1

Recently, I am trying to derive an algorithm where one step requires me to solve the equation of the form $axe^x+bx+c=0$ where $a, b, c$ are all constant and $x$ is a scalar variable.

Any help would be appreciated.

Jingx
  • 13
  • 1
    I'd recommend adding your attempt or thought process for the questions. – reyna Oct 19 '20 at 14:21
  • 2
    Thanks @zaira. I tried to solve this question using https://math.stackexchange.com/q/1238668 as a hint. But I cannot write the equation into the form of $u e^u = w$ as suggested by https://math.stackexchange.com/q/1238745. And now I have no idea how to solve this question. – Jingx Oct 19 '20 at 14:49
  • I shall continue tomorrow morning (dinner time here) – Claude Leibovici Oct 19 '20 at 15:06
  • @ClaudeLeibovici: enjoy your meal. –  Oct 19 '20 at 15:11
  • Rather than leaving such context in the comments and being vague in the question, you should [edit] the question to include these details. Others may be interested, or such context may help drive answers. This also includes other information, such as the possible values of $a,b,c$ for example. Your tags include [tag:matrix-exponential], which suggests these may be matrices. If so, what are their dimensions? – Simply Beautiful Art Oct 19 '20 at 21:47

2 Answers2

0

In the most general case, there is no explicit solution to this equation except the generalized Lambert equation since you can write $$e^{-x}=-\frac{a x}{b x+c}$$ (have a look at equation $(4)$ in the linked paper.

This being said, it is nice but not very useful from a practical point of view; so, numerical methods are required and, as a consequence, reasonable starting points.

Without any loss of generality, assuming $a \neq 0$ we can divide all coefficient by $a$ or , simpler, let $a=1$. Now, consider that we look for the zero('s) of function $$f(x)=xe^x+bx+c$$ for which $$f'(x)=e^x (x+1)+b \qquad \text{and} \qquad f''(x)=e^x (x+2)$$ The first derivative cancels at $$x_1=W(-eb)-1$$ where appears Lambert function. In the real domain, this point exists if $b \leq \frac 1{e^2}$.

At this point, we have $$f(x_1)=b W(-e b)+\frac{b}{W(-eb)}-2 b+c $$ $$f''(x_1)= - b \left(1+\frac{1}{W(-eb)}\right)$$ We must notice that $$f(x_1)-c=b W(-e b)+\frac{b}{W(-eb)}-2 b$$ is always negative.

So, if $c <0$, no root. If $c=0$, two possible roots located at $x=0$ and $x=-\log(-b)$ (if the last does exist). If $c<0$, two roots.

There is also an inflection point at $$x_2=-2 \implies f(x_2)=-2 b+c-\frac{2}{e^2}$$ So, if $c<2b+\frac{2}{e^2}$, another possible root.

There two roots that we can locate more or less since the will be on each side of $x_1$. Building a Taylor series around $x_1$, we have as estimates $$x_\pm=x_1 \pm \sqrt{-2 \frac {f(x_1)}{f''(x_1)}}$$

For example, let us try for $b=-0.1$ and $c=-1$. This will give $x_+=1.36$ and $x_-=-2.92$. So, using Newton method, the following iterates $$\left( \begin{array}{cc} 0 & -2.92363 \\ 1 & -7.17566 \\ 2 & -9.92501 \\ 3 & -10.0045 \end{array} \right)$$ $$\left( \begin{array}{cc} 0 & 1.36058 \\ 1 & 0.90268 \\ 2 & 0.65533 \\ 3 & 0.59170 \\ 4 & 0.58807 \\ 5 & 0.58806 \end{array} \right)$$

0

Not an answer, just a hint.

The equation can be written in the form

$$xe^x=mx+p,$$ which represents the intersection between a transcendental curve and an arbitrary straight line. The curve has an inflection point and there can be up to three solutions.

enter image description here

To discuss the number of roots, you can rely on the tangents that have the slope $m$. Unfortunately, this leads to a nasty equation,

$$(x+1)e^x=m$$ that requires the Lambert function, with

$$x=W(em)-1.$$ There can be two solutions. The intercept of the tangent is then

$$xe^x-mx$$ which you compare to your $p$ to determine the number of solutions and to isolate them between the contact points.