Mapping Techniques for Isogeometric Analysis of Elliptic Boundary Value Problems Containing Singularities

Size: px
Start display at page:

Download "Mapping Techniques for Isogeometric Analysis of Elliptic Boundary Value Problems Containing Singularities"

Transcription

1 Mapping Techniques for Isogeometric Analysis of Elliptic Boundary Value Problems Containing Singularities Jae Woo Jeong Department of Mathematics, Miami University, Hamilton, OH 450 Hae-Soo Oh Department of Mathematics and Statistics, University of North Carolina at Charlotte, Charlotte, NC Sunbu Kang Korea Air Force Academy, Cheongwon, Chungbuk, South Korea and University of North Carolina at Charlotte, Charlotte, NC Hyunju Kim Department of Mathematics and Statistics, University of North Carolina at Charlotte, Charlotte, NC Comput. Methods Appl. Mech. Engrg. (202), Abstract The Method of Auxiliary Mapping (MAM), introduced by Babuška and Oh [2], is an effective method for dealing with singularities in elasticity [6]. MAM was extended to boundary element method (BEM) in the framework of mesh free particle methods [7], reproducing polynomial particle methods [20], and also to infinite domain problems [8]. Similarly, we consider NURBS geometrical mappings that are able to generate crack singularities for isogeometric analysis of elliptic boundary value problems. However, the mapping techniques proposed in this paper are different from MAM. In order to generate singular shape functions, MAM uses conformal mappings that locally change the physical domain, whereas the NURBS mappings used for design of engineering system are not allowed to alter the physical domain for isogeometric analysis. Moreover, unlike MAM, the proposed method makes it possible to independently control the radial and angular direction of the function to be approximated as far as the point singularities are concerned. We prove error estimates in Sobolev norms and demonstrate that the proposed mapping technique is highly effective for isogeometric analysis of elliptic boundary value problems with singularities. Mesh refinements to deal with singularities are compared with the mapping technique in the isogeometic analysis framework. Corresponding author. Tel.: ; Fax: ; Supported in part by NSF grant DMS

2 Keywords: Isogeometric analysis; B-spline basis functions; NURBS geometrical mapping; Knot vectors; Control points; Mapping techniques. Introduction NURBS (non-uniform rational B-spline) basis functions that are non interpolant are tools for engineering designs, whereas piecewise polynomial basis functions that are interpolant are shape functions for finite element analysis. Thus, there are barriers between computer aided design (CAD) and finite element analysis (h-fem [5, 9] and p-fem [26]) that make it difficult to synthesize them together. Most recently, introducing NURBS basis functions to finite element analysis, Hughes et al. [0] combined two tasks (engineering design and analysis) into one process. They coined this concept isogeometric analysis. Since then quite a few papers ([3, 7] and references within) have been published to show the theories as well as the computations where isogeometric finite element analysis is a powerful new approach. In the literature, X-FEM in which linear finite element basis functions are enriched with singular functions were successful in dealing with singularity problems. Notably, X-FEM is built into isogeometric analysis of elastic domain with cracks [8]. In order to handle the singularities arising in the partial differential equations, Babuška and Oh [2] introduced the mapping techniques called the Method of Auxiliary Mapping (MAM) into conventional p-fem. It was proved that MAM is successful in dealing with singularities in elasticity [3, 4, 6], oscillating singularities [22], and delamination crack singularities [5]. Recently, Oh et al. extended MAM to the meshfree particle methods [9, 20], the boundary element method in the framework of reproducing polynomial particle methods [7], and also to infinite domain problems [8, 23]. However, MAM has no relevance in the h-fem that uses linear basis functions. In a similar spirit to MAM, we introduce mapping techniques into isogeometric analysis to deal with point singularities (cracks and jump boundary data) that arise in elliptic boundary value problems. The proposed mapping technique implemented in isogeometric analysis is different from MAM in the following aspects: MAM uses conformal mappings that locally change the physical domain (possibly altering design). In contrast, the proposed mapping technique uses NURBS geometrical mappings with proper selection of control points and hence does not alter the system design. Unlike MAM, this mapping technique makes it possible to independently control the radial and angular direction of the function to be approximated as far as the point singularities are concerned. The proposed mapping method in the framework of isogeometric analysis yields the same accuracy as MAM in p-fem at a much lower degree of freedom. This is actually a salient feature of isogeometric analysis []. In this paper, we use NURBS basis functions only for the constructions of geometrical mappings that precisely map the parameter space onto a physical domain, however we employ B-spline 2

3 basis functions (continuous piecewise polynomials) for analysis. In other words, by the partition of unity property, B-spline basis functions are NURBS basis functions with unit weights, and hence the finite element analysis of this paper is an isogeometric analysis. However, we do not use full-blown NURBS basis functions (piecewise rational functions whose support consist of a union of several knot spans). Instead we use B-spline basis functions that are interpolant at each knot. Therefore the proofs of error estimates of this paper are easier than those presented in [3] which is a highly technical paper dealing with error estimates of the approximations by rational basis functions whose supports are different from that of FEM. Another popular approach to deal with singularity in the conventional FEM is geometric mesh refinements ([6], [26]). However, it is not simple to implement mesh refinement methods in the isogeometric analysis as shown in [4]. We compare mesh refinement methods with our mapping technique in Example 5.4. It is important to note that the proposed mapping technique is not properly working with neither the B-spline functions by the k-refinement nor the NURBS functions. Moreover, the mapping technique has no effects unless the polynomial degrees of B-spline basis functions are 2 in each variable. This paper is organized as follows: Definitions and terminologies used in this paper are introduced in section 2. In section 3, we construct NURBS geometrical mappings that map the parameter space onto a cracked physical domain and investigate the properties of the geometrical mappings. In section 4, we prove error estimates of the proposed mapping technique. Numerical tests that support the theoretical results are presented in section 5. We apply the proposed methods to the crack singularity, to the jump data singularity (the Motz problem), and to general monotone singularities of type r /q ψ(θ) for q = 2, 4, 5, 8, 0. Finally, the concluding remarks are stated in section 6, in which we describe how to apply the mapping technique for an enrichment of NURBS functions to be used for isogeometric analysis. 2 Preliminaries 2. B-splines, control points, knot vectors, weights, and NURBS In this section, we briefly review definitions and terminologies that are used throughout this paper. We follow those in the books [7, 24], and we thus refer to these texts for details. A knot vector Ξ = {ξ, ξ 2,, ξ m } is a nondecreasing sequence of real numbers in the parameter space [0, ], and the components ξ i are called knots. An open knot vector of order p + is a knot vector that satisfies ξ = = ξ p+ < ξ p+2 ξ m p < ξ m p = = ξ m, in which the first and the last p + knots are repeated and the interior knots can be repeated at most p times. The B-spline functions N i,k (ξ) of order k = p + corresponding to the knot vector Ξ = {ξ, ξ 2,, ξ m } are piecewise polynomials of degree p which are constructed recursively by 3

4 Figure : B-spline functions N i,3 (ξ), i =, 2,, 7 of order k = 3 corresponding to the knot vector Ξ = {0, 0, 0, 0.2, 0.5, 0.8, 0.8,,, }. the formula (Cox-de Boor): { if ξ i ξ < ξ i+, N i, (ξ) = for im, 0 otherwise, N i,t (ξ) = ξ ξ i N i,t (ξ) + ξ i+t ξ N i+,t (ξ), for i m, 2 t k. ξ i+t ξ i ξ i+t ξ i+ (There is a terminology conflict between the design and analysis community. Designers will say a quadratic polynomial has degree 2 and order 3 [, 24]). For example, the piecewise quadratic polynomial B-spline functions N i,3 (ξ) corresponding to the knot vector Ξ = {0, 0, 0, 0.2, 0.5, 0.8, 0.8,,, } are depicted in Fig.. The B-spline functions are useful in design as well as finite element analysis because they have the following properties: variation diminishing, convex hull, non-negativity, piecewise polynomial, compact support, and partition of unity. A B-spline curve is defined as follows: C(ξ) = m k i= N i,k (ξ)b i, where B i are control points that make B-spline functions draw a desired curve as shown in Fig. 2(a). Let Ξ η = {η,, η n } be an open knot vector and let p η and k = p η +, respectively, be the polynomial degree and order of B-spline functions M j,k (η). Then a B-spline surface is defined by S(ξ, η) = m k i= n k j= N i,k (ξ)m j,k (η)b i,j, where B i,j are control points that make a bidirectional control net as shown in Fig. 2(b). 4

5 (a) B-spline Curve (b) B-spline Surface Figure 2: (a) B-spline curve and control points. (b) B-spline surface and control net. Let {w i : i =,, m k} be the set of weights. Then the corresponding NURBS basis functions are defined by R i,k (ξ) = N m k i,k(ξ)w i, W (ξ) = N s,k (ξ)w s > 0. W (ξ) The NURBS basis functions are now piecewise rational functions. A NURBS curve corresponding to the control points {B i : i =,, m k}, NURBS basis functions {R i,k (ξ) : i =,, m k}, and the weights {w i : i =,, m k} is C(ξ) = m k i= s= R i,k (ξ)b i. Let {w i,j : i =,, m k, j =,, n k } be the set of weights. Then NURBS basis functions corresponding to the open knot vectors Ξ ξ and Ξ η and the weights {w i,j } are defined by where W (ξ, η) = R i,j (ξ, η) = N i,k(ξ)m j,k (η)w i,j, W (ξ, η) m k s= n k t= N s,k (ξ)m t,k (η)w s,t > 0. Let {B i,j : i =,, m k, j =,, n k } be a set of control points in R d, d 2. Then a NURBS surface corresponding to the control points {B i,j }, NURBS basis functions {R i,j (ξ, η)}, and the weights {w i,j } is S(ξ, η) = m k i= n k j= R i,j (ξ, η)b i,j. 5

6 2.2 Weak solution in Sobolev space Let Ω be a connected open subset of R d. We define the vector space C m (Ω) to consist of all those functions φ which, together with all their partial derivatives α φ(= α α d d φ) of orders α = α + +α d m, are continuous on Ω. A function φ C m (Ω) is said to be a C m -function. If Ψ is a function defined on Ω, we define the support of Ψ as supp Ψ = {x Ω Ψ(x) 0}. For an integer k 0, we also use the usual Sobolev space denoted by H k (Ω). For u H k (Ω), the norm and the semi-norm, respectively, are u k,ω = /2 α u 2 dx, u k,,ω = max α k {ess.sup α u(x) : x Ω} ; α k u k,ω = α =k Ω Ω α u 2 dx /2, u k,,ω = max α =k {ess.sup α u(x) : x Ω}. Suppose we are concerned with an elliptic boundary value problem on a domain Ω with Dirichlet boundary condition g(x, y) along the boundary Ω. Let W = {w H (Ω) : w Ω = g} and V = {w H (Ω) : w Ω = 0}. The variational formulation of the Dirichlet boundary value problem can be written as: Find u W such that B(u, v) = L(v), for all v V, () where B is a continuous bilinear form that is V-elliptic ([5]) and L is a linear functional. The solution to () is called a weak solution which is equivalent to the strong (classical) solution corresponding elliptic PDE whenever u is smooth enough. The energy norm of the trial function u is defined by [ ] /2 u eng = B(u, u). 2 Let W h W, V h V be finite dimensional subspaces. Since the NURBS basis functions do not satisfy the Kronecker delta property, in this paper we approximate the nonhomogenuous Dirichlet boundary condition by the least squares method as follows: g h W h such that g g h 2 dγ = minimum. Ω We can write the Galerkin form (a discrete variational equation) of () as follows: Given g h, find u h = w h + g h, where w h V h, such that B(u h, v h ) = L(v h ), for all v h V h, which can be rewritten as: Find the trial function w h V h such that B(w h, v h ) = L(v h ) B(g h, v h ), for all test functions v h V h. (2) 6

7 3 NURBS geometrical mappings that generate singular basis functions The geometrical mappings we are concerned with are the NURBS curves or the NURBS surfaces defined in the previous section. In order to clearly address how the proposed NURBS geometrical mapping handles the singularity problems in higher dimensional cases, we first show how a B-spline geometrical mapping F handles effectively one-dimensional singularities. Let Ξ η = {0,, 0,,, } be an open knot vector of order k = p η +. Then the B-spline functions M j,k (η) corresponding to Ξ η are ( ) pη M j,k (η) = η j ( η) pη j+ for j =,, k. j Here, M j,k, j =,, k, are also called the Bernstein polynomials of degree p η. Let B j = (0, 0), for j =,, k, and B k = (0, c) be control points, for a constant c. Then the B-spline geometrical mapping F(η) = M j,k (η)b j = (0, cη pη ) k j= maps the parameter space [0, ] onto the physical space {0} [0, c] R 2 and its inverse is η = F (0, y) = (/c) /pη y /pη. Thus, the approximation space V h = span{m i,k F i =,, k }, where k is an integer greater than or equal to k and M i,k are the Bernstein polynomials (B-spline functions) of degree k, contains the following singular as well as smooth functions: y l/pη, l = 0,,, k. In other words, the geometrical mapping F is able to generate the singularity of type r λ, where 0 < λ = /p η <. For example, if p = 2, then the Bernstein polynomials of degree 2 are and M,3 = ( η) 2, M 2,3 = 2η( η), M 3,3 = η 2. B = (0, 0), B 2 = (0, 0), B 3 = (0, /2) are control points. Then the geometrical mapping obtained by these control points and its inverse, respectively, are F(η) = (0, 2 ) η2 and F (0, y) = 2 y. 7

8 Suppose Sη h = span{m j,5 j =,, 5} where M j,5 are the Bernstein polynomials corresponding the the open knot vector Ξ = {0, 0, 0, 0, 0,,,,, } of order 5, then Sη h contains, η,, η 4. Hence the approximation space Vy h = span{m j,5 F : j =,, 5} for isogeometric analysis contains, y, y, y 3/2, y 2. However, the approximation space V h y = span{r j,5 (NURBS) F : j =,, 5} may not be able to generate these singular functions. Thus, the mapping technique is not applicable to the NURBS based isogeometric analysis. We are now going to extend this argument to NURBS geometrical mappings from the parameter space Ω = [0, ] [0, ] onto a cracked unit disk Ω R 2 in the next two subsections. 3. NURBS geometrical mapping to deal with crack singularity of type r /2 ψ(θ) Suppose the physical domain Ω is a unit disk with a crack along the positive x-axis as shown in Fig. 3(b). We now consider a NURBS isogeometric mapping from the parameter space Ω = [0, ] [0, ] to the physical domain Ω. Consider the knot vectors: Ξ ξ = {0, 0, 0, 4, 4, 2, 2, 34, 34 },,,, Ξ η = {0, 0, 0,,, }. Let N i,3, i =,, 9 be the B-splines corresponding to the knot vector Ξ ξ and let M j,3, j =,, 3 be the B-splines corresponding to the knot vector Ξ η. Then these B-spline functions are { ( 4ξ) 2 if ξ [0, N,3 (ξ) = 4 ] 0 if ξ / [0, 4 ] N 2,3 (ξ) = N 3,3 (ξ) = N 5,3 (ξ) = N 7,3 (ξ) = N 9,3 (ξ) = { 8ξ( 4ξ) if ξ [0, 4 ] 0 if ξ / [0, 4 ] (4ξ) 2 if ξ [0, 4 ] { (2 4ξ) 2 if ξ [ 4, 2(4ξ )(2 4ξ) if ξ [ 2 ] N 4,3 (ξ) = 4, 2 ] 0 if ξ / [0, 2 ] 0 if ξ / [ 4, 2 ] (4ξ ) 2 if ξ [ 4, 2 ] { (3 4ξ) 2 if ξ [ 2, 3 2(4ξ 2)(3 4ξ) if ξ [ 4 ] N 6,3 (ξ) = 2, 3 4 ] 0 if ξ / [ 4, 3 4 ] 0 if ξ / [ 2, 3 4 ] (4ξ 2) 2 if ξ [ 2, 3 4 ] (4 4ξ) 2 if ξ [ 3 4, ] N 8,3 (ξ) = 0 if ξ / [ 2 {, ] (4ξ 3) 2 if ξ [ 3 4, ] 0 if ξ / [ 3 4, ] { 8(4ξ 3)( ξ) if ξ [ 3 4, ] 0 if ξ / [ 3 4, ] M,3 (η) = ( η) 2, M 2,3 (η) = 2η( η), M 3,3 (η) = η 2, for η [0, ]. (4) Consider the control points B i,j and the weights w i,j for i 9, j 3, that are listed in Table. With the B-spline functions shown in (3) and (4), the 27 control points and (3) 8

9 Table : Control points B i,j and weights w i,j. i j B i,j w i,j i j B i,j w i,j i j B i,j w i,j (0, 0) 2 (0, 0) 3 (, 0) 2 (0, 0) (0, 0) (, ) 2 3 (0, 0) 3 2 (0, 0) 3 3 (0, ) 4 (0, 0) (0, 0) (, ) 2 5 (0, 0) 5 2 (0, 0) 5 3 (, 0) 6 (0, 0) (0, 0) (, ) 2 7 (0, 0) 7 2 (0, 0) 7 3 (0, ) 8 (0, 0) (0, 0) (, ) 2 9 (0, 0) 9 2 (0, 0) 9 3 (, 0) weights, we now construct a NURBS geometrical mapping from the parameter space Ω onto Ω as follows: 9 3 F(ξ, η) = R i,j (ξ, η)b i,j. i= j= Here R i,j (ξ, η), i 9, j 3, are NURBS basis functions defined by where R i,j (ξ, η) = N i,3(ξ)m j,3 (η)w i,j, W (ξ, η) W (ξ, η) = 9 s= t= 3 N s,3 (ξ)m t,3 (η)w s,t. Noting that from Table, w s,j = if s =, 3, 5, 7, 9 and w s,j = / 2 if s = 2, 4, 6, 8, and using the partition of unity property: 3 t= M t,3(η) =, we have [ 9 3 ] W (ξ, η) = N s,3 (ξ) M t,3 (η)w s,t s= t= = [N 2,3 (ξ) + N 4,3 (ξ) + N 6,3 (ξ) + N 8,3 (ξ)] / 2 w(ξ), + [N,3 (ξ) + N 3,3 (ξ) + N 5,3 (ξ) + N 7,3 (ξ) + N 9,3 ( ξ)] which becomes a function of ξ only. Since B i, = B i,2 = (0, 0) for all i by the choice of the control points B i,j and the weights w i,j in Table, we have F(ξ, η) = η 2 9 i= N i,3 (ξ)w i,3 B i,3 := (x(ξ, η), y(ξ, η)), w(ξ) 9 (5)

10 where the coordinate functions are as follows: x(ξ, η) = η2 [N,3 + N 2,3 / 2 N 4,3 / 2 N 5,3 N 6,3 / 2 + N 8,3 / ] 2 + N 9,3 (ξ) w(ξ) ( ) X(ξ) = η 2, w(ξ) y(ξ, η) = η2 w(ξ) = η 2 ( Y (ξ) w(ξ) [ N 2,3 / 2 N 3,3 N 4,3 / 2 + N 6,3 / 2 + N 7,3 + N 8,3 / ] 2 (ξ) ). Moreover, the weight function w(ξ) is bounded away from zero: (6) W (ξ, η) w(ξ). Lemma 3.. Suppose u(r, θ) = rψ(θ) for a smooth function ψ. Then û(ξ, η) := u F(ξ, η) = ηψ(ξ), where Ψ(ξ) C 0 [0, ] and Ψ(ξ) C (0, ) unless ξ = /4, /2, 3/4. Proof. From Fig. 3 and the weights w i,j in Table, we have We thus have which implies = x(ξ, η) 2 + y(ξ, η) 2 = η 4. ( ) X(ξ) 2 + w(ξ) From (7), the pull back of r onto Ω becomes Since ( ) Y (ξ) 2, (7) w(ξ) w(ξ) 2 = X(ξ) 2 + Y (ξ) 2. [ (X(ξ) ) 2 r F(ξ, η) = η + w(ξ) ( ) ] Y (ξ) 2 /2 = η. w(ξ) ( ) ( ) 2π + tan y(ξ,η) x(ξ,η) = 2π + tan Y (ξ) X(ξ) if ξ [0, 4 ), 3 2 π ( ) ( ) if ξ = 4, θ(ξ, η) = π + tan y(ξ,η) x(ξ,η) = π + tan Y (ξ) X(ξ) if ξ ( 4, 3 4 ), 2 π ( ) ( ) if ξ = 3 4, tan y(ξ,η) x(ξ,η) = tan Y (ξ) X(ξ) if ξ ( 3 4, ]. and X(ξ) and Y (ξ) are smooth, and X(ξ) 0 unless ξ = /4, /2, 3/4, the pull back of ψ, Ψ(ξ) ψ(θ(ξ, η)), is smooth unless ξ = /4, /2, 3/4. Since F and ψ are continuous, so does Ψ. 0

11 (a) Parameter space with coarse mesh (b) Physical domain with control points Figure 3: The parameter space and the physical domain for the NURBS mapping F. Lemma 3. shows that the pull back of a singular function r /2 ψ(θ) through the NURBS mapping F becomes a piecewise smooth function on the parameter space Ω. For the error analysis in the forthcoming section, we estimate an upper bound of the determinant of the Jacobian of F: det(j(f)) = x y ξ η x y η ξ = 2η3 w(ξ) 2 X (ξ)y (ξ) X(ξ)Y (ξ), 3 X (ξ)y (ξ) X(ξ)Y [ (ξ) = (X (ξ)y (ξ) X(ξ)Y (ξ))χ [i/4,(i+)/4) (ξ) ], i=0 where X(ξ) and Y (ξ) are defined by (6). Here χ [i/4,(i+)/4) (ξ) = for ξ [i/4, (i + )/4) and χ [i/4,(i+)/4) (ξ) = 0 otherwise, i = 0,, 3. From those quadratic B-spline functions N i,3 (ξ) defined by (3), one can see that X (ξ)y (ξ) X(ξ)Y (ξ) is a bounded cubic polynomial on each subinterval [i/4, (i + )/4). We actually get 2( + 2) (X (ξ)y (ξ) X(ξ)Y (ξ))χ [0,/4) (ξ) 4 2, 2( + 2) (X (ξ)y (ξ) X(ξ)Y (ξ))χ [/4,/2) (ξ) 4 2, 2( + 2) (X (ξ)y (ξ) X(ξ)Y (ξ))χ [/2,3/4) (ξ) 4 2, 2( + 2) (X (ξ)y (ξ) X(ξ)Y (ξ))χ [3/4,) (ξ) 4 2. (8) We thus have det(j(f)) 64 2(3 2 2) 6. (9) Remark 3.. () Unlike the usual NURBS geometrical mapping in the isogeometric analysis literature, the NURBS mapping constructed above is not one-to-one along η = 0 where the determinant of Jacobian matrix J(F) is zero. (2) In order to make det(j(f)) approach zero slower than η does as η goes to 0, it is possible to use ηe ωη in place of η for the construction of the geometrical mapping F with damping factor ω. However, we tested this mapping to see that the alternative non-nurbs mapping does not yield better results in dealing with the singularity of type r λ ψ(θ), where 0 < λ /2.

12 (3) Since N i,3 (ξ), i 9, are interpolant when ξ = 0, /4, /2, 3/4, we have F(0, η) = η 2 B,3 = (η 2, 0), F(/4, η) = η 2 B 3,3 = (0, η 2 ), F(/2, η) = η 2 B 5,3 = ( η 2, 0), F(3/4, η) = η 2 B 7,3 = (0, η 2 ). 3.2 NURBS mapping to deal with stronger singularity of type r λ ψ(θ) with 0 < λ < /2 Suppose λ = /q for a positive integer q that is greater than 2. Then we consider the two open knot vectors: Ξ ξ = {0, 0, 0, 4, 4, 2, 2, 34, 34 },,,, Ξ η = {η,, η q+, η q+2,, η 2(q+) }, where η j = 0 if j q + ; η j = if q + < j. We select 9(q + ) control points so that B i,j = (0, 0), if j q +, and the corresponding weights w i,j in a similar way to those in Table. Then the corresponding NURBS geometrical mapping F : Ω Ω is of the form F(ξ, η) = (x(ξ, η), y(ξ, η)), where ( ) ( ) X(ξ) Y (ξ) x(ξ, η) = η q, y(ξ, η) = η q, (0) w(ξ) w(ξ) where X(ξ), Y (ξ), w(ξ) are the same as those in (6). It has the following property: r λ ψ(θ) F(ξ, η) = ηψ(ξ), where Ψ(ξ) is piecewise smooth by a similar argument to Lemma 3.. That is, the pull back of a singular function onto the parameter space through the NURBS mapping F becomes a continuous piecewise smooth function. Lemma 3.2. Suppose ψ(θ) is a smooth function such that Ψ(ξ) = (ψ F)(ξ, η) is smooth except ξ = 0, /4, /2, 3/4,. Then, for each non-negative integer p, there is a constant C(p) such that d p dξ p Ψ(ξ) C(p) ψ p,,[0,2π], j =,, 4, () ((j )/4,j/4) where C() 7.8 and C(2) 3.25, and the constant C(p) is increasing as p is getting larger. Proof. By the definition of the geometrical map F and Eq. (7), we have θ(ξ) = C j + tan Y (ξ) X(ξ), where C j is a constant that depends on the region [(j )/4, j/4], j =, 2, 3, 4. Since Ψ(ξ) = ψ(c j + tan Y (ξ) X(ξ) ), by the Faà di Bruno s Formula [2] (a formula for derivatives of a composite function), we have d p dξ p Ψ(ξ) = p! s!s 2! s p! ψ(s) (θ) ( θ (ξ) 2! ) s ( θ ) ( ) (ξ) s2 θ (p) sp (ξ) 2! p!

13 where the sum runs all possible solutions of the system s +s 2 + +s p = s, s +2s 2 + +ps p = p with s i 0. Therefore, for each j =,, 4, we get d p dξ p Ψ 0,,([j ]/4,j/4) p C(s,, s p ) ψ s, C(p) ψ p,. s=0 For example, if ψ(θ) = sin(θ/2), by (7), X 2 +Y 2 (3+2 2)/8 and by (8), Y X X Y 4 2, and hence Y X X Y X 2 +Y Error estimates The NURBS geometrical mapping F : Ω Ω constructed with coarse mesh on Ω (Fig. 3(a)) does not change as the mesh on Ω is further refined. Let S h S(Ξ ξ, Ξ η, p ξ, p η ) = span{n i,pξ +(ξ)m j,pη+(η) i m, j n}, Sξ h S(Ξ ξ, p ξ ) = span{n i,pξ +(ξ) i m}, V h = span{[n i,pξ +(ξ)m j,pη+(η)] F i m, j n}, Vξ h = span{[n i,p ξ +(ξ)η l ] F i m}, for a fixed l, Q h = mesh on (0, ) 2, K = F(Q), Q Q h. (2) Here p ξ and p η, respectively, are the polynomial degrees of B-spline basis functions in the ξ- and the η-directions. Let us note that these approximation spaces are for the isogeometric analysis of elliptic problems, but not for the construction of F. In this section, we assume the following:. The degree p without suffix means the degree of polynomial in the ξ-variable. M j,n (η) are Bernstein polynomials for all j whose supports are [0, ] in the construction of the geometrical mapping F. 2. u is the weak solution of () and u h is the Galerkin approximate solution of (2). Each knot value of the open knot Ξ ξ has multiplicity p. That is, each member in S h ξ is a C0 -function. Because of these facts, the error estimates of our isogeometric analysis now becomes similar to those of FEM. Lemma 4.. Suppose v C[0, h ] is (p + 2)-times differentiable on (0, h ). If are p + distinct numbers, then there exists 0 = x 0 < x < x 2 < < x p = h Π p v S p (x) span{x i ( x) p i : i = 0,, 2,, p} 3

14 such that Moreover, if h <, Π p v(0) = v(0) and Π p v(h ) = v(h ). (3) v(x) Π p v(x) h p+ v p+, /(p + )!; d dx [v(x) Π pv(x)] hp ( v p+, + v p+2, ) /p!. (4) Proof. If a projection operator from C[0, ] onto S p (x) is defined by the p-th Lagrange polynomial that interpolates v C[0, ] at x i, i = 0,, 2,, p: Π p v = p v(x l )L p,l (x), l=0 where L p,l (x) = p (x x i ) i=0,i l (x l x i ), then, for each x (0, h ), there exists c(x) (0, h ) such that v(x) Π p v(x) = v(p+) (c(x)) β(x), with β(x) = (x x 0 ) (x x p ). (5) (p + )! Since β(x) h p+, we have v (p+) v(x) Π p v(x) 0, h p+ (p + )! 0, Since f(x) := [v(x) Π p v(x)] (p + )! and β(x) are differentiable for all x (0, h ), we have which implies f(x + h) f(x) h. [ ] = β(x + h) v (p+) (c(x + h)) v (p+) (c(x)) /h + v (p+) (c(x)) [β(x + h) β(x)] /h, (6) d dx [v(x) Π pv(x)] h p [ v p+, ]/(p!), if lim h 0 β(x + h) = 0. (7) [ ] If lim β(x + h) 0, in the relation (6), then lim v (p+) (c(x + h)) v (p+) (c(x)) /h exists, h 0 h 0 and hence d [ ] dx [v(x) Π pv(x)] h p+ v (p+2) 0, + (p + )h p v(p+) 0, /(p + )! h p [ v (p+2) 0, + v (p+) 0, ] /(p!) (since h < ). 4

15 Let T k : [0, ] I k = [ξ k, ξ k+ ] be the linear transformation. operator Π [k] p : H (I k ) S p [k] (ξ) Sξ h I k is defined by Then a local projection Π [k] p w = Π p (w T k ) T k for w H (I k ). Let us note that S p [k] (ξ) T k is the space of Bernstein polynomials of degree p because of the choice of the open knot vector Ξ ξ with multiplicity p at each knot. Theorem 4.. Let 0 < λ /2. Suppose u(r, θ) = N k=0 c kr (λ+k) ψ k (θ) with smooth functions ψ k (θ) solves the Poisson equation. Assume that each node in the open knot vector Ξ ξ for the approximation space S h ξ has the multiplicty p ξ. (i) If F is the NURBS geometrical mapping from the parameter space Ω onto the physical domain Ω, defined by (0) with q = /λ, then u F = N c l η +l/λ Ψ l (ξ) = l=0 N c l η +ql Ψ l (ξ), l=0 for piecewise smooth functions Ψ l (ξ). (ii) If u h V h is an isogeometric finite element solution of u, then we have [ N ] u u h,ω C h p c l ( Ψ l p+, + Ψ l p+2, ) /(p!). (8) l=0 (iii) Assume u H0 (Ω). If h 0 (0, ) is determined so that the projection P h of S onto S h and the complement operator I P h are continuous with respect to a weighted Sobolev norm (Theorem 22.6 of [5]), h min{h 0, /e}, and p = p ξ 2, p η max{2, qn + }, then we have [ N ] u u h,ω C h p+ c l ( Ψ l p+, + Ψ l p+2, ) /(p!), l=0 [ N ] u u h 0,Ω C 0 h p+ c l ( Ψ l p+, + Ψ l p+2, ) /(p!). l=0 Ψ l p+i, := k Ψ l p+i,,ik, where Ψ l is smooth on I k, for each k and k I k = [0, ]. Here p = p ξ is the polynomial degree of N i,p+ (ξ) for an approximation of the angular direction (the polynomial degree p η 2 for an approximation of the radial direction is held fixed), and h = max{ ξ i+ ξ i } is the maximum length of knot spans of the open knot vector Ξ ξ and the constants C, C 0, C are independant of h and p. Proof. (i) An error estimate on each element K = F(Q), Q Q h can be summed together to have a global error estimate on Ω. Thus we have an error estimate on an element K in the following. Without loss of generality, we may assume that λ = /2 and u(r, θ) is a sum of two terms: u(r, θ) = r /2 ψ (θ) + r 3/2 ψ 2 (θ). (9) 5

16 Using the arguments similar to the proof of Lemma 3. we have u(r, θ) F = ηψ (ξ) + η 3 Ψ 2 (ξ), (20) where the continuous functions Ψ (ξ), Ψ 2 (ξ) are smooth unless ξ = /4, 2/4, 3/4. (ii) In what follows, we let p η = p 3 and p ξ = p 2.. Let T = {Q k = [ξ k, ξ k+ ] [0, ] k =,, N ξ } be a partition of ˆΩ, such that {i/4} [0, ], i = 0,, 2, 3, 4, are edges of some elements in T. (η η i ) 2. L p,s (η) = p i=0,i s (η s η i ), where η i = i/ p, i = 0,,, p. For each k =,, N ξ, we let L [k] p,l (ξ) = p (ξ ξi k) i=0,i l (ξl k ξk), where ξk i = ξ k + i(ξ k+ ξ k )/p, i = 0,,, p. i 3. The p-th and p-th Lagrange interpolating polynomials, respectively, are Π η (Φ(η)) = p s=0 Φ(η s )L p,s (η) on [0, ] and p Π ξ (Ψ(ξ)) Ik := Π [k] ξ (Ψ) = Ψ(ξ s [k] )L k p,s(ξ), on I k = [ξ k, ξ k+ ]. s=0 Π ξ is the interpolant defined on C[0, ] such that Π ξ Ik = Π [k] ξ for k =,, N ξ. Suppose Φ(η) is a polynomial of order p and Ψ(ξ) is a piecewise smooth function. Π η Φ = Φ and since Π ξη = Π ξ Π η, we have Then ΦΨ Π ξη ΦΨ = ΦΨ Π η ΦΠ ξ Ψ = Φ[Ψ Π ξ Ψ]. (2) Using this equation, Lemma 3.2, and Lemma 4., we obtain the following inequalities that will be used throughout the proof: [Ψ Π [k] ξ Ψ]2 dξ h 2(p+) { Ψ p+, /(p + )!} 2 [meas(i k )]; I k Q k [ΦΨ Π ξη ΦΨ] 2 dξdη = 0 Φ 2 dη [Ψ Π [k] ξ Ψ]2 dξ I k h 2(p+) { Ψ p+, /(p + )!} 2 Φ 2 0,[0,] [meas(i k)]; ΦΨ Π ξη ΦΨ 0, Φ 0, Ψ Π ξ Ψ 0, C h p+ { Ψ p+, /(p + )!} ; d ΦΨ Π ξη ΦΨ, Φ, Ψ Π ξ Ψ 0, + Φ 0, dξ [Ψ Π ξψ] 0, C 2 h p Φ, ( Ψ p+, + Ψ p+2, ) /p!. Let us note that k meas(i k) =. Hence, the terms, meas(i k ), will be cancelled out after summing the error bounds over Q k. (22) 6

17 Let û(ξ, η) = u(r, θ) F(ξ, η), Π ξη û Qk = ηπ [k] ξ Ψ (ξ) + η 3 Π [k] ξ Ψ 2(ξ), Π xy u K = Π ξη û Q F (x, y). Then, since the interpolating operator Π ξη is linear, (23) (û Π ξη û) Qk = η(ψ Π [k] ξ Ψ ) + η 3 (Ψ 2 Π [k] ξ Ψ 2). (24) First, we estimate an upper bound of u Π xy u 0,K. Using (9), (22) and (24) yields u Π xy u 2 0,K = {(û Π ξη û) Qk } 2 det(j(f)) dξdη Q k [ 6 η(ψ Π [k] ξ Ψ ) + η 3 (Ψ 2 Π [k] 2 ξ 2)] Ψ dξdη Q k 32 { [ ] Ψ (ξ) Π [k] 2 [ ] } 3 ξ Ψ (ξ) + Ψ 2 (ξ) Π [k] 2 ξ Ψ 2(ξ) dξ I k C 02 h 2(p+) ( Ψ 2 p+, + Ψ 2 2 p+, ) /(p + )![meas(ik )]. (25) Second, we estimate an upper bound of u Π xy u,k. Since the geometrical mapping F is unchanged at all levels of refinement (the h-p-k refinement, see page 45 of [7]), we do not lose a generality even though we work with a fixed geometrical mapping. Using the geometrical mapping defined by (6) and the property (7) of F, we have [ ] q q 2 = det(j(f)) [J(F) ] T [J(F) ] q 2 q 22 { (X ) 4 2 ( ) } Y 2 {( ) ( ) Y Y ( ) ( ) X X } +, 2 + η w w w w w w = = w 2 2 X Y XY {( ) ( ) Y Y 2 + w w ( X w 2w 2 η X Y XY, 0 0, { [(Y ηw 2 ) ] 2 2 X Y XY + w ) ( ) X } { [(Y, η w w [( ) X ] 2 } w. ) ] 2 + [( ) X ] 2 } w From (3) and (5), X (ξ)w(ξ) X(ξ)w (ξ) and Y (ξ)w(ξ) Y (ξ)w (ξ) are piecewise cubic polynomials and (2 + 2)/4 w(ξ). Moreover, the common factor of denominators is bounded away from zero: from (8), we have 4.8 X (ξ)y (ξ) X(ξ)Y (ξ). Thus we have max{ ηq, q 22 } C. 7

18 Let (u Π xy u) F = (u Πxy u), xy = [ x, y ]T, and ξη = [ ξ, η ]T. Then using the inequalities in (4), we have u Π xy u 2,K = [ xy (u Π xy u)] T [ xy (u Π xy u)] dxdy K [ T [ ] = ξη(u Πxy u)] det(j(f)) [J(F) ] T [J(F) ] ξη(u Πxy u) dξdη Q ( [ ] C 2 [ ξη η(ψ Π [k] ξ Ψ ) + η 3 (Ψ 2 Π [k] T ξ 2))] Ψ q q 2 Q q 2 q 22 ( [ ξη η(ψ Π [k] ξ Ψ ) + η 3 (Ψ 2 Π [k] dξdη ξ 2))] Ψ [ { d C 3 η 2 q Q dξ (Ψ Π [k] ξ Ψ ) + η 2 d } 2 dξ (Ψ 2 Π [k] ξ Ψ 2) { } + q 22 (Ψ Π [k] ξ Ψ ) + 3η 2 (Ψ 2 Π [k] 2 ] ξ Ψ 2) dξdη C 4 B 2 [meas(i k )]h 2p, (26) where B = ( Ψ p+, + Ψ p+2, + Ψ 2 p+, + Ψ 2 p+2, ) /(p!). Adding (25) and (26), we have u Π xy u,k C 5 h p meas(i k )B. (27) Since Π xy û K = ηπ [k] ξ Ψ (ξ) + η 3 Π [k] ξ Ψ 2(ξ) S h Qk and V h = S h F, using the well known argument (Céa s Lemma, [5]) with (27), we have Summing (28) over all K Q F, we have u u h,k C 6 h p meas(i k )B. (28) u u h,ω C 7 h p B, (29) where the constant C 7 is independent of h and p. (iii) For a maximum norm estimate, we use those arguments in Section 22 of Ciarlet[5]: let P h be the projection of H 0 (ˆΩ) onto S h such that ˆB((I P h )û, ˆv h ) = 0, for all ˆv h S h, (30) where I is an identity operator on S. By Theorem 22.6 of [5], there is h 0 (0, ) such that for all h min{h 0, /e}, ln h /2 P h û 0, + h P h û, C 3 ( û 0, + h ln h û, ), ln h /2 Iû 0, + h Iû, ln h 0 /2 ( û 0, + h ln h û, ), (3) for all û H 0 (ˆΩ) W, (ˆΩ). Here W, (ˆΩ) is the set of functions v such that v, <. By a similar manner to the proof of Theorem 22.7 of [5], we apply (3) to û û h = û P h û = (I P h )û = (I P h )(û Π ξη û) 8

19 to obtain the following ln h /2 û û h 0, + h û û h, (C 3 + ln h /2 ) ( û Π ξη û 0, + h ln h û Π ξη û, ). (32) Moreover, by (22.8) of [5], if, for each v S h, v Qk (ξ, η) is a polynomial of order 2, one can drop ln h terms from the above inequality (32). Thus, using û Π ξη û = η(ψ (ξ) Π ξ Ψ (ξ)) + η 3 (Ψ 2 (ξ) Π ξ Ψ 2 (ξ)) and those estimates in (22) yields û û h 0,,Ω C 4 ( û Π ξη û 0, + h û Π ξη û, ) C 5 h p+ ( Ψ p+, + Ψ p+2, + Ψ 2 p+, + Ψ 2 p+2, ) /(p!). (33) The L 2 -norm estimate is followed by the maximum norm estimate: since meas(ˆω) =, we have u u h 0,Ω C 03 û û h 0,,Ω C 0 h p+. (34) Under the same assumptions of Theorem 4. and with the same notations of the proof, using results in Schwab ([25]), we prove the following error estimates: Let ϕ : Ω = [, ] I k = [ξ k, ξ k+ ] [0, ] be the linear mapping defined by ξ = ϕ(t) = (h/2)(t + ) + ξ k, where h = ξ k+ ξ k. Let Ψ(t) = Ψ(ξ) ϕ(t). Let 0 j m be integers. Then V m j ( Ω) is the space of all w L 2 ( Ω) for which m w 2 Vj m( Ω) = ( t 2 ) i w (i) (t) 2 dt <. i=j Ω Suppose Ψ(t) H ( Ω) V m 0 ( Ω) for some m, then by Theorem 3.4 and Corollary 3.5 of Schwab ([25]), there exist Π p Ψ S p ( Ω) (which denotes the space of polynomials of degree p) such that Π p Ψ(±) = Ψ(±) and Ψ Π p C Ψ 0, Ω 2 (m)p m Ψ V m s ( Ω), Ψ [Π p Ψ] 0, Ω C (m)p m Ψ V m s ( Ω), where 0 s min(p, m). We assume m p. Then scaling back to I k yields Π p Ψ(±) = Ψ(±), (35) Ψ [Π p Ψ] 0,Ik C (m)( h 2 )m p m Ψ V m m (I k ), (36) Ψ Π p Ψ 0,Ik C 2 (m)( h 2 )m+ p m Ψ V m m (I k ). (37) Let Π p Ψ ϕ = Π p Ψ Ik = Π [k] p Ψ, then Π p Ψ S h ξ. If u(r, θ) F = ηψ (ξ) + η 3 Ψ 2 (ξ), we have ηπ p Ψ (ξ) + η 3 Π p Ψ 2 (ξ) S h = span{n(ξ)η j N(ξ) Sξ h, j = 0,, 3}. 9

20 Now using (35)-(37), we have (ηψ + η 3 Ψ 2 ) (ηπ p Ψ + η 3 Π p Ψ 2 ) 2 0,Q k = η(ψ Π p Ψ ) + η 3 (Ψ 2 ΠΨ 2 ) 2 0,Q k (Ψ Π p Ψ ) 2 0,I k + (Ψ 2 ΠΨ 2 ) 2 0,I k C (m) 2 ( h 2 )2(m+) p 2m ( Ψ 2 V m m (I k) + Ψ 2 2 V m m (I k) ); (38) and from (26) [ Q k Q k ξη C 3 T [ (u Πxy u)] det(j(f)) [J(F) ] T [J(F) ] [ ξη ( η(ψ Π [k] p Ψ ) + η 3 (Ψ 2 Π [k] ( [ ξη Q k [ η 2 q + q 22 η(ψ Π [k] p Ψ ) + η 3 (Ψ 2 Π [k] ξη [ T q q p Ψ 2 ))] 2 q 2 q 22 p Ψ 2 ))] dξdη { d dξ (Ψ Π [k] p Ψ ) + η 2 d dξ (Ψ 2 Π [k] p Ψ 2 ) { (Ψ Π [k] p Ψ ) + 3η 2 (Ψ 2 Π [k] } 2 ] p Ψ 2 ) dξdη ] C 5 (m)( h [ 2 )2m p 2m Ψ 2 Vm (I k) + Ψ 2 2 Vm (I k) + C 6 (m)( h [ ] 2 )2(m+) p 2m Ψ 2 Vm (I k) + Ψ 2 2 Vm (I. k) } 2 ] (u Πxy u) dξdη ] (39) By applying the inequalities (38) and (39) and Céa s Lemma([5]) on each Q k and then summing over all K k = F(Q k ), we have the following: Theorem 4.2. Under the same assumption of the previous theorem, if m p, we have [ ] N u u h,ω C(m) hm 2 m p m d dξ Ψ l(ξ) V m (I k ). (40) Remark 4.. From Lemma 3.2, for each l, Ψ l (ξ) V m m (I k ) in (40), is sharply increased as m is increased. Thus, in Theorem 4.2, if m( p) is large, then h should be small for an optimal result. However, the error estimate (40) shows that the error goes to zero as p is increased even if the regularity m of Ψ l is fixed to be low. Thus, the estimate (40) has an advantage over the estimate (8). 5 Numerical tests In order to show that the proposed mapping techniques are effective in dealing with singularity problems, our mapping method is applied to the elliptic boundary value problems with singularity of type k l=0 r λ ψ(θ), where 0 < λ <, and ψ is a smooth function. 20

21 Table 2: The degree of polynomials and the open knot vectors for the geometrical mapping F. variables polynomial degrees open knot vectors ξ p ξ = 2 Ξ ξ = {0, 0, 0, 4, 4, 2, 2, 3 4, 3 4,,, } η p η = 2 Ξ η = {0, 0, 0,,, } For example, the crack singularity and the jump-boundary data singularity have λ = /2 and the interface problems and the elasticity problems with exotic boundary conditions could have λ close to 0. Throughout this section, we measure the error (u u h ) of the computed solutions obtained by the proposed mapping method in the following norms: u u h,rel (%) = u uh 00, u u h u L 2,rel(%) = u uh L 2 u L 2 [ ] u u u h 2 eng u h 2 2 eng eng,rel (%) = 00. u 2 eng 00, 5. The crack singularity of type r /2 In this subsection,. the parameter space is Ω = [0, ] [0, ], the physical space is Ω = {(r, θ) : r, 0 < θ < 2π} with crack along the positive x-axis, 2. the geometrical mapping F is the NURBS surface corresponding to the knot vectors in Table 2 and the 27 control points and the 27 weights in Table. We emphasize that our geometrical mapping F does not change as the degrees of B-splines are elevated or the knots are inserted. We consider two Poisson s equations on Ω: One with homogeneous Dirichlet BC (Example 5.) and another with non-homogeneous Dirichlet BC (Example 5.2). Example 5.. Let u (r, θ) = [ r( r) sin ( ) θ + sin 2 ( )] 3θ 2 and f = u. Then u is the analytic solution of the Poisson equation with homogeneous boundary condition: u = f in Ω with u = 0 on Γ = Ω, (4) that has the crack singularity at (0, 0). By Lemma 3., the pull back of u (x, y) onto the parameter space is of the form û = (u F)(ξ, η) = ηψ (θ) + η 3 Ψ 2 (θ) 2

22 and hence we need to have order elevations up to degree p η = 3 (or higher) in η-direction by choosing the open knot vector Ξ η = {0, 0, 0, 0,,,, }. Note that the knot vector for the mapping F is unaltered (that is, those in Table 2). To improve the isogeometric analysis of (4) in the angular direction we elevate the degrees of polynomials with the mesh fixed (the p-version): [0, ] = [0, /4] [/4, /2] [/2, 3/4] [3/4, ]. For an isogeometric finite element solution of (4), the relative error (%) in the maximum norm as well as the L 2 -norm with respect to a p-refinement are listed in Table 3. The computed strain energy and the relative error (%) in the energy norm with respect to a p-refinement are listed in Table 4. Moreover, the relative error (%) in the maximum norm and the L 2 -norm listed in Table 3 are depicted in Fig. 4. The relative error (%) in the energy norm of Table 4 are depicted in Fig. 5(a). We use the tolerance and double precision for our calculation. In Table 4, the strain energy of the computed solution reaches the true strain energy when p = 3. The computed energies corresponding to p = 4, 5 and 6 in Table 4 are spurious because these are larger than the the true energy outside the tolerance. However, the computed energies shown in Table 3 corresponding to p = 3, 4, 5, 6 agree with the true energy within the tolerance. In this section, all of the figures are plotted in semi-log scales except four figures that are plotted in log-log scale (Figs. 5(b), 6, 2 and 3). The energy norm is equivalent to the - norm for u. The degrees of freedom (DOF) is the number of B-spline basis functions that are employed for the finite element analysis. In order to numerically justify the error analysis of the previous section, we plot the relative error of the computed solutions obtained by the p-refinements (Figs. 4 and 5(a)) as well as the h-refinements (Figs. 5(b) and 6). By Theorem 4., we have log u u h (p + ) log h + log C, log u u h 0 (p + ) log h + log C 0, log u u h p log h + log C, (42) where we choose h = /4 for the p-version and p(= p ξ ) = 2 or 3 for the h-version, from which we observe the following: The slopes of convergence profiles in the maximum norm, the L 2 -norm, and the -norm for the relative error against the p-degrees (the p-refinement) are about log(/4) = 0.602, as shown in Figs. 4 and 5(a). The slopes of convergence profiles of the relative error against the mesh size h in the angular direction (the h-refinement) are p in the -norm; (p + ) in the L 2 -norm and the maximum norm, as shown in Figs. 5(b) and 6. The corresponding numerical data of the h-refinement are listed in Tables 7, 8, and 9 in appendix. 22

23 Table 3: The relative error (%) in the maximum norm and the L 2 -norm of the computed solutions of the Poisson equation (4) with homogeneous Dirichlet boundary condition. The slopes are computed by using two consecutive nodes on the line in Fig. 4. (p ξ, p η ) dof u u h,rel (%) slope u u h L 2,rel(%) slope (2, 3) E E + 0 (3, 3) E E (4, 3) E E (5, 3) E E (6, 3) E E (7, 3) E E (8, 3) E E (9, 3) 70.59E E (0, 3) E E (, 3) E E (2, 3) E E (3, 3) E E (4, 3) E E (5, 3) 8.263E E (6, 3) E E From Theorem 4., the slope of the convergence profile in the -norm can be estimated as follows: log u u h with (p + ) log u u h with p [(p + ) log h + log C ] [p log h + log C ] = log h = 0.602, (43) when the relative error in -norm is plotted against the p degrees. However, the computed slopes of the convergence profile are slightly better than the estimated slopes in Tables 3 and 4. Together with the h-convergence in Figs. 5(b) and 6 and Tables 7-9, we can see that numerical tests support our theory on the error analysis. The performance of the k-refinement combined with mapping technique is shown in Table 0 in appendix. We do not observe a big advantage of the k-refinement over the p-refinement in the presence of singularity. Let us note that in the support of the B-spline functions used in our p-refinements, û is smooth, however, in the support of the B-spline function by the k-refinements, û is not smooth. In general, if the η direction is also k-refined, the mapping technique are not able to generate η, η 3. Hence, the corresponding approximation space do not contain the singular functions r /2 ψ (θ), r 3/2 ψ 2 (θ) which resemble the true solution u. However, the B-spline functions obtained by the p-refinement generate the complete polynomials. Thus, the corresponding approximation space on Ω contains the singular functions that resemble the true solution. 23

24 Relative error in maximum norm (%) Relative error in L2 norm (%) Polynomial degree (a) Rel error in max-norm: p-version, homogenuous BC Polynomial degree (b) Rel error in L 2 -norm: p-version, homogenous BC Figure 4: Relative error of computed solutions of the Poisson equation (4) with homogeneous Dirichlet boundary condition: (a) The relative error in the maximum norm (%), (b) The relative error in the L 2 norm (%). Relative error in energy norm (%) Relative error in energy norm (%) Quadratic Cubic Polynomial degree (a) Rel error in -norm: p-version, homogenuous BC 0 3 h size 0 (b) Rel error in -norm: h-version, homogenuous BC Figure 5: Relative error in energy norm (%) of the Poisson equation (4) with homogeneous Dirichlet boundary condition: (a) Relative error in -norm versus polynomial degrees p. The slope of the line is about log(/4) (b) Relative error versus the mesh size h, hence the slopes of the lines are 2 and 3, respectively, according as p = 2 and p = 3. 24

25 Table 4: The computed strain energy and the relative error in energy norm (%) with respect to a p-refinement for the Poisson equation (4) with homogeneous Dirichlet boundary condition. The computed energy becomes the true energy when p = 3. The energies corresponding to p = 4, 5, 6 are spurious because these energies are larger than the true energy. (p ξ, p η ) dof u h 2 eng u u h eng,rel(%) slope (2, 3) E E + 0 (3, 3) E E (4, 3) E E (5, 3) E E (6, 3) E E (7, 3) E E (8, 3) E E (9, 3) E E (0, 3) E E (, 3) E E (2, 3) E E (3, 3) E E (4, 3) E E 06 (5, 3) E E 06 (6, 3) E E E

26 Relative error in maximum norm (%) Quadratic Cubic 3 4 Relative error in L2 norm (%) Quadratic Cubic 3 4 h size 0 (a) Rel error in max-norm: h-version, homogenuous BC h size 0 (b) Rel error in L 2 -norm: h-version, homogenuous BC Figure 6: (a) The relative errors in the maximum norm (%) and (b) The relative errors in the L 2 -norm (%) of computed solutions of the Poisson equation (4) with homogeneous Dirichlet boundary condition. The plot of the relative errors versus the mesh sizes h in log-log scale make straight lines whose slopes are 3 and 4, respectively, according as p = 2 and p = 3. Next, we test the proposed mapping method to the Poisson equation with non-homogeneous boundary condition. Example 5.2. Let u 2 (r, θ) = r( + e r ) [ sin ( ) θ + cos 2 ( ) θ + sin 2 ( ) 3θ + cos 2 ( )] 3θ 2 and f = u 2. Then u 2 is the analytic solution of the Poisson equation with non-homogeneous boundary condition: u = f in Ω with u = u 2 on Γ = Ω, (44) that has a crack singularity at (0, 0). This example is different from Example 5. in the following aspects:. The true solution contains the term e r and e r F can not be a polynomial in the parameter space, hence it would be better to have refinement in the radial direction as well as in the angular direction. Thus, the error analysis of Theorem 4. does not fit to this case. 2. The problem has non-homogenous Dirichlet BC along the entire boundary and B-spline basis functions do not have the Kronecker delta property. Hence, we use the least squares method to impose the non-homogenous Dirichlet BC. We expect additional errors in imposing BC. Moreover, the energy norm is not equivalent to the -norm for u 2. 26

27 0 0 Relative error (%) Maximum norm L2 norm Energy norm Polynomial degree Figure 7: The relative error (%) in the maximum norm, the L 2 -norm, and the energy norm of the computed solutions of the Poisson equation (44) with non-homogeneous Dirichlet boundary condition. Therefore, log(/4) is not an expected slope of the convergence profile for Example 5.2. Theorem 4. is concerned with the p-refinement in the ξ-direction only for the error bound because there are no errors in the η-direction for isogeometric analysis. However, for the numerical solutions of Example 5.2, we use the p-refinement in the ξ-direction as well as in the η-direction. Therefore, the error bound in the energy norm for Example 5.2 should be stated in terms of p ξ, p η and h. Nevertheless, we observe that the convergence profile in the maximum norm, the L 2 -norm and the energy norm depicted in Fig. 7 are almost the same as those of Example 5.. Numerical data corresponding to Fig. 7 are in Tables and 2 in appendix. Next we test our mapping methods to other prominent singularity problems. However, the error analysis of Theorem 4. is not applicable to these cases. Further investigation of NURBS mapping for error analysis of the general cases in which the p-refinement in the radial direction as well as in the angular direction is required. 5.2 The Motz problem The Motz problem is a well known benchmarking problem containing a point singularity ([, 4, 7] and references within). It contains a jump boundary data singularity of type O(r /2 ) at the origin (0, 0). 27

28 (,) u n = 0 Γ 3 (,) u n = 0 Γ 4 Γ 2 u = 500 (,0) Γ 5 Γ (0,0) (,0) u u = 0 n = 0 (a) Domain & jump boundary data at (0, 0) (b) control points Figure 8: (a) The domain of the Motz problem and boundary conditions. (b) The control points for NURBS geometrical mapping to deal with the Motz problem. The eight physical elements are formed by solid curves and lines, and 8 control points are repeated at the origin (0, 0). Example 5.3. (Motz problem) Let Ω = [, ] [0, ]. Consider the following Laplace s equation with mixed boundary conditions: u = 0 in Ω, u = 500 on Γ 2, u = 0 on Γ 5, u n = 0 on Γ Γ 3 Γ 4, where Γ = [0, ] {0}, Γ 2 = {} [0, ], Γ 3 = [, ] {}, Γ 4 = { } [0, ], and Γ 5 = [, 0] {0}, as shown in Fig. 8. The true solution of the Motz problem can be expressed asymptotically as follow: u 3 (r, θ) = A k r (/2+k) cos ((/2 + k)θ). (45) k=0 Oh et al.([7]) introduced a benchmarking numerical solution of this problem by accurately estimating the first 50 coefficients of the solution (45). We use this computed solution (the partial sum of the first 50 terms of (45)) as the true solution of the Motz problem for the errors of the computed solutions. The control points and the mesh on the physical domain are illustrated in Fig. 8(b) in which to generate the singularity of the form r 2 +k ψ(θ), 8 control points are located at the origin (0, 0). The relative error (%) in the maximum norm, the L 2 -norm and the energy norm are shown in Table 5 and are depicted in Fig. 9. We observe that the proposed mapping method yields as accurate numerical solutions as those in [4, 7] at lower DOF. The control points and the corresponding weights to construct the geometrical mapping to deal with the Motz problem are listed in Table 3 in appendix. 28

29 Table 5: The relative error(%) in the maximum norm, the L 2 -norm and the energy norm of the computed solutions (with respect to a p-refinement) of the Motz problem are listed. (p ξ, p η ) dof u 3 u h 3,rel(%) u 3 u h 3 L 2,rel(%) u 3 u h 3 eng,rel(%) (2, 2) E E E + 00 (3, 3) E E E 0 (4, 4) E E E 02 (5, 5) 24.6E E E 02 (6, 6) E 04.05E E 03 (7, 7) E E E 04 (8, 8) E E E 04 (9, 9) E E E 05 (0, 0) E E E 05 (, ) E E 08.62E 05 (2, 2) E E E 05 (3, 3) E E E 05 (4, 4) E E E 05 (5, 5) E E E 05 (6, 6) E E E Relative errors (%) Maximum Norm L2 Norm Energy Norm Polynomial degree Figure 9: The relative error (%) in the maximum norm, the L 2 -norm, and the energy norm of computed solutions of the Motz problem are depicted. 29

The Closed Form Reproducing Polynomial Particle Shape Functions for Meshfree Particle Methods

The Closed Form Reproducing Polynomial Particle Shape Functions for Meshfree Particle Methods The Closed Form Reproducing Polynomial Particle Shape Functions for Meshfree Particle Methods by Hae-Soo Oh Department of Mathematics, University of North Carolina at Charlotte, Charlotte, NC 28223 June

More information

Isogeometric Analysis:

Isogeometric Analysis: Isogeometric Analysis: some approximation estimates for NURBS L. Beirao da Veiga, A. Buffa, Judith Rivas, G. Sangalli Euskadi-Kyushu 2011 Workshop on Applied Mathematics BCAM, March t0th, 2011 Outline

More information

Constructions of C 1 -Basis Functions for Numerical Solutions of the Fourth-Order Partial Differential Equations

Constructions of C 1 -Basis Functions for Numerical Solutions of the Fourth-Order Partial Differential Equations Constructions of C 1 -Basis Functions for Numerical Solutions of the Fourth-Order Partial Differential Equations Hae-Soo Oh 1 and Jae Woo Jeong 2 1 Department of Mathematics and Statistics, University

More information

Condition number estimates for matrices arising in the isogeometric discretizations

Condition number estimates for matrices arising in the isogeometric discretizations www.oeaw.ac.at Condition number estimates for matrices arising in the isogeometric discretizations K. Gahalaut, S. Tomar RICAM-Report -3 www.ricam.oeaw.ac.at Condition number estimates for matrices arising

More information

ISOGEOMETRIC COLLOCATION METHOD FOR ELASTICITY PROBLEMS CONTAINING SINGULARITIES. Puja Rattan

ISOGEOMETRIC COLLOCATION METHOD FOR ELASTICITY PROBLEMS CONTAINING SINGULARITIES. Puja Rattan ISOGEOMETRIC COLLOCATION METHOD FOR ELASTICITY PROBLEMS CONTAINING SINGULARITIES by Puja Rattan A dissertation proposal submitted to the faculty of The University of North Carolina at Charlotte in partial

More information

Benchmarking high order finite element approximations for one-dimensional boundary layer problems

Benchmarking high order finite element approximations for one-dimensional boundary layer problems Benchmarking high order finite element approximations for one-dimensional boundary layer problems Marcello Malagù,, Elena Benvenuti, Angelo Simone Department of Engineering, University of Ferrara, Italy

More information

Isogeometric mortaring

Isogeometric mortaring Isogeometric mortaring E. Brivadis, A. Buffa, B. Wohlmuth, L. Wunderlich IMATI E. Magenes - Pavia Technical University of Munich A. Buffa (IMATI-CNR Italy) IGA mortaring 1 / 29 1 Introduction Splines Approximation

More information

The Reproducing Singularity Particle Shape Functions for Problems Containing Singularities

The Reproducing Singularity Particle Shape Functions for Problems Containing Singularities The Reproducing Singularity Particle Shape Functions for Problems Containing Singularities by Hae-Soo Oh Department of Mathematics, University of North Carolina at Charlotte, Charlotte, NC 28223 Jae Woo

More information

Extraction Formulas of Stress Intensity Factors for fourth-order problems containing crack singularities

Extraction Formulas of Stress Intensity Factors for fourth-order problems containing crack singularities Extraction Formulas of Stress Intensity Factors for fourth-order problems containing crack singularities Seokchan Kim, Birce Palta 2, Hae-Soo Oh 2, 1 Department of Mathematics Changwon National University,

More information

Reproducing polynomial particle methods for boundary integral equations

Reproducing polynomial particle methods for boundary integral equations DOI 10.1007/s00466-011-0581-x ORIGINAL PAPER Reproducing polynomial particle methods for boundary integral equations Hae-Soo Oh Christopher Davis June G. Kim YongHoon Kwon Received: 18 November 2010 /

More information

Isogeometric Analysis with Geometrically Continuous Functions on Two-Patch Geometries

Isogeometric Analysis with Geometrically Continuous Functions on Two-Patch Geometries Isogeometric Analysis with Geometrically Continuous Functions on Two-Patch Geometries Mario Kapl a Vito Vitrih b Bert Jüttler a Katharina Birner a a Institute of Applied Geometry Johannes Kepler University

More information

Yongdeok Kim and Seki Kim

Yongdeok Kim and Seki Kim J. Korean Math. Soc. 39 (00), No. 3, pp. 363 376 STABLE LOW ORDER NONCONFORMING QUADRILATERAL FINITE ELEMENTS FOR THE STOKES PROBLEM Yongdeok Kim and Seki Kim Abstract. Stability result is obtained for

More information

Numerical Solutions to Partial Differential Equations

Numerical Solutions to Partial Differential Equations Numerical Solutions to Partial Differential Equations Zhiping Li LMAM and School of Mathematical Sciences Peking University Numerical Methods for Partial Differential Equations Finite Difference Methods

More information

INTRODUCTION TO FINITE ELEMENT METHODS

INTRODUCTION TO FINITE ELEMENT METHODS INTRODUCTION TO FINITE ELEMENT METHODS LONG CHEN Finite element methods are based on the variational formulation of partial differential equations which only need to compute the gradient of a function.

More information

arxiv: v1 [math.na] 29 Feb 2016

arxiv: v1 [math.na] 29 Feb 2016 EFFECTIVE IMPLEMENTATION OF THE WEAK GALERKIN FINITE ELEMENT METHODS FOR THE BIHARMONIC EQUATION LIN MU, JUNPING WANG, AND XIU YE Abstract. arxiv:1602.08817v1 [math.na] 29 Feb 2016 The weak Galerkin (WG)

More information

1 Discretizing BVP with Finite Element Methods.

1 Discretizing BVP with Finite Element Methods. 1 Discretizing BVP with Finite Element Methods In this section, we will discuss a process for solving boundary value problems numerically, the Finite Element Method (FEM) We note that such method is a

More information

Numerical methods for PDEs FEM convergence, error estimates, piecewise polynomials

Numerical methods for PDEs FEM convergence, error estimates, piecewise polynomials Platzhalter für Bild, Bild auf Titelfolie hinter das Logo einsetzen Numerical methods for PDEs FEM convergence, error estimates, piecewise polynomials Dr. Noemi Friedman Contents of the course Fundamentals

More information

CLASSIFICATION AND PRINCIPLE OF SUPERPOSITION FOR SECOND ORDER LINEAR PDE

CLASSIFICATION AND PRINCIPLE OF SUPERPOSITION FOR SECOND ORDER LINEAR PDE CLASSIFICATION AND PRINCIPLE OF SUPERPOSITION FOR SECOND ORDER LINEAR PDE 1. Linear Partial Differential Equations A partial differential equation (PDE) is an equation, for an unknown function u, that

More information

Chapter 5 A priori error estimates for nonconforming finite element approximations 5.1 Strang s first lemma

Chapter 5 A priori error estimates for nonconforming finite element approximations 5.1 Strang s first lemma Chapter 5 A priori error estimates for nonconforming finite element approximations 51 Strang s first lemma We consider the variational equation (51 a(u, v = l(v, v V H 1 (Ω, and assume that the conditions

More information

PhD dissertation defense

PhD dissertation defense Isogeometric mortar methods with applications in contact mechanics PhD dissertation defense Brivadis Ericka Supervisor: Annalisa Buffa Doctor of Philosophy in Computational Mechanics and Advanced Materials,

More information

A WEAK GALERKIN MIXED FINITE ELEMENT METHOD FOR BIHARMONIC EQUATIONS

A WEAK GALERKIN MIXED FINITE ELEMENT METHOD FOR BIHARMONIC EQUATIONS A WEAK GALERKIN MIXED FINITE ELEMENT METHOD FOR BIHARMONIC EQUATIONS LIN MU, JUNPING WANG, YANQIU WANG, AND XIU YE Abstract. This article introduces and analyzes a weak Galerkin mixed finite element method

More information

Scientific Computing WS 2018/2019. Lecture 15. Jürgen Fuhrmann Lecture 15 Slide 1

Scientific Computing WS 2018/2019. Lecture 15. Jürgen Fuhrmann Lecture 15 Slide 1 Scientific Computing WS 2018/2019 Lecture 15 Jürgen Fuhrmann juergen.fuhrmann@wias-berlin.de Lecture 15 Slide 1 Lecture 15 Slide 2 Problems with strong formulation Writing the PDE with divergence and gradient

More information

Discrete Maximum Principle for a 1D Problem with Piecewise-Constant Coefficients Solved by hp-fem

Discrete Maximum Principle for a 1D Problem with Piecewise-Constant Coefficients Solved by hp-fem The University of Texas at El Paso Department of Mathematical Sciences Research Reports Series El Paso, Texas Research Report No. 2006-10 Discrete Maximum Principle for a 1D Problem with Piecewise-Constant

More information

u xx + u yy = 0. (5.1)

u xx + u yy = 0. (5.1) Chapter 5 Laplace Equation The following equation is called Laplace equation in two independent variables x, y: The non-homogeneous problem u xx + u yy =. (5.1) u xx + u yy = F, (5.) where F is a function

More information

Scientific Computing WS 2017/2018. Lecture 18. Jürgen Fuhrmann Lecture 18 Slide 1

Scientific Computing WS 2017/2018. Lecture 18. Jürgen Fuhrmann Lecture 18 Slide 1 Scientific Computing WS 2017/2018 Lecture 18 Jürgen Fuhrmann juergen.fuhrmann@wias-berlin.de Lecture 18 Slide 1 Lecture 18 Slide 2 Weak formulation of homogeneous Dirichlet problem Search u H0 1 (Ω) (here,

More information

On an Approximation Result for Piecewise Polynomial Functions. O. Karakashian

On an Approximation Result for Piecewise Polynomial Functions. O. Karakashian BULLETIN OF THE GREE MATHEMATICAL SOCIETY Volume 57, 010 (1 7) On an Approximation Result for Piecewise Polynomial Functions O. arakashian Abstract We provide a new approach for proving approximation results

More information

A local-structure-preserving local discontinuous Galerkin method for the Laplace equation

A local-structure-preserving local discontinuous Galerkin method for the Laplace equation A local-structure-preserving local discontinuous Galerkin method for the Laplace equation Fengyan Li 1 and Chi-Wang Shu 2 Abstract In this paper, we present a local-structure-preserving local discontinuous

More information

Applied/Numerical Analysis Qualifying Exam

Applied/Numerical Analysis Qualifying Exam Applied/Numerical Analysis Qualifying Exam August 9, 212 Cover Sheet Applied Analysis Part Policy on misprints: The qualifying exam committee tries to proofread exams as carefully as possible. Nevertheless,

More information

Numerical Solutions to Partial Differential Equations

Numerical Solutions to Partial Differential Equations Numerical Solutions to Partial Differential Equations Zhiping Li LMAM and School of Mathematical Sciences Peking University The Residual and Error of Finite Element Solutions Mixed BVP of Poisson Equation

More information

The Reproducing Singularity Particle Shape Functions for Problems Containing Singularities

The Reproducing Singularity Particle Shape Functions for Problems Containing Singularities The Reproducing Singularity Particle Shape Functions for Problems Containing Singularities by Hae-Soo Oh Department of Mathematics, University of North Carolina at Charlotte, Charlotte, NC 28223 Jae Woo

More information

Variational Formulations

Variational Formulations Chapter 2 Variational Formulations In this chapter we will derive a variational (or weak) formulation of the elliptic boundary value problem (1.4). We will discuss all fundamental theoretical results that

More information

We consider the problem of finding a polynomial that interpolates a given set of values:

We consider the problem of finding a polynomial that interpolates a given set of values: Chapter 5 Interpolation 5. Polynomial Interpolation We consider the problem of finding a polynomial that interpolates a given set of values: x x 0 x... x n y y 0 y... y n where the x i are all distinct.

More information

CHAPTER 3 Further properties of splines and B-splines

CHAPTER 3 Further properties of splines and B-splines CHAPTER 3 Further properties of splines and B-splines In Chapter 2 we established some of the most elementary properties of B-splines. In this chapter our focus is on the question What kind of functions

More information

LECTURE # 0 BASIC NOTATIONS AND CONCEPTS IN THE THEORY OF PARTIAL DIFFERENTIAL EQUATIONS (PDES)

LECTURE # 0 BASIC NOTATIONS AND CONCEPTS IN THE THEORY OF PARTIAL DIFFERENTIAL EQUATIONS (PDES) LECTURE # 0 BASIC NOTATIONS AND CONCEPTS IN THE THEORY OF PARTIAL DIFFERENTIAL EQUATIONS (PDES) RAYTCHO LAZAROV 1 Notations and Basic Functional Spaces Scalar function in R d, d 1 will be denoted by u,

More information

We denote the space of distributions on Ω by D ( Ω) 2.

We denote the space of distributions on Ω by D ( Ω) 2. Sep. 1 0, 008 Distributions Distributions are generalized functions. Some familiarity with the theory of distributions helps understanding of various function spaces which play important roles in the study

More information

From Completing the Squares and Orthogonal Projection to Finite Element Methods

From Completing the Squares and Orthogonal Projection to Finite Element Methods From Completing the Squares and Orthogonal Projection to Finite Element Methods Mo MU Background In scientific computing, it is important to start with an appropriate model in order to design effective

More information

[2] (a) Develop and describe the piecewise linear Galerkin finite element approximation of,

[2] (a) Develop and describe the piecewise linear Galerkin finite element approximation of, 269 C, Vese Practice problems [1] Write the differential equation u + u = f(x, y), (x, y) Ω u = 1 (x, y) Ω 1 n + u = x (x, y) Ω 2, Ω = {(x, y) x 2 + y 2 < 1}, Ω 1 = {(x, y) x 2 + y 2 = 1, x 0}, Ω 2 = {(x,

More information

Numerische Mathematik

Numerische Mathematik Numer. Math. (997) 76: 479 488 Numerische Mathematik c Springer-Verlag 997 Electronic Edition Exponential decay of C cubic splines vanishing at two symmetric points in each knot interval Sang Dong Kim,,

More information

Basic Concepts of Adaptive Finite Element Methods for Elliptic Boundary Value Problems

Basic Concepts of Adaptive Finite Element Methods for Elliptic Boundary Value Problems Basic Concepts of Adaptive Finite lement Methods for lliptic Boundary Value Problems Ronald H.W. Hoppe 1,2 1 Department of Mathematics, University of Houston 2 Institute of Mathematics, University of Augsburg

More information

Quintic deficient spline wavelets

Quintic deficient spline wavelets Quintic deficient spline wavelets F. Bastin and P. Laubin January 19, 4 Abstract We show explicitely how to construct scaling functions and wavelets which are quintic deficient splines with compact support

More information

Traces, extensions and co-normal derivatives for elliptic systems on Lipschitz domains

Traces, extensions and co-normal derivatives for elliptic systems on Lipschitz domains Traces, extensions and co-normal derivatives for elliptic systems on Lipschitz domains Sergey E. Mikhailov Brunel University West London, Department of Mathematics, Uxbridge, UB8 3PH, UK J. Math. Analysis

More information

Non-Conforming Finite Element Methods for Nonmatching Grids in Three Dimensions

Non-Conforming Finite Element Methods for Nonmatching Grids in Three Dimensions Non-Conforming Finite Element Methods for Nonmatching Grids in Three Dimensions Wayne McGee and Padmanabhan Seshaiyer Texas Tech University, Mathematics and Statistics (padhu@math.ttu.edu) Summary. In

More information

A proof for the full Fourier series on [ π, π] is given here.

A proof for the full Fourier series on [ π, π] is given here. niform convergence of Fourier series A smooth function on an interval [a, b] may be represented by a full, sine, or cosine Fourier series, and pointwise convergence can be achieved, except possibly at

More information

A very short introduction to the Finite Element Method

A very short introduction to the Finite Element Method A very short introduction to the Finite Element Method Till Mathis Wagner Technical University of Munich JASS 2004, St Petersburg May 4, 2004 1 Introduction This is a short introduction to the finite element

More information

Simple Examples on Rectangular Domains

Simple Examples on Rectangular Domains 84 Chapter 5 Simple Examples on Rectangular Domains In this chapter we consider simple elliptic boundary value problems in rectangular domains in R 2 or R 3 ; our prototype example is the Poisson equation

More information

FEM Convergence for PDEs with Point Sources in 2-D and 3-D

FEM Convergence for PDEs with Point Sources in 2-D and 3-D FEM Convergence for PDEs with Point Sources in 2-D and 3-D Kourosh M. Kalayeh 1, Jonathan S. Graf 2 Matthias K. Gobbert 2 1 Department of Mechanical Engineering 2 Department of Mathematics and Statistics

More information

Space-time Finite Element Methods for Parabolic Evolution Problems

Space-time Finite Element Methods for Parabolic Evolution Problems Space-time Finite Element Methods for Parabolic Evolution Problems with Variable Coefficients Ulrich Langer, Martin Neumüller, Andreas Schafelner Johannes Kepler University, Linz Doctoral Program Computational

More information

Basic Principles of Weak Galerkin Finite Element Methods for PDEs

Basic Principles of Weak Galerkin Finite Element Methods for PDEs Basic Principles of Weak Galerkin Finite Element Methods for PDEs Junping Wang Computational Mathematics Division of Mathematical Sciences National Science Foundation Arlington, VA 22230 Polytopal Element

More information

Chapter 4: Interpolation and Approximation. October 28, 2005

Chapter 4: Interpolation and Approximation. October 28, 2005 Chapter 4: Interpolation and Approximation October 28, 2005 Outline 1 2.4 Linear Interpolation 2 4.1 Lagrange Interpolation 3 4.2 Newton Interpolation and Divided Differences 4 4.3 Interpolation Error

More information

On the relationship of local projection stabilization to other stabilized methods for one-dimensional advection-diffusion equations

On the relationship of local projection stabilization to other stabilized methods for one-dimensional advection-diffusion equations On the relationship of local projection stabilization to other stabilized methods for one-dimensional advection-diffusion equations Lutz Tobiska Institut für Analysis und Numerik Otto-von-Guericke-Universität

More information

SUPERCONVERGENCE PROPERTIES FOR OPTIMAL CONTROL PROBLEMS DISCRETIZED BY PIECEWISE LINEAR AND DISCONTINUOUS FUNCTIONS

SUPERCONVERGENCE PROPERTIES FOR OPTIMAL CONTROL PROBLEMS DISCRETIZED BY PIECEWISE LINEAR AND DISCONTINUOUS FUNCTIONS SUPERCONVERGENCE PROPERTIES FOR OPTIMAL CONTROL PROBLEMS DISCRETIZED BY PIECEWISE LINEAR AND DISCONTINUOUS FUNCTIONS A. RÖSCH AND R. SIMON Abstract. An optimal control problem for an elliptic equation

More information

Robust exponential convergence of hp-fem for singularly perturbed systems of reaction-diffusion equations

Robust exponential convergence of hp-fem for singularly perturbed systems of reaction-diffusion equations Robust exponential convergence of hp-fem for singularly perturbed systems of reaction-diffusion equations Christos Xenophontos Department of Mathematics and Statistics University of Cyprus joint work with

More information

The Weighted Riesz Galerkin Method for Elliptic Boundary Value Problems on Unbounded Domains

The Weighted Riesz Galerkin Method for Elliptic Boundary Value Problems on Unbounded Domains The Weighted Riesz Galerkin Method for Elliptic Boundary Value Problems on Unbounded Domains Hae Soo Oh 1, Bongsoo Jang and Yichung Jou Dept. of Mathematics, University of North Carolina at Charlotte,

More information

We have to prove now that (3.38) defines an orthonormal wavelet. It belongs to W 0 by Lemma and (3.55) with j = 1. We can write any f W 1 as

We have to prove now that (3.38) defines an orthonormal wavelet. It belongs to W 0 by Lemma and (3.55) with j = 1. We can write any f W 1 as 88 CHAPTER 3. WAVELETS AND APPLICATIONS We have to prove now that (3.38) defines an orthonormal wavelet. It belongs to W 0 by Lemma 3..7 and (3.55) with j =. We can write any f W as (3.58) f(ξ) = p(2ξ)ν(2ξ)

More information

FEM Convergence for PDEs with Point Sources in 2-D and 3-D

FEM Convergence for PDEs with Point Sources in 2-D and 3-D FEM Convergence for PDEs with Point Sources in -D and 3-D Kourosh M. Kalayeh 1, Jonathan S. Graf, and Matthias K. Gobbert 1 Department of Mechanical Engineering, University of Maryland, Baltimore County

More information

QUASI-OPTIMAL RATES OF CONVERGENCE FOR THE GENERALIZED FINITE ELEMENT METHOD IN POLYGONAL DOMAINS. Anna L. Mazzucato, Victor Nistor, and Qingqin Qu

QUASI-OPTIMAL RATES OF CONVERGENCE FOR THE GENERALIZED FINITE ELEMENT METHOD IN POLYGONAL DOMAINS. Anna L. Mazzucato, Victor Nistor, and Qingqin Qu QUASI-OPTIMAL RATES OF CONVERGENCE FOR THE GENERALIZED FINITE ELEMENT METHOD IN POLYGONAL DOMAINS By Anna L. Mazzucato, Victor Nistor, and Qingqin Qu IMA Preprint Series #2408 (September 2012) INSTITUTE

More information

2 A Model, Harmonic Map, Problem

2 A Model, Harmonic Map, Problem ELLIPTIC SYSTEMS JOHN E. HUTCHINSON Department of Mathematics School of Mathematical Sciences, A.N.U. 1 Introduction Elliptic equations model the behaviour of scalar quantities u, such as temperature or

More information

L. Levaggi A. Tabacco WAVELETS ON THE INTERVAL AND RELATED TOPICS

L. Levaggi A. Tabacco WAVELETS ON THE INTERVAL AND RELATED TOPICS Rend. Sem. Mat. Univ. Pol. Torino Vol. 57, 1999) L. Levaggi A. Tabacco WAVELETS ON THE INTERVAL AND RELATED TOPICS Abstract. We use an abstract framework to obtain a multilevel decomposition of a variety

More information

Numerical Methods for Two Point Boundary Value Problems

Numerical Methods for Two Point Boundary Value Problems Numerical Methods for Two Point Boundary Value Problems Graeme Fairweather and Ian Gladwell 1 Finite Difference Methods 1.1 Introduction Consider the second order linear two point boundary value problem

More information

Partially Penalized Immersed Finite Element Methods for Parabolic Interface Problems

Partially Penalized Immersed Finite Element Methods for Parabolic Interface Problems Partially Penalized Immersed Finite Element Methods for Parabolic Interface Problems Tao Lin, Qing Yang and Xu Zhang Abstract We present partially penalized immersed finite element methods for solving

More information

Lecture 9 Approximations of Laplace s Equation, Finite Element Method. Mathématiques appliquées (MATH0504-1) B. Dewals, C.

Lecture 9 Approximations of Laplace s Equation, Finite Element Method. Mathématiques appliquées (MATH0504-1) B. Dewals, C. Lecture 9 Approximations of Laplace s Equation, Finite Element Method Mathématiques appliquées (MATH54-1) B. Dewals, C. Geuzaine V1.2 23/11/218 1 Learning objectives of this lecture Apply the finite difference

More information

LECTURE 1: SOURCES OF ERRORS MATHEMATICAL TOOLS A PRIORI ERROR ESTIMATES. Sergey Korotov,

LECTURE 1: SOURCES OF ERRORS MATHEMATICAL TOOLS A PRIORI ERROR ESTIMATES. Sergey Korotov, LECTURE 1: SOURCES OF ERRORS MATHEMATICAL TOOLS A PRIORI ERROR ESTIMATES Sergey Korotov, Institute of Mathematics Helsinki University of Technology, Finland Academy of Finland 1 Main Problem in Mathematical

More information

Introduction. J.M. Burgers Center Graduate Course CFD I January Least-Squares Spectral Element Methods

Introduction. J.M. Burgers Center Graduate Course CFD I January Least-Squares Spectral Element Methods Introduction In this workshop we will introduce you to the least-squares spectral element method. As you can see from the lecture notes, this method is a combination of the weak formulation derived from

More information

Scientific Computing I

Scientific Computing I Scientific Computing I Module 8: An Introduction to Finite Element Methods Tobias Neckel Winter 2013/2014 Module 8: An Introduction to Finite Element Methods, Winter 2013/2014 1 Part I: Introduction to

More information

An introduction to Birkhoff normal form

An introduction to Birkhoff normal form An introduction to Birkhoff normal form Dario Bambusi Dipartimento di Matematica, Universitá di Milano via Saldini 50, 0133 Milano (Italy) 19.11.14 1 Introduction The aim of this note is to present an

More information

Space-time isogeometric analysis of parabolic evolution equations

Space-time isogeometric analysis of parabolic evolution equations www.oeaw.ac.at Space-time isogeometric analysis of parabolic evolution equations U. Langer, S.E. Moore, M. Neumüller RICAM-Report 2015-19 www.ricam.oeaw.ac.at SPACE-TIME ISOGEOMETRIC ANALYSIS OF PARABOLIC

More information

Institut de Recherche MAthématique de Rennes

Institut de Recherche MAthématique de Rennes LMS Durham Symposium: Computational methods for wave propagation in direct scattering. - July, Durham, UK The hp version of the Weighted Regularization Method for Maxwell Equations Martin COSTABEL & Monique

More information

Mesh Grading towards Singular Points Seminar : Elliptic Problems on Non-smooth Domain

Mesh Grading towards Singular Points Seminar : Elliptic Problems on Non-smooth Domain Mesh Grading towards Singular Points Seminar : Elliptic Problems on Non-smooth Domain Stephen Edward Moore Johann Radon Institute for Computational and Applied Mathematics Austrian Academy of Sciences,

More information

Local discontinuous Galerkin methods for elliptic problems

Local discontinuous Galerkin methods for elliptic problems COMMUNICATIONS IN NUMERICAL METHODS IN ENGINEERING Commun. Numer. Meth. Engng 2002; 18:69 75 [Version: 2000/03/22 v1.0] Local discontinuous Galerkin methods for elliptic problems P. Castillo 1 B. Cockburn

More information

Convergence and optimality of an adaptive FEM for controlling L 2 errors

Convergence and optimality of an adaptive FEM for controlling L 2 errors Convergence and optimality of an adaptive FEM for controlling L 2 errors Alan Demlow (University of Kentucky) joint work with Rob Stevenson (University of Amsterdam) Partially supported by NSF DMS-0713770.

More information

Convergence Order Studies for Elliptic Test Problems with COMSOL Multiphysics

Convergence Order Studies for Elliptic Test Problems with COMSOL Multiphysics Convergence Order Studies for Elliptic Test Problems with COMSOL Multiphysics Shiming Yang and Matthias K. Gobbert Abstract. The convergence order of finite elements is related to the polynomial order

More information

Partial Differential Equations

Partial Differential Equations Part II Partial Differential Equations Year 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2015 Paper 4, Section II 29E Partial Differential Equations 72 (a) Show that the Cauchy problem for u(x,

More information

MA8502 Numerical solution of partial differential equations. The Poisson problem: Mixed Dirichlet/Neumann boundary conditions along curved boundaries

MA8502 Numerical solution of partial differential equations. The Poisson problem: Mixed Dirichlet/Neumann boundary conditions along curved boundaries MA85 Numerical solution of partial differential equations The Poisson problem: Mied Dirichlet/Neumann boundar conditions along curved boundaries Fall c Einar M. Rønquist Department of Mathematical Sciences

More information

Littlewood-Paley theory

Littlewood-Paley theory Chapitre 6 Littlewood-Paley theory Introduction The purpose of this chapter is the introduction by this theory which is nothing but a precise way of counting derivatives using the localization in the frequency

More information

1. Introduction. We consider the model problem that seeks an unknown function u = u(x) satisfying

1. Introduction. We consider the model problem that seeks an unknown function u = u(x) satisfying A SIMPLE FINITE ELEMENT METHOD FOR LINEAR HYPERBOLIC PROBLEMS LIN MU AND XIU YE Abstract. In this paper, we introduce a simple finite element method for solving first order hyperbolic equations with easy

More information

A BIVARIATE SPLINE METHOD FOR SECOND ORDER ELLIPTIC EQUATIONS IN NON-DIVERGENCE FORM

A BIVARIATE SPLINE METHOD FOR SECOND ORDER ELLIPTIC EQUATIONS IN NON-DIVERGENCE FORM A BIVARIAE SPLINE MEHOD FOR SECOND ORDER ELLIPIC EQUAIONS IN NON-DIVERGENCE FORM MING-JUN LAI AND CHUNMEI WANG Abstract. A bivariate spline method is developed to numerically solve second order elliptic

More information

CONVERGENCE THEORY. G. ALLAIRE CMAP, Ecole Polytechnique. 1. Maximum principle. 2. Oscillating test function. 3. Two-scale convergence

CONVERGENCE THEORY. G. ALLAIRE CMAP, Ecole Polytechnique. 1. Maximum principle. 2. Oscillating test function. 3. Two-scale convergence 1 CONVERGENCE THEOR G. ALLAIRE CMAP, Ecole Polytechnique 1. Maximum principle 2. Oscillating test function 3. Two-scale convergence 4. Application to homogenization 5. General theory H-convergence) 6.

More information

Partial Differential Equations

Partial Differential Equations M3M3 Partial Differential Equations Solutions to problem sheet 3/4 1* (i) Show that the second order linear differential operators L and M, defined in some domain Ω R n, and given by Mφ = Lφ = j=1 j=1

More information

Finite Element Method for Ordinary Differential Equations

Finite Element Method for Ordinary Differential Equations 52 Chapter 4 Finite Element Method for Ordinary Differential Equations In this chapter we consider some simple examples of the finite element method for the approximate solution of ordinary differential

More information

Finite Elements. Colin Cotter. February 22, Colin Cotter FEM

Finite Elements. Colin Cotter. February 22, Colin Cotter FEM Finite Elements February 22, 2019 In the previous sections, we introduced the concept of finite element spaces, which contain certain functions defined on a domain. Finite element spaces are examples of

More information

Sample Exam 1 KEY NAME: 1. CS 557 Sample Exam 1 KEY. These are some sample problems taken from exams in previous years. roughly ten questions.

Sample Exam 1 KEY NAME: 1. CS 557 Sample Exam 1 KEY. These are some sample problems taken from exams in previous years. roughly ten questions. Sample Exam 1 KEY NAME: 1 CS 557 Sample Exam 1 KEY These are some sample problems taken from exams in previous years. roughly ten questions. Your exam will have 1. (0 points) Circle T or T T Any curve

More information

Point estimates for Green s matrix to boundary value problems for second order elliptic systems in a polyhedral cone

Point estimates for Green s matrix to boundary value problems for second order elliptic systems in a polyhedral cone Maz ya, V. G., Roßmann, J.: Estimates for Green s matrix 1 ZAMM Z. angew. Math. Mech. 00 2004 0, 1 30 Maz ya, V. G.; Roßmann, J. Point estimates for Green s matrix to boundary value problems for second

More information

Lecture Note III: Least-Squares Method

Lecture Note III: Least-Squares Method Lecture Note III: Least-Squares Method Zhiqiang Cai October 4, 004 In this chapter, we shall present least-squares methods for second-order scalar partial differential equations, elastic equations of solids,

More information

Superconvergence of discontinuous Galerkin methods for 1-D linear hyperbolic equations with degenerate variable coefficients

Superconvergence of discontinuous Galerkin methods for 1-D linear hyperbolic equations with degenerate variable coefficients Superconvergence of discontinuous Galerkin methods for -D linear hyperbolic equations with degenerate variable coefficients Waixiang Cao Chi-Wang Shu Zhimin Zhang Abstract In this paper, we study the superconvergence

More information

Maximum norm estimates for energy-corrected finite element method

Maximum norm estimates for energy-corrected finite element method Maximum norm estimates for energy-corrected finite element method Piotr Swierczynski 1 and Barbara Wohlmuth 1 Technical University of Munich, Institute for Numerical Mathematics, piotr.swierczynski@ma.tum.de,

More information

Elimination of self-straining in isogeometric formulations of curved Timoshenko beams in curvilinear coordinates

Elimination of self-straining in isogeometric formulations of curved Timoshenko beams in curvilinear coordinates Available online at www.sciencedirect.com ScienceDirect Comput. Methods Appl. Mech. Engrg. 309 (2016) 680 692 www.elsevier.com/locate/cma Elimination of self-straining in isogeometric formulations of curved

More information

A posteriori error estimation for elliptic problems

A posteriori error estimation for elliptic problems A posteriori error estimation for elliptic problems Praveen. C praveen@math.tifrbng.res.in Tata Institute of Fundamental Research Center for Applicable Mathematics Bangalore 560065 http://math.tifrbng.res.in

More information

On the positivity of linear weights in WENO approximations. Abstract

On the positivity of linear weights in WENO approximations. Abstract On the positivity of linear weights in WENO approximations Yuanyuan Liu, Chi-Wang Shu and Mengping Zhang 3 Abstract High order accurate weighted essentially non-oscillatory (WENO) schemes have been used

More information

Abstract. 1. Introduction

Abstract. 1. Introduction Journal of Computational Mathematics Vol.28, No.2, 2010, 273 288. http://www.global-sci.org/jcm doi:10.4208/jcm.2009.10-m2870 UNIFORM SUPERCONVERGENCE OF GALERKIN METHODS FOR SINGULARLY PERTURBED PROBLEMS

More information

Ultraconvergence of ZZ Patch Recovery at Mesh Symmetry Points

Ultraconvergence of ZZ Patch Recovery at Mesh Symmetry Points Ultraconvergence of ZZ Patch Recovery at Mesh Symmetry Points Zhimin Zhang and Runchang Lin Department of Mathematics, Wayne State University Abstract. The ultraconvergence property of the Zienkiewicz-Zhu

More information

for compression of Boundary Integral Operators. Steven Paul Nixon B.Sc.

for compression of Boundary Integral Operators. Steven Paul Nixon B.Sc. Theory and Applications of the Multiwavelets for compression of Boundary Integral Operators. Steven Paul Nixon B.Sc. Institute for Materials Research School of Computing, Science & Engineering, University

More information

Velocity averaging a general framework

Velocity averaging a general framework Outline Velocity averaging a general framework Martin Lazar BCAM ERC-NUMERIWAVES Seminar May 15, 2013 Joint work with D. Mitrović, University of Montenegro, Montenegro Outline Outline 1 2 L p, p >= 2 setting

More information

Mathematical Methods for Physics and Engineering

Mathematical Methods for Physics and Engineering Mathematical Methods for Physics and Engineering Lecture notes for PDEs Sergei V. Shabanov Department of Mathematics, University of Florida, Gainesville, FL 32611 USA CHAPTER 1 The integration theory

More information

A Mixed Nonconforming Finite Element for Linear Elasticity

A Mixed Nonconforming Finite Element for Linear Elasticity A Mixed Nonconforming Finite Element for Linear Elasticity Zhiqiang Cai, 1 Xiu Ye 2 1 Department of Mathematics, Purdue University, West Lafayette, Indiana 47907-1395 2 Department of Mathematics and Statistics,

More information

The variational collocation method

The variational collocation method The variational collocation method Hector Gomez a,, Laura De Lorenzis b a Departamento de Métodos Matemáticos, Universidade da Coruña, Campus de A Coruña, 15071, A Coruña, Spain. b Institut für Angewandte

More information

R T (u H )v + (2.1) J S (u H )v v V, T (2.2) (2.3) H S J S (u H ) 2 L 2 (S). S T

R T (u H )v + (2.1) J S (u H )v v V, T (2.2) (2.3) H S J S (u H ) 2 L 2 (S). S T 2 R.H. NOCHETTO 2. Lecture 2. Adaptivity I: Design and Convergence of AFEM tarting with a conforming mesh T H, the adaptive procedure AFEM consists of loops of the form OLVE ETIMATE MARK REFINE to produce

More information

High Frequency Scattering by Convex Polygons Stephen Langdon

High Frequency Scattering by Convex Polygons Stephen Langdon Bath, October 28th 2005 1 High Frequency Scattering by Convex Polygons Stephen Langdon University of Reading, UK Joint work with: Simon Chandler-Wilde Steve Arden Funded by: Leverhulme Trust University

More information

Numerical Solutions of Laplacian Problems over L-Shaped Domains and Calculations of the Generalized Stress Intensity Factors

Numerical Solutions of Laplacian Problems over L-Shaped Domains and Calculations of the Generalized Stress Intensity Factors WCCM V Fifth World Congress on Computational Mechanics July 7-2, 2002, Vienna, Austria Eds.: H.A. Mang, F.G. Rammerstorfer, J. Eberhardsteiner Numerical Solutions of Laplacian Problems over L-Shaped Domains

More information

PARTITION OF UNITY FOR THE STOKES PROBLEM ON NONMATCHING GRIDS

PARTITION OF UNITY FOR THE STOKES PROBLEM ON NONMATCHING GRIDS PARTITION OF UNITY FOR THE STOES PROBLEM ON NONMATCHING GRIDS CONSTANTIN BACUTA AND JINCHAO XU Abstract. We consider the Stokes Problem on a plane polygonal domain Ω R 2. We propose a finite element method

More information

Applications of the periodic unfolding method to multi-scale problems

Applications of the periodic unfolding method to multi-scale problems Applications of the periodic unfolding method to multi-scale problems Doina Cioranescu Université Paris VI Santiago de Compostela December 14, 2009 Periodic unfolding method and multi-scale problems 1/56

More information