# Category Archives: Mathematics

## “Second Moments in the Generalized Gauss Circle Problem” (with T. Hulse, C. Ieong Kuan, and A. Walker)

This is joint work with Thomas Hulse, Chan Ieong Kuan, and Alexander Walker. This is a natural successor to our previous work (see their announcements: one, two, three) concerning bounds and asymptotics for sums of coefficients of modular forms.

We now have a variety of results concerning the behavior of the partial sums

$$S_f(X) = \sum_{n \leq X} a(n)$$

where $f(z) = \sum_{n \geq 1} a(n) e(nz)$ is a GL(2) cuspform. The primary focus of our previous work was to understand the Dirichlet series

$$D(s, S_f \times S_f) = \sum_{n \geq 1} \frac{S_f(n)^2}{n^s}$$

completely, give its meromorphic continuation to the plane (this was the major topic of the first paper in the series), and to perform classical complex analysis on this object in order to describe the behavior of $S_f(n)$ and $S_f(n)^2$ (this was done in the first paper, and was the major topic of the second paper of the series). One motivation for studying this type of problem is that bounds for $S_f(n)$ are analogous to understanding the error term in lattice point discrepancy with circles.

That is, let $S_2(R)$ denote the number of lattice points in a circle of radius $\sqrt{R}$ centered at the origin. Then we expect that $S_2(R)$ is approximately the area of the circle, plus or minus some error term. We write this as

$$S_2(R) = \pi R + P_2(R),$$

where $P_2(R)$ is the error term. We refer to $P_2(R)$ as the “lattice point discrepancy” — it describes the discrepancy between the number of lattice points in the circle and the area of the circle. Determining the size of $P_2(R)$ is a very famous problem called the Gauss circle problem, and it has been studied for over 200 years. We believe that $P_2(R) = O(R^{1/4 + \epsilon})$, but that is not known to be true.

The Gauss circle problem can be cast in the language of modular forms. Let $\theta(z)$ denote the standard Jacobi theta series,

$$\theta(z) = \sum_{n \in \mathbb{Z}} e^{2\pi i n^2 z}.$$

Then

$$\theta^2(z) = 1 + \sum_{n \geq 1} r_2(n) e^{2\pi i n z},$$

where $r_2(n)$ denotes the number of representations of $n$ as a sum of $2$ (positive or negative) squares. The function $\theta^2(z)$ is a modular form of weight $1$ on $\Gamma_0(4)$, but it is not a cuspform. However, the sum

$$\sum_{n \leq R} r_2(n) = S_2(R),$$

and so the partial sums of the coefficients of $\theta^2(z)$ indicate the number of lattice points in the circle of radius $\sqrt R$. Thus $\theta^2(z)$ gives access to the Gauss circle problem.

More generally, one can consider the number of lattice points in a $k$-dimensional sphere of radius $\sqrt R$ centered at the origin, which should approximately be the volume of that sphere,

$$S_k(R) = \mathrm{Vol}(B(\sqrt R)) + P_k(R) = \sum_{n \leq R} r_k(n),$$

giving a $k$-dimensional lattice point discrepancy. For large dimension $k$, one should expect that the circle problem is sufficient to give good bounds and understanding of the size and error of $S_k(R)$. For $k \geq 5$, the true order of growth for $P_k(R)$ is known (up to constants).

Therefore it happens to be that the small (meaning 2 or 3) dimensional cases are both the most interesting, given our predilection for 2 and 3 dimensional geometry, and the most enigmatic. For a variety of reasons, the three dimensional case is very challenging to understand, and is perhaps even more enigmatic than the two dimensional case.

Strong evidence for the conjectured size of the lattice point discrepancy  comes in the form of mean square estimates. By looking at the square, one doesn’t need to worry about oscillation from positive to negative values. And by averaging over many radii, one hopes to smooth out some of the individual bumps. These mean square estimates take the form

\begin{align} \int_0^X P_2(t)^2 dt &= C X^{3/2} + O(X \log^2 X) \\ \int_0^X P_3(t)^2 dt &= C’ X^2 \log X + O(X^2 (\sqrt{ \log X})). \end{align}

These indicate that the average size of $P_2(R)$ is $R^{1/4}$. and that the average size of $P_3(R)$ is $R^{1/2}$. In the two dimensional case, notice that the error term in the mean square asymptotic has pretty significant separation. It has essentially a $\sqrt X$ power-savings over the main term. But in the three dimensional case, there is no power separation. Even with significant averaging, we are only just capable of distinguishing a main term at all.

It is also interesting, but for more complicated reasons, that the main term in the three dimensional case has a log term within it. This is unique to the three dimensional case. But that is a description for another time.

In a paper that we recently posted to the arxiv, we show that the Dirichlet series

$$\sum_{n \geq 1} \frac{S_k(n)^2}{n^s}$$

and

$$\sum_{n \geq 1} \frac{P_k(n)^2}{n^s}$$

for $k \geq 3$ have understandable meromorphic continuation to the plane. Of particular interest is the $k = 3$ case, of course. We then investigate smoothed and unsmoothed mean square results.  In particular, we prove a result stated  following.

Theorem

\begin{align} \int_0^\infty P_k(t)^2 e^{-t/X} &= C_3 X^2 \log X + C_4 X^{5/2} \\ &\quad + C_kX^{k-1} + O(X^{k-2} \end{align}

In this statement, the term with $C_3$ only appears in dimension $3$, and the term with $C_4$ only appears in dimension $4$. This should really thought of as saying that we understand the Laplace transform of the square of the lattice point discrepancy as well as can be desired.

We are also able to improve the sharp second mean in the dimension 3 case, showing in particular the following.

Theorem

There exists $\lambda > 0$ such that

$$\int_0^X P_3(t)^2 dt = C X^2 \log X + D X^2 + O(X^{2 – \lambda}).$$

We do not actually compute what we might take $\lambda$ to be, but we believe (informally) that $\lambda$ can be taken as $1/5$.

The major themes behind these new results are already present in the first paper in the series. The new ingredient involves handling the behavior on non-cuspforms at the cusps on the analytic side, and handling the apparent main terms (int his case, the volume of the ball) on the combinatorial side.

There is an additional difficulty that arises in the dimension 2 case which makes it distinct. But soon I will describe a different forthcoming work in that case.

## How fat would we have to get to balance carbon emissions?

Let’s consider a ridiculous solution to a real problem. We’re unearthing tons of carbon, burning it, and releasing it into the atmosphere.

Disclaimer: There are several greenhouse gasses, and lots of other things that we’re throwing wantonly into the environment. Considering them makes things incredibly complicated incredibly quickly, so I blithely ignore them in this note.

Such rapid changes have side effects, many of which lead to bad things. That’s why nearly 150 countries ratified the Paris Agreement on Climate Change.1 Even if we assume that all these countries will accomplish what they agreed to (which might be challenging for the US),2

most nations and advocacy groups are focusing on increasing efficiency and reducing emissions. These are good goals! But what about all the carbon that is already in the atmosphere?3

You know what else is a problem? Obesity! How are we to solve all of these problems?

Looking at this (very unscientific) graph,4 we see that the red isn’t keeping up! Maybe we aren’t using the valuable resource of our own bodies enough! Fat has carbon in it — often over 20% by weight. What if we took advantage of our propensity to become propense? How fat would we need to get to balance last year’s carbon emissions?

That’s what we investigate here.

Posted in Data, Mathematics, Story | | 1 Comment

## Slides from a Dissertation Defense

I just defended my dissertation. Thank you to Jeff, Jill, and Dinakar for being on my Defense Committee. In this talk, I discuss some of the ideas and follow-ups on my thesis. I’ll also take this moment to include the dedication in my thesis.

Here are the slides from my defense.

After the defense, I gave Jeff and Jill a poster of our family tree. I made this using data from Math Genealogy, which has so much data.

Posted in Mathematics | Tagged , , , | 2 Comments

## Smooth Sums to Sharp Sums 1

In this note, I describe a combination of two smoothed integral transforms that has been very useful in my collaborations with Alex Walker, Chan Ieong Kuan, and Tom Hulse. I suspect that this particular technique was once very well-known. But we were not familiar with it, and so I describe it here.

In application, this is somewhat more complicated. But to show the technique, I apply it to reprove some classic bounds on $\text{GL}(2)$ $L$-functions.

This note is also available as a pdf. This was first written as a LaTeX document, and then modified to fit into wordpress through latex2jax.

## Introduction

Consider a Dirichlet series
$$D(s) = \sum_{n \geq 1} \frac{a(n)}{n^s}. \notag$$
Suppose that this Dirichlet series converges absolutely for $\Re s > 1$, has meromorphic continuation to the complex plane, and satisfies a functional equation of shape
$$\Lambda(s) := G(s) D(s) = \epsilon \Lambda(1-s), \notag$$
where $\lvert \epsilon \rvert = 1$ and $G(s)$ is a product of Gamma factors.

Dirichlet series are often used as a tool to study number theoretic functions with multiplicative properties. By studying the analytic properties of the Dirichlet series, one hopes to extract information about the coefficients $a(n)$. Some of the most common interesting information within Dirichlet series comes from partial sums
$$S(n) = \sum_{m \leq n} a(m). \notag$$
For example, the Gauss Circle and Dirichlet Divisor problems can both be stated as problems concerning sums of coefficients of Dirichlet series.

One can try to understand the partial sum directly by understanding the integral transform
$$S(n) = \frac{1}{2\pi i} \int_{(2)} D(s) \frac{X^s}{s} ds, \notag$$
a Perron integral. However, it is often challenging to understand this integral, as delicate properties concerning the convergence of the integral often come into play.

Instead, one often tries to understand a smoothed sum of the form
$$\sum_{m \geq 1} a(m) v(m) \notag$$
where $v(m)$ is a smooth function that vanishes or decays extremely quickly for values of $m$ larger than $n$. A large class of smoothed sums can be obtained by starting with a very nicely behaved weight function $v(m)$ and take its Mellin transform
$$V(s) = \int_0^\infty v(x) x^s \frac{dx}{x}. \notag$$
Then Mellin inversion gives that
$$\sum_{m \geq 1} a(m) v(m/X) = \frac{1}{2\pi i} \int_{(2)} D(s) X^s V(s) ds, \notag$$
as long as $v$ and $V$ are nice enough functions.

In this note, we will use two smoothing integral transforms and corresponding smoothed sums. We will use one smooth function $v_1$ (which depends on another parameter $Y$) with the property that
$$\sum_{m \geq 1} a(m) v_1(m/X) \approx \sum_{\lvert m – X \rvert < X/Y} a(m). \notag$$
And we will use another smooth function $v_2$ (which also depends on $Y$) with the property that
$$\sum_{m \geq 1} a(m) v_2(m/X) = \sum_{m \leq X} a(m) + \sum_{X < m < X + X/Y} a(m) v_2(m/X). \notag$$
Further, as long as the coefficients $a(m)$ are nonnegative, it will be true that
$$\sum_{X < m < X + X/Y} a(m) v_2(m/X) \ll \sum_{\lvert m – X \rvert < X/Y} a(m), \notag$$
which is exactly what $\sum a(m) v_1(m/X)$ estimates. Therefore
$$\label{eq:overall_plan} \sum_{m \leq X} a(m) = \sum_{m \geq 1} a(m) v_2(m/X) + O\Big(\sum_{m \geq 1} a(m) v_1(m/X) \Big).$$

Hence sufficient understanding of $\sum a(m) v_1(m/X)$ and $\sum a(m) v_2(m/X)$ allows one to understand the sharp sum
$$\sum_{m \leq X} a(m). \notag$$

Posted in Expository, Math.NT, Mathematics | | 3 Comments

While idly thinking while heading back from the office, and then more later while thinking after dinner with my academic little brother Alex Walker and my future academic little sister-in-law Sara Schulz, we began to think about $2017$, the number.

## General Patterns

• 2017 is a prime number. 2017 is the 306th prime. The 2017th prime is 17539.
• As 2011 is also prime, we call 2017 a sexy prime.
• 2017 can be written as a sum of two squares,
$$2017 = 9^2 +44^2,$$
and this is the only way to write it as a sum of two squares.
• Similarly, 2017 appears as the hypotenuse of a primitive Pythagorean triangle,
$$2017^2 = 792^2 + 1855^2,$$
and this is the only such right triangle.
• 2017 is uniquely identified as the first odd prime that leaves a remainder of $2$ when divided by $5$, $13$, and $31$. That is,
$$2017 \equiv 2 \pmod {5, 13, 31}.$$
• In different bases,
\begin{align} (2017)_{10} &= (2681)_9 = (3741)_8 = (5611)_7 = (13201)_6 \notag \\ &= (31032)_5 = (133201)_4 = (2202201)_3 = (11111100001)_2 \notag \end{align}
The base $2$ and base $3$ expressions are sort of nice, including repetition.
Posted in Mathematics | Tagged , | 1 Comment

## Revealing zero in fully homomorphic encryption is a Bad Thing

When I was first learning number theory, cryptography seemed really fun and really practical. I thought elementary number theory was elegant, and that cryptography was an elegant application. As I continued to learn more about mathematics, and in particular modern mathematics, I began to realize that decades of instruction and improvement (and perhaps of more useful points of view) have simplified the presentation of elementary number theory, and that modern mathematics is less elegant in presentation.

Similarly, as I learned more about cryptography, I learned that though the basic ideas are very simple, their application is often very inelegant. For example, the basis of RSA follows immediately from Euler’s Theorem as learned while studying elementary number theory, or alternately from Lagrange’s Theorem as learned while studying group theory or abstract algebra. And further, these are very early topics in these two areas of study!

But a naive implementation of RSA is doomed (For that matter, many professional implementations have their flaws too). Every now and then, a very clever expert comes up with a new attack on popular cryptosystems, generating new guidelines and recommendations. Some guidelines make intuitive sense [e.g. don’t use too small of an exponent for either the public or secret keys in RSA], but many are more complicated or designed to prevent more sophisticated attacks [especially side-channel attacks].

In the summer of 2013, I participated in the ICERM IdeaLab working towards more efficient homomorphic encryption. We were playing with existing homomorphic encryption schemes and trying to come up with new methods. One guideline that we followed is that an attacker should not be able to recognize an encryption of zero. This seems like a reasonable guideline, but I didn’t really understand why, until I was chatting with others at the 2017 Joint Mathematics Meetings in Atlanta.

It turns out that revealing zero isn’t just against generally sound advice. Revealing zero is a capital B capital T Bad Thing.

## Basic Setup

For the rest of this note, I’ll try to identify some of this reasoning.

In a typical cryptosystem, the basic setup is as follows. Andrew has a message that he wants to send to Beatrice. So Andrew converts the message into a list of numbers $M$, and uses some sort of encryption function $E(\cdot)$ to encrypt $M$, forming a ciphertext $C$. We can represent this as $C = E(M)$. Andrew transmits $C$ to Beatrice. If an eavesdropper Eve happens to intercept $C$, it should be very hard for Eve to recover any information about the original message from $C$. But when Beatrice receives $C$, she uses a corresponding decryption function $D(\cdot)$ to decrypt $C$, $M = d(C)$.

Often, the encryption and decryption techniques are based on number theoretic or combinatorial primitives. Some of these have extra structure (or at least they do with basic implementation). For instance, the RSA cryptosystem involves a public exponent $e$, a public mod $N$, and a private exponent $d$. Andrew encrypts the message $M$ by computing $C = E(M) \equiv M^e \bmod N$. Beatrice decrypts the message by computing $M = C^d \equiv M^{ed} \bmod N$.

Notice that in the RSA system, given two messages $M_1, M_2$ and corresponding ciphertexts $C_1, C_2$, we have that

E(M_1 M_2) \equiv (M_1 M_2)^e \equiv M_1^e M_2^e \equiv E(M_1) E(M_2) \pmod N. \notag

The encryption function $E(\cdot)$ is a group homomorphism. This is an example of extra structure.

A fully homomorphic cryptosystem has an encryption function $E(\cdot)$ satisfying both $E(M_1 + M_2) = E(M_1) + E(M_2)$ and $E(M_1M_2) = E(M_1)E(M_2)$ (or more generally an analogous pair of operations). That is, $E(\cdot)$ is a ring homomorphism.

This extra structure allows for (a lot of) extra utility. A fully homomorphic $E(\cdot)$ would allow one to perform meaningful operations on encrypted data, even though you can’t read the data itself. For example, a clinic could store (encrypted) medical information on an external server. A doctor or nurse could pull out a cellphone or tablet with relatively little computing power or memory and securely query the medical data. Fully homomorphic encryption would allow one to securely outsource data infrastructure.

A different usage model suggests that we use a different mental model. So suppose Alice has sensitive data that she wants to store for use on EveCorp’s servers. Alice knows an encryption method $E(\cdot)$ and a decryption method $D(\cdot)$, while EveCorp only ever has mountains of ciphertexts, and cannot read the data [even though they have it].

## Why revealing zero is a Bad Thing

Let us now consider some basic cryptographic attacks. We should assume that EveCorp has access to a long list of plaintext messages $M_i$ and their corresponding ciphertexts $C_i$. Not everything, but perhaps from small leaks or other avenues. Among the messages $M_i$ it is very likely that there are two messages $M_1, M_2$ which are relatively prime. Then an application of the Euclidean Algorithm gives a linear combination of $M_1$ and $M_2$ such that

M_1 x + M_2 y = 1 \notag

for some integers $x,y$. Even though EveCorp doesn’t know the encryption method $E(\cdot)$, since we are assuming that they have access to the corresponding ciphertexts $C_1$ and $C_2$, EveCorp has access to an encryption of $1$ using the ring homomorphism properties:
\label{eq:encryption_of_one}
E(1) = E(M_1 x + M_2 y) = x E(M_1) + y E(M_2) = x C_1 + y C_2.

By multiplying $E(1)$ by $m$, EveCorp has access to a plaintext and encryption of $m$ for any message $m$.

Now suppose that EveCorp can always recognize an encryption of $0$. Then EveCorp can mount a variety of attacks exposing information about the data it holds.

For example, EveCorp can test whether a particular message $m$ is contained in the encrypted dataset. First, EveCorp generates a ciphertext $C_m$ for $m$ by multiplying $E(1)$ by $m$, as in \eqref{eq:encryption_of_one}. Then for each ciphertext $C$ in the dataset, EveCorp computes $C – C_m$. If $m$ is contained in the dataset, then $C – C_m$ will be an encryption of $0$ for the $C$ corresponding to $m$. EveCorp recognizes this, and now knows that $m$ is in the data. To be more specific, perhaps a list of encrypted names of medical patients appears in the data, and EveCorp wants to see if JohnDoe is in that list. If they can recognize encryptions of $0$, then EveCorp can access this information.

And thus it is unacceptable for external entities to be able to consistently recognize encryptions of $0$.

Up to now, I’ve been a bit loose by saying “an encryption of zero” or “an encryption of $m$”. The reason for this is that to protect against recognition of encryptions of $0$, some entropy is added to the encryption function $E(\cdot)$, making it multivalued. So if we have a message $M$ and we encrypt it once to get $E(M)$, and we encrypt $M$ later and get $E'(M)$, it is often not true that $E(M) = E'(M)$, even though they are both encryptions of the same message. But these systems are designed so that it is true that $C(E(M)) = C(E'(M)) = M$, so that the entropy doesn’t matter.

This is a separate matter, and something that I will probably return to later.

## Math 100 Fall 2016: Concluding Remarks

It is that time of year. Classes are over. Campus is emptying. Soon it will be mostly emptiness, snow, and grad students (who of course never leave).

I like to take some time to reflect on the course. How did it go? What went well and what didn’t work out? And now that all the numbers are in, we can examine course trends and data.

Since numbers are direct and graphs are pretty, let’s look at the numbers first.

## Math 100 grades at a glance

Let’s get an understanding of the distribution of grades in the course, all at once.

These are classic box plots. The center line of each box denotes the median. The left and right ends of the box indicate the 1st and 3rd quartiles. As a quick reminder, the 1st quartile is the point where 25% of students received that grade or lower. The 3rd quartile is the point where 75% of students received that grade or lower. So within each box lies 50% of the course.

Each box has two arms (or “whiskers”) extending out, indicating the other grades of students. Points that are plotted separately are statistical outliers, which means that they are $1.5 \cdot (Q_3 – Q_1)$ higher than $Q_3$ or lower than $Q_1$ (where $Q_1$ denotes the first quartile and $Q_3$ indicates the third quartile).

Within each blob, you’ll notice an embedded box-and-whisker graph. The white dots indicate the medians, and the thicker black parts indicate the central 50% of the grade. The width of the colored blobs roughly indicate how many students scored within that region. [As an aside, each blob actually has the same area, so the area is a meaningful data point].

Posted in Brown University, Math 100, Mathematics, Teaching | Tagged , , , , | 1 Comment

## Computing $\pi$

This note was originally written in the context of my fall Math 100 class at Brown University. It is also available as a pdf note.

While investigating Taylor series, we proved that
\label{eq:base}
\frac{\pi}{4} = 1 – \frac{1}{3} + \frac{1}{5} – \frac{1}{7} + \frac{1}{9} + \cdots

Let’s remind ourselves how. Begin with the geometric series

\frac{1}{1 + x^2} = 1 – x^2 + x^4 – x^6 + x^8 + \cdots = \sum_{n = 0}^\infty (-1)^n x^{2n}. \notag

(We showed that this has interval of convergence $\lvert x \rvert < 1$). Integrating this geometric series yields

\int_0^x \frac{1}{1 + t^2} dt = x – \frac{x^3}{3} + \frac{x^5}{5} – \frac{x^7}{7} + \cdots = \sum_{n = 0}^\infty (-1)^n \frac{x^{2n+1}}{2n+1}. \notag

Note that this has interval of convergence $-1 < x \leq 1$.

We also recognize this integral as

\int_0^x \frac{1}{1 + t^2} dt = \text{arctan}(x), \notag

one of the common integrals arising from trigonometric substitution. Putting these together, we find that

\text{arctan}(x) = x – \frac{x^3}{3} + \frac{x^5}{5} – \frac{x^7}{7} + \cdots = \sum_{n = 0}^\infty (-1)^n \frac{x^{2n+1}}{2n+1}. \notag

As $x = 1$ is within the interval of convergence, we can substitute $x = 1$ into the series to find the representation

\text{arctan}(1) = 1 – \frac{1}{3} + \frac{1}{5} – \frac{1}{7} + \cdots = \sum_{n = 0}^\infty (-1)^n \frac{1}{2n+1}. \notag

Since $\text{arctan}(1) = \frac{\pi}{4}$, this gives the representation for $\pi/4$ given in \eqref{eq:base}.

However, since $x=1$ was at the very edge of the interval of convergence, this series converges very, very slowly. For instance, using the first $50$ terms gives the approximation

\pi \approx 3.121594652591011. \notag

The expansion of $\pi$ is actually

\pi = 3.141592653589793238462\ldots \notag

So the first $50$ terms of \eqref{eq:base} gives two digits of accuracy. That’s not very good.

I think it is very natural to ask: can we do better? This series converges slowly — can we find one that converges more quickly?

## Series Convergence Tests with Prototypical Examples

This is a note written for my Fall 2016 Math 100 class at Brown University. We are currently learning about various tests for determining whether series converge or diverge. In this note, we collect these tests together in a single document. We give a brief description of each test, some indicators of when each test would be good to use, and give a prototypical example for each. Note that we do justify any of these tests here — we’ve discussed that extensively in class. [But if something is unclear, send me an email or head to my office hours]. This is here to remind us of the variety of the various tests of convergence.

A copy of just the statements of the tests, put together, can be found here. A pdf copy of this whole post can be found here.

In order, we discuss the following tests:

1. The $n$th term test, also called the basic divergence test
2. Recognizing an alternating series
3. Recognizing a geometric series
4. Recognizing a telescoping series
5. The Integral Test
6. P-series
7. Direct (or basic) comparison
8. Limit comparison
9. The ratio test
10. The root test

## The $n$th term test

### Statement

Suppose we are looking at $\sum_{n = 1}^\infty a_n$ and

\lim_{n \to \infty} a_n \neq 0. \notag

Then $\sum_{n = 1}^\infty a_n$ does not converge.

### When to use it

When applicable, the $n$th term test for divergence is usually the easiest and quickest way to confirm that a series diverges. When first considering a series, it’s a good idea to think about whether the terms go to zero or not. But remember that if the limit of the individual terms is zero, then it is necessary to think harder about whether the series converges or diverges.

### Example

Each of the series

\sum_{n = 1}^\infty \frac{n+1}{2n + 4}, \quad \sum_{n = 1}^\infty \cos n, \quad \sum_{n = 1}^\infty \sqrt{n} \notag

diverges since their limits are not $0$.

## Recognizing alternating series

### Statement

Suppose $\sum_{n = 1}^\infty (-1)^n a_n$ is a series where

1. $a_n \geq 0$,
2. $a_n$ is decreasing, and
3. $\lim_{n \to \infty} a_n = 0$.

Then $\sum_{n = 1}^\infty (-1)^n a_n$ converges.

Stated differently, if the terms are alternating sign, decreasing in absolute size, and converging to zero, then the series converges.

### When to use it

The key is in the name — if the series is alternating, then this is the goto idea of analysis. Note that if the terms of a series are alternating and decreasing, but the terms do not go to zero, then the series diverges by the $n$th term test.

### Example

Suppose we are looking at the series

\sum_{n = 1}^\infty \frac{(-1)^n}{\log(n+1)} = \frac{-1}{\log 2} + \frac{1}{\log 3} + \frac{-1}{\log 4} + \cdots \notag

The terms are alternating.
The sizes of the terms are $\frac{1}{\log (n+1)}$, and these are decreasing.
Finally,

\lim_{n \to \infty} \frac{1}{\log(n+1)} = 0. \notag

Thus the alternating series test applies and shows that this series converges.

## A Notebook Preparing for a Talk at Quebec-Maine

This is a notebook containing a representative sample of the code I used to  generate the results and pictures presented at the Quebec-Maine Number Theory Conference on 9 October 2016. It was written in a Jupyter Notebook using Sage 7.3, and later converted for presentation on this site.
There is a version of the notebook available on github. Alternately, a static html version without WordPress formatting is available here. Finally, this notebook is also available in pdf form.
The slides for my talk are available here.

# Testing for a Generalized Conjecture on Iterated Sums of Coefficients of Cusp Forms¶

Let $f$ be a weight $k$ cusp form with Fourier expansion

$$f(z) = \sum_{n \geq 1} a(n) e(nz).$$

Deligne has shown that $a(n) \ll n^{\frac{k-1}{2} + \epsilon}$. It is conjectured that

$$S_f^1(n) := \sum_{m \leq X} a(m) \ll X^{\frac{k-1}{2} + \frac{1}{4} + \epsilon}.$$

It is known that this holds on average, and we recently showed that this holds on average in short intervals.
(See HKLDW1, HKLDW2, and HKLDW3 for details and an overview of work in this area).
This is particularly notable, as the resulting exponent is only 1/4 higher than that of a single coefficient.
This indicates extreme cancellation, far more than what is implied merely by the signs of $a(n)$ being random.

It seems that we also have

$$\sum_{m \leq X} S_f^1(m) \ll X^{\frac{k-1}{2} + \frac{2}{4} + \epsilon}.$$

That is, the sum of sums seems to add in only an additional 1/4 exponent.
This is unexpected and a bit mysterious.

The purpose of this notebook is to explore this and higher conjectures.
Define the $j$th iterated sum as

$$S_f^j(X) := \sum_{m \leq X} S_f^{j-1} (m).$$

Then we numerically estimate bounds on the exponent $\delta(j)$ such that

$$S_f^j(X) \ll X^{\frac{k-1}{2} + \delta(j) + \epsilon}.$$

In [1]:
# This was written in SageMath 7.3 through a Jupyter Notebook.

# sage plays strangely with ipython. This re-allows inline plotting
from IPython.display import display, Image


We first need a list of coefficients of one (or more) cusp forms.
For initial investigation, we begin with a list of 50,000 coefficients of the weight $12$ cusp form on $\text{SL}(2, \mathbb{Z})$, $\Delta(z)$, i.e. Ramanujan’s delta function.
We will use the data associated to the 50,000 coefficients for pictoral investigation as well.

We will be performing some numerical investigation as well.
For this, we will use the first 2.5 million coefficients of $\Delta(z)$

In [2]:
# Gather 10 coefficients for simple checking
check_10 = delta_qexp(11).coefficients()
print check_10

fiftyk_coeffs = delta_qexp(50000).coefficients()
print fiftyk_coeffs[:10] # these match expected

twomil_coeffs = delta_qexp(2500000).coefficients()
print twomil_coeffs[:10] # these also match expected

[1, -24, 252, -1472, 4830, -6048, -16744, 84480, -113643, -115920]
[1, -24, 252, -1472, 4830, -6048, -16744, 84480, -113643, -115920]
[1, -24, 252, -1472, 4830, -6048, -16744, 84480, -113643, -115920]

In [3]:
# Function which iterates partial sums from a list of coefficients

def partial_sum(baselist):
ret_list = [baselist[0]]
for b in baselist[1:]:
ret_list.append(ret_list[-1] + b)
return ret_list

print check_10
print partial_sum(check_10) # Should be the partial sums

[1, -24, 252, -1472, 4830, -6048, -16744, 84480, -113643, -115920]
[1, -23, 229, -1243, 3587, -2461, -19205, 65275, -48368, -164288]

In [4]:
# Calculate the first 10 iterated partial sums
# We store them in a single list list, sums_list
# the zeroth elelemnt of the list is the array of initial coefficients
# the first element is the array of first partial sums, S_f(n)
# the second element is the array of second iterated partial sums, S_f^2(n)

fiftyk_sums_list = []
fiftyk_sums_list.append(fiftyk_coeffs) # zeroth index contains coefficients
for j in range(10):                    # jth index contains jth iterate
fiftyk_sums_list.append(partial_sum(fiftyk_sums_list[-1]))

print partial_sum(check_10)
print fiftyk_sums_list[1][:10]         # should match above

twomil_sums_list = []
twomil_sums_list.append(twomil_coeffs) # zeroth index contains coefficients
for j in range(10):                    # jth index contains jth iterate
twomil_sums_list.append(partial_sum(twomil_sums_list[-1]))

print twomil_sums_list[1][:10]         # should match above

[1, -23, 229, -1243, 3587, -2461, -19205, 65275, -48368, -164288]
[1, -23, 229, -1243, 3587, -2461, -19205, 65275, -48368, -164288]
[1, -23, 229, -1243, 3587, -2461, -19205, 65275, -48368, -164288]


As is easily visible, the sums alternate in sign very rapidly.
For instance, we believe tha the first partial sums should change sign about once every $X^{1/4}$ terms in the interval $[X, 2X]$.
In this exploration, we are interested in the sizes of the coefficients.
But in HKLDW3, we investigated some of the sign changes of the partial sums.

Now seems like a nice time to briefly look at the data we currently have.
What do the first 50 thousand coefficients look like?
So we normalize them, getting $A(n) = a(n)/n^{5.5}$ and plot these coefficients.

In [5]:
norm_list = []
for n,e in enumerate(fiftyk_coeffs, 1):
normalized_element = 1.0 * e / (1.0 * n**(5.5))
norm_list.append(normalized_element)
print norm_list[:10]

1

In [6]:
# Make a quick display
normed_coeffs_plot = scatter_plot(zip(range(1,60000), norm_list), markersize=.02)
normed_coeffs_plot.save("normed_coeffs_plot.png")
display(Image("normed_coeffs_plot.png"))


Since some figures will be featuring prominently in the talk I’m giving at Quebec-Maine, let us make high-quality figures now.

1. 00000000000000, -0.530330085889911, 0.598733612492945, -0.718750000000000, 0.691213333204735, -0.317526448138560, -0.376547696558964, 0.911504835123284, -0.641518061271148, -0.366571226366719
Posted in Math.NT, Mathematics, Open, Programming, sagemath | | 1 Comment