FEM Problems

Below are the definitions of the types which specify problems. Some general notes are:

(t,x) vs (t,x,y): Mathematically one normally specifies equations in 2D as $f(t,x,y)$. However, in this code we use x as a vector. Thus you can think of $x$=x[:,1] and $y$=x[:,2]. Thus input equations are of the form f(x,t) no matter the dimension. If time is not included in the problem (for example, a Poisson equation problem), then we use f(x). An example is the equation $u(x,y)= sin(2πx)cos(2πy)/(8π^2)$ would be specified as sol(x) = sin(2π.*x[:,1]).*cos(2π.*x[:,2])/(8π*π).
Linearity: If the equation has a linear term, they are specified with functions f(t,x). If it is nonlinear, it is specified with functions f(t,x,u). The boundary conditions are always (t,x)
Stochastic: By default the equation is deterministic. For each equation, one can specify a σ term which adds a stochastic $σ(t,x,u)dW_t$ term to the equation (or with $σ(t,x)dW_t$ if linear, must match f). $dW_t$ corresponds to the type of noise which is chosen. By default this is space-time Gaussian white noise.

Poisson Equation Problem

Wraps the data that defines a 2D linear Poisson equation problem:

\[-Δu = f\]

with bounday conditions gD on the Dirichlet boundary and gN on the Neumann boundary. Linearity is determined by whether the forcing function f is a function of one variable (x) or two (u,x) (with x=[:,1] and y=[:,2]).

If the keyword σ is given, then this wraps the data that defines a 2D stochastic heat equation

\[-Δu = f + σdW\]

Constructors

PoissonProblem(f,analytic,Du,mesh): Defines the Dirichlet problem with analytical solution analytic, solution gradient Du = [u_x,u_y], and forcing function f

PoissonProblem(u0,f,mesh): Defines the problem with initial value u0 (as a function) and f. If your initial data is a vector, wrap it as u0(x) = vector.

Note: If all functions are of (x), then the program assumes it's linear. Write your functions using the math to program syntax translation: $x$= x[:,1] and $y$= x[:,2]. Use f=f(u,x) and σ=σ(u,x) (if specified) for nonlinear problems (with the boundary conditions still (x)). Systems of equations can be specified with u_i = u[:,i] as the ith variable. See the example problems for more help.

Keyword Arguments

gD = Dirichlet boundary function
gN = Neumann boundary function
σ = The function which multiplies the noise $dW$. By default σ=0.
noisetype = A string which specifies the type of noise to be generated. By default noisetype=:White for Gaussian Spacetime White Noise.
numvars = The number of variables in the Poisson system. Automatically calculated in many cases.
D = Vector of diffusion coefficients. Defaults is D=ones(1,numvars).

Heat Equation Problem

Wraps the data that defines a 2D heat equation problem:

\[u_t = Δu + f\]

with bounday conditions gD on the Dirichlet boundary and gN on the Neumann boundary. Linearity is determined by whether the forcing function f is a function of two variables (t,x) or three (t,x,u) (with x=[:,1] and y=[:,2]).

If the keyword σ is given, then this wraps the data that defines a 2D stochastic heat equation.

\[u_t = Δu + f + σdW_t\]

Constructors

HeatProblem(analytic,Du,f,mesh): Defines the Dirichlet problem with solution analytic, solution gradient Du = [u_x,u_y], and the forcing function f.
HeatProblem(u0,f,mesh): Defines the problem with initial value u0 (as a function) and f. If your initial data is a vector, wrap it as u0(x) = vector.

Note: If all functions are of (t,x), then the program assumes it's linear. Write your functions using the math to program syntax translation: $x$= x[:,1] and $y$= x[:,2]. Use f=f(t,x,u) and σ=σ(t,x,u) (if specified) for nonlinear problems (with the boundary conditions still (t,x)). Systems of equations can be specified with u_i = u[:,i] as the ith variable. See the example problems for more help.

Keyword Arguments

gD = Dirichlet boundary function
gN = Neumann boundary function
σ = The function which multiplies the noise dW. By default σ=0.
noisetype = A string which specifies the type of noise to be generated. By default noisetype=:White for Gaussian Spacetime White Noise.
numvars = Number of variables in the system. Automatically calculated from u0 in most cases.
D = Array which defines the diffusion coefficients. Default is D=ones(1,numvars).

Example Problems

Examples problems can be found in DiffEqProblemLibrary.jl.

To use a sample problem, you need to do:

# Pkg.add("DiffEqProblemLibrary")
using DiffEqProblemLibrary

Poisson Equation

DiffEqProblemLibrary.prob_poisson_birthdeathinteractingsystem — Constant.

Nonlinear Poisson equation with $f(u)=1-u/2$ and $f(v)=.5u-v$ and initial condition homogenous 1/2. Corresponds to the steady state of a humogenous reaction-diffusion equation with the same $f$.

source

DiffEqProblemLibrary.prob_poisson_noisywave — Constant.

Problem with deterministic solution: $u(x,y)= \sin(2πx)\cos(2πy)/(8π^2)$ and additive noise $σ(x,y)=5$

source

DiffEqProblemLibrary.prob_poisson_birthdeathsystem — Constant.

Nonlinear Poisson equation with $f(u)=1-u/2$ and $f(v)=1-v$ and initial condition homogenous 1/2. Corresponds to the steady state of a humogenous reaction-diffusion equation with the same $f$.

source

DiffEqProblemLibrary.prob_poisson_wave — Constant.

Problem defined by the solution: $u(x,y)= \sin(2πx)\cos(2πy)/(8π^2)$

source

DiffEqProblemLibrary.prob_poisson_birthdeath — Constant.

Nonlinear Poisson equation with $f(u)=1-u/2$. Corresponds to the steady state of a humogenous reaction-diffusion equation with the same $f$.

source

Heat Equation

DiffEqProblemLibrary.prob_femheat_birthdeathsystem — Constant.

Homogenous reaction-diffusion which starts at 1/2 and solves the system $f(u)=1-u/2$ and $f(v)=1-v$

source

DiffEqProblemLibrary.prob_femheat_birthdeathinteractingsystem — Constant.

Homogenous reaction-diffusion which starts with 1/2 and solves the system $f(u)=1-u/2$ and $f(v)=.5u-v$

source

DiffEqProblemLibrary.prob_femheat_diffuse — Constant.

Example problem defined by the solution:

\[u(x,y,t)=\exp(-10((x-\frac{1}{2})^2 + (y-\frac{1}{2})^2 )-t)\]

This is a Gaussian centered at $(\frac{1}{2},\frac{1}{2})$ which diffuses over time.

source

DiffEqProblemLibrary.prob_femheat_stochasticbirthdeath — Constant.

Homogenous stochastic reaction-diffusion problem which starts with 0 and solves with $f(u)=1-u/2$ with noise $σ(u)=10u^2$

source

DiffEqProblemLibrary.prob_femheat_moving — Constant.

Example problem defined by the solution:

\[u(x,y,t)=\frac{1}{10}(1-\exp(-100(t-\frac{1}{2})^2))\exp(-25((x-t+0.5)^2 + (y-t+0.5)^2))\]

This will have a mound which moves across the screen. Good animation test.

source

DiffEqProblemLibrary.prob_femheat_pure — Constant.

Example problem which starts with a Dirac δ cenetered at (0.5,0.5) and solves with $f=gD=0$. This gives the Green's function solution.

source

DiffEqProblemLibrary.prob_femheat_birthdeath — Constant.

Homogenous reaction-diffusion problem which starts with 0 and solves with $f(u)=1-u/2$

source