Is there a formula or algorithm to remove infinitesimal and oscillating parts from an expression while keeping finite and infinite ones?

59 Views Asked by At

Below, we interpret divergent integrals as germs of partial integrals at infinity:

$$\int_0^\infty f(x) dx=\operatorname{bigpart} \int_0^\omega f(x) dx$$

where $\operatorname{bigpart}$ means taking finite and infinite parts of $\int_0^\omega f(x) dx$ at $\omega\to\infty$ while throwing away infinitesimal and oscillating (with zero average) parts. The question is, how can we define this operator?

For instance,

$$\int_0^\omega 1 dx=\omega.$$

It is infinite, so our desired result is $\int_0^\infty 1 dx=\omega$. Thus, $\omega$ plays the role of an infinite constant.

$$\int_0^{\omega } \exp (x) \sin (2 x) \, dx=\frac{2}{5}-\frac{2}{5} e^{\omega } (\cos (2 \omega )-\sin (\omega ) \cos (\omega )).$$

Here, $2/5$ is finite part. The term $-\frac{2}{5} e^{\omega } (\cos (2 \omega )-\sin (\omega ) \cos (\omega ))$ is oscillating with zero average, so should be taken to be equal to zero. So, the desired result is $2/5$.

$$\int_0^{\omega } \cos ^2(x) \, dx=\frac{\omega }{2}+\frac{1}{4} \sin (2 \omega ).$$

The part $\frac{1}{4} \sin (2 \omega )$ is oscillating with zero average, so the desired result is the infinite part $\frac{\omega }{2}$.

$$\int_0^{\omega } \exp (\log (x)+x) \, dx=e^{\omega } (\omega -1)+1.$$

Here we have infinite and finite parts, so the desired result should be kept intact: $\int_0^\infty \exp (\log (x)+x) dx=e^{\omega } (\omega -1)+1$.

On the other hand,

$$\int_0^{\omega } \exp (\log (x)-x) \, dx=e^{-\omega } (-\omega -1)+1.$$

The term $e^{-\omega } (-\omega -1)$ is infinitesimal, so we throw it away, and our desired result is $\int_0^{\infty } \exp (\log (x)-x) \, dx=1$.

Is there a way to define such function $\operatorname{bigpart}$ consistently and automatize the process in a CAS system?

1

There are 1 best solutions below

0
On

As far as the oscillating part is concerned, you provided a solution yourself in terms of averages. As far as decomposition into infinite + real + infinitesimal, this can be done in nonstandard analysis more generally than the context of integration as follows. Consider the standard part function $[\;]\colon\mathbb R\to [0,1)$. It has a natural extension to the hyperreals: $[\;]^\ast\colon \mathbb R^\ast \to [0,1)^\ast$. As usual, we will drop the asterisk on functions (but not on sets). Then given an arbitrary hyperreal $x$, we can decompose it as $x= [x]+r$ where $r$ is a finite hyperreal (between $0$ and $1$). Since $r$ is finite, we can decompose $r$ into its standard part st$(r)$ and an infinitesimal $\alpha$. In sum, we get $$x=[x]+\text{st}(r)+\alpha$$ (where $[x]$ could be infinite).