# System of PDEs which is tricky for PDEPE

54 views (last 30 days)
Matthew Hunt on 23 May 2018
Answered: Matthew Hunt on 15 Jun 2018
I have a system of PDEs, mainly diffusion equations of the form:
T_t-(k(x)(T_x)_x=a*E^2
c_t-(D(x)*c_x)_x=d*(E_x+c_x-T_x)
(epsilon*E)_x=-b*c
Where a and d are constants and _t,_x represent partial differentiation w.r.t. t and x respectively. In terms of pdepe, I would have: f=0 for the third equation. Would this cause errors in the code?
I'm also struggling to see how I can input the boundary conditions.
Any suggestions?

Torsten on 24 May 2018
A boundary condition for E can only be given in one end point of the interval, thus at x=0 or x=1.
Matthew Hunt on 24 May 2018
That is the boundary condition, you can ignore one of them.
Bill Greene on 24 May 2018
Do you mean T=0 at the boundaries? What about the initial conditions on the three variables? At t=0, the boundary and initial conditions must be consistent and also satisfy your differential equations.

Torsten on 23 May 2018
Edited: Torsten on 23 May 2018
"pdepe" is designed to solve systems of parabolic-elliptic pdes. Your third pde is hyperbolic in nature. Thus "pdepe" is not suited to solve your system.
You will have to discretize your equations in space and solve the resulting system of ordinary differential equations using ODE15S.
Look up "method-of-lines" for more details.
Best wishes
Torsten.

Matthew Hunt on 23 May 2018
Cheers,
I had a feeling that this might be the case. You will note however that it is possible to solve the hyperbolic equation completely thus giving E as a function of c essentially. Can I drop that equation somehow and use the integral of c as part of the code for is it too hard to do that?
Torsten on 23 May 2018
Since "pdefun" is called only for a single value of x, you don't have the complete vector for c available. Thus no chance to calculate E, I guess.

Torsten on 24 May 2018
You could try to use "pdepe" with the third equation differentiated:
(epsilon*E)_xx=-b*c_x
with Dirichlet boundary condition
E = E0
at one end of the interval and
(epsilon*E)_x + b*c=0
at the opposite end.
Best wishes
Torsten.

Matthew Hunt on 24 May 2018
E is an electric field, there is no magnetic field, so there exists a potential for it E=phi_x, say. However, the derivative of E also appears in the diffusion equation for c, and this will entail a new form of "f" in the pdepe routine which does not fit in with the boundary conditions for c.
I was thinking about using a crank-Nicholson scheme for these equations.
Torsten on 24 May 2018
But the E_x and T_x terms in the diffusion equation for c will be put in the source term s for this equation, not in the f-term.
Matthew Hunt on 24 May 2018
That still won't work as it won't fit the boundary equation for E.

Matthew Hunt on 25 May 2018
So I managed a work around to get pdepe to work with my system and the code I'm using is:
%This is the simplistic model of a lithium ion battery model which I
%cobbled together knowing very little about batteries. The constants and so
%forth come from experiment.
%Define some global constants and parameters:
global rho; rho=1;global c_th; c_th=1;
global a_1; a_1=1;global a_2; a_2=1;global b; b=1; global c_1; c_1=1;global c_2; c_2=1;
global I_app; I_app=1.5;
global k;global D; global epsilon;
k=1; %Thermal conductivity, most likely a function.
epsilon=1; %Electrical conductivity, most likely a function
D=1; %lithium ion diffusion, most likely a function
m=0;
x=linspace(0,1,100);
t=linspace(0,1,50);
%Initial conditions;
global T_0; global c_0; global phi_0;
T_0=x./x;
c_0=smooth_step(x,0,1/3);
E_0=I_app/(epsilon)-(b/epsilon)*cumtrapz(x,c_0);
phi_0=cumtrapz(x,E_0);
sol = pdepe(m,@battery_GE,@battery_ic,@battery_bc,x,t);
T = sol(:,:,1);
E = sol(:,:,2);
c = sol(:,:,3);
figure;
surf(x,t,T);
title('Temperature');
xlabel('Distance x');
ylabel('Time t');
figure;
surf(x,t,E);
title('Electric Field');
xlabel('Distance x');
ylabel('Time t');
figure;
surf(x,t,c);
title('Li ion concentration');
xlabel('Distance x');
ylabel('Time t');
function [CC, FF, SS]=battery_GE(x,t,u,DuDx)
CC=[rho*c_th; 0; 1];
FF=[a_1*k; epsilon; c_1*D].*DuDx;
SS=[a_2*sigma*DuDx(2); -b*u(3);c_2*(DuDx(3)-(b*u(3)/epsilon)-DuDx(1))];
function u0=battery_ic(x)
u0=[T_0;phi_0;c_0];
function [pl,ql,pr,qr] = pdex1bc(xl,ul,xr,ur,t)
T_L=1;
T_R=2;
pl=[T_L; I_app; 0];
ql=[0; sigma/epsilon 1];
pr=[T_R; -I_app; 0];
qr=[0; sigma/epsilon 1];
I get the following error, which I don't understand because I made T_0 a global variable:
Undefined function or variable 'T_0'.
Error in battery_model>battery_ic (line 54)
u0=[T_0;phi_0;c_0];
Error in pdepe (line 229)
temp = feval(ic,xmesh(1),varargin{:});
Error in battery_model (line 24)
sol = pdepe(m,@battery_GE,@battery_ic,@battery_bc,x,t);
I'm not sure what is going on now.

Bill Greene on 25 May 2018
What is the definition of smooth_step?
Matthew Hunt on 25 May 2018
function y=smooth_step(x,a,b)
%This is a smooth approximation of the Step function done via the erf
%function
z=zeros(1,length(x));
if (a==x(1))
m=find(x>b,1,'first');
z(1:m-1)=1;
y=min(z,0.5*(1+erf(-30*(x-b))));
elseif(b==x(end))
n=find(x<=a,1,'last');
z(n+1:end)=1;
y=min(z,0.5*(1+erf(30*(x-a))));
else
y=0.5*min(1+erf(30*(x-a)),1+erf(-30*(x-b)));
end

Precise Simulation on 9 Jun 2018
Systems of PDEs might also be easier to solve with the FEATool FEM Toolbox which features a GUI and easy syntax for defining custom PDEs and equations. This example of heat transport and diffusion might be a good start, otherwise you could perhaps use the built-in convection and diffusion physics mode and modify it according to your equations.

Matthew Hunt on 15 Jun 2018
I have some values which allow the code to run BUT I have oscillations in the diffusion which I should not get. I am assuming that this is a numerical instability, is it caused by too large a timestep?