Tutorial/Laplace equation with Neumann boundary condition

Jump to: navigation, search

The considered problem

We propose here to solve a first very simple academic problem with the help of GMSH and GetDP. We considered the unit square $\Omega = [0,1]\times[0,1]$ and we seek $u$, solution of the following problem \begin{equation} \begin{cases}\label{eq:problemU} -\Delta u + u = f & \text{in } \Omega,\\ \displaystyle{\frac{\partial u}{\partial \mathbf{n}} = 0} & \text{on }\partial\Omega, \end{cases} \end{equation} where $\mathbf{n}$ is the unit outward normal of $\Omega$ and $\displaystyle{\Delta = \frac{\partial^2}{\partial x_1^2} + \frac{\partial^2}{\partial x_2^2} }$ is the Laplace operator.

Domain Omega.

To simplify, we suppose that the function $f$ is defined by $$ \forall (x,y)\in [0,1]^2,\qquad f(x,y) = (1+2\pi^2)\cos(\pi x)\cos(\pi y). $$ With this choice of $f$, one can easily show that the unique solution of the problem \eqref{eq:problemU} reads as $$ \forall (x,y)\in[0,1]^2, \qquad u(x,y) = \cos(\pi x)\cos(\pi y). $$ In order to solve problem \eqref{eq:problemU} with the finite element method, we write the weak formulation of problem \eqref{eq:problemU}: \begin{equation}\label{eq:WeakFormulation} \left\{\begin{array}{l} \text{Find } u\in H^1(\Omega) \text{ such that, }\\ \displaystyle{\forall v\in H^1(\Omega), \qquad \int_{\Omega} \nabla u\cdot\nabla v \;{\rm d}\Omega + \int_{\Omega}uv \;{\rm d}\Omega - \int_{\Omega}fv\;{\rm d}\Omega = 0}, \end{array}\right. \end{equation} where $H^1(\Omega)$ is the classical Sobolev space and the functions $v$ are the test functions.

Outline of the program

We give here a (very) detailled solution which is composed by 3 different files:

This (auxiliary) file contains the index number associated with the geometry. This ensures that GMSH and GetDP use the same numbering of the domains.
GMSH file, used to build the domain (the square $\Omega$). The extension ".geo" is mainly used to design a GMSH file
GetDP file, contains the weak formulation \eqref{eq:WeakFormulation} of the problem \eqref{eq:problemU}. The extension ".pro" is associated with GetDP files.

param.geo: the auxiliary file

// File "param.geo"

//Numbers that caracterise the interior of the square (Omega) and its boundary (Gama):
Omega = 1000;

// Three remarks on these numbers :
// - They are arbitrary choosen.
// - They are placed in a separated file to be readable by both GMSH and GetDP.
// - "Gamma" is a special word used by GMSH/GetDP, that is why the boundary is named "Gama", with one "m"...
// Do not forget to let a blank line at the end, this could make GMSH crash...

Direct link to file `LaplacianNeumann/GMSH_GETDP/param.geo'

LaplacianNeumann.geo: creation of the geometry with GMSH

// File "LaplacianNeumann.geo".

// We include the file containing the numbering of the geometry.
// This is usefull at the end of this file, and used to "synchronise" GMSH and GetDP
Include "param.geo";

//Caracteristic length of the finite elements (reffinement is also possible after the mesh is built):
lc = 0.05;
// This parameter could be placed for instance in "param.geo", to separate more easyly the geometry
// 	and the discretization parameters.

// The parameters of the border of the domain :
x_max = 1;
x_min = 0;
y_max= 1;
y_min = 0;

//Creation of the 4 angle points of the domain Omega (=square)
p1 = newp; Point(p1) = {x_min,y_min,0,lc};
p2 = newp; Point(p2) = {x_min,y_max,0,lc};
p3 = newp; Point(p3) = {x_max,y_max,0,lc};
p4 = newp; Point(p4) = {x_max,y_min,0,lc};
// Remarks:
// -"newp" is a GMSH function that give the first available number for describing a point.
// 	For any other entity, like Line, Surface, etc. We recommand the use of "newreg" (see below).
// - By default, GMSH create a 3D domain. The z-coordinate must always be precised.

//The four edges of the square
L1 = newreg; Line(L1) = {p1,p2};
L2 = newreg; Line(L2) = {p2,p3};
L3 = newreg; Line(L3) = {p3,p4};
L4 = newreg; Line(L4) = {p4,p1};

// Line Loop (= boundary of the square)
Bound = newreg; Line Loop(Bound) = {L1,L2,L3,L4};

//Surface of the square
SurfaceOmega = newreg; Plane Surface(SurfaceOmega) = {Bound};

// To conclude, we define the physical entities, that is "what GetDP could see/use".
// "Omega" is a number imported from the file "param.geo".
Physical Surface(Omega) = {SurfaceOmega};
// Do not forget to let a blank line at the end, this could make GMSH crash...

Direct link to file `LaplacianNeumann/GMSH_GETDP/LaplacianNeumann.geo'

LaplacianNeumann.pro: weak formulation

// File LaplacianNeumann.pro

// As in the .geo file, we include the file containing the numbering of the geometry.
Include "param.geo";

// Group
//We now build the "Groups", that is, the geometrical entities, the different domains of computation.
//Here we only need the interior of the open domain Omega
Omega = Region[{Omega}];

// Function
// Pi is a special value in GetDp (=3.1415...)
Coef = Pi;
// Definition of the source function f(x,y)
f[] = (1+2*Coef*Coef)*Cos[Coef*X[]]*Cos[Coef*Y[]];
- the argument (as "x" and "y") are not writen between the bracket [].
Indeed, between the bracket is writen the domain of definition of the function.
- The argument "x" and "y" are here designed by the GetDP inner functions X[] and Y[], which give
respectively the x-coordinate and the y-coordinate of a considered point.
- To define a function globaly (i.e: not only on a subdomain), we write:
f[] = ...
- In our example, we could also have written
f[Omega] = (1+2*Coef*Coef)*Cos[Coef*X[]]*Cos[Coef*Y[]];

Jacobian {
  { Name JVol ;
    Case {
      { Region All ; Jacobian Vol ; }
  { Name JSur ;
    Case {
      { Region All ; Jacobian Sur ; }
  { Name JLin ;
    Case {
      { Region All ; Jacobian Lin ; }

Remark: roughly speacking, we make use of...:
- Jacobian "Vol" as far as the integration domain is of same dimension than the problem (e.g: calculation in a 3D domain, in a 3D problem,
a 2D domain (surface) in a 2 problem, a 1D domain (line) in a 1D problem)
- Jacobian "Sur" when the domain of integration has one dimension less than the global problem (e.g:
calculation on a surface (2D) in a 3D problem, calculation on a line (1D) in a 2D problem).
- Jacobian "Lin" when the domain of integration has 2 dimensions less than the problem. That is,
for example, a calculation on a line (1D) in a 3D problem.
- Here, we just have define some Jacobian, you will see later that we just use the Jacobian "JVol"

//Integration (parameters)
Integration {
  { Name I1 ;
    Case {
      { Type Gauss ;
        Case {
          { GeoElement Point ; NumberOfPoints  1 ; }
          { GeoElement Line ; NumberOfPoints  4 ; }
          { GeoElement Triangle ; NumberOfPoints  6 ; }
          { GeoElement Quadrangle ; NumberOfPoints 7 ; }
          { GeoElement Tetrahedron ; NumberOfPoints 15 ; }
          { GeoElement Hexahedron ; NumberOfPoints 34 ; }

// There is no Constrain because of the Neumann boundary condition
// We go directly to the FuntionSpace

  { Name Vh; Type Form0;
      {Name wn; NameOfCoef vn; Function BF_Node;
    Support Omega; Entity NodesOf[All];}

The space of approximation is called "Vh".
Its type is "Form0", that means "scalar".
The basis functions are called "wn", and the associated coefficients "vn".
In other words, a function "v" of "Vh" can be written as
v(x,y) = sum_{n} vn.wn(x,y)
The functions "wn" are nodale functions P1 ("BF_Node"), supported on the domain Omega ("Support Omega").

//(Weak) Formulation

  {Name LaplacianNeumann; Type FemEquation;
// We decided to call the formulation "LaplacianNeumann".
// Its type is "FemEquation" ("Finite Element Method")
      {Name u; Type Local; NameOfSpace Vh;}
// Here, we introduce a quantity "u", which belongs to the function space Vh, defined above.
      Galerkin{ [Dof{Grad u}, {Grad u}];
        In Omega; Jacobian JVol; Integration I1;}

      Galerkin{ [  Dof{u}, {u}];
       In Omega; Jacobian JVol; Integration I1;}
      Galerkin{ [-f[], {u}];
        In Omega; Jacobian JVol; Integration I1;}
The variationnal formulation is written between the acolades {}, after "Equation".
Let us first give some vocabulary:
- "Galerkin" : GetDP syntaxic word. This could be translated mathematicaly by "integration" (see below)
- Dof{u}: "Degree Of Freedom". This is used to specify that the quantity is the unknown.
			If "Dof" is not written, then "u" is seen as a test function and not as the unknown.
			(Be carreful, the unknown and the test functions has the same name in GetDP !
			The "Dof" is here to distinguish the unknown to the test-functions)

Now, let us detail the program written between the acolades {}, :
          Galerkin{ [Dof{Grad u}, {Grad u}];
          In Omega; Jacobian JVol; Integration Int;}
This could be translated mathematicaly by :
            Integration on "Omega" of Grad(u).Grad(v)
where "u" is the unknown and "v" a test function.
Note the use of the Jacobian JVol(2D problem and integration on a 2D domain).
Moreover the number of integrations point is given in "I1" (see above "Integration").
The total variationnal formulation can be read as :

search "Dof{u}" in V_h such that, for every "{u}" in V_h,
\int_{\Omega} Grad( Dof{u}) . Grad( {u}) d\Omega + \int_{\Omega} Dof{u}.{u} d\Omega - ...
 		... - \int_{\Omega} f.{u} d\Omega = 0
(this is exactly the weak formulation of our problem !)

- Between two "Galerkin", a positive sign "+" is implicitly written
- The sum of all integrals "Galerkin" is equal to 0 (do not forget the "minus" sign of the right hand side)
- Why do we use a volumic jacobian even in a 2D problem? The problem is a 2 dimensional problem and 
the integral is defined on a 2D domain (Omega). If the integral was writen on, e.g, the boundary, 
then the Jacobian "Jsur" would have been used.

// Resolution (of the problem)

  {Name LaplacianNeumann;
// We chose the name LaplacianNeumann for the resolution
// Remark: in GetDP, every entity has a name. The same same can be used for different entities but of different kind, of course
// Here we chose the same name as the formulation, but this is just a choice, no obligation !
      {Name Syst; NameOfFormulation LaplacianNeumann;}
// A system is linked to a weak formulation
// Here, we only have one weak formulation, which is "LaplacianNeumann"
      Generate[Syst]; Solve[Syst]; SaveSolution[Syst];
/* When we launch GetDP, the program will ask the user to chose a Resolution.
When calling the Resolution "LaplacianNeumann", GetDP will...
- Generate the system
- Solve the system
- Save the solution
Note that GetDP respects the order of the operation !

//Post Processing

  {Name LaplacianNeumann; NameOfFormulation LaplacianNeumann;
      {Name u; Value {Local{[{u}];In Omega;Jacobian JVol;}}}

The name of the PostProcessing is LaplacianNeumann.
It call the weak formulation LaplacianNeumann, then take the solution "u" and compute it on all the domain Omega, by the operation:
    Value {Local{[{u}];In Carre; Jacobian JVol;}}}
This means that "u" is the interpolate of the solution "Dof{u}" computed in the weak formulation "LaplacianNeumann"

Remark: Again, we chose the same name, but this is just a choice, this has no influence on GetDP resolution.

//Post Operation

  {Name Map_u; NameOfPostProcessing LaplacianNeumann;
      Print[u, OnElementsOf Omega, File "u_Neumann.pos"];

The only PostOperation we write is to display "u" introduced in the PostProcessing.
This PostOperation is called "Map_u".
When launching GetDP, it will ask the user to chose :
- the Resolution
- the PostOperation

Direct link to file `LaplacianNeumann/GMSH_GETDP/LaplacianNeumann.pro'

How to launch

All the files (.geo and .pro) must be located in the same directory.

Meshing the domain
  • Graphical way: launch GMSH and open "LaplacianNeumann.geo". Then chose the "mesh" menu and click on "2D". Finally, do not forget to save the mesh by clicking on the "Save" button
  • With a terminal: go to the directory and then type:
gmsh -LaplacianNeumann.geo -algo iso -2
Note that here we chose the isobare algorithm ("iso"). One can also chose the following algorithms:

After the mesh is built, a file LaplacianNeumann.msh should have been created in the directory.

Solving the problem with GetDP

In a Terminal, type (in the right directory)

getdp LaplacianNeumann.pro -solve -pos

GetDP will then propose the Resolution ("-solve") and the PostOperation ("-pos"). This should build a file called "u_Neumann.pos".

Remarks :

the choice can be pre-selected by typing:
getdp LaplacianNeumann.pro -solve#1 -pos#1
There exists other options, like the choice of the linear solver, \ldots
Showing the result

Finally, open the file "u_Neumann.pos" with GMSH, with the graphical way ("open \ldots') of with a terminal (by typing "gmsh u_Neumann.pos" in a terminal in the right directory).


Here is the result obtained with GMSH/GetDP. On the left is an example of the mesh of $\Omega$ and on the right, the modulus of the solution $u$.

Mesh of Omega. Modulus of u.

All files

A zipfile containing all files (.geo, .pro, .mesh and .pos) can be downloaded here.