9. Proof of the Optimal Function Inequalities
- No associated Lean code or declarations.
Consider the function A:(0,\infty)\to\C defined as
A(t):=-t^2\phi_0(i/t)-\frac{36}{\pi^2}\,\psi_I(it).
Then
A(t) < 0
for all t > 0.
- No associated Lean code or declarations.
Consider the function B:(0,\infty)\to\C defined as
B(t) := -t^2\phi_0(i/t)+\frac{36}{\pi^2}\,\psi_I(it).
Then
B(t) > 0
for all t > 0.
Here we formalize the proof of the inequalities by Lee (2024). First, we can rewrite the inequality in Lemma 9.1 as follows.
Define two quasi-modular forms by
F(z) = (E_2(z) E_4(z) - E_6(z))^2
G(z) = H_2(z)^{3} (2 H_{2}(z)^{2} + 5 H_{2}(z) H_{4}(z) + 5 H_{4}(z)^{2}).
Lean code for Definition9.3●2 definitions
-
defdefined in SpherePacking/ModularForms/FG.leancomplete
def F : UpperHalfPlane → ℂ
def F : UpperHalfPlane → ℂ
Definition of $F$ and $G$ and auxiliary functions for the inequality between them on the imaginary axis.
-
defdefined in SpherePacking/ModularForms/FG.leancomplete
def G (τ : UpperHalfPlane) : ℂ
def G (τ : UpperHalfPlane) : ℂ
- No associated Lean code or declarations.
We have
\phi_0 = \frac{F}{\Delta}
\psi_S = -\frac{1}{2} \frac{G}{\Delta}.
Equation eqn:phi0-F is clear. Equation eqn:psiS-G is already shown in
Lemma Lemma 8.19.
- No associated Lean code or declarations.
Inequality eqn:ineqA and eqn:ineqB are equivalent to
F(it) + \frac{18}{\pi^2} G(it) > 0
F(it) - \frac{18}{\pi^2} G(it) > 0
respectively.
By eqn:psiS-define from Definition 8.18,
\psi_I(it) = (\psi_S|_{-2}S)(it) = (it)^{2}\psi_S\left(-\frac{1}{it}\right) = -t^2 \psi_S\left(\frac{i}{t}\right).
Combined with Lemma Lemma 9.4 we can rewrite
eqn:ineqA as
A(t) = -t^2 \phi_0\left(\frac{i}{t}\right) + \frac{36}{\pi^2} \psi_S\left(\frac{i}{t}\right) < 0 \Leftrightarrow \frac{F(it)}{\Delta(it)} + \frac{18}{\pi^2} \frac{G(it)}{\Delta(it)} > 0
for t > 0, which is equivalent to eqn:ineqAnew by Corollary
Corollary 7.25.
Equivalences of eqn:ineqB and eqn:ineqBnew follows similarly; just change
the sign.
Now, the first inequality eqn:ineqAnew follows from the positivity of each
F(it) and G(it).
-
F_imag_axis_pos[complete] -
G_imag_axis_pos[complete]
For all t > 0, we have F(it) > 0 and G(it) > 0.
Lean code for Lemma9.6●2 theorems
Associated Lean declarations
-
F_imag_axis_pos[complete]
-
G_imag_axis_pos[complete]
-
F_imag_axis_pos[complete] -
G_imag_axis_pos[complete]
-
theoremdefined in SpherePacking/ModularForms/FG.leancomplete
theorem F_imag_axis_pos : ResToImagAxis.Pos F
theorem F_imag_axis_pos : ResToImagAxis.Pos F
`F(it) > 0` for all `t > 0`. Blueprint: F = 9*(D E₄)² and D E₄ > 0 on imaginary axis.
-
theoremdefined in SpherePacking/ModularForms/FG.leancomplete
theorem G_imag_axis_pos : ResToImagAxis.Pos G
theorem G_imag_axis_pos : ResToImagAxis.Pos G
`G(it) > 0` for all `t > 0`. Blueprint: Lemma 8.6 - follows from H₂(it) > 0 and H₄(it) > 0. G = H₂³ (2H₂² + 5H₂H₄ + 5H₄²) is positive since all factors are positive.
By Ramanujan's identity Theorem 7.48, we have
F(z) = 9 E_4'(z)^2, and hence
F(it) = 9E_4'(it)^2 = 9 \left(240\sum_{n \geq 1} n \sigma_3(n) e^{-2 \pi n t} \right)^{2} > 0.
The inequality G(it) > 0 follows from positivity of H_2(it) and
H_4(it) in Corollary 7.42.
Equation eqn:ineqAnew holds.
Lean code for Corollary9.7●1 theorem, incomplete
Associated Lean declarations
-
FG_inequality_1[sorry in proof]
-
FG_inequality_1[sorry in proof]
-
theoremdefined in SpherePacking/ModularForms/FG.leancontains sorry
theorem FG_inequality_1 {t : ℝ} (ht : 0 < t) : FReal t + 18 * Real.pi ^ (-2) * GReal t > 0
theorem FG_inequality_1 {t : ℝ} (ht : 0 < t) : FReal t + 18 * Real.pi ^ (-2) * GReal t > 0
Main inequalities between $F$ and $G$ on the imaginary axis.
To prove the second inequality, we need some identities satisfied by F and
G.
F and G satisfy the following differential equations:
\partial_{12}\partial_{10} F - \frac{5}{6} E_{4} F = 7200 \Delta (-E_{2}')
\partial_{12}\partial_{10} G - \frac{5}{6} E_{4} G = -640 \Delta H_{2}.
Lean code for Lemma9.8●2 theorems
-
theoremdefined in SpherePacking/ModularForms/FG.leancomplete
theorem MLDE_F : serre_D 12 (serre_D 10 F) = 5 * 6⁻¹ * E₄.toFun * F + 7200 * Δ_fun * negDE₂
theorem MLDE_F : serre_D 12 (serre_D 10 F) = 5 * 6⁻¹ * E₄.toFun * F + 7200 * Δ_fun * negDE₂
Modular linear differential equation satisfied by $F$.
-
theoremdefined in SpherePacking/ModularForms/FG.leancomplete
theorem MLDE_G : serre_D 12 (serre_D 10 G) = 5 * 6⁻¹ * E₄.toFun * G - 640 * Δ_fun * H₂
theorem MLDE_G : serre_D 12 (serre_D 10 G) = 5 * 6⁻¹ * E₄.toFun * G - 640 * Δ_fun * H₂
Modular linear differential equation satisfied by $G$.
Both identities can be shown by direct computations.
By Ramanujan's identities (Theorem Theorem 7.48) and the
product rule for Serre derivatives (Theorem
Theorem 7.51), we have
\partial_{5} (E_2 E_4 - E_6)
= (E_2 E_4 - E_6)' - \frac{5}{12} E_2 (E_2 E_4 - E_6)
= \frac{E_2^2 - E_4}{12} \cdot E_4 + E_2 \cdot \frac{E_2 E_4 - E_6}{3} - \frac{E_2 E_6 - E_4^2}{2} - \frac{5}{12}E_2 (E_2 E_4 - E_6)
= -\frac{5}{12} (E_2 E_6 - E_4^2)
and
\partial_{7} (E_2 E_6 - E_4^2)
= (E_2 E_6 - E_4^2)' - \frac{7}{12} E_2 (E_2 E_6 - E_4^2)
= \frac{E_2^2 - E_4}{12} \cdot E_6 + E_2 \cdot \frac{E_2 E_6 - E_4^2}{2} - 2 E_4 \cdot \frac{E_2 E_4 - E_6}{3} - \frac{7}{12} E_2 (E_2 E_6 - E_4^2)
= -\frac{7}{12} E_4 (E_2 E_4 - E_6).
Using these, we compute
\partial_{10} F = \partial_{10} (E_2 E_4 - E_6)^2
= 2 (E_2 E_4 - E_6) \partial_{5} (E_2 E_4 - E_6)
= -\frac{6}{5} (E_2 E_4 - E_6) (E_2 E_6 - E_4^2),
and then
\partial_{12}\partial_{10} F
= -\frac{5}{6} \partial_{12} ((E_2 E_4 - E_6) (E_2 E_6 - E_4))
= -\frac{5}{6} (\partial_{5}(E_2 E_4 - E_6)) (E_2 E_6 - E_4^2) - \frac{5}{6} (E_2 E_4 - E_6) (\partial_{7} (E_2 E_6 - E_4))
= \frac{25}{72} (E_2 E_6 - E_4^2)^2 + \frac{35}{72} E_4 (E_2 E_4 - E_6)^2.
Hence
\partial_{12}\partial_{10}F - \frac{5}{6} E_4 F
= \frac{25}{72} ((E_2 E_6 - E_4^2)^2 - E_4 (E_2 E_4 - E_6)^2)
= \frac{25}{72} (- E_2^2 E_4^3 + E_2^2 E_6^2 + E_4^4 - E_4 E_6^3)
= -\frac{25}{72} (E_4^3 - E_6^2) (E_2^2 - E_4)
= 7200 \Delta (-E_2').
This proves equation eqn:ddf. The second is proved similarly,
using Proposition Lemma 7.50 and
Lemma Lemma 7.41.
- No associated Lean code or declarations.
Equation eqn:ddf is positive and equation eqn:ddg is negative on the
positive imaginary axis.
From equation eqn:E2 in Definition 7.17 and
Lemma Corollary 7.25,
7200 (-E_2'(it)) \Delta(it) = 7200 \cdot 24 \left(\sum_{n \ge 1} n \sigma_1(n) e^{-2 \pi n t}\right) \cdot \Delta(it) > 0.
Negativity of equation eqn:ddg, namely
-640 \Delta(it) H_2(it) < 0, follows from
Corollary 7.42 and Corollary 7.25.
The second inequality follows from the following two observations.
Since G(it) > 0 for all t > 0, we can define the quotient
Q(t) := \frac{F(it)}{G(it)}
as a function on (0, \infty).
-
FmodG_rightLimitAt_zero[sorry in proof]
We have
\lim_{t \to 0^+} Q(t) = \frac{18}{\pi^2}.
Lean code for Lemma9.10●1 theorem, incomplete
Associated Lean declarations
-
FmodG_rightLimitAt_zero[sorry in proof]
-
FmodG_rightLimitAt_zero[sorry in proof]
-
theoremdefined in SpherePacking/ModularForms/FG.leancontains sorry
theorem FmodG_rightLimitAt_zero : Filter.Tendsto FmodGReal (nhdsWithin 0 (Set.Ioi 0)) (nhdsWithin (18 * Real.pi ^ (-2)) Set.univ)
theorem FmodG_rightLimitAt_zero : Filter.Tendsto FmodGReal (nhdsWithin 0 (Set.Ioi 0)) (nhdsWithin (18 * Real.pi ^ (-2)) Set.univ)
$\lim_{t \to 0^+} F(it) / G(it) = 18 / \pi^2$.
We have
\lim_{t \to 0^+} Q(t) = \lim_{t \to 0^+} \frac{F(it)}{G(it)} = \lim_{t \to \infty} \frac{F(i/t)}{G(i/t)}.
By using the transformation laws of Eisenstein series eqn:E2-S-transform,
eqn:Ek-trans-S (for k = 4, 6) and the thetanull functions,
eqn:H2-transform-S, eqn:H4-transform-S, we get
F\left(\frac{i}{t}\right) = t^{12} F(it) - \frac{12t^{11}}{\pi} (E_2(it)E_4(it) - E_6(it))E_4(it) + \frac{36t^{10}}{\pi^2}E_4(it)^2
and
G\left(\frac{i}{t}\right) = t^{10} H_{4}(it)^{3}(2H_{4}(it)^{2} + 5 H_{4}(it)H_{2}(it) + 5 H_{2}(it)^{2}).
Since F, E_2 E_4 - E_6 and H_2 are cusp forms, we have
\lim_{t \to \infty}t^k A(it) = 0 when A(z) is one of these forms and
k \geq 0.
From \lim_{t \to \infty} E_4(it) = 1 = \lim_{t \to \infty}H_{4}(it), we get
\begin{aligned}
\lim_{t \to \infty} \frac{F(i/t)}{G(i/t)}
&= \lim_{t \to \infty} \frac{t^{12} F(it) - \frac{12t^{11}}{\pi} (E_2(it)E_4(it) - E_6(it))E_4(it) + \frac{36t^{10}}{\pi^2}E_4(it)^2}{t^{10} H_{4}(it)^{3}(2H_{4}(it)^{2} + 5 H_{4}(it)H_{2}(it) + 5 H_{2}(it)^{2})} \\
&= \lim_{t \to \infty} \frac{t^{2} F(it) - \frac{12t}{\pi} (E_2(it)E_4(it) - E_6(it))E_4(it) + \frac{36}{\pi^2}E_4(it)^2}{H_{4}(it)^{3}(2H_{4}(it)^{2} + 5 H_{4}(it)H_{2}(it) + 5 H_{2}(it)^{2})} \\
&= \frac{18}{\pi^2}.
\end{aligned}
Let F be a quasimodular form whose vanishing order at infinity is
n_0 > 0, that is,
F(z) = \sum_{n \geq n_0} a_n q^{n} with a_{n_0} \ne 0. Then
\lim_{t \to \infty} \frac{F'(it)}{F(it)} = n_0.
By Lemma Lemma 7.44,
\lim_{t \to \infty} \frac{F'(it)}{F(it)}
= \lim_{t \to \infty} \frac{\sum_{n \ge n_0} n a_n e^{-2 \pi n t}}{\sum_{n \ge n_0} a_n e^{-2 \pi n t}}
= \lim_{t \to \infty} \frac{n_0 a_{n_0} e^{-2 \pi n_0 t} + O(e^{-2 \pi (n_0 + 1) t})}{a_{n_0} e^{-2 \pi n_0 t} + O(e^{-2 \pi (n_0 + 1) t})}
= n_0.
-
FmodG_strictAntiOn[complete]
The function t \mapsto Q(t) is strictly decreasing.
Lean code for Lemma9.12●1 theorem
Associated Lean declarations
-
FmodG_strictAntiOn[complete]
-
FmodG_strictAntiOn[complete]
-
theoremdefined in SpherePacking/ModularForms/FG.leancomplete
theorem FmodG_strictAntiOn : StrictAntiOn FmodGReal (Set.Ioi 0)
theorem FmodG_strictAntiOn : StrictAntiOn FmodGReal (Set.Ioi 0)
**Proposition 8.12**: `FmodGReal` is strictly decreasing on `(0, ∞)`.
It is enough to show that
\frac{\dd}{\dd t} \left(\frac{F(it)}{G(it)}\right) < 0
\Leftrightarrow (- 2\pi) \frac{F'(it)G(it) - F(it) G'(it)}{G(it)^{2}} < 0
\Leftrightarrow F'(it) G(it) - F(it) G'(it) > 0
\Leftrightarrow (\partial_{10}F)(it) G(it) - F(it) (\partial_{10}G)(it) > 0.
Let \mathcal{L}_{1, 0} := (\partial_{10}F) G - F (\partial_{10} G) = F'G - FG'.
Then
\frac{\mathcal{L}_{1, 0}}{FG} = \frac{F'G - FG'}{FG} = \frac{F'}{F} - \frac{G'}{G}.
The vanishing orders of F and G at infinity are 2 and
\frac{3}{2} respectively, so by
Lemma 9.11 we have
\lim_{t \to \infty} \frac{\mathcal{L}_{1, 0}(it)}{F(it) G(it)}
= \lim_{t \to \infty} \left(\frac{F'(it)}{F(it)} - \frac{G'(it)}{G(it)}\right)
= 2 - \frac{3}{2} = \frac{1}{2} > 0,
so \mathcal{L}_{1, 0}(it) > 0 for sufficiently large t.
Its Serre derivative is positive by Corollary 9.9:
\partial_{22} \mathcal{L}_{1, 0} = (\partial_{12} \partial_{10} F) G - F (\partial_{12}\partial_{10} G)
= \Delta (7200 (-E_{2}') G + 640 H_2 F) > 0.
Hence \mathcal{L}_{1, 0}(it) > 0 by
Theorem 7.52, and the
monotonicity follows.
-
FG_inequality_2[sorry in proof]
Equation eqn:ineqBnew holds.
Lean code for Corollary9.13●1 theorem, incomplete
Associated Lean declarations
-
FG_inequality_2[sorry in proof]
-
FG_inequality_2[sorry in proof]
-
theoremdefined in SpherePacking/ModularForms/FG.leancontains sorry
theorem FG_inequality_2 {t : ℝ} (ht : 0 < t) : FReal t - 18 * Real.pi ^ (-2) * GReal t < 0
theorem FG_inequality_2 {t : ℝ} (ht : 0 < t) : FReal t - 18 * Real.pi ^ (-2) * GReal t < 0
We have
\frac{F(it)}{G(it)} = Q(t) < \lim_{u \to 0^+} Q(u) = \frac{18}{\pi^2},
and by Lemma Lemma 9.6 the desired inequality follows.
The function
g(x):=\frac{\pi\,i}{8640}a(x)+\frac{i}{240\pi}\,b(x)
satisfies conditions eqn:g1--eqn:g3.
First, we prove the first sign condition. By Propositions
Lemma 8.14 and Lemma 8.28 we know
that for r>\sqrt{2},
g(r)=\frac{\pi}{2160}\,\sin(\pi r^2/2)^2\,\int\limits_0^\infty A(t)\,e^{-\pi r^2 t}\,dt
where
A(t)=-t^2\phi_0(i/t)-\frac{36}{\pi^2}\,\psi_I(it).
From Proposition Lemma 9.1 we know that A(t)<0 for
t\in(0,\infty), so this identity implies the first sign condition.
/- source paragraph break -/
Next, we prove the Fourier-side sign condition. By Propositions
Lemma 8.15 and
Lemma 8.29 we
know that for r>0,
\widehat{g}(r)=\frac{\pi}{2160}\,\sin(\pi r^2/2)^2\,\int\limits_0^\infty B(t)\,e^{-\pi r^2 t}\,dt
where
B(t)=-t^2\phi_0(i/t)+\frac{36}{\pi^2}\,\psi_I(it).
/- source paragraph break -/
Finally, the normalization g(0)=\widehat g(0)=1 follows readily from
Propositions Lemma 8.16 and Lemma 8.30.
This finishes the proof.