Abstract
We discuss two theorems in analytic number theory and combinatory analysis that have seen increased use in recent years. A corollary to a Tauberian theorem of Ingham allows one to quickly prove asymptotic formulas for arithmetic sequences, so long as the corresponding generating function exhibits exponential growth of a certain form near its radius of convergence. Two common methods for proving the required analytic behavior are modular transformations and Euler–Maclaurin summation. However, these results are sometimes stated without certain technical conditions that are necessary for the complex analytic techniques that appear in Ingham’s proof. We carefully examine the precise statements and proofs of these results, and find that in practice, the technical conditions are satisfied for those cases appearing in recent applications. We also generalize the classical approach of Euler–Maclaurin summation in order to prove asymptotic expansions for series with complex values, simple poles, or multi-dimensional summation indices.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and statement of results
In mathematics, one often encounters sequences \(\{b_n\}_{n\in \mathbb {N}_0}\) whose terms enumerate the objects in some family of interest. Although the problem of finding closed-form expressions for the \(b_n\) is often intractable, for some applications it is sufficient to determine the asymptotic behavior of \(b_n\). A powerful technique is to consider the generating function of the sequence as a complex analytic power series, as its asymptotic analytic behavior can provide information about the asymptotic behavior of the \(b_n\). In this article we revisit Ingham’s Tauberian theorem [13], which was devised to carry out the above idea for a class of sequences related to modular forms and the combinatorics of integer partitions.
Recall that a partition of a non-negative integer n is a weakly decreasing sequence of positive integers that sum to n, and that the partition function p(n) denotes the number of partitions of n. For example, \(p(5)=7\) and the relevant partitions are: (5), (4, 1), (3, 2), (3, 1, 1), (2, 2, 1), (2, 1, 1, 1), and (1, 1, 1, 1, 1). The function p(n) does not have a closed form, nor does it satisfy any finite order recurrence. However, its asymptotic behavior was proven by Hardy and Ramanujan [12], who showed that
In fact, they obtained a much stronger result by introducing what is now known as the Hardy–Ramanujan Circle Method, which uses modular transformations to obtain a divergent series whose truncations approximate p(n) with a very small error (a later refinement of Rademacher [17] gave a convergent series representation for p(n)).
Ingham [13] showed that (1.1) can also be derived from a certain Tauberian theorem (see Sect. 4 below). This approach has recently seen increased use in combinatorics and number theory, including applications in plane partitions [11], t-core partitions [18], overpartitions [8, 9], partitions arising from permutation groups [10], families of partitions with certain “gap” conditions [14], and bounds for the coefficients of modular functions [7]. In usage, Ingham’s theorem is often stated as follows: Suppose that \(B(q)=\sum _{n\ge 0}b_nq^n\) is a power series with weakly increasing non-negative coefficients and radius of convergence 1. If \(\lambda \), \(\beta \), and \(\gamma \) are real numbers with \(\gamma >0\) such that \(B(\mathrm{e}^{-t}) \sim \lambda t^\beta \mathrm{e}^{\frac{\gamma }{t}}\) as \(t\rightarrow 0^+\), then
However, this is not quite correct as written, as it is missing an important technical condition from Ingham’s work. In particular, the analytic behavior of \(B(\mathrm{e}^{-z})\) for \(z \rightarrow 0^+\) along the real axis is not sufficient in general to determine the asymptotic behavior of the coefficients \(b_n\), as one also needs to consider \(B(\mathrm{e}^{-z})\) for complex values of z (see Sect. 3.2 below for some counterexamples). The full statement of Ingham’s theorem from [13] is given in Theorem 4.1 below, and the following result includes all necessary conditions for \(B(\mathrm{e}^{-z})\). The general statement also includes an additional logarithmic term that has been needed in some recent applications (see for example [5]).
Theorem 1.1
Suppose that \(B(q)=\sum _{n\ge 0}b_nq^n\) is a power series with non-negative real coefficients and radius of convergence at least one. If \(\lambda \), \(\alpha \), \(\beta \), and \(\gamma \) are real numbers with \(\gamma >0\) such that
with \(z=x+iy\) (\(x,y\in \mathbb {R}, x>0\)) in each region of the form \(|y|\le \Delta x\) for \(\Delta >0\), then
Furthermore, if the \(b_n\) are weakly increasing, then
Remark
-
1.
The conclusion of Theorem 1.1 forces B(q) to have radius of convergence exactly one. We also note that the second condition in (1.2) does not follow from the first using a simple term-by-term estimate, as
$$\begin{aligned} \left| B\left( \mathrm{e}^{-z}\right) \right| \le \sum _{n \ge 0} b_n \mathrm{e}^{-n {\text {Re}}(z)} = B \left( \mathrm{e}^{-{\text {Re}}(z)}\right) , \end{aligned}$$but \(\mathrm{e}^{\frac{\gamma }{{\text {Re}}(z)}}\) is not \(O(\mathrm{e}^{\frac{\gamma }{|z|}})\) for complex \(z \rightarrow 0\). In fact, in Sect. 3.2 we see that the second condition is essential in general.
-
2.
If in each region \(|y|\le \Delta x\) we have
$$\begin{aligned} B\left( \mathrm{e}^{-z}\right) \sim \lambda {\text {Log}}\left( \frac{1}{z} \right) ^\alpha z^\beta \mathrm{e}^{\frac{\gamma }{z}}, \end{aligned}$$(1.5)then the second bound in (1.2) is automatically satisfied. As explained in Sect. 3.1, this case holds for any modular form with a pole at \(z = 0\). Here and throughout we follow the standard convention that \({\text {Log}}\) denotes the principal branch of the logarithm, so that for \(z \ne 0\), \({\text {Log}}(z) = \log |z| + {\text {Arg}}(z)\), with \({\text {Arg}}(z) \in (-\pi , \pi ].\)
The appeal of Ingham’s Tauberian theorem is that it yields asymptotics for sequences with very little effort, particularly in comparison to the Circle Method, which typically requires modular transformations and bounds along various arcs near the complex unit circle. Fortunately, although the bound along the restricted angle \(\Delta \) in Theorem 1.1 has not always been mentioned explicitly, the conclusion of the theorem statement still applies in all published applications that we are aware of. Indeed, one of the purposes of this article is to show that the extra condition is often guaranteed by whatever method is used to determine the asymptotic growth of \(B(\mathrm{e}^{-t})\). For example, as discussed in Sect. 4, if the growth is determined by applying transformations of a modular form, then the required bound in the restricted angle is always satisfied as well.
Another common method for determining the growth of \(B(\mathrm{e}^{-t})\) is to find an asymptotic expansion of \(B(\mathrm{e}^{-t})\) for t near zero. The classical Euler–Maclaurin summation formula is (see e.g. [16, Eq. (2.10.1)])
where \(M,N \in \mathbb {N}\), \(B_n(x)\) is the n-th Bernoulli polynomial, \(B_n\) the n-th Bernoulli number, and f is continuous on the interval [0, M] and 2N-times continuously differentiable inside the interval. In [21], Zagier gave an elegant account of how this formula implies asymptotic expansions of the form (\(N \in \mathbb {N}_0\))
where \(a\in \mathbb {R}^+\) and \(f:(0,\infty )\rightarrow \mathbb {C}\) is a \(C^\infty \) function such that f(x) and all of its derivatives are of “rapid decay” as \(x\rightarrow 0\). For example, this approach has been used to determine the asymptotic behavior of partial theta functions \(\sum _{n \ge 0} (-1)^n q^{an^2 + bn}\) as \(q \rightarrow 1^-\) in [14, 20].
In consideration of Theorem 1.1, the immediate question is to what extent do we also have expansions of this form when f is a function of a complex variable. To be precise, we say that a function f is of sufficient decay in a domain \(D\subset \mathbb {C}\) if there exists some \(\varepsilon > 0\) such that \(f(w) \ll w^{-1-\varepsilon }\) as \(|w| \rightarrow \infty \) in D. Our first result shows that Euler–Maclaurin summation gives an asymptotic expansion that converges uniformly on domains that preclude a tangential approach to 0.
Theorem 1.2
Suppose that \(0\le \theta < \frac{\pi }{2}\) and let \(D_\theta := \{ r\mathrm{e}^{i\alpha } : r\ge 0 \text{ and } |\alpha |\le \theta \}\). Let \(f:\mathbb {C}\rightarrow \mathbb {C}\) be holomorphic in a domain containing \(D_\theta \), so that in particular f is holomorphic at the origin, and assume that f and all of its derivatives are of sufficient decay. Then for \(a\in \mathbb {R}\) and \(N\in \mathbb {N}_0\),
uniformly, as \(w\rightarrow 0\) in \(D_\theta \).
Remark
We see in the proof of Theorem 1.2 that the decay assumption can be slightly relaxed, as the primary technical condition is that \(|f^{(n)}(w)|\) is integrable. However, in practice f(w) often has much stronger decay (for example, \(f(w) = g(w) \mathrm{e}^{-w}\) for a rational function g).
The next result extends Theorem 1.2 to the case that the function has a simple pole at zero. In order to state it we need the constants
which are defined for \(a\in \mathbb {R}, a \not \in -\mathbb {N}_0\). We note that \(C_a=-\gamma -\psi (a)\), where \(\psi (a):= \frac{\Gamma '(a)}{\Gamma (a)}\) is the digamma function [1, Eq. 6.3.16], and \(\gamma \) is the Euler–Mascheroni constant.
Theorem 1.3
Suppose that \(0\le \theta < \frac{\pi }{2}\), let \(f:\mathbb {C}\rightarrow \mathbb {C}\) be holomorphic in a domain containing \(D_\theta \), except for a simple pole at the origin, and assume that f and all of its derivatives are of sufficient decay in \(D_\theta \). If \(f(w) = \sum _{n\ge -1} b_{n}w^n\) near 0, then for \(a\in \mathbb {R}, a \not \in -\mathbb {N}_0\), and \(N\in \mathbb {N}_0\), then uniformly, as \(w\rightarrow 0\) in \(D_\theta \),
There are also applications where one needs asymptotic expansions of the form (1.6) for sums over multiple indices in \(\mathbb {N}\) (e.g. [6, Sect. 4]). This requires a multi-dimensional generalization of Theorem 1.2. While the two-dimensional version of this formula has appeared in a small number of recent articles, and the authors stated the general form in [6], we are not aware of any recorded proofs. Writing vectors in bold letters and their components with subscripts here and throughout the paper, we say that a multivariable function f in s variables is of sufficient decay in D if there exist \(\varepsilon _j>0\) such that \(f(\varvec{x})\ll (x_1+1)^{-1-\varepsilon _{1}}\cdots (x_s+1)^{-1-\varepsilon _{s}}\) uniformly as \(|x_1|+\cdots +|x_s|\rightarrow \infty \) in D.
Theorem 1.4
Suppose that \(0 \le \theta _j < \frac{\pi }{2}\) for \(1 \le j \le s\), and that \(f:\mathbb {C}^s\rightarrow \mathbb {C}\) is holomorphic in a domain containing \(D_{\varvec{\theta }}:=D_{\theta _1}\times \cdots \times D_{\theta _s}\). If f and all of its derivatives are of sufficient decay in \(D_{\varvec{\theta }}\), then for \(\varvec{a}\in \mathbb {R}^s\) and \(N\in \mathbb {N}_0\) we have
uniformly, as \(w\rightarrow 0\) in \(D_{\varvec{\theta }}\), where \(\mathcal N_N:=\{0,1,\dotsc ,N-1\}\).
We are writing Theorem 1.4 in a compact form, so to illustrate the unpacked statement we note that the two-dimensional case is
The remainder of this article is organized as follows. In the following section we recall known facts for Bernoulli polynomials. In Sect. 3, we give examples of a few applications related to Theorems 1.1 and 1.2. In particular, we demonstrate why the additional growth constraint in the right half-plane is necessary for Theorem 1.1 and why the limit in Theorem 1.2 must be taken non-tangentially to the imaginary axis. In Sect. 4, we state Ingham’s Tauberian theorem and use it to prove Theorem 1.1. In Sect. 5 we extend the classical use of Euler–Maclaurin summation to complex functions, proving Theorems 1.2, 1.3, and 1.4. We conclude in Sect. 6 with a brief discussion comparing Ingham’s Tauberian theorem and Wright’s Circle Method.
2 Preliminaries
We begin by recalling basic properties of the Bernoulli polynomials (see [1, Sect. 23.1]), which have the exponential generating function
For \(n \in \mathbb {N}_0\setminus \{1\}\), the Bernoulli numbers are defined by
whereas in order to avoid confusion for \(n=1\), we typically simply directly plug in
The polynomials satisfy many useful identities, including
We also need the Euler polynomials, which have the generating function
These are related to the Bernoulli polynomials by the identity
3 Examples
In this section, we consider some applications for Theorems 1.1 and 1.2. In these examples, we carefully examine the technical issues that arise in applying and using these theorems.
3.1 Partitions and weakly holomorphic modular forms
To illustrate the use of Theorem 1.1, we first revisit one of the motivating examples in [13]. Euler’s partition generating function is
where \(\eta (\tau ):=q^{\frac{1}{24}}\prod _{n\ge 1}(1-q^n)\) is Dedekind’s \(\eta \)-function. Here, and in the other examples, q and \(\tau \) are related by \(q:=\mathrm{e}^{2\pi i\tau }\). The Dedekind \(\eta \)-function satisfies the modular transformation [2, Theorem 3.1]
which implies that for \(z\in \mathbb {C}\) with \({\text {Re}}(z) > 0\),
We now easily see that Theorem 1.1 can be applied, since if \(z=x+iy\) \((x>0)\) with \(|y|\le \Delta x\), then
Thus, in these regions of restricted angle, we have (see (1.5))
And indeed, Theorem 1.1 does give the correct asymptotic formula, as (1.4) implies (1.1).
Finally, we also note that (3.2) does not hold in the whole right half-plane \({\text {Re}}(z) > 0\). For example, if z approaches 0 tangentially along the path \(z=x+ix^{\frac{1}{3}}\), then
and
as \(x\rightarrow 0^+\). Thus, \(|\mathrm{e}^{-\frac{1}{z}}|\rightarrow 1\) and we can no longer isolate the leading asymptotic term in (3.1). This can also be seen numerically. In Table 1, we give a numerical approximation of the size of \(P(\mathrm{e}^{-z}) \sqrt{\frac{2\pi }{z}}\mathrm{e}^{-\frac{\pi ^2}{6z}}-1\) along three different paths, with the first two being non-tangential and the third being tangential. As expected, the error tends to zero for the first two paths, but not the third.
The principle is similar for any other modular form, as the modular inversion map \(\tau \mapsto -\frac{1}{\tau }\) always gives an expansion in terms of \(\mathrm{e}^{-cz}\) for some \(c>0\), which rapidly tends to 0 so long as the angle of z is restricted.
3.2 A counterexample for the real-analytic version of Ingham’s theorem
In this section, we give a detailed presentation of an example that demonstrates the necessity of the second condition in (1.2).
3.2.1 General discussion
The importance of the example under consideration was highlighted by Ingham, who stated in note 2) on page 1088 of [13] that “In Theorem \(1'\) we may regard...(ii) (for every \(\Delta \)) as Tauberian conditions which convert the generally false inference...into a true proposition. An example...has been constructed by Avakumović and Karamata (353, e).”
More specifically, we work with the special case \(\gamma = \frac{3}{2}\) of Avakumović and Karamata’s example e) [3] (after making some minor modifications to obtain a power series instead of the continuous Laplace transform used in their original construction).
We define the coefficients
and the corresponding series \(F(q) := \sum _{n \ge 0} A(n) q^n\).
Proposition 3.1
As \(t \rightarrow 0^+\),
and
As a consequence, we see that Theorem 1.1 is false in general if one only considers the asymptotic behavior of the series along the real line. In particular, the A(n) are weakly increasing and \(F(\mathrm{e}^{-t})\) satisfies the first condition in (1.2). However, a short calculation shows that (1.4) does not hold; otherwise, the conclusion would be that \(A(n) \sim B(n)\), with
But (3.4) and (3.5) show that this expression does not accurately describe the behavior of A(n), either from above or below, as
3.2.2 Proof of Proposition 3.1
We first verify the asymptotic formulas for the coefficients. By construction, if s(n) is an increasing sequence, then any maxima of \(\frac{A(n)}{s(n)}\) occur at the values \(n = m^3\), thus
This proves (3.4).
Similarly, the minima of \(\frac{A(n)}{B(n)}\) occur at \(n = (m+1)^3-1\). To simplify the calculations, note that the expression \(n^{\frac{1}{12}}\mathrm{e}^{-2 \sqrt{n} + 3 n^{\frac{1}{6}} }\) is asymptotically equal to the same expression with \(n \mapsto n-1.\) We can therefore plug in \((m+1)^3\) instead of \((m+1)^3-1\). Thus
As \(m \rightarrow \infty \), we have that \((m+1)^{\frac{1}{4}} m^{-\frac{1}{4}} \rightarrow 1\). The exponential term has the overall exponent
which goes to 0 as \(m\rightarrow \infty \). This proves (3.5).
It is more involved to determine the asymptotic behavior of F. By definition,
We see below that the final exponential term is asymptotically negligible, and we next show that when this final term is removed, the resulting sum indeed gives the claimed asymptotic formula.
Proposition 3.2
As \(t \rightarrow 0^+\), we have
Proof
We roughly follow the arguments on pp. 354–355 of [3], with some additional details added. We begin by showing the final asymptotic equality, as it is useful throughout the rest of the proof. Using the substitution \(w = \sqrt{t} (u^{\frac{3}{2}} - \frac{1}{t})\), the integral becomes
For the sum, noting that \(\mathrm{e}^{2m^\frac{3}{2} - m^3 t} = \mathrm{e}^{\frac{1}{t}} \mathrm{e}^{-t(m^\frac{3}{2} - \frac{1}{t})^2}\), we approximate \(\sum _{m \ge 1} g(m^{\frac{3}{2}})\), where
We prove the integral approximation by showing that the summands g(m) are unimodal, with a peak near \(m \sim t^{-\frac{2}{3}}\). The growth rate of these terms is determined by the derivative of g(x), which is
The terms in front are always positive for \(x > 0\), so the sign of \(g'(x)\) is determined by the quadratic expression in the parentheses. The roots of this expression are
the minimum of g occurs at \(x_1\), and the maximum at \(x_2\).
However, the behavior near \(x_1\) does not have any effect on F(t), since, as \(t\rightarrow 0\) \(x_1 \sim \frac{1}{12}\). Specifically, this means that for sufficiently small t, the minimum of g(x) occurs at some \(x < 1\), and thus the sum beginning at \(m = 1\) is monotonically increasing until \(m_2 := \lfloor x_2^{\frac{2}{3}} \rfloor \), and monotonically decreasing beginning from \(m_2 + 1\). Moreover, we need the observation that \( x_2\sim \frac{1}{t}. \)
The standard integral comparison criterion for monotonic functions now implies that
From this, we obtain that
Thus
Using that \(g(x_2)=o(1)\) and the integral evaluation \(\int _1^\infty g(x^\frac{3}{2}) \mathrm{d}x \sim \frac{2 \sqrt{\pi }}{3}\), the bound in (3.7) shows that the sum and integral are asymptotically equal. \(\square \)
We now prove that the final sum in (3.6) is asymptotically smaller than the remaining terms.
Lemma 3.3
([3], p. 350) As \(t \rightarrow 0^+\), we have
Proof
We show that the sum in the numerator is termwise smaller than the denominator for \(m > t^{-\frac{1}{2} - \varepsilon }\) (for some \(\varepsilon > 0\)), and the sum over this initial segment of ms is itself asymptotically negligible. More precisely, since \(g(m^\frac{3}{2})\) is increasing in this range we obtain for \(0<\varepsilon <\frac{1}{6}\),
This also gives
since Proposition 3.2 shows that the left-hand side is asymptotically \(\frac{2 \sqrt{\pi }}{3}\).
Continuing, since each term in the numerator of the lemma statement is smaller than the corresponding denominator term, (3.8) also implies that
The final sum can then be compared termwise to the denominator sum, that is,
Using (3.8) and Proposition 3.2 gives that
\(\square \)
Finally, the proof of Proposition 3.1 is completed by combining Proposition 3.2 and Lemma 3.3. In particular, by plugging these in to (3.6), we find that the main asymptotic term is
3.3 Eisenstein series
Here we give an example to demonstrate that the expansion in Theorem 1.2 may fail when w is allowed to approach 0 along tangential paths in the right half-plane. For this we use the Eisenstein series of weight four for the full modular group. However, examples of this type generally arise from any modular form to which Theorem 1.2 can be applied. Set
From the modular transformation,
for \({\text {Im}}(\tau )>0\), we find that, for \({\text {Re}}(w)>0\),
Thus, when \(w\rightarrow 0\) on paths non-tangential to the imaginary axis, we have for each \(N\in \mathbb {N}_0\) that
As explained in Example 3 of [21], one can also deduce (3.10) directly from Theorem 1.2 by taking \(f(w):=\frac{w^3 \mathrm{e}^{-w}}{1-\mathrm{e}^{-w}}\) and \(a=1\), writing
However, along paths tangential to the imaginary axis, (3.10) may fail. For example, along the path \(w=x+ix^{\frac{1}{3}}\), (3.3) shows that every point along the unit circle is a limit point of \(\mathrm{e}^{-\frac{4\pi ^2}{w}}\). In particular, since \(g_3(q)\rightarrow \infty \) as \(q\rightarrow 1^-\), we see that \(\limsup _{w\rightarrow 0} | g_3(\mathrm{e}^{-\frac{4\pi ^2}{w}}) | = \infty \) on the path \(w=x+ix^\frac{1}{3}\). Thus, (3.9) implies that (3.10) cannot hold along this path. This gives an example where Theorem 1.2 fails without the additional assumption that \(w\in D_\theta \). Again this is visible from numerical data. In Table 2, we give a numerical approximation of the size of \(g_3(\mathrm{e}^{-w}) - \frac{\pi ^{4}}{15w^4} + \frac{1}{240}\) along three different paths.
3.4 Asymptotic expansions valid along any path
In contrast to the previous example, there are also functions where the asymptotic expansion of Theorem 1.2 is valid on general paths. For example, taking \(f(w):=\mathrm{e}^{-w}\) and \(a=0\) in Theorem 1.2 gives
as \(w\rightarrow 0\) in any \(D_\theta \). The left-hand side of (3.11) can be summed as a geometric series when \({\text {Re}}(w)>0\),
The right hand-side of (3.11) can be interpreted in terms of a truncation of the generating function for Bernoulli numbers, specifically
From this we see that (3.11) is valid with \(w\rightarrow 0\) along any path, as \(\frac{w}{1-\mathrm{e}^{-w}}\) is analytic in \(|w| < 2\pi \)
That the asymptotic expansion of Theorem 1.2 is valid for general paths to 0 for some functions and not others should come as no surprise. The series \(\sum _{m\ge 0}f(w(m+a))\) defines a holomorphic function in the right half-plane and the series diverges at \(w=0\). The analytic behavior of such a function can be anything from a simple pole at \(w=0\) to the imaginary axis being a natural boundary in the sense that the singularities are dense along the axis.
4 Ingham’s Tauberian theorem and the proof of Theorem 1.1
In this section we discuss Ingham’s Tauberian theorem and use it to prove Theorem 1.1. We start by recalling the following theorem, which is due to Ingham [13, Theorem 1].
Theorem 4.1
Let D be a connected open subset of \(\mathbb {C}\) that contains (0, h] (for some \(h\in \mathbb {R}^+\)). For \(t\in (0,h]\), we let \(\delta (t)\) denote the distance from t to the complement of D. Suppose that \(\varphi \) and \(\chi \) are functions on D that satisfy the following conditions:
- (a):
-
\(\varphi \) and \(\chi \) are holomorphic on D, and are positive on (0, h];
- (b):
-
\(-t\varphi ^\prime (t) \rightarrow \infty \) as \(t \rightarrow 0^+\);
- (c):
-
\(-\dfrac{\delta (t) \varphi ^\prime (t)}{t\sqrt{\varphi ^{\prime \prime }(t)}}\rightarrow \infty \) as \(t \rightarrow 0^+\), and
- (d):
-
\(\varphi ^{\prime \prime }(t+z) = O\left( \varphi ^{\prime \prime }(t) \right) \) and \(\chi (t+z) = O\left( \chi (t) \right) \) uniformly in z for \(|z|<\delta (t)\) as \(t \rightarrow 0^+.\)
Let \(A:[0,\infty )\rightarrow \mathbb {R}\) be a weakly increasing function with \(A(0)=0\). Set
and assume that f(z) exists for \({\text {Re}}(z)>0\). Suppose that the following conditions hold:
- (i):
-
\(f(z)\sim \chi (z)\mathrm{e}^{\varphi (z)}\) as \(z\rightarrow 0\) with z in D;
- (ii):
-
for each fixed \(\Delta >0\), \(f(z)=O(\chi (|z|)\mathrm{e}^{\varphi (|z|)})\) as \(z\rightarrow 0\) where \(|{\text {Im}}(z)|\le \Delta {\text {Re}}(z)\).
Then
where \(\psi \) is the inverse function of \(-\varphi ^\prime \).
Ingham also discussed a special case that eliminates the need for calculating an exact asymptotic formula for f throughout the complex domain D (although it is still necessary to bound the asymptotic order of f as in condition (ii)). The following result is [13, Theorem \(1'\)].
Theorem 4.2
Suppose that conditions (a), (b), (c), and (d) of Theorem 4.1 are satisfied, and additionally, \(-t^k\varphi ^\prime (t)\) decreases to 0 for some fixed \(k\in \mathbb {R}\). Then condition (i) may be replaced by
- (i\('\)):
-
\(f(t)\sim \chi (t)\mathrm{e}^{\varphi (t)}\) as \(t\rightarrow 0^+\).
We are now ready to prove Theorem 1.1.
Proof of Theorem 1.1
We first show that (1.4) follows from (1.3) applied to the series \(C(q):=(1-q)B(q)\). The monotonicity of \(b_n\) implies that C(q) has non-negative coefficients. Furthermore, as \(z \rightarrow 0\) we have
The theorem is trivially true for \(\lambda =0\), so we assume \(\lambda >0\). To prove (1.3), we apply Theorems 4.1 and 4.2 with
We let D consist of those points z which satisfy \(|\mathrm{Arg}(z)|< \frac{\pi }{4}\). In particular, in this region \(\delta (t)=\frac{t}{\sqrt{2}};\) this follows from applying the Law of Sines to calculate the distance from t to the ray along \(\mathrm{Arg}(z) = \frac{\pi }{4}.\) To verify that the conditions on \(\varphi \) and \(\chi \) are satisfied, we differentiate
It is then obvious that conditions (a), (b), and (c) hold, as well as the extra growth condition from Theorem 4.2. If \(|z| < \frac{t}{\sqrt{2}}\) then \((1-\frac{1}{\sqrt{2}})t< |z+t| < (1+\frac{1}{\sqrt{2}})t\), and and thus as \(t\rightarrow 0^+\) we have \(\varphi ^{\prime \prime }(z+t)=O(\varphi ^{\prime \prime }(t))\) uniformly in z. Furthermore,
and thus condition (d) of Theorem 4.1 holds.
By the definition of the Riemann–Stieltjes integral, we have
By the assumptions on B in Theorem 1.1, f(z) exists for \({\text {Re}}(z)>0\) and conditions (ii) of Theorem 4.1 and (i’) of Theorem 4.2 are satisfied. Thus all the hypotheses of Theorem 4.1 are satisfied. We note that \(\psi (x)=\sqrt{\frac{\gamma }{x}}\), and thus we have
A short calculation using the expression on the right-hand side shows that \(A(N+1) \sim A(N)\), which implies (1.3) on plugging in \(x = N+1\) to the left-hand side. \(\square \)
5 The Euler–Maclaurin summation formula and the proofs of Theorems 1.2, 1.3, and 1.4
5.1 The one-dimensional case
In this sub-section, we prove Theorems 1.2 and 1.3. We make use of Taylor’s theorem in the following form: Suppose that f is holomorphic in a neighborhood containing \(C_R(0)\), the circle of radius R centered at the origin. If \(|z|<R\), then
In particular, for z sufficiently small,
We are now ready to prove Theorem 1.2.
Proof of Theorem 1.2
The assumption that f has sufficient decay ensures that the sum converges, and also implies that \(wf(w) \rightarrow 0\) uniformly as \(|w| \rightarrow \infty \) in \(D_\theta \). Finally, if \(n \in \mathbb {N}_0\) is fixed and \(|\alpha |\le \theta \), then we have
where throughout the proof, all integrals are taken along straight lines. The claim in (5.3) follows by splitting the integral as
where \(C_1(n)\) is a constant such that \(|w|\ge C_1(n)\) implies that \(|f^{(n)}(w)|\le C_2(n)|w|^{-1-\varepsilon _n}\) for some \(C_2(n)\). The second piece is then clearly uniformly bounded, and the first piece is as well, due to the fact that the region \(|{\text {Arg}}(w)|\le \theta \) and \(|w|\le C_1(n)\) is compact (and \(f^{(n)}(w)\) is continuous).
We now present the fundamental identities that underlie Euler–Maclaurin summation, which follow from integration by parts and properties of Bernoulli polynomials. Throughout, supposing that \(\mu \in \mathbb {R}\) is fixed, we take w sufficiently small so that f is holomorphic in a region containing the line segment \([w(\mu -1),0]\). We use (2.2), (2.3), and the fact that \(B_0(x)=1\) to obtain
Next, for \(n \ge 1\), we use (2.1) and (2.3) to conclude that
Using induction on \(N\in \mathbb {N}\), one may then show that
We take w sufficiently small, so that f is holomorphic in a region containing the line segment [wa, 0]. Summing (5.4) with \(\mu = m+a\), for \(m\in \{1,2,\dotsc ,M\}\), gives that
where the N-th periodic Bernoulli polynomial is defined by \(\widetilde{B}_N(x) := B_N(x - \lfloor x\rfloor )\). Substituting \(z=wx\) in the integrals, we obtain
which in the limit \(M\rightarrow \infty \) becomes
We now claim that
In particular, since f has no poles in \(D_\theta \), the Residue theorem implies that for \(r > 0\) we have (writing \(\alpha :={\text {Arg}}(w)\))
The second integral vanishes as \(r \rightarrow \infty \), since
This yields (5.6).
Moreover, for w sufficiently small, we have that
Thus (5.5) becomes
In order to obtain the desired expression, we plug (5.1) into the second sum (5.7), finding that
Here, R is chosen such that \(C_R(0)\) is contained in the domain in which f is holomorphic. The first sum above is then further simplified using (2.4), as this implies that
Thus (5.7) becomes
We now claim that all terms except the first and the second on the right-hand side are \(O(w^N)\). For the third term, we use (5.2) to find that
For the fourth term, we only need to show that the integral is uniformly bounded as \(w \rightarrow 0\) in \(D_\theta \). Since f is holomorphic on \(C_R(0)\), \(f^{(n)}(z)\) is uniformly bounded, and \(|\frac{1}{z^{N-n}}| = \frac{1}{R^{N-n}}\). Furthermore, \(\frac{1}{z-wa}\) is uniformly bounded on \(|aw| < \frac{R}{2}\). This yields the claim for the fourth term.
Finally, the fact that \(\widetilde{B}_N(x)\) is periodic implies that \(\widetilde{B}_N(\frac{z}{w}-a)\) is bounded on the ray from the origin through w. We therefore have the following bound on the fifth term,
where the final bound follows from (5.3). This completes the proof. \(\square \)
We record one particularly useful corollary to Theorem 1.2 that allows for alternating signs.
Corollary 5.1
Suppose that \(0\le \theta < \frac{\pi }{2}\). Let \(f:\mathbb {C}\rightarrow \mathbb {C}\) be holomorphic in a domain containing \(D_\theta \), and assume that f and all of its derivatives are of sufficient decay as \(|w|\rightarrow \infty \) in \(D_\theta \). Then for \(a \in \mathbb {R}\) and \(N\in \mathbb {N}_0\),
uniformly, as \(w\rightarrow 0\) in \(D_\theta \). Recall that the Euler polynomials are defined in (2.5).
Proof
We write
and thus we may apply Theorem 1.2 with \(w\mapsto 2w\) and \(a \mapsto \frac{a}{2} + \frac{r}{2}\). This yields
where in the second equality we use (2.6).
Remark
The expansion in Corollary 5.1 can alternatively be proven using the Euler–Boole summation formula (see [4, Eq. (5)]), namely
where \(0\le a<1\) and \(\widetilde{E}_{n}(x)\) are the periodic Euler functions defined through the Euler polynomials by \(\widetilde{E}_n(x):=E_n(x)\) for \(0 \le x<1\), and \(\widetilde{E}_n(x+1):=-\widetilde{E}_n(x)\).
More generally, one sees that the method used in the proof of Corollary 5.1 applies to sums of the form \(\sum _m \chi (m)f(wm)\), where \(\chi \) is periodic. Of specific interest are the cases when the average of \(\chi \) is zero, such as with non-principal Dirichlet characters, as the integral terms of Theorem 1.2 then cancel, leaving an asymptotic expansion that converges at \(w=0\).
In the case that there is a simple pole at zero, the main new ingredient in the proof is the use of series representations for the complex logarithm digamma function in order to identify the asymptotic contribution of the pole, as we use Theorem 1.2 to obtain the remainder of the asymptotic expansion.
Proof of Theorem 1.3
Set
Using the series expansion of the exponential function, we obtain that
In particular, g has a simple pole at the origin with residue \(b_{-1}\), and thus h is holomorphic at the origin. By Theorem 1.2, we have that
The claim follows once we show that for \({\text {Re}}(w)>0\),
Since for \({\text {Re}}(w)>0\), we have
we may instead prove the equivalent identity,
To see (5.10), we first note that
Thus we have
for some constant C, and we see that the left hand-side provides an analytic continuation of the right hand-side in a neighborhood of \(w=0\). However, the left hand-side is clearly zero when \(w=0\). To evaluate the limit of the right hand-side, as \(w\rightarrow 0\) with \({\text {Re}}(w)>0\), we first note that
where in the final equality we use that
for \({\text {Re}}(w)>0\), since both \(1-\mathrm{e}^{-w}\) and \(\frac{1}{w}\) lie in the right half-plane. A short calculation with l’Hospital’s rule shows that the first two terms in (5.11) tend to zero (as \(w \rightarrow 0^+\)), and since the convergence of the series is uniform in w, for \({\text {Re}}(w)>0\), we have
Therefore (5.10), and as such (5.9), holds. \(\square \)
5.2 The multivariable Euler–Maclaurin summation formula
We now turn to the multivariable version of the Euler–Maclaurin asymptotic expansion stated in Theorem 1.4. The following proposition is a refined version that enables a proof by induction.
Proposition 5.2
Suppose that \(0 \le \theta _j < \frac{\pi }{2}\) for \(1 \le j \le s\), and that \(f:\mathbb {C}^s\rightarrow \mathbb {C}\) is holomorphic in a domain containing \(D_{\varvec{\theta }}\). If f and all of its derivatives are of sufficient decay in \(D_{\varvec{\theta }}\), then for \(1 \le r < s\), \(\varvec{a}\in \mathbb {R}^s\) and \(N\in \mathbb {N}_0\), we have
where \(\varvec{x}\in \mathbb {R}^s\), \(\varvec{x}_{r+1,s} := (x_{r+1},\ldots , x_s)\), \(\varvec{0}_j\) is the zero vector of length j, and \(g_{r,w}:\mathbb {C}^{s-r}\rightarrow \mathbb {C}\) is such that \(g_{r,w}(\varvec{x}_{r+1,s})\ll 1\) uniformly in \(\varvec{x}_{r+1,s}\) as \(w\rightarrow 0\) in \(D_{\theta _1}\cap \cdots \cap D_{\theta _r}\) and
uniformly in w and \((x_{r+j+1},\dotsc ,x_s)\) for \(1\le j\le s-r\). When \(r=s\), (5.12) holds with \(\varvec{x}_{r+1,s}\) being the empty vector and \(g_{s,w}\) a function of w such that \(g_{s,w}\ll 1\) as \(w\rightarrow 0\) in \(D_{\theta _1}\cap \cdots \cap D_{\theta _r}\).
Proof
Throughout the proof \(s \in \mathbb {N}\) is fixed, and we proceed by induction on r. At various points in the proof, we consider \(f^{(n_1, \dots , n_s)}\), with each \(n_k \le N\), and restrict this derivative to a function of the single variable \(x_j\) (with all other variables held fixed). We choose \(R > 0\) such that all such functions are holomorphic in a neighborhood containing \(C_R(0)\). This is possible because the decay assumption implies that R can be chosen uniformly for each individual function, and there are finitely many in total.
The base case of \(r=1\) is (5.8) with \(f(x) \mapsto f(x,\varvec{x}_{2,s})\) and the resulting \(g_{1,w}(\varvec{x}_{2,s})\) is
where \(R>0\) is sufficiently small. Due to the assumption of sufficient decay, the integrals converge uniformly in \(\varvec{x}_{2,s}\) and thus we see that the second and third terms above meet the conditions required of \(g_{1,w}\). For the first term, (5.2) gives
which also meets the conditions required of \(g_{1,w}\) due to the assumption of sufficient decay.
We now fix r with \(2\le r\le s\), assume that (5.12) is true for \(r-1\), and verify that it holds for r. We let \(\varvec{x}_{r-1}:=(x_1,\dotsc ,x_{r-1})\) and \(\varvec{a}_{r-1}:=(a_1,\ldots ,a_{r-1})\). By assumption, we can take (5.12) with \(r-1\) and integrate with respect to \(x_{r}\), and also divide by w, which yields
where
The restrictions on \(g_{r-1,w}\) give that \(h_{1,w}(\varvec{x}_{r+1,s})\) satisfies the conditions required of \(g_{r,w}\).
However, for fixed \(\varvec{m}\in \mathbb {N}_0^{r-1}\), by (5.8) we have
This yields that
where
We find that \(h_{2,w}\) satisfies the conditions of \(g_{r,w}\) by reasoning similar to that used for \(h_{1,w}\).
For fixed \(n_r\), applying (5.12) with \(r-1\) gives
where each \(g_{n_r,r-1,w}(\varvec{x}_{r+1,s})\) satisfies the conditions of \(g_{r,w}\). Plugging this back into (5.14) yields
where \(h_{3,w}(\varvec{x}_{r+1,s})\) satisfies the conditions required of \(g_{r,w}\), since it is the sum of \(h_{2,w}\) and the finitely many \(g_{n_r,r-1,w}\).
We insert (5.15) back into (5.13) to find that
where \(g_{r,w}(\varvec{x}_{r+1,s}) := h_{1,w}(\varvec{x}_{r+1,s}) + h_{3,w}(\varvec{x}_{r+1,s})\). Upon inspection, we find that the third, fourth, fifth, and sixth terms in the right hand-side combine exactly as stated in (5.12), so that the proof is complete. \(\square \)
6 Concluding remarks
There is another variant of the Circle Method due to Wright that is closely related to Ingham’s Tauberian theorem. Recall that in Theorem 1.1, the analytic behavior of B(q) in a small region near \(q = 1\) is sufficient to determine the asymptotic main terms of the coefficients \(b_n\). In particular, for a small, fixed \(t > 0\), the conditions in (1.2) require an asymptotic formula for \(B(\mathrm{e}^{-t})\), and uniform bounds for B(q) along a small arc of radius \(\mathrm{e}^{-t}\).
In contrast, Wright’s Circle Method requires an asymptotic formula for B(q) near \(q = 1\) (the “Major arc”), as well as bounds along the remainder of the circle of radius \(\mathrm{e}^{-t}\) (the “Minor arc”). However, the conclusion is also stronger, as one obtains an asymptotic expansion for the coefficients \(b_n\), so long as one has an asymptotic expansion for \(B(\mathrm{e}^{-t})\). Wright first introduced this approach in [19], and applied it to another example in [20]; in the latter case, he also used Euler–Maclaurin summation to derive the asymptotic expansion for B.
In their comprehensive article [15], Ngo and Rhoades gave a generalized version of Wright’s Circle Method. Specifically, Proposition 1.8 in [15] requires that
in the restricted angle \(y \le \Delta |x|\), as well as
for some \(d > 0\) in the remainder of the circle \(|z| = \mathrm{e}^{-x}\). In that case, the resulting asymptotic expansion for the coefficients is
where \(\beta _{s,r}\) are certain combinatorial coefficients. Furthermore, they showed that this result applies to a wide class of functions following essentially the same arguments we discussed in Sect. 4. In particular, they proved that (6.1) holds if \(B(q) = L(q) \xi (q)\), where \(\xi (q)\) essentially behaves like a modular form, and L(q) has an asymptotic expansion that is derived using Euler–Maclaurin summation.
References
Abramowitz, M., Stegun, I. (eds.): Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables. National Bureau of Standards Applied Mathematics Series, vol. 55. Courier Corporation, North Chelmsford (1964)
Apostol, T.: Modular Functions and Dirichlet Series in Number Theory Series. Graduate Texts in Mathematics, vol. 41. Springer, Berlin (1990)
Avakumović, V., Karamata, J.: Über einige Taubersche Sätze, deren Asymptotik von Exponentialcharakter ist. 1. Math. Z. 41, 345–356 (1936)
Borwein, J., Calkin, N., Manna, D.: Euler–Boole summation revisited. Am. Math. Monthly 116, 387–412 (2009)
Bringmann, K., Jennings-Shaffer, C., Mahlburg, K.: The asymptotic distribution of the rank for unimodal sequences. arXiv:1910.10790
Bringmann, K., Jennings-Shaffer, C., Mahlburg, K., Rhoades, R.: Peak Positions of Strongly Unimodal Sequences. Trans. Am. Math. Soc. 372, 7087–7109 (2019)
Brisebarre, N., Philibert, G.: Effective lower and upper bounds for the Fourier coefficients of powers of the modular invariant \(j\). J. Ramanujan Math. Soc. 20, 255–282 (2005)
Chen, S.-C.: Congruences and asymptotics of Andrews’ singular overpartitions. J. Number Theory 164, 343–358 (2016)
Choi, Y.-S., Kim, B., Lovejoy, J.: Overpartitions into distinct parts without short sequences. J. Number Theory 175, 117–133 (2017)
Cotron, T., Dicks, R., Fleming, S.: Asymptotics and congruences for partition functions which arise from finitary permutation groups. Res. Number Theory 2, 20 (2016)
Han, G.-N., Xiong, H.: Some useful theorems for asymptotic formulas and their applications to skew plane partitions and cylindric partitions. Adv. Appl. Math. 96, 18–38 (2018)
Hardy, G., Ramanujan, S.: Asymptotic formulæ in combinatory analysis. Proc. Lond. Math. Soc 17, 75–115 (1918)
Ingham, A.: A Tauberian theorem for partitions. Ann. Math. 42, 1075–1090 (1941)
Kim, B., Kim, E.: On the subpartitions of the ordinary partitions II. Electron. J. Comb. 21, Paper 4.21, 11pp (2014)
Ngo, H., Rhoades, R.: Integer partitions, probabilities and quantum modular forms. Res. Math. Sci. 4, 17 (2017)
NIST Digital Library of Mathematical Functions. Olver, F.W.J., Olde Daalhuis, A.B., Lozier, D.W., Schneider, B.I., Boisvert, R.F., Clark, C.W., Miller, B.R., Saunders, B.V. eds. http://dlmf.nist.gov/, Release 1.0.23 of 2019-06-15
Rademacher, H.: On the partition function p(n). Proc. Lond. Math. Soc. 43, 241–254 (1937)
Rolen, L.: On \(t\)-core towers and \(t\)-defects of partitions. Ann. Comb. 21, 119–130 (2017)
Wright, E.: Asymptotic partition formulae II. Weighted partitions. Proc. Lond. Math. Soc. 36, 117–141 (1933)
Wright, E.: Stacks. II. Q. J. Math. 22, 107–116 (1971)
Zagier, D.: The Mellin Transform and Related Analytic Techniques. Appendix to E. Zeidler, Quantum Field Theory I: Basics in Mathematics and Physics. A Bridge Between Mathematicians and Physicists, pp. 305–323. Springer, Berlin (2006)
Acknowledgements
The authors thank Karl Dilcher for bringing Euler–Boole summation to our attention. Moreover we thank the referee for useful comments on an earlier version of this paper.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Additional information
Dedicated to the life and memory of Richard Askey
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The first author was supported by the European Research Council (ERC) under the European Union’s Horizon 2020 Research and Innovation Programme (Grant agreement no. 101001179).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Bringmann, K., Jennings-Shaffer, C. & Mahlburg, K. On a Tauberian theorem of Ingham and Euler–Maclaurin summation. Ramanujan J 61, 55–86 (2023). https://doi.org/10.1007/s11139-020-00377-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11139-020-00377-5