2
$\begingroup$

Let us consider the sum of a function series $$ \sum _{n=1}^{\infty } \frac{z^n}{\left(z^n+1\right)^n}.$$

This series absolutely converges if Abs[z] < 1 or Abs[z] > 1 or z==1 in view of the results of

DiscreteMaxLimit[(Abs[z]^n/Abs[(1 + z^n)]^n)^(1/n), n -> Infinity, 
Assumptions -> Abs[z] < 1]

Abs[z]

and

DiscreteMaxLimit[(Abs[z]^n/Abs[(1 + z^n)]^n)^(1/n), n -> Infinity, 
Assumptions -> Abs[z] > 1]

0

and the root test. Its sum presents two different analytic functions having the unit disk $\{z:|z| < 1\}$ and its exterior $\{z:|z| > 1\}$ as the natural domains of existence (see Encyclopedia of Mathematics for info) since the term $\frac{z^n}{\left(z^n+1\right)^n}$ has $n$ poles of order $n$ uniformly distributed on the unit circumference.

ComplexPlot3D[Sum[z^n/(1 + z^n)^n, {n, 1, 20}], {z, 2}]

enter image description here

illustrates it.

In 14.0 on Windows 10 I try to plot this function for real values of $x$ from $0$ to $2$.

Here are my not very successful attempts taking a lot of time.

Plot[Sum[x^n/(1 + x^n)^n, {n, 1, Infinity}], {x, 0, 2}, PlotRange -> All]

enter image description here

Plot[NSum[x^n/(1 + x^n)^n, {n, 1, Infinity}], {x, 0, 2}, 
PlotRange -> All, WorkingPrecision -> 25]

enter image description here

Replacing the function under consideration by its partial sums, I obtain

Plot[{Sum[x^n/(1 + x^n)^n, {n, 1, 100}], Sum[x^n/(1 + x^n)^n, {n, 1, 1000}], 
Sum[x^n/(1 + x^n)^n, {n, 1, 10000}]}, {x, 0, 2}, PlotRange -> All]

enter image description here

and

General::munfl: 0.0000408571^71 is too small to represent as a normalized machine number; precision may be lost.

The latest approach is not reliable (think of Sum[x^n,{n,1,Infinity}] and Sum[x^n,{n,1,1000}]). The hard piece is that

Plot[{Sum[x^n/(1 + x^n)^n, {n, 1, 100}],  Sum[x^n/(1 + x^n)^n, {n, 1, 1000}], 
 Sum[x^n/(1 + x^n)^n, {n, 1, 10000}], 
 58 - 163 x + 156 x^2 - 50 x^3, 
 Sum[x^n/(1 + x^n)^n, {n, 1, 20000}]]}, {x, 0.95, 1.05}, PlotRange -> All]

enter image description here

General::munfl: 0.950002^13812 is too small to represent as a normalized machine number; precision may be lost.

Plot[NSum[x^n/(1 + x^n)^n, {n, 1, Infinity}], {x, 0.95, 1.05}, 
PlotRange -> All, WorkingPrecision -> 15, MaxRecursion -> 4, PerformanceGoal -> "Quality"]

enter image description here

NumericalMath`NSequenceLimit::seqlim: The general form of the sequence could not be determined, and the result may be incorrect.

We see the last two plots differ for $x<1$. 58 - 163 x + 156 x^2 - 50 x^3 is nothing but the wrong result of Series[Sum[x^k/(1 + x^k)^k, {k, 1, Infinity}], {x, 1, 3}] // Normal: if this series were convergent , then there is an analytic continuation through 1.

So the question is: how to reliably plot the function under consideration on the real range $[0,2]$?

$\endgroup$
4
  • 4
    $\begingroup$ (1) Use 50 terms. (2) Show the error bound is smaller than you can discern on a plot. $\endgroup$ Commented Feb 27 at 21:27
  • $\begingroup$ @DanielLichtblau: Your instructions are of this kind: to walk on a tightrope, move so that your center of mass is constantly projected onto the rope. I will be waiting for rich-in-content and serious replies. $\endgroup$
    – user64494
    Commented Feb 28 at 4:44
  • $\begingroup$ In the question even the replies of Mathematica were edited by an unknown to me person. For example, "General::munfl: 0.0000408571^71 is too small to represent as a normalized machine number; precision may be lost". In view of it I rolled back the text. $\endgroup$
    – user64494
    Commented Feb 28 at 4:51
  • $\begingroup$ @DanielLichtblau: NMaximize[{Sum[x^n/(1 + x^n)^n, {n, 51, Infinity}], x >= 0 && x <= 2}, x] returns the input and a warning "NMaximize::nnum: The function value -\!(*UnderoverscriptBox[([Sum]), (n = 51), ([Infinity])](*SuperscriptBox[(0.09146359257542153\), \(n\)]\ \*SuperscriptBox[\((1 + \*SuperscriptBox[\(0.09146359257542153), (n)])), (-n)])) is not a number at {x} = {0.0914636}.'. $\endgroup$
    – user64494
    Commented Feb 28 at 8:35

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.