1
$\begingroup$

Cross-posted from Math SE.

The linked question explores the concept of a "generalized limit" that assigns values to sequences which diverge in the Cauchy sense. It asks the following question:

Does there exist an "explicitly definable" generalized limit $\operatorname{Lim} : X^\mathbb{N} \rightharpoonup X$ that is stronger than the linear and stable closure of the Cauchy limit?

I'd be curious to know what such a $\operatorname{Lim}$ might be.

As Gerald pointed out, the concept of an almost convergent sequence is relevant.

$\endgroup$
6
  • $\begingroup$ What does it mean for a generalised limit to be 'stronger' than something? $\endgroup$
    – LSpice
    Commented Nov 3, 2019 at 21:29
  • 1
    $\begingroup$ @LSpice $\operatorname{Lim}_1$ is stronger than $\operatorname{Lim}_2$ iff $\operatorname{Lim}_2 \subseteq \operatorname{Lim}_1$. $\endgroup$
    – user76284
    Commented Nov 3, 2019 at 22:01
  • $\begingroup$ Ah, so the $\rightharpoonup$ notation indicates a partial function? $\endgroup$
    – LSpice
    Commented Nov 4, 2019 at 2:13
  • $\begingroup$ @LSpice That's right. $\endgroup$
    – user76284
    Commented Nov 4, 2019 at 2:14
  • 1
    $\begingroup$ It might be of interest to you that there is a vast literature on the topic of generalized limits going back at least to Hardy's "Divergent series". The most recent one that I know of is "Classical and modern methods in summability" by J. Boos which contains references to earlier work. There seems to be a cultural divide between those who discuss convergence of sequences, respectively summability of series but they are mathematically equivalent. $\endgroup$ Commented Mar 29, 2021 at 10:12

2 Answers 2

1
$\begingroup$

Maybe this is not what generalization you are looking for, because it extends the range of possible values of generalized limits beyond the complex numbers but it can possibly be interesting to you (and any generalized limit has a regularized part which may be of interest).

The approach is based on evaluating the integral of the derivative of the function at a point in the following way:

$$\operatorname{gen}\lim_{x\to u^+}f(x)=f(a)+\int_u^a f'(x)dx$$

$$\operatorname{gen}\lim_{x\to u^-}f(x)=f(a)+\int_a^u f'(x)dx$$

After we made an association between the limit and the corresponding integral, all we need to do is to ascribe a value to the divergent integral.

The simplest method is to sum up the integral using Cesaro summation, this will give us the average value of the function as it approaches the point of interest.

But in cases when the integral grows to infinity Cesaro summation won't work.

So, in this case we would want to express the infinitely-large divergent integrals through other divergent integrals, and also hopefully find the regularized values of them.

For instance, it is quite natural that if we assign some value to the integral $\int_0^\infty dx$ then $\operatorname{gen}\lim_{x\to\infty} x$ would be that same value and $\operatorname{gen}\lim_{x\to\infty} 2x$ would be twice that much.

That said I refer you to this post regarding divergent integrals.

To summarize that approach, it turned out that employing some parallels in fourier analysis and the theory of hyperfunctions, it is possible to write down in closed form a large range of generalized limits.

if we define

$$\tau=\int_0^\infty dx=\pi\delta(0)$$

$$\omega_+=\tau+1/2$$

and

$$\omega_-=\tau-1/2$$

then there is a lot of limits that can be described in these terms.

For instance,

$$\omega_-^n=\operatorname{gen}\lim_{x\to\infty}B_n(x)$$

$$\omega_+^n=\operatorname{gen}\lim_{x\to\infty}B_n(x+1)$$

$$\operatorname{gen}\lim_{x\to\infty}x^n=0^n+i^{n-1}\pi n\delta^{(n-1)}(0)=\frac{\omega _+^{n+1}-\omega _-^{n+1}}{n+1}$$

$$\operatorname{gen}\lim_{x\to 0^+} \frac1{x^n}=0^n+\frac{i^{n-1}\pi\delta^{(n-1)}(0)}{(n-1)!}=\frac{\omega _+^{n+1}-\omega _-^{n+1}}{(n+1)!}$$

$$\operatorname{gen}\lim_{x\to\infty}\frac{x^3}3=\int_0^\infty x^2dx=\tau^3 +\frac\tau{4}=\pi \delta''(0)$$

$$\operatorname{gen}\lim_{x\to\infty}\frac{x^2}2=\int_0^{\infty } x \, dx = \frac{\tau ^2}{2}+\frac{1}{24} = i\pi\delta'(0)$$

It is also possible to evaluate the function's poles using generalized limits. The limits would be polynomials of $\tau$ of the same order as the function's pole, with Cauchy's principle value representing the regularized part.

$$\operatorname{gen}\lim_{x\to0^\pm}\Gamma(0)=-\gamma\pm\tau$$

$$\operatorname{gen}\lim_{x\to{-1}^\pm}\Gamma(x)=\gamma-1\mp\tau$$

$$\operatorname{gen}\lim_{x\to{-2}^\pm}\Gamma(x)=\frac{3}{4}-\frac{\gamma }{2}\pm\frac\tau 2$$

$$\operatorname{gen}\lim_{x\to{-3}^\pm}\Gamma(x)=\frac{\gamma }{6}-\frac{11}{36}\mp\frac\tau 6$$

$$\operatorname{gen}\lim_{x\to1^\pm}\zeta(x)=\gamma\pm\tau$$

Here is a table with some other values.

$\endgroup$
6
  • 2
    $\begingroup$ One has to be very conservative extending the range of possible limits beyond the complex numbers, or else you could do something like taking the codomain to be the quotient of $X^{\mathbb N}$ by the ideal of sequences tending to $0$. (Of course you haven't done this, but what I mean is that it might be easy to sneak up accidentally on doing this by innocent-seeming steps.) $\endgroup$
    – LSpice
    Commented Nov 4, 2019 at 16:59
  • $\begingroup$ @LSpice can you please provide an example of what your concern can lead to. $\endgroup$
    – Anixx
    Commented Nov 5, 2019 at 14:45
  • 1
    $\begingroup$ What I meant is that the natural projection $X^{\mathbb N} \to X^{\mathbb N}/c_0(X)$ is a reasonable, if not very interesting, kind of "generalised limit with generalised values", and, via the inclusion of constant sequences $X \hookrightarrow X^{\mathbb N}$, agrees with the usual limit on convergent sequences; but it seems likely it's not what @user76284 had in mind. (It's also not what you said! I just meant that some care needs to be taken when extending the admissible values of the limit.) $\endgroup$
    – LSpice
    Commented Nov 5, 2019 at 17:01
  • $\begingroup$ @LSpice as an algebraist, what do you think about algebraic properties of the linked system? I posted a question about this here: math.stackexchange.com/questions/3390772/… $\endgroup$
    – Anixx
    Commented Nov 5, 2019 at 19:20
  • 3
    $\begingroup$ With all due respect, I think that advertising about it in questions to which it does not seem directly related doesn't belong here; but the voting on your answer shows that that might be a minority opinion. $\endgroup$
    – LSpice
    Commented Nov 5, 2019 at 21:03
0
$\begingroup$

Your condition 2 adds nothing. Given any $\text{Lim}$ you can write it as $\lim \circ f$ where $f$ maps each sequence $\mathbf x$ to the constant sequence with value $\text{Lim}(\mathbf{x})$.

$\endgroup$
7
  • $\begingroup$ But is there an explicit such $f$? $\endgroup$
    – user76284
    Commented Nov 3, 2019 at 18:54
  • 1
    $\begingroup$ I guess $f$ is explicit exactly when $\text{Lim}$ was already explicit. $\endgroup$ Commented Nov 3, 2019 at 18:55
  • $\begingroup$ Maybe consult the material on Banach limits, in particular connect your ideas to the notion "almost convergent" found there. en.wikipedia.org/wiki/Banach_limit $\endgroup$ Commented Nov 3, 2019 at 18:58
  • $\begingroup$ To give an idea of what I meant by an "explicit" sequence transformation, one example is the Cesàro transformation described in the second half of the original question. $\endgroup$
    – user76284
    Commented Nov 3, 2019 at 19:01
  • 1
    $\begingroup$ @user76284, re: your comment, while it is a good idea to edit your question, it is not generally considered polite to do so in a way that completely invalidates existing answers. $\endgroup$
    – LSpice
    Commented Nov 3, 2019 at 21:28

You must log in to answer this question.

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