We introduce three families of discrete hyperbolic cosecant probability distributions, supported either bilaterally on Z or unilaterally on N0, inspired by series evaluations of Srinivasa Ramanujan. We determine their exact normalization constants in terms of singular values of complete elliptic integrals and the Gaußian constant G=ϖ/π with the lemniscate constant Γ2(1/4)/(22π) by explicitly calculating some series, e.,g. ∑κ∈Z2κcosech(κπ)=G2+π1. Furthermore, by connecting these distributions to Fourier transforms of powers of the hyperbolic secant and applying the Poisson summation formula, we obtain closed-form evaluations for even moments and hyperbolic secant lattice sums, e.,g. ∑k∈Zsech4(kπ)=32(G4+G2+π1). This machinery proves conjectures on lattice sums arising in electrostatics proposed by Vigren and Dieckmann (Symmetry, 2020).
Introduction
We introduce three types of discrete hyperbolic cosecant probability distributions, denoted either by Cosech in the bilateral case supported on Z, or by Cosech in the unilateral case supported on the non-negative natural numbers N0; this family of distributions is inspired by some series identities provided by Srinivasa Ramanujan. These distributions offer a view of the classical results through the lenses of probability theory and lemniscatic geometry.
Ramanujan’s formulae in his second notebook have a very general setup. They contain terms of Jacobi’s complete elliptic integral of the first and second kind, K and E, viz.
Here, k∈(0,1) is called the elliptic modulus with complementary modulus k′, such that k2+k′2=1, cf. [7 p. 31]. For r>0, let kr∈(0,1) be the unique elliptic modulus with kr′:=1−kr2, Kr:=K(kr), and Kr′=K(kr′), such that Kr′/Kr=r holds. The elliptic lambda-star function then yields λ∗(r)=kr. According to [4], in case of r∈Q+, the corresponding elliptic integrals Kr and Kr′ can be expressed in closed form using a finite number of values of the gamma function. For r∈N, kr is called an elliptic singular modulus and the corresponding elliptic integrals Kr can often be calculated explicitly. Cf. [2 p. 298] for the values of Kr for the singular elliptic moduli k1,…,k16; numerous further values of singular elliptic moduli are stated in the exercises in the same book. For the calculation of Er:=E(kr), we use the elliptic alpha function α(r) which is implicitly defined via Equation (5.1.2) in [2 p. 152]; it yields
Er=4Krrπ+(1−rα(r))Kr(1)
and the Tables 5.2a., 5.2b., and 5.3. on p. 172–173 in the same book provide many special values for α(r) for r∈Q+. For elliptic singular values kr, Ramanujan’s notation in the Chapters 17 and 18 of his second notebook can be translated into the common language via x=kr2, y=rπ, and z=2Kr/π, cf. Entry 6 of Chapter 17 in [9 p. 102].
Let ϖ:=Γ2(1/4)/(4π) denote the lemniscate constant. A particularly important elliptic singular modulus occurs in the lemniscatic case where r=1. Then, we get K1′=K1=ϖ/2 and λ∗(1)=k1=k1′=1/2. Hence, in the lemniscatic case, we may substitute x=1/2, y=π and z=2G into Ramanujan’s equations. Finally, Gauß’ constant G:=ϖ/π yields particularly simple expressions for Kr, e.g. for its values at singular elliptic moduli kr for r=s2 with s∈{1,…,5}. In this case, also the normalization constants for our discrete type I and type III hyperbolic cosecant distribution are given in notably simple closed forms.
The Discrete Hyperbolic Cosecant Distribution – Type I
We introduce the discrete type I hyperbolic cosecant distribution of the first, second, and third kind which are mainly based on the hyperbolic cosecant series identities given in [9] and [3].
We begin with the basic case, viz. the bilateral lemniscatic Cosech distribution. Let w(x):=xcosech(xπ/2) be the underlying continuous real weight function; we replace its removable singularity in 0 by setting w(0):=limx→0w(x)=2/π. The discrete weight functions wX1, wY1, and wZ1 are obtained by sampling the continuous weight function w on the corresponding even, odd, and integer lattices, respectively. Specifically, for κ∈Z, define wX1(κ):=w(2κ), wY1(κ):=w(2κ+1), and wZ1(κ):=w(κ) for the type I Cosech distributions of the first, second, and third kind, respectively. The proofs for the normalization constants of the lemniscatic type I Cosech distribution are based on quasi-modularity and a duplication formula for the Eisenstein series E2 for the first kind; we cite Ramanujan’s formula (xiii) in Entry 15 in Chapter 17 of his second notebook for the second kind, cf. [9 p. 133]; finally, a simple parity split yields the normalization constant for the third kind. For the type I Cosech distribution of the first kind, the underlying weight function is
wX1(x):=w(2x). Its simple Fourier transform, viz.
will be particularly useful in the context of Poisson summation. We therefore call the CosechX1 distribution simply the discrete hyperbolic cosecant distribution Cosech.
Proposition and Definition 2.1. [Lemniscatic Type I Cosech Distributions.] Let κ∈Z. For the rows 2–4 of Table 1, let CA denote the normalization constant of the listed CosechA distribution with weight function wA(κ). Then the associated sequence (pA(κ))κ∈Z:=(CA⋅wA(κ))κ∈Z is a probability mass function. We name CosechX1, CosechY1, and CosechZ1 the discrete lemniscatic type I hyperbolic cosecant distribution of the first, second, and third kind, respectively. We omit adjectives when the context is clear. We call the CosechX1 distribution simply the discrete hyperbolic cosecant distribution Cosech.
Type
Distribution
Normalizing Constant
Weight Function
I
Cosech = CosechX1
πG2+1π
2κcosech(κπ)
I
CosechY1
2G21
(2κ+1)cosech((2κ+1)2π)
I
CosechZ1
πG2+1+2πG2π
κcosech(κ2π)
I
CosechXr
π+4α(r)Kr2rπ2
2κcosech(rκπ)
I
CosechYr
4krKr2π2
(2κ+1)cosech(r(2κ+1)2π)
I
CosechZr
π+4α(r)Kr2+4rkrKr2rπ2
κcosech(rκ2π)
I
CosechX1
πG2+32π
(2n+2)cosech((2n+2)2π)
I
CosechY1
G22
(2n+1)cosech((2n+1)2π)
I
CosechZ1
(1+2)πG2+32π
(n+1)cosech((n+1)2π)
I
CosechXr
4α(r)Kr2+3π2rπ2
(2n+2)cosech(r(2n+2)π)
I
CosechYr
2krKr2π2
(2n+1)cosech(r(2n+1)2π)
I
CosechZr
4α(r)Kr2+4rkrKr2+3π2rπ2
(n+1)cosech(r(n+1)2π)
Table 1. Discrete bilateral (lemniscatic) type I Cosech distributions of the first (X), second (Y), and third (Z) kind for κ∈Z and r>0 (r=1); further the discrete unilateral (lemniscatic) type I Cosech distributions of the first (X), second (Y), and third (Z) kind for n∈N0 and r>0 (r=1).
Proof. The lemniscatic normalization constants could easily be deduced from the general elliptic formulae presented in Proposition 2.3, by setting Kr′/Kr=r=1. But we want to examine some connections of the Cosech distribution to Eisenstein series and Jacobi theta functions, and further to Ramanujan’s work; hence, now we give separate proofs for the lemniscatic case.
Ad CosechX1: Unfortunately, the normalization constant CX1 is not provided by Entry 15 in Chapter 17 of Ramanujan’s second notebook, cf. [9 p. 132–133]. Instead, Ramanujan states here formulae for odd moments ≥3 of the Cosech distribution. We calculate the normalization constant CX1 as follows: First, we express wX1(n) in terms of two Eisenstein series, viz. E2(i) and E2(i/2). Due to the quasi-modularity of E2, we can directly evaluate E2(i)=3/π and write E2(i/2) in terms of E2(2i). Finally, we determine the value of E2(2i) by showing a duplication formula. This includes Jacobi theta functions whose values we can directly compute. In detail: For every n∈N, by definition we have
Let τ∈H={z∈C:ℑ(z)>0} lie in the upper half-plane and define the nome qτ=eiπτ. The Eisenstein series E2 has the Lambert-series expansion
E2(τ)=1−24n=1∑∞1−qτ2nnqτ2n.
Hence,
n=1∑∞2ncosech(nπ)=61(E2(i)−E2(2i)).
Let G2(τ)=2ζ(2)E2(τ) be a scaled version of E2 by the factor ζ(2)=π2/6 of the Riemann zeta function. The quasi-modular transformation formula
G2(−τ1)=τ2G2(τ)−2πiτ⟺E2(−τ1)=τ2E2(τ)+πi6τ
is given by Equation (50) on p. 69 in [1]. This formula directly yields the values E2(i)=3/π, and E2(i/2)=12/π−4E2(2i). Hence, we get
n=1∑∞2ncosech(nπ)=32E2(2i)−2π3.
Now, we represent the Eisenstein series E2(2τ) by a duplication formula which includes E2(τ) and the Jacobi theta functions ϑ3 and ϑ4. Those are defined by
ϑ3(qτ)=k∈Z∑qτk2,ϑ4(qτ)=k∈Z∑(−1)kqτk2,
cf. [11 p. 464]. The duplication formula reads as follows.
Lemma 2.2 (Duplication formula). Letτ∈H={z∈C:ℑ(z)>0}and define the nomeqτ=eiπτ. Then
E2(2τ)=21E2(τ)+41(ϑ34(qτ)+ϑ44(qτ))
holds.
Proof. According to [2 Proposition 3.2, pp. 70–71], the fourth powers of the Jacobi theta functions admit the Lambert-series expansions
Since E2(τ)=1−24∑n=1∞1−qτ2nnqτ2n and E2(2τ)=1−24∑n=1∞1−qτ4nnqτ4n, we conclude
ϑ34(qτ)+ϑ44(qτ)=4E2(2τ)−2E2(τ),
and the claim follows.
Now, we use the well-known identities ϑ32(qτ)=2K(k)/π and ϑ42(qτ)=2k′K(k)/π, e. g. see Equation (5) in [Zucker, 1984]. In the lemniscatic case, where λ∗(1)=k1=1/2 and K(1/2)=ϖ/2, this results in ϑ34(qi)=2G2 and ϑ44(qi)=G2. Together with the Eisenstein value E2(i)=3/π from above, we get E2(2i)=3/(2π)+3G2/4. Hence,
n=1∑∞2ncosech(nπ)=2ππG2−1.
Since by definition, we have wX1(0)=2/π, we finally deduce
Ad CY1: In the lemniscatic case, we translate Ramanujan’s formulae in Ramanujan’s Entry 15 in Chapter 17 of his second notebook, cf. [9 p. 132–133] into the common language via x=1/2, y=π, and z=2G. Now, Equation (xiii) in this Entry reads as
due to absolute convergence of each series. Hence, we get
CZ1=(1+2)πG2+1π
as stated.
Next, we observe the discrete type I Cosech distribution in the general elliptic setting.
Proposition and Definition 2.3. [Type I Cosech Distributions.] Let κ∈Z and r>0. For the rows 5–7 in Table 1, let CA denote the normalization constant of the listed CosechA distribution with weight function wA(κ). Then the associated sequence (pA(κ))κ∈Z:=(CA⋅wA(κ))κ∈Z is a probability mass function.
We name CosechXr, CosechYr, and CosechZr the discrete type I hyperbolic cosecant distribution of the first, second, and third kind, respectively. We omit adjectives when the context is clear.
Proof. Ad CXr: For the calculation of the normalization constant CXr, we follow the method in [3]. We use the Jacobi elliptic function dn and the Lambert-series expansion of dn2 for the nome e−πK′/K, viz.
dn2(u,k)=KE+K22π2n=1∑∞1−q2nnqncos(Kuπn),
cf. Equation (14) in [3 p. 294]. Since dn(0,k)=1, a modification of equation (3) yields the identities
cf. Equation (25) in [3 p. 294]. We choose the elliptic lambda-star function λ∗(r)=kr such that K(kr′)/K(kr)=r. By definition of the weight function in 0 and due to Equation (1), we get
Ad CYr: Equation (xiii) of Ramanujan’s Entry 15 in Chapter 17 of his second notebook, cf. [9 p. 133], can be translated into the common language via x=kr2, y=πr, and z=2Kr/π, cf. [9 p. 102], viz.
In [9] and [3], the hyperbolic series are formulated as a summation over N0. Hence, the proof of Proposition 2.3 directly yields a unilateral version Cosech of our type I Cosech distribution. For the lemniscatic case, we simply insert r=1, α(1)=1/2, k1=1/2, and K1=ϖ/2 into the general elliptic formulae.
Corollary and Definition 2.4. [(Lemniscatic) Type I Cosech Distribution.] Let n∈N0 and r>0 (r=1). For the rows 8–14 of Table 1, let CA denote the normalization constants of the listed CosechA distribution with weight function wA(n). Then the associated sequence (pA(n))n∈N0:=(CA⋅wA(n))n∈N0 is a probability mass function.
We name CosechXr, CosechYr, and CosechZr the discrete unilateral (lemniscatic for r=1) type I hyperbolic cosecant distribution of the first, second, and third kind, respectively. We omit adjectives when the context is clear.
The Discrete Hyperbolic Cosecant Distribution – Type II
Now, we introduce the discrete type II hyperbolic cosecant distribution, building on the foundational work of [9], [3], and [Zucker, 1984].
For the type I Cosech distribution, the underlying continuous weight function w(x)=xcosech(xπ/2) has a removable singularity at 0. The x factor is necessary to make w positive and even on its domain Z. In contrast, for the type II Cosech distribution, we use the underlying continuous real weight function wII(x):=cosech(xπ/2)/x which has a non-removable double pole at 0. Hence, we first consider the unilateral type II Cosech distributions on N0 after index shift. We prove the general elliptic formulae and for the lemniscatic case where r=1, we insert the elliptic singular modulus k1=1/2 into the formulae.
Proposition and Definition 3.1. [(Lemniscatic) Type II Cosech Distributions.] Let n∈N0 and r>0 (r=1). For the rows 2–7 of Table 2, let CA denote the normalization constant of the listed CosechA distribution with weight function wA(n). Then the associated sequence (pA(n))n∈N0:=(CA⋅wA(n))n∈N0 is a probability mass function.
We name CosechUr, CosechVr, and CosechWr the discrete (lemniscatic for r=1) unilateral type II hyperbolic cosecant distribution of the first, second, and third kind, respectively. We omit adjectives when the context is clear.
Type
Distribution
Normalizing Constant
Weight Function
II
CosechU1
π−ln(8)24
cosech((n+1)π)/(2n+2)
II
CosechV1
ln(1+2)2=arsinh(1)2
cosech((2n+1)2π)/(2n+1)
II
CosechW1
π−ln(8)+12ln(1+2)24
cosech((n+1)2π)/(n+1)
II
CosechUr
rπ+2ln(kr)−2ln(2+2kr)−2ln(2−2kr)24
cosech(r(n+1)π)/(2n+2)
II
CosechVr
artanh(kr)2
cosech(r(2n+1)2π)/(2n+1)
II
CosechWr
rπ+2ln(4kr)+4ln(1+kr)−8ln(2−2kr)24
cosech(r(n+1)2π)/(n+1)
II
CosechV1
arsinh(1)1
cosech((2κ+1)2π)/(2κ+1)
II
CosechVr
ln(1+kr)−ln(1−kr)2
cosech(r(2κ+1)2π)/(2κ+1)
Table 2. Discrete unilateral (lemniscatic) type II Cosech distributions of the first (U), second (V), and third (W) kind for n∈N0 and r>0 (r=1); further the discrete bilateral (lemniscatic) type II Cosech distributions of the second (V) kind for κ∈Z and r>0 (r=1).
Proof. Ad CUr: We follow the technique in [Zucker, 1984]. Define the nome q=e−πK′/K with the complete elliptic integral of the first kind K and further the infinite products
Q0:=n=1∏∞(1−q2n),Q3:=n=1∏∞(1−q2n−1).
Equation (5) in [Zucker, 1984, p. 406] yields the equations
AdCVr: The suitable formula is given by the Example after Entry 7 of Chapter 18 in Ramanujan’s second notebook, cf. [9 p. 156]. With x=k2 and y=πK′/K, Ramanujan states
Ad CWr: We use the transformations stated in Equation (7) from [Zucker, 1984, p. 409]. We choose the elliptic lambda-star function λ∗(r)=kr such that Kr′/Kr=r. Hence, Equation (6) yields
Since it leaves the 0 out, we can also define a bilateral type II Cosech of the second kind. By symmetry, we just have to take the normalization constant CVr=CVr′/2.
Corollary and Definition 3.2. [Type II Cosech Distributions.] Let κ∈Z. For the rows 8–9 of Table 2, let CA denote the normalization constant of the listed CosechA distribution with weight function wA(κ). Then the associated sequence (pA(κ))κ∈Z:=(CA⋅wA(κ))κ∈Z is a probability mass function.
We name CosechVr and CosechV1 the discrete bilateral (lemniscatic for r=1) type II hyperbolic cosecant distribution of the second kind, respectively. We omit adjectives when the context is clear.
The Discrete Hyperbolic Cosecant Distribution – Type III
Next, we introduce a further version of the Cosech distribution: The type III hyperbolic cosecant distribution. Like for the type II Cosech distribution, the underlying continuous weight function wIII(x):=cosech2(xπ) has a non-removable double pole at 0. Therefore, we again first concentrate on the unilateral type III Cosech distributions on N0 after index shift. We use techniques from [9], [2], and [12] to prove the general elliptic case for r>0. For the lemniscatic case, we insert r=1, k1=1/2, K1=ϖ/2, and α(1)=1/2 into the general formulae.
Proposition and Definition 4.1. [(Lemniscatic) Type III Cosech Distributions.] Let n∈N0 and r>0 (r=1). For the rows 2–6 of Table 3, let CA denote the normalization constant of the listed CosechA distribution with weight function wA(n). Then the associated sequence (pA(n))n∈N0:=(CA⋅wA(n))n∈N0 is a probability mass function.
Type
Distribution
Normalizing Constant
Weight Function
III
CosechR1
π−36π
cosech2((2n+2)2π)
III
CosechS1
πG2−12π
cosech2((2n+1)2π)
III
CosechT1
π−6+3πG26π
cosech2((n+1)2π)
III
CosechRr
rπ2−3π+4Kr2(3α(r)−(1+kr2)r)6rπ2
cosech2(r(2n+2)2π)
III
CosechSr
4Kr2α(r)−π2rπ2
cosech2(r(2n+1)2π)
III
CosechTr
rπ2−6π+4Kr2(6α(r)−(1+kr2)r)6rπ2
cosech2(r(n+1)2π)
III
CosechS1
πG2−1π
cosech2((2κ+1)2π)
III
CosechSr
4Kr2α(r)−πrπ2
cosech2(r(2κ+1)2π)
Table 3. Discrete unilateral (lemniscatic) type III Cosech distributions of the first (R), second (S), and third (T) kind for n∈N and r>0 (r=1); further the discrete bilateral (lemniscatic) type III Cosech distributions of the second (S) kind for κ∈Z and r>0 (r=1).
We name CosechRr, CosechSr, and CosechTr the discrete (lemniscatic for r=1) unilateral type III hyperbolic cosecant distribution of the first, second, and third kind, respectively. We omit adjectives when the context is clear.
Proof. Ad CRr: For the calculation of the normalizing constant CRr for the CosechRr distribution we can use the technique in [12 p. 197] or complete Exercise 7 of chapter 3.7 in [2 p. 96]. We do the latter. Let c:=Kr′/Kr with nome q=e−πc, where k=k(c) is determined implicitly. Further, define
For c>0, the series and its termwise derivative converge uniformly on every compact subinterval of (0,∞). Hence, the derivative of the series yields
F′(c)=−2πn=1∑∞cosech2(cnπ).
Differentiating the right-hand side with respect to c results in
F′(c)=−12π−61dcdln(2kk′K3).
Equating the two expressions for F′(c), we obtain
n=1∑∞cosech2(cnπ)=61+3π1dcdln(2kk′K3).
We now compute the derivative on the right-hand side. Since K is a function of k, and k is a function of c, the expression f(K,k) depends on c through k(c). Thus, by the chain rule,
dcdf(K(k(c)),k(c))=(∂K∂fdkdK+∂k∂f)dcdk.
The expression in parentheses is the total derivative with respect to k. Hence,
dcdln(2kk′K3)=dkdln(2kk′K3)dcdk.
The standard derivative formula, given in equation 19.4.1 of [8], is
Since its support excludes 0, we can also define the bilateral type III Cosech of the second kind. By symmetry, we just have to take the normalization constant CSr=CSr/2.
Corollary and Definition 4.2. [Type III Cosech Distributions.] Let κ∈Z. For the rows 8–9 of Table 3, let CA denote the normalization constant of the listed CosechA distribution with weight function wA(κ). Then the associated sequence (pA(κ))κ∈Z:=(CA⋅wA(κ))κ∈Z is a probability mass function.
We name CosechSr and CosechS1 the discrete bilateral (lemniscatic for r=1) type III hyperbolic cosecant distribution of the second kind, respectively. We omit adjectives when the context is clear.
Moments of the Discrete Hyperbolic Cosecant Distribution
We want to determine some moments of the discrete hyperbolic cosecant distribution Cosech. According to Table 1, its probability mass function is given by
(p(κ))κ∈Z=πG2+1π2κcosech(κπ).
Ramanujan’s series identities directly yield the even moments up to the eighth order.
Proposition 5.1. Let X∼Cosech be a discrete random variable which is distributed according to the hyperbolic cosecant distribution. The odd moments vanish and we have
(i)
E[X2]=Var(X)=πG2+1πG4≈0.4781246,
(ii)
E[X4]=πG2+13πG6≈0.9991876,
(iii)
E[X6]=πG2+118πG8≈4.1762161,
(iv)
E[X8]=πG2+1189πG10≈30.5461835,
with the Gaußian constant G=ω/π.
Proof. Let C=π/(πG2+1). We get E[Xj]=0 for odd j∈N due to symmetry. The other formulae directly follow by Ramanujan’s Entry 15 (i)-(iv) in Chapter 17 of his second notebook, cf. [9 pp. 132-133]: For j∈{2,4,6,8} we write
We want to evaluate some exact expressions for series of even powers of the hyperbolic secant via the Poisson summation. For that purpose we first prove a formula for their Fourier transform.
Lemma 5.2. Let m∈N be a positive integer and let ξ∈R. Then the Fourier transformation formula
Proof. We use the hyperbolic substitution u=u(x):=(1+tanh(xπ))/2, and hence 2u−1=tanh(xπ), which implies x=ln(u/(1−u))/(2π). Now, 4u(1−u)=1−tanh2(xπ)=sech2(xπ) leads to
sech2m(xπ)=4mum(1−u)m.
For the boundaries, we get limx→−∞u(x)=0 and limx→∞u(x)=1. Hence, by continuity, u∈(0,1) and thus, u/(1−u)∈(0,∞). Therefore, we can use the unique, real natural logarithm to get
e−2πixξ=exp(−iξln(1−uu))=u−iξ(1−u)iξ.
Further, u′(x)=πsech2(xπ)/2 results in
dx=πsech2(xπ)2du=2πu(1−u)du.
We use Euler’s beta function B:{(a,b)∈C2:ℜ(a),ℜ(b)>0}→C with its key property
Proof. Ad (i): For the application of the Poisson summation formula, we first calculate the Fourier transformation of sech2(kπ). According to Lemma 5.2, one has
F[x↦sech2(xπ)](ξ)=π2∣Γ(1+iξ)∣2=2ξcosech(ξπ).
Now, the Poisson summation formula and Equation (4) in the proof of Proposition 2.1 yield
Ad (ii) [Probabilistic proof]: Let X∼Cosech be a discrete random variable which is distributed according to the hyperbolic cosecant distribution with normalization constant C=π/(πG2+1). Therefore, by using the duality of the Fourier transformation from (i), the Poisson summation formula yields
Now, our (4) in the proof of Proposition 2.1 provides
κ∈Z∑2κcosech(κπ)=G2+π1.
Further, Ramanujan’s series identity (i) of Entry 15 in Chapter 17 of his second notebook, cf. [9 p. 132], which we already used in the proof of Proposition 5.1 with x=1/2, y=π, and z=2G, yields
Remark 5.4. Identity (ii) in Theorem 5.3 provides an analytic proof for a conjecture on electrostatic lattice sums proposed by Vigren and Dieckmann [Vigren and Dieckmann, [10] via experimental mathematics. Further identities and physical applications in this direction will be investigated in subsequent work.
Leonard Jobst Eberhard Pleschberger, Neptunstraße 18, Düsseldorf D-40223 & Sechshauser Straße 24/1/15, A-1150 Vienna.
[1][Apostol, 1990] Apostol, Tom M.: Modular Functions and Dirichlet Series in Number Theory (2nd edition). Springer.
[2][Borwein and Borwein, 1987] Borwein, Jonathan M.; Borwein, Peter B.: Pi and the AGM. A Study in Analytic Number Theory and Computational Complexity. Wiley.
[3][Bruckman, 1977] Bruckman, P. S.: On the evaluation of certain infinite series by elliptic functions. Fibonacci Quart. 15 (1977), 293–310.DOI
[4][Chowla and Selberg, 1967] Chowla, S., and Selberg, A. (1967). On Epstein’s Zeta-function. Journal für die reine und angewandte Mathematik, 227, 86–110.DOI
[7][Jacobi, 1829] Jacobi, C. G. J. (1829). Fundamenta nova theoriae functionum ellipticarum. Publisher Sumtibus fratrum Borntreager
[8][NIST, 2026] National Institute of Standards and Technology (NIST). Digital Library of Mathematical Functions (DLMF). Version 1.2.8; Release date: September 2026.
[9][Ramanujan and Berndt, 1991] Ramanujan, Srinivasa; Berndt, Bruce (1991). Ramanujan’s Notebooks. Part III. Springer.
[10][Vigren and Dieckmann, 2020] Vigren, Erik; Dieckmann, Andreas (2020). Simple Solutions of Lattice Sums for Electric Fields Due to Infinitely Many Parallel Line Charges. Symmetry (2020), 12(6), 1040.DOI
[11][Whittaker and Watson, 1920] Whittaker, Edmund T.; Watson, George N. (1920). A Course of Modern Analysis; an Introduction to the General Theory of Infinite Processes and of Analytic Functions; with an Account of the Principal Transcendental Functions (3rd Edition). Cambridge University Press.
[12][Zucker, 1979] Zucker, I. J.: The Summation of Series of Hyperbolic Functions. SIAM J. Math. Anal. 10(1) (1979), 192–206.DOI