# Trouble with shooting method for a 4th-order stiff ODE

The ODE I need to solve is

$$\left(y^3y^{\prime\prime\prime}\right)^\prime+\frac{5}{8}xy^\prime-\frac{1}{2}y+\frac{c}{y}=0$$ where $\prime$ denotes differentiation, $c$ is a constant and $0<c\le1$. It needs 4 boundary conditions (BCs): $y^\prime(0)=y^{\prime\prime\prime}(0)=0$ are given, and the other two values of $y(0)$ and $y^{\prime\prime}(0)$ are determined by shooting for BCs at infinity (or $x_\text{Max}$).

Inspired by @ bbgodfrey's idea in this answer, I know that the far-field asymptotic behavior is given by DSolveing the remainder of the ODE after the nonlinear derivative term are neglected:

DSolve[5/8 x y'[x] - 1/2 y[x] + c/y[x] == 0, y[x], x]


Thus the far-field asymptotic behavior is $y= kx^{4/5}$, where $k$ is a constant. Accordingly, we refer to a solution with $y\sim x^{4/5}$ asymptotic behavior as a $x^{4/5}$ solution. Here, I want to search for this kind of solutions by shooting method in which the ODE is integrated from $x_0=0$ to some large value of $x_\text{Max}$, say $x_\text{Max}=10$.

The shooting parameters $y(0)$ and $y^{\prime\prime}(0)$ will be adjusted so that $y-\frac{5}{4}xy^\prime=0$ or $y+\frac{25}{4}x^2y^{\prime\prime}=0$ at $x=x_\text{Max}$ is satisfied. The two BCs were chosen to be independent to involve low-order derivatives and to require $y\propto x^{4/5}$ at the end of the integration interval.

Thanks to @ xzczd's answer to this question, an example of the code is:

c = 1;
(*c=1/10;*)
ode = D[y[x]^3*y'''[x], x] + 5/8 x y'[x] - 1/2 y[x] + c/y[x] == 0;
bc1 = y[x] - 5/4 x y'[x] == 0;
bc2 = y[x] + 25/4 x^2 y''[x] == 0;


Thank again to this answer, I can obtain the series solution at $x_0=0$ with seriesDSolve

solSeries = seriesDSolve[ode, y, {x, 0, 5},
{y -> a, y' -> 0, y'' -> b, y''' -> 0}]


We thus obtain a set of boundary conditions at $x=x_0$ by

newbclist =  Thread[(Derivative[#][y][x0] ==
(D[Normal@solSeries, {x, #}] /. x -> x0) &) /@ Range[0, 3]]


Then, give the interval, solve the ODE and plot the parameter plane

x0 = 0; xMax = 6;
sol = ParametricNDSolveValue[{ode, newbclist}, y, {x, x0, xMax}, {a, b}]
ContourPlot[{bc1, bc2} /. x -> xMax /. y -> sol[a, b] // Evaluate,
{a, 0.2, 0.6}, {b, -0.05, 1}]


Unfortunately, Mathematica gives something like this:

ParametricNDSolveValue::ndsz: At x$803007 == 0.27144285125721357, step size is effectively zero; singularity or stiff system suspected. as encounters in this problem. After some experiments, I found the situation appears to be worse for a larger parameter$c$, i.e.$c\rightarrow 1$. And I also played with the intervals of$a$and$b$but it seems no help. Three ugly results for your reference (sorry!...) 1. When I use c = 1/10, xMax = 0.5 with {a, -0.2, 1}, {b, -0.1, 1}, ContourPlot gives 2. When I use c = 1/10, xMax = 0.5 with {a, 0.2, 0.6}, {b, -0.05, 1}, ContourPlot gives 3. When I change to bc1 = 1/2 y[x] - 5/8 x y'[x] - c/y[x] == 0; with c = 1, xMax = 6 and {a, -0.2, 1}, {b, -0.1, 1}, ContourPlot gives To sum up, my questions are: (1) After I find the parameter$y(0)$and$y^{\prime\prime}(0)$, how can I figure out the associated constant$k$in the far-field asymptotic behavior$y= kx^{4/5}$? (2) Is it possible to estimate the parameters$y(0)$and$y^{\prime\prime}(0)$automatically to start the integration, as shown in this answer? • I think it's just the nature of your equation: Some choices of {a, b} can't lead to a solution extending to infinity. Just observe this: mid = sol[0.6, 0.6]; {{lb, rb}} = mid["Domain"]; Plot[mid[t], {t, lb, rb}, PlotRange -> All] – xzczd Jan 16 '16 at 4:29 • Thanks, @ xzczd, That really is the question. I am wondering is there any technique to find out the ranges of$a$and$b$so that the solutions can shoot up to an appropriate$x_\text{Max}$. In other words, I just try to explore the existence of some isolated parameter pairs of$(a,b)$for certain given$c$and$x_\text{Max}$. That is, the values of$y(0)$and$y''(0)$are determined by shooting for boundary condition at$x_\text{Max}$. – W. Robin Jan 16 '16 at 6:56 ## 1 Answer Automatically looking for proper$y(0)$and$y''(0)$isn't easy, at least I can't think out a solution, still, I don't think you need such fascinating technique to solve the new equation, you just need to slightly modify the code for plotting to filter out those improper$(a,b)$: Quiet@ContourPlot[ If[y["Domain"][[1, -1]] < xMax, 1, #] == 0 & /@ Subtract @@@ {bc1, bc2} /. x -> xMax /. y -> sol[a, b] // Evaluate, {a, 1, 1.6}, {b, -0.1, 0.2}, PlotPoints -> 50] The plot range is obtained through not too much trial and error. "How can I figure out the associated constant$k$in the far-field asymptotic behavior$y=kx^{4/5}$?" Just use the obtained$a$as b.c. of the far-field asymptotic equation. "I tried the following direct shooting method… It gives some errors… " I guess it's because bc1 and bc2 are not b.c in usual meaning, after all, they're equal to y[xMax] == (y[xMax] /. DSolve[#, y, x]) & /@ {bc1, bc2} (* {y == {E^(4/5) C}, y == {C Cos[2/5] + C Sin[2/5]}} *)  • @ xzczd, I just noticed your reply. I will learn something from you and reply you later. Many thanks! – W. Robin Jan 16 '16 at 16:49 • @ xzczd, 3 questions, I'm a beginner. 1. Do you remember what value of the$c$and$xMax\$ were used in your answer? 2. I found that you have Map a pure function If[y["Domain"][[1, -1]] < xMax, 1, #] == 0 & to each element in the 1st level of Subtract @@@ {bc1, bc2}. However, applying Subtract to {bc1,bc2} by @@@ is just equivalent to {bc1,bc2}, if I didn't make a mistake on the computation sequence? That is, you apply True or False to {bc1,bc2}? 3. What is the meaning of # in If expression, I know it is commonly used the formal para. for a pure function. Thanks again! – W. Robin Jan 17 '16 at 3:48
• @W.Robin 1. c = 1; xMax = 6; 2. Observe Subtract @@@ {xx == yy, zz == ww} 3. Yeah, # owns no other meaning, it's the argument of pure function. If you still have trouble in understanding its usage here, just execute If[y["Domain"][[1, -1]] < xMax, 1, #] == 0 & /@ Subtract @@@ {bc1, bc2} outside of ContourPlot and observe the result. – xzczd Jan 17 '16 at 3:58
• @ xzczd, now I understand a little bit: using Subtract @@@ {bc1, bc2} just extract the LHS of each BC. because the 1st level of bc1 or bc2 is just Equal in which the 2nd argument is 0! Am I right? – W. Robin Jan 17 '16 at 4:31
• @W.Robin The warning isn't a big deal, it's simply because when y["Domain"][[1, -1]] execute for the first time, y["Domain"] isn't yet a list, in v10 a new function Indexed is introduced to handle this problem more properly, but since I (and you) are still in v9, so I simply used Quiet`. This doesn't influence the result at all. – xzczd Jan 17 '16 at 5:56