Bell Series

The study of arithmetic functions, which map integers to complex numbers, is a cornerstone of number theory. However, the behavior of these functions can often appear chaotic and unpredictable. This raises a fundamental challenge: how can we uncover the hidden structure within these seemingly random sequences of values? The answer lies in a powerful "divide and conquer" strategy known as local analysis, which simplifies the problem by examining a function's behavior at the building blocks of integers—the powers of single primes. The primary tool for this analysis is the Bell series.

This article delves into the theory and application of Bell series as a transformative concept in number theory. It provides a comprehensive guide to understanding this elegant mathematical device. First, in "Principles and Mechanisms," we will introduce the formal definition of a Bell series and explore its magical property of turning Dirichlet convolution into simple multiplication. We will then see how these "local fingerprints" reassemble to form the global picture of a function through Dirichlet series and Euler products. Following this, the section on "Applications and Interdisciplinary Connections" will showcase the power of this tool in action. We will see how Bell series act as a Rosetta Stone for number-theoretic identities, serve as a computational laboratory for dissecting functions like Euler's totient function, and provide a remarkably simple method for the art of functional inversion.

The world of integers is governed by a remarkable and rigid structure: every number is either a prime or a unique product of primes. This, the ​​Fundamental Theorem of Arithmetic​​, is the bedrock upon which number theory is built. It suggests a powerful strategy for studying the often-chaotic behavior of arithmetic functions—functions that take an integer as input, like $f(n)$. Instead of trying to understand the function's behavior across all numbers at once, what if we could "divide and conquer"? What if we could understand the function by examining its behavior on the fundamental building blocks—the powers of a single prime, $p^k$? This is the essence of ​​local analysis​​, a method that allows us to isolate the complexity of number theory one prime at a time.

To enact this strategy, we need a tool, a mathematical device that can neatly package all the information about how a function $f$ acts on the powers of a single prime $p$. This tool is the ​​Bell series​​. For a given arithmetic function $f$ and a prime $p$, we define its Bell series as the formal power series:

$B_p(f;x) = \sum_{k=0}^{\infty} f(p^k) x^k$

Think of this as a "fingerprint" or a "spectrum" of the function $f$ at the prime $p$. The sequence of values $f(1), f(p), f(p^2), f(p^3), \dots$ contains everything there is to know about $f$'s behavior at $p$. The Bell series is simply a ​​generating function​​ for this sequence—a single, compact object that holds all this information.

Let's see what these fingerprints look like. For the simplest multiplicative function, the constant function $1(n)=1$ for all $n$, the values at prime powers are always $1, 1, 1, \dots$. Its Bell series is therefore the familiar geometric series:

$B_p(1;x) = \sum_{k=0}^{\infty} 1 \cdot x^k = \frac{1}{1-x}$

This is true for any prime $p$. The function is so simple its fingerprint is the same everywhere. The same elegance appears for the power function $\text{id}^{\alpha}(n) = n^{\alpha}$, whose values on prime powers are $1, p^\alpha, p^{2\alpha}, \dots$. Its Bell series is another geometric series:

$B_p(\text{id}^{\alpha};x) = \sum_{k=0}^{\infty} (p^{\alpha})^k x^k = \sum_{k=0}^{\infty} (p^{\alpha}x)^k = \frac{1}{1-p^{\alpha}x}$

These two functions are ​​completely multiplicative​​, meaning $f(mn) = f(m)f(n)$ for all integers $m$ and $n$. This property simplifies their behavior on prime powers to the elegant rule $f(p^k) = (f(p))^k$, which is precisely why their Bell series are simple geometric series.

But what about functions that aren't completely multiplicative? Consider the divisor function, $d(n)$, which counts the number of divisors of $n$. The divisors of $p^k$ are $1, p, \dots, p^k$, so there are $k+1$ of them. Thus, $d(p^k)=k+1$. The sequence of values is $1, 2, 3, \dots$. The Bell series is:

$B_p(d;x) = \sum_{k=0}^{\infty} (k+1) x^k = \frac{1}{(1-x)^2}$

This is astonishing! The seemingly random values of the divisor function, when viewed locally at a prime $p$, are encoded by a beautifully simple rational function. We are finding profound order in the chaos.

The true power of Bell series is revealed when we consider ​​Dirichlet convolution​​, the fundamental way arithmetic functions interact. The convolution of two functions, $f$ and $g$, is defined as:

$(f*g)(n) = \sum_{d|n} f(d) g(n/d)$

This operation is central to number theory, but its definition involves a sum over divisors that can be quite complicated to work with directly. Let's see what happens when we apply our "divide and conquer" strategy and look at the convolution on a prime power, $p^k$. The divisors of $p^k$ are just $p^0, p^1, \dots, p^k$. The formula becomes:

$(f*g)(p^k) = \sum_{j=0}^{k} f(p^j) g(p^{k-j})$

If you've ever multiplied two polynomials or power series, this should look incredibly familiar. This is the ​​Cauchy product​​ formula for the coefficients of a product of two series. This observation leads to the central, magical property of Bell series:

$B_p(f*g; x) = B_p(f; x) B_p(g; x)$

This is a revolutionary result. It tells us that the messy, complex operation of Dirichlet convolution in the world of numbers becomes simple, familiar multiplication in the world of Bell series. It's like finding a new coordinate system where a difficult problem becomes trivial.

Let's put this magic to the test. We know that the divisor function can be written as the convolution of the constant function $1(n)$ with itself: $d = 1 * 1$. Using our new rule, its Bell series must be:

$B_p(d; x) = B_p(1*1; x) = B_p(1; x) B_p(1; x) = \left(\frac{1}{1-x}\right) \left(\frac{1}{1-x}\right) = \frac{1}{(1-x)^2}$

This confirms our earlier calculation, but now we have a deeper structural understanding of where this result comes from. Similarly, the sum-of-divisors function $\sigma(n) = \sum_{d|n} d$ can be written as $\sigma = 1 * \text{id}$. Its Bell series is therefore:

$B_p(\sigma; x) = B_p(1; x) B_p(\text{id}; x) = \frac{1}{1-x} \cdot \frac{1}{1-px}$ This allows us to effortlessly find the Bell series for a vast family of important functions.

The true power of this method shines when proving identities. Consider the identity $\lambda * \mu^2 = \varepsilon$, where $\lambda$ is the Liouville function, $\mu$ is the Möbius function, and $\varepsilon$ is the identity for convolution ($\varepsilon(1)=1$, $\varepsilon(n)=0$ for $n>1$). Proving this directly is a chore. Using Bell series, it's a delight. One can calculate that $B_p(\lambda; x) = \frac{1}{1+x}$ and $B_p(\mu^2; x) = 1+x$. Their product is simply $1$, which is the Bell series for $\varepsilon$. Since this is true for every prime $p$, the identity must hold for all $n$.

We have seen how Bell series provide a "local" picture of a function at each prime. But how do we reassemble these local fingerprints to recover the "global" function? The key lies in the multiplicative nature of the functions we study. For a ​​multiplicative function​​ $f$ (where $f(mn)=f(m)f(n)$ for coprime $m,n$), its value at any integer $n = p_1^{k_1} \cdots p_r^{k_r}$ is simply the product of its values on the prime-power components:

$f(n) = f(p_1^{k_1}) \cdots f(p_r^{k_r})$

This property of "independence across primes" translates directly into the world of ​​Dirichlet series​​, $D_f(s) = \sum_{n=1}^\infty \frac{f(n)}{n^s}$. Because of multiplicativity, this sum can be factored into a product over all primes, an ​​Euler product​​:

$D_f(s) = \prod_{p} \left( \sum_{k=0}^{\infty} \frac{f(p^k)}{p^{ks}} \right)$

Look closely at the term inside the product. It is $\sum_{k=0}^{\infty} f(p^k) (p^{-s})^k$. This is nothing but the Bell series $B_p(f;x)$ evaluated at $x = p^{-s}$! This gives us the grand, unifying formula that connects the local and global pictures:

$D_f(s) = \prod_p B_p(f; p^{-s})$

This is the beautiful synthesis. The Dirichlet series, a global object containing information about $f(n)$ for all $n$, is the product of all its local fingerprints. The local algebraic structure of convolution being turned into multiplication of Bell series at each prime is the very reason that Dirichlet series of convolutions factor into products globally ($D_{f*g} = D_f D_g$). For example, by computing the simple local factor for the function $f(n)=\mu(n)^2$ to be $1+p^{-s}$, we can assemble the global Euler product $\prod_p (1+p^{-s})$, which elegantly simplifies to the analytic object $\frac{\zeta(s)}{\zeta(2s)}$.

This framework is not just beautiful; it is immensely powerful. We can use it to solve problems that would be otherwise intractable. One of the most striking applications is finding the ​​Dirichlet inverse​​ of a function. Given a function $f$, its inverse $g$ is the function such that $f*g = \varepsilon$. Finding $g$ from this convolution equation is a daunting task.

But in the world of Bell series, the equation becomes trivial:

$B_p(f;x) B_p(g;x) = B_p(\varepsilon;x) = 1$

This means we can solve for the Bell series of the unknown inverse function $g$ with simple algebra:

$B_p(g;x) = \frac{1}{B_p(f;x)}$

Let's perform this amazing feat for the divisor function $d(n)$. We want to find its inverse, let's call it $g$. We know $B_p(d;x) = \frac{1}{(1-x)^2}$. Therefore, the Bell series of its inverse must be:

$B_p(g;x) = \frac{1}{1/(1-x)^2} = (1-x)^2 = 1 - 2x + x^2$

By definition, $B_p(g;x) = g(1) + g(p)x + g(p^2)x^2 + \dots$. By simply comparing the coefficients of this polynomial with the series definition, we can read off the values of $g$ on prime powers: $g(1)=1$, $g(p)=-2$, $g(p^2)=1$, and $g(p^k)=0$ for all $k \ge 3$. This holds for any prime $p$!

Now we can use multiplicativity to find the value of $g$ for any number. For instance, what is $g(2100)$? First, we factor $2100 = 2^2 \cdot 3^1 \cdot 5^2 \cdot 7^1$. Since $g$ is multiplicative, we have:

$g(2100) = g(2^2) \cdot g(3^1) \cdot g(5^2) \cdot g(7^1)$

Using the values we just discovered:

$g(2100) = (1) \cdot (-2) \cdot (1) \cdot (-2) = 4$

We have solved a highly non-trivial deconvolution problem with little more than high-school algebra, a testament to the power of choosing the right perspective. By breaking a problem down to its prime components, the Bell series allows us to see the simple, elegant algebraic structure hidden within the complex world of numbers.

Now that we have acquainted ourselves with the principles of Bell series, let us embark on a journey to see what they can do. Like a new lens for a telescope, this tool doesn't just show us what we already knew; it reveals new structures, simplifies what was once complex, and connects seemingly distant islands of thought into a single, beautiful continent. The true power of a mathematical idea is measured by its ability to solve problems, create connections, and deepen our understanding of the world.

An old idea in science is that if you can't solve a problem, you should try translating it into a different language. In number theory, we have several "languages" for talking about arithmetic functions. One is the direct, hands-on language of ​​divisor sums​​. Another is the algebraic language of ​​Bell series​​. A third is the powerful analytic language of ​​Dirichlet series​​. The remarkable thing is that Bell series act as a kind of Rosetta Stone, allowing us to translate between these worlds with startling ease.

Consider an identity like the one relating the sum-of-divisors function $\sigma_k$ to its components, $\sigma_k = \operatorname{id}^k * 1$.

• In the language of divisor sums, we prove this by picking a prime power, $p^a$, and laboriously summing up all the terms in the convolution. It's direct, physical, and a bit like counting grains of sand. With some clever telescoping, we can show that $\sum_{d|p^a} J_k(d) = p^{ak}$, proving an analogous identity for Jordan's totient function. It works, but it feels like manual labor.

• In the language of Dirichlet series, the entire identity is captured by a single, global equation: $D_{\sigma_k}(s) = D_{\operatornameid^k}(s) D_1(s)$, which translates to the majestic relationship between Riemann zeta functions: $\sum_{n \geq 1} \frac{\sigma_k(n)}{n^s} = \zeta(s-k)\zeta(s)$. This is incredibly elegant but highly abstract, relying on deep theorems from complex analysis.

• The Bell series provides the perfect bridge. At each prime $p$, the convolution identity becomes a simple product of power series: $B_p(\sigma_k; x) = B_p(\operatorname{id}^k; x) B_p(1; x)$. We are no longer summing over all divisors, nor are we dealing with complex functions. We are simply multiplying two high-school level geometric series: $\frac{1}{1-p^k x} \cdot \frac{1}{1-x}$.

This reveals the hierarchy of abstraction: the Bell series method is an algebraic simplification of the direct sum, and it provides the "local" building blocks (the Euler factors) for the global Dirichlet series. It is the crucial middle ground that connects the concrete to the abstract.

With Bell series, we can put on our lab coats and "dissect" arithmetic functions to see what they're made of. The process often reveals their innermost secrets with an almost embarrassing simplicity.

Let's start with the divisor function, $d(n)$, which counts the number of divisors of $n$. It is defined by the convolution $d = 1 * 1$. In the world of Bell series, convolution is multiplication. The Bell series for the constant-one function $1$ is just the geometric series $\sum_{a=0}^\infty x^a = \frac{1}{1-x}$. So, the Bell series for $d$ must be $(\frac{1}{1-x})^2 = \frac{1}{(1-x)^2}$. If we expand this using the binomial theorem, we get $\sum_{k=0}^\infty (k+1)x^k$. By comparing the coefficients of $x^k$, we have, without any messy divisor-summing, discovered the famous formula for prime powers: $d(p^k) = k+1$. It feels like a magic trick!

This pattern holds for more complex functions. Consider Euler's totient function, $\phi(n)$. A fundamental identity states that $\sum_{d|n} \phi(d) = n$, which in the language of convolution is $\phi * 1 = \operatorname{id}$. By using Möbius inversion, this implies $\phi = \operatorname{id} * \mu$. What does this look like locally? The Bell series for the mysterious Möbius function $\mu$ is, astonishingly, just the polynomial $1-x$. The series for $\operatorname{id}$ is $\frac{1}{1-px}$. Their product, the Bell series for $\phi$, is thus $\frac{1-x}{1-px}$. Expanding this gives the coefficients $p^k - p^{k-1}$. The well-known formula for $\phi(p^k)$ simply falls out of this algebraic manipulation. This same technique can be extended to find the formula for the more general Jordan totient function $J_k = \operatorname{id}^k * \mu$, or to analyze the self-convolution of the Liouville function $\lambda * \lambda$, showing the universality of the method.

One of the most profound applications of Bell series is in finding the Dirichlet inverse of a function. The inverse $f^{-1}$ of a function $f$ is defined by the property $f * f^{-1} = \varepsilon$, where $\varepsilon$ is the identity element of convolution ($\varepsilon(1)=1$ and $\varepsilon(n)=0$ for $n>1$). Finding this inverse can be a daunting task.

Once again, we translate the problem. The Bell series for $\varepsilon$ is simply $1$. So the convolution identity becomes $B_p(f; x) B_p(f^{-1}; x) = 1$. This is a revelation! To find the Bell series of the inverse function, we simply compute the multiplicative inverse of the original Bell series in the ring of formal power series: $B_p(f^{-1}; x) = \frac{1}{B_p(f; x)}$.

What was a complicated global problem of untangling sums over divisors has become a local, algebraic problem of inverting a power series at each prime. Since the inverse of a multiplicative function is also multiplicative, we can find the value of $f^{-1}(n)$ by calculating its value at each prime power factor of $n$ and multiplying the results.

Let's see this in action. What is the Dirichlet inverse of the function $f(n) = n^k$? The Bell series for $f$ is $B_p(f; x) = \frac{1}{1-p^k x}$. Its inverse is simply $B_p(f^{-1}; x) = 1 - p^k x$. Reading off the coefficients, we find that $f^{-1}(1)=1$, $f^{-1}(p) = -p^k$, and $f^{-1}(p^a)=0$ for all $a \ge 2$. This pattern perfectly matches the function $\mu(n)n^k$. And so we have discovered another beautiful identity: the inverse of $\operatorname{id}^k$ is the function $n \mapsto \mu(n)n^k$.

Why are Bell series and their global cousins, Dirichlet series, so special? Why don't other types of generating functions exhibit this wonderful multiplicative behavior? The answer lies in the nature of the "basis" used to form the series.

A Dirichlet series is of the form $\sum f(n) n^{-s}$. The function $n \mapsto n^{-s}$ is completely multiplicative. This is the secret ingredient that allows the series to be factored into an Euler product over primes, with each factor in the product being precisely the Bell series evaluated at $x = p^{-s}$.

Now consider another type of generating function, the Lambert series, defined as $L_f(q) = \sum_{n \geq 1} f(n) \frac{q^n}{1 - q^n}$. A bit of manipulation shows this is equal to the ordinary power series $\sum_{m \geq 1} (f * 1)(m) q^m$. While the coefficients $(f * 1)(m)$ are nicely multiplicative (if $f$ is), the series itself has no simple Euler product. The reason is that the basis element $q^n$ is not multiplicative in $n$ (that is, $q^{mn} \neq q^m q^n$). This failure of the basis to respect multiplication prevents a global product factorization.

Even so, the Bell series formalism still provides clarity on the local level. The coefficients of the Lambert series at prime powers, $(f * 1)(p^v)$, can be computed directly from the Bell series product $B_p(f;x) B_p(1;x)$. This shows that even when a global multiplicative structure is absent, the "local" viewpoint of Bell series remains a powerful tool for understanding the function's behavior at each prime.

In the end, Bell series offer more than just a computational shortcut. They provide a profound insight into the structure of numbers. They teach us that the intricate, global tapestry of an arithmetic function is woven from simple, independent threads—one for each prime number. By studying these threads one at a time, we can understand the entire design, revealing the inherent beauty and unity that lie at the heart of number theory.