8
$\begingroup$

Mathematica 11.3.

I am not sure if this solution given by Mathematica is correct. But I'd like to ask the experts.

eqn = 2*Sqrt[x] y'[x]==Sqrt[1-y[x]];
sol = DSolve[eqn,y,x]

Mathematica graphics

(eqn/.sol)//Simplify

Now the above is supposed to give True if the solution satisfies the ODE. But it does not

Mathematica graphics

I found the above is true only under the following condition

 Assuming[ (C[1] + Sqrt[x]) < 0, Simplify[(eqn /. sol)]]

Mathematica graphics

Therefore, I think the solution given by Mathematica could not be considered valid in general. It is true only under specific condition.

Question is: Is Mathematica solution to the above ODE correct, and if so, why it does not give True then when substituting the solution back to the ODE?

http://reference.wolfram.com/language/tutorial/DSolveSolutionVerification.html

Mathematica graphics

Appendix

The step by step solution to the above problem is below.

The ODE to solve is \begin{align*} 2\,\sqrt {x}{\frac {\rm d}{{\rm d}x}}y \left( x \right) =\sqrt {1-y \left( x \right) } \end{align*} Since the ODE is $y'=1/2\,{\frac {\sqrt {1-y}}{\sqrt {x}}}$, then this is separable. It can be written as $$ y' = f(x) g(y) $$ Where $f(x)={\frac {1}{\sqrt {x}}}$ and $g(y)=1/2\,\sqrt {1-y}$, therefore $$ y' = \left({\frac {1}{\sqrt {x}}}\right)\left(1/2\,\sqrt {1-y}\right) $$
Hence
\begin{align*} \left(2\,{\frac {1}{\sqrt {1-y}}}\right)\mathop{\mathrm{d}y} &= {\frac {1}{\sqrt {x}}}\mathop{\mathrm{d}x}\\ \int \left(2\,{\frac {1}{\sqrt {1-y}}}\right)\mathop{\mathrm{d}y} &= \int {\frac {1}{\sqrt {x}}}\mathop{\mathrm{d}x}\\ -4\,\sqrt {1-y}+{\it C}&=2\,\sqrt {x}\\ \end{align*}

The above is the final solution. I will leave it implicit.

$\endgroup$
  • $\begingroup$ From your manual implicit solution it follows that,$-4\sqrt{1-y}=2\sqrt{x}+2C1 \implies y=1-\frac{x}{4}-C1 \frac{\sqrt{x}}{2}-\frac{C1^2}{4}$. And that is what Mathematica outputs $\endgroup$ – Subho Jun 9 '18 at 2:35
  • $\begingroup$ @Subho95 But as I mentioned, Mathematica itself says this solution does not satisfy the ODE. I also checked using Maple's odetest. Maple say this solution do not satisfy the ODE. $\endgroup$ – Nasser Jun 9 '18 at 2:51
9
$\begingroup$

The explicit solution given by DSolve is actually Solved from the implicit solution. This can be verified by:

Trace[DSolve[eqn, y, x], Solve[_, y[x]], TraceInternal -> True] // Flatten
(* {HoldForm[Solve[-2 Sqrt[1 - y[x]] == Sqrt[x] + C[1], y[x]]]} *)
% // ReleaseHold
(* {{{y[x] -> 1/4 (4 - x - 2 Sqrt[x] C[1] - C[1]^2)}}} *)

So the question boils down to "why doesn't Solve give the condition for the solution", and this has been explained in Possible Issues of document of Solve:

Solve gives generic solutions; solutions involving equations on parameters are not given… With MaxExtraConditions -> All, Solve also gives non-generic solutions.

Solve[-2 Sqrt[1 - y[x]] == Sqrt[x] + C[1], y[x], MaxExtraConditions -> All]
(* Alternatively: *)
Solve[-2 Sqrt[1 - y[x]] == Sqrt[x] + C[1], y[x], Method -> Reduce]
(* {{y[x] -> ConditionalExpression[1/4 (4 - x - 2 Sqrt[x] C[1] - C[1]^2), 
    Sqrt[x] + C[1] + Sqrt[(Sqrt[x] + C[1])^2] == 0]}} *)
$\endgroup$
  • $\begingroup$ Actually, the main question I have, which in the post, is that verifying the solution against the ODE does not give True. This is the method documentation says to use to verify the solution. Hence if the solution is supposed to be "correct", why it does not verify the ODE. So may be this should be mentioned in the documentation, that extra conditions could be needed to verify the solution. $\endgroup$ – Nasser Jun 9 '18 at 3:56
  • 2
    $\begingroup$ @Nasser Perhaps WRI considers this as a performance issue to be fixed in the future so they decided not to document it? Compare the following: Solve[-Sqrt[y] == Sqrt[x], y] and Solve[-Sqrt[y] == 1 + Sqrt[x], y]. Anyway, I think reporting it to WRI isn't a bad idea. $\endgroup$ – xzczd Jun 9 '18 at 4:15
  • 1
    $\begingroup$ good example. You get {} from both by adding Reals !Mathematica graphics $\endgroup$ – Nasser Jun 9 '18 at 4:19
3
$\begingroup$

Picking up from my earlier comment, note that you first have to define a branch for square-root(positive/negative). The conventional branch is positive. Once that is agreed upon, eyeball the implicit solution in the following form:

$$-4\sqrt{1-y}=2\sqrt{x}+2C_1$$

The LHS is non-positive, and so should be the RHS. This information gets lost when one squares both sides to get the explicit solution.

This means that $\sqrt{x}+C_1 \leq 0\; \forall C_1\; \forall (x,y)$ satisfying the above equation.

Note that here $C_1=-\frac{C}{2}$ as you mentioned in the implicit solution in the question.

$\endgroup$
3
$\begingroup$

What we learn here is, always try to verify solutions. You might get conditons for variables and parameters.

test = (eqn /. First@sol) // Simplify

red = test // Reduce[#, Reals] &

(*   (C[1] < 0 && 0 <= x <= C[1]^2) || (C[1] == 0 && x == 0)   *)

RegionPlot[red /. C[1] -> a, {x, -2, 5}, {a, -6, 6}, 
    FrameLabel -> {x, "C[1]"}]

enter image description here

$\endgroup$

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.