Is there a link between antiderivative and measures?

71 Views Asked by At

I struggle to link the concepts I learnt in measure theory and the one I learnt in analysis talking about antiderivative.

Especially, this equality disturbs me (not in the sense that I don't agree with it, but just that it seems to be contradicting antiderivatives) :

$$ f(x_0) = \int f(x) d( \delta_{x_0} ) $$

Are antiderivatives particularly specific to analysis over classical fields like $\mathbb R^n$ and $\mathbb C^n $?

Is it still true that in general an integral is continuous ? That was a given in my head and I have the impression that in fact it is not true in general...

To make that last sentence clearer :

I know that $$f(x) = \int^x_{-\infty} g(y) dy $$ are continuous functions.

Depending on the continuity of $g$ wrt $z$, $$f(z) = \int g(x,z) dx $$ is also continuous. What about the case where the variable appears in the measure?

I believe the question of antiderivative and continuity are linked. Please let me know if it is unclear.

1

There are 1 best solutions below

0
On BEST ANSWER

I'm not entirely sure what you mean when you ask if "an integral is continuous". Do you mean that, given a regular measure $\mu$ on $\mathbb{R}$ and $x_0\in\mathbb{R}$, the function $x\mapsto \mu([x_0,x])$ is continuous on $[x_0,\infty)$? This is not true (e.g. take the point-mass $\delta_1$ and $x_0=0$).


Now on to you question.

In principle, measures are simply a notion of "size" on a set. Nothing about derivatives.

So let us consider a probability space $(X,\mu)$. Could be $[0,1]$ with the usual Lebesgue measure, for example. More generally, one common interpretation is that $X$ is a set of possible states of some physical system (e.g. could be $\mathbb{S}^2\times[0,1]$, whose elements $(x,t)$ have a point $x$ in a sphere/the earth and a point $t$ in time).

Then consider some positive $L^1$ function $f$ on $X$. This is just something that we can measure on our system (e.g. $f(x,t)$ is the temperature of point $x$ of the sphere at time $t$). We can consider the function $$F\colon A\mapsto\int_A f(x)d\mu(x)$$ defined for measurable sets $A$. This gives us "how much of whatever $f$ measures is there in $A$".

It just so happens that the map above is also a measure on $X$. It can also be seen as a "weighted version of $\mu$ by $f$". Since the set "$A$" is the argument of the function $F$, we can give it a better name by just ommiting $A$ in the formula: Thus $$\int fd\mu$$ is a measure, which takes a set $A$ to $\int_A f(x)d\mu(x)$.


We can go in the opposite direction via the Radon-Nikodym theorem: If $\nu$ and $\mu$ are nice enough measures on $\nu$ and $\nu\ll\mu$, then there exists a function $f$ on $X$ such that $\nu=\int fd\mu$. This function $f$ is unique (up to null sets), and is denoted by $\frac{d\nu}{d\mu}$. It is called the Radon-Nikodym derivative of $\nu$ wrt $\mu$.

The following formulas are all valid, as long as everything is sufficiently regular:

  • $\dfrac{d\left(\int fd\mu\right)}{d\mu}=f$;
  • $\int 1d\mu=\mu$;
  • $\int\left(\dfrac{d\nu}{d\mu}\right)d\mu=\nu$.

so Radon-Nikodym derivatives are really the opposite of integrals.


Ok. So now we have a new version of derivatives, which have nothing to do, it seems, with the derivatives in Calculus. Let's see what we can do:

Consider a sufficiently regular measure $\mu$ on $\mathbb{R}$. Let us associate to $\mu$ the function $$F_\mu(x)=\begin{cases}\mu(0,x]&\text{ if }x>0\\-\mu(x,0]&\text{ if }x<0\end{cases}$$ The measure $\mu$ is completely determined by the function $F_\mu$ (in the sense that $F_\mu=F_{\mu'}$ implies $\mu=\mu'$).

Nice. So (sufficiently regular) measures on $\mathbb{R}$ are functions.

Let's do that with the simplest measures we can think of: Let $f\colon\mathbb{R}\to\mathbb{R}$ be continuous, and take $\lambda$ the Lebesgue measure on $\mathbb{R}$. Construct the measure $\int fd\lambda$. By the discussion above, this measure is determined by the function $$F(x)=\int_0^x f(t)dt$$

But we have seen above that the Radon-Nikodym derivative of $\int fd\lambda$ is $f$. How does that translate to the function $F$? The answer is given by the Fundamental Theorem of Calculus:

Fundamental Theorem of Calculus evolved: If $\lambda$ is the Lebesgue measure on $\mathbb{R}$ and $\mu\ll\lambda$ is another sufficiently regular measure, then the Radon-Nikodym derivative $\frac{d\mu}{d\lambda}$ is the derivative in the sense of calculus of the function $F_\mu$ above ($\lambda$-almost everywhere).