eliptic grid generation

Upload: padmanathan

Post on 02-Jun-2018

232 views

Category:

Documents


1 download

TRANSCRIPT

  • 8/10/2019 Eliptic Grid Generation

    1/40

    Elliptic Mesh Generation

    An introduction to mesh generation

    Part IV : elliptic meshing

    Jean-Franois Remacle

    Department of Civil Engineering,

    Universit catholique de Louvain, Belgium

    Jean-Franois Remacle Mesh Generation

    http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    2/40

    Elliptic Mesh Generation

    Curvilinear Meshes

    Basic conceptA curvilinear mesh is build using a parametric space

    Sponge analogy

    Let us consider the real coordonates (x,y)and a parametric space (u,v).

    The transformationx =x(u,v)and y =y(u,v)has to be build.

    Jean-Franois Remacle Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    3/40

    Elliptic Mesh Generation

    Curvilinear Meshes

    The transformation transforms a cartesian mesh in the parametric space into

    a structured mesh in the real space with the following properties:

    Two non intersecting lines in the parametric space do not cross in thereal space,

    The mesh adapts to the boundary of on the real space,

    The mesh has not to be orthogonal in the real space.

    Jean-Franois Remacle Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    4/40

    Elliptic Mesh Generation

    Curvilinear Meshes

    The transformation transforms a cartesian mesh in the parametric space into

    a structured mesh in the real space with the following properties:

    Two non intersecting lines in the parametric space do not cross in thereal space,

    The mesh adapts to the boundary of on the real space,

    The mesh has not to be orthogonal in the real space.

    Jean-Franois Remacle Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    5/40

    Elliptic Mesh Generation

    Curvilinear Meshes

    The transformation transforms a cartesian mesh in the parametric space into

    a structured mesh in the real space with the following properties:

    Two non intersecting lines in the parametric space do not cross in thereal space,

    The mesh adapts to the boundary of on the real space,

    The mesh has not to be orthogonal in the real space.

    Jean-Franois Remacle Mesh Generation

    Elli ti M h G ti

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    6/40

    Elliptic Mesh Generation

    Curvilinear Meshes

    The transformation transforms a cartesian mesh in the parametric space into

    a structured mesh in the real space with the following properties:Two non intersecting lines in the parametric space do not cross in the

    real space,

    The mesh adapts to the boundary of on the real space,

    The mesh has not to be orthogonal in the real space.

    The advantage of curvilinear meshes are numerous. They are usually high

    quality/very regular meshes. Yet, the sponge analogy cannot be applied

    easily to complex domains.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    7/40

    Elliptic Mesh Generation

    Transfinite Interpolation

    Let us consider a simply connected 2D manifold that is bounded by 4

    curves ci(t), i=1. . . ,4.

    We consider the four corners Pi,j that are the intersections if curves ciand cj.

    We look for a transformation that maps the unit square in the

    parametric space into the domain of interest.

    We suppose that curves c1 and c3 will be the mapping of the vertical

    segmentsv =0 andv =1.

    We suppose that curvesc2 and

    c4 will be the mapping of the horizontal

    segmentsu =0 andu=1.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    8/40

    Elliptic Mesh Generation

    Transfinite Interpolation

    Let us try the following intuitive formula

    S(u,v) = (1 v)c1(u) +vc3(u) + (1 u)c4(v) +uc2(v)

    Let us see if we get back the boundaries of the domain

    S(u, 0) = c1(u) + (1 u)c4(0) +uc2(0)

    S(u, 1) = c3(u) + (1 u)c4(1) +uc2(1)

    S(0,v) = (1 v)c1(0) +vc3(0) +c4(v)

    S(1,v) = (1 v)c1(1) +vc3(1) +c2(v)

    Let us identify the four corners

    c1(0) = c4(0) = P1,4 c1(1) = c2(0) = P1,2

    c2(1) = c3(1) = P2,3 c3(0) = c4(1) = P3,4

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    9/40

    Elliptic Mesh Generation

    Transfinite Interpolation

    Taking into account the corners

    S(u, 0) = c1(u) + (1 u)P1,4+uP1,2 = c1(u)

    S(u, 1) = c3(u) + (1 u)P3,4+uP2,3 = c3(u)

    S(0,v) = (1 v)P1,4+vP3,4+c4(v) = c4(v)

    S(1,v) = (1 v)P1,2+vP2,3+c2(v) = c2(v)

    Let us try an alternative formula

    S(u,v) = (1 v)c1(u) +vc3(u) + (1 u)c2(v) +uc4(v) (1 u)(1 v)P1,4+uvP2,3+u(1 v)P1,2+ (1 u)vP3,4

    .

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    10/40

    Elliptic Mesh Generation

    Transfinite Interpolation

    S(u,v) = (1 v)c1(u) +vc3(u) + (1 u)c2(v) +uc4(v) (1 u)(1 v)P1,4+uvP2,3+u(1 v)P1,2+ (1 u)vP3,4

    .

    Taking into account corners

    S(u, 0) = c1(u) + (1 u)P1,4+uP1,2 (1 u)P1,4uP1,2= c1(u)

    S(u, 1) = c3(u) + (1 u)P3,4+uP2,3 (1 u)P3,4uP2,3= c3(u)

    S(0,v) = (1 v)

    P1,4+v

    P3,4+

    c4(v) (1 v)P1,4v

    P3,4=

    c4(v)

    S(1,v) = (1 v)P1,2+vP2,3+c2(v) (1 v)P1,2vP2,3= c2(v)

    This formula is correct, it is called the transfinite interpolation formula.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    11/40

    p

    Transfinite Interpolation

    Using a transfinite interpolation does not guarantee that the mapping ispositive, i.e. the relation det |iSj| > 0 is not guaranteed, see some examples

    withgmsh.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    12/40

    Transfinite Interpolation

    Transfinite interpolation exists for 3-sides shapes,

    S(u,v) = uc2(v) + (1 v)c1(u) +vc3(v) u(1 v)P1,2+uvP2,3

    .

    Transfinite interpolation exists for 3D shapes.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    13/40

    Elliptic meshes

    It is not always simple to devise an explicit formula for the mesh

    mapping,

    With the transfinite interpolation, mesh lines may cross each others,

    leading to a wrong mesh

    PDE based mesh have been developped in order to avoid those

    difficulties

    Two kind of approaches are possible, using either an hyperbolic PDE,

    or using anellipticPDE. This last approach is now studied.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    14/40

    Elliptic meshes : Thermal Analogy

    Let us consider the same 4-sided domain as the one we used for

    transfinite meshing,

    We build mesh lines using a thermal analogy : mesh lines are

    superimposed with the isolines of temperature.

    Steady state heat equation has to be solved.

    Two sides of the domain are adiabatical and a constant temperature isimposed on the two other sides.

    !

    " "

    !

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    15/40

    Elliptic meshes : Thermal Analogy

    We have to build up a PDE system that will lead to two families of

    (orthogonal) mesh lines. Fo that purpose, two thermal problems will be

    defined:1 The first family is constructed using the isolines of temperatures1 of

    a thermal problem in which two opposite sides are adiabatical, i.e.

    n1=0. The temperature is set to be constant on each of the two

    other sides: 1=1 on the first side and1=0 on the other side. The

    temperature field1 has to satisfy the Laplace equation inside thedomain :

    21=0

    2 The second family is constructed using the isolines of temperatures 2of the adjoint thermal problem in which adiabatic sides are replaced by

    constant temperature sides and where constant temperature sides arereplaced by adiabatic sides. The co-temperature field2 satisfies the

    Laplace equation as well :

    22=0

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    16/40

    Elliptic meshes : Thermal Analogy

    The two families of mesh lines will have the right properties, thanks to the

    properties of the solutions of the Laplace equation:

    The maximum principle : any solution of the Laplace equation has its

    maximum on the boundaries (no local maxima is allowed),

    The solution is unique,The solution belongs toH1, it is smooth,

    Iso-contours cannot intersect,

    Iso-contours stay inside the domain (obvious for a PDE, not for a

    general mapping),

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    17/40

    Elliptic meshes : Thermal Analogy

    The two families of mesh lines will have the right properties, thanks to the

    properties of the solutions of the Laplace equation:

    The maximum principle : any solution of the Laplace equation has its

    maximum on the boundaries (no local maxima is allowed),

    The solution is unique,

    The solution belongs toH1, it is smooth,

    Iso-contours cannot intersect,

    Iso-contours stay inside the domain (obvious for a PDE, not for a

    general mapping),

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    18/40

    Elliptic meshes : Formulation

    The resulting mesh equations are two boundary value problems. We

    have to find1 and2 solution of

    2i=0 , i=1,2.

    boundary conditions are

    Boundary 1 2c1 1(x,y) =0 n2(x,y) =0

    c2 n1(x,y) =0 2(x,y) =0

    c3 1(x,y) =1 n2(x,y) =0c4 n1(x,y) =0 2(x,y) =1

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    19/40

    Elliptic meshes : Formulation

    Problem : the equations have to be solved in the real space.

    Disappointingly, solving such PDE requires... a mesh!

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    20/40

    Elliptic meshes : Formulation

    This issue is addresded by the inversion of variablesx and y and 1 and

    2,

    The problem is solved in the space of temperatures 1 and2 and the

    unknowns are the positions of the mesh in the real space x and y,

    For a pair of temperatures(1,2), we look the point(x,y)in the

    physical space where this temperature occurs.

    Because of the properties of the Laplace operator, the exist such a

    point(x,y)for any pair(1,2)for which 0 1 1 and for which

    0 2 1

    This is equivalent to the following change of variables

    x=x1=x1(1,2) and y=x2=x2(1,2)

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    21/40

    Elliptic meshes : Formulation

    This issue is addresded by the inversion of variablesx and y and 1 and

    2,

    The problem is solved in the space of temperatures 1 and2 and the

    unknowns are the positions of the mesh in the real space x and y,

    For a pair of temperatures(1,2), we look the point(x,y)in the

    physical space where this temperature occurs.

    Because of the properties of the Laplace operator, the exist such a

    point(x,y)for any pair(1,2)for which 0 1 1 and for which

    0 2 1

    This is equivalent to the following change of variables

    x=x1=x1(1,2) and y=x2=x2(1,2)

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elli i h F l i

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    22/40

    Elliptic meshes : Formulation

    This issue is addresded by the inversion of variablesx and y and 1 and

    2,

    The problem is solved in the space of temperatures 1 and2 and the

    unknowns are the positions of the mesh in the real space x and y,

    For a pair of temperatures(1,2), we look the point(x,y)in the

    physical space where this temperature occurs.

    Because of the properties of the Laplace operator, the exist such a

    point(x,y)for any pair(1,2)for which 0 1 1 and for which

    0 2 1

    This is equivalent to the following change of variables

    x=x1=x1(1,2) and y=x2=x2(1,2)

    Jean-Franois Remacle Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    23/40

    Elliptic Mesh Generation

    Elli ti h F l ti

  • 8/10/2019 Eliptic Grid Generation

    24/40

    Elliptic meshes : Formulation

    This issue is addresded by the inversion of variablesx and y and 1 and

    2,

    The problem is solved in the space of temperatures 1 and2 and the

    unknowns are the positions of the mesh in the real space x and y,

    For a pair of temperatures(1,2), we look the point(x,y)in the

    physical space where this temperature occurs.

    Because of the properties of the Laplace operator, the exist such a

    point(x,y)for any pair(1,2)for which 0 1 1 and for which

    0 2 1

    This is equivalent to the following change of variables

    x=x1=x1(1,2) and y=x2=x2(1,2)

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elli ti h F l ti

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    25/40

    Elliptic meshes : Formulation

    We use the chain derivative rule to compute

    xj=

    2

    j=1

    j

    j

    xiand

    2

    xixk=

    2

    j=1

    2

    l=1

    2

    jl

    l

    xk

    j

    xi

    In expanded form

    2

    x21= 2

    21

    1

    x1

    2

    + 2 2

    12

    1

    x1

    2

    x1

    + 2

    22

    2

    x1

    2

    .

    2

    x22= 2

    21

    1

    x2

    2+ 2

    2

    12

    1

    x2

    2

    x2

    + 2

    22

    2

    x2

    2.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elli ti h F l ti

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    26/40

    Elliptic meshes : Formulation

    The Laplace operator is written as

    2 = 2

    21

    1

    x1

    2

    +

    1

    x2

    2

    + 2 2

    12

    1

    x1

    2

    x1+1

    x2

    2

    x2

    +

    + 2

    22

    2

    x1

    2+

    2

    x2

    2.

    Therefore, in 2D, the PDEs that have to be solved can be written as

    2x12

    12x221

    + 2

    2x1

    122x2

    12

    +

    2x12

    22x222

    =0.

    with

    =

    1

    x1

    2+

    1

    x2

    2, =

    1

    x1

    2

    x1+1

    x2

    2

    x2, =

    2

    x1

    2+

    2

    x2

    2

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Formulation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    27/40

    Elliptic meshes : Formulation

    Note that

    =J2

    x11

    2

    +x12

    2

    =1

    x1

    2

    +1

    x2

    2

    ,

    = J2x11

    x21

    +x12

    x22

    =1

    x1

    2

    x1+1

    x2

    2

    x2,

    =J2

    x21

    2+

    x22

    2=

    2

    x1

    2+

    2

    x2

    2,

    with

    J=det[jxi]

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Formulation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    28/40

    Elliptic meshes : Formulation

    The equations that have to be solved for the mesh problem are coupled and

    non-linear! They can only be solved using numerical methods. There exists

    several techniques to solve such systems. We are not going to detail all the

    methods that are available (this is could be the subject of another course).

    We are going to illustrate some basic principles. Two items will be detailed

    hereThe numerical discretization of the mesh equations in the parametric

    space(1,2),Their resolution.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Formulation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    29/40

    Elliptic meshes : Formulation

    We assume that the mesh in the parametric space is a mnfinite

    difference mesh. The mesh is uniform in the sense that the spacings and

    in both directions are constant.

    f

    fi+1,j fi1,j2

    f

    fi,j+1 fi,j12

    2f

    2

    fi+1,j 2fi,j +fi1,j2

    2f

    2

    fi,j+1 2fi,j +fi,j12

    2f

    fi+1,j+1 fi+1,j1 fi1,j+1+fi1,j1

    4

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Formulation

    http://goforward/http://find/http://goback/
  • 8/10/2019 Eliptic Grid Generation

    30/40

    Elliptic meshes : Formulation

    We recall the mesh equations

    2x1212x22

    1

    + 2 2x112

    2x212

    + 2x122

    2x22

    2

    =0.Their discrete version can be written, for any vertex (i,j)of the mesh in the

    parametric space, as

    [xi+1,j 2xi,j+xi1,j ]+ [xi,j+1 2xi,j+ xi,j1]

    2 [xi+1,j+1 xi1,j+1 xi+1,j1+xi1,j1] = 0

    [yi+1,j 2yi,j+ yi

    1,j] +

    [yi,j+1 2yi,j+yi,j

    1]2 [yi+1,j+1 yi1,j+1 yi+1,j1+yi1,j1] = 0

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Resolution

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    31/40

    Elliptic meshes : Resolution

    With the numerical evaluation of coefficients , and

    = (xi,j+1 xi,j1)

    2 + (yi,j+1 yi,j1)2

    (2)2

    = (xi+1,j xi1,j)

    2 + (yi+1,j yi1,j)2

    (2)2

    = (xi+1,j xi1,j)(xi,j+1 xi,j1)

    (4)2

    +(yi+1,j yi1,j)(yi,j+1 yi,j1)

    (4)2

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Resolution

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    32/40

    Elliptic meshes : Resolution

    Applying the formulation to the mnnodes of the finite difference mesh,

    we obtain a coupled system of non linear equations. In fact, , and

    are all functions of the unknowns, i.e. the position of the nodes in the realspace. Several methods are available for solving such a system. They can be

    classified in two categories:

    1 direct methods,

    2 iterative methods.

    The choice of the method is essentially based on the following criterion

    The CPU time cost and the memory cost,

    The ability of the method to converge to the solution,

    The relative ease of programming !

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Resolution

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    33/40

    Elliptic meshes : Resolution

    Direct methods are not the choice that is usually made for building elliptic

    meshes. Direct methods have the advantage of ensuring the convergence in

    a finite number of step (yet this not true for non-linear systems). Theirprincipal drawback is their memory signature and their (relative)

    complexity.

    Iterative methods have the following advantages:

    1 They are usually easier to code,

    2 They are well adapted to structured grids,

    The following smoothers are often used for building elliptic meshes

    Point Jacobi iteration,

    Block Jacobi iteration (the block may be a line or a column of nodes),

    ADIs, i.e. alternated direction implicit !

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Point Jacobi Iteration

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    34/40

    Elliptic meshes : Point Jacobi Iteration

    We look for the values of the variables x and y, notedx andy at node

    (i,j)verifying

    xi+1,j 2x

    i,j+ x+

    i1,j

    +

    xi,j+1 2x

    i,j+ x+

    i,j1

    2xi+1,j+1 xi1,j+1 x

    +i+1,j1+x

    +i1,j1

    = 0

    yi+1,j 2y

    i,j+y+

    i1,j

    +

    yi,j+1 2y

    i,j+ y+

    i,j1

    2yi+1,j+1 yi1,j+1 y

    +

    i+1,j1+y+

    i1,j1

    = 0

    Here,x andy are for the values ofx and y at the previous sweep and x

    andy are the values at the present sweep. The linearization of the

    coefficients consist in using previous values ofx and y in , and .

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Over relaxation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    35/40

    Elliptic meshes : Over relaxation

    We know thatx and y are updated by doing

    x=x+ (xx) and y=y+ (yy).

    Over relaxing the system consist in doing

    x+ =x+(xx) and y+ =y+(yy).

    with 1. Current values are computed as

    xi,j=xi,j+Cxi,j/ and y

    i,j=yi,j+Cyi,j/.

    with the corrections

    Cxi,j=x+

    i,jxi,j and Cyi,j=y+

    i,jyi,j.

    Jean-Franois Remacle Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    36/40

    Elliptic Mesh Generation

    Elliptic meshes : Boundary conditions

  • 8/10/2019 Eliptic Grid Generation

    37/40

    p y

    We do not want to move the boundary nodes. The way to implement it is

    very simple : do not apply any corrections to those nodes

    Cx1, j = Cxm, j = 0

    Cy1, j = Cym, j = 0

    Those are Dirichlet boundary conditions.

    Jean-Franois Remacle Mesh Generation

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    38/40

    Elliptic Mesh Generation

    Elliptic meshes : Line Jacobi Iteration

  • 8/10/2019 Eliptic Grid Generation

    39/40

    p

    Iterative process can be accelerated using a more implicit method. We

    cound compute one line at a time

    valeurs courantes

    valeurs ltape n + 1

    valeurs ltape n

    j+1

    i-1 i i+1

    j

    j-1

    The resulting system is tridiagonal and ad hoc linear solvers can be used for

    accelerating the process.

    Jean-Franois Remacle Mesh Generation

    Elliptic Mesh Generation

    Elliptic meshes : Line Jacobi Iteration

    http://find/
  • 8/10/2019 Eliptic Grid Generation

    40/40

    p

    The system of equations can be assembled solved using the following finite

    difference formula.

    xi+1,j 2x

    i,j+ x

    i1,j

    +

    xi,j+1 2x

    i,j+ x+i,j1

    2xi+1,j+1 xi1,j+1 x

    +i+1,j1+x

    +i1,j1

    = 0

    yi+1,j 2y

    i,j+ y

    i1,j

    +

    yi,j+1 2y

    i,j+ y+i,j1

    2yi+1,j+1 yi1,j+1 y

    +

    i+1

    ,j1

    +y+i1

    ,j1

    = 0

    Jean-Franois Remacle Mesh Generation

    http://find/