Sum of weighted exponentials

71 Views Asked by At

Consider $x\in (-\infty, \infty)$, $0\leq d_i<1$, $\mu_i\in \mathbb{R}$ and $\sigma_i\geq 0$ for $i=1,...,s$, $\sum_{i=1}^s d_i=1$.

Assume $$ \sum_{i=1}^s d_i\frac{1}{1+e^{\frac{-(x-\mu_i)}{\sigma_i}}}=0 $$

Statement: if $(\mu_i,\sigma_i)_{i=1}^s$ are distinct across $i$ and $\sigma_i>0$ $\forall i=1,...s$, then $d_i=0$ $\forall i=1,...s$.

I do not understand this statement. What is see is that the equation above can be true iff $d_i=0$ $\forall i=1,...s$ or $x=-\infty$. Can you help me to see the rationale behind the statement? What am I doing wrong?

1

There are 1 best solutions below

0
On BEST ANSWER

Notice that $x=-\infty$ is not a real number.

The statement that you wrote is the following:

IF

  • $x\in (-\infty,+\infty)$, $0\leq d_i<1, \mu_i\in \mathbb{R},\sigma_i>0$ for $i=1,\ldots,s$
  • $\sum_{i=1}^s d_i=1$
  • $\displaystyle{\sum_{i=1}^s d_i \dfrac{1}{1+e^{\frac{-(x-\mu_i)}{\sigma_i}}}=0}$
  • The pairs $(\mu_i,\sigma_i)$ are all pairwise distinct.

THEN the only possibility is $d_i=0$.


This statement as stated is complete non-sense, because if $\sum_{i=1}^s d_i=1$ then it is impossible that $d_i=0$ for all $i=1,\ldots,s$.

Given that one of your tags says algebra-precalculus, I will assume that you can answer this using very elementary properties of the reals. Probably, the easiest similar statement is the following:

IF

  • $0\leq d_i,\mu_i,\sigma_i\in \mathbb{R}, \sigma_i\neq 0$ for $i=1,\ldots,s$
  • and $\displaystyle{\sum_{i=1}^s d_i \dfrac{1}{1+e^{\frac{-(x-\mu_i)}{\sigma_i}}}=0}$ (This equation is the most exotic part of your question, so I guess we should use it)

THEN, the only possibility is $d_i=0$ for all $i=1,\ldots,s$.

This statement can be easily prove if you have in mind that $e^y>0$ for any real number $y$. Therefore, by doing $\alpha_i=e^{\frac{-(x-\mu_i)}{\sigma_i}}$ for all $i=1,\ldots,s$, the equation that you have is a sum $$\displaystyle{\sum_{i=1}^s d_i\cdot \alpha_i=0}$$ with $S=\alpha_i>0$ and $d_i\geq 0$.

If $d_i>0$ for some $i$, then $S\geq \alpha_i\cdot d_i>0$ because the product of two positive number is positive. But that would contradict the hypothesis that $S=0$.

Therefore, the only option is $d_i=0$ for all $i=1,2,\ldots,s$.