12
$\begingroup$

Is there an accepted analytic continuation of $\sum_{n=1}^m \frac{1}{n}$? Even a continuation to positive reals would be of interested, though negative and complex arguments would also be interesting.

I don't have a specific application in mind, but I'd very much like to understand how / if such a continuation could be accomplished. I've Googled but haven't come up with anything meaningful - perhaps because it's not possible?

ADDENDUM

@Noble below suggests $\frac{\Gamma'(x)}{\Gamma(x)}$. But this produces the following mismatched plots:

enter image description here

Can anyone explain?

$\endgroup$
3
  • 3
    $\begingroup$ As is the case with the Gamma function, it would help to specify what properties you want the analytic continuation to retain. For instance, Euler tells us that $H_n=\int_0^1 \frac {1-x^n}{1-x}dx$. Replacing $n$ by a continuous parameter gives you an analytic continuation (at least for $n>0$). Is it useful? $\endgroup$
    – lulu
    Jan 1, 2019 at 14:04
  • $\begingroup$ @Richard Burke-Ward Here's the explanation: the correct comparison is: Plot[{EulerGamma + Gamma'[z + 1]/Gamma[z + 1], HarmonicNumber[z]}, {z, -1, 5}] $\endgroup$ Jan 2, 2019 at 9:11
  • $\begingroup$ Related: math.stackexchange.com/q/3159749/8530 $\endgroup$ Aug 23, 2023 at 17:39

2 Answers 2

18
$\begingroup$

Let's try it in an elementary manner

  1. We can use the defining recursion of the harmonic number valid for $n\in Z^{+}$

$$H_{n} = H_{n-1} + \frac{1}{n}, H_{1}=1\tag{1a}$$

also for any complex $z$

$$H_{z} = H_{z-1} + \frac{1}{z}, H_{1}=1\tag{1b}$$

For instance for $z=1$ we obtain $$H_{1} = H_{0} + \frac{1}{1}$$

from which we conclude that $H_{0}=0$.

If we try to find $H_{-1}$ we encounter the problem that from $H_0 = 0 = \lim_{z\to0}(H_{-1+z} + \frac{1}{z})$ we find that $H_{z} \simeq \frac{1}{z}$ for $z\simeq 0$. In other words, $H_{z}$ has a simple pole at $z=-1$.

Hence we cannot continue in this manner to go to further into the region of negative $z$, so let us move to the following general approach.

  1. Starting with this formula for the harmonic number which is valid for $n\in Z^{+}$

$$H_{n} = 1+\frac{1}{2}+ ... + \frac{1}{n}\\\\=\frac{1}{1}+ \frac{1}{2}+ ... + \frac{1}{n} +\frac{1}{1+n}+\frac{1}{n+2} + ... \\\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;-\frac{1}{1+n}- \frac{1}{n+2} + ...\\=\sum_{k=0}^\infty \left(\frac{1}{k}-\frac{1}{ (k+n)}\right)\tag{2}$$

The sum can be written as

$$H_{n}= \sum_{k=1}^\infty \frac{n}{k (k+n)}\tag{3}$$

and this can be extended immediately to complex values $z$ in place of $n$

$$H_{z}= \sum_{k=1}^\infty \frac{z}{k (k+z)}=\sum_{k=1}^\infty \left(\frac{1}{k} -\frac{1}{k+z}\right)\tag{4}$$

This sum is convergent (the proof is left to the reader) for any $z$ except for $z=-1, -2, ...$ where $H_{z}$ has simple poles with residue $-1$.

Hence $(4)$ gives the analytic continuation.

For instance close to $z=0$ we have as in 1. that

$$H_{z} \simeq z \sum_{k=1}^\infty \frac{1}{k^2} = z\;\zeta(2) =z\;\frac{\pi^2}{6}\to 0 $$

We can also derive an integral representation from the second form of $(4)$ writing

$$\frac{1}{k} -\frac{1}{k+z} =\int_0^1 (x^{k-1}-x^{z+k-1})\,dx $$

Performing the sum under the integral is just doing a geometric sum and gives

$$H_{z} = \int_0^1 \frac{1-x^{z}}{1-x}\,dx \tag{5}$$

  1. $H_{z}$ at negativ half integers ($z = -\frac{1}{2}, -\frac{3}{2}, ...$)

These can be calculated from $(1b)$ as soon as $H_{\frac{1}{2}}$ is known.

So let us calculate $H_\frac{1}{2}$.

Consider

$$H_{2n} = \frac{1}{1} + \frac{1}{2} + \frac{1}{3} + ... + \frac{1}{2n}$$

Splitting even and odd terms gives

$$H_{2n}= \frac{1}{1} + \frac{1}{3} + \frac{1}{5} + ... + \frac{1}{2n-1}\\+ \frac{1}{2} + \frac{1}{4} + ... + \frac{1}{2n}\\= \sum_{k=1}^n \frac{1}{2k-1} + \frac{1}{2} H_{n}\tag{6}$$

Now for the sum of the odd terms we write as in $(1)$

$$O_{n} = \sum_{k=1}^\infty \left(\frac{1}{2k-1} - \frac{1}{2(n+k)-1}\right)\tag{7}$$

This can be anlytically continued to any complex $n\to z$.

Replacing as before the summand by an integral and doing the summation under the integral gives

$$O_{z} = \int_0^1 \frac{1-x^{2z}}{1-x^2}\,dx\tag{8} $$

Substituting $x \to \sqrt{t}$ we find

$$O_{z} = \frac{1}{2}\int_0^1 \frac{1}{\sqrt{t}}\frac{1-t^{z}}{1-t}\,dt\\= \frac{1}{2}\int_0^1 \frac{1-t^{z-\frac{1}{2}}}{1-t}\,dt- \frac{1}{2}\int_0^1 \frac{1-t^{-\frac{1}{2}}}{1-t}\,dt\\ =\frac{1}{2}H_{z-\frac{1}{2}}+\log{2}\tag{9}$$

Hence $(6)$ can be written as

$$H_{2z} = \frac{1}{2} H_{z} +\frac{1}{2} H_{z-\frac{1}{2}}+\log{2}\tag{10} $$

Letting $z=1$ this gives

$$H_{2} = \frac{1}{2} H_{1} +\frac{1}{2} H_{\frac{1}{2}}+\log{2} $$

from which we deduce finally

$$H_{\frac{1}{2}} = 2(1-\log{2})\simeq 0.613706 \tag{11}$$

EDIT

Altenatively, the calculation of $H_{\frac{1}{2}}$ can be done using $(5)$ with the substitution $(x\to t^2)$:

$$H_{\frac{1}{2}} = \int_0^1 \frac{1-x^{\frac{1}{2}}}{1-x}\,dx = 2\int_0^1 t \frac{(1-t)}{{1-t^2}}\,dt = 2\int_0^1 t \frac{(1-t)}{(1+t)(1-t)}\,dt \\=2\int_0^1 \frac{t}{{1+t}}\,dt=2\int_0^1 \frac{1+t}{{1+t}}\,dt -2\int_0^1 \frac{1}{{1+t}}\,dt = 2 - 2 \log(2)$$

and we have recovered $(11)$.

As an exercise calculate $H_{\frac{1}{n}}$ for $n =3, 4,...$.

I found that Mathematica returns explicit expression up to $n=12$ except for the case $n=5$. I have not yet understood the reason for this exception. Maybe someone else can explain it?

EDIT 2

Mathematica does readily calculate explicit expression using the formula

$$H_{\frac{1}{n}}=\sum_{m=1}^{\infty} \left(\frac{1}{m} - \frac{1}{m+\frac{1}{n}}\right)$$

For the previouly missing case $n=5$ we find

$$H_{\frac{1}{5}} = \frac{1}{4}\left(20-2\pi \sqrt{1+\frac{2}{\sqrt{5}}}-2 \sqrt{5}\; \; \text{arccoth}(\sqrt{5})-5 \log(5) \right)$$

For $n=7$ the result is a typical combination of trigonometric and log functions

$$H_{\frac{1}{7}} = 7 - \frac{\pi}{2} \cot ( \frac{\pi }{7}) - \log(14) - 2 \cos(\frac{\pi}{7}) \log(\cos(\frac{\pi}{14})) - 2 \sin(\frac{\pi}{14}) \log(\cos(\frac{3 \pi}{14})) \\+ 2 \sin(\frac{3 \pi}{14})\log(\sin(\frac{\pi}{7})) $$

The general case for $n=1,2,3,...$ can be done by starting from the integral representation $(5)$ with $z=\frac{1}{n}$ using the substitution $x\to t^n, dx \to n t^{n-1} dt$ so that

$$H_{\frac{1}{n}}=n \int_{0}^{1} t^{n-1}\frac{1-t}{1-t^n}\;dt$$

Partial fraction decomposition leads to a simplification where the integrals can be solved elementarily. I shall leave the details to another occasion.

$\endgroup$
2
  • $\begingroup$ I think there's a typo in (2) and similar, the sums should start from 1, not 0. $\endgroup$
    – Bladewood
    Jan 2, 2019 at 1:32
  • $\begingroup$ @Bladewood You are right. Thanks. Have corrected it. $\endgroup$ Jan 2, 2019 at 8:46
10
$\begingroup$

I am not sure if this is what you meant, but Wolfram Alpha has an analytic formula for the $n^{\text{th}}$ harmonic number:

enter image description here

Here, the digamma function is $\psi_0(x)=\frac{\Gamma'(x)}{\Gamma(x)}$, which I believe is defined for all numbers in the complex plane except for negative real integers.

$\endgroup$
3
  • $\begingroup$ Hi @Noble. The two don't seem to match when plotted - see my addendum to the OP. Many thanks though! $\endgroup$ Jan 1, 2019 at 21:04
  • 4
    $\begingroup$ @RichardBurke-Ward I am sorry, but you plotted the wrong function. You plotted just $\phi_0(n)$ vs $H_n$. However, what I proposed was $\gamma+\phi_0(n+1)$ vs $H_n$. Here is the correct plot on Wolfram Alpha: wolframalpha.com/input/… As you can see, the two plots coincide. $\endgroup$ Jan 1, 2019 at 21:48
  • $\begingroup$ Understood. Appreciated, both of you. $\endgroup$ Jan 2, 2019 at 13:39

You must log in to answer this question.

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