Here are three (possibly different) definitions of $p$-adic integrals that I have encountered during my self-studies.
First of all, here is what Vladimirov, Volovich and Zelenov write at the beginning of their book on mathematical physics:
As the field $\mathbb Q_p$ is a locally compact commutative group with
respect to addition then in $\mathbb Q_p$ there exists the Haar
measure, a positive measure $dx$ which is invariant to shifts. We
normalize the measure $dx$ such that $\int_{|x|_p \le 1} dx = 1.$
Under such agreement the measure is unique.For any compact $K \subseteq \mathbb Q_p$ the measure $dx$ defines a positive linear
continuous functional on $C(K)$ by the formula $\int_K f(x) dx$. A
function $f \in \mathcal L^1_{\textrm{loc}}$ is called integrable on
$\mathbb Q_p$ if there exists $$\lim_{N \to \infty} \int_{B(N)} f(x)dx = \lim_{N \to \infty} \sum_{\gamma = -N}^\infty \int_{S(-\gamma)} f(x) dx.$$
Their denote by $B(N)$ the set $\{x \in \mathbb Q_p : |x|_p \le p^N \}$ and $S(\gamma)$ is defined as $B(\gamma) \setminus B(\gamma-1)$.
Next we have the famous Volkenborn integral, described as follows by Robert:
We say that $f$ is strictly differentiable at a point $a \in X$ – and
denote this property by $f \in \mathcal S^1(a)$ – if the difference
quotients $[f(x) – f(y)]/(x-y)$ have a limit as $(x, y) \to (a,a)$
($x$ and $y$ remaining distinct). By the way, $\mathcal S^1(X) := \bigcap_{a \in X} \mathcal S^1(a)$.
The Volkenborn integral of a function $f \in \mathcal S^1(\mathbb Z_p)$ is by definition $$\int_{\mathbb Z_p} f(x) dx = \lim_{n \to
\infty} \frac{1}{p^n} \sum_{j=0}^{p^n-1} f(j).$$
Finally a quote from Koblitz book "$p$-adic numbers, analysis and $\zeta$-functions":
Now let $X$ be a compact-open subset of $\mathbb Q_p$. A $p$-adic distribution $\mu$ on $X$ is a $\mathbb Q_p$-linear vector space homomorphism from the $\mathbb Q_p$-vector space of locally constant functions on $X$ to $\mathbb Q_p$.
Later he states that for $p$-adic measures (distributions that are bounded on compact-open subsets by some constant) and continuous functions $f$ there is a reasonable way to define $\mu(f) =: \int_X f \mu$.
Cassels is confusing me even more as he mentions Shnirelman. So, here are my actual questions:
- Do these ultrametric integrals have real analogues like: being a limit of Riemannian sums or an operation inverse to differentiation?
- Are they compatible to each other? Is any of them a generalization of the others?
- What are the positive and negative attributes of the cited definitions?
- Where can I find an exhaustive table of integrals? I'm mainly interested in something similar to https://www.tug.org/texshowcase/cheat.pdf.
- Can we somehow imitate the Lebesgue integration theory?
Best Answer
Not really an full answer, but some comments (that hopefully answer some of your queries).
There seems to be a big confusion here : what do we want to integrate, i.e. to define $\int_{\mathbb{Z}_p} f(x)dx$, what is the 'nature' of $f$, and of the result ?
For the first one, (it's the one I am familiar with), the Haar measure on $\mathbb{Z}_p$ is in particular a map $\mu$ that assign to a Borel subset (say $E$) of $\mathbb{Z}_p$ real number $\mu(E)$. Take $f=1_E$ the characteristic function of $E$, then $$\int_{\mathbb{Z}_p} 1_E d\mu=\mu(E) \in \mathbb{R},$$ by definition. The result is then a real number. That tells us that measure theory is about integration of functions with value in $\mathbb{R}$ !
That means that in this context, for example, $\int_{\mathbb{Z}_p} xd\mu(x)$ has no meaning from the point of view of this definition of integral.
If one is to look for an analogue of Riemann sums, we may notice that the sequence $(n)_{n\in \mathbb{N}}$ is equidistributed on $\mathbb{Z}_p$ with respect to $\mu$ (An ergodic theorist would say that the transformation $T:\mathbb{Z}_p\to \mathbb{Z}_p, T(x)=x+1$ is uniquely ergodic). The analogue of Weyl's criterion holds: for $f$ a real-valued continuous function, bounded on $\mathbb{Z}_p$, then $$\int_{\mathbb{Z}_p} fd\mu=\lim_{N\to +\infty} \frac1N \sum_{n=0}^{N-1} f(n).$$
Another thing: the formula for integration on $\mathbb{Q}_p$ in the OP is in this case an analogue of the definition of the improper integral: $$\int_{-\infty}^{\infty} fdx := \lim_{T\to +\infty} \int_{-T}^T f(x)dx,$$ which allows to make sense of functions whose integral in the measure theory sense has no meaning, like $f(x)=\sin(x)/x$.
About the Volkenborn integral, although this is not said, if we are to believe https://en.wikipedia.org/wiki/Volkenborn_integral, is made to integrate functions $f$ with values in $\mathbb{C}_p$ (but let's reduce it to functions with values in $\mathbb{Q}_p$ to be able to compare with the latter definition). The definition given in the OP $$\int_{\mathbb{Z}_p} f(x)dx:=\lim_{N\to +\infty} \frac1{p^N} \sum_{n=0}^{p^N-1} f(n),$$ looks pretty much the same than the analogue of Riemann sums above, but with one big difference: one looks only at the subsequence $(p^N)_{N\geq 0}$. Indeed, an easy but instructive example is to compute these Riemann sums for $f(x)=x$. Then $$\frac1{n} \sum_{k=0}^{n-1} f(k)=\frac{n-1}2 \in \mathbb{Q}_p,$$ which does not converge (recall the topology is the $p$-adic one), but does for the subsequence $(p^k)_{k\geq 0}$, to $-1/2$. This gives us $\int_{\mathbb{Z}_p} xdx=-1/2$, as said in the above wikipedia link (which contains, btw, a few formulas as required).
The third definition deals with $\mathbb{Q}_p$-linear vector space homomorphism of locally constant functions to $\mathbb{Q}_p$. So clearly here we are also looking at integration of function $f:\mathbb{Z}_p\to \mathbb{Q}_p$. The Volkenborn integral is an example, locally constant functions being strictly differentiable. But it's not the only one (EDIT : I previously stated that the Volkenborn integral is invariant by translations, which was wrong, as noted by Dap). So this definition is more general (the dirac measures works, for example).
I hope this clarifies a little bit...