3
$\begingroup$

I have the following inhomogeneous differential equation (sorry, I cannot come up with a simpler example)

Hom = -(-1 + 
     z) z^4 (8 (-5 + 6 z) f0[z] + (20 - 70 z + 52 z^2) f0'[
      z] + (-1 + z) z (2 (-5 + 7 z) f0''[z] + (-1 + z) z f0'''[z]))
InHom = -((24 z (24 + z (-48 + z (23 + 2 z))))/(-1 + z)) + (
  48 Log[1 - z] (z (18 + z (-24 + 5 z)) + 3 (-2 + z) (-1 + z) Log[1 - z]))/z
myeq = Hom + InHom

DSolve gives me a solution but it doesn't work, meaning that

DSolve[myeq == 0, f0[z], z][[1, 1, 2]];
myeq /. f0 -> ((% /. z -> #) &) // FullSimplify
% === 0

gives False. Any idea why this is happening?

$\endgroup$
10
  • 2
    $\begingroup$ Could be a bug... but the issue unfortunately seems to depend sensitively on the form of the rather complicated equation you are solving. $\endgroup$ – Will.Mo Nov 26 '19 at 18:15
  • $\begingroup$ @Bill, the range of $z$ i'm interested is $0<z<1$, I tried to use Assuming but it doesn't change. The solution I'm interested in should be $0$ at $z=0$ and for small $z$ behave linearly in $z$. $\endgroup$ – bnado Nov 26 '19 at 19:57
  • $\begingroup$ Yeah, it says that the solution diverges as $z^{-5}$ but I don't trust that. For example, I can easily solve the differential equation as a power series. For example Nmax = 15; powser = ((1/z^3 myeq) /. f0 -> ((Sum[\[Alpha][n] #^n, {n, 1, Nmax}]) &)); li = CoefficientList[Series[powser, {z, 0, Nmax}], z] // Normal; lis = Solve[li == 0 li]; fsol = (Sum[\[Alpha][n] z^n, {n, 1, Nmax}]) /. lis[[1]]; Series[myeq /. f0 -> (fsol /. z -> # &), {z, 0, Nmax}] gives the first Nmax terms of the solution I'm interested in. It does not diverge at $z=0$ $\endgroup$ – bnado Nov 26 '19 at 21:00
  • $\begingroup$ Your checking is wrong. Guess what aaaa === 0 will evaluate to? $\endgroup$ – xzczd Nov 27 '19 at 4:06
  • $\begingroup$ In fact, a simpler example is Hom + InHom[[2]]. $\endgroup$ – bbgodfrey Nov 27 '19 at 5:27
3
$\begingroup$

DSolve gives incorrect answer

With Hom, InHom, and myeq as defined in the question, DSolve indeed gives an incorrect answer,

sol = DSolveValue[myeq == 0, f0, z];
FullSimplify[(myeq == 0) /. f0 -> %]
(* ((-1 + z) (z^2 - 5 (-2 + z) z Log[1 - z] + (14 + z (-14 + 3 z)) Log[1 - z]^2))/z == 0 *)

which is not, in general, True, as can be seen by evaluating the expression for a few values of z. Note that none of the constants of integration appear in this expression, indicating that the error is not associated with the homogeneous equation. This also can be demonstrated by solving the homogeneous equation directly.

DSolveValue[Hom == 0, f0, z]
FullSimplify[(Hom == 0) /. f0 -> %]
(* Function[{z}, C[1]/z^4 - (C[2] Log[1 - z])/z^4 + 
   (C[3] (1/(1 - z) - z - 1/4 (-1 - 2 Log[-1 + z])^2))/z^4] *)
(* True *)

Obtaining correct answer

Somewhat surprisingly (to me, at least), the correct answer can be obtained as follows. Restructure InHom as

InHomsim = Collect[InHom, _Log, Simplify]
(* -((24 z (24 + z (-48 + z (23 + 2 z))))/(-1 + z)) + 
   48 (18 + z (-24 + 5 z)) Log[1 - z] + (144 (-2 + z) (-1 + z) Log[1 - z]^2)/z *)

Next, solve

sol1 = DSolveValue[Hom + InHomsim[[1]] == 0, f0, z];
FullSimplify[(Hom + InHomsim[[1]] == 0) /. f0 -> %]
(* True *)
sol2 = DSolveValue[Hom + InHomsim[[2]] == 0, f0, z];
FullSimplify[(Hom + InHomsim[[2]] == 0) /. f0 -> %]
(* True *)
sol3 = DSolveValue[Hom + InHomsim[[3]] == 0, f0, z];
FullSimplify[(Hom + InHomsim[[3]] == 0) /. f0 -> %]
(* True *)

Thus, correct solutions for each of the three additive components of InHomsim can be obtained without difficulty. Simply add them to obtain the correct answer to the original equation.

Collect[(sol1[[2]] + sol2[[2]] + sol3[[2]]) /. C[i_] -> C[i]/3, C[_], FullSimplify];
soltr = Function[{z}, Evaluate@%]
FullSimplify[(myeq == 0) /. f0 -> soltr]
(* Function[{z}, C[1]/z^4 - (C[2] Log[1 - z])/z^4 + 
   (C[3] (1/(1 - z) - z - 1/4 (1 + 2 Log[-1 + z])^2))/z^4 + 
   (1/((-1 + z) z^4)) 4 (-6 (7 + 6 \[Pi]^2 (-1 + z) - z (7 + 4 z)) - 
   2 (-1 + z) Log[1 - z]^3 - 6 Log[1 - z] (-16 + z (15 + 2 z) + 
   (-1 + z) (-23 + Log[-1 + z]) Log[-1 + z]) + (-1 + z) Log[-1 + z] 
   (12 + Log[-1 + z] (-69 + 4 Log[-1 + z])) - 6 (-1 + z) Log[1 - z]^2 (5 + Log[z]) + 
   12 (-1 + z) Log[1 - z] PolyLog[2, z] - 12 (-1 + z) PolyLog[3, 1 - z])] *)
(* True *)

Evidently, DSolve can handle each of the three components of InHomsim but not the whole expression at once.

Why the error?

DSolve also can solve the general equation, Hom + g[z] == 0, correctly.

solgen = DSolveValue[Hom + g[z] == 0, f0, z];
FullSimplify[(Hom + g[z] == 0) /. f0 -> %]
(* True *)

Not surprisingly solgen contains three integrals over g[z] , two of them fairly complicated. Because the g[z] in the question (namely InHom) and (as it happens) the kernels of the integrals both contain expressions with branch cuts, DSolve needs to select the proper contours along which to perform the integrals. Probably, DSolve chose incorrect contours.

$\endgroup$
1
  • $\begingroup$ thanks, it does work now. I thought that Assuming would take care of the branch cuts but it's not the case. Only one thing, you have a sol1 written instead of a sol3 in your answer. Thanks again! $\endgroup$ – bnado Nov 27 '19 at 21:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.