I know that if you have a bump function $B$

$$B(x) = \begin{cases} e^\frac{-1}{x}, & x>0 \\[2ex] 0, & x\le0 \end{cases}$$

and some other arbitrary smooth function $f$ (i.e. $f(x)=x^2$) and create a secondary function to bind both $B$ and $f$ to make $f=B$ for some interval and $f\neq B$ anywhere outside of that interval (as done in the comments here).

However, if you have two functions, $f$ and $g$ (say $g(x)=x$), is there a way to use a bump function to smoothly "connect" these two functions (you probably need to make sure there is some sort of intersection between $f$ and $g$ anyways, $g(x)=x$ may be a poor example).

The behavior I am looking for is that, when $B\le0$, $B=f$ and when $B\ge1$, $B=g$, and across the interval $(0,1)$ $B$ is a $C^\infty$ function where $B\neq f$ and $B\neq g$. I'm sure there is some sort of restriction between what $f$ and $g$ can be, which may not be properly represented in this example.


1 Answer 1


The function that's more useful for your purpose is the so-called bump function:

$$f : \mathbb{R} \to \mathbb{R}, x \mapsto \begin{cases} \exp\left( -\frac{1}{1 - x^2} \right) & \text{ if } |x| < 1, \\ 0 & \text{ else}. \end{cases}$$

The reason this one is nicer is because it is also a smooth function, but this one is compactly supported, so outside of the small region $(-1,1)$, this is zero, so it's a useful tool for making tiny fixes to your function around small areas. It basically behaves the same way that your function $e^{-\tfrac{1}{x}}$ behaves at zero, but at two different points.

What's more, consider what happens if you integrate this function (and for reasons you'll see in a moment, we also normalize the integral

$$F : \mathbb{R} \to \mathbb{R}, x \mapsto \frac{\int_{-\infty}^x f(y) dy }{\int_{-\infty}^\infty f(y) dy}.$$

This is clearly again a smooth function, and it has the property that if $x < -1$, then $F(x) = 0$, and if $x > 1$, then $F(x) = 1$. Hence, if you have any other function $h : \mathbb{R} \to \mathbb{R}$, the function $F \cdot h$ will have

$$(F \cdot h )|_{(-\infty,-1)} = 0, \quad (F \cdot h )|_{(1,\infty)} = h|_{(1,\infty)}. $$

Hence, if you want to glue two functions $g$ and $h$ together, multiply one of them with $F$, multiply the other one with a horizontally flipped $F$, and add them together! (of course, this $F$ does the connecting across the interval $(-1,1)$ and not $(0,1)$ as you asked, but that's just a matter of translating and rescaling the function).

If you would instead like to stay with your $e^{-1/x}$-function, you can also look into the answers to this question, there are some constructions which are even more elementary :)

Note in particular: you do not need any specific requirements for your functions, you can glue all two functions together that you like!


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .