Approximation algorithms for trilinear optimization with nonconvex constraints and its extensions

Size: px
Start display at page:

Download "Approximation algorithms for trilinear optimization with nonconvex constraints and its extensions"

Transcription

1 Approximation agorithms for triinear optimization with nonconvex constraints and its extensions Yuning Yang Schoo of Mathematics Science and LPMC Nankai University Tianjin , P.R. China Qingzhi Yang Schoo of Mathematics Science and LPMC Nankai University Tianjin , P.R. China Apri 2, 2011 Abstract In this paper, we study triinear optimization probems with nonconvex constraints under some assumptions. We first consider the semidefinite reaxation (SDR) of the origina probem. Then motivated by So [3], we reduce the probem to that of determining the L 2-diameters of certain convex bodies, which can be approximatey soved in deterministic poynomia-time. After the reaxed probem being soved, the feasibe soution of the origina probem with a good approximation ratio can be obtained from the feasibe soution of the reaxed probem by state-of-art agorithms. Last we consider a cass of biquadratic optimization probems, which has a cose reationship with the triinear optimization probems. Key words: triinear optimization, biquadratic optimization, semidefinite reaxation, approximation soution, convex bodies. MSC 74B99, 15A18, 15A69 Corresponding author: Qingzhi Yang. This work was supported by the Nationa Natura Science Foundation of China (Grant No ) and Scientific Research Foundation for the Returned Overseas Chinese Schoars, State Education Ministry. 1

2 1 Introduction The poynomia optimization is a hot topic in the recent years and studied extensivey in the iterature, e.g., the mutiinear and homogeneous optimization over unit spheres [3, 15], quadratic constraints [15] and discrete sets [17]; the biquadratic optimization over unit spheres [5, 20, 21] and quadratic constraints [20, 6]; and other genera forms [12, 16]. In this paper, we study the triinear optimization of the form (T LP ) maximize Bxyu subject to x S R m, y T R n, u 2 = 1, u R where S R m and T R n are certain bounded nonconvex sets, respectivey. B = (b ijk ) R m n is a 3-th order tensor. (T LP ) generaizes the triinear optimization probems over unit spheres studied by [3], and over discrete sets studied by He et a [17]. We consider two cases of the constraints S and T in this paper: 1. S = {x x A i x 1, i = 0,..., p, } and T = {y y B j y 1, j = 1,..., q, }, where A i and B j are symmetric positive semidefinite matrices, i = 1,..., p, j = 1,..., q whie A 0 may be indefinite. 2. S = {x x 2 F 1 } and T = {y y 2 F 2 }, where F 1 and F 2 are some cosed convex subsets; The second case incudes some we-known probems, e.g., when F 1 and F 2 are the standard simpexes, it hods that m i=1 x2 i = 1 and n j=1 y2 j = 1. Then the probem is the triinear optimization with unit sphere constraints, which is consider by He et a [15] and So [3]; when F 1 = {e} and F 2 = {e} where e denotes the a-one vector, i.e., x { 1, 1} m and y { 1, 1} n, or, F 1 is the singeton point set and F 2 is the standard simpex, they are the mixed-variabes forms studied by He et a [17]. Thus we give a unified framework to treat these probems. The first case is reated to the biquadratic form studied by Zhang et a [20] under some assumptions, and we wi discuss it in the ast section. We wi show that the bounds obtained there can be improved by our approach. To sove the probem, we first consider the semidefinite reaxation of the origina probem. Then motivated by So [3], we reduce the reaxed probem to that of determining the L 2 -diameters of certain convex bodies, which can be approximatey soved in poynomia-time by using powerfu resuts from the agorithmic theory of convex bodies [24, 1], with an approximation ratio of Ω( ). We say a poynomia-time approximation agorithm appied to a probem P to find the maxima vaue P max has an approximation ratio α (0, 1], if can find a ower bound p, such that { P max p αp max if P max 0 αp P max p if P max < 0. It is worth pointing out that in [3], So used this approach to sove the mutiinear optimization probem over unit spheres, who got the best approximation ratio known to date. After the reaxed probem being soved, the feasibe soution of the origina probem with a good approximation ratio can be obtained from the feasibe soution of the reaxed probem by state-of-art agorithms. 2

3 The rest of this paper is organized as foows. In Section 2, we review the approach using by So [3] and introduce some concerned definitions, which makes the content more cear and sef-contained. In section 3 and Section 4, we discuss the two cases above. The reationship between the triinear form and the biquadratic form is studied in Section 5. We first add a comment on the notation that is used in the seque. Q denotes the rationa number fied. Vectors are written as owercase etters (x, y,...), matrices correspond to itaic capitas (A, B,...), and tensors are written as caigraphic capitas (A, B, ). The spaces of n-dimensiona rea vectors are denoted by R n, and the spaces of n n rea (symmetric) matrices are denoted by R n n (S n n ). For two rea matrices A and B with same dimension, A B stands for the usua matrix inner product, i.e., A B = tr(a B), where tr( ) denotes the trace of a matrix. In addition, A F denotes the Frobenius norm of A, i.e., A F = (A A) 1/2. The symbo. denotes the entry-wise production of two vectors with same dimension foowing that of MATLAB. For a vector x R n, x 2 = x. x. The symbo A B means that (A B) is in the positive semidefinite cone. v( ) denotes the maxima vaue of a probem. 2 A powerfu approach to sove the triinear optimization probem over unit spheres Let us review the approach proposed by So [3] to sove the triinear optimization probem over unit spheres. Briefy speaking, the probem is reduced to that of maximizing a certain norm over the sphere, which, by standard duaity arguments, is equivaent to determining the L 2 -diameter of a certain convex body. Let us reca the probem: (T LP ) maximize Bxyu subject to x 2 = y 2 = u 2 = 1, x R m, y R n, u R. Note that for any given u R, this probem is nothing but finding the argest singuar vaue of matrix B(u), where B(u) R m n is the matrix such that B(u) ij = B ijk u k. (2.1) k=1 Let B(u) 2 be the argest singuar vaue of B(u), then the maxima vaue of the origina probem is equa to max u 1 B(u) 2. As such, the author first showed that the function u u B B(u) 2 defines a norm on R. Then he caimed that the unit ba of the norm B given by B B = {u R u B 1} is a 0-centered, centray symmetric convex body. Moreover, this body is we-bounded [24], i.e. a rationa number R 1 is given expicity such that B B B 2(R 1 ) and a rationa number R 2 is given expicity such that B 2(R 2 ) B B, where B 2(R) denotes the -dimensiona Eucidean ba centered at the origin with radius R > 0. Note that if a convex body K satisfying the above conditions (n-dimensiona, a 0 -centered, we-bounded with radiuses R 1 and R 2 known expicity), then it can be denoted by a quintupe (K; n, R 1, R 2, a 0 ), see Definition of [24]. 3

4 Here we use (B B ;, R 1, R 2, 0) to denote B B. The we-boundness of B B pays an important roe in the whoe method. We wi mention it ater. Now consider the poar of B B, where the poar of a convex body K R is defined as K = {y R y x 1 for a x K}. It is known that if (K; n, R, r, 0) is a centray symmetric convex body, then so is (K ; n, 1/r, 1/R, 0), see e.g. [24] p By this property, we can use (B B ;, 1/R 2, 1/R 1, 0) to represent B B. The purpose of considering the poar of B B is that finding the maxima vaue of (T LP ) is equivaent to cacuating the L 2 -diameter of B B : v(t LP ) = max u 1 u B = max (max u v) u 1 v BB = max v B ( max u 1 u v) = max v B v 2 = 1 2 diam 2(B B) where diam 2 (K) represents the L 2 -diameter of a convex body K, i.e., diam 2 (K) = sup x,y K x y 2, where the ast equaity comes from the centray symmetry of BB. Thus the probem reduces to cacuate diam 2 (BB ). By the NP-hardness of (T LP ), we can not hope to cacuate it in poynomia-time. However, there are agorithms to approximate it in poynomia-time, see e.g. [1]. The foowing theorem points out that for a convex body K R n, diam 2 (K) can be og n approximated within a ratio Ω( n ) in poynomia-time if certain probems associated with K can be soved in poynomia-time: Theorem 2.1 (Theorem 2 of [3]) Given an integer n 1, one can construct in deterministic poynomia-time a centray symmetric poytope P R n such that 1. B n 2 (1) P B n 2 (O( n og n )) 2. for any we-bounded centray symmetric convex body K in R n, one has og n Ω( n ) diam 2(K) diam P (K) diam 2 (K), where diam P (K) is the diameter of K with respect to the poytopa norm P induced by P (i.e., for any x R n, one has x P = min{λ 0 x λp }, and P is the unit ba of the induced norm). Moreover, if K is equipped with a weak membership orace, then for any given rationa number ɛ > 0, the quantity diam P (K) can be computed to an accuracy of ɛ in deterministic orace-poynomia-time, and a vector x K(ɛ) is deivered with x P (1/2)diam P (K) ɛ. In this theorem, some definitions shoud be emphasized: An Orace can be regarded as a device that soves certain probems, ([1], p.26.) or simpy speaking, it is a subroutine (back box) that can be used by any agorithm ([24]). A Weak 4

5 Membership Orace soves the Weak Membership Probem (WMEM) ([1], p.51) for a body K R n : given a vector y Q n and a rationa number ɛ > 0, either (i) assert that y K(ɛ), or (ii) assert that y K( ɛ), where K(ɛ) and K( ɛ) are the outer ba and inner ba of K given respectivey by K(ɛ) = K + B n 2 (ɛ), K( ɛ) = {x R n x + B n 2 (ɛ) K}. Weak refers to the fact that we have to aow for a rounding error, since ony finite precision is avaiabe. An agorithm has orace-poynomia-time compexity if its runtime is poynomia in both the input size and the number of cas to the orace [3]. An important consequence is that any orace-poynomia-time agorithm can be turned into a genuiney poynomia agorithm for any situation in which the orace s action can be carried out in poynomia-time [1]. Thus Theorem 2.1 impies that if the WMEM of BB can be computed in poynomiatime, then diam P (BB ) can be approximated to arbitrary accuracy in poynomia-time. From [1], it is known that if we want to approximate diam P (BB ) to arbitrary accuracy in poynomia-time, a Weak Separation Orace associated with BB which soves the Weak Separation Probem (WSEP) shoud be given. One shoud note that if the WSEP associated with BB can be soved in poynomia-time, then the WMEM of B B can aso be soved in poynomia-time, but not vice versa. However, by the we-boundness of BB, these two probems are the same, in the sense of poynomia-time-compexity, see e.g., [24]. Actuay, by the we-boundness of K, and the reationship between K and its poar K, we ony need to check whether the WMEM associated with B B can be soved in poynomia-time. To see this, we need two other probems associated with a body K: Weak Optimization Probem (WOPT) and Weak Vaidity Probem (WVAL). The WOPT is the strongest one in these probems, since other probems can be soved in orace-poynomia-time if a body K is given by a weak optimization orace, but not vice versa. Nevertheess, the foowing emma shows the equivaence of WMEM and WOPT in the sense of poynomia-time-compexity, if the body is we-bounded: Lemma 2.1 (Coroary of [24]) There exists an orace-poynomia-time agorithm that soves the weak optimization probem for every centered body (K; n, R, r, a 0 ) given by a weak membership orace. These emma shows the equivaence of WOPT, WMEM and WVAL. The foowing emma shows the reationship between WVAL of K and WMEM of K Lemma 2.2 (Lemma of [24]) There exists an orace-poynomia-time agorithm that soves the weak membership probem for K, where K is a 0-centered convex body given by a weak vaidity orace. 5

6 Since B B and BB are 0-centered, centray symmetric and we-bounded, we can deduce the foowing emma Lemma 2.3 If the weak membership probem of B B can carry out its computation in poynomiatime, then so is the weak membership probem of its poar B B. Hence, it suffices to prove the poynomia-time-compexity of WMEM of B B for approximating diam 2 (BB ). For (T LP ), this condition hods. Finay, So showed that how to construct a feasibe soution of (T LP ) that attains the approximation ratio Ω( ). To begin, et ɛ > 0 be such that ɛ < γ 4 max i,j,k b ijk γ 4 v(t LP ), where γ 2 5 is the constant behind the Ω-notation in Theorem 2.1. By such theorem, we can in deterministic-poynomia-time compute a vector u BB (ɛ) such that u P 1 2 diam P (B B) ɛ. Then the author proved the foowing hods u 2 γ v(t LP ) ɛ. (2.2) Let u = u and (x, y) = arg max u Bxyu 2 x 2=1, y 2=1 we have (x, y, u) is feasibe for (T LP ) and Bxyu = u B. Moreover, it foows from the choice of ɛ and (2.2) that u ɛ u 2 u BB. Thus Bxyu = u B = max u v v BB u (u ɛ u) u 2 = u 2 ɛ γ v(t LP ). 2 This competes the whoe summary of the method of [3]. Not ike the SDR-based approach proposed by Luo and Zhang [23], and Ling et a [5], or the SOS method (cf. Lasserre [9, 10]), or other reaxation methods proposed by He et a [16, 15], So provides a totay different approach for the poynomia optimization probems over unit spheres, where the main toos come from the agorithmic theory of convex bodies, with the best-known approximation ratio to date. Looking backward, the foowing are some key points of this powerfu method Condition.1 1. For any fixed u R, the probem (T LP ) defines a norm on R. 2. The unit ba of the norm is 0-centered and we-bounded. 6

7 3. The WMEM associated with this unit ba can be soved in deterministic-poynomia-time. Naturay, we wi ask a question: can we appy this method to more genera constrained probems? E.g., the probem introduced in the beginning of this paper (T LP ) maximize Bxyu subject to x S R m, y T R n, u 2 = 1, u R where S and T may be some nonconvex bounded sets, or even discrete sets. We aways assume the probem under consideration has an optima soution. We sti use (T LP ) to denote this probem. Note that in the approach, we do not need to know the detai of the norm, which impies that if (T LP ) defines a norm on R, then computing v(t LP ) is aso equivaent to cacuating the L 2 -diameter of the unit ba of the norm. Furthermore, if the unit ba is webounded and the WMEM probem associated with the ba can be soved in poynomia-time, then we can appy the approach to (T LP ). The foowing theorem summarizes it Theorem 2.2 If (T LP ) satisfies Condition.1, then there is a deterministic-poynomia-time agorithm that, given an instance of (T LP ), returns a feasibe soution (x, y, u), such that Bxyu Ω( )v(t LP ). In the next section we wi consider two cases that satisfy these conditions. Here we give some other form of the objective function Bxyu for further anaysis. Let B(u) be defined in (2.1). Then Bxyu = B(u) xy = x B(u)y. Denote B R mn the unfoded matrix of B such that B (i 1)n+j,k = B ijk. Then Bxyu is equivaent to (x y) Bu, where denotes the Kronecker products such that x y = (x 1 y 1, x 1 y 2,..., x 1 y n, x 2 y 1,..., x 2 y n,..., x m y n ). We suppose mn in the foowing of this paper. 3 Quadratic constraints In this section we assume that S = {x R m x A i x 1, i = 0,..., p} and T = {y R n y B j y 1}, where A i, B j are a symmetric positive semidefinite matrices, i = 1,..., p, j = 1,..., q, whie A 0 may be indefinite. We assume that there exists u k 0, k = 0,..., p and v 0, = 1,..., q such that p q u i A i 0, v j B j 0. (3.3) i=0 The probem we consider in this section is j=1 (T LP ) maximize Bxyu = (x y) Bu (3.4) subject to x A i x 1, i = 0,..., p, y B j y 1, j = 1,..., q, u 2 = 1, u R 7

8 For a fixed u R, this mode reduces to the probem of maximizing a homogeneous quadratic form over quadratic constraints, which has been studied intensivey in the iterature [4, 18]. Even in this case, the probem is NP-hard [18], so we can not hope to find a maxima soution in poynomia-time. Instead, we want to use the approach of Section 2 to approximatey sove it. But we can not directy appy the approach on this probem. The reason wi be expained after Proposition 4.8. Instead, we wi consider its semidefinite reaxation. To begin with, et B(u) = 1 2 [ 0 B(u) B(u) 0 be a (m + n) (m + n) symmetric matrix and z = [x, y ], and et C i R (m+n) (m+n), i = 0,..., p + q be [ ] A i 0 if i = 0,..., p 0 0 C i = [ ] 0 0 if i = p + 1,..., p + q. 0 B (i p) Then (3.4) can be formuated as the homogeneous form with quadratic constraints The SDP reaxation is ] maximize z B(u)z (3.5) subject to z C i z 1, i = 0,..., p + q, u 2 = 1, u R. (RT LP ) maximize B(u) Z (3.6) subject to C i Z 1, i = 0,..., p + q, Z 0, u 2 = 1, u R. It is cear that v(rt LP ) v(t LP ). In what foows, we mainy prove that (RT LP ) satisfies Condition.1. To this end, we first show that the maxima vaue of (RT LP ) is bounded away from zero, where the bound is expicity known Proposition 3.1 If B = 0, then v(rt LP ) âˆb max ijk b ijk > 0, where â = 1 max0 i p max 1 k m (A i ) kk, ˆb 1 = max1 i q max 1 k n (B j ) kk. Proof. By (3.3), the denominator of â (resp. ˆb) must be positive, hence â and ˆb are wedefined. Without oss of generaity suppose b 111 = arg max b ijk. Let u = [sign(b 111 ), 0,..., 0], x = [a, 0,..., 0] and y = [b, 0,..., 0]. Then (x, y) S T and we have v(rt LP ) v(t LP ) Bxyu = âˆb max ijk b ijk > 0, as desired. For any fixed u, (RT LP ) is a we-formuated semidefinite programming g(u) := maximize B(u) Z (3.7) subject to C i Z 1, i = 0,..., p + q, Z 0. 8

9 By (3.3), it hods that p u i C i + i=0 q v j C j+p 0. (3.8) j=1 Under this condition, the SDP reaxation satisfies the dua Sater condition. Thus the primadua optima soutions exist and the prima-dua optima objective vaues are attainabe, and (3.7) can be soved by the interior-point method in poynomia-time. Now we caim that g(u) defines a norm on R Proposition 3.2 g(u) defines a norm on R. Proof. Firsty, for any u R, Bxyu is inear on x and y, hence we have max (x,y) S T Bxyu 0 for fixed u, which impies g(u) 0 for any u. Next we prove g(u) = g( u). Let Z = rz i=1 zi (z i ) be such that B(u) Z = g(u) where Z is feasibe for (3.7). For i = 1,..., r z, denote z i = [x i, y i ] where x i R m and y i R n satisfying z i = [x, y ]. Let Z = rz i=1 zi (z i ). Then we see that Z is aso feasibe for (4.11) and B( u) Z = g(u), which shows that g( u) g(u); using a simiar argument, we can aso deduce g(u) g( u). Thus we have g(u) = g( u) for any u R. Secondy, it is cear that g(α 1 u 1 + α 2 u 2 ) α 1 g(u 1 ) + α 2 g(u 2 ) for any u 1, u 2 α 1, α 2 0. Thus g(u) defines a semi-norm on R. R and Thirdy, denote L = {u g(u) = 0}. To compete the proof, we need to show that L = {0}. The equation g(u) = 0 impies that for any feasibe soution of (3.7), the objective vaue is identicay zero, which means that for any feasibe pair (x, y) S T, Bxyu = 0. If Bu 0 and without oss suppose (Bu) 1 0, then et x = [â sign((bu) 1 ), 0,..., 0], y = [ˆb, 0,..., 0] where â and ˆb are defined in Proposition 3.1. We have Bxyu > 0, which deduces a contradiction. Thus g(u) = 0 impies Bu = 0. This means that L is the nuspace of matrix B. As that of [3], armed with Proposition 3.1, we see that any optima soution (Z, u ) of (RT LP ) must satisfy u L where L R is the orthogona compement of L. With the assumption that mn, we can without oss assume that matrix B R mn has fu coumn rank, otherwise we can reduce the dimension of B and obtain a probem that is equivaent to (RT LP ). So Bu = 0 impies u = 0 and hence L = {0}. This competes the proof. Denote the norm by u B. It is cear that v(rt LP ) = max u 2=1 u B = max u 2 1 u B. Consider the unit ba of the norm u B given by B B = {u R u B 1}, which is 0-centered, centray symmetric and convex. As that of Section 2, computing v(rt LP ) is equivaent to cacuating the L 2 -diameter of the unit ba. The foowing concusion shows that B B is a we-bounded convex body, whose outer ba s radius and inner ba s radius are expicit known. Proposition 3.3 There exist rationa numbers 0 < C 1 C 2 <, whose encoding engths are poynomiay bounded by the input size of the probem, such that B 2(C 1 ) B B B 2(C 2 ). Proof. To prove this proposition, we first give a bound on the feasibe set of (3.7). Denote Ω = {Z R (m+n) (m+n) C i Z 1, i = 0,..., p + q, Z 0}. 9

10 Let C = p i=0 u ic i + q j=1 v jc j+p. From (3.8) we know that C 0 and we have Ω Ω := {Z R (m+n) (m+n) C Z p u i + i=0 q v j, Z 0}. By the symmetry of C, there is an orthogona matrix such that C = QΛQ where Λ is diagona and positive definite. Then it is equivaent that Ω = {Z R (m+n) (m+n) Λ Z p u i + i=0 j=1 q v j, Z 0}, which shows that for any Z Ω, we have m+n i=1 λ ic ii p i=0 u i + q j=1 v j, where λ i is an eigenvaue of C (it is aso the i-th diagona entry of Λ). By the positive definiteness of C and the positive semidefiniteness of Z, we have for any index i {1,..., m + n}, m+n λ min Z ii λ min i=1 Z ii Λ Z j=1 p u i + i=0 q v j, where λ min is the smaest eigenvaue of C. Thus we have for any Z Ω, Z ij Z ii Z jj ( p u i + i=0 j=1 q v j )/λ min, which is expicit bounded. Let e k be the k-th basis vector in R, k = 1,...,. Then we have 0 e k B = ( 1 i m,m+1 j m+n 1 i m,m+1 j m+n p u i + i=0 q v j ) j=1 j=1 B(e k ) ij Z ij = B ijk Zii Z jj 1 i m,1 j n where Z is a maximizer of (3.7). For k = 1,...,, denote c k = ( p u i + i=0 q v j ) j=1 1 i m,1 j n 1 i m,m+1 j m+n B ijk /λ min, B ijk /λ min, c max = max k=1 c k. Then e k 1 /c max B B. Let C 1 = (. We have c max) B 2(C 1 ) = {u R u 2 B ijk Z ij 1 ( c max ) } {u R u 1 1 } B B, c max where the first incusion comes from the fact that u 1 u 2 for any u R and the ast incusion comes from the equivaence between {u R u 1 1 c max } and the convex 1 hu c max {±e 1,..., ±e }. Thus we have proved that B B contains a ba with radius C 1, whose encoding ength is poynomia bounded by the input size of (RT LP ). On the other hand, for any u R \ {0}, it foows from Proposition 3.2 that B(u) 0. Suppose B(u) 11 = arg max B(u) ij 0. Let x = [sign(b(u) 11 )â, 0,..., 0], y = [ˆb, 0,..., 0] 10

11 where â and ˆb are defined in Proposition 3.1. Then we have (x, y) S T and obviousy u B B(u) xy λmin (B = âˆb Bu âˆb B) u 2. mn Since B R mn has fu coumn rank by the proof of Proposition 3.2 and mn, we have λ min (B B) > 0. Hence B B B2(C 2 ), where C 2 = (ab) mn/ λ min (B B). Thus we have proved B B is contained in a ba with radius C 2, whose encoding ength is aso bounded by the input size of (RT LP ). This competes the proof. The ony thing eft is to prove that the weak membership probem associated with B B can be soved in poynomia-time. We caim that this concusion hods Proposition 3.4 The weak membership probem associated with BB can be soved in deterministic poynomia-time. Proof. By the anaysis of Section 2, since B B is we-bounded, we ony need to prove that the weak membership probem associated with B B can be soved in deterministic poynomia-time. From the definition of the WMEM probem, we see that a key point to make the concusion hod is the poynomia-time computationa compexity of (3.7). For a given u Q \{0} and a rationa number ɛ > 0, we can use the interior-point method to compute a Z Ω in poynomia-time, such that u B ω(u) B(u) Z u B ɛ u 2, where Ω is the feasibe soution set of (3.7). Simiar to the proof of Proposition 3 of [3], if ω(u) > 1, then u B > 1, which impies u B B ( ɛ); if ω(u) 1, then u B B (ɛ). This competes the proof. We mention here that for fixed u R, since (T LP ) is NP-hard, it can not be soved in poynomia-time, which, by the proof above, does not meet the requirement of the poynomiatime sovabiity of the WMEM probem. That is why we must consider the reaxed probem of (T LP ). Now we have proved that Condition.1 hods for this case. Then we can in poynomia-time find a soution of (RT LP ) with ratio Ω( ). Foowing Section 2, we wi show that how to construct such a feasibe soution of (RT LP ). Let ɛ be chosen such that ɛ < γ âˆb max 4 b ijk < γ v(rt LP ), ijk 4 where â and ˆb are defined in Proposition 3.2. We can in deterministic poynomia-time compute a vector u BB (ɛ) such that u 2 γ v(rt LP ) ɛ and u ɛu/ u 2 B B. Let û = u u 2, and Z be a maximizer of (3.7) with respect to û. We 11

12 have that (û, Z ) is feasibe for (RT LP ) and B(û) Z = û B = max û v v BB u 2 ɛ γ v(rt LP ). 2 Thus we have in poynomia-time found a feasibe soution of (RT LP ) with approximation ratio Ω( ). The task eft is to extract a rank-one feasibe soution of (T LP ) from Z. Note again that for fixed u, it reduces to the probem studied by He et a [18], who used a randomized procedure to get a rank-one soution from a maximizer of the semidefinite reaxation probem. Since Z is a maximizer of the semidefinite probem (3.7) with respect to û, we can appy their method on our case. Combing their method and the previous anaysis, we have Theorem 3.3 There exists a deterministic poynomia-time agorithm that, given any instance of (RT LP ), returns a feasibe soution pair (û, Z ), such that B(û) Z Ω( ) v(rt LP ); where Z is a maximizer of (3.7) with respect to û. There exists a feasibe soution (x, y, û) of (T LP ) such that the probabiity Bxyû > B(û) Z α is at east (p+q)µe α/2, where α > 2 og[(174(p+q)µ] where µ = min{(p+q), max p+q i=0 rank(c iz )}. Moreover, we can find such a feasibe soution in randomized poynomia-time time. To sum up, we can in poynomia-time find a feasibe soution of (3.4) with approximation ratio Ω( og[175(p+q)u] ). Note that if A 0 is positive semidefinite, then foowing Nemirovski [4], the feasibe soution (x, y, û) of (T LP ) returned by the randomized agorithm satisfies Prob{Bxyû > B(û) Z } > 1 2(p + q)µ e α /2 α where α > 2 og[(2(p + q)µ ] where µ = min{(p + q), max p+q i=0 rank(c i)}. 4 cosed convex set In this section, we consider this case: S = {x R m x 2 F 1 }, T = {y R n y 2 F 2 }, where F 1 and F 2 are some cosed convex subsets. Through out this section, we assume the foowing hod for S and T. 12

13 Assumption There exist R 1 (m, n) > 0 and R 2 (m, n) > 0 such that F 1 B m 2 (R 1 ) and F 2 B n 2 (R 2 ), where R 1 and R 2 are poynomias with variabes m, n; 2. there exists a positive number β, for any given subindex pair (i, j), there exists (x, y) S T, such that x i β, y j β; 3. F 1 and F 2 are fu dimensiona, i.e., there exists at east a pair (x, y) S T such that x i 0 and y j 0, i = 1,..., m, j = 1,..., n; 4. R 1, R 2 and β are known expicity. These assumptions, athough seems restrictive, they do not excude some common cases. E.g. F 1 = {e}. In this case, S is exacty the binary set S = {1, 1} m ; F 1 = {y R m + m i=1 y i = 1}. In this case, S is exacty the unit sphere which reduces to that of [3]. It is easy to see these two case satisfy Assumption 4.1. The probem can be written as the foowing triinear form (T LP ) maximize Bxyu = B(u) xy = (x y) Bu (4.9) subject to x 2 F 1 R m, y 2 F 2 R n, u 2 = 1, u R Even for a fixed u R, this probem is sti NP-hard [19]. As the previous section, the WMEM associated with this probem can not be soved in poynomia-time. For this reason, we sti consider its semidefinite reaxation. To begin with, et B(u) = [ 0 B(u) 0 0 ] R (m+n) (m+n) and z = [x, y ] R m+n, F := F 1 F 2. (T LP ) can be rewritten as Its SDR is maximize B(u) zz (4.10) subject to z 2 F, u 2 = 1 (RT LP ) maximize B(u) Z subject to diag(z) F, Z 0, u 2 = 1 where diag(z) R m+n denotes the diagona entries of Z. Note that for any fixed u, (RT LP ) is a we-formuated convex optimization probem. Obviousy we have v(rt LP ) v(t LP ). Moreover, it hods that Proposition 4.5 v(rt LP ) v(t LP ) β 2 max ijk b ijk > 0, where β is given in Assumption 4.1. Proof. Without oss of generaity suppose b 111 = arg max ijk b ijk 0. Let u = [sign(b 111 ), 0,..., 0] R. 13

14 By Assumption 4.1 (2), there exists a pair (ˆx, ŷ) S T such that ˆx 1, ŷ 1 β. Let x = ˆx. [1, σ x ] and y = ŷ. [1, σ y ] where (σ x, σ y ) { 1, 1} (m 1) { 1, 1} (n 1). Traversing a the eements of { 1, 1} (m 1) { 1, 1} (n 1), we have (σ x,σ y) { 1,1} (m 1) { 1,1} (n 1) [1, σ x ] [1, σ y ] = [2 m+n 2, 0,..., 0] R m n, and so Bxyu = 2 m+n 2 b 111 x 1 y 1 2 m+n 2 β 2 b 111, x,y which means that there is at east a pair (x, y) S T satisfying Bxyu β 2 b 111 = β 2 max i,j,k b ijk. It foows that v(rt LP ) v(t LP ) β 2 max i,j,k b ijk > 0. Now, Denote g(u) := maximize B(u) Z (4.11) subject to diag(z) F, Z 0 To ensure the WMEM of the associated convex body can be soved in poynomia-time, we sha make an assumption Assumption 4.2 For any fixed u, (4.11) can be soved in poynomia-time by the interior-point method or the eipsoid method. Note that in many cases, this semidefinite probem can be soved by the interior-method or the eipsoid method in poynomia-time. We prove the foowing, which shows that (4.11) satisfies Condition.1 (1) Proposition 4.6 g(u) defines a norm on R. Proof. As that of Proposition 3.2, it is straightforward to show that g(u) 0 for any u R and g(α 1 u 1 + α 2 u 2 ) α 1 g(u 1 ) + α 2 g(u 2 ) for any u i R and α i 0, i = 1, 2. The ony nontrivia task is to show that g(u) = 0 ony if u = 0. Denote L = {u g(u) = 0}. As the previous section, we need to show that L = {0}. The equation g(u) = 0 aso means that for such u and for any (x, y) S T, Bxyu = 0. Under Assumption 4.1 (3), there exists a pair (ˆx, ŷ) such that a the components are not zero. Without oss of generaity suppose ˆx > 0, ŷ > 0. This impies that for a x = ˆx. σ x and y = ŷ. σ y where (σ x, σ y ) { 1, 1} m { 1, 1} n, the components of a these vectors are not zero. Since there are 2 m 1 ineary independent vectors in { 1, 1} m and 2 n 1 ineary independent vectors in { 1, 1} n, we have totay 2 m+n 2 ineary independent vectors of x y R mn. Reca that (x y) Bu 0 for a these x y, we have Bu 0, which means that L is the nuspace of matrix B. As that of Proposition 3.2, we can without oss of generaity suppose B has fu coumn rank, which deduces u = 0. Thus g(u) defines a norm on R, as needed. Denote the norm by u B. Consider the unit ba of the norm u B given by B B = {u R u B 1}, 14

15 We wi prove that B B is a we- which is aso 0-centered, centray symmetric and convex. bounded Proposition 4.7 There exist rationa numbers 0 < C 1 C 2 <, whose encoding engths are poynomiay bounded by the input size of the probem, such that B 2(C 1 ) B B B 2(C 2 ). Proof. Foowing the previous section, et e k be the k-th basis vector in R, k = 1,..., r. Then we have 0 e k B = B(e k ) ij Zij = B ijk Zij 1 i m,m+1 j m+n 1 i m,m+1 j m+n R 1 R 2 1 i m,1 j n B ijk Zii Z jj B ijk, 1 i m,m+1 j m+n where Z is a maximizer of (4.11) with respect to u = e k, and the ast inequaity comes from Assumption 4.1 (1). Denote c k = R 1 R 2 i,j B ijk, c max = max k=1 c 1 k and C 1 = (, we c max) have B2(C 1 ) = {u R 1 u 2 ( c max ) } {u R u 1 1 } B B. c max Thus we have proved that B B contains a ba with radius C 1, whose encoding ength is poynomia bounded by the input size of (T LP ). On the other hand, for any u R \ {0}, we have Bu R mn \ {0} by Proposition 4.6, which means that B(u) R m n \{0}. Without oss of generaity suppose B(u) 11 = arg max B(u) ij = arg max k=1 B ijku k. By Assumption 4.1 (2), there exists a pair (ˆx, ŷ) S T such that ˆx 1, ŷ 1 β > 0. Let x = ˆx. [1, σ x ] and y = ŷ. [1, σ y ] where (σ x, σ y ) { 1, 1} (m 1) { 1, 1} (n 1). As the proof of Proposition 4.5, there exists at east a pair (x, y) S T such that B(u) xy = (x y) Bu β 2 max i,j B(u) ij = β 2 Bu. Thus we have u B B(u) xy β 2 Bu β 2 λmin (B B) u 2. mn Since B R mn has fu coumn rank by Proposition 4.6 and mn, we have λ min (B B) > 0. Hence B B B2(C 2 ), where C 2 = mn β 2. Thus we have proved B λ min(b B is contained in a B) ba with radius C 2, whose encoding ength is aso bounded by the input size of (T LP ). We caim that Condition.1 (3) hods Proposition 4.8 The weak membership probem associated with BB can be soved in deterministicpoynomia-time. Proof. We ony need to prove that the weak membership probem associated with B B can be soved in deterministic poynomia-time. By Assumption 4.2, (4.11) can be sove in poynomiatime. For a given u Q \ {0} and a rationa number ɛ > 0 (without oss of generaity suppose ɛ < u 2 ), if (4.11) can be soved by the eipsoid method [25], then we can in deterministic poynomia-time compute a matrix Z R (m+n) (m+n), such that ɛ d(z, K) max{2 B(u) F, 1} u 2 15

16 and ω(u) B(u) Z u B ɛ u B ɛ, max{2 B(u) F, 1} u 2 u 2 where K := {Z R (m+n) (m+n) diag(z) F, Z 0} is the feasibe soution set of (4.11), where d(z, K) denotes the Eucidean distance between Z and K. If ω(u) 1, it foows from [3] that u B B (ɛ); if ω(u) > 1, by assumption, there exists a Z K such that Z Z F ɛ max{2 B(u) F,1} u 2. Then and so u + ɛ which impies u B B ( ɛ). u B B(u) Z = ω(u) + B(u) (Z Z) ɛ 1 B(u) F max{2 B(u) F, 1} u 2 1 ɛ 2 u 2 u u 2 B = (1 + ɛ u 2 ) u B (1 + ɛ u 2 )(1 ɛ 2 u 2 ) > 1, If (4.11) can be sove by the interior-point method in deterministic poynomia-time, then it foows from Proposition 4.8 that the concusion hods. This competes the proof. Now we have proved that Condition.1 hods for this case. Hence we can in poynomia-time find a soution of (RT LP ) with ratio Ω( ). Let ɛ be chosen such that ɛ < γ β 2 max 4 b ijk < γ v(rt LP ). ijk 4 Then we can in deterministic poynomia-time compute a vector u BB (ɛ) such that u 2 γ v(rt LP ) ɛ. Let û = u u 2, and Z be a maximizer of (4.11) with respect to û, we have that (û, Z ) is feasibe for (RT LP ) and B(û) Z = max û v v BB u 2 ɛ γ v(rt LP ). 2 Thus we have found a feasibe soution (û, Z ) of (RT LP ) with approximation ratio Ω( ) in poynomia-time. Then we must extract a z R m+n from Z such that z = [x, y ] is a feasibe soution of (4.9). Note that when u is given and F = {e}, (4.10) reduces to a quadratic integer programming, where Aon and Naor [14] provided a randomized poynomiatime agorithm to get a binary vector from Z, with approximation ratio ρ = 2 n(1+ 2) π > Actuay, their approach can be generaized to fit our case. Motivated by Zhang [19], et d = diag(z ) R m+n and X = (D ) + Z (D ) + + D, (4.12) 16

17 where d is the root of diagona entries of Z, D is the diagona matrix representation of d, (D ) + stands for the pseudo-inverse of D, i.e., it is a diagona and { (D ) + ii = (d i ) 1, d i > 0, 0, d i = 0, and D denotes a binary diagona matrix where D ii = 1 if Z ii = 0 and D ii = 0 otherwise. Then we have (d ) 2 F, diag(x ) = e, X 0 and Z = D X D where e denotes the a-one vector. Thus we can denote X = V V where V = [v 1,..., v m+n ] where v i are a unit vector, i = 1,..., m + n. To continue, we reca a emma given in [14] Lemma 4.4 (Lemma 4.2 of [14]) For any set {u i 1 i n} {v j 1 j m} of unit vectors in a Hibert space H, and for c = sinh 1 (1) = n(1 + 2), there is a set {u i 1 i n} {v j 1 j m} of unit vectors in a Hibert space H, such that if z is chosen randomy and uniformy in the unit sphere of H then for a 1 i n, 1 j m. π 2 E([sign(z u i)] [sign(z v j)]) = cu i v j (4.13) Based on this emma, given X R (m+n) (m+n) such that X = V V, we can find v 1,..., v m+n in another space in poynomia-time satisfying Lemma 4.4 as that of [14]. Let ξ be chosen randomy and uniformy in the unit sphere of the same space as v i and et x R m+n where x i = sign(ξ v i ), we have for i j E(x ix j) = 2c π v i v j = 2c π X ij. By noticing that the diagona entries of B(u) are zero, we get E(B(u) D x (x ) D ) = 2 n(1 + 2) π B(u) D X D = 2 n(1 + 2) B(u) Z. π Let [x, y ] = z = D x. Then (x, y) is a feasibe soution of (T LP ). Summarizing the whoe anaysis of this section, we get the foowing concusion Theorem 4.4 Suppose Assumption 4.1 and 4.2 hod. Then there is a deterministic poynomiatime agorithm that, given an instance of (RT LP ), returns a feasibe soution (û, Z ), such that B(û) Z Ω( )v(rt LP ); (4.14) there is a randomized poynomia-time agorithm that, given the feasibe soution (û, Z ) of (RT LP ) produced by the agorithm above, returns a feasibe soution (x, y, û) of (T LP ), such that E(Bxyû) = ρb(û) Z ρω( )v(t LP ), where ρ = 2 n(1+ 2) π > To sum up, we can in poynomia-time find a feasibe soution of (T LP ) with approximation ratio Ω( ). 17

18 We shoud point out that, if both F 1 and F 2 are the singeton point set {e}, or one of them is {e} whie another is the standard simpex, then (T LP ) reduces to the triinear mix-variabes forms studied by He et a [17]. If min{m, n}, then we improve their ratio from Ω( 1 ) to Ω( ). 5 The biquadratic case In this section, we mainy discuss a specia cass of biquadratic form (BQP ) maximize f(x, y) := Axxyy subject to x S, y T, where Axxyy = m i,j=1 n k,=1 A ijkx i x j y k y, A R m m n n is a 4-th order partiay symmetric tensor, i.e. a ijk = a jik = a ijk. The biquadratic optimization has many appications in quantum physics, see e.g., [2, 8, 7, 11, 13]. Some specia forms of (5.15) are studied in [5, 20, 6]. Now we expore the structure of A. Note that A can be unfoded into a matrix A R mn mn, with A (i 1)n+k,(j 1)n+ = A ijk. By such unfoding, A is a symmetric matrix. Denote its eigenvaue decomposition by A = r s=1 λ sη s (η s ) where r is the rank of A. In what foows, we make the foowing assumption Assumption 5.3 The unfoded matrix A is positive semidefinite. Note that a partiay symmetric tensor A is positive semidefinite if its unfoded matrix is positive semidefinite, but not vice versa, see, e.g., [21]. If the unfoded matrix of a partiay symmetric tensor A is positive semidefinite, then we ca A strongy positive semidefinite. Under this assumption, for s = 1,..., r, et b s = λ s η s. Then f(x, y) = Axxyy = r s=1 ((bs ) (x y)) 2. Denote matrix B = [b 1, b 2,..., b r ] R mn r. It is cear that the foowing proposition hods for B Proposition 5.9 B has fu coumn rank, i.e. rank(b) = r. From inear agebra, we immediatey have another concusion Proposition 5.10 For any u R r, Bu = 0 if and ony if u = 0. Denote 3-th order tensor B R m n r where B ijk = B (i 1)n+j,k. Then B is the foded tensor of matrix B. Now consider the foowing triinear form (T LP ) maximize g(x, y, u) := Bxyu = (x y) Bu subject to x S, y T, u 2 = 1 Actuay (T LP ) is equivaent to (BQP ). we state the proposition here, which is aso mentioned in He et a [15] 18

19 Proposition 5.11 v(bqp ) = v 2 (T LP ). Moreover, for any feasibe soution (x, y, u) of (T LP ), it hods that Axxyy (Bxyu) 2. Proof. Suppose (x, y, u) is a feasibe soution of (T LP ). For such given (x, y), u = Bxy/ Bxy 2 maximizes g(x, y, u) where (Bxy) R r with (Bxy) k = i,j=1 B ijkx i y j. Thus Bxy 2 = (Bxy) u = g(x, y, u) g(x, y, u). On the other hand, by the construction of B, we have Bxy 2 2 = r ((b s ) (x y)) 2 = Axxyy. s=1 Thus for any feasibe soution (x, y, u) of (T LP ), (x, y) is a feasibe soution pair of (BQP ) satisfying Axxyy (Bxyu) 2. Next suppose (x, y ) is an optima soution of (BQP ). By etting u = Bx y / Bx y 2, we have v(bqp ) = f(x, y ) = g 2 (x, y, u) v 2 (T LP ). Combining above, we have the concusion hods. By the anaysis above, when A is strongy positive semidefinite, we can reate the biquadratic optimization probem (BQP ) with the triinear optimization probem mainy studied in this paper. It is cear that the foowing theorem hods Theorem 5.5 Suppose A is strongy positive semidefinite. 1. If S = {x x A i x 1, i = 0,..., p, } and T = {y y B j y 1, j = 1,..., q, }, then there is a randomized poynomia-time agorithm that returns a feasibe soution (x, y) of (BQP ) og r with approximation ratio Ω( og 2 [175(p+q)u] r ); 2. if S = {x x 2 F 1 } and T = {y y 2 F 2 }, then there is a randomized poynomiatime agorithm that returns a feasibe soution (x, y) of (BQP ) with approximation ratio Ω( og r r ), where r is the rank of the unfoded matrix A R mn mn. Note that for case 1, Zhang et a [20] got an approximation ratio under some different assumptions ( ) 1 Ω (p + q) 2 (1 + 2 og(100p 2. )) og(100q) max{m, n}(max{m, n} 1) Thus in our case, we get a better approximation ratio than that. References [1] A. Brieden, P. Gritzmann, R. Kannan, V. Kee. L. Lovász and M. Simonovits, Deterministic and randomized poynomia-time approximation of radii, Mathematika 48 (2001)

20 [2] A. Einstein, B. Podosky and N. Rosen, Can quantum-mechanica description of physica reaity be considered compete? Phys.Rev. 47 (1935) [3] A. M. So, Deterministic approximation agorithms for sphere constrained homogeneous poynomia optimization probems, working paper, [4] A. Nemirovski, C. Roos and T. Teraky. On maximization of quadratic form over intersection of eipsoids with common center. Mathematica Programming, 86 (1999) [5] C. Ling, J. Nie, L. Qi and Y. Ye, Biquadratic optimization over unit spheres and semidefinite programming reaxations, SIAM J. Optim. 20 (2009) [6] C. Ling, X. Zhang and L. Qi, Semidefinite Reaxation Approximation for Mutivariate Bi-quadratic Optimization with Quadratic Constraints, preprint (2011). [7] D. Han, H. H. Dai, L. Qi, Conditions for strong eipticity of anisotropic eastic materias. J. East. 97 (2009) [8] G. Dah, J. M. Leinaas, J. Myrheim and E. Ovrum, A tensor product matrix aproximation probem in quantum physics. Linear Agebra App. 420 (2007) [9] J. B. Lasserre, Goba optimization with poynomias and the probem of moments. SIAM J. Optim. 11 (2001) [10] J. B. Lasserre, Poynomias nonnegative on a grid and discrete representations. Trans. Am. Math. Soc. 354 (2001) [11] J. K. Knowes and E. Sternberg, On the eipticity of the equations of the equations for finite eastostatics for a specia materia. J. East. 5 (1975) [12] J. Nie, An Exact Jacobian SDP Reaxation for Poynomia Optimization, (2010). [13] L. Qi, H. H. Dai and D. Han, Conditions for strong eipticity and M-eigenvaues. Front. Math. China 4 (2009) [14] N. Aon and A. Naor, Approximating the Cut-Norm via Grothendieck s Inequaity, SIAM J. Comput. 35 (2006) [15] S. He, Z. Li and S. Zhang, Approximation agorithms for homogeneous poynomia optimization with quadratic constraints, Math. Program. Ser. B 125 (2010) [16] S. He, Z. Li and S. Zhang, Genera Constrained Poynomia Optimization: an Approximation Approach, Technica Report SEEM , Department of Systems Engineering & Engineering Management, The Chinese University of Hong Kong, (2009). [17] S. He, Z. Li and S. Zhang, Approximation Agorithms for Discrete Poynomia Optimization, working paper, (2010). [18] S. He, Z-Q. Luo, J. Nie and S. Zhang, Semidefnite Reaxation Bounds for Indefinite Homogeneous Quadratic Optimization, SIAM J. Optim. 2 (2008) [19] S. Zhang, Quadratic maximization and semidefinite reaxation, Math. Program. Ser. A, 87 (2000),

21 [20] X. Zhang, C. Ling and L. Qi, Semidefinite reaxation bounds for bi-quadratic optimization probems with quadratic constraints, J. Gob. Optim. (2010) 49 (2010) [21] Y. Wang, L. Qi and X. Zhang, A practica method for computing the argest M-eigenvaue of a fourth-order partiay symmetric tensor, Numer. Linear Agebra App. 16 (2009) [22] Y. Wang and M. Aron, A reformuation of the strong eipticity conditions for unconstrained hypereastic media. J. East. 44 (1996) [23] Z. Luo and S. Zhang, A Semidefinite Reaxation Scheme for Mutivariate Quartic Poynomia Optimization With Quadratic Constraints, [24] M. Grötsche, L. Lovász and A. Schrijver, Geometric Agorithms and Combinatoria Optimization. Springer, Berin, (1988). [25] M. Grötsche, L. Lovász and A. Schrijver, The eipsoid method and its consequence in combinatoria optimization, Combinatorica 1 (1981)

Lecture Note 3: Stationary Iterative Methods

Lecture Note 3: Stationary Iterative Methods MATH 5330: Computationa Methods of Linear Agebra Lecture Note 3: Stationary Iterative Methods Xianyi Zeng Department of Mathematica Sciences, UTEP Stationary Iterative Methods The Gaussian eimination (or

More information

The Group Structure on a Smooth Tropical Cubic

The Group Structure on a Smooth Tropical Cubic The Group Structure on a Smooth Tropica Cubic Ethan Lake Apri 20, 2015 Abstract Just as in in cassica agebraic geometry, it is possibe to define a group aw on a smooth tropica cubic curve. In this note,

More information

Problem set 6 The Perron Frobenius theorem.

Problem set 6 The Perron Frobenius theorem. Probem set 6 The Perron Frobenius theorem. Math 22a4 Oct 2 204, Due Oct.28 In a future probem set I want to discuss some criteria which aow us to concude that that the ground state of a sef-adjoint operator

More information

Convergence Property of the Iri-Imai Algorithm for Some Smooth Convex Programming Problems

Convergence Property of the Iri-Imai Algorithm for Some Smooth Convex Programming Problems Convergence Property of the Iri-Imai Agorithm for Some Smooth Convex Programming Probems S. Zhang Communicated by Z.Q. Luo Assistant Professor, Department of Econometrics, University of Groningen, Groningen,

More information

Global Optimality Principles for Polynomial Optimization Problems over Box or Bivalent Constraints by Separable Polynomial Approximations

Global Optimality Principles for Polynomial Optimization Problems over Box or Bivalent Constraints by Separable Polynomial Approximations Goba Optimaity Principes for Poynomia Optimization Probems over Box or Bivaent Constraints by Separabe Poynomia Approximations V. Jeyakumar, G. Li and S. Srisatkunarajah Revised Version II: December 23,

More information

The Symmetric and Antipersymmetric Solutions of the Matrix Equation A 1 X 1 B 1 + A 2 X 2 B A l X l B l = C and Its Optimal Approximation

The Symmetric and Antipersymmetric Solutions of the Matrix Equation A 1 X 1 B 1 + A 2 X 2 B A l X l B l = C and Its Optimal Approximation The Symmetric Antipersymmetric Soutions of the Matrix Equation A 1 X 1 B 1 + A 2 X 2 B 2 + + A X B C Its Optima Approximation Ying Zhang Member IAENG Abstract A matrix A (a ij) R n n is said to be symmetric

More information

ORTHOGONAL MULTI-WAVELETS FROM MATRIX FACTORIZATION

ORTHOGONAL MULTI-WAVELETS FROM MATRIX FACTORIZATION J. Korean Math. Soc. 46 2009, No. 2, pp. 281 294 ORHOGONAL MLI-WAVELES FROM MARIX FACORIZAION Hongying Xiao Abstract. Accuracy of the scaing function is very crucia in waveet theory, or correspondingy,

More information

SUPPLEMENTARY MATERIAL TO INNOVATED SCALABLE EFFICIENT ESTIMATION IN ULTRA-LARGE GAUSSIAN GRAPHICAL MODELS

SUPPLEMENTARY MATERIAL TO INNOVATED SCALABLE EFFICIENT ESTIMATION IN ULTRA-LARGE GAUSSIAN GRAPHICAL MODELS ISEE 1 SUPPLEMENTARY MATERIAL TO INNOVATED SCALABLE EFFICIENT ESTIMATION IN ULTRA-LARGE GAUSSIAN GRAPHICAL MODELS By Yingying Fan and Jinchi Lv University of Southern Caifornia This Suppementary Materia

More information

Separation of Variables and a Spherical Shell with Surface Charge

Separation of Variables and a Spherical Shell with Surface Charge Separation of Variabes and a Spherica She with Surface Charge In cass we worked out the eectrostatic potentia due to a spherica she of radius R with a surface charge density σθ = σ cos θ. This cacuation

More information

Homogeneity properties of subadditive functions

Homogeneity properties of subadditive functions Annaes Mathematicae et Informaticae 32 2005 pp. 89 20. Homogeneity properties of subadditive functions Pá Burai and Árpád Száz Institute of Mathematics, University of Debrecen e-mai: buraip@math.kte.hu

More information

Mat 1501 lecture notes, penultimate installment

Mat 1501 lecture notes, penultimate installment Mat 1501 ecture notes, penutimate instament 1. bounded variation: functions of a singe variabe optiona) I beieve that we wi not actuay use the materia in this section the point is mainy to motivate the

More information

Algorithms to solve massively under-defined systems of multivariate quadratic equations

Algorithms to solve massively under-defined systems of multivariate quadratic equations Agorithms to sove massivey under-defined systems of mutivariate quadratic equations Yasufumi Hashimoto Abstract It is we known that the probem to sove a set of randomy chosen mutivariate quadratic equations

More information

Uniprocessor Feasibility of Sporadic Tasks with Constrained Deadlines is Strongly conp-complete

Uniprocessor Feasibility of Sporadic Tasks with Constrained Deadlines is Strongly conp-complete Uniprocessor Feasibiity of Sporadic Tasks with Constrained Deadines is Strongy conp-compete Pontus Ekberg and Wang Yi Uppsaa University, Sweden Emai: {pontus.ekberg yi}@it.uu.se Abstract Deciding the feasibiity

More information

CS229 Lecture notes. Andrew Ng

CS229 Lecture notes. Andrew Ng CS229 Lecture notes Andrew Ng Part IX The EM agorithm In the previous set of notes, we taked about the EM agorithm as appied to fitting a mixture of Gaussians. In this set of notes, we give a broader view

More information

An explicit Jordan Decomposition of Companion matrices

An explicit Jordan Decomposition of Companion matrices An expicit Jordan Decomposition of Companion matrices Fermín S V Bazán Departamento de Matemática CFM UFSC 88040-900 Forianópois SC E-mai: fermin@mtmufscbr S Gratton CERFACS 42 Av Gaspard Coriois 31057

More information

A Brief Introduction to Markov Chains and Hidden Markov Models

A Brief Introduction to Markov Chains and Hidden Markov Models A Brief Introduction to Markov Chains and Hidden Markov Modes Aen B MacKenzie Notes for December 1, 3, &8, 2015 Discrete-Time Markov Chains You may reca that when we first introduced random processes,

More information

Robust Sensitivity Analysis for Linear Programming with Ellipsoidal Perturbation

Robust Sensitivity Analysis for Linear Programming with Ellipsoidal Perturbation Robust Sensitivity Anaysis for Linear Programming with Eipsoida Perturbation Ruotian Gao and Wenxun Xing Department of Mathematica Sciences Tsinghua University, Beijing, China, 100084 September 27, 2017

More information

MATH 172: MOTIVATION FOR FOURIER SERIES: SEPARATION OF VARIABLES

MATH 172: MOTIVATION FOR FOURIER SERIES: SEPARATION OF VARIABLES MATH 172: MOTIVATION FOR FOURIER SERIES: SEPARATION OF VARIABLES Separation of variabes is a method to sove certain PDEs which have a warped product structure. First, on R n, a inear PDE of order m is

More information

Week 6 Lectures, Math 6451, Tanveer

Week 6 Lectures, Math 6451, Tanveer Fourier Series Week 6 Lectures, Math 645, Tanveer In the context of separation of variabe to find soutions of PDEs, we encountered or and in other cases f(x = f(x = a 0 + f(x = a 0 + b n sin nπx { a n

More information

Volume 13, MAIN ARTICLES

Volume 13, MAIN ARTICLES Voume 13, 2009 1 MAIN ARTICLES THE BASIC BVPs OF THE THEORY OF ELASTIC BINARY MIXTURES FOR A HALF-PLANE WITH CURVILINEAR CUTS Bitsadze L. I. Vekua Institute of Appied Mathematics of Iv. Javakhishvii Tbiisi

More information

BALANCING REGULAR MATRIX PENCILS

BALANCING REGULAR MATRIX PENCILS BALANCING REGULAR MATRIX PENCILS DAMIEN LEMONNIER AND PAUL VAN DOOREN Abstract. In this paper we present a new diagona baancing technique for reguar matrix pencis λb A, which aims at reducing the sensitivity

More information

Bourgain s Theorem. Computational and Metric Geometry. Instructor: Yury Makarychev. d(s 1, s 2 ).

Bourgain s Theorem. Computational and Metric Geometry. Instructor: Yury Makarychev. d(s 1, s 2 ). Bourgain s Theorem Computationa and Metric Geometry Instructor: Yury Makarychev 1 Notation Given a metric space (X, d) and S X, the distance from x X to S equas d(x, S) = inf d(x, s). s S The distance

More information

Primal and dual active-set methods for convex quadratic programming

Primal and dual active-set methods for convex quadratic programming Math. Program., Ser. A 216) 159:469 58 DOI 1.17/s117-15-966-2 FULL LENGTH PAPER Prima and dua active-set methods for convex quadratic programming Anders Forsgren 1 Phiip E. Gi 2 Eizabeth Wong 2 Received:

More information

JENSEN S OPERATOR INEQUALITY FOR FUNCTIONS OF SEVERAL VARIABLES

JENSEN S OPERATOR INEQUALITY FOR FUNCTIONS OF SEVERAL VARIABLES PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Voume 128, Number 7, Pages 2075 2084 S 0002-99390005371-5 Artice eectronicay pubished on February 16, 2000 JENSEN S OPERATOR INEQUALITY FOR FUNCTIONS OF

More information

Appendix of the Paper The Role of No-Arbitrage on Forecasting: Lessons from a Parametric Term Structure Model

Appendix of the Paper The Role of No-Arbitrage on Forecasting: Lessons from a Parametric Term Structure Model Appendix of the Paper The Roe of No-Arbitrage on Forecasting: Lessons from a Parametric Term Structure Mode Caio Ameida cameida@fgv.br José Vicente jose.vaentim@bcb.gov.br June 008 1 Introduction In this

More information

u(x) s.t. px w x 0 Denote the solution to this problem by ˆx(p, x). In order to obtain ˆx we may simply solve the standard problem max x 0

u(x) s.t. px w x 0 Denote the solution to this problem by ˆx(p, x). In order to obtain ˆx we may simply solve the standard problem max x 0 Bocconi University PhD in Economics - Microeconomics I Prof M Messner Probem Set 4 - Soution Probem : If an individua has an endowment instead of a monetary income his weath depends on price eves In particuar,

More information

Smoothers for ecient multigrid methods in IGA

Smoothers for ecient multigrid methods in IGA Smoothers for ecient mutigrid methods in IGA Cemens Hofreither, Stefan Takacs, Water Zuehner DD23, Juy 2015 supported by The work was funded by the Austrian Science Fund (FWF): NFN S117 (rst and third

More information

Maximizing Sum Rate and Minimizing MSE on Multiuser Downlink: Optimality, Fast Algorithms and Equivalence via Max-min SIR

Maximizing Sum Rate and Minimizing MSE on Multiuser Downlink: Optimality, Fast Algorithms and Equivalence via Max-min SIR 1 Maximizing Sum Rate and Minimizing MSE on Mutiuser Downink: Optimaity, Fast Agorithms and Equivaence via Max-min SIR Chee Wei Tan 1,2, Mung Chiang 2 and R. Srikant 3 1 Caifornia Institute of Technoogy,

More information

Integrating Factor Methods as Exponential Integrators

Integrating Factor Methods as Exponential Integrators Integrating Factor Methods as Exponentia Integrators Borisav V. Minchev Department of Mathematica Science, NTNU, 7491 Trondheim, Norway Borko.Minchev@ii.uib.no Abstract. Recenty a ot of effort has been

More information

arxiv: v1 [math.fa] 23 Aug 2018

arxiv: v1 [math.fa] 23 Aug 2018 An Exact Upper Bound on the L p Lebesgue Constant and The -Rényi Entropy Power Inequaity for Integer Vaued Random Variabes arxiv:808.0773v [math.fa] 3 Aug 08 Peng Xu, Mokshay Madiman, James Mebourne Abstract

More information

TARGET FOLLOWING ALGORITHMS FOR SEMIDEFINITE PROGRAMMING

TARGET FOLLOWING ALGORITHMS FOR SEMIDEFINITE PROGRAMMING TARGET FOLLOWING ALGORITHMS FOR SEMIDEFINITE PROGRAMMING CHEK BENG CHUA C & O Research Report: CORR 006 10 May 10, 006 Abstract. We present a target-foowing framework for semidefinite programming, which

More information

Research Article Numerical Range of Two Operators in Semi-Inner Product Spaces

Research Article Numerical Range of Two Operators in Semi-Inner Product Spaces Abstract and Appied Anaysis Voume 01, Artice ID 846396, 13 pages doi:10.1155/01/846396 Research Artice Numerica Range of Two Operators in Semi-Inner Product Spaces N. K. Sahu, 1 C. Nahak, 1 and S. Nanda

More information

Tight Approximation Algorithms for Maximum Separable Assignment Problems

Tight Approximation Algorithms for Maximum Separable Assignment Problems MATHEMATICS OF OPERATIONS RESEARCH Vo. 36, No. 3, August 011, pp. 416 431 issn 0364-765X eissn 156-5471 11 3603 0416 10.187/moor.1110.0499 011 INFORMS Tight Approximation Agorithms for Maximum Separabe

More information

MARKOV CHAINS AND MARKOV DECISION THEORY. Contents

MARKOV CHAINS AND MARKOV DECISION THEORY. Contents MARKOV CHAINS AND MARKOV DECISION THEORY ARINDRIMA DATTA Abstract. In this paper, we begin with a forma introduction to probabiity and expain the concept of random variabes and stochastic processes. After

More information

Higher dimensional PDEs and multidimensional eigenvalue problems

Higher dimensional PDEs and multidimensional eigenvalue problems Higher dimensiona PEs and mutidimensiona eigenvaue probems 1 Probems with three independent variabes Consider the prototypica equations u t = u (iffusion) u tt = u (W ave) u zz = u (Lapace) where u = u

More information

THE REACHABILITY CONES OF ESSENTIALLY NONNEGATIVE MATRICES

THE REACHABILITY CONES OF ESSENTIALLY NONNEGATIVE MATRICES THE REACHABILITY CONES OF ESSENTIALLY NONNEGATIVE MATRICES by Michae Neumann Department of Mathematics, University of Connecticut, Storrs, CT 06269 3009 and Ronad J. Stern Department of Mathematics, Concordia

More information

Efficiently Generating Random Bits from Finite State Markov Chains

Efficiently Generating Random Bits from Finite State Markov Chains 1 Efficienty Generating Random Bits from Finite State Markov Chains Hongchao Zhou and Jehoshua Bruck, Feow, IEEE Abstract The probem of random number generation from an uncorreated random source (of unknown

More information

A. Distribution of the test statistic

A. Distribution of the test statistic A. Distribution of the test statistic In the sequentia test, we first compute the test statistic from a mini-batch of size m. If a decision cannot be made with this statistic, we keep increasing the mini-batch

More information

Polyhedral results for a class of cardinality constrained submodular minimization problems

Polyhedral results for a class of cardinality constrained submodular minimization problems Poyhedra resuts for a cass of cardinaity constrained submoduar minimization probems Jiain Yu and Shabbir Ahmed Georgia Institute of Technoogy, Atanta, GA 30332 August 28, 2014 Abstract Motivated by concave

More information

(f) is called a nearly holomorphic modular form of weight k + 2r as in [5].

(f) is called a nearly holomorphic modular form of weight k + 2r as in [5]. PRODUCTS OF NEARLY HOLOMORPHIC EIGENFORMS JEFFREY BEYERL, KEVIN JAMES, CATHERINE TRENTACOSTE, AND HUI XUE Abstract. We prove that the product of two neary hoomorphic Hece eigenforms is again a Hece eigenform

More information

Physics 505 Fall Homework Assignment #4 Solutions

Physics 505 Fall Homework Assignment #4 Solutions Physics 505 Fa 2005 Homework Assignment #4 Soutions Textbook probems: Ch. 3: 3.4, 3.6, 3.9, 3.0 3.4 The surface of a hoow conducting sphere of inner radius a is divided into an even number of equa segments

More information

BASIC NOTIONS AND RESULTS IN TOPOLOGY. 1. Metric spaces. Sets with finite diameter are called bounded sets. For x X and r > 0 the set

BASIC NOTIONS AND RESULTS IN TOPOLOGY. 1. Metric spaces. Sets with finite diameter are called bounded sets. For x X and r > 0 the set BASIC NOTIONS AND RESULTS IN TOPOLOGY 1. Metric spaces A metric on a set X is a map d : X X R + with the properties: d(x, y) 0 and d(x, y) = 0 x = y, d(x, y) = d(y, x), d(x, y) d(x, z) + d(z, y), for a

More information

Approximation algorithms for nonnegative polynomial optimization problems over unit spheres

Approximation algorithms for nonnegative polynomial optimization problems over unit spheres Front. Math. China 2017, 12(6): 1409 1426 https://doi.org/10.1007/s11464-017-0644-1 Approximation algorithms for nonnegative polynomial optimization problems over unit spheres Xinzhen ZHANG 1, Guanglu

More information

Pricing Multiple Products with the Multinomial Logit and Nested Logit Models: Concavity and Implications

Pricing Multiple Products with the Multinomial Logit and Nested Logit Models: Concavity and Implications Pricing Mutipe Products with the Mutinomia Logit and Nested Logit Modes: Concavity and Impications Hongmin Li Woonghee Tim Huh WP Carey Schoo of Business Arizona State University Tempe Arizona 85287 USA

More information

Generalized Bell polynomials and the combinatorics of Poisson central moments

Generalized Bell polynomials and the combinatorics of Poisson central moments Generaized Be poynomias and the combinatorics of Poisson centra moments Nicoas Privaut Division of Mathematica Sciences Schoo of Physica and Mathematica Sciences Nanyang Technoogica University SPMS-MAS-05-43,

More information

6 Wave Equation on an Interval: Separation of Variables

6 Wave Equation on an Interval: Separation of Variables 6 Wave Equation on an Interva: Separation of Variabes 6.1 Dirichet Boundary Conditions Ref: Strauss, Chapter 4 We now use the separation of variabes technique to study the wave equation on a finite interva.

More information

Smoothness equivalence properties of univariate subdivision schemes and their projection analogues

Smoothness equivalence properties of univariate subdivision schemes and their projection analogues Numerische Mathematik manuscript No. (wi be inserted by the editor) Smoothness equivaence properties of univariate subdivision schemes and their projection anaogues Phiipp Grohs TU Graz Institute of Geometry

More information

VALIDATED CONTINUATION FOR EQUILIBRIA OF PDES

VALIDATED CONTINUATION FOR EQUILIBRIA OF PDES SIAM J. NUMER. ANAL. Vo. 0, No. 0, pp. 000 000 c 200X Society for Industria and Appied Mathematics VALIDATED CONTINUATION FOR EQUILIBRIA OF PDES SARAH DAY, JEAN-PHILIPPE LESSARD, AND KONSTANTIN MISCHAIKOW

More information

MIXING AUTOMORPHISMS OF COMPACT GROUPS AND A THEOREM OF SCHLICKEWEI

MIXING AUTOMORPHISMS OF COMPACT GROUPS AND A THEOREM OF SCHLICKEWEI MIXING AUTOMORPHISMS OF COMPACT GROUPS AND A THEOREM OF SCHLICKEWEI KLAUS SCHMIDT AND TOM WARD Abstract. We prove that every mixing Z d -action by automorphisms of a compact, connected, abeian group is

More information

Restricted weak type on maximal linear and multilinear integral maps.

Restricted weak type on maximal linear and multilinear integral maps. Restricted weak type on maxima inear and mutiinear integra maps. Oscar Basco Abstract It is shown that mutiinear operators of the form T (f 1,..., f k )(x) = R K(x, y n 1,..., y k )f 1 (y 1 )...f k (y

More information

XSAT of linear CNF formulas

XSAT of linear CNF formulas XSAT of inear CN formuas Bernd R. Schuh Dr. Bernd Schuh, D-50968 Kön, Germany; bernd.schuh@netcoogne.de eywords: compexity, XSAT, exact inear formua, -reguarity, -uniformity, NPcompeteness Abstract. Open

More information

Cryptanalysis of PKP: A New Approach

Cryptanalysis of PKP: A New Approach Cryptanaysis of PKP: A New Approach Éiane Jaumes and Antoine Joux DCSSI 18, rue du Dr. Zamenhoff F-92131 Issy-es-Mx Cedex France eiane.jaumes@wanadoo.fr Antoine.Joux@ens.fr Abstract. Quite recenty, in

More information

Akaike Information Criterion for ANOVA Model with a Simple Order Restriction

Akaike Information Criterion for ANOVA Model with a Simple Order Restriction Akaike Information Criterion for ANOVA Mode with a Simpe Order Restriction Yu Inatsu * Department of Mathematics, Graduate Schoo of Science, Hiroshima University ABSTRACT In this paper, we consider Akaike

More information

LECTURE NOTES 8 THE TRACELESS SYMMETRIC TENSOR EXPANSION AND STANDARD SPHERICAL HARMONICS

LECTURE NOTES 8 THE TRACELESS SYMMETRIC TENSOR EXPANSION AND STANDARD SPHERICAL HARMONICS MASSACHUSETTS INSTITUTE OF TECHNOLOGY Physics Department Physics 8.07: Eectromagnetism II October, 202 Prof. Aan Guth LECTURE NOTES 8 THE TRACELESS SYMMETRIC TENSOR EXPANSION AND STANDARD SPHERICAL HARMONICS

More information

A NOTE ON QUASI-STATIONARY DISTRIBUTIONS OF BIRTH-DEATH PROCESSES AND THE SIS LOGISTIC EPIDEMIC

A NOTE ON QUASI-STATIONARY DISTRIBUTIONS OF BIRTH-DEATH PROCESSES AND THE SIS LOGISTIC EPIDEMIC (January 8, 2003) A NOTE ON QUASI-STATIONARY DISTRIBUTIONS OF BIRTH-DEATH PROCESSES AND THE SIS LOGISTIC EPIDEMIC DAMIAN CLANCY, University of Liverpoo PHILIP K. POLLETT, University of Queensand Abstract

More information

LECTURE NOTES 9 TRACELESS SYMMETRIC TENSOR APPROACH TO LEGENDRE POLYNOMIALS AND SPHERICAL HARMONICS

LECTURE NOTES 9 TRACELESS SYMMETRIC TENSOR APPROACH TO LEGENDRE POLYNOMIALS AND SPHERICAL HARMONICS MASSACHUSETTS INSTITUTE OF TECHNOLOGY Physics Department Physics 8.07: Eectromagnetism II October 7, 202 Prof. Aan Guth LECTURE NOTES 9 TRACELESS SYMMETRIC TENSOR APPROACH TO LEGENDRE POLYNOMIALS AND SPHERICAL

More information

Formulas for Angular-Momentum Barrier Factors Version II

Formulas for Angular-Momentum Barrier Factors Version II BNL PREPRINT BNL-QGS-06-101 brfactor1.tex Formuas for Anguar-Momentum Barrier Factors Version II S. U. Chung Physics Department, Brookhaven Nationa Laboratory, Upton, NY 11973 March 19, 2015 abstract A

More information

arxiv: v1 [math.co] 17 Dec 2018

arxiv: v1 [math.co] 17 Dec 2018 On the Extrema Maximum Agreement Subtree Probem arxiv:1812.06951v1 [math.o] 17 Dec 2018 Aexey Markin Department of omputer Science, Iowa State University, USA amarkin@iastate.edu Abstract Given two phyogenetic

More information

On the Goal Value of a Boolean Function

On the Goal Value of a Boolean Function On the Goa Vaue of a Booean Function Eric Bach Dept. of CS University of Wisconsin 1210 W. Dayton St. Madison, WI 53706 Lisa Heerstein Dept of CSE NYU Schoo of Engineering 2 Metrotech Center, 10th Foor

More information

Iterative Decoding Performance Bounds for LDPC Codes on Noisy Channels

Iterative Decoding Performance Bounds for LDPC Codes on Noisy Channels Iterative Decoding Performance Bounds for LDPC Codes on Noisy Channes arxiv:cs/060700v1 [cs.it] 6 Ju 006 Chun-Hao Hsu and Achieas Anastasopouos Eectrica Engineering and Computer Science Department University

More information

VALIDATED CONTINUATION FOR EQUILIBRIA OF PDES

VALIDATED CONTINUATION FOR EQUILIBRIA OF PDES VALIDATED CONTINUATION FOR EQUILIBRIA OF PDES SARAH DAY, JEAN-PHILIPPE LESSARD, AND KONSTANTIN MISCHAIKOW Abstract. One of the most efficient methods for determining the equiibria of a continuous parameterized

More information

FRST Multivariate Statistics. Multivariate Discriminant Analysis (MDA)

FRST Multivariate Statistics. Multivariate Discriminant Analysis (MDA) 1 FRST 531 -- Mutivariate Statistics Mutivariate Discriminant Anaysis (MDA) Purpose: 1. To predict which group (Y) an observation beongs to based on the characteristics of p predictor (X) variabes, using

More information

Partial permutation decoding for MacDonald codes

Partial permutation decoding for MacDonald codes Partia permutation decoding for MacDonad codes J.D. Key Department of Mathematics and Appied Mathematics University of the Western Cape 7535 Bevie, South Africa P. Seneviratne Department of Mathematics

More information

Homework #04 Answers and Hints (MATH4052 Partial Differential Equations)

Homework #04 Answers and Hints (MATH4052 Partial Differential Equations) Homework #4 Answers and Hints (MATH452 Partia Differentia Equations) Probem 1 (Page 89, Q2) Consider a meta rod ( < x < ), insuated aong its sides but not at its ends, which is initiay at temperature =

More information

Semidefinite relaxation and Branch-and-Bound Algorithm for LPECs

Semidefinite relaxation and Branch-and-Bound Algorithm for LPECs Semidefinite reaxation and Branch-and-Bound Agorithm for LPECs Marcia H. C. Fampa Universidade Federa do Rio de Janeiro Instituto de Matemática e COPPE. Caixa Posta 68530 Rio de Janeiro RJ 21941-590 Brasi

More information

Jackson 4.10 Homework Problem Solution Dr. Christopher S. Baird University of Massachusetts Lowell

Jackson 4.10 Homework Problem Solution Dr. Christopher S. Baird University of Massachusetts Lowell Jackson 4.10 Homework Probem Soution Dr. Christopher S. Baird University of Massachusetts Lowe PROBLEM: Two concentric conducting spheres of inner and outer radii a and b, respectivey, carry charges ±.

More information

Efficient Generation of Random Bits from Finite State Markov Chains

Efficient Generation of Random Bits from Finite State Markov Chains Efficient Generation of Random Bits from Finite State Markov Chains Hongchao Zhou and Jehoshua Bruck, Feow, IEEE Abstract The probem of random number generation from an uncorreated random source (of unknown

More information

Reflection principles and kernels in R n _+ for the biharmonic and Stokes operators. Solutions in a large class of weighted Sobolev spaces

Reflection principles and kernels in R n _+ for the biharmonic and Stokes operators. Solutions in a large class of weighted Sobolev spaces Refection principes and kernes in R n _+ for the biharmonic and Stokes operators. Soutions in a arge cass of weighted Soboev spaces Chérif Amrouche, Yves Raudin To cite this version: Chérif Amrouche, Yves

More information

Selmer groups and Euler systems

Selmer groups and Euler systems Semer groups and Euer systems S. M.-C. 21 February 2018 1 Introduction Semer groups are a construction in Gaois cohomoogy that are cosey reated to many objects of arithmetic importance, such as cass groups

More information

Math 220B - Summer 2003 Homework 1 Solutions

Math 220B - Summer 2003 Homework 1 Solutions Math 0B - Summer 003 Homework Soutions Consider the eigenvaue probem { X = λx 0 < x < X satisfies symmetric BCs x = 0, Suppose f(x)f (x) x=b x=a 0 for a rea-vaued functions f(x) which satisfy the boundary

More information

An Extension of Almost Sure Central Limit Theorem for Order Statistics

An Extension of Almost Sure Central Limit Theorem for Order Statistics An Extension of Amost Sure Centra Limit Theorem for Order Statistics T. Bin, P. Zuoxiang & S. Nadarajah First version: 6 December 2007 Research Report No. 9, 2007, Probabiity Statistics Group Schoo of

More information

UNIFORM CONVERGENCE OF MULTIPLIER CONVERGENT SERIES

UNIFORM CONVERGENCE OF MULTIPLIER CONVERGENT SERIES royecciones Vo. 26, N o 1, pp. 27-35, May 2007. Universidad Catóica de Norte Antofagasta - Chie UNIFORM CONVERGENCE OF MULTILIER CONVERGENT SERIES CHARLES SWARTZ NEW MEXICO STATE UNIVERSITY Received :

More information

1D Heat Propagation Problems

1D Heat Propagation Problems Chapter 1 1D Heat Propagation Probems If the ambient space of the heat conduction has ony one dimension, the Fourier equation reduces to the foowing for an homogeneous body cρ T t = T λ 2 + Q, 1.1) x2

More information

Construction of Supersaturated Design with Large Number of Factors by the Complementary Design Method

Construction of Supersaturated Design with Large Number of Factors by the Complementary Design Method Acta Mathematicae Appicatae Sinica, Engish Series Vo. 29, No. 2 (2013) 253 262 DOI: 10.1007/s10255-013-0214-6 http://www.appmath.com.cn & www.springerlink.com Acta Mathema cae Appicatae Sinica, Engish

More information

C. Fourier Sine Series Overview

C. Fourier Sine Series Overview 12 PHILIP D. LOEWEN C. Fourier Sine Series Overview Let some constant > be given. The symboic form of the FSS Eigenvaue probem combines an ordinary differentia equation (ODE) on the interva (, ) with a

More information

THE THREE POINT STEINER PROBLEM ON THE FLAT TORUS: THE MINIMAL LUNE CASE

THE THREE POINT STEINER PROBLEM ON THE FLAT TORUS: THE MINIMAL LUNE CASE THE THREE POINT STEINER PROBLEM ON THE FLAT TORUS: THE MINIMAL LUNE CASE KATIE L. MAY AND MELISSA A. MITCHELL Abstract. We show how to identify the minima path network connecting three fixed points on

More information

Symbolic models for nonlinear control systems using approximate bisimulation

Symbolic models for nonlinear control systems using approximate bisimulation Symboic modes for noninear contro systems using approximate bisimuation Giordano Poa, Antoine Girard and Pauo Tabuada Abstract Contro systems are usuay modeed by differentia equations describing how physica

More information

2M2. Fourier Series Prof Bill Lionheart

2M2. Fourier Series Prof Bill Lionheart M. Fourier Series Prof Bi Lionheart 1. The Fourier series of the periodic function f(x) with period has the form f(x) = a 0 + ( a n cos πnx + b n sin πnx ). Here the rea numbers a n, b n are caed the Fourier

More information

Statistics for Applications. Chapter 7: Regression 1/43

Statistics for Applications. Chapter 7: Regression 1/43 Statistics for Appications Chapter 7: Regression 1/43 Heuristics of the inear regression (1) Consider a coud of i.i.d. random points (X i,y i ),i =1,...,n : 2/43 Heuristics of the inear regression (2)

More information

Explicit overall risk minimization transductive bound

Explicit overall risk minimization transductive bound 1 Expicit overa risk minimization transductive bound Sergio Decherchi, Paoo Gastado, Sandro Ridea, Rodofo Zunino Dept. of Biophysica and Eectronic Engineering (DIBE), Genoa University Via Opera Pia 11a,

More information

Research Article On the Lower Bound for the Number of Real Roots of a Random Algebraic Equation

Research Article On the Lower Bound for the Number of Real Roots of a Random Algebraic Equation Appied Mathematics and Stochastic Anaysis Voume 007, Artice ID 74191, 8 pages doi:10.1155/007/74191 Research Artice On the Lower Bound for the Number of Rea Roots of a Random Agebraic Equation Takashi

More information

#A48 INTEGERS 12 (2012) ON A COMBINATORIAL CONJECTURE OF TU AND DENG

#A48 INTEGERS 12 (2012) ON A COMBINATORIAL CONJECTURE OF TU AND DENG #A48 INTEGERS 12 (2012) ON A COMBINATORIAL CONJECTURE OF TU AND DENG Guixin Deng Schoo of Mathematica Sciences, Guangxi Teachers Education University, Nanning, P.R.China dengguixin@ive.com Pingzhi Yuan

More information

PSEUDO-SPLINES, WAVELETS AND FRAMELETS

PSEUDO-SPLINES, WAVELETS AND FRAMELETS PSEUDO-SPLINES, WAVELETS AND FRAMELETS BIN DONG AND ZUOWEI SHEN Abstract The first type of pseudo-spines were introduced in [1, ] to construct tight frameets with desired approximation orders via the unitary

More information

Math 124B January 17, 2012

Math 124B January 17, 2012 Math 124B January 17, 212 Viktor Grigoryan 3 Fu Fourier series We saw in previous ectures how the Dirichet and Neumann boundary conditions ead to respectivey sine and cosine Fourier series of the initia

More information

Throughput Rate Optimization in High Multiplicity Sequencing Problems

Throughput Rate Optimization in High Multiplicity Sequencing Problems Throughput Rate Optimization in High Mutipicity Sequencing Probems Aexander Grigoriev A.Grigoriev@KE.unimaas.n Maastricht University, 6200 MD, Maastricht, The Netherands Joris van de Kundert J.vandeKundert@MATH.unimaas.n

More information

Approximated MLC shape matrix decomposition with interleaf collision constraint

Approximated MLC shape matrix decomposition with interleaf collision constraint Approximated MLC shape matrix decomposition with intereaf coision constraint Thomas Kainowski Antje Kiese Abstract Shape matrix decomposition is a subprobem in radiation therapy panning. A given fuence

More information

Available online at ScienceDirect. IFAC PapersOnLine 50-1 (2017)

Available online at   ScienceDirect. IFAC PapersOnLine 50-1 (2017) Avaiabe onine at www.sciencedirect.com ScienceDirect IFAC PapersOnLine 50-1 (2017 3412 3417 Stabiization of discrete-time switched inear systems: Lyapunov-Metzer inequaities versus S-procedure characterizations

More information

SVM: Terminology 1(6) SVM: Terminology 2(6)

SVM: Terminology 1(6) SVM: Terminology 2(6) Andrew Kusiak Inteigent Systems Laboratory 39 Seamans Center he University of Iowa Iowa City, IA 54-57 SVM he maxima margin cassifier is simiar to the perceptron: It aso assumes that the data points are

More information

NOISE-INDUCED STABILIZATION OF STOCHASTIC DIFFERENTIAL EQUATIONS

NOISE-INDUCED STABILIZATION OF STOCHASTIC DIFFERENTIAL EQUATIONS NOISE-INDUCED STABILIZATION OF STOCHASTIC DIFFERENTIAL EQUATIONS TONY ALLEN, EMILY GEBHARDT, AND ADAM KLUBALL 3 ADVISOR: DR. TIFFANY KOLBA 4 Abstract. The phenomenon of noise-induced stabiization occurs

More information

(This is a sample cover image for this issue. The actual cover is not yet available at this time.)

(This is a sample cover image for this issue. The actual cover is not yet available at this time.) (This is a sampe cover image for this issue The actua cover is not yet avaiabe at this time) This artice appeared in a journa pubished by Esevier The attached copy is furnished to the author for interna

More information

POLYNOMIAL OPTIMIZATION WITH APPLICATIONS TO STABILITY ANALYSIS AND CONTROL - ALTERNATIVES TO SUM OF SQUARES. Reza Kamyar. Matthew M.

POLYNOMIAL OPTIMIZATION WITH APPLICATIONS TO STABILITY ANALYSIS AND CONTROL - ALTERNATIVES TO SUM OF SQUARES. Reza Kamyar. Matthew M. Manuscript submitted to AIMS Journas Voume X, Number 0X, XX 200X doi:10.3934/xx.xx.xx.xx pp. X XX POLYNOMIAL OPTIMIZATION WITH APPLICATIONS TO STABILITY ANALYSIS AND CONTROL - ALTERNATIVES TO SUM OF SQUARES

More information

An approximate method for solving the inverse scattering problem with fixed-energy data

An approximate method for solving the inverse scattering problem with fixed-energy data J. Inv. I-Posed Probems, Vo. 7, No. 6, pp. 561 571 (1999) c VSP 1999 An approximate method for soving the inverse scattering probem with fixed-energy data A. G. Ramm and W. Scheid Received May 12, 1999

More information

Asymptotic Properties of a Generalized Cross Entropy Optimization Algorithm

Asymptotic Properties of a Generalized Cross Entropy Optimization Algorithm 1 Asymptotic Properties of a Generaized Cross Entropy Optimization Agorithm Zijun Wu, Michae Koonko, Institute for Appied Stochastics and Operations Research, Caustha Technica University Abstract The discrete

More information

Approximate Bandwidth Allocation for Fixed-Priority-Scheduled Periodic Resources (WSU-CS Technical Report Version)

Approximate Bandwidth Allocation for Fixed-Priority-Scheduled Periodic Resources (WSU-CS Technical Report Version) Approximate Bandwidth Aocation for Fixed-Priority-Schedued Periodic Resources WSU-CS Technica Report Version) Farhana Dewan Nathan Fisher Abstract Recent research in compositiona rea-time systems has focused

More information

Stochastic Automata Networks (SAN) - Modelling. and Evaluation. Paulo Fernandes 1. Brigitte Plateau 2. May 29, 1997

Stochastic Automata Networks (SAN) - Modelling. and Evaluation. Paulo Fernandes 1. Brigitte Plateau 2. May 29, 1997 Stochastic utomata etworks (S) - Modeing and Evauation Pauo Fernandes rigitte Pateau 2 May 29, 997 Institut ationa Poytechnique de Grenobe { IPG Ecoe ationae Superieure d'informatique et de Mathematiques

More information

Approximation and Fast Calculation of Non-local Boundary Conditions for the Time-dependent Schrödinger Equation

Approximation and Fast Calculation of Non-local Boundary Conditions for the Time-dependent Schrödinger Equation Approximation and Fast Cacuation of Non-oca Boundary Conditions for the Time-dependent Schrödinger Equation Anton Arnod, Matthias Ehrhardt 2, and Ivan Sofronov 3 Universität Münster, Institut für Numerische

More information

A BUNDLE METHOD FOR A CLASS OF BILEVEL NONSMOOTH CONVEX MINIMIZATION PROBLEMS

A BUNDLE METHOD FOR A CLASS OF BILEVEL NONSMOOTH CONVEX MINIMIZATION PROBLEMS SIAM J. OPTIM. Vo. 18, No. 1, pp. 242 259 c 2007 Society for Industria and Appied Mathematics A BUNDLE METHOD FOR A CLASS OF BILEVEL NONSMOOTH CONVEX MINIMIZATION PROBLEMS MIKHAIL V. SOLODOV Abstract.

More information

Minimizing Total Weighted Completion Time on Uniform Machines with Unbounded Batch

Minimizing Total Weighted Completion Time on Uniform Machines with Unbounded Batch The Eighth Internationa Symposium on Operations Research and Its Appications (ISORA 09) Zhangiaie, China, September 20 22, 2009 Copyright 2009 ORSC & APORC, pp. 402 408 Minimizing Tota Weighted Competion

More information

CONJUGATE GRADIENT WITH SUBSPACE OPTIMIZATION

CONJUGATE GRADIENT WITH SUBSPACE OPTIMIZATION CONJUGATE GRADIENT WITH SUBSPACE OPTIMIZATION SAHAR KARIMI AND STEPHEN VAVASIS Abstract. In this paper we present a variant of the conjugate gradient (CG) agorithm in which we invoke a subspace minimization

More information

Math-Net.Ru All Russian mathematical portal

Math-Net.Ru All Russian mathematical portal Math-Net.Ru A Russian mathematica porta D. Zaora, On properties of root eements in the probem on sma motions of viscous reaxing fuid, Zh. Mat. Fiz. Ana. Geom., 217, Voume 13, Number 4, 42 413 DOI: https://doi.org/1.1547/mag13.4.42

More information