# PDE of real-world system, integral boundary condition

I've stripped all the physical-significance for clarity, but I know that u[x,t] will be everywhere positive and continuous.

here are the equations in Mathematica code:

eqs = {D[u[x, t], {x, 2}] == D[u[x, t], t],
u[x, 0] == c1,
(D[u[x, t], x] /. x -> 1) == 0,
u[0, t] == 3 c2 - c2 Integrate[u[x, t], {x, 0, 1}]} /. {c1 -> 1, c2 -> 1}


The dependent variable u represents pressure, x represents distance and t time.

The last item in eqs represents a material-balance on the gas in the system - the integral is the amount of gas distributed in the region of interest - it accounts for gas in the region of interest, stating that

 gas in region-of-interest +
gas in the isobaric region at the surface (not of interest)

=
amount of gas in the entire system at time t=0 (represented by u[x,0]==c1 and the addition of 3 C2 - I stripped numerous physical symbols out so while the equations appear flaky, it's the shape that is of interest, not the internal consistency of the simplified equations).


When I try to solve this in Mathematica, regardless of whether I use Integrate or NIntegrate I get errors:

NDSolve[eqs, {x, 0, 1}, {t, 0, 1}]


When using Integrate.... I get:

Equation or list of equations expected instead of .... (integral here)


When using NIntegrate I get:

The integrand u[x,t] has evaluated to non-numerical values


(this error message is no surprise - just including it for completeness).

I modeled this using years ago for some graduate research, and got good solutions (I built the physical system in a lab and measured transient pressure profiles etc. to validate model results, so it's a real-world problem. Now I want to see if I can solve it again with Mathematica).

The actual BC is this: P[x,t] is the dependent variable; x & t are independent variables, everything else is a known constant.

No postings I can find here or on Wolfram.com indicate how or if MMA can do this.

I would like to know if anyone has solved a similar problem (boundary condition containing an integral of the solution itself) using MMA, and how it might be done.

• NDSolve cannot handle integro-differential equations. Learn more at Advanced Numerical Differential Equation Solving in the Wolfram Language. – bbgodfrey Mar 29 '15 at 5:36
• Also, you have two typos. u[0, t] = 3 c2 - c2 ... should be u[0, t] == 3 c2 - c2 ... and NDSolve[eqs, {x, 0, 1}, {t, 0, 1}] should be NDSolve[eqs, u, {x, 0, 1}, {t, 0, 1}], although fixing them leads to other error messages. – bbgodfrey Mar 29 '15 at 5:45
• Finally, setting c2 -> 0 instead of 1 allows NDSolve to produce an answer, although this is not the problem you wish to solve. In my experience, you need to discretize your integral boundary condition to make headway. – bbgodfrey Mar 29 '15 at 5:53
• thanks for comments. The typos are transcription errors; the equation in MMA is fully formed and correct – Paul_A Mar 29 '15 at 17:48

Because NDSolve cannot accommodate the x=0 boundary condition, it is necessary to perform this computation by discretizing the PDE in x. The resulting do-it-yourself procedure is discussed in Introduction to Method of Lines.

For illustrative purposes, assume that x is divided into five equal segments.

n = 5; h = 1/n;

with a variable defined at each node, at {0, .2, .4, .6, .8, 1}

U[t_] = Table[u[i][t], {i, 0, n}]
(* {u[t], u[t], u[t], u[t], u[t], u[t]} *)


The PDE then decomposes into n ODEs plus one algebraic equation. (The first and last equations are modified to take account of the boundary conditions at x = 0 and x = 1.)

Thread[D[U[t], t] == Join[{0}, ListCorrelate[{1, -2, 1}/h^2, U[t]],
2 {u[n - 1][t] - u[n][t]}/h^2]];
eqns = ReplacePart[%, 1 -> u[t] == (3 - (Total[U[t]] - u[t]/2 - u[n][t]/2) h) c2]
(* {u[t] == c2 (3 +
1/5 (-(1/2) u[t] - u[t] - u[t] - u[t] - u[t] - 1/2 u[t])),
Derivative[u][t] == 25 u[t] - 50 u[t] + 25 u[t],
Derivative[u][t] == 25 u[t] - 50 u[t] + 25 u[t],
Derivative[u][t] == 25 u[t] - 50 u[t] + 25 u[t],
Derivative[u][t] == 25 u[t] - 50 u[t] + 25 u[t],
Derivative[u][t] == 50 (u[t] - u[t])} *)


Initial conditions at t = 0 are given by

initc = ReplacePart[Thread[U == Table[c1, {n + 1}]], 1 -> eqns[] /. t -> 0]
(* {u == c2 (3 +
1/5 (-(1/2) u - u - u - u - u - 1/2 u)),
u == c1, u == c1, u == c1, u == c1, u == c1} *)


Note that u is given by u[t]/.t -> 0 for consistency. NDSolve objects, if u is set to anything else.

Finally, the n + 1 equations are solved and plotted, below with n = 250 for accuracy.

lines = NDSolve[{eqns, initc} /. {c1 -> 1, c2 -> 1}, U[t], {t, 0, 1}];
ParametricPlot3D[Evaluate[Table[{i h, t, First[u[i][t] /. lines]}, {i, 0, n}]],
{t, 0, 1}, PlotRange -> All, AxesLabel -> {"x", "t", "u"}] The steady state converges on 1.5 as n becomes large.