Can anyone help me solve a Non-Homogenous PDE for my assignment?

Pauly Man · Apr 27, 2003

I have an assignment due at the end of the week, and I was wondering if someone could check my working for me, as I am prone to making errors. Also, in Step Five I am unsure how to solve for T(t), can anyone point me in the right direction?

∂u/∂t = (c/r)*(r(∂u/∂r)) + Q(r,t)
0 < r < a ; c > 0

u(0,t) is finite
u(a,t) = 0
u(r,0) = 0

Q(r,t) = qJ₀((μ₁r)/a)

Where μ₁ is the first crossing of J₀(x), which is a Bessel Function of the first kind of order zero, q is a constant.

Solve for u(r,t).

Step One-

Use separation of variables on the homogenous equation, (ie. Set Q(r,t) = 0):

u(r,t) = R(r)T(t)

⇒ ∂u/∂t = RT'
⇒ ∂u/∂r = R'T
⇒ ∂²u/∂r² = R"T

⇒ (1/c)*(T'/T) = (R"/R) + (1/r)*(R'/R) = -λ

Step Two-

Solve for R(r) first:

r²R" + rR' + λr²R = 0
let λ = k²
and ρ = kr

⇒ dR/r = (dR/dρ)*(dρ/dr) = k(dR/dρ)
⇒ d²R/dr² = k²(d²R/dρ²)
⇒ ρ²R"(ρ) + ρR'(ρ) + ρ²R(ρ) = 0

which is Bessels equation, with ν = 0.

⇒ R(ρ) = AJ₀(ρ) + BY₀(ρ)
⇒ R(r) = AJ₀(kr) + BY₀(kr)
where Y(x) is a Bessel Function of the second kind.

Now Y(x) approaches infinity as x approaches zero, therefore if R(r) is to satisfy the first boundaryb condition;

u(0,t) finite;

then B must equal zero.

⇒ R(r) = AJ₀(kr)

The second boundary condition is;

u(a,t) = R(a)T(t) = 0;

⇒ R(a) = 0 and/or T(t) = 0
Now T(t) = 0 is a trivial solution, as this implies u(r,t) = 0. So we look at R(a) = 0;

⇒ R(a) = AJ₀(ka) = 0
A = 0 is another trivial solution, so we look at;

J₀(ka) = 0

We define μ_n to be the nth zero of J₀, then

k_n = μ_n/a

Setting A = 1 we get;
⇒ R(r) = J₀((μ_nr)/a)

Step Three-

Expand Q(r,t) as a Fourier series of R_n(r):

Let Q(r,t) = ∑ b_n(t)R_n(r)

Use the orthogonality condition:

∫ rJ₀(k_nr)J₀(k_mr)dr = 0 ; n ≠ m

∫ rJ₀(k_nr)J₀(k_mr)dr = (a²/2)*J₁²(k_ma) ; n = m

⇒ b_m(t) = 0 ; m ≠ 1

⇒ b_m(t) = (2q/a²J₁²(μ_m))*∫ re^rtJ₀²((μ₁r)/a)dr ; m = 1

(So far so good, I think. I have a solution for R(r) with no unknowns, and have expanded Q(r,t) out in terms of the eigenfunctions R(r), and have "solved" for the coefficient b_m).

Step Four-

Substitute the solution;

u(r,t) = ∑R_n(r)T_n(t);

into the original PDE;

∑R_nT'_n = c∑(R"_n + (1/r)R'_n)T_n + ∑b_nR_n

Now R' + (1/r)R' = -λR

⇒ ∑(T'_n + c&lambda_nT_n)R_n = ∑b_nR_n

Use the orthogonality condition for R_n to get;

∑T'_n + c&lambda_nT_n = ∑b_n

⇒ ∑T'_n + c&lambda_nT_n = 0 ; n ≠ 1

⇒ ∑T'_n + c&lambda_nT_n = (2q/a²J₁²(μ_m))*∫ re^rtJ₀²((μ₁r)/a)dr ; n = 1

Step Five-

Now I have to solve for T(t) using the equations above. For n > 1 the solution is easy to find;

T'_n + cλ_nT_n = 0

⇒ T_n(t) = C_ncos((√c)(μ_n/a)) + D_nsin((√c)(μ_n/a))

For n = 1 the I am unsure how to find the solution. The DE to solve is given below, and I've never had to solve a DE like it before:

T₁' + c&lambda_nT_n = (2q/a²J₁²(μ_m))*∫ re^rtJ₀²((μ₁r)/a)dr

I can find the homogenous solution, however I can't find the particular solution. HELP!

The final solution-

The final solution is:

u(r,t) = J₀((μ₀r)/a)[C₀cos((√c)(μ₀/a)) + D₀sin((√c)(μ₀/a))] + (2 to infinity)∑J₀((μ_nr)/a)[C_ncos((√c)(μ_n/a)) + D_nsin((√c)(μ_n/a))] + J₀((μ₁r)/a)[(Particular Solution of T₁(t))

quantumdude · Apr 27, 2003

Sit tight--I'm printing this out and taking it home.

Till tomorrow...

Pauly Man · Apr 28, 2003

I realized last nite that I for some reason solved the equation for T(t) as a second order DE, when it is clearly a first order DE. So ignore the final solution and part of step five.

For n ≥ 1-

T'_n + cλ_nT_n = 0

T(t) = e^{-(cλ_nt)}

quantumdude · Apr 28, 2003

OK.

I had some trouble getting through your initial post because of what I am convinced are typos.

For instance, think that this:

Originally posted by Pauly Man
∂u/∂t = (c/r[/color])*(r(∂u/∂r)) + Q(r,t)

Should be this:

∂u/∂t=(c∂/∂r[/color])*(r(∂u/∂r))+Q(r,t)

At least, that would be consistent with what you write later.

Also this:

⇒ dR/r[/color] = (dR/dρ)*(dρ/dr) = k(dR/dρ)

Should presumably be this:

⇒ dR/dr[/color] = (dR/dρ)*(dρ/dr) = k(dR/dρ)

Also, I don't know what

⇒

is supposed to be. It shows up as a rectangle.

If I'm right about those typos, then you are OK through Steps 1 and 2. I am still checking Step 3.

Pauly Man · Apr 29, 2003

Originally posted by Tom

∂u/∂t=(c∂/∂r[/color])*(r(∂u/∂r))+Q(r,t)

At least, that would be consistent with what you write later.

Yep, there was a typo there. Although it should still be c/r, and not c:

∂u/∂t=((c/r)∂/∂r[/color])*(r(∂u/∂r))+Q(r,t)

Also this:

Originally posted by Tom

Should presumably be this:

⇒ dR/dr[/color] = (dR/dρ)*(dρ/dr) = k(dR/dρ)

Yep, that is correct.

Originally posted by Tom

Also, I don't know what

⇒

is supposed to be. It shows up as a rectangle.

That is supposed to be a "implies" symbol, it doesn't seem to work on all broswers and OS's. Sorry.

Originally posted by Tom

If I'm right about those typos, then you are OK through Steps 1 and 2. I am still checking Step 3.

Cool.

Pauly Man · Apr 30, 2003

I think I've solved it!

I found out today that the reason I was having so much difficulty with the problem was that the assignment had a "typo". Q(r,t) is not given by:

Q(r,t) = qJ₀((μ₁r)/a)e^rt

But by:

Q(r,t) = qJ₀((μ₁r)/a)e^{ν[/color]t}

This makes the problem a heck of a lot easier to solve.

Step one and step two remain unchanged, step three changes only with the final answer:

b_m(t) = 0 ; m ≠ 1
b_m(t) = (qe^νtJ₀²(μ₁))/(J₀²(μ_m)) ; m =1

Step four reamins unchanged until the end result:

where T'_n + ck_n²T_n = b_n(t) , where b_n is given above.

Step five becomes:

Step Five-

For n ≥ 1-

T_n(t) = D_ne^-ck_n²t

For n = 1-

Let α = ck_n²
Let β = q(J₀²(μ₁)/J₀²(μ_n))

then;

T'₁ + αT₁ = βe^νt

The integrating factor is:

I.F- e^αt

So;

e^αtT₁(t) - T₁(0) = (β/(α + ν))(e^{(α + ν)t} - 1)

Now, T_n(0) = 0 from the initial condition. SO the above simplifies. Now we add the T(t)'s together:

T_n(t) = (β/(α + ν))(e^νt - e^-αt) + (2 to infinity)∑D_ne^-ck_n²t

Now let's look at the initial condition to determine D_n-

T_n(0) = (β/(α + ν))(1 - 1) + (2 to infinity)∑D_n = 0

T_n(0) = (2 to infinity)∑D_n = 0

So D_n = 0

Therefore T_n(t) is given by:

T_n(t) = T₁(t) = (q(J₀²(μ₁)/J₀²(μ₁))
/(ck_n² + ν))(e^νt - e^-ck_n²t)

Which upon simplification becomes:

T_n(t) = T₁(t) = (q/(ck_n² + ν))(e^νt - e^-ck_n²t)

Therefore the final solution is given by:

u(r,t) = ∑R_n(r)T_n(t) = R₁(r)T₁(t)

u(r,t) = J₀((μ₁r)/a)(q/(c(μ₁/a)² + ν))(e^νt - e^{-c(μ₁/a)²t})

This solution satifsies the boundary and initial conditions, and so I hopw that it is right.

Another God · Apr 30, 2003

sweet jesus!

quantumdude · Apr 30, 2003

I'll check it and get back to you tomorrow.

Pauly Man · Apr 30, 2003

Thanx for all the help Tom, I really appreciate it.

Most of the class has the same result, so I'm crossing fingers.

Pauly Man · Apr 30, 2003

Originally posted by Another God
sweet jesus!

LOL. Looks daunting I agree.

quantumdude · May 1, 2003

Looks good to me.

Pauly Man · May 4, 2003

Thanx for all the help Tom.

Can anyone help me solve a Non-Homogenous PDE for my assignment?

Similar threads

Chain falling out of a horizontal tube onto a table

Rolling without slipping on a curved surface

Is it possible for a vertical rod balancing on a table to lose contact by striking the top of the rod?

Coulomb's force vs the Lorentz force

Earthed plates confusion

Insights Thinking Outside The Box Versus Knowing What’s In The Box

Insights Why Entangled Photon-Polarization Qubits Violate Bell’s Inequality

Insights Quantum Entanglement is a Kinematic Fact, not a Dynamical Effect

Insights What Exactly is Dirac’s Delta Function? - Insight

Insights Relativator (Circular Slide-Rule): Simulated with Desmos - Insight

Insights Fixing Things Which Can Go Wrong With Complex Numbers