Matrix Rigidity of Random Toeplitz Matrices

Size: px
Start display at page:

Download "Matrix Rigidity of Random Toeplitz Matrices"

Transcription

1 Matrix Rigidity of Radom Toeplitz Matrices Oded Goldreich Avishay Tal May 7, 2015 Abstract We prove that radom -by- Toeplitz (alteratively Hakel) matrices over F 2 have rigidity Ω( 3 r 2 log ) for rak r, with high probability. This improves, for r = o(/ log log log ), over the Ω( 2 r log( r )) boud that is kow for may explicit matrices. Our result implies that the explicit triliear [] [] [2] fuctio defied by F (x, y, z) = i,j x iy j z i+j has complexity Ω( 3/5 ) i the multiliear circuit model suggested by Goldreich ad Wigderso (ECCC, 2013), which yields a exp( 3/5 ) lower boud o the size of the socalled caoical depth-three circuits for F. We also prove that F has complexity Ω( 2/3 ) if the multiliear circuits are further restricted to be of depth 2. I additio, we show that a matrix whose etries are sampled from a 2 -biased distributio has complexity Ω( 2/3 ), regardless of depth restrictios, almost matchig the O( 2/3 ) upper boud for ay matrix by Goldreich ad Wigderso. We tur this radomized costructio ito a explicit 4-liear costructio with similar lower bouds, usig the quadratic small-biased costructio of Mossel et al. (RS&A, 2006). Keywords: Matrix rigidity, multi-liear fuctios, multi-liear circuits. Weizma Istitute of Sciece, Rehovot, Israel. oded.goldreich@weizma.ac.il. Partially supported by the Mierva Foudatio with fuds from the Federal Germa Miistry for Educatio ad Research. Weizma Istitute of Sciece, Rehovot, Israel. avishay.tal@weizma.ac.il. Supported by a Adams Fellowship of the Israel Academy of Scieces ad Humaities, by a ISF grat ad by the I-CORE Program of the Plaig ad Budgetig Committee.

2 Cotets 1 Itroductio Matrix Rigidity The Project of Goldreich-Wigderso Resolvig the Foregoig Ope Problems Overview of the Proof of Theorem Orgaizatio Prelimiaries 5 3 Mai Results 6 4 The Structure of Matrices of Small Biliear Circuits The Structure of Matrices Associated with Depth Two Biliear Circuits The Structure of Matrices Associated with Geeral Biliear Circuits Substructures Testig AN Complexity ad AN2 Complexity Lower Bouds for the AN-Complexity of Small-Biased Matrices Lower Bouds for the AN2-Complexity of Radom Toeplitz Matrices Explicit 4-Liear Fuctios with AN-Complexity Ω( 2/3 ) Digest ad Ope Problems Digest Ope Problems Refereces 21 Appedices 22 A.1 Geeralizatio to Larger Fields A.2 The Structure of Matrices Associated with Geeral Biliear Circuits

3 1 Itroductio This paper cocers the costructio of rigid matrices, a cetral ope problem posed by Valiat [Val77], ad its applicatio to lower bouds o caoical depth-three Boolea circuits (where caoical is as defied by Goldreich ad Wigderso [GW13]). I particular, we improve the kow lower boud o matrix rigidity, but the improvemet is for a rage of parameters that is ot the oe motivated by Valiat s problem, but rather the oe that arises from [GW13]. Ideed, this improvemet resolves ope problems posed by Goldreich ad Wigderso [GW13]. 1.1 Matrix Rigidity The Matrix Rigidity Problem (i.e., providig explicit matrices of high rigidity) is oe of the most allurig problems i arithmetic circuits lower bouds. Itroduced i 1977 by Valiat [Val77], the problem was origially motivated by provig lower bouds for the computatio of liear trasformatios. Loosely speakig, a matrix is called rigid if it caot be writte as a sum of a low rak matrix ad a sparse matrix. Needless to say, the actual defiitio specifies both parameters. Defiitio 1.1 (Matrix Rigidity, [Val77]). A matrix A over a field F has rigidity s for rak r if every matrix of rak at most r (over F) disagrees with A o more tha s etries. Valiat showed that ay matrix with rigidity 1+δ for rak ω(/ log log ), where δ is some costat greater tha 0, caot be computed by a liear circuit of size O() ad depth O(log ). Valiat also proved that almost all -by- matrices, over a fiite field F (e.g., the two-elemet field F 2 ), have rigidity Ω(( r) 2 / log ) for rak r. Sice the, comig up with a explicit 1 rigid matrix has remaied a challege. The best techiques to date provide explicit -by- matrices of rigidity 2 r log( r ) for rak r (see [Lok09] for a survey about matrix rigidity). To the best of our kowledge, this state of affairs also holds for simple radomized costructios that use O() radom bits. The commo belief is that rigidity bouds for such radomized costructios ca be used for provig lower bouds for explicit computatioal problems that are related to the origial oes. For example, a adequate rigidity lower boud for radom Toeplitz (or Hakel) matrices 2 would yield a lower boud o the complexity of computig explicit biliear trasformatios. Ideed, this is aalogous to Adreev s proof of formula lower bouds [Ad87], where a lower boud for a radomized fuctio is trasformed ito a lower boud for a explicit fuctio (which takes the radom bits of the costructio as part of its iput). Our mai result is the followig Theorem 1.2 (o the rigidity of radom Toeplitz/Hakel matrices). Let A F2 be a radom Toeplitz/Hakel matrix. The, for every r [, /32], with probability 1 o(1), the matrix A has rigidity Ω( 3 ) for rak r. r 2 log Our bouds are asymptotically better tha Ω( 2 r log( r )) for rak r = o( log log log ), alas Valiat s origial motivatio refers to r > / log log. For rak r = 0.5+ε, where ε (0, 0.5), our boud yields a sigificat improvemet (i.e., 3 = 2 2ε 1.5 ε = 2 r 2 r ), ad this is actually the rage that is relevat for the project of Goldreich ad Wigderso [GW13]. 1 For a ifiite I N, the sequece of matrices, {A } I such that A is a matrix, is called explicit if there exists a poly()-time algorithm that o iput I outputs the matrix A (ad outputs if I). 2 Recall that a Toeplitz matrix T = (T i,j) has costat diagoals (i.e., T i,j = T i+1,j+1 for every i, j). Hakel matrices are obtaied by turig Toeplitz matrices upside dow; that is, a Hakel matrix H = (H i,j) has costat skew-diagoals (i.e., H i,j = H i+1,j 1 for every i, j). Hece, ay claim regardig oe family traslates to a equivalet claim regardig the other family. 1

4 1.2 The Project of Goldreich-Wigderso The project started by Goldreich ad Wigderso [GW13] provides aother motivatio for the study of matrix rigidity. I fact, the problem of improvig the rigidity bouds for radom Toeplitz matrices was posed explicitly there. Specifically, provig a rigidity boud of 1.5+Ω(1) for rak 0.5+Ω(1) for radom Toeplitz matrices was proposed there as a possible ext step. Lower Bouds for Depth Three Caoical Circuits. Håstad [Hås86] showed that ay depththree Boolea circuit computig the -way parity fuctio must be of size at least exp( ); to date, Håstad s result is the best lower boud for a explicit fuctio i the model of depth-three Boolea circuits 3. The work of Goldreich ad Wigderso [GW13] put forward a model of depth three caoical circuits, with the uderlyig log-term goal to exhibit better lower bouds for geeral depth-three Boolea circuits. Caoical circuits are restricted type of such depth-three circuits, which ca be illustrated by cosiderig the smallest kow depth-three circuits for -way parity. The latter Õ(2 )-size circuits are obtaied by combiig a CNF that computes a -way parity with DNFs that compute -way parities of disjoit blocks of the iput bits. The costructio, ad its optimality by [Hås86], suggests the followig scheme for obtaiig Boolea circuits that compute multiliear fuctios. First, costruct a arithmetic circuit that uses arbitrary multiliear gates of parameterized arity, ad the covert it to a Boolea circuit whose size is expoetial i the maximum betwee the arity ad the umber of gates i the arithmetic circuit. The arithmetic model is outlied ext. Lower Bouds for Multiliear Circuits. Suppose we wish to compute a t-liear fuctio that depeds o t blocks of iputs, x (1),..., x (t), each of legth ; that is, the fuctio is liear i each of the x (j) s. We cosider circuits that use arbitrary multiliear gates of parameterized arity. That is, the circuits are directed acyclic graphs, where each iteral ode computes a multiliear fuctio of its iputs. We further restrict our circuit such that each iteral gate computes a multiliear formal polyomial i the iputs x (1)..., x (t). We say that such a multiliear circuit is of AN-complexity 4 m if m equals the maximum betwee the umber of the circuit gates ad the maximal arity of the gates. For a multiliear fuctio F, we deote by C(F ) the miimal AN-complexity of a multiliear circuit which compute the fuctio F. (We will abuse otatio ad refer to the AN-complexity of a tesor/matrix as the AN-complexity of the correspodig multiliear fuctio.) I the example of parity, we have a bottom layer of gates each takig iputs ad computig their parity. Above these gates, we have a gate which takes the results ad computes their parity. Overall, we got a (multi)-liear circuit of AN-complexity + 1. Goldreich ad Wigderso showed that ay multiliear circuit of AN-complexity m yields a depth-three Boolea circuit of size exp(m) computig the same fuctio (see [GW13, Prop. 2.9]). I fact, the Boolea circuits have much more structure, ad are referred to by Goldreich ad Wigderso as caoical circuits. Thus, a prelimiary step towards beatig the exp(ω( )) lower boud o the size of depth-three Boolea circuits for explicit O(1)-liear fuctios, 5 will be to beat the Ω( ) AN-complexity lower boud for such fuctios i the model of multiliear circuits. Agai, as i Valiat s questio, if we just ask about the existece of hard t-liear fuctios, the most t-liear fuctios caot be computed by a multiliear circuit of AN-complexity smaller tha (t) t/(t+1) : See [GW13, Thm. 4.1], which uses a coutig argumet. The more importat ad 3 That is, circuits of ubouded fa-i OR ad AND gates with leaves that are variables or their egatios. 4 where AN stads for Arity ad Number of gates. 5 Ideed, this suggestio presumes that there exist O(1)-liear fuctios that require depth-three Boolea circuits of size exp(ω( )), which is also a ope problem suggested i [GW13]. 2

5 challegig problem is to came up with a explicit t-liear fuctio for which such bouds, or eve just ω( ) lower bouds, ca be proved. Reductio to (Structured) Rigidity. Goldreich ad Wigderso reduces the problem of provig lower bouds for biliear circuits to the problem of rigidity [GW13, Sec. 4.2]. They show that if a biliear circuit is of AN-complexity m/2, the its correspodig matrix is ot m 3 rigid for rak m (i.e., it ca be expressed as a sum of a m 3 -sparse matrix ad a matrix of rak at most m). Hece, ay matrix that has rigidity m 3 for rak m correspods to a biliear fuctio that caot be computed by a biliear circuit of AN-complexity at most m/2. Furthermore, Goldreich ad Wigderso show that the sparse matrix arisig from their reductio has a additioal structure (to be specified later). This leads to a weaker otio of rigidity (see [GW13, Thm. 4.12] which establishes a separatio), called structured rigidity, for which it is potetially easier to prove lower bouds. Ope Problems i Goldreich-Wigderso. Oe ope problem posed by Goldreich ad Wigderso is provig that radom Toeplitz matrices have rigidity m 3 (or just structured rigidity m 3 ) for rak m = 0.5+Ω(1). This would yield a AN-complexity lower boud of m for the correspodig biliear fuctio (via the reductio i [GW13, Thm. 4.4]) 6 as well as a similar lower boud for the followig explicit triliear fuctio (via [GW13, Prop. 4.6]): F tet (x, y, z) = i 1,i 2,i 3 []: 3 j=1 i j /2 /2 x i1 y i2 z i3. (1) 1.3 Resolvig the Foregoig Ope Problems We resolve the aforemetioed ope problem [GW13, Prob. 4.8] by provig that radom Toeplitz matrices have rigidity m 3 for rak m = Θ( 3/5 ), with high probability. This follows from our log 1/5 mai theorem (Theorem 1.2) by choosig r = m. Furthermore, we ca get rid of the logarithmic factor i the Ω otatio, by provig a slightly better lower boud for structured rigidity. Theorem 1.3 (o the structured rigidity of radom Toeplitz/Hakel matrices). Let A F 2 be a radom Toeplitz/Hakel matrix. The, for every r [, /32], the matrix A has structured rigidity Ω( 3 /r 2 ) for rak r. This implies (usig [GW13, Thm. 4.10] ad [GW13, Prop. 4.6]) that the AN-complexity of a radom Toeplitz matrix is Ω( 3/5 ), ad ditto for the explicit triliear fuctio F tet from Eq. (1). This resolves Problems 4.7 ad 4.2 i [GW13], resp. I additio, we show that aother explicit triliear fuctio has AN-complexity Ω( 3/5 ). Corollary 1.4 (AN-complexity lower boud for a explicit triliear fuctio). Let F : {0, 1} {0, 1} {0, 1} 2 {0, 1} be the triliear fuctio defied by F (x, y, z) = j=1 x iy j z i+j. The, C(F ) = Ω( 3/5 ). New Challeges. The most atural questio that arises from the foregoig results is to tighte the lower boud; that is, to show that radom Toeplitz matrices have AN-complexity Ω( 2/3 ) as cojectured by [GW13]. This would be the best possible, sice ay biliear fuctio ca be computed by a biliear circuit of AN-complexity O( 2/3 ); more geerally, by [GW13, Thm. 3.1], 6 For structured rigidity, we use [GW13, Thm. 4.10]. 3

6 for ay t 2, ay t-liear fuctio ca be computed by a t-liear circuit of AN-complexity O((t) t/(t+1) ). Aother atural follow up questio is to exhibit a explicit O(1)-liear fuctio havig AN-complexity Ω( α ) for some costat α > 3/5; of course, the larger α, the better. Our progress o these ope problems is captured by the followig two results. Theorem 1.5 (depth-two AN-complexity lower boud for radom Toeplitz matrices). Let F be a biliear fuctio that correspods to a radom Toeplitz matrix. The, with probability 1 o(1), the fuctio F caot be computed by multiliear circuits of depth two havig AN-complexity 2/3 /(log ) 1/3. Theorem 1.5 establishes the desired AN-complexity lower boud for radom Toeplitz matrices, but oly for depth-two multiliear circuits. We ote that the AN-complexity upper boud of [GW13, Thm. 3.1] holds via depth-two circuits, ad so Theorem 1.5 is almost optimal with respect to depth-two multiliear circuits. Theorem 1.5 implies that the triliear fuctio F (x, y, z) = j=1 x iy j z i+j caot be computed by multiliear circuits of depth two ad AN-complexity 2/3 /(log ) 1/3. Theorem 1.6 (improved AN-complexity lower boud for 4-liear fuctios). There exists a explicit 4-liear fuctio havig AN-complexity Ω( 2/3 /(log ) 1/3 ). Theorem 1.6 is proved by first showig that, with high probability, biliear fuctios associated with matrices that are sampled from a 2 -biased sample space (over {0, 1} 2 ) have AN-complexity Ω( 2/3 ). Note that by the aforemetioed upper boud, this lower boud is tight (up to logarithmic factors). Next, we ote that samplig such matrices ca be doe usig O() radom bits [NN93, AGHP92, MST06], which matches the amout of radomess used for samplig a radom Toeplitz matrix. Furthermore, i the explicit small-biased costructio of Mossel et al. [MST06], each bit i the sampled strig is a biliear fuctio of the radom bits, allowig us to give a explicit 4-liear fuctio with AN-complexity Ω( 2/3 ). 1.4 Overview of the Proof of Theorem 1.2 We give a overview of the proof of Theorem 1.2 (for the case of Hakel matrices). Recall that we wish to show that a radom Hakel matrix has rigidity 3 /r 2 log for rak r, with high probability. Let A be a radom -by- Hakel matrix, of the form A i,j = a i+j for idepedet radom bits a 2,..., a 2. Suppose that A ca be expressed as a sum of a s-sparse matrix S ad a matrix of rak at most r. Cosider a partitio of A ad S ito (/2r) (/2r) submatrices, each of size 2r 2r, such that a geeric submatrix cosists of 2r cosecutive colums ad 2r equally spaced rows (i.e., rows that are at distace /2r apart). The, there exists a pair of correspodig submatrices A ad S such that S is of sparsity s = (2r)2 s (ad, of course, A S has rak at most r). Next, 2 we ote that ay of the above submatrices of A are of the form b 1 b 2 b 3... b 2r A = b k+1 b k+2 b k+3... b k+2r b (2r 1)k+1 b (2r 1)k+2 b (2r 1)k+3... b (2r 1)k+2r where k = /2r ( 2r, by the assumptio r ), ad b 1,..., b (2r 1)k+2r is a cosecutive subsequece of a 2,..., a 2. Notice that A is a 2r 2r submatrix that depeds o (2r 1)k + 2r = Θ() radom bits. 4

7 I our mai lemma, we show that for ay fixed matrix S (eve if S is ot sparse) the submatrix matrix A S is of rak greater tha r with probability at least 1 2 Ω(), where the probability is take over the choice of A (equiv., over the choice of b 1,..., b (2r 1)k+2r ). Hece, takig a uio boud over all possible s -sparse submatrices we get that, with probability at least 1 ( 2r 2r s ) 2 Ω(), the submatrix A has rigidity s for rak r. Pickig s = o( 3 r 2 log ) implies that s = o(/ log ), which completes the proof (by applyig a uio boud o all submatrices, ad iferrig that, with high probability, matrix A is of rigidity s for rak r). 1.5 Orgaizatio Our mai results (i.e., Theorems 1.2 ad 1.3 ad Corollary 1.4) are proved i Sectio 3, which follows a short prelimiary sectio (Sec. 2). Next, Theorems 1.5 ad 1.6 are proved, i two steps. I Sectio 4 we idetify structural properties of matrices that correspod to biliear fuctios of low AN (ad AN2) complexity. These properties correspod to (eve more) restricted otios of structured rigidity, ad i Sectio 5 we show that (with high probability) matrices draw from the two relevat distributios do ot satisfy these properties. We coclude, with a techical digest (Sectio 6.1) ad a list of some ope problems (Sectio 6.2). 2 Prelimiaries We deote by [] = {1,..., }. For, k N, we deote by ( k) = k ( i=0 i), ad use the boud ( ) k (2) k. For a matrix A, we deote its i-th row by A i, ad its j-th colum by A (j). We deote by wt(a) the umber of o-zero etries i the matrix A, ad say that A is s-sparse if wt(a) = s. A Hakel matrix over a field F is a square matrix with costat skew-diagoals; that is, ay matrix A F of the form A i,j = a i+j for some a 2,..., a 2 F. A Toeplitz matrix over a field F is a square matrix with costat diagoals, i.e. ay matrix A F of the form A i,j = a i j for some a ( 1),..., a 1 F. Note that a Hakel matrix is a upside-dow Toeplitz matrix. Throughout the paper, uless specified otherwise, we talk about matrices over the field F 2, ad matrix rak refers to its rak over F 2. Defiitio 2.1 (Structured Rigidity, [GW13, Def. 4.9]). We say that a matrix A has structured rigidity (m 1, m 2, m 3 ) for rak r if for every matrix R of rak at most r ad for every X 1,... X m1, Y 1,..., Y m1 [] such that X 1 = = X m1 = m 2 ad Y 1 = = Y m1 = m 3 it holds that A R m 1 k=1 (X k Y k ), where M S meas that all o-zero etries of the matrix M reside i the set S [] []. We say that a matrix A has structured rigidity m 3 for rak r if A has structured rigidity (m, m, m) for rak r. Ideed, ay matrix that has rigidity s for rak r, also has structured rigidity s for rak r, but the other directio does ot hold (see [GW13, Thm. 4.12]). Defiitio 2.2. A multiliear circuit o t blocks of iputs x (1),..., x (t) {0, 1} is a directed acyclic graph whose odes are associated with arbitrary multiliear gates such that ay two gates with directed paths from the same block of iputs are ot multiplied together by aother gate. Defiitio 2.3 (the AN-complexity of multiliear circuits with geeral gates, [GW13, Def. 2.2]). The arity of a multiliear circuit is the maximum arity of its (geeral) gates. The AN-complexity of a multiliear circuit is the maximum betwee its arity ad its umber of gates (where we cout oly the geeral gates ad ot the leaves, i.e., variables). The AN-complexity of a multiliear fuctio F, 5

8 deoted C(F ), is the miimum AN-complexity of a multiliear circuit that computes F. The AN2- complexity of a multiliear fuctio F, deoted C 2 (F ), is the miimum complexity of a depth-two multiliear circuit that computes F. Theorem 2.4 ([GW13, Thm. 4.10]). If A is a -by- matrix that has structured rigidity m 3 for rak m, the the correspodig biliear fuctio F satisfies C(F ) m/2. 3 Mai Results We prove our results bottom-up, startig with the mai lemma, as metioed i the proof overview. Lemma 3.1 (Mai Lemma). Let m, k N, 16 k m. Let A F m m 2 be the radom matrix a 1 a 2 a 3... a m a k+1 a k+2 a k+3... a k+m a (m 1)k+1 a (m 1)k+2 a (m 1)k+3... a (m 1)k+m where a 1,..., a (m 1)k+m are uiform idepedet radom bits, ad let S F m m 2 be some fixed matrix. The, Pr A [rak(s + A) m/2] 2 km/16. Note that for k = 1 the matrix i Lemma 3.1 is a radom Hakel matrix, ad for k = m it is a totally radom matrix. The requiremet k 16 is ot essetial i the lemma; it is used to make expressios icer. For k 1 ad rak r m/2 the proof gives Pr A [rak(s+a) r] ( m r) 2 mk/8. Proof. For a fixed S ad a radom A as above, let B = S + A. If r = rak(b) m/2, the oe ca costruct a basis B i1, B i2,..., B ir of the row space of B by the followig iterative process: Let i 1 be the first ozero row of B, let i 2 > i 1 be the first row i B that is ot spaed by row i 1, let i 3 > i 2 be the first row i B that is ot spaed by rows i 1 ad i 2, etc. We get that i 1 < i 2 < < i r ad 1. For j < i 1 the j-th row of B is the all zeroes row. 2. For i t 1 < j < i t the j-th row of B is spaed by rows i 1,..., i t 1 of B. 3. For i r < j the j-th row of B is spaed by rows i 1,..., i r of B. More cocisely, deotig by I = {i 1,..., i r }, we get j [m] \ I : B j spa{b i : i I, i < j}. (2) We boud the probability that such a sequece I = {i 1,..., i r } exists, where r m/2. We will uio boud over all possible sequeces I, ad for ay fixed sequece of legth at most m/2, we shall show that (2) holds with very low probability. Give such a sequece I, let J = [m] I be its complemet. Settig = m/k, we ca select a icreasig sequece of J / idices i J such that each two idices differ by at least. 7 Take j 1 < j 2 < < j t to be such a sequece of idices, where t J m/2 m/k k 4. For l [t], let E l be the evet that row j l is spaed by the rows idexed by I [j l 1]. The, Pr [Eq. (2) holds for I] Pr[E 1, E 2,..., E t ] = Pr[E 1 ] Pr[E 2 E 1 ] Pr[E t E 1,..., E t 1 ] (3) 7 Oe ca costruct such a set greedily: choose the miimal idex j i J, remove all idices i J [j, j + 1]. Repeat util J is empty. 6

9 Next, we show that for each l [t], we have Pr[E l E 1,..., E l 1 ] 2 m/2. However, istead of coditioig o E 1,..., E l 1, we shall coditio o a set of the radom bits, to be specified ext, that determie rows B 1,..., B jl 1 o oe had, but are idepedet from the radom row B jl o the other had. Sice j l j l 1 + m/k by our desig, we get (j l 1)k (j l 1 1)k + m. Hece, the radom bits a 1,..., a (jl 1)k, which determie B 1,..., B jl 1, leave the radom row B jl = (a (jl 1)k+1,..., a (jl 1)k+m) totally udetermied. Coditioig o the worst-case assigmet for the former radom variables (uder which E 1,..., E l 1 holds) yields a upper boud o Pr[E l E 1,..., E l 1 ]. Thus, it is eough to show that Pr[E l a 1,..., a (jl 1)k] 2 m/2 for ay possible fixed choice of values to a 1,..., a (jl 1)k. To avoid multiple subscripts, we set for the rest of the proof j j l. Let us remark that after fixig a 1,..., a (j 1)k, rows 1,..., j m/k are completely fixed, rows j m/k + 1,..., j 1 are partially fixed, ad row j is etirely udetermied. Based o that, we shall show that Pr[E l a 1,..., a (j 1)k ] 2 m/2. (4) Let I := I [j 1], ad fix a liear combiatio of the rows idexed by I, i.e., i I c ib i, amog all 2 I such liear combiatios. We show that the probability that B j = i I c i B i. (5) is 2 m. (This is similar, up to mior differeces, to the folklore result that ay fixed liear combiatio of rows i a radom Toeplitz matrix is distributed uiformly over F m 2 see [Gol08, Prop. 8.25]. We give the details for completeess.) The probability that the first bit of B j equals the first bit of the liear combiatio i (5) is exactly 1/2, sice B j,1 = S j,1 + a (j 1)k+1, ad all etries {B i,1 } i I ivolve oly bits from a 1,..., a (j 2)k+1, which were already fixed (sice (j 2)k + 1 (j 1)k). Fixig a (j 1)k+1 such that equality o the first bit holds, the secod bit B j,2 equals the resultig liear combiatio with probability 1/2 as well. This happes sice B j,2 equals S j,2 + a (j 1)k+2, where a (j 1)k+2 was t already fixed, ad all etries {B i,2 } i I ivolve oly bits from a 2,..., a (j 2)k+2, which were already fixed (sice (j 2)k + 2 (j 1)k + 1). Ad so o, every bit i the j-th row of B equals the resultig liear combiatio with probability 1/2, coditioed o the fixig of the previous bits. Overall, B j = i I c ib i with probability 2 m for a fixed choice of coefficiets {c i } i I. 8 Takig a uio boud over all possible coefficiets {c i } i I gives Pr[E l E 1,..., E l 1 ] 2 I 2 m 2 m/2. Pluggig this boud ito Eq. (3) we get Pr [Eq. (2) holds for I] Pr[E 1 ] Pr[E 2 E 1 ] Pr[E t E 1,..., E t 1 ] ( 2 m/2) t 2 mk/8. where i the last iequality we used t k/4. Takig a uio boud over all possible sequeces I of legth at most m/2, whose umber is defiitely less tha 2 m, ad usig k 16, we get Pr[rak(S + A) m/2] 2 m 2 mk/8 2 mk/16. We cotiue with our mai theorem. Theorem 3.2 (radom Hakel matrices are rigid). Let A F 2 be a radom Hakel matrix A i,j = a i+j where a 2,..., a 2 are uiform idepedet radom bits. The, for every r /32, with probability 1 o(1), the matrix A has rigidity 3 for rak r. 160r 2 log 8 Alteratively, coditioed o a 1,..., a (j 1)k ad the choice of the liear combiatio, there exist exactly oe choice for a (j 1)k+1,..., a (j 1)k+m that satisfies Eq. (5). 7

10 Before provig Theorem 3.2, we state a immediate corollary of it. Corollary 3.3. Let A F 2 be a radom Hakel matrix. The, there exists a uiversal costat c > 0 such that for every ε > 0 1. With probability 1 o(1), the matrix A is c 2 2ε / log rigid for rak 1/2+ε. 2. With probability 1 o(1), the matrix A is m 3 rigid for rak m = c 3/5 log 1/5. 3. With probability 1 o(1), the matrix A is c 1+2ε / log rigid for rak 1 ε. Proof of Theorem 3.2. Suppose towards cotradictio that A ca be represeted as a sum of a matrix R of rak at most r, ad a s-sparse matrix S, where s 3 /160r 2 log. Let m = 2r, ad assume for coveiece that k = /m is a iteger. Cosider the followig partitio of A s etries ito (/m) 2 submatrices, each of dimesio m m. For i [/m] ad j [/m], let I i = {i, i + k,..., i + (m 1)k}, J j = {(j 1)m + 1, (j 1)m + 2,..., jm}. (6) Deote by A i,j (R i,j, S i,j, resp.) the matrix A (R, S, resp.) restricted to rows I i ad colums J j. See Figure 1 for a example of such a submatrix. The mai observatio is that for each a 2 a 3 a 4 a 5 a 6 a 7 a 8 a 9 a 3 a 4 a 5 a 6 a 7 a 8 a 9 a 10 a 4 a 5 a 6 a 7 a 8 a 9 a 10 a 11 a 5 a 6 a 7 a 8 a 9 a 10 a 11 a 12 a 6 a 7 a 8 a 9 a 10 a 11 a 12 a 13 a 7 a 8 a 9 a 10 a 11 a 12 a 13 a 14 a 8 a 9 a 10 a 11 a 12 a 13 a 14 a 15 a 9 a 10 a 11 a 12 a 13 a 14 a 15 a 16 Figure 1: A submatrix A 1,1 of the matrix A, for m = 4 ad k = 2. (i, j) [/m] 2, the matrix A i,j is of the form eeded by the mai lemma. Aother observatio is that sice the submatrices S i,j partitios the sparse matrix S, oe of them has sparsity at most s s m2. I additio, sice rak of a submatrix may oly decrease, for every i, j, it holds that 2 rak(r i,j ) rak(r) r. We say that A i,j is simple if it ca be represeted as a sum of a s -sparse matrix ad a matrix of rak at most r. By the above discussio, A ca be represeted as S + R where S is s-sparse ad R is of rak at most r, oly if there exists a submatrix A i,j that is simple. We shall show that the latter occurs with very low probability: Pr [ i, j : A i,j is simple ] i,j i,j Pr[A i,j is simple] S F m m 2 : wt(s) s Pr[rak(A i,j + S) m/2] (Uio Boud) (Uio Boud) ( ) ( ) 2 m 2 m s 2 mk/16 (Lemma 3.1) < 2 (2m 2 ) s 2 /16. (km = ) Fially, usig s 40 log, which follows from s 3 160r 2 log, we get that Pr[ i, j : Ai,j is simple] = o(1), which completes the proof. 8

11 Note that the proof works as log as the umber of possibilities for a s -sparse matrix S i,j is smaller tha 2 /16 / 2. Our ext theorem exploits the fact that there is a smaller umber of possibilities for submatrices of structured sparse matrices (as i Defiitio 2.1). I fact, this is the oly property of S that the foregoig proof uses. This yields the followig improved boud. Theorem 3.4 (radom Hakel matrices are structured rigid). Let A F 2 be a radom Hakel matrix. The, for every r /32, ad s 3 /1000r 2, with probability 1 o(1), the matrix A has structured rigidity s for rak r. Before provig Theorem 3.4 we state three corollaries of it. The first corollary is immediate by choosig r = 3/5. Corollary 3.5. Let A F 2 be a radom Hakel matrix. The, there exists a uiversal costat c > 0 such that with probability 1 o(1), the matrix A has structured rigidity c 9/5 for rak 3/5. The secod corollary follows from the first corollary ad Theorem 2.4. Corollary 3.6. Let A F 2 be a radom Hakel matrix, ad let F (x, y) = j=1 A i,jx i y j. The, with probability 1 o(1), it holds that C(F ) = Ω( 3/5 ). The last corollary shows that there exists a explicit triliear form with AN-complexity Ω( 3/5 ). This is the first improvemet over the trivial Ω( ) lower boud for explicit tesors, ad i doig so it solves Problem 4.2 from [GW13] i the affirmative. Goldreich ad Wigderso [GW13, Prop. 4.6] show that if some Toeplitz matrix have AN-complexity Ω(m), the F tet defied i Eq. (1) has ANcomplexity Ω(m) has well. We follow their method, but preset a simpler argumet for a differet triliear fuctio. Corollary 3.7. Let F : {0, 1} {0, 1} {0, 1} 2 {0, 1} be the triliear fuctio defied by F (x, y, z) = j=1 z i+jx i y j. The, C(F ) = Ω( 3/5 ). Proof. Accordig to Corollary 3.6, there exists a Hakel matrix A, defied by some diagoal values a 2,..., a 2, such that the biliear form i,j a i+jx i y j has AN-complexity Ω( 3/5 ). Let C be a triliear circuit computig F with miimal AN-complexity, ad deote its complexity by m. Fixig the values of the variables z i to a i, for all i {2,..., 2}, we get a biliear circuit i x ad y of AN-complexity at most m. Thus, m = Ω( 3/5 ). We retur to prove Theorem 3.4. Proof of Theorem 3.4. The proof follows the lies of the proof of Theorem 3.2. We let m = 2r, k = /m, ad t = s 1/3. We assume towards cotradictio that A = S + R, where R is of rak at most r, ad S is a sum of t matrices S 1,..., S t F 2, such that the oes i each matrix S l are a subset of some X l Y l, where X l, Y l t. Deote by T the -by- matrix over F 2 with T i,j = 1 iff (i, j) is cotaied i at least oe X l Y l. It is clear from T s defiitio that the oes i S are a subset of the oes i T. As i Theorem 3.2, we partitio A, R, S, ad also T, to (/m) 2 submatrices, accordig to the partitio of row idices I 1,..., I /m ad colum idices J 1,..., J /m, defied as i the proof of Theorem 3.2 (see Eq. 6). For a radom (i, j) [/m] 2, it holds that [ E wt(t i,j ) ] t 3 m2 i,j 2, E i,j [ t ] X l I i t t m, E i,j [ t ] Y l J j t t m. (7) 9

12 We say that a submatrix T i,j is good if wt(t i,j ) 4t 3 m2 2, t X l I i 4t t m, t Y l J j 4t t m. (8) Usig Markov s iequality, each of the above three evets happe with probability at least 3/4. Usig uio boud (o the complemet evets) with probability at least 1/4 all evets occur simultaeously, makig T i,j good. Next, we cout the umber of possible good submatrices T i,j. Each such submatrix is uiquely determied by the sets X 1,..., X t ad Y 1,..., Y t, where X l = X l I i ad Y l = Y l J j. Furthermore, a collectio (X 1,..., X t) such that l X l 4t2 m correspods to a set X I i [t] of size at most 4t 2 m such that (p, l) X iff p X l (ad similarly for (Y 1,..., Y t )). Hece, the umber of possible good submatrices is at most { X I i [t] : X 4t2 m } 2 ( = mt 4t 2 m/ ) 2 ( ) 2 (2mt) 4t2 m/ 16t 2m/. We say that S i,j is good if T i,j is good, ad we say that A i,j is simple if it is the sum of a good S i,j ad a matrix of rak at most r. Next, we cout the umber of possible good submatrices S i,j. Sice the oes of S i,j are a subset of the oes i T i,j, the umber of possibilities for S i,j is at most 16t2 m/ 2 wt(t i,j) 16t2 m/ 2 4t3 m 2 / 2. Usig the boud o the umber of possible good submatrices S i,j, we may boud the probability that some A i,j is simple: Pr [ i, j : A i,j is simple ] Pr[rak(A i,j + S i,j ) m/2] (Uio Boud) i,j Recall that m = 2r ad k = /m to get which is o(1) for t 3 S i,j good ( ) 2 16t 2 m/ 2 4t3 m 2 / 2 2 mk/16 (Lemma 3.1) m Pr [ i, j : A i,j is simple ] 2 2 log + 32 log t2 r/ + 16t 3 r 2 / 2 /16, 3 ad r. 1000r 2 Geeralizatio to Larger Fields. The choice of field F 2 was ot crucial i the proofs of Lemma 3.1, Theorem 3.2 ad Theorem 3.4. Oe ca sytactically replace the field size 2 by ay prime power q, keepig the proofs itact. Furthermore, i Theorem 3.2, we slightly beefit from takig a larger field. For details see Appedix A.1. 4 The Structure of Matrices of Small Biliear Circuits I this sectio we shall further refie the structure of matrices associated with small biliear circuits, beyod the structure captured by Defiitio 2.1 ad Theorem 2.4. We begi with the structure that arises from depth-2 biliear circuits, ad the cotiue to the structure arisig from geeral biliear circuits. Our aalysis follows the proof of [GW13, Thm. 4.4], ad it ca be viewed as relatig to fier otios of structured rigidity (tha the oe of Defiitio 2.1). We the follow the first part of the proof of Theorem 3.4, ad fid submatrices with correspodig (rigidity-like) parameters. 10

13 4.1 The Structure of Matrices Associated with Depth Two Biliear Circuits We say a row/colum i a matrix is m-sparse if it cotais at most m o-zero etries. Likewise, a liear fuctio l(x) (resp. l (y)) is m-sparse if it depeds o at most m etries i x (resp. y). Lastly, recall that by Defiitio 2.3, C 2 (F ) is the miimal AN-complexity of a depth-two biliear circuit computig F. Propositio 4.1 (Structure of fuctios computed by depth two biliear circuits). If C 2 (F ) = m, the F ca be expressed as L i(x)l i (y) + Q l (x, y) where L 1,..., L m are m-sparse liear fuctios, L 1,..., L m are geeral liear fuctios, ad each Q l is a biliear fuctio of at most m variables from x ad at most m variables from y. The matrix associated with F has the form A = CL row + S l (9) where C is a geeral m matrix, L row is a m matrix with m-sparse rows, ad each S l is a matrix whose oes reside i a m m rectagle. Propositio 4.1 is proved explicitly i the warm-up part of the proof of [GW13, Thm. 4.4]. 4.2 The Structure of Matrices Associated with Geeral Biliear Circuits Propositio 4.2 (Structure of fuctios computed by geeral biliear circuits). If C(F ) = m, the F ca be expressed as L i (x)l i(y) + M i(x)m i (y) + Q l (x, y) where L 1,..., L m ad M 1,..., M m are m-sparse liear fuctios, L 1,..., L m ad M 1,..., M m are geeral liear fuctios, ad each Q l is a biliear fuctio of at most m variables from x ad at most m variables from y. The matrix associated with F has the form A = L col B + CL row + S l (10) where L col is a m matrix with m-sparse colums, B is a geeral m matrix, C is a geeral m matrix, L row is a m matrix with m-sparse rows, ad each S l is a matrix whose oes reside i a m m rectagle. Propositio 4.2 is oly implicit i the proof of [GW13, Thm. 4.4], ad we iclude its proof i Appedix A Substructures I this subsectio, similarly to the first part of the proof of Theorem 3.4, we fid a submatrix (of a matrix associated with a biliear circuit) that has average rigidity-like parameters. Startig with Propositio 4.2, for C(F ) = m, we write the matrix A associated with F as A = L col B + CL row + m S l such that the o-zero etries of S l are a subset of X l Y l, where X l, Y l m. Deote 11

14 by T = m X l Y l, ad ote that T m 3. Let I 1,..., I /2m ad J 1,..., J /2m be some fixed equipartitio of the row idices ad colum idices of A, respectively, where each I i ad J j is of size 2m. This partitio aturally defies (/2m) 2 submatrices as follows. For ay (i, j) we deote by A i,j (resp. S i,j l ) the matrix A (resp. S l) restricted to rows I i ad colums J j. For ay i (resp. j) we deote by L i col ad C i (resp. B j ad L j row) the matrices L col ad C (resp,. B ad L row ) restricted to I i (resp. J j ). The, oe ca write A i,j = L i col Bj + C i L j row + S i,j l, (11) where S i,j l T (I i J j ). Next, we show that there exists a choice of (i, j) with favorable properties (to be exploited i the ext sectio) of the submatrices of L i col, Lj row ad o the subsets {X l I i } l, {Y l J j } l, ad T (I i J j ). Propositio 4.3 (Structure of submatrix of matrices associated with small biliear circuits). There exists a (i, j) such that: (1) T (I i J j ) 24m5, (2) m X l I i 12m3, (3) m Y l J j 12m 3, (4) wt(li col ) 12m3 2, ad (5) wt(lj row) 12m3. If C 2 (F ) = m, the the same statemet holds, except that we ca replace L col ad B with the 0 m ad 0 m matrices, respectively. Proof. For a uiformly radom (i, j) [/2m] 2, it holds that E [ T (I i J j ) ] m 3 (2m)2 i,j [ 2 m ] E X l I i m m 2m i,j [ m ] E Y l J j m m 2m i,j E i,j [wt(li col )] m m 2m E i,j [wt(lj row)] m m 2m Usig Markov s iequality, each of the followig bad evets occur with probability at most 1/6 T (I i J j ) 6m 3 (2m)2 2 X l I i 6m m 2m Y l J j 6m m 2m wt(l i col ) 6m m 2m wt(l j row) 6m m 2m By uio boud, with probability at least 1 5/6 over the choice of (i, j), oe of the bad evets occur, which completes the proof. 12

15 We wish to express the structure captured by Eq. (11) i terms of liear equatios o the etries of the matrix A i,j l Si,j l. To do so we eed the followig defiitio. Defiitio 4.4 (Orthogoal Complemet of a Matrix). Let m. If A is a m matrix, ad B is a ( m) matrix of rak m such that BA = 0 the we say that B is a left orthogoal complemet of A. If A is a m matrix, ad B is a ( m) matrix of rak m such that AB = 0 the we say that B is a right orthogoal complemet of A. It is well kow that ay matrix over a field has a orthogoal complemet. Now, suppose that A i,j l S l i,j = L i col Bj + C i L j row (as i Eq. (11)). Let D be a m 2m matrix which is a left orthogoal complemet of L i col, ad let E be a 2m m matrix which is a right orthogoal complemet of L j row. The, D (A i,j l S l i,j ) E = 0 m m. (12) I the case of depth-2 circuits we have A i,j l S l i,j = C i L j row. Usig E, the right orthogoal complemet of L j row as above, we ca write (A i,j l S l i,j ) E = 0 2m m. (13) I the ext sectio, we shall desig tests based o Equatios (12) ad (13). Remark: Note that there are may possible choices of a orthogoal complemet of a give matrix. Therefore, we shall refer to the left (right, resp.) orthogoal complemet of A as some caoical choice of a left (right, resp.) orthogoal complemet of A, say the first such matrix accordig to lexicographical order (over a fiite field F). 5 Testig AN Complexity ad AN2 Complexity We would like to desig a test such that matrices associated with (AN or AN2) complexity at most m will surely pass the test, whereas the matrices we are iterested i will fail it. (Sice the test is merely a metal experimet, i.e., we do ot ited to actually ru it, the test could be iefficiet.) The poit is that ay matrix o which the test fails must have complexity greater tha m. We will show that a radom Toeplitz matrix, as well as a matrix whose etries are sampled from a 2 - biased distributio, will fail the test with overwhelmig probability, thus provig complexity lower bouds for such matrices. We will preset two tests: Oe for AN-complexity failig most matrices take from a small-biased space, ad oe for AN2-complexity failig most Toeplitz matrices. 5.1 Lower Bouds for the AN-Complexity of Small-Biased Matrices For i [/2m] ad j [/2m], let 9 I i = {i, i + (/2m),..., i + (2m 1) (/2m)}, J j = {(j 1) (2m) + 1, (j 1) (2m) + 2,..., j (2m)}. (14) 9 The specific choice for I i ad J j is ot crucial for our argumet i this subsectio, however it will be importat i the ext subsectio. Hece, sice we eed to pick some partitio, we might as well choose this oe. 13

16 ad deote by A i,j the 2m-by-2m sub-matrix of A obtaied by restrictig A to rows I i ad colums J j. Cosider the followig test, where A i,j is viewed as idexed by [2m] [2m] rather tha by I i J j. Test 1 AN-Complexity Test Iput: Matrix A F 2 ad parameter m [] 1: for i = 1,..., /2m ad j = 1,..., /2m do 2: for all subsets {X i l }m of [2m] such that l Xi l 12m3 3: for all subsets {Y j l }m of [2m] such that l Y j l 12m3 4: Let T := m Xi l Y j 5: if T 24m5 the 2 6: for all matrices L i col l. do do 12m3 of dimesio 2m m ad sparsity at most do 7: Let D be the left orthogoal complemet of L i col. 8: for all matrices L j row of dimesio m 2m ad sparsity at most 12m3 do 9: Let E be the right orthogoal complemet of L j row. 10: if there exists N F 2m 2m 2 such that N T, ad D(A i,j N)E = 0 m m the 11: retur Pass. 12: retur Fail. The followig is a immediate corollary of Propositio 4.3 ad Eq. (12). Corollary 5.1. Every matrix associated with a biliear circuit of AN-complexity at most m passes Test 1 with parameter m. I this subsectio we cosider a distributio of matrices whose etries are chose from a small biased sample space. Specifically, we shall use a sample space over strigs of legth N = 2 i order to defie -by- matrices. We shall show that almost all such matrices fail Test 1 with parameter m. But we eed a few prelimiaries first. Prelimiaries. Recall the defiitio of a ε-biased distributio from [NN93]. Defiitio 5.2 (small-biased distributio). A distributio X over {0, 1} N is said to be ε-biased if for every o-empty set S [N], it holds that E x X [( 1) i S x i ] ε. We shall use the followig property of ε-biased distributios (implicit i [NN93]). Lemma 5.3 ([AGHP92, Lem. 1]). Let X be a ε-biased distributio over {0, 1} N. Let l 1,..., l t be liearly idepedet liear fuctios o x 1,..., x N. The, the probability that all liear fuctios equal 0 simultaeously is at most ε + 2 t. We shall also use the followig simple fact from liear algebra. Fact 5.4. Let t,, m N such that t m. Let l 1,..., l t be a sequece of liearly idepedet liear fuctios (over F) o x 1,..., x. The, l 1,..., l t spa at least t m liearly idepedet fuctios that ivolve oly the variables x m+1,..., x. Proof. Thik of the liear fuctios as vectors i F, ad let V = spa{l 1,..., l t }. Cosider the subspace U = spa{e m+1,..., e }, where e i F is the uit vector with 1 i the i-th coordiate ad 0 elsewhere. The, dim(u V ) dim(u) + dim(v ) = ( m) + t = t m, whereas U V is the spa of l 1,..., l t that is supported oly o the last m coordiates. 14

17 Actual Results. We are ow ready to aalyze the probability that a matrix sampled from a small biased space passes Test 1. The core of the aalysis refers to a sigle applicatio of Step 10, which refers to a specific choice of i, j, {Xl i}m, {Y j l }m as well as Li col, Lj row (which i tur, fixes D ad E as well). Lemma 5.5 (core of the aalysis of Test 1). Fix i, j, {Xl i}m j ad {Yl }m that pass the check of Step 5, ad fix L i col ad L j row (which i tur, fixes D ad E as well). The, a matrix A whose etries are sampled from a ε-biased distributio satisfies the coditio i Step 10 with probability at most ε + 2 m2 +24m 5 / 2. Proof. For a fixed choice of i, j, {X i l }m, {Y j l }m, Li col ad L j row as above, we cosider a specific submatrix of dimesio 2m 2m of A, deoted A i,j. Note that the correspodig left (resp. right) orthogoal complemet of L i col (resp. L j row) is a m-by-2m (resp. 2m-by-m) matrix of rak m, deoted by D (resp. E). Recall that A i,j is a submatrix whose etries are sampled accordig to a ε-biased distributio. Our goal is to show that the equatio D(A i,j N)E = 0 (checked i Step 10) implies a lot of liearly idepedet liear equatios o the etries of A i,j. Let Z be a 2m 2m matrix of (2m) 2 Boolea variables, where we will later take Z to be A i,j N. Iterpret the equatios DZE = 0 m m as m 2 liear equatios o the (2m) 2 variables i Z. For i [m] ad j [m], we have a equatio of the form D i ZE (j) = 0, where D i is the i-th row of D ad E (j) is the j-th colum of E. We ca write D i ZE (j) = 2 2 k=1 D i,k Z k,l E l,j = k,l (D i E (j) ) k,l Z k,l ; that is, the coefficiets of the equatio are the tesor product of the vector D i with the vector E (j). Thikig of these m 2 liear equatios o (2m) 2 variables as a big matrix of dimesio m 2 (2m) 2, we ote that this matrix of liear equatios is the tesor product of D ad E, sice the (i, j) row equals to D i E (j) (viewed as a (2m) 2 -bit log vector). It is a kow fact that the rak of the tesor product of ay two matrices is the product of their rak; hece, we get rak(d E ) = rak(d) rak(e ) = m 2. I other words, we have a liearly idepedet set of m 2 liear equatios o the variables Z. However, we wat to get liear equatios over the variables of A, where Z = A N. Say that Z k,l is a oisy variable if (k, l) T. It will be eough to show that there are may idepedet liear equatios which ivolve oly o-oisy variables of the matrix. Sice the umber of oisy variables is T, by Fact 5.4 we ca fid at least m 2 T idepedet liear equatios that do ot ivolve oisy variables. Overall, we got m 2 T idepedet liear equatios o A i,j. By Lemma 5.3, a submatrix A i,j whose etries are sampled accordig to a ε-biased distributio satisfies all m 2 T equatios with probability at most ε + 2 m2 + T. Lastly, the fact that {Xl i}m j ad {Yl }m passed the check of Step 5 meas that T 24m 5 / 2, which fiishes the proof. Theorem 5.6 (Almost all ε-biased matrices have high AN-complexity). A matrix A whose etries are sampled from a ε biased distributio fails Test 1 with parameter m (which implies that the correspodig biliear fuctio has AN-complexity greater tha m), with probability at least ( ) ( ) 2 2m 2 4 ( 1 2m 12m 3 ε + 2 m2 +24m 5 / 2) /. 15

18 I particular, for ε = 2 ad m = 2/3 10(log ) 1/3, this probability is at least 1 2 /2, for sufficietly large. Proof. We use a uio boud over all possible i, j, {Xl i}m, {Y j l }m, Li col ad L j row that ca be selected by the test, ad employ Lemma 5.5 for each possibility. The umber of optios for choosig (i, j) is (/2m) 2 ; the umber of optios for choosig {Xl i}m j (resp., {Yl }m ) is at most ( 2m 2 12m /) ; 3 the umber of optios for choosig L i col (resp., Lj row) is at most ( 2m 2 12m /) Lower Bouds for the AN2-Complexity of Radom Toeplitz Matrices The followig is a degeerate versio of Test 1. Recall the defiitio of I i ad J j from Eq. (14), ad the defiitio of A i,j. Test 2 AN-2-Complexity Test Iput: Matrix A F 2 ad parameter m [] 1: for i = 1,..., /2m ad j = 1,..., /2m do 2: for all subsets {X i l }m of [2m] such that l Xi l 12m3 3: for all subsets {Y j l }m of [2m] such that l Y j l 12m3 4: Let T := m Xi l Y j l. do do 5: if T 24m5 2 the 6: for all matrices L j row of dimesio m 2m ad sparsity at most 12m3 do 7: Let E be the right orthogoal complemet of L j row. 8: if there exists N F2 2m 2m such that N T, ad (A i,j N)E = 0 2m m the 9: retur Pass. 10: retur Fail. The followig is a immediate corollary of Propositio 4.3 ad Eq. (13). Corollary 5.7. Every matrix associated with a biliear circuit of AN2-complexity at most m passes Test 2 with parameter m. Lemma 5.8 (core of the aalysis of Test 2). Fix i, j, {Xl i}m j ad {Yl }m that pass the check of Step 5, ad fix L j row (which i tur, fixes E as well). The, a radom Hakel matrix A satisfies the coditio i Step 8 with probability at most 2 /2+6m3 / Proof. For a fixed choice of i, j, {Xl i}m, {Y j l }m ad Lj row, we cosider a specific submatrix of dimesio 2m 2m of A, deoted A i,j. Note that the correspodig right orthogoal complemet of L j row is a 2m-by-m matrix of rak m, deoted by E. By the defiitio of I i ad J j i Eq. (14), A i,j is of the form a 1 a 2 a 3... a 2m a k+1 a k+2 a k+3... a k+2m a (2m 1)k+1 a (2m 1)k+2 a (2m 1)k+3... a (2m 1)k+2m where k = /(2m) ad a 1,..., a (2m 1)k+2m are uiform idepedet radom bits. Our goal will be to show that the equatio (A i,j N) E = 0 2m m implies a lot of liearly idepedet liear equatios o the radom variables a 1,..., a (2m 1)k+2m. 16

19 First thik of a geeric 2m 2m matrix Z as a matrix of (2m) 2 variables, ad iterpret the equatios ZE = 0 2m m as liear equatios o Z. For each row l [2m], we have m equatios correspodig to Z l E = 0 1 m, which are liearly idepedet. Deote by T l the itersectio of T with the idices correspodig to the l-th row of the submatrix, i.e. T l = T ({l} [2m]). Say that Z l,l is a oisy variable if (l, l ) T. By Fact 5.4, we ca get at least m T l idepedet liear equatios o the l-th row of Z that do ot ivolve oisy variables. Summig over all l s we have at least 2m (m T l ) = 2m 2 T idepedet liear equatios that do ot ivolve the oisy etries of the matrix, ad such that each equatio ivolves oly variables from oe row of Z. Take Z = A i,j N; sice we got equatios o Z that do ot ivolve oisy etries, these are actually equatios o A i,j as well. The mai difficulty is that we wat to exhibit liearly idepedet liear equatios o the variables a 1,... a (2m 1)k+2m, but the equatios we got may ot be liearly idepedet oce we idetify multiple etries i the matrix A i,j with the same variable. 10 To solve this issue, we shall look for a set of equatios which remais liearly idepedet after this idetificatio. Let l = m T l be umber of liearly idepedet equatios we got o the l-th row. Let s = (2m) 2 /, ad cosider all rows startig from some idex r [s], ad takig jumps of s. The, by the pigeo-hole priciple there exists a r [s] such that l (2m 2 T )/s. l:l r mod s A key poit is that by our choice of s, the l-th row ad the (l + s)-th row of A i,j deped o disjoit sets of radom variables, sice s k (2m)2 2m = 2m. Thus, the sets of variables out of a 1,..., a (2m 1)k+2m that participate i rows with idex i {l : l r mod s} are pairwise disjoit, ad the equatios we got o these rows are liearly idepedet as equatios over the variables a 1,..., a (2m 1)k+2m. Sice we got at least (2m 2 T )/s idepedet liear equatios o completely radom bits, all equatios hold simultaeously with probability at most 2 ( 2m2 + T )/s. The fact that {Xl i}m j ad {Yl }m passed the check i Step 5 meas that T 24m5 / 2, ad usig s = 2m 2 /, we get a probability boud of 2 ( 2m2 +24m 5 / 2 ) 4m 2, which completes the proof. Theorem 5.9 (Almost all radom Hakel matrices have high AN2-complexity). A radom Hakel matrix A fails Test 2 with parameter m (which implies it has direct complexity at least m) with probability at least ( ) ( ) 2 2m m 12m 3 2 /2+6m3 /. / I particular, for m = 2/3, this probability is at least 1 2 /4, for large eough. 10(log ) 1/3 Proof. We use a uio boud over all the ( ) 2 ( 2m 2 )3 2m 12m 3 / possible ways to pick i, j, {X i l } m, {Y j l }m ad L j row, ad employ Lemma 5.8 to boud each possibility. Explicit 3-Liear Fuctios with C 2 = Ω( 2/3 ). The followig is a corollary of Theorem 5.9. Corollary Let F : {0, 1} {0, 1} {0, 1} 2 {0, 1} be the triliear fuctio defied by F (x, y, z) = j=1 z i+jx i y j. The, C 2 (F ) = Ω( 2/3 / log 1/3 ). We omit the proof, sice it is idetical to that of Corollary I fact, we caot expect this set of equatios to be liearly idepedet simply because there are too may equatios (i.e., more equatios tha variables). 17

Matrix Rigidity of Random Toeplitz Matrices

Matrix Rigidity of Random Toeplitz Matrices Matrix Rigidity of Radom Toeplitz Matrices Oded Goldreich Avishay Tal Jue 12, 2016 Abstract A matrix A is said to have rigidity s for rak r if A differs from ay matrix of rak r o more tha s etries. We

More information

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014. Product measures, Toelli s ad Fubii s theorems For use i MAT3400/4400, autum 2014 Nadia S. Larse Versio of 13 October 2014. 1. Costructio of the product measure The purpose of these otes is to preset the

More information

Basics of Probability Theory (for Theory of Computation courses)

Basics of Probability Theory (for Theory of Computation courses) Basics of Probability Theory (for Theory of Computatio courses) Oded Goldreich Departmet of Computer Sciece Weizma Istitute of Sciece Rehovot, Israel. oded.goldreich@weizma.ac.il November 24, 2008 Preface.

More information

Convergence of random variables. (telegram style notes) P.J.C. Spreij

Convergence of random variables. (telegram style notes) P.J.C. Spreij Covergece of radom variables (telegram style otes).j.c. Spreij this versio: September 6, 2005 Itroductio As we kow, radom variables are by defiitio measurable fuctios o some uderlyig measurable space

More information

On Random Line Segments in the Unit Square

On Random Line Segments in the Unit Square O Radom Lie Segmets i the Uit Square Thomas A. Courtade Departmet of Electrical Egieerig Uiversity of Califoria Los Ageles, Califoria 90095 Email: tacourta@ee.ucla.edu I. INTRODUCTION Let Q = [0, 1] [0,

More information

Problem Set 2 Solutions

Problem Set 2 Solutions CS271 Radomess & Computatio, Sprig 2018 Problem Set 2 Solutios Poit totals are i the margi; the maximum total umber of poits was 52. 1. Probabilistic method for domiatig sets 6pts Pick a radom subset S

More information

If a subset E of R contains no open interval, is it of zero measure? For instance, is the set of irrationals in [0, 1] is of measure zero?

If a subset E of R contains no open interval, is it of zero measure? For instance, is the set of irrationals in [0, 1] is of measure zero? 2 Lebesgue Measure I Chapter 1 we defied the cocept of a set of measure zero, ad we have observed that every coutable set is of measure zero. Here are some atural questios: If a subset E of R cotais a

More information

Advanced Stochastic Processes.

Advanced Stochastic Processes. Advaced Stochastic Processes. David Gamarik LECTURE 2 Radom variables ad measurable fuctios. Strog Law of Large Numbers (SLLN). Scary stuff cotiued... Outlie of Lecture Radom variables ad measurable fuctios.

More information

The Growth of Functions. Theoretical Supplement

The Growth of Functions. Theoretical Supplement The Growth of Fuctios Theoretical Supplemet The Triagle Iequality The triagle iequality is a algebraic tool that is ofte useful i maipulatig absolute values of fuctios. The triagle iequality says that

More information

Resolution Proofs of Generalized Pigeonhole Principles

Resolution Proofs of Generalized Pigeonhole Principles Resolutio Proofs of Geeralized Pigeohole Priciples Samuel R. Buss Departmet of Mathematics Uiversity of Califoria, Berkeley Győrgy Turá Departmet of Mathematics, Statistics, ad Computer Sciece Uiversity

More information

A Note on Matrix Rigidity

A Note on Matrix Rigidity A Note o Matrix Rigidity Joel Friedma Departmet of Computer Sciece Priceto Uiversity Priceto, NJ 08544 Jue 25, 1990 Revised October 25, 1991 Abstract I this paper we give a explicit costructio of matrices

More information

Optimally Sparse SVMs

Optimally Sparse SVMs A. Proof of Lemma 3. We here prove a lower boud o the umber of support vectors to achieve geeralizatio bouds of the form which we cosider. Importatly, this result holds ot oly for liear classifiers, but

More information

An Introduction to Randomized Algorithms

An Introduction to Randomized Algorithms A Itroductio to Radomized Algorithms The focus of this lecture is to study a radomized algorithm for quick sort, aalyze it usig probabilistic recurrece relatios, ad also provide more geeral tools for aalysis

More information

Lecture 2. The Lovász Local Lemma

Lecture 2. The Lovász Local Lemma Staford Uiversity Sprig 208 Math 233A: No-costructive methods i combiatorics Istructor: Ja Vodrák Lecture date: Jauary 0, 208 Origial scribe: Apoorva Khare Lecture 2. The Lovász Local Lemma 2. Itroductio

More information

Math 61CM - Solutions to homework 3

Math 61CM - Solutions to homework 3 Math 6CM - Solutios to homework 3 Cédric De Groote October 2 th, 208 Problem : Let F be a field, m 0 a fixed oegative iteger ad let V = {a 0 + a x + + a m x m a 0,, a m F} be the vector space cosistig

More information

Lecture Notes for Analysis Class

Lecture Notes for Analysis Class Lecture Notes for Aalysis Class Topological Spaces A topology for a set X is a collectio T of subsets of X such that: (a) X ad the empty set are i T (b) Uios of elemets of T are i T (c) Fiite itersectios

More information

Infinite Sequences and Series

Infinite Sequences and Series Chapter 6 Ifiite Sequeces ad Series 6.1 Ifiite Sequeces 6.1.1 Elemetary Cocepts Simply speakig, a sequece is a ordered list of umbers writte: {a 1, a 2, a 3,...a, a +1,...} where the elemets a i represet

More information

4 The Sperner property.

4 The Sperner property. 4 The Sperer property. I this sectio we cosider a surprisig applicatio of certai adjacecy matrices to some problems i extremal set theory. A importat role will also be played by fiite groups. I geeral,

More information

The Boolean Ring of Intervals

The Boolean Ring of Intervals MATH 532 Lebesgue Measure Dr. Neal, WKU We ow shall apply the results obtaied about outer measure to the legth measure o the real lie. Throughout, our space X will be the set of real umbers R. Whe ecessary,

More information

Math 155 (Lecture 3)

Math 155 (Lecture 3) Math 55 (Lecture 3) September 8, I this lecture, we ll cosider the aswer to oe of the most basic coutig problems i combiatorics Questio How may ways are there to choose a -elemet subset of the set {,,,

More information

(A sequence also can be thought of as the list of function values attained for a function f :ℵ X, where f (n) = x n for n 1.) x 1 x N +k x N +4 x 3

(A sequence also can be thought of as the list of function values attained for a function f :ℵ X, where f (n) = x n for n 1.) x 1 x N +k x N +4 x 3 MATH 337 Sequeces Dr. Neal, WKU Let X be a metric space with distace fuctio d. We shall defie the geeral cocept of sequece ad limit i a metric space, the apply the results i particular to some special

More information

The multiplicative structure of finite field and a construction of LRC

The multiplicative structure of finite field and a construction of LRC IERG6120 Codig for Distributed Storage Systems Lecture 8-06/10/2016 The multiplicative structure of fiite field ad a costructio of LRC Lecturer: Keeth Shum Scribe: Zhouyi Hu Notatios: We use the otatio

More information

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence Chapter 3 Strog covergece As poited out i the Chapter 2, there are multiple ways to defie the otio of covergece of a sequece of radom variables. That chapter defied covergece i probability, covergece i

More information

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4.

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4. 4. BASES I BAACH SPACES 39 4. BASES I BAACH SPACES Sice a Baach space X is a vector space, it must possess a Hamel, or vector space, basis, i.e., a subset {x γ } γ Γ whose fiite liear spa is all of X ad

More information

Polynomial identity testing and global minimum cut

Polynomial identity testing and global minimum cut CHAPTER 6 Polyomial idetity testig ad global miimum cut I this lecture we will cosider two further problems that ca be solved usig probabilistic algorithms. I the first half, we will cosider the problem

More information

THE ASYMPTOTIC COMPLEXITY OF MATRIX REDUCTION OVER FINITE FIELDS

THE ASYMPTOTIC COMPLEXITY OF MATRIX REDUCTION OVER FINITE FIELDS THE ASYMPTOTIC COMPLEXITY OF MATRIX REDUCTION OVER FINITE FIELDS DEMETRES CHRISTOFIDES Abstract. Cosider a ivertible matrix over some field. The Gauss-Jorda elimiatio reduces this matrix to the idetity

More information

ACO Comprehensive Exam 9 October 2007 Student code A. 1. Graph Theory

ACO Comprehensive Exam 9 October 2007 Student code A. 1. Graph Theory 1. Graph Theory Prove that there exist o simple plaar triagulatio T ad two distict adjacet vertices x, y V (T ) such that x ad y are the oly vertices of T of odd degree. Do ot use the Four-Color Theorem.

More information

Discrete Mathematics for CS Spring 2007 Luca Trevisan Lecture 22

Discrete Mathematics for CS Spring 2007 Luca Trevisan Lecture 22 CS 70 Discrete Mathematics for CS Sprig 2007 Luca Trevisa Lecture 22 Aother Importat Distributio The Geometric Distributio Questio: A biased coi with Heads probability p is tossed repeatedly util the first

More information

The Random Walk For Dummies

The Random Walk For Dummies The Radom Walk For Dummies Richard A Mote Abstract We look at the priciples goverig the oe-dimesioal discrete radom walk First we review five basic cocepts of probability theory The we cosider the Beroulli

More information

Math Solutions to homework 6

Math Solutions to homework 6 Math 175 - Solutios to homework 6 Cédric De Groote November 16, 2017 Problem 1 (8.11 i the book): Let K be a compact Hermitia operator o a Hilbert space H ad let the kerel of K be {0}. Show that there

More information

Axioms of Measure Theory

Axioms of Measure Theory MATH 532 Axioms of Measure Theory Dr. Neal, WKU I. The Space Throughout the course, we shall let X deote a geeric o-empty set. I geeral, we shall ot assume that ay algebraic structure exists o X so that

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS MASSACHUSTTS INSTITUT OF TCHNOLOGY 6.436J/5.085J Fall 2008 Lecture 9 /7/2008 LAWS OF LARG NUMBRS II Cotets. The strog law of large umbers 2. The Cheroff boud TH STRONG LAW OF LARG NUMBRS While the weak

More information

Riesz-Fischer Sequences and Lower Frame Bounds

Riesz-Fischer Sequences and Lower Frame Bounds Zeitschrift für Aalysis ud ihre Aweduge Joural for Aalysis ad its Applicatios Volume 1 (00), No., 305 314 Riesz-Fischer Sequeces ad Lower Frame Bouds P. Casazza, O. Christese, S. Li ad A. Lider Abstract.

More information

Lecture 1: Basic problems of coding theory

Lecture 1: Basic problems of coding theory Lecture 1: Basic problems of codig theory Error-Correctig Codes (Sprig 016) Rutgers Uiversity Swastik Kopparty Scribes: Abhishek Bhrushudi & Aditya Potukuchi Admiistrivia was discussed at the begiig of

More information

Disjoint Systems. Abstract

Disjoint Systems. Abstract Disjoit Systems Noga Alo ad Bey Sudaov Departmet of Mathematics Raymod ad Beverly Sacler Faculty of Exact Scieces Tel Aviv Uiversity, Tel Aviv, Israel Abstract A disjoit system of type (,,, ) is a collectio

More information

CALCULATION OF FIBONACCI VECTORS

CALCULATION OF FIBONACCI VECTORS CALCULATION OF FIBONACCI VECTORS Stuart D. Aderso Departmet of Physics, Ithaca College 953 Daby Road, Ithaca NY 14850, USA email: saderso@ithaca.edu ad Dai Novak Departmet of Mathematics, Ithaca College

More information

CS284A: Representations and Algorithms in Molecular Biology

CS284A: Representations and Algorithms in Molecular Biology CS284A: Represetatios ad Algorithms i Molecular Biology Scribe Notes o Lectures 3 & 4: Motif Discovery via Eumeratio & Motif Represetatio Usig Positio Weight Matrix Joshua Gervi Based o presetatios by

More information

Machine Learning for Data Science (CS 4786)

Machine Learning for Data Science (CS 4786) Machie Learig for Data Sciece CS 4786) Lecture & 3: Pricipal Compoet Aalysis The text i black outlies high level ideas. The text i blue provides simple mathematical details to derive or get to the algorithm

More information

1 Review and Overview

1 Review and Overview DRAFT a fial versio will be posted shortly CS229T/STATS231: Statistical Learig Theory Lecturer: Tegyu Ma Lecture #3 Scribe: Migda Qiao October 1, 2013 1 Review ad Overview I the first half of this course,

More information

Random Walks on Discrete and Continuous Circles. by Jeffrey S. Rosenthal School of Mathematics, University of Minnesota, Minneapolis, MN, U.S.A.

Random Walks on Discrete and Continuous Circles. by Jeffrey S. Rosenthal School of Mathematics, University of Minnesota, Minneapolis, MN, U.S.A. Radom Walks o Discrete ad Cotiuous Circles by Jeffrey S. Rosethal School of Mathematics, Uiversity of Miesota, Mieapolis, MN, U.S.A. 55455 (Appeared i Joural of Applied Probability 30 (1993), 780 789.)

More information

Commutativity in Permutation Groups

Commutativity in Permutation Groups Commutativity i Permutatio Groups Richard Wito, PhD Abstract I the group Sym(S) of permutatios o a oempty set S, fixed poits ad trasiet poits are defied Prelimiary results o fixed ad trasiet poits are

More information

b i u x i U a i j u x i u x j

b i u x i U a i j u x i u x j M ath 5 2 7 Fall 2 0 0 9 L ecture 1 9 N ov. 1 6, 2 0 0 9 ) S ecod- Order Elliptic Equatios: Weak S olutios 1. Defiitios. I this ad the followig two lectures we will study the boudary value problem Here

More information

A Note on the Symmetric Powers of the Standard Representation of S n

A Note on the Symmetric Powers of the Standard Representation of S n A Note o the Symmetric Powers of the Stadard Represetatio of S David Savitt 1 Departmet of Mathematics, Harvard Uiversity Cambridge, MA 0138, USA dsavitt@mathharvardedu Richard P Staley Departmet of Mathematics,

More information

Square-Congruence Modulo n

Square-Congruence Modulo n Square-Cogruece Modulo Abstract This paper is a ivestigatio of a equivalece relatio o the itegers that was itroduced as a exercise i our Discrete Math class. Part I - Itro Defiitio Two itegers are Square-Cogruet

More information

Determinants of order 2 and 3 were defined in Chapter 2 by the formulae (5.1)

Determinants of order 2 and 3 were defined in Chapter 2 by the formulae (5.1) 5. Determiats 5.. Itroductio 5.2. Motivatio for the Choice of Axioms for a Determiat Fuctios 5.3. A Set of Axioms for a Determiat Fuctio 5.4. The Determiat of a Diagoal Matrix 5.5. The Determiat of a Upper

More information

6.3 Testing Series With Positive Terms

6.3 Testing Series With Positive Terms 6.3. TESTING SERIES WITH POSITIVE TERMS 307 6.3 Testig Series With Positive Terms 6.3. Review of what is kow up to ow I theory, testig a series a i for covergece amouts to fidig the i= sequece of partial

More information

Lecture 12: November 13, 2018

Lecture 12: November 13, 2018 Mathematical Toolkit Autum 2018 Lecturer: Madhur Tulsiai Lecture 12: November 13, 2018 1 Radomized polyomial idetity testig We will use our kowledge of coditioal probability to prove the followig lemma,

More information

Random Models. Tusheng Zhang. February 14, 2013

Random Models. Tusheng Zhang. February 14, 2013 Radom Models Tusheg Zhag February 14, 013 1 Radom Walks Let me describe the model. Radom walks are used to describe the motio of a movig particle (object). Suppose that a particle (object) moves alog the

More information

Measure and Measurable Functions

Measure and Measurable Functions 3 Measure ad Measurable Fuctios 3.1 Measure o a Arbitrary σ-algebra Recall from Chapter 2 that the set M of all Lebesgue measurable sets has the followig properties: R M, E M implies E c M, E M for N implies

More information

Pairs of disjoint q-element subsets far from each other

Pairs of disjoint q-element subsets far from each other Pairs of disjoit q-elemet subsets far from each other Hikoe Eomoto Departmet of Mathematics, Keio Uiversity 3-14-1 Hiyoshi, Kohoku-Ku, Yokohama, 223 Japa, eomoto@math.keio.ac.jp Gyula O.H. Katoa Alfréd

More information

Notes 27 : Brownian motion: path properties

Notes 27 : Brownian motion: path properties Notes 27 : Browia motio: path properties Math 733-734: Theory of Probability Lecturer: Sebastie Roch Refereces:[Dur10, Sectio 8.1], [MP10, Sectio 1.1, 1.2, 1.3]. Recall: DEF 27.1 (Covariace) Let X = (X

More information

Discrete-Time Systems, LTI Systems, and Discrete-Time Convolution

Discrete-Time Systems, LTI Systems, and Discrete-Time Convolution EEL5: Discrete-Time Sigals ad Systems. Itroductio I this set of otes, we begi our mathematical treatmet of discrete-time s. As show i Figure, a discrete-time operates or trasforms some iput sequece x [

More information

The natural exponential function

The natural exponential function The atural expoetial fuctio Attila Máté Brookly College of the City Uiversity of New York December, 205 Cotets The atural expoetial fuctio for real x. Beroulli s iequality.....................................2

More information

Math 220A Fall 2007 Homework #2. Will Garner A

Math 220A Fall 2007 Homework #2. Will Garner A Math 0A Fall 007 Homewor # Will Garer Pg 3 #: Show that {cis : a o-egative iteger} is dese i T = {z œ : z = }. For which values of q is {cis(q): a o-egative iteger} dese i T? To show that {cis : a o-egative

More information

Randomized Algorithms I, Spring 2018, Department of Computer Science, University of Helsinki Homework 1: Solutions (Discussed January 25, 2018)

Randomized Algorithms I, Spring 2018, Department of Computer Science, University of Helsinki Homework 1: Solutions (Discussed January 25, 2018) Radomized Algorithms I, Sprig 08, Departmet of Computer Sciece, Uiversity of Helsiki Homework : Solutios Discussed Jauary 5, 08). Exercise.: Cosider the followig balls-ad-bi game. We start with oe black

More information

# fixed points of g. Tree to string. Repeatedly select the leaf with the smallest label, write down the label of its neighbour and remove the leaf.

# fixed points of g. Tree to string. Repeatedly select the leaf with the smallest label, write down the label of its neighbour and remove the leaf. Combiatorics Graph Theory Coutig labelled ad ulabelled graphs There are 2 ( 2) labelled graphs of order. The ulabelled graphs of order correspod to orbits of the actio of S o the set of labelled graphs.

More information

CHAPTER 5. Theory and Solution Using Matrix Techniques

CHAPTER 5. Theory and Solution Using Matrix Techniques A SERIES OF CLASS NOTES FOR 2005-2006 TO INTRODUCE LINEAR AND NONLINEAR PROBLEMS TO ENGINEERS, SCIENTISTS, AND APPLIED MATHEMATICIANS DE CLASS NOTES 3 A COLLECTION OF HANDOUTS ON SYSTEMS OF ORDINARY DIFFERENTIAL

More information

Bertrand s Postulate

Bertrand s Postulate Bertrad s Postulate Lola Thompso Ross Program July 3, 2009 Lola Thompso (Ross Program Bertrad s Postulate July 3, 2009 1 / 33 Bertrad s Postulate I ve said it oce ad I ll say it agai: There s always a

More information

Lecture 3 The Lebesgue Integral

Lecture 3 The Lebesgue Integral Lecture 3: The Lebesgue Itegral 1 of 14 Course: Theory of Probability I Term: Fall 2013 Istructor: Gorda Zitkovic Lecture 3 The Lebesgue Itegral The costructio of the itegral Uless expressly specified

More information

w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ.

w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ. 2 5. Weighted umber of late jobs 5.1. Release dates ad due dates: maximimizig the weight of o-time jobs Oce we add release dates, miimizig the umber of late jobs becomes a sigificatly harder problem. For

More information

Mathematical Foundations -1- Sets and Sequences. Sets and Sequences

Mathematical Foundations -1- Sets and Sequences. Sets and Sequences Mathematical Foudatios -1- Sets ad Sequeces Sets ad Sequeces Methods of proof 2 Sets ad vectors 13 Plaes ad hyperplaes 18 Liearly idepedet vectors, vector spaces 2 Covex combiatios of vectors 21 eighborhoods,

More information

Quantum Computing Lecture 7. Quantum Factoring

Quantum Computing Lecture 7. Quantum Factoring Quatum Computig Lecture 7 Quatum Factorig Maris Ozols Quatum factorig A polyomial time quatum algorithm for factorig umbers was published by Peter Shor i 1994. Polyomial time meas that the umber of gates

More information

Statistics 511 Additional Materials

Statistics 511 Additional Materials Cofidece Itervals o mu Statistics 511 Additioal Materials This topic officially moves us from probability to statistics. We begi to discuss makig ifereces about the populatio. Oe way to differetiate probability

More information

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables Chapter 7 Limit Theorems Throughout this sectio we will assume a probability space (, F, P), i which is defied a ifiite sequece of radom variables (X ) ad a radom variable X. The fact that for every ifiite

More information

Section 4.3. Boolean functions

Section 4.3. Boolean functions Sectio 4.3. Boolea fuctios Let us take aother look at the simplest o-trivial Boolea algebra, ({0}), the power-set algebra based o a oe-elemet set, chose here as {0}. This has two elemets, the empty set,

More information

On the Linear Complexity of Feedback Registers

On the Linear Complexity of Feedback Registers O the Liear Complexity of Feedback Registers A. H. Cha M. Goresky A. Klapper Northeaster Uiversity Abstract I this paper, we study sequeces geerated by arbitrary feedback registers (ot ecessarily feedback

More information

subcaptionfont+=small,labelformat=parens,labelsep=space,skip=6pt,list=0,hypcap=0 subcaption ALGEBRAIC COMBINATORICS LECTURE 8 TUESDAY, 2/16/2016

subcaptionfont+=small,labelformat=parens,labelsep=space,skip=6pt,list=0,hypcap=0 subcaption ALGEBRAIC COMBINATORICS LECTURE 8 TUESDAY, 2/16/2016 subcaptiofot+=small,labelformat=pares,labelsep=space,skip=6pt,list=0,hypcap=0 subcaptio ALGEBRAIC COMBINATORICS LECTURE 8 TUESDAY, /6/06. Self-cojugate Partitios Recall that, give a partitio λ, we may

More information

It is always the case that unions, intersections, complements, and set differences are preserved by the inverse image of a function.

It is always the case that unions, intersections, complements, and set differences are preserved by the inverse image of a function. MATH 532 Measurable Fuctios Dr. Neal, WKU Throughout, let ( X, F, µ) be a measure space ad let (!, F, P ) deote the special case of a probability space. We shall ow begi to study real-valued fuctios defied

More information

Stochastic Matrices in a Finite Field

Stochastic Matrices in a Finite Field Stochastic Matrices i a Fiite Field Abstract: I this project we will explore the properties of stochastic matrices i both the real ad the fiite fields. We first explore what properties 2 2 stochastic matrices

More information

Machine Learning Theory Tübingen University, WS 2016/2017 Lecture 11

Machine Learning Theory Tübingen University, WS 2016/2017 Lecture 11 Machie Learig Theory Tübige Uiversity, WS 06/07 Lecture Tolstikhi Ilya Abstract We will itroduce the otio of reproducig kerels ad associated Reproducig Kerel Hilbert Spaces (RKHS). We will cosider couple

More information

Geometry of LS. LECTURE 3 GEOMETRY OF LS, PROPERTIES OF σ 2, PARTITIONED REGRESSION, GOODNESS OF FIT

Geometry of LS. LECTURE 3 GEOMETRY OF LS, PROPERTIES OF σ 2, PARTITIONED REGRESSION, GOODNESS OF FIT OCTOBER 7, 2016 LECTURE 3 GEOMETRY OF LS, PROPERTIES OF σ 2, PARTITIONED REGRESSION, GOODNESS OF FIT Geometry of LS We ca thik of y ad the colums of X as members of the -dimesioal Euclidea space R Oe ca

More information

Algebra of Least Squares

Algebra of Least Squares October 19, 2018 Algebra of Least Squares Geometry of Least Squares Recall that out data is like a table [Y X] where Y collects observatios o the depedet variable Y ad X collects observatios o the k-dimesioal

More information

Lecture 9: Expanders Part 2, Extractors

Lecture 9: Expanders Part 2, Extractors Lecture 9: Expaders Part, Extractors Topics i Complexity Theory ad Pseudoradomess Sprig 013 Rutgers Uiversity Swastik Kopparty Scribes: Jaso Perry, Joh Kim I this lecture, we will discuss further the pseudoradomess

More information

The random version of Dvoretzky s theorem in l n

The random version of Dvoretzky s theorem in l n The radom versio of Dvoretzky s theorem i l Gideo Schechtma Abstract We show that with high probability a sectio of the l ball of dimesio k cε log c > 0 a uiversal costat) is ε close to a multiple of the

More information

Lecture 3: August 31

Lecture 3: August 31 36-705: Itermediate Statistics Fall 018 Lecturer: Siva Balakrisha Lecture 3: August 31 This lecture will be mostly a summary of other useful expoetial tail bouds We will ot prove ay of these i lecture,

More information

A class of spectral bounds for Max k-cut

A class of spectral bounds for Max k-cut A class of spectral bouds for Max k-cut Miguel F. Ajos, José Neto December 07 Abstract Let G be a udirected ad edge-weighted simple graph. I this paper we itroduce a class of bouds for the maximum k-cut

More information

Entropy Rates and Asymptotic Equipartition

Entropy Rates and Asymptotic Equipartition Chapter 29 Etropy Rates ad Asymptotic Equipartitio Sectio 29. itroduces the etropy rate the asymptotic etropy per time-step of a stochastic process ad shows that it is well-defied; ad similarly for iformatio,

More information

Solution. 1 Solutions of Homework 1. Sangchul Lee. October 27, Problem 1.1

Solution. 1 Solutions of Homework 1. Sangchul Lee. October 27, Problem 1.1 Solutio Sagchul Lee October 7, 017 1 Solutios of Homework 1 Problem 1.1 Let Ω,F,P) be a probability space. Show that if {A : N} F such that A := lim A exists, the PA) = lim PA ). Proof. Usig the cotiuity

More information

Lecture 2: April 3, 2013

Lecture 2: April 3, 2013 TTIC/CMSC 350 Mathematical Toolkit Sprig 203 Madhur Tulsiai Lecture 2: April 3, 203 Scribe: Shubhedu Trivedi Coi tosses cotiued We retur to the coi tossig example from the last lecture agai: Example. Give,

More information

4.3 Growth Rates of Solutions to Recurrences

4.3 Growth Rates of Solutions to Recurrences 4.3. GROWTH RATES OF SOLUTIONS TO RECURRENCES 81 4.3 Growth Rates of Solutios to Recurreces 4.3.1 Divide ad Coquer Algorithms Oe of the most basic ad powerful algorithmic techiques is divide ad coquer.

More information

LECTURE 8: ORTHOGONALITY (CHAPTER 5 IN THE BOOK)

LECTURE 8: ORTHOGONALITY (CHAPTER 5 IN THE BOOK) LECTURE 8: ORTHOGONALITY (CHAPTER 5 IN THE BOOK) Everythig marked by is ot required by the course syllabus I this lecture, all vector spaces is over the real umber R. All vectors i R is viewed as a colum

More information

Lecture 4: Unique-SAT, Parity-SAT, and Approximate Counting

Lecture 4: Unique-SAT, Parity-SAT, and Approximate Counting Advaced Complexity Theory Sprig 206 Lecture 4: Uique-SAT, Parity-SAT, ad Approximate Coutig Prof. Daa Moshkovitz Scribe: Aoymous Studet Scribe Date: Fall 202 Overview I this lecture we begi talkig about

More information

The Borel hierarchy classifies subsets of the reals by their topological complexity. Another approach is to classify them by size.

The Borel hierarchy classifies subsets of the reals by their topological complexity. Another approach is to classify them by size. Lecture 7: Measure ad Category The Borel hierarchy classifies subsets of the reals by their topological complexity. Aother approach is to classify them by size. Filters ad Ideals The most commo measure

More information

Notes for Lecture 11

Notes for Lecture 11 U.C. Berkeley CS78: Computatioal Complexity Hadout N Professor Luca Trevisa 3/4/008 Notes for Lecture Eigevalues, Expasio, ad Radom Walks As usual by ow, let G = (V, E) be a udirected d-regular graph with

More information

Lesson 10: Limits and Continuity

Lesson 10: Limits and Continuity www.scimsacademy.com Lesso 10: Limits ad Cotiuity SCIMS Academy 1 Limit of a fuctio The cocept of limit of a fuctio is cetral to all other cocepts i calculus (like cotiuity, derivative, defiite itegrals

More information

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables Chapter 7 Limit theorems Throughout this sectio we will assume a probability space (Ω, F, P), i which is defied a ifiite sequece of radom variables (X ) ad a radom variable X. The fact that for every ifiite

More information

MA131 - Analysis 1. Workbook 3 Sequences II

MA131 - Analysis 1. Workbook 3 Sequences II MA3 - Aalysis Workbook 3 Sequeces II Autum 2004 Cotets 2.8 Coverget Sequeces........................ 2.9 Algebra of Limits......................... 2 2.0 Further Useful Results........................

More information

6 Integers Modulo n. integer k can be written as k = qn + r, with q,r, 0 r b. So any integer.

6 Integers Modulo n. integer k can be written as k = qn + r, with q,r, 0 r b. So any integer. 6 Itegers Modulo I Example 2.3(e), we have defied the cogruece of two itegers a,b with respect to a modulus. Let us recall that a b (mod ) meas a b. We have proved that cogruece is a equivalece relatio

More information

(3) If you replace row i of A by its sum with a multiple of another row, then the determinant is unchanged! Expand across the i th row:

(3) If you replace row i of A by its sum with a multiple of another row, then the determinant is unchanged! Expand across the i th row: Math 5-4 Tue Feb 4 Cotiue with sectio 36 Determiats The effective way to compute determiats for larger-sized matrices without lots of zeroes is to ot use the defiitio, but rather to use the followig facts,

More information

Assignment 5: Solutions

Assignment 5: Solutions McGill Uiversity Departmet of Mathematics ad Statistics MATH 54 Aalysis, Fall 05 Assigmet 5: Solutios. Let y be a ubouded sequece of positive umbers satisfyig y + > y for all N. Let x be aother sequece

More information

Largest families without an r-fork

Largest families without an r-fork Largest families without a r-for Aalisa De Bois Uiversity of Salero Salero, Italy debois@math.it Gyula O.H. Katoa Réyi Istitute Budapest, Hugary ohatoa@reyi.hu Itroductio Let [] = {,,..., } be a fiite

More information

MAT1026 Calculus II Basic Convergence Tests for Series

MAT1026 Calculus II Basic Convergence Tests for Series MAT026 Calculus II Basic Covergece Tests for Series Egi MERMUT 202.03.08 Dokuz Eylül Uiversity Faculty of Sciece Departmet of Mathematics İzmir/TURKEY Cotets Mootoe Covergece Theorem 2 2 Series of Real

More information

lim za n n = z lim a n n.

lim za n n = z lim a n n. Lecture 6 Sequeces ad Series Defiitio 1 By a sequece i a set A, we mea a mappig f : N A. It is customary to deote a sequece f by {s } where, s := f(). A sequece {z } of (complex) umbers is said to be coverget

More information

Math 216A Notes, Week 5

Math 216A Notes, Week 5 Math 6A Notes, Week 5 Scribe: Ayastassia Sebolt Disclaimer: These otes are ot early as polished (ad quite possibly ot early as correct) as a published paper. Please use them at your ow risk.. Thresholds

More information

Machine Learning for Data Science (CS 4786)

Machine Learning for Data Science (CS 4786) Machie Learig for Data Sciece CS 4786) Lecture 9: Pricipal Compoet Aalysis The text i black outlies mai ideas to retai from the lecture. The text i blue give a deeper uderstadig of how we derive or get

More information

A Proof of Birkhoff s Ergodic Theorem

A Proof of Birkhoff s Ergodic Theorem A Proof of Birkhoff s Ergodic Theorem Joseph Hora September 2, 205 Itroductio I Fall 203, I was learig the basics of ergodic theory, ad I came across this theorem. Oe of my supervisors, Athoy Quas, showed

More information

IP Reference guide for integer programming formulations.

IP Reference guide for integer programming formulations. IP Referece guide for iteger programmig formulatios. by James B. Orli for 15.053 ad 15.058 This documet is iteded as a compact (or relatively compact) guide to the formulatio of iteger programs. For more

More information

Week 5-6: The Binomial Coefficients

Week 5-6: The Binomial Coefficients Wee 5-6: The Biomial Coefficiets March 6, 2018 1 Pascal Formula Theorem 11 (Pascal s Formula For itegers ad such that 1, ( ( ( 1 1 + 1 The umbers ( 2 ( 1 2 ( 2 are triagle umbers, that is, The petago umbers

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013 MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013 Fuctioal Law of Large Numbers. Costructio of the Wieer Measure Cotet. 1. Additioal techical results o weak covergece

More information

Recitation 4: Lagrange Multipliers and Integration

Recitation 4: Lagrange Multipliers and Integration Math 1c TA: Padraic Bartlett Recitatio 4: Lagrage Multipliers ad Itegratio Week 4 Caltech 211 1 Radom Questio Hey! So, this radom questio is pretty tightly tied to today s lecture ad the cocept of cotet

More information