Möbius Summation in Number Theory

• Möbius summation is a method leveraging the Möbius function to derive inversion formulas and explicit summatory representations in number theory.
• It employs analytic techniques such as contour integration and Mellin transforms to connect arithmetic sums with the distribution of primes and zeta zeros.
• Recent developments extend these methods to algebraic identities and higher moment estimates, enhancing sieve techniques and L-function analysis.

Möbius summation refers to analytic and arithmetic methods leveraging the Möbius function, μ(n), in summatory processes, inversion formulas, and explicit computations of arithmetic sums. The Möbius summation techniques are foundational in analytic number theory, encoding core properties of arithmetic functions and L-functions, and underpinning results on the distribution of primes, zero distributions of zeta and L-functions, and explicit formulae tying the additive and multiplicative structures of integers.

1. Möbius Function and Key Summatory Functions

The Möbius function μ(n) is defined by μ(1) = 1; μ(n) = (−1)^ω(n) if n is squarefree with ω(n) distinct prime divisors; and μ(n) = 0 otherwise. The central object of Möbius summation is its summatory function, the Mertens function:

$M(x) = \sum_{n \le x} \mu(n)$

Related variants include:

• Weighted sum: $m(x) = \sum_{n \le x} \mu(n)/n$
• First centered moment: $m_1(x) = m(x) - M(x)/x$
• Logarithmic transform: $$\check m(x) = \sum_{n \le x} \mu(n)/n \cdot \log(x/n)$$

Truncated sums of the form $M(n, z) = \sum_{d\mid n,\, d \le z} \mu(d)$ and their moments also play a role in sieve methods and Dirichlet convolution logic (Bretèche et al., 2019, Daval, 2020).

2. Explicit and Asymptotic Summation Formulas

Classical explicit formulas for $M(x)$ relate it to the nontrivial zeros of the Riemann zeta function via Perron's formula and contour integration:

$$M(x) = \sum_{|\Im \rho| < T} \frac{x^\rho}{\rho \zeta'(\rho)} + O\left( x^{1/2} e^{C (\log\log x)^2} \right)$$

for suitable $T \sim x^{1/2}$ (Inoue, 2018).

Extensions cover arithmetic progressions and number fields, generalizing Möbius summation to sums like $$M(x;q,a) = \sum_{n \le x, n \equiv a \bmod q} \mu(n)$$ and $$M_K(x) = \sum_{N\mathfrak{a} \le x} \mu_K(\mathfrak{a})$$ for Dedekind zeta functions $\zeta_K(s)$. Explicit error bounds are derived using zero-density estimates for Dirichlet L-functions and Euler product factorizations (Inoue, 2018).

New convolutional identities express $M(x)$ via auxiliary sums: $$M(x) = G(x) + \sum_{p \le x} G\left( \left\lfloor x/p \right\rfloor \right)$$ where $G(x) = \sum_{n\le x} g(n)$ and $g(n)$ is the Dirichlet inverse of $h(n) = \omega(n) + 1$ (Schmidt, 2021). These allow exact representations with explicit finite sums and provide new methods to bound and analyze $M(x)$.

3. Möbius–Poisson and Operator-Theoretic Summation

The Möbius–Poisson summation formula generalizes Poisson summation by replacing characters with the Möbius function through the Dirichlet series $1/\zeta(s)$. For a suitable test function $F(x)$,

$$\sum_{n=1}^{\infty} \mu(n) F(n y) = \frac{1}{2\pi i} \int_{\Re(s) = c} \Gamma(-s) \frac{1}{\zeta(-s)} F^{*}(s)(0) y^{-s} ds$$

where $F^*(s)$ is the Mellin transform (Jean-Christophe, 2016, Yakubovich, 2014).

Residue calculus and contour shifting yield Euler–Möbius–Poisson asymptotic expansions, making the method an analytic dual to the Euler–Maclaurin formula. This contour approach systematically generates summatory and asymptotic formulas paired by extracting poles from zeta and gamma factors.

Operator-theoretic analysis links Möbius summation to Riesz-type operators, with connections to the Riemann Hypothesis: If the Mellin transform of the Möbius operator $O f (x) = \sum_{n=1}^{\infty} \mu(n) f(n x)$ has no zeros in the critical strip, then RH holds (Yakubovich, 2014).

4. Conditional and Unconditional Bounds for M(x)

The magnitude and oscillation of $M(x)$ are known to intimately reflect the zero-distribution of $\zeta(s)$. Under the Riemann Hypothesis, Soundararajan (0705.0723) proved: $$M(x) \ll x^{1/2} \exp\left( \frac{\log x}{(\log\log x)^{14}} \right)$$ improving over previous conditional results and tying the growth of $M(x)$ to finer statistical behavior of zeta zeros (zero clustering in short intervals).

This is achieved using a combination of Selberg’s entire function approximation, explicit formulae for $\zeta(s)$, and delicate contour deformation optimized via “V-typical” zero distribution. It is generally conjectured that $M(x) = O(x^{1/2 + \varepsilon})$ for any $\varepsilon > 0$, and even more precisely that the maximal order is roughly $x^{1/2} (\log\log\log x)^{5/4}$ (0705.0723). Unconditional upper bounds lag behind, typically of the form $$M(x) = O(x \exp(-c (\log x)^{3/5} / (\log_2 x)^{1/5}))$$.

For weighted sums and logarithmic variants, explicit “box-type” inequalities translate best-known $M(x)$ bounds into sharp estimates for $m(x)$ and related functions (Daval, 2020).

5. Algebraic and Structural Identities

A notable aspect of recent Möbius summation literature is the proliferation of algebraic identities reducing high-dimensional moment sums or large range summations to lower-dimensional, more tractable forms. For example, for $M(N^2) = \sum_{n \le N^2} \mu(n)$: $M(N^2) = 2 M(N) - \mathbf{m}^T A \mathbf{m}$ where $\mathbf{m} = (\mu(1), \ldots, \mu(N))^T$ and $A$ is an explicit matrix counting lattice points (Huxley et al., 2018). Spectral decomposition of $A$ reveals a dominant eigenvalue with size $\sim (\pi^2/6) N^2$ and precise structure, facilitating efficient computation and error analysis. Further, Dirichlet convolution approaches allow iterative inversion or transformation between Möbius-weighted and other arithmetic sums.

6. Truncated Sums, Higher Moments, and Sieve Applications

Truncated Möbius sums $M(n, z)$ and their moments, such as $S(x, z) = \sum_{n \le x} M(n, z)^2$, admit precise asymptotics: $$S(x, z) = Lx + O\left( \frac{x}{\mathcal{L}(3\xi)^c} \right)$$ with explicit Euler product for $L$ and slow-varying logarithmic factor $$\mathcal{L}(y) = \exp((\log y)^{3/5}/(\log_2 y)^{1/5})$$ (Bretèche et al., 2019). These results provide the quantitative backbone for sieve theory, minor-arc analysis, and Goldbach-type applications.

Similar integral transforms permit real-analytic conversions between the main summatory, weighted, and logarithmic versions of Möbius sums, establishing equivalences and improving upon classical results via the use of carefully constructed weight kernels and Mellin transforms (Daval, 2020).

7. Deep Connections and Research Directions

Möbius summation lies at the intersection of additive and multiplicative number theory, spectral analysis of arithmetic matrices, operator theory, and probabilistic models of arithmetic functions. Recently developed identities link the Mertens function to smoother auxiliary sums involving the Liouville function and prime-counting function, offering alternative probabilistic and analytic perspectives (Schmidt, 2021).

The methodology underpins key equivalences to the Riemann Hypothesis: e.g., analytic properties of the Möbius–Poisson summatory operator correspond precisely to zero-free regions for $\zeta(s)$ (Yakubovich, 2014, Jean-Christophe, 2016). The operator and spectral approaches complement classical probabilistic models, while convolutional identities offer new computational and theoretical tools.

Unresolved questions include unconditional progress toward square-root cancellation in $M(x)$, the extension of explicit formulas to non-Abelian fields (hindered by general zero-density problems), the analogues for higher moments or general $L$-functions, and refined understanding of the statistical fluctuations of Möbius summatory functions in short intervals, all of which remain active areas of research (0705.07231805.05015Bretèche et al., 2019).