteral ~x) lterals of x. The operaton of producng the resolvent of K and K s called resoluton. Defnton 14. General resoluton s a proof system of propostonal logc that has only one nference rule. Ths rule s to resolve two exstng clauses to produce a new one. Gven a CNF formula F, a proof L(F) of unsatsfablty of F n the general resoluton system conssts of a sequence of resolutons resultng n the dervaton of an empty clause (.e. a clause wthout lterals). General resoluton s complete, whch means that gven an unsatsfable formula F there s always a proof L(F) that derves an empty clause. Defnton 15. Let F be a set of clauses. Denote by supp(f) the set of varables whose lterals occur n clauses of F. The followng Proposton 3 and also Propostons 4,5,6 proved n subsectons 4.2, 4.3 are used n the proof of Proposton 7 that s the man result of ths paper. Proposton 3. Let F be a set of clauses that mples a clause K. Then there s a sequence of resolutons of at most 3 supp(f) steps that results n the dervaton of a clause that mples K. Proof. Denote by F the formula that s obtaned from F by makng the assgnments that set the lterals of K to 0 (and removng the satsfed clauses and the lterals set to 0). It s not hard to see that F s unsatsfable snce t mples an empty clause. So there s a resoluton proof L(F ) that results n deducng an empty clause. Then by replacng each clause of F nvolved n L(F ) wth ts parent clause from F we get a sequence of resolutons resultng n deducng a clause that mples K. The number of resolvents n L(F ) cannot be more than 3 supp(f ) (.e. the total number of clauses of supp(f ) varables) and so t cannot be more than 3 supp(f). 4.2 Correlaton functons Defnton 16. Let S be a CS of crcuts N 1 and C be a varable of S. A functon Cf (v 1 (C),v 2 (C)) s called a correlaton functon for encodngs q 1 and q 2 of the values of C (used when producng N 1 ) f : supp(cf ) v 1 (C) v 2 (C). Cf(z 1, z 2 )=1 for any assgnment z 1 to v 1 (C) and z 2 to v 2 (C) such that z 1 =q 1 (c) and z 2 =q 2 (c) where c D(C). Otherwse Cf(z 1, z 2 )=0. From now on we wll say that Cf(v 1 (C),v 2 (C)) s the correlaton functon for the varable C meanng that t s the correlaton functon for encodngs q 1 (C),q 2 (C) of the varable C. Remark 4. If C s a prmary nput varable of S, then Cf(v 1 (C),v 2 (C)) Eq(v 1 (C),v 2 (C)). The functon Eq(z 1, z 2 ) s equal to 1 ff z 1 =q 1 (c), z 2 = q 2 (c), where c D(C) and q 1 (c) = q 2 (c). Otherwse, Eq(v 1 (C),v 2 (C)) s equal to 0. Indeed, as t follows from Remark 1, sets v 1 (C) and v 2 (C) have the same number of codng varables and any assgnment to v 1 (C) or v 2 (C) s the code of a value c D(C). Besdes, from the defnton of CS t follows that q 1 (C)=q 2 (C). So every permssble assgnment (x,y) to the varables of v 1 (C),v 2 (C) can be represented as (q 1 (c),q 2 (c)), were c D(C) and q 1 (c) = q 2 (c). Proposton 4. Let S be a CS of crcuts N 1. Let C=G(A,B) be a block of S. Let F be the CNF formula specfyng the mter of N 1 bult as descrbed n Secton 3. Let F(I 1 (G)) and F(I 2 (G)) be the part of F specfyng the mplementaton I 1 (G) and I 2 (G) of G n N 1 respectvely. Then P mples Cf(v 1 (C),v 2 (C)). Here P = Correlaton Implementaton and, Correlaton = Cf(v 1 (A),v 2 (A)) Cf(v 1 (B),v 2 (B)), Implementaton = F(I 1 (G)) F(I 2 (G)). Proof. To prove that P mples Cf(v 1 (C),v 2 (C)) one needs to show that any assgnment that sets P to 1 also sets Cf(v 1 (C),v 2 (C)) to 1. It s not hard to see that the support of all the functons of the expresson P Cf(v 1 (C),v 2 (C)) s a subset of supp(f(i 1 (G)) supp(f(i 2 (G)). Let h=(x 1, x 2, y 1, y 2, z 1, z 2 ) be an assgnment that sets P to 1 where x 1, x 2, y 1, y 2, z 1, z 2 are assgnments to v 1 (A), v 2 (A), v 1 (B), v 2 (B), v 1 (C), v 2 (C) respectvely. Then h has to set to 1 all the functons the conjuncton of whch forms P. Snce h has to set the functon Correlaton to 1, then x 1 =q 1 (a), x 2 =q 2 (a) where a D(A) and y 1 =q 1 (b), y 2 =q 2 (b), where b D(B). So h=(q 1 (a),q 2 (a), q 1 (b),q 2 (b), z 1, z 2 ). Snce h sets the functon Implementaton to 1, then z 1 has to be equal to q 1 (c), c=g(a,b) and z 2 has to be equal to q 2 (c). So h s equal to (q 1 (a),q 2 (a),q 1 (b),q 2 (b),q 1 (c),q 2 (c)) and hence t sets the correlaton functon Cf(v 1 (C),v 2 (C)) to Complexty of formulas from M(p) Proposton 5. Let A,B,C be Boolean functons and A B C. Then for any functon A such that A A, t s true that A B C. Proof. Let x be an assgnment that sets A B to 1. Then A (x)=1 and B(x)=1. Snce A A, then A(x)=1. Then A(x) B(x) = 1 and so C(x)=1. Proposton 6. Let X 1 and X 2 be sets of Boolean varables, F(X 1,X 2 ) and H(X 2 ) be CNF formulas and F mply H. Then n at most 3 supp(f) resoluton steps one can derve a CNF formula H that mples H(X 2 ) such that supp(h ) supp(h). Proof. Let K be a clause of H. From Proposton 3 t follows that n at most 3 supp(f) steps one can derve a clause K that mples K. Snce K K, then supp(k ) supp(k). So n at most H 3 supp(f) steps, where H s the number of clauses n H, one can derve a CNF H mplyng H such that supp(h ) supp(h). (The fact that H mples H follows from Proposton 5.) However, f
5 one does not produce the same resolvent twce, the total number of resoluton steps when dervng H cannot be more than 3 supp(f) (because t s the total number of clauses of supp(f) varables). Proposton 7. Let F be a formula of M(p) specfyng the mter of crcuts N 1 obtaned from a CS S of granularty p. The unsatsfablty of F can be proven by a resoluton proof of no more than d n 3 6p resoluton steps where n s the number of blocks n S and d s a constant. Proof. From Proposton 4 t follows that one can deduce correlaton functons for all the varables of S startng wth blocks of topologcal level 1 and proceedng n topologcal order. Indeed, let C=G(A,B) be a block of topologcal level 1. Then A and B are prmary nput varables and the correlaton functons for them are equal to Eq(v 1 (A),v 2 (A)) and Eq(v 1 (B),v 2 (B)) respectvely (see Remark 4). The correlaton functon Cf(v 1 (C),v 2 (C)) s mpled by P =F(I 1 (G)) F(I 2 (G)) Eq(v 1 (A),v 2 (A)) Eq(v 1 (B),v 2 (B)). So a functon mplyng Cf(v 1 (C),v 2 (C)) can be derved from P by resoluton. From Proposton 5 t follows that to apply Proposton 4, nstead of the functon Cf(v 1 (C),v 2 (C)), one can use any functon mplyng t. After correlaton functons are computed for all the varables of level 1, the same procedure can be appled to varables of topologcal level 2 and so on. If S conssts of n blocks, then n n steps one can deduce correlaton functons for the prmary output varables of S. At each step the correlaton functon s computed for a varable C=G(A,B) of S. The complexty of ths step s no more than 3 6p. Indeed, the support of all functons mentoned n Proposton 4 needed for computng Cf(v 1 (C),v 2 (C)) s a subset of A=supp(F(I 1 (G))) supp(f(i 2 (G))). The total number of gates n I 1 (G) and I 2 (G) s bounded by 2p, each gate havng 2 nputs and 1 output. So the total number of varables n A cannot be more than 6p. Then from Proposton 6 t follows that n at most 3 6p steps one can deduce CNFs mplyng Cf(v 1 (C),v 2 (C)). Then the total number of resoluton steps one needs to deduce functons mplyng the correlaton functons for the prmary output varables of S s bounded by n 3 6p. Now we show that from the correlaton functons for prmary output varables of S, one can deduce an empty clause n the number of resoluton steps lnear n n p. Let C be a prmary output varable specfyng the output of a block G of N. Let I 1 (G) and I 2 (G) be the mplementatons of G n N 1 respectvely. Let D(C) =2 k (By Assumpton 2 the multplcty of C s a power of 2.) Then length(q 1 (C))= length(q 2 (C))=k. (By Assumpton 3, values of S are encoded by a mnmal length encodng.) Now we show that there s always a correlaton functon Cf(v 1 (C),v 2 (C)) that mples the CNF consstng of k pars of two lteral clauses specfyng the equvalence of correspondng outputs of I 1 (G) and I 2 (G). Let f 1 and f 2 be two Boolean varables of v 1 (C) and v 2 (C) respectvely that specfy correspondng outputs of N 1. Snce S s a CS of N 1, then q 1 (C)=q 2 (C). So any assgnment q 1 (c),q 2 (c) to v 1 (C) and v 2 (C) that satsfes Cf(v 1 (C),v 2 (C)) also satsfes clauses K =f 1 ~f 2 and K =~f 1 f 2. So K and K are mpled by Cf(v 1 (C),v 2 (C)) and so clauses mplyng them can be deduced by the procedure descrbed n the proof of Proposton 4. (The resoluton steps one needs to deduce equvalence clauses are already counted n the expresson n 3 6p ) Usng each par of equvalence clauses K and K (or clauses mplyng them) and the clauses specfyng the gate g=xor(f 1,f 2 ) of the mter, one can deduce a sngle lteral clause ~g. Ths clause requres settng the output of ths XOR gate to 0. Each such a clause can be deduced n the number of resolutons bounded by a constant and the total number of such clauses cannot be more than n p. Fnally, from these unt clauses and the clauses specfyng the fnal OR gate of the mter, the empty clause can be deduced n the number of resolutons bounded by n p. So the empty clause s deduced n no more than n 3 6p + d n p steps where d s a constant. Fnally, one can pck a constant d such n 3 6p + d n p d n 3 6p Remark 5. The essence of the resoluton proof descrbed n Proposton 7 s to compute correlaton functons nductvely movng from nputs to outputs. It s not hard to see that ths computaton s not restrcted to general resoluton. Indeed, all the terms of the expresson P=Cf(v 1 (A), v 2 (A)) Cf(v 1 (B), v 2 (B)) F(I 1 (G)) F(I 2 (G)) are just functons and so can be represented n any possble way (.e. not only as CNF formulas). Besdes, Proposton 7 can be proven n terms of exstental quantfcaton ntroduced by Defnton 17. Indeed, from Proposton 8 below t follows that Cf(v 1 (C),v 2 (C)) (or a functon mplyng t) can be obtaned from P by exstentally quantfyng away all the varables except those of v 1 (C) v 2 (C). Exstental quantfcaton of a functon can be done n many ways, for example, by usng BDDs. So, summarzng, Proposton 7 can be formulated and proven n terms of functons and exstental quantfcaton.e. ndependently of a proof system. Defnton 17. Let f be a Boolean functon. We wll say that functon f* s obtaned from f by exstentally quantfyng away Boolean varable x f f* = f(,x=0, ) f(,x=1,.). Proposton 8. Let X 1 and X 2 be two dsjont sets of Boolean varables. Let F(X 1,X 2 ) and H(X 2 ) be two Boolean functons and F mply H. Let F*(X 1 \ {x}, X 2 ) be obtaned from F(X 1,X 2 ) by exstentally quantfyng away the varable x. Then F*(X 1 \ {x},x 2 ) also mples H(X 2 ). Proof. Denote by X 1 the set X 1 \{x}. Let (z, z 1,z 2 ) be a boolean vector representng an assgnment to the varables of X 1 X 2. Here z s a Boolean value assgned to the varable x and z 1, z 2 are Boolean vectors representng assgnments to the varables of X 1 and X 2 respectvely. Suppose that F*(z 1, z 2 ) = 1. Accordng to Defnton 17, F*(X 1,X 2 ) = F(0, X 1,X 2 ) F(1, X 1,X 2 ) and so ether F(0, z 1, z 2 ) or F(0, z 1, z 2 ) has to be equal to 1. Snce F(x,X 1,X 2 ) mples H(X 2 ) then H(z 2 ) = 1. So from F*=1 t follows that H=1. Hence F* mples H. Remark 6. In Proposton 7 we gve a worst case estmate for the complexty of correlaton functon computaton. In practce, ths complexty can be much lower. In a sense, the best way to nterpret the theory developed n ths secton s that the complexty of equvalence checkng of crcuts N 1 wth a CS S s lnear n the number of blocks n S. Remark 7. In ths report, for the sake of clarty, we assumed that every block of a specfcaton has two nputs and one output (Assumpton 1). However, one can easly extend Proposton 7 to the case of a specfcaton S where a block may have an arbtrary (but fnte) number of nputs. (We stll assume that every gate of crcuts N 1 mplementng S have two nputs and one output). Indeed, let G be a block of S wth n nputs and let C, A 1,.., A n be varables assocated wth ts output
6 and n nputs respectvely. Then one can prove (n the same manner as n Proposton 4) that correlaton functon Cf(v 1 (C),v 2 (C)) s mpled by the expresson P = Correlaton Implementaton. Here Implementaton= F(I 1 (G)) F(I 2 (G)) s the same as n Proposton 4 and Correlaton = Cf(v 1 (A 1 ), v 2 (A 1 )) Cf(v 1 (A n ), v 2 (A n )). So to compute the correlaton functon for the output of an n-nput block one needs to compute n correlaton functons correspondng to n nput varables. Other than that, the proof of Proposton 7 does not change. 4.4 A few words about flterng functons In subsecton 4.3, we reproduced the result of [2] wthout ntroducng flterng functons. To make thngs even more clear, n ths subsecton we gve an nformal explanaton of why flterng functons can be dropped. We also explan under what crcumstances flterng functons mght come useful. Here s the defnton of flterng functons from [2]. Defnton 18. Let N be an mplementaton of a specfcaton S. Let C be a varable of S assocated wth the output of a block G. A functon Ff s called a flterng functon f: supp(ff ) v(c). If an assgnment z to the varables of v(c) s a code q(c), c D(C), then Ff(z)=1. Otherwse, Ff(z)=0. Let N 1 be crcuts wth a CS S. Let v 1 (C) and v 2 (C) be the codng varables of the varable C of S correspondng to the mplementatons I 1 (G) and I 2 (G) n N 1 respectvely. From Defnton 18 of flterng functons and Defnton 16 of correlaton functons t follows that Ff(v 1 (C)) Ff(v 2 (C)) Cf(v 1 (C),v 2 (C)) = Cf(v 1 (C),v 2 (C)). On the other hand, n Proposton 7 of [2] (used to prove the man result.e. Proposton 8) flterng functons Ff(v 1 (C)) Ff(v 2 (C)) appear only n conjuncton wth the correlaton functon Cf(v 1 (C),v 2 (C)). So flterng functons can be removed from the proof. The reason why flterng functons appeared n [2] was that orgnally the defnton of correlaton functons used n the manuscrpt was as follows. Defnton 19. Let S be a CS of crcuts N 1 and C be a varable of S. A functon Cf s called a correlaton functon for encodngs q 1 and q 2 of the values of C (used when obtanng N 1 ) f : supp(cf ) v 1 (C) v 2 (C). Cf(z 1, z 2 )=0 for any assgnment z 1 to v 1 (C) and z 2 to v 2 (C) such that z 1 =q 1 (c) and z 2 =q 2 (c*) where c,c* D(C) and c c*. It s not hard to see that Defnton 19, only partally defnes Cf. Defnton 19 can be vewed s a relaxaton of Defnton 16, meanng that the correlaton functon specfed by the latter s an mplementaton of the correlaton functon specfed by the former. If correlaton functons are specfed by Defnton 19, then to prove Proposton 7 one needs flterng functons. (Because now Ff(v 1 (C)) Ff(v 2 (C)) Cf(v 1 (C),v 2 (C)), n general, s not equvalent to Cf(v 1 (C),v 2 (C)).) Later, the defnton of correlaton functons was changed to the one used n [2] whch made flterng functons redundant. Defnton 16 s preferable from a theoretcal pont of vew because t reduces the number of objects employed n our theory. However, n practcal mplementatons of the algorthm of equvalence checkng descrbed n Secton 5, the use of flterng and correlaton functons specfed by Defnton 18 and Defnton 19 respectvely, nstead of correlaton functons specfed by Defnton 16, may make sense. The reason s that Defnton 16 mxes up two unary and one bnary relaton specfed over the set of output assgnments of subcrcuts I 1 (G) and I 2 (G). On the other hand, n practcal applcatons one may want to compute them separately. (The unary relatons are specfed by the flterng functons Ff(v 1 (C)) and Ff(v 2 (C)) that sngle out output assgnments of I 1 (G) and I 2 (G) that are codes of C. The bnary relaton s gven by the correlaton functon Cf(v 1 (C),v 2 (C)) specfed by Defnton 19. ) 5. Algorthm of Equvalence Checkng wth a Known Specfcaton In ths secton we descrbe an algorthm for equvalence checkng of crcuts wth a known specfcaton. Ths algorthm s dentcal to the one ntroduced n [4] wth a few exceptons. Frst, we formulate ths algorthm n terms of correlaton functons only (omttng flterng functons). Second, we emphasze the fact that ths equvalence checkng procedure needs only an mplct representaton of a CS of crcuts N 1, N 2. Ths representaton s gven as a parttonng of N 1, N 2 nto subcrcuts. In Secton 4 we consdered equvalence checkng n general resoluton that s a non-determnstc proof system. Ths means that ths proof s guded by an oracle that ponts to the next par of clauses to be resolved. Now we summarze the results of Secton 4 n a determnstc procedure of equvalence checkng of crcuts N 1 wth a CS S of granularty p. The dea s that f a CS S of N 1 s known, then S tself can be vewed as an oracle. Ths oracle s powerful enough to make equvalence checkng of N 1 effcent. (However, f S s unknown t s unlkely that there s an effcent algorthm for equvalence checkng of N 1 even f there exsts a CS of N 1 of small granularty.) For the sake of smplcty, we wll assume that all the prmary nput and output varables of S are bnary. (A more general case mpled by Assumpton 2 and Assumpton 3 s not much dfferent but makes explanaton more wordy.) We wll also assume that N 1 have only one prmary output. Besdes, we gve the descrpton of the algorthm that s ndependent of the proof system (see Remark 5). Note that n the proof of Proposton 7 we never used an explct representaton of blocks of S. We only needed to know how gates of N 1 are assgned to subcrcuts that are mplementatons of blocks of S. So n the algorthm descrpton shown n Fgure 2, a k-block specfcaton S of N 1 s represented mplctly as a parttonng of these two crcuts nto k subcrcuts N 1 1,..,N k 1, N 1 2,..,N k 2. We assume that N 1 and N 2 are mplementatons of the same block of specfcaton S. We also
7 assume that subcrcuts are numbered n the topologcal order of blocks n S. That s f > j, then the topologcal level of the block mplemented by N 1 s greater or equal to the topologcal level of the block mplemented by N 1 j. /* Part(N 1 )= {N 1 1,..,N 1 k }, Part(N 2 )= {N 2 1,.. k } */ check_for_equvalence(n 1, N 2, Part(N 1 ),Part(N 2 )) { /* check that specfcaton s correct topolgcally */ f (check_parttonns(part(n 1 ),Part(N 2 )) == ncorrect ) return( unsolved ); /* compute correlaton functons */ for (=1; <= k ; ++) {Correlaton = comp_np_corr_func(n 1 ); Cf(N 1, N 2 ) = comp_out_corr_func(n 1, Correlaton); } /* check the correlaton functon of the last par of subcrcuts */ f (Cf(N 1 k, N 2 k ) mples equvalence_ functon) return( equvalent ); else return( unsolved ); } Fgure 2. Pseudocode of equvalence checkng algorthm The pseudocode of our algorthm for equvalence checkng s gven n Fgure 2. The procedure check_parttons checks that specfcaton S represented by Part(N 1 ),Part(N 2 ) s correct topologcally. Namely, t checks that f outputs of subcrcut N 1 are (not) connected to nputs of subcrcut N j 1, then outputs of subcrcut N 2 should (not) be connected to nputs of N j 2. If ths s not true, the check_parttons procedure returns result ncorrect. In the man loop we compute the correlaton functons Cf(N 1,N 2 ) n topologcal order. (Note that n Fgure 2 we denote correlaton functon dfferently to emphasze the fact that specfcaton S s represented mplctly. Here Cf(N 1,N 2 ) denotes what we prevously denoted as Cf(v 1 (C),v 2 (C)) where C s the varable assocated wth the output of the block of S mplemented by N 1 and N 2.). Before computng Cf(N 1,N 2 ) the procedure comp_np_corr_func forms the expresson Correlaton. Ths expresson s a conjuncton of the correlaton functons correspondng to subcrcuts whose outputs are connected to nputs of N 1 and N 2. the correlaton functons correspondng to the prmary nputs of N 1 (f any) that are n the fann of N 1 or N 2. If, for example, nputs of N 1 are connected only to outputs of subcrcuts N j 1 and N m 1 (and so nputs of N 2 are connected only to j outputs of N 2 and N m 2 ), then Correlaton = Cf(N j 1,N j 2 ) Cf(N m 1,N m 2 ). On the other hand, f an nput x 1 of N 1 s a prmary nput of N 1 (and so the correspondng nput x 2 of N 2 s a prmary nput of N 2 ), then n the conjuncton of terms specfyng Correlaton there s term Eq(x 1,x 2 ) descrbng the equvalence of x 1 and x 2. The functon comp_out_corr_func computes the correlaton functon Cf(N 1,N 2 ) by exstentally quantfyng the functon P=Implementaton Correlaton. The functon Implementaton = F(N 1 ) F(N 2 ) descrbes consstent assgnments to the varables of N 1 and N 2. The functon Cf(N 1,N 2 ) s obtaned from P by exstentally quantfyng away all the varables of N 1 except the ones correspondng to outputs of N 1 and N 2. Fnally, the algorthm checks f the correlaton functon of subcrcuts N k 1 and N k 2 (whose prmary outputs are prmary outputs of N 1 ) mples the equvalence functon. If yes, then N 1 are equvalent. Otherwse, the algorthm returns the unsolved answer. The complexty of the algorthm shown n Fgure 2 s the same as n general resoluton.e. d n 3 6p where d s a constant. That s for the class of formulas M(p) wth the fxed value of p, the complexty of ths algorthm s lnear n crcut sze. 6. Conclusons In ths report, we prove the results of paper [2] wthout usng the noton of flterng functons. Ths allows us to smplfy the formulaton of our theory for equvalence checkng of crcuts wth a common specfcaton. Besdes, we gve a modfed descrpton of the algorthm for equvalence checkng of crcuts wth a known specfcaton. In ths descrpton we use only correlaton functons (omttng fterng functons) and emphasze the fact that CS s represented mplctly. 7. References [1] Brand, D., Verfcaton of large syntheszed desgns. Proceedngs of ICCAD-1993,pp [2] E.Goldberg, Y.Novkov. Equvalence Checkng of Dssmlar Crcuts. Internatonal Workshop on Logc and Synthess, May 28-30,2003,USA. [3] E.Goldberg, What Sat-solvers can and cannot do. pp n Advanced Formal Verfcaton, edted by Rolf Drechsler,2004, Kluwer Academc Publshers. [4] E.Goldberg, Y.Novkov. On complexty of equvalence checkng. Techncal Report, CDNL-TR , August, 2003.
Affine transformations and convexity
Affne transformatons and convexty The purpose of ths document s to prove some basc propertes of affne transformatons nvolvng convex sets. Here are a few onlne references for background nformaton: http://math.ucr.edu/
More informationProblem Set 9 Solutions
Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem
More informationCalculation of time complexity (3%)
Problem 1. (30%) Calculaton of tme complexty (3%) Gven n ctes, usng exhaust search to see every result takes O(n!). Calculaton of tme needed to solve the problem (2%) 40 ctes:40! dfferent tours 40 add
More informationMore metrics on cartesian products
More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of
More informationa b a In case b 0, a being divisible by b is the same as to say that
Secton 6.2 Dvsblty among the ntegers An nteger a ε s dvsble by b ε f there s an nteger c ε such that a = bc. Note that s dvsble by any nteger b, snce = b. On the other hand, a s dvsble by only f a = :
More informationFoundations of Arithmetic
Foundatons of Arthmetc Notaton We shall denote the sum and product of numbers n the usual notaton as a 2 + a 2 + a 3 + + a = a, a 1 a 2 a 3 a = a The notaton a b means a dvdes b,.e. ac = b where c s an
More informationInner Product. Euclidean Space. Orthonormal Basis. Orthogonal
Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,
More informationFINITELY-GENERATED MODULES OVER A PRINCIPAL IDEAL DOMAIN
FINITELY-GENERTED MODULES OVER PRINCIPL IDEL DOMIN EMMNUEL KOWLSKI Throughout ths note, s a prncpal deal doman. We recall the classfcaton theorem: Theorem 1. Let M be a fntely-generated -module. (1) There
More informationReport on Image warping
Report on Image warpng Xuan Ne, Dec. 20, 2004 Ths document summarzed the algorthms of our mage warpng soluton for further study, and there s a detaled descrpton about the mplementaton of these algorthms.
More informationThe Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction
ECONOMICS 5* -- NOTE (Summary) ECON 5* -- NOTE The Multple Classcal Lnear Regresson Model (CLRM): Specfcaton and Assumptons. Introducton CLRM stands for the Classcal Lnear Regresson Model. The CLRM s also
More informationThe Minimum Universal Cost Flow in an Infeasible Flow Network
Journal of Scences, Islamc Republc of Iran 17(2): 175-180 (2006) Unversty of Tehran, ISSN 1016-1104 http://jscencesutacr The Mnmum Unversal Cost Flow n an Infeasble Flow Network H Saleh Fathabad * M Bagheran
More informationThe Order Relation and Trace Inequalities for. Hermitian Operators
Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence
More informationModule 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur
Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:
More informationNP-Completeness : Proofs
NP-Completeness : Proofs Proof Methods A method to show a decson problem Π NP-complete s as follows. (1) Show Π NP. (2) Choose an NP-complete problem Π. (3) Show Π Π. A method to show an optmzaton problem
More informationStanford University CS254: Computational Complexity Notes 7 Luca Trevisan January 29, Notes for Lecture 7
Stanford Unversty CS54: Computatonal Complexty Notes 7 Luca Trevsan January 9, 014 Notes for Lecture 7 1 Approxmate Countng wt an N oracle We complete te proof of te followng result: Teorem 1 For every
More informationWeek 5: Neural Networks
Week 5: Neural Networks Instructor: Sergey Levne Neural Networks Summary In the prevous lecture, we saw how we can construct neural networks by extendng logstc regresson. Neural networks consst of multple
More information3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X
Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number
More informationMin Cut, Fast Cut, Polynomial Identities
Randomzed Algorthms, Summer 016 Mn Cut, Fast Cut, Polynomal Identtes Instructor: Thomas Kesselhem and Kurt Mehlhorn 1 Mn Cuts n Graphs Lecture (5 pages) Throughout ths secton, G = (V, E) s a mult-graph.
More informationFor now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.
Neural Networks : Dervaton compled by Alvn Wan from Professor Jtendra Malk s lecture Ths type of computaton s called deep learnng and s the most popular method for many problems, such as computer vson
More informationVapnik-Chervonenkis theory
Vapnk-Chervonenks theory Rs Kondor June 13, 2008 For the purposes of ths lecture, we restrct ourselves to the bnary supervsed batch learnng settng. We assume that we have an nput space X, and an unknown
More informationDifference Equations
Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1
More informationPolynomials. 1 More properties of polynomials
Polynomals 1 More propertes of polynomals Recall that, for R a commutatve rng wth unty (as wth all rngs n ths course unless otherwse noted), we defne R[x] to be the set of expressons n =0 a x, where a
More informationSolutions of Boolean equations by orthonormal expansion
Solutons of Boolean equatons by orthonormal expanson Vrendra Sule Department of Electrcal Engneerng Indan Insttute of Technology Bombay, Powa Mumba 400076, Inda (vrs@ee.tb.ac.n) December 29, 2012 1 Introducton
More informationPulse Coded Modulation
Pulse Coded Modulaton PCM (Pulse Coded Modulaton) s a voce codng technque defned by the ITU-T G.711 standard and t s used n dgtal telephony to encode the voce sgnal. The frst step n the analog to dgtal
More informationErrors for Linear Systems
Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch
More informationAPPENDIX A Some Linear Algebra
APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,
More informationU.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016
U.C. Berkeley CS94: Spectral Methods and Expanders Handout 8 Luca Trevsan February 7, 06 Lecture 8: Spectral Algorthms Wrap-up In whch we talk about even more generalzatons of Cheeger s nequaltes, and
More informationLecture 20: Lift and Project, SDP Duality. Today we will study the Lift and Project method. Then we will prove the SDP duality theorem.
prnceton u. sp 02 cos 598B: algorthms and complexty Lecture 20: Lft and Project, SDP Dualty Lecturer: Sanjeev Arora Scrbe:Yury Makarychev Today we wll study the Lft and Project method. Then we wll prove
More informationTime-Varying Systems and Computations Lecture 6
Tme-Varyng Systems and Computatons Lecture 6 Klaus Depold 14. Januar 2014 The Kalman Flter The Kalman estmaton flter attempts to estmate the actual state of an unknown dscrete dynamcal system, gven nosy
More informationCanonical transformations
Canoncal transformatons November 23, 2014 Recall that we have defned a symplectc transformaton to be any lnear transformaton M A B leavng the symplectc form nvarant, Ω AB M A CM B DΩ CD Coordnate transformatons,
More informationTHE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens
THE CHINESE REMAINDER THEOREM KEITH CONRAD We should thank the Chnese for ther wonderful remander theorem. Glenn Stevens 1. Introducton The Chnese remander theorem says we can unquely solve any par of
More informationComputing Correlated Equilibria in Multi-Player Games
Computng Correlated Equlbra n Mult-Player Games Chrstos H. Papadmtrou Presented by Zhanxang Huang December 7th, 2005 1 The Author Dr. Chrstos H. Papadmtrou CS professor at UC Berkley (taught at Harvard,
More informationC/CS/Phy191 Problem Set 3 Solutions Out: Oct 1, 2008., where ( 00. ), so the overall state of the system is ) ( ( ( ( 00 ± 11 ), Φ ± = 1
C/CS/Phy9 Problem Set 3 Solutons Out: Oct, 8 Suppose you have two qubts n some arbtrary entangled state ψ You apply the teleportaton protocol to each of the qubts separately What s the resultng state obtaned
More informationNotes on Frequency Estimation in Data Streams
Notes on Frequency Estmaton n Data Streams In (one of) the data streamng model(s), the data s a sequence of arrvals a 1, a 2,..., a m of the form a j = (, v) where s the dentty of the tem and belongs to
More informationNumerical Heat and Mass Transfer
Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and
More informationOn a direct solver for linear least squares problems
ISSN 2066-6594 Ann. Acad. Rom. Sc. Ser. Math. Appl. Vol. 8, No. 2/2016 On a drect solver for lnear least squares problems Constantn Popa Abstract The Null Space (NS) algorthm s a drect solver for lnear
More informationLearning Theory: Lecture Notes
Learnng Theory: Lecture Notes Lecturer: Kamalka Chaudhur Scrbe: Qush Wang October 27, 2012 1 The Agnostc PAC Model Recall that one of the constrants of the PAC model s that the data dstrbuton has to be
More information2.3 Nilpotent endomorphisms
s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms
More informationLinear, affine, and convex sets and hulls In the sequel, unless otherwise specified, X will denote a real vector space.
Lnear, affne, and convex sets and hulls In the sequel, unless otherwse specfed, X wll denote a real vector space. Lnes and segments. Gven two ponts x, y X, we defne xy = {x + t(y x) : t R} = {(1 t)x +
More informationLinear Approximation with Regularization and Moving Least Squares
Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...
More information= z 20 z n. (k 20) + 4 z k = 4
Problem Set #7 solutons 7.2.. (a Fnd the coeffcent of z k n (z + z 5 + z 6 + z 7 + 5, k 20. We use the known seres expanson ( n+l ( z l l z n below: (z + z 5 + z 6 + z 7 + 5 (z 5 ( + z + z 2 + z + 5 5
More informationCollege of Computer & Information Science Fall 2009 Northeastern University 20 October 2009
College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:
More informationStanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011
Stanford Unversty CS359G: Graph Parttonng and Expanders Handout 4 Luca Trevsan January 3, 0 Lecture 4 In whch we prove the dffcult drecton of Cheeger s nequalty. As n the past lectures, consder an undrected
More informationUncertainty in measurements of power and energy on power networks
Uncertanty n measurements of power and energy on power networks E. Manov, N. Kolev Department of Measurement and Instrumentaton, Techncal Unversty Sofa, bul. Klment Ohrdsk No8, bl., 000 Sofa, Bulgara Tel./fax:
More informationTuring Machines (intro)
CHAPTER 3 The Church-Turng Thess Contents Turng Machnes defntons, examples, Turng-recognzable and Turng-decdable languages Varants of Turng Machne Multtape Turng machnes, non-determnstc Turng Machnes,
More informationDecision Diagrams Derivatives
Decson Dagrams Dervatves Logc Crcuts Desgn Semnars WS2010/2011, Lecture 3 Ing. Petr Fšer, Ph.D. Department of Dgtal Desgn Faculty of Informaton Technology Czech Techncal Unversty n Prague Evropský socální
More informationModule 9. Lecture 6. Duality in Assignment Problems
Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept
More informationBezier curves. Michael S. Floater. August 25, These notes provide an introduction to Bezier curves. i=0
Bezer curves Mchael S. Floater August 25, 211 These notes provde an ntroducton to Bezer curves. 1 Bernsten polynomals Recall that a real polynomal of a real varable x R, wth degree n, s a functon of the
More informationFormulas for the Determinant
page 224 224 CHAPTER 3 Determnants e t te t e 2t 38 A = e t 2te t e 2t e t te t 2e 2t 39 If 123 A = 345, 456 compute the matrx product A adj(a) What can you conclude about det(a)? For Problems 40 43, use
More informationfind (x): given element x, return the canonical element of the set containing x;
COS 43 Sprng, 009 Dsjont Set Unon Problem: Mantan a collecton of dsjont sets. Two operatons: fnd the set contanng a gven element; unte two sets nto one (destructvely). Approach: Canoncal element method:
More informationComplete subgraphs in multipartite graphs
Complete subgraphs n multpartte graphs FLORIAN PFENDER Unverstät Rostock, Insttut für Mathematk D-18057 Rostock, Germany Floran.Pfender@un-rostock.de Abstract Turán s Theorem states that every graph G
More information8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS
SECTION 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS 493 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS All the vector spaces you have studed thus far n the text are real vector spaces because the scalars
More informationIntroduction to Vapor/Liquid Equilibrium, part 2. Raoult s Law:
CE304, Sprng 2004 Lecture 4 Introducton to Vapor/Lqud Equlbrum, part 2 Raoult s Law: The smplest model that allows us do VLE calculatons s obtaned when we assume that the vapor phase s an deal gas, and
More informationYong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )
Kangweon-Kyungk Math. Jour. 4 1996), No. 1, pp. 7 16 AN ITERATIVE ROW-ACTION METHOD FOR MULTICOMMODITY TRANSPORTATION PROBLEMS Yong Joon Ryang Abstract. The optmzaton problems wth quadratc constrants often
More informationExample: (13320, 22140) =? Solution #1: The divisors of are 1, 2, 3, 4, 5, 6, 9, 10, 12, 15, 18, 20, 27, 30, 36, 41,
The greatest common dvsor of two ntegers a and b (not both zero) s the largest nteger whch s a common factor of both a and b. We denote ths number by gcd(a, b), or smply (a, b) when there s no confuson
More informationPerron Vectors of an Irreducible Nonnegative Interval Matrix
Perron Vectors of an Irreducble Nonnegatve Interval Matrx Jr Rohn August 4 2005 Abstract As s well known an rreducble nonnegatve matrx possesses a unquely determned Perron vector. As the man result of
More informationMA 323 Geometric Modelling Course Notes: Day 13 Bezier Curves & Bernstein Polynomials
MA 323 Geometrc Modellng Course Notes: Day 13 Bezer Curves & Bernsten Polynomals Davd L. Fnn Over the past few days, we have looked at de Casteljau s algorthm for generatng a polynomal curve, and we have
More informationNUMERICAL DIFFERENTIATION
NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the
More informationTOPICS MULTIPLIERLESS FILTER DESIGN ELEMENTARY SCHOOL ALGORITHM MULTIPLICATION
1 2 MULTIPLIERLESS FILTER DESIGN Realzaton of flters wthout full-fledged multplers Some sldes based on support materal by W. Wolf for hs book Modern VLSI Desgn, 3 rd edton. Partly based on followng papers:
More informationStructure and Drive Paul A. Jensen Copyright July 20, 2003
Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.
More informationVolume 18 Figure 1. Notation 1. Notation 2. Observation 1. Remark 1. Remark 2. Remark 3. Remark 4. Remark 5. Remark 6. Theorem A [2]. Theorem B [2].
Bulletn of Mathematcal Scences and Applcatons Submtted: 016-04-07 ISSN: 78-9634, Vol. 18, pp 1-10 Revsed: 016-09-08 do:10.1805/www.scpress.com/bmsa.18.1 Accepted: 016-10-13 017 ScPress Ltd., Swtzerland
More informationAdditional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty
Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,
More informationEcon107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)
I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes
More informationLecture 12: Discrete Laplacian
Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly
More information4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA
4 Analyss of Varance (ANOVA) 5 ANOVA 51 Introducton ANOVA ANOVA s a way to estmate and test the means of multple populatons We wll start wth one-way ANOVA If the populatons ncluded n the study are selected
More informationBasic Regular Expressions. Introduction. Introduction to Computability. Theory. Motivation. Lecture4: Regular Expressions
Introducton to Computablty Theory Lecture: egular Expressons Prof Amos Israel Motvaton If one wants to descrbe a regular language, La, she can use the a DFA, Dor an NFA N, such L ( D = La that that Ths
More informationLecture 4: Universal Hash Functions/Streaming Cont d
CSE 5: Desgn and Analyss of Algorthms I Sprng 06 Lecture 4: Unversal Hash Functons/Streamng Cont d Lecturer: Shayan Oves Gharan Aprl 6th Scrbe: Jacob Schreber Dsclamer: These notes have not been subjected
More informationCOMPLEX NUMBERS AND QUADRATIC EQUATIONS
COMPLEX NUMBERS AND QUADRATIC EQUATIONS INTRODUCTION We know that x 0 for all x R e the square of a real number (whether postve, negatve or ero) s non-negatve Hence the equatons x, x, x + 7 0 etc are not
More information10-701/ Machine Learning, Fall 2005 Homework 3
10-701/15-781 Machne Learnng, Fall 2005 Homework 3 Out: 10/20/05 Due: begnnng of the class 11/01/05 Instructons Contact questons-10701@autonlaborg for queston Problem 1 Regresson and Cross-valdaton [40
More informationCHAPTER III Neural Networks as Associative Memory
CHAPTER III Neural Networs as Assocatve Memory Introducton One of the prmary functons of the bran s assocatve memory. We assocate the faces wth names, letters wth sounds, or we can recognze the people
More informationLecture 3. Ax x i a i. i i
18.409 The Behavor of Algorthms n Practce 2/14/2 Lecturer: Dan Spelman Lecture 3 Scrbe: Arvnd Sankar 1 Largest sngular value In order to bound the condton number, we need an upper bound on the largest
More informationLecture 10 Support Vector Machines II
Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed
More informationThe Expectation-Maximization Algorithm
The Expectaton-Maxmaton Algorthm Charles Elan elan@cs.ucsd.edu November 16, 2007 Ths chapter explans the EM algorthm at multple levels of generalty. Secton 1 gves the standard hgh-level verson of the algorthm.
More informationEEE 241: Linear Systems
EEE : Lnear Systems Summary #: Backpropagaton BACKPROPAGATION The perceptron rule as well as the Wdrow Hoff learnng were desgned to tran sngle layer networks. They suffer from the same dsadvantage: they
More informationLecture 7: Gluing prevarieties; products
Lecture 7: Glung prevaretes; products 1 The category of algebrac prevaretes Proposton 1. Let (f,ϕ) : (X,O X ) (Y,O Y ) be a morphsm of algebrac prevaretes. If U X and V Y are affne open subvaretes wth
More informationSpeeding up Computation of Scalar Multiplication in Elliptic Curve Cryptosystem
H.K. Pathak et. al. / (IJCSE) Internatonal Journal on Computer Scence and Engneerng Speedng up Computaton of Scalar Multplcaton n Ellptc Curve Cryptosystem H. K. Pathak Manju Sangh S.o.S n Computer scence
More informationA CLASS OF RECURSIVE SETS. Florentin Smarandache University of New Mexico 200 College Road Gallup, NM 87301, USA
A CLASS OF RECURSIVE SETS Florentn Smarandache Unversty of New Mexco 200 College Road Gallup, NM 87301, USA E-mal: smarand@unmedu In ths artcle one bulds a class of recursve sets, one establshes propertes
More information12 MATH 101A: ALGEBRA I, PART C: MULTILINEAR ALGEBRA. 4. Tensor product
12 MATH 101A: ALGEBRA I, PART C: MULTILINEAR ALGEBRA Here s an outlne of what I dd: (1) categorcal defnton (2) constructon (3) lst of basc propertes (4) dstrbutve property (5) rght exactness (6) localzaton
More informationGeneralized Linear Methods
Generalzed Lnear Methods 1 Introducton In the Ensemble Methods the general dea s that usng a combnaton of several weak learner one could make a better learner. More formally, assume that we have a set
More informationprinceton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg
prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there
More informationChapter Newton s Method
Chapter 9. Newton s Method After readng ths chapter, you should be able to:. Understand how Newton s method s dfferent from the Golden Secton Search method. Understand how Newton s method works 3. Solve
More informationGeneral theory of fuzzy connectedness segmentations: reconciliation of two tracks of FC theory
General theory of fuzzy connectedness segmentatons: reconclaton of two tracks of FC theory Krzysztof Chrs Ceselsk Department of Mathematcs, West Vrgna Unversty and MIPG, Department of Radology, Unversty
More informationGraph Reconstruction by Permutations
Graph Reconstructon by Permutatons Perre Ille and Wllam Kocay* Insttut de Mathémathques de Lumny CNRS UMR 6206 163 avenue de Lumny, Case 907 13288 Marselle Cedex 9, France e-mal: lle@ml.unv-mrs.fr Computer
More informationThe Geometry of Logit and Probit
The Geometry of Logt and Probt Ths short note s meant as a supplement to Chapters and 3 of Spatal Models of Parlamentary Votng and the notaton and reference to fgures n the text below s to those two chapters.
More informationSimultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals
Smultaneous Optmzaton of Berth Allocaton, Quay Crane Assgnment and Quay Crane Schedulng Problems n Contaner Termnals Necat Aras, Yavuz Türkoğulları, Z. Caner Taşkın, Kuban Altınel Abstract In ths work,
More informationBOUNDEDNESS OF THE RIESZ TRANSFORM WITH MATRIX A 2 WEIGHTS
BOUNDEDNESS OF THE IESZ TANSFOM WITH MATIX A WEIGHTS Introducton Let L = L ( n, be the functon space wth norm (ˆ f L = f(x C dx d < For a d d matrx valued functon W : wth W (x postve sem-defnte for all
More informationOn the Multicriteria Integer Network Flow Problem
BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 5, No 2 Sofa 2005 On the Multcrtera Integer Network Flow Problem Vassl Vasslev, Marana Nkolova, Maryana Vassleva Insttute of
More informationVQ widely used in coding speech, image, and video
at Scalar quantzers are specal cases of vector quantzers (VQ): they are constraned to look at one sample at a tme (memoryless) VQ does not have such constrant better RD perfomance expected Source codng
More informationMATH 241B FUNCTIONAL ANALYSIS - NOTES EXAMPLES OF C ALGEBRAS
MATH 241B FUNCTIONAL ANALYSIS - NOTES EXAMPLES OF C ALGEBRAS These are nformal notes whch cover some of the materal whch s not n the course book. The man purpose s to gve a number of nontrval examples
More informationWeek 2. This week, we covered operations on sets and cardinality.
Week 2 Ths week, we covered operatons on sets and cardnalty. Defnton 0.1 (Correspondence). A correspondence between two sets A and B s a set S contaned n A B = {(a, b) a A, b B}. A correspondence from
More information20. Mon, Oct. 13 What we have done so far corresponds roughly to Chapters 2 & 3 of Lee. Now we turn to Chapter 4. The first idea is connectedness.
20. Mon, Oct. 13 What we have done so far corresponds roughly to Chapters 2 & 3 of Lee. Now we turn to Chapter 4. The frst dea s connectedness. Essentally, we want to say that a space cannot be decomposed
More informationCompilers. Spring term. Alfonso Ortega: Enrique Alfonseca: Chapter 4: Syntactic analysis
Complers Sprng term Alfonso Ortega: alfonso.ortega@uam.es nrque Alfonseca: enrque.alfonseca@uam.es Chapter : Syntactc analyss. Introducton. Bottom-up Analyss Syntax Analyser Concepts It analyses the context-ndependent
More informationn α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0
MODULE 2 Topcs: Lnear ndependence, bass and dmenson We have seen that f n a set of vectors one vector s a lnear combnaton of the remanng vectors n the set then the span of the set s unchanged f that vector
More information42. Mon, Dec. 8 Last time, we were discussing CW complexes, and we considered two di erent CW structures on S n. We continue with more examples.
42. Mon, Dec. 8 Last tme, we were dscussng CW complexes, and we consdered two d erent CW structures on S n. We contnue wth more examples. (2) RP n. Let s start wth RP 2. Recall that one model for ths space
More informationLecture 5 Decoding Binary BCH Codes
Lecture 5 Decodng Bnary BCH Codes In ths class, we wll ntroduce dfferent methods for decodng BCH codes 51 Decodng the [15, 7, 5] 2 -BCH Code Consder the [15, 7, 5] 2 -code C we ntroduced n the last lecture
More informationA new construction of 3-separable matrices via an improved decoding of Macula s construction
Dscrete Optmzaton 5 008 700 704 Contents lsts avalable at ScenceDrect Dscrete Optmzaton journal homepage: wwwelsevercom/locate/dsopt A new constructon of 3-separable matrces va an mproved decodng of Macula
More informationLecture 3: Probability Distributions
Lecture 3: Probablty Dstrbutons Random Varables Let us begn by defnng a sample space as a set of outcomes from an experment. We denote ths by S. A random varable s a functon whch maps outcomes nto the
More informationBayesian Networks. Course: CS40022 Instructor: Dr. Pallab Dasgupta
Bayesan Networks Course: CS40022 Instructor: Dr. Pallab Dasgupta Department of Computer Scence & Engneerng Indan Insttute of Technology Kharagpur Example Burglar alarm at home Farly relable at detectng
More informationLecture Notes on Linear Regression
Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume
More informationEdge Isoperimetric Inequalities
November 7, 2005 Ross M. Rchardson Edge Isopermetrc Inequaltes 1 Four Questons Recall that n the last lecture we looked at the problem of sopermetrc nequaltes n the hypercube, Q n. Our noton of boundary
More information