Using bounds for an analytically intractable differential equations problem

627 Views Asked by At

Exercise 2.8.6 from Strogatz's Nonlinear Dynamics and Chaos with Applications asks us to consider the initial value problem $\dot x = x+e^{-x}$, $x(0)=0$ $\left(\dot x\, \text{means}\, \frac{dx}{dt}\right)$, which cannot be solved analytically.

Part (b) of the exercise states as follows:

Using some analytical arguments, obtain rigorous bounds for the value of $x$ at $t=1$. In other words, prove that $a<x(1)<b$ for $a$, $b$ to be determined. By being clever, try to make $a$ and $b$ as close together as possible.

I was given a hint from my professor that, in other words, we want to approximate $$\psi(x) \leq x+ e^{-x} \leq \phi(x) $$ for $\phi(x),\,\psi(x) \geq 0$, and that essentially what we want to do is solve the following three IVPs: $$(1) \qquad\frac{dx}{dt} = \psi(x) \\ \qquad x(t_{0})=x_{0}, $$ $$ (2) \qquad \frac{dx}{dt} = x+e^{-x} \\ \qquad x(t_{0}) = x_{0},$$ and $$ (3) \qquad \frac{dx}{dt} = \phi(x) \\ \qquad x(t_{0})=x_{0} .$$ Then, if we call the solution to $(1)$ $a(t)$, the solution to $(2)$ $x(t)$, and the solution to $(3)$ $b(t)$, the problem boils down to showing that $$a(1) \leq x(1) \leq b(1). $$

In that regard, I am very close to finding $b(1)$. To get there, I considered the Taylor expansion for $$e^{-x} = \sum_{n=0}^{\infty} (-1)^{n} \frac{x^{n}}{n!} =1-x + \frac{x^{2}}{2} - \frac{x^{3}}{3!}+ \cdots$$ so that $x+e^{-x} = x + \sum_{n=0}^{\infty} (-1)^{n}\frac{x^{n}}{n!}.$ As a bound, I decided to show that $x+e^{-x} \leq 1 + \frac{x^{2}}{2}$ $\forall x \geq 0.$

To that effect, I let $h(x) = x + e^{-x} - 1 - \frac{x^{2}}{2}$, and then I considered $h^{\prime}(x) = 1 - e^{-x} - x$ and $h^{\prime \prime}(x) = e^{-x} - 1$.

Since $h^{\prime \prime}(x) > 0$ for $x < 0$, $h^{\prime}(x)$ is increasing on $(-\infty, 0)$.

Since $h^{\prime \prime}(x) < 0$ for $x > 0$, $h^{\prime}(x)$ is decreasing on $(0, \infty)$.

But, $h^{\prime}(x) = 0$ for $x = 0$, so $h^{\prime}(x) < 0$ for $x<0$ and $h^{\prime}(x) < 0 $ for $x > 0$ as well - the only time $h^{\prime}(x)$ is not negative is when $x=0$, and there it is simply $=0$.

Therefore, $h$ is everywhere decreasing, and since $h(x)=0$, $h > 0$ for $x<0$ and $h < 0$ for $x > 0$, so we have that $h(x) = x + e^{-x} - 1 - \frac{x^{2}}{2} \leq 0$ for $x \geq 0$, which implies that $x + e^{-x} \leq 1 + \frac{x^{2}}{2} $, as desired.

So, then I let $\phi(x) = 1 - \frac{x^{2}}{2}$, and I need to solve Equation $(3)$ above.

Using separation of variables, I have that $\frac{dx}{dt} = 1 - \frac{x^{2}}{2}$ becomes $\displaystyle \int \frac{1}{\displaystyle 1 - \left(\frac{x}{\sqrt{2}} \right)^{2}} dx = \int dt$, which comes out to be $$\ln \left(1 + \frac{x}{\sqrt{2}} \right) - \ln \left(1 - \frac{x}{\sqrt{2}} \right) = \sqrt{2}t + C, $$

And then, exponentiating both sides, we get $$\frac{1+\frac{x}{\sqrt{2}}}{1-\frac{x}{\sqrt{2}}}= Ce^{\sqrt{2}t} $$

From here, I don't know what to do. The problem tells us what the initial condition when $t=0$ is - it tells us that $x(0) = 0$, but it doesn't tell us what it is for $t=1$. And since we're trying to find bounds for $x(1)$, I'm assuming that for Equation (3) (actually in all three of the equations), we want $t_{0} = 1$, right? The only problem is that I can't figure out what $x_{0}$ is in that case, so how do I solve for $C$?

Also, if you could help me out on the left bound (i.e., solving for $\psi(x)$), I would appreciate that very much!

Thank you ahead of time for your help.

2

There are 2 best solutions below

0
On BEST ANSWER

As $e^x\ge 1+x$, you get $\dot x\ge 1$, which is easier to solve than your lower bound.

In your lower bound computation, inserting $x(0)=0$ gives $C=0$. Then solving for $x$ is just arithmetic/algebraic manipulations.

For the upper bound, as long as $x<3$ we get by the Leibniz test $e^{-x}\le 1-x+\frac12x^2$. You showed that this is even valid for the whole positive half axis. Thus $$ \dot x\le 1+\frac12x^2 $$ Both differential inequalities can be solved via standard integration methods an give an interval of length about $0.2$ for $x(1)$.

One can improve the lower bound to get a similar formula as for the upper bound by using $$e^{-x}=(e^{-x/2})^2\ge\Bigl(1-\frac x2\Bigr)^2$$ to get $$ \dot x\ge 1+\frac14x^2 $$ As we know that $\dot y=1+Ay^2$ has the solution $y(t)=\frac1{\sqrt A}\tan{\sqrt{A}t}$, you can get the bounds $$ 2\tan\frac12\le x(1)\le\sqrt2\tan\frac1{\sqrt2} $$ which falls between $1.09$ and $1.21$.

0
On

As @Evgeny proposed use the inequality

$$x \leq \dot{x} \leq x + 1.$$

Then apply the following Theorem (Petrovitsch, 1901):

If $u$ satisfies the differential inequality $u′(t)≥f(u(t),t)$, and $y$ is the solution to the ODE $y′(t)=f(y(t),t)$ under the boundary condition $u(t_0)=y(t_0)$, then: $\forall t<t_0, u(t)\leq y(t) \,\, \forall t>t_0, u(t)\geq y(t)$.