11

As might be known, the function $L(x) = x+\exp(x)\log(x)$ plays a prominent role in the Lagarias formulation of the Riemann hypothesis:

$$\sigma(n) \le H_n + \exp(H_n) \log(H_n)$$

My question is, what "abstract" properties does this function have (like for example the $\exp$-function has the property $\exp(x+y) = \exp(x)\exp(y)$) and do these properties characterize this function?

Thanks for your help! (Also I am not sure how to properly tag this question)

Related: A group theoretic interpretation of Lagarias inequality

Edit: Another related question: Dou you know of any other situation where the function $L(x)$ occurs? That would also be quite interesting!

Second Edit: I found a very exciting connection to "Logarithmic numbers" as defined by J. M. Gandhi:

http://oeis.org/A002741

The numbers $\frac{d^n}{dx^n} L(x)$ at $x=1$ are related to the Logarithmic numbers as defined by Gandhi.

There are two papers by Gandhi, on this topic:

http://oeis.org/A002741/a002741.pdf which is a little bit hard to read because it is scanned, and https://www.tandfonline.com/doi/abs/10.1080/00029890.1966.11970871 where there is made a connection to $\sigma(n)$ and the logarithmic numbers.

Third edit (18.05.2019): I think I found a very interesting property which seems to always hold:

If $A$ is a normal ($A^TA=AA^T$) and non-singular matrix and such that $\frac{1}{|A|}\cdot A$ is a doubly stochastic, positive matrix, then we have:

$$L(|A|) = |L(A)|$$ and $L(A)$ is a normal, non-singular matrix such that $\frac{1}{|L(A)|}\cdot L(A)$is doubly stochastic, positive matrix. where $|.|$ denotes the spectral norm.

To be more concrete I will tell how I construct the matrix for a given finite group $G$:

Let $\rho$ be the regular representation of $G$. $S \subset G$ a generating set, $|g| := |g|_S=$ word length with respect to $S$. Then I construct such a matrix, where we have some ordering $g_i$ of the group $G$:

$$ a_{i,j} = \frac{1}{1+|g_i g_j^{-1}|}$$

This is a group matrix as defined by Dedekind and Frobenius. Let $H_G:= \sum_{g \in G} \frac{1}{|g|+1}$ be the harmonic number associated to $S$ and $G$.

Here are my conjectures concerning this matrix some of which I can prove:

  1. $H_G = |A|$ [proved by Perron-Frobenius theorem]

  2. (If 1. is true, then by definition of $A$ we must have that $1/H_G A = 1/|A| A $ is a doubly stochastic matrix [thats clear by 1. and definition of $A_G$.]

  3. $A = \sum_{g \in G} \frac{1}{1+|g|} \rho(g)$ is the Birkhoff-Neunmann decomposition induced by the doubly stochastic matrix [proved by definition of $A_G$ and $\rho$]

  4. Using 2. I can prove that $A$ is a normal matrix

  5. $A$ is non-singular. [that remains mysterious]

My updated question is, if (any of) this can be proven (or if it is known, then any reference would also be nice)?

Thanks for your help!

  • 10
    One way to characterize it is $x L'' - (2 x - 1) L' + (x - 1) L =x^2 - 3 x + 1$, $L(1)=1$, $L''(1)=e$ –  Apr 30 '19 at 11:14
  • @MattF. I think you forgot the condition $L'(1) = e+1$. Thanks for your help. –  Apr 30 '19 at 11:28
  • WolframAlpha confirms the uniqueness of the characterization I proposed: https://www.wolframalpha.com/input/?i=x+L%27%27%5Bx%5D+-+(2x-1)L%27%5Bx%5D+%2B+(x-1)L%5Bx%5D+%3D%3D+x%5E2-3+x%2B1,+L%5B1%5D%3D%3D1,+L%27%27%5B1%5D%3D%3DE. The alternative suggestion has $L(1)=e$. –  Apr 30 '19 at 12:42
  • @MattF. You are right. The uniqueness of your characterization is given. –  Apr 30 '19 at 13:38
  • 1
    Can you clarify your third edit? What is the Euclidean norm of a matrix? Frobenius/Hilbert-Schmidt norm? Or operator/spectral norm? Or something else? I've tried a few guesses, but numerics quickly disprove $L(|A|) = |L(A)|$ for every guess that I've made for what the norm should be. – Nathaniel Johnston May 18 '19 at 15:42
  • @NathanialJohnston: Spectral norm. Can you give your examples? The matrices I am looking at seem to have this property. –  May 18 '19 at 15:52
  • Oh I see. You're right, that property holds then simply by usual functional calculus. $|A|$ is the largest eigenvalue of $A$, whereas $|L(A)|$ is the largest eigenvalue of $L(A)$. But the eigenvalues of $L(A)$ are just $L$ applied to the eigenvalues of $A$. Since $L$ is monotone, the largest eigenvalue of $L(A)$ is just $L(|A|)$. This works as long as $A$ has positive real eigenvalues (but if that's not the case then even defining $L(A)$ in the first place is going to be somewhat icky, so I hope it's OK). – Nathaniel Johnston May 18 '19 at 16:14
  • @NathanielJonston: Ok, thanks for your comment. Would you mind writing your comment in a bit more detail as an answer? (If it is not asked too much than maybe with reference?) –  May 18 '19 at 16:17
  • @NathanielJohnston: Some of the matrices I am considering do not have real eigenvalues, but still $L(|A|) = |L(A)|$. –  May 18 '19 at 16:43
  • @orgesleka -- It definitely can't be true if you allow arbitrary complex eigenvalues. Try the diagonal matrix $A$ with diagonal entries $1$ and $i$. Then $L(|A|) = 1$ but $|L(A)| \approx 2.27$. – Nathaniel Johnston May 18 '19 at 20:13
  • @NathanielJohnston: Ok, I will update the question, with further properties. –  May 18 '19 at 20:19
  • I think (from the last paragraph of the article due to Lagarias) that maybe it is interesting consider $l(x)=e^x\log x$. For $L(x)$ or $l(x)$, maybe can be interesting to state Möbius inversion, and related asymptotic formulas, section Generalizations of the Wikipedia Möbius inversion formula as starting point. I refer also (section 3) of Manuel Benito and Luis M. Navas and Juan Luis Varona, Möbius inversion from the point of view of arithmetical semigroup flows, Biblioteca de la Revista Matemática Iberoamericana, Proceedings of the Segundas Jornadas de Teoría de Números (Madrid, 2007). – user142929 Oct 12 '19 at 20:35
  • 1
    @user142929 thanks for your comment –  Oct 14 '19 at 05:47

0 Answers0