Introduction Quality criteria Recent constructions Sequences Final remarks A Belgian view on lattice rules Ronald Cools Dept. of Computer Science, KU Leuven Linz, Austria, October 14–18, 2013
Introduction Quality criteria Recent constructions Sequences Final remarks Atomium – Brussels built in 1958 height ≈ 103 m figure = 2 e coin 5 · 10 6 in circulation Body centered cubic lattice
Introduction Quality criteria Recent constructions Sequences Final remarks Introduction Given is an integral � I [ f ] := w ( x ) f ( x ) d x Ω where Ω ⊆ R s and w ( x ) ≥ 0 , ∀ x ∈ R s . Search an approximation for I [ f ] n � w j f ( y ( j ) ) I [ f ] ≃ Q [ f ] := j =1 with w j ∈ R and y ( j ) ∈ R s . Webster: quadrature: the process of finding a square equal in area to a given area. cubature: the determination of cubic contents. If s = 1 then Q is called a quadrature formula. If s ≥ 2 then Q is called a cubature formula.
Introduction Quality criteria Recent constructions Sequences Final remarks n � w j f ( y ( j ) ) Q [ f ] := j =1 Cubature/quadrature formulas are basic integration rules → choose points y ( j ) and weights w j independent of integrand f . It is difficult (time consuming) to construct basic integration rules, but the result is usually hard coded in programs or tables.
Introduction Quality criteria Recent constructions Sequences Final remarks n � w j f ( y ( j ) ) Q [ f ] := j =1 Cubature/quadrature formulas are basic integration rules → choose points y ( j ) and weights w j independent of integrand f . It is difficult (time consuming) to construct basic integration rules, but the result is usually hard coded in programs or tables. Restriction to unit cube: given is � 1 � 1 � I [ f ] = · · · f ( x 1 , . . . , x s )d x 1 · · · d x s = [0 , 1) s f ( x )d x 0 0
Introduction Quality criteria Recent constructions Sequences Final remarks Taxonomy: two major classes polynomial based methods 1 incl. methods exact for algebraic or trigonometric polynomials number theoretic methods 2 incl. Monte Carlo and quasi-Monte Carlo methods As in zoology, some species are difficult to classify.
Introduction Quality criteria Recent constructions Sequences Final remarks Taxonomy: two major classes polynomial based methods 1 incl. methods exact for algebraic or trigonometric polynomials number theoretic methods 2 incl. Monte Carlo and quasi-Monte Carlo methods As in zoology, some species are difficult to classify. For example Definition An s -dimensional lattice rule is a cubature formula which can be expressed in the form d 1 d 2 d t �� j 1 z 1 �� 1 + j 2 z 2 + . . . + j t z t � � � Q [ f ] = . . . f , d 1 d 2 . . . d t d 1 d 2 d t j 1 =1 j 2 =1 j t =1 where d i ∈ N 0 and z i ∈ Z s for all i .
Introduction Quality criteria Recent constructions Sequences Final remarks Alternative formulation: Definition A multiple integration lattice Λ is a subset of R s which is discrete and closed under addition and subtraction and which contains Z s as a subset. Definition A lattice rule is a cubature formula where the n points are the points of a multiple integration lattice Λ that lie in [0 , 1) s and the weights are all equal to 1 /n . n = n ( Q ) = # { Λ ∩ [0 , 1) s } .
Introduction Quality criteria Recent constructions Sequences Final remarks Example The Fibonnaci lattice with n = F j and z = (1 , F j − 1 ) � � has points x ( j ) = F j , jF j − 1 j F j n − 1 ⇒ lattice rule Q [ f ] = 1 �� ( j, jF j − 1 ) �� � f n n j =0 Example: the lattice rule with n = d 1 = F 7 = 13 and z 1 = (1 , 8) 1 s 0.5 0 s 0 0.5 1
Introduction Quality criteria Recent constructions Sequences Final remarks Example The Fibonnaci lattice with n = F j and z = (1 , F j − 1 ) � � has points x ( j ) = F j , jF j − 1 j F j n − 1 ⇒ lattice rule Q [ f ] = 1 �� ( j, jF j − 1 ) �� � f n n j =0 Example: the lattice rule with n = d 1 = F 7 = 13 and z 1 = (1 , 8) 1 s 0.5 s 0 s 0 0.5 1
Introduction Quality criteria Recent constructions Sequences Final remarks Example The Fibonnaci lattice with n = F j and z = (1 , F j − 1 ) � � has points x ( j ) = F j , jF j − 1 j F j n − 1 ⇒ lattice rule Q [ f ] = 1 �� ( j, jF j − 1 ) �� � f n n j =0 Example: the lattice rule with n = d 1 = F 7 = 13 and z 1 = (1 , 8) 1 s s 0.5 s 0 s 0 0.5 1
Introduction Quality criteria Recent constructions Sequences Final remarks Example The Fibonnaci lattice with n = F j and z = (1 , F j − 1 ) � � has points x ( j ) = F j , jF j − 1 j F j n − 1 ⇒ lattice rule Q [ f ] = 1 �� ( j, jF j − 1 ) �� � f n n j =0 Example: the lattice rule with n = d 1 = F 7 = 13 and z 1 = (1 , 8) 1 s s s s s s 0.5 s s s s s s 0 s 0 0.5 1
Introduction Quality criteria Recent constructions Sequences Final remarks Polynomials Let α = ( α 1 , α 2 , . . . , α s ) ∈ Z s and | α | := � s j =1 | α j | . algebraic polynomial s a α x α = x α j � � � p ( x ) = a α j , with α j ≥ 0 j =1 trigonometric polynomial s a α e 2 πiα · x = � � � e 2 πix j α j t ( x ) = a α j =1 The degree of a polynomial = max a α � =0 | α | . P s d = all algebraic polynomials in s variables of degree at most d . T s d = all trigonometric polynomials in s variables of degree at most d .
Introduction Quality criteria Recent constructions Sequences Final remarks Quality criteria? Definition A cubature formula Q for an integral I has algebraic (trigonometric) degree d if it is exact for all polynomials of algebraic (trigonometric) degree at most d .
Introduction Quality criteria Recent constructions Sequences Final remarks Quality criteria? Definition A cubature formula Q for an integral I has algebraic (trigonometric) degree d if it is exact for all polynomials of algebraic (trigonometric) degree at most d . How many points are needed in a cubature formula to obtain a specified degree of precision?
Introduction Quality criteria Recent constructions Sequences Final remarks The dimensions of the vector spaces of polynomials are: � s + d � dim P s d = d s � � � � s d dim T s � 2 j . d = j j j =0 We will use the symbol V s d to refer to one of the vector spaces P s d or T s d .
Introduction Quality criteria Recent constructions Sequences Final remarks Theorem If a cubature formula is exact for all polynomials of V s 2 k , then the number of points n ≥ dim V s k . Algebraic degree: For s = 2 (Radon, 1948); general s (Stroud, 1960) Trigonometric degree: (Mysovskikh, 1987) A. Stroud ■✳P✳ ▼②s♦✈s❦✐❤ J. Radon
Introduction Quality criteria Recent constructions Sequences Final remarks Theorem If a cubature formula is exact for all polynomials of degree d > 0 and has only real points and weights, then it has at least dim V s k positive weights, k = ⌊ d 2 ⌋ . Algebraic degree: (Mysovskikh, 1981) Trigonometric degree: (C. 1997) ⇒ minimal formulas have only positive weights. Corollary If a cubature formula of trigonometric degree 2 k has n = dim T s k points, then all weights are equal. This is a reason to restrict searches to n Q [ f ] = 1 � f ( x j ) . n j =1
Introduction Quality criteria Recent constructions Sequences Final remarks Improved bound for odd degrees For algebraic degree, the improved lower bound for odd degrees takes into account the symmetry of the integration region. E.g., centrally symmetric regions such as a cube → (M¨ oller, 1973) H.M. M¨ oller Result for trigonometric degree is very similar.
Introduction Quality criteria Recent constructions Sequences Final remarks Improved bound for odd degrees G k := span of trigonometric monomials of degree ≤ k with the same parity as k . Theorem ((Noskov, 1985), (Mysovskikh, 1987)) The number of points n of a cubature formula for the integral over [0 , 1) s which is exact for all trigonometric polynomials of degree at most d = 2 k + 1 satisfies n ≥ 2 dim G k .
Introduction Quality criteria Recent constructions Sequences Final remarks Definition A cubature formula is called shift symmetric if it is invariant w.r.t. the group of transformations � x �→ x , x �→ { x + (1 2 , . . . , 1 � 2) } ( This is the ‘central symmetry’ for the trig. case. ) Theorem (Beckers & C., 1993) If a shift symmetric cubature formula of degree 2 k + 1 has n = 2 dim G k points, then all weights are equal. Conjecture (C., 1997) Any cubature formula that attains the lower bound is shift symmetric. This became a Theorem (Osipov, 2001).
Introduction Quality criteria Recent constructions Sequences Final remarks Known minimal formulas for trigonometric degree for all s degree 1 degree 2 (Noskov, 1988) degree 3 (Noskov, 1988) for s = 2 all even degrees (Noskov, 1988) all odd degrees (Reztsov, 1990) (Beckers & C., 1993) (C. & Sloan, 1996) for s = 3 degree 5 (Frolov, 1977) M. Beckers ❆✳ ❘❡③❝♦✈ I.H. Sloan ▼✳❱✳ ◆♦s❦♦✈
Introduction Quality criteria Recent constructions Sequences Final remarks All known minimal formulas of trigonometric degree are lattice rules, except...
Recommend
More recommend
