Parabolic Pdes - Explicit Method: Heat Flow and Diffusion

Lecture 35 Parabolic PDEs - Explicit Method
Heat Flow and Diusion

The conduction of heat and diusion of a chemical happen to be modeled by the same dierential equation. The reason for this is that they both involve similar processes. Heat conduction occurs when hot, fast moving molecules bump into slower molecules and transfer some of their energy. In a solid this involves moles of molecules all moving in dierent, nearly random ways, but the net eect is that the energy eventually spreads itself out over a larger region. The diusion of a chemical in a gas or liquid simliarly involves large numbers of molecules moving in dierent, nearly random ways. These molecules eventually spread out over a larger region. In three dimensions, the equation that governs both of these processes is the heat/diusion equation ut = cu , where c is the coecient of conduction or diusion, and u(x, y, z) = uxx + uyy + uzz . The symbol in this context is called the Laplacian. If there is also a heat/chemical source, then it is incorporated a function g(x, y, z, t) in the equation as ut = cu + g. In some problems the z dimension is irrelevent, either because the object in question is very thin, or u does not change in the z direction. In this case the equation is ut = cu = c(uxx + uyy ). Finally, in some cases only the x direction matters. In this case the equation is just ut = cuxx , or ut = cuxx + g(x, t), if there is a heat source represented by a function g(x, t). In this lecture we will learn a straight-forward technique for solving (35.1) and (35.2). It is very similar to the nite dierence method we used for nonlinear boundary value problem in the previous lecture. It is worth mentioning a related equation ut = c(u ) for > 1, (35.2) (35.1)
124
125
which is called the porus-media equation. This equation models diusion in a solid, but porus, material, such as sandstone or an earthen structure. We will not solve this equation numerically, but the methods introduced here would work. Many equations that involve 1 time derivative and 2 spatial derivatives are parabolic and the methods introduced in this lecture will work for most of them.
Explicit Method Finite Dierences

The one dimensional heat/diusion equation ut = cuxx , has two independent variables, t and x, and so we have to discretize both. Since we are considering 0 x L, we subdivide [0, L] into m equal subintervals, i.e. let h = L/m and (x0 , x1 , x2 , . . . , xm1 , xm ) = (0, h, 2h, . . . , L h, L). Similarly, if we are interested in solving the equation on an interval of time [0, T ], let k = T /n and (t0 , t1 , t2 , . . . , tn1 , tn ) = (0, k, 2k, . . . , T k, T ). We will then denote the approximate solution at the grid points by uij u(xi , tj ). The equation ut = cuxx can then be replaced by the dierence equations c ui,j+1 ui,j = 2 (ui1,j 2ui,j + ui+1,j ). (35.3) k h Here we have used the forward dierence for ut and the central dierence for uxx . This equation can be solved for ui,j+1 to produce ui,j+1 = rui1,j + (1 2r)ui,j + rui+1,j for 1 i m 1, 0 j n 1, where r= (35.4)
ck . (35.5) h2 The formula (35.4) allows us to calculate all the values of u at step j + 1 using the values at step j. Notice that ui,j+1 depends on ui,j , ui1,j and ui+1,j . That is u at grid point i depends on its previous value and the values of its two nearest neighbors at the previous step (see Figure 35.1).
Initial Condition
To solve the partial dierential equation (35.1) or (35.2) we need an initial condition. This represents the state of the system when we begin, i.e. the initial temperature distribution or initial concentration prole. This is represented by u(x, 0) = f (x). To implement this in a program we let ui,0 = f (xi ).
126
LECTURE 35. PARABOLIC PDES - EXPLICIT METHOD
tj+1
ui,j+1
T
tj
xi1
xi
xi+1
Figure 35.1: The value at grid point i depends on its previous values and the previous values of its nearest neighbors.
Boundary Conditions
To solve the partial dierential equation (35.1) or (35.2) we also need boundary conditions. Just as in the previous section we will have to specify something about the ends of the domain, i.e. at x = 0 and x = L. One possibility is xed boundary conditions, which we can implement just as we did for the ODE boundary value problem. A second possibility is called variable boundary conditions. This is represented by timedependent functions, u(0, t) = g1 (t) and u(L, t) = g2 (t). In a heat problem, g1 and g2 would represent heating or cooling applied to the ends. These are easily implemented in a program by letting u0,j = g1 (tj ) and um,j = g2 (tj ). A third possibility is insulating boundary conditions, which we also did for the ODE boundary value problem. To implement this condition, we need to copy values from inside the region to ctional points just outside the region. See Figure 35.2 for an illustration.
127
z z z z j
tj+1
T s T s T s d d d d d d d d d z dz dz dj T s T s T s d d d d d d d d d z dz dz dj T s T s T s d d d d d d d d d z dz dz dj
tj
xm1
xm
xm+1
Figure 35.2: Illustration of information ow for the explicit method with an insulated boundary condition. The open circles are the ctional points and the curved arrows represent copying the value.
Implementation
The following program (also available on the web page) implements the explicit method. It incorporates variable boundary conditions at both ends. To run it you must dene functions f , g1 and g2 . Notice that the main loop has only one line. The values of u are kept as a matrix. It is often convenient to dene a matrix of the right dimension containing all zeros, and then ll in the calculated values as the program runs. Run this program using L = 2, T = 20, f (x) = .5x, g1 (t) = 0, and g2 (t) = cos(t). Note that g1 (t) must be input as g1 = inline(0*t).
128
LECTURE 35. PARABOLIC PDES - EXPLICIT METHOD
function [t x u] = myheat(f,g1,g2,L,T,m,n,c) % function [t x u] = myheat(f,g1,g2,L,T,m,n,c) % solve u_t = c u_xx for 0<=x<=L, 0<=t<=T % BC: u(0, t) = g1(t); u(L,t) = g2(t) % IC: u(x, 0) = f(x) % Inputs: % f -- inline function for IC % g1,g2 -- inline functions for BC % L -- length of rod % T -- length of time interval % m -- number of subintervals for x % n -- number of subintervals for t % c -- rate constant in equation % Outputs: % t -- vector of time points % x -- vector of x points % u -- matrix of the solution, u(i,j)~=u(x(i),t(j)) h r x t = = = = L/m; k = T/n; c*k/h^2; rr = 1 - 2*r; linspace(0,L,m+1); linspace(0,T,n+1);
%Set up the matrix for u: u = zeros(m+1,n+1); % assign initial conditions u(:,1) = f(x); % assign boundary conditions u(1,:) = g1(t); u(m+1,:) = g2(t); % find solution at remaining time steps for j = 1:n u(2:m,j+1) = r*u(1:m-1,j) + rr*u(2:m,j) + r*u(3:m+1,j); end % plot the results mesh(x,t,u)
Exercises
35.1 Modify the program myheat.m to have an insulated boundary at x = L (rather than u(L, t) = g2 (t)). Run the program with L = 2, T = 20, c = .5, g1 (t) = sin(t) and f (x) = sin(x/4). Set m = 20 and experiment with n. Get a plot when the program is stable and one when it isnt. Turn in your program and plots.

Parabolic Pdes - Explicit Method: Heat Flow and Diffusion

Uploaded by

Parabolic Pdes - Explicit Method: Heat Flow and Diffusion

Uploaded by

Lecture 35 Parabolic PDEs - Explicit Method

Heat Flow and Diusion

Explicit Method Finite Dierences

LECTURE 35. PARABOLIC PDES - EXPLICIT METHOD

LECTURE 35. PARABOLIC PDES - EXPLICIT METHOD

You might also like