Introduction
The study of moments of the Riemann zeta function has a long history in analytic number theory, dating back to the classical work of Hardy and Littlewood [24 ] . In particular, the Lindelöf hypothesis is equivalent to the assertion that, for every positive integer s s s and every fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 ,
lim T → ∞ 1 T ∫ 1 T ∣ ζ ( σ + i t ) ∣ 2 s d t = ∑ n = 1 ∞ d s ( n ) 2 n 2 σ , (1) \lim_{T \to\infty} \frac{1}{T} \int_{1}^{T} \lvert\zeta(\sigma+ \mathrm{i}t) \rvert^{2s} \,\mathrm{d}t = \sum_{n=1}^{\infty} \frac{d_s(n)^2}{n^{2\sigma}}, \tag*{(1)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ( σ + i t ) ∣ 2 s d t = n = 1 ∑ ∞ n 2 σ d s ( n ) 2 , (1)
where d s ( n ) d_s(n) d s ( n ) denotes the s s s -fold divisor function; see [45 Theorem 13.2] . The case s = 1 s = 1 s = 1 of (1) is known unconditionally by the classical mean-value theorem for Dirichlet series; see also [22 ] . The case s = 2 s = 2 s = 2 was established by Hardy and Littlewood [23 ] . However, for s = 3 s = 3 s = 3 , the identity remains unproved throughout the entire range σ > 1 / 2 \sigma> 1/2 σ > 1/2 . From (1) , we see that, under the Lindelöf hypothesis, for every positive integer s s s ,
lim T → ∞ 1 T ∫ 1 T ∣ ζ ( σ + i t ) ∣ 2 s d t ∼ a s ( 2 σ − 1 ) s 2 , σ ↓ 1 / 2 , (2) \lim_{T \to\infty} \frac{1}{T} \int_{1}^{T} \lvert\zeta(\sigma+ \mathrm{i}t) \rvert^{2s} \,\mathrm{d}t \sim\frac{a_s}{(2\sigma- 1)^{s^2}}, \qquad\sigma\downarrow1/2, \tag*{(2)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ( σ + i t ) ∣ 2 s d t ∼ ( 2 σ − 1 ) s 2 a s , σ ↓ 1/2 , (2)
where a s a_s a s is the arithmetic factor defined by
a s = ∏ p ( 1 − p − 1 ) s 2 ∑ k = 0 ∞ ( ( s ) k k ! ) 2 p − k , (3) a_s = \prod_p (1-p^{-1})^{s^2} \sum_{k=0}^{\infty} \left(\frac{(s)_k}{k!}\right)^2 p^{-k}, \tag*{(3)} a s = p ∏ ( 1 − p − 1 ) s 2 k = 0 ∑ ∞ ( k ! ( s ) k ) 2 p − k , (3)
with ( s ) k = s ( s + 1 ) ⋯ ( s + k − 1 ) (s)_k = s(s+1)\cdots(s+k-1) ( s ) k = s ( s + 1 ) ⋯ ( s + k − 1 ) for k ≥ 1 k \ge1 k ≥ 1 and ( s ) 0 = 1 (s)_0 = 1 ( s ) 0 = 1 denoting the rising factorial. In particular, the asymptotic formula (2) holds unconditionally for s = 1 s = 1 s = 1 and s = 2 s = 2 s = 2 .
Recently, Simm and Wei [42 ] studied the analogous problem for the moments of the derivative of the Riemann zeta function, ζ ′ \zeta' ζ ′ . More precisely, they proposed the following conjecture, analogous to (2) , concerning the asymptotic behaviour of the moments of ζ ′ ( σ + i t ) \zeta'(\sigma+ \mathrm{i}t) ζ ′ ( σ + i t ) as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
Conjecture 1.1. For every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 , one has
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t ∼ a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) ( 2 σ − 1 ) s 2 + 2 s , σ ↓ 1 / 2 , (4) \lim_{T\to\infty}\frac{1}{T}\int_{1}^{T}\left|\zeta'(\sigma+it)\right|^{2s}\,\mathrm{d}t \sim a_s\frac{\mathrm{e}^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2)}{(2\sigma-1)^{s^2+2s}},\qquad\sigma\downarrow1/2, \tag*{(4)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t ∼ a s ( 2 σ − 1 ) s 2 + 2 s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , σ ↓ 1/2 , (4) where a s a_s a s is the arithmetic factor defined by (3) .
Compared with (2) , this conjecture predicts that differentiation changes the exponent of ( 2 σ − 1 ) − 1 (2\sigma-1)^{-1} ( 2 σ − 1 ) − 1 from s 2 s^2 s 2 to s 2 + 2 s s^2+2s s 2 + 2 s and introduces an additional factor involving the confluent hypergeometric function arising from random matrix theory. The conjecture is known to hold for all positive integers s s s . The case s = 1 s=1 s = 1 follows from the mean-value theorem for Dirichlet series, while the case s = 2 s=2 s = 2 was proved unconditionally in [42 Theorem 1.7] . For every integer s ≥ 3 s\ge3 s ≥ 3 , it was established under the Lindelöf hypothesis in [42 Theorem 1.8] . A difficulty in studying these derivative moments is that differentiation introduces logarithmic factors into the coefficients of the associated Dirichlet series, which are no longer multiplicative. Consequently, the Euler product approach used for the moments of the Riemann zeta function cannot be applied directly.
The conjecture above is motivated by the work of Simm and Wei [42 ] , who studied complex moments of the derivatives of CUE characteristic polynomials at points inside the unit disc. By analyzing the asymptotic behaviour of these moments as the spectral parameter approaches the unit circle, they obtained an asymptotic formula with the same leading order s 2 + 2 s s^2+2s s 2 + 2 s as in Conjecture 1.1 . This suggests a corresponding asymptotic formula for the moments of the derivative of the Riemann zeta function, with an additional arithmetic factor a s a_s a s . The idea of using characteristic polynomials of random unitary matrices to model moments of the Riemann zeta function goes back to Keating and Snaith [32 ] , who proposed conjectures for the moments of the Riemann zeta function on the critical line based on calculations in the CUE. This approach was subsequently extended to moments of the derivative of the Riemann zeta function; see, for e.g., [5 , 7 , 11 , 14 , 16 , 19 , 28 , 29 , 46 ] . For related results on moments of higher-order derivatives, see, e.g., [6 , 21 , 33 , 34 ] .
In the present paper, we establish the asymptotic formula predicted by Conjecture 1.1 under the Lindelöf hypothesis for every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 . Moreover, the result holds unconditionally when 0 < Re ( s ) < 2 0<\operatorname{Re}(s)<2 0 < Re ( s ) < 2 . Our approach is based on probabilistic models of the Riemann zeta function. The classical work of Bohr and Jessen [12 , 13 ] established the existence of limiting value distributions for log ζ ( σ + i t ) \log\zeta(\sigma+it) log ζ ( σ + i t ) for fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 . This result was further developed by Jessen and Wintner [31 ] , who investigated the associated limiting distributions using infinite convolutions and Fourier-analytic methods. Bagchi [10 ] subsequently established a functional limit theorem for the vertical translates ζ ( ⋅ + i t ) \zeta(\cdot+it) ζ ( ⋅ + i t ) , with the limiting distribution described by a random Euler product. The random Euler product arising in this limiting distribution is constructed by replacing the prime-dependent phases p − i t p^{-it} p − i t with independent random variables X ( p ) X(p) X ( p ) uniformly distributed on the unit circle. Extending these variables completely multiplicatively gives a Steinhaus random multiplicative function X ( n ) X(n) X ( n ) and the associated random Euler product
ζ ( w , X ) = ∏ p ( 1 − X ( p ) p − w ) − 1 , Re ( w ) > 1 / 2. \zeta(w,X)=\prod_p(1-X(p)p^{-w})^{-1},\qquad\operatorname{Re}(w)>1/2. ζ ( w , X ) = p ∏ ( 1 − X ( p ) p − w ) − 1 , Re ( w ) > 1/2.
The study of random multiplicative functions has a long history, dating back to the early work of Wintner [47 ] . Various aspects of the partial sums of random multiplicative functions, including their almost sure behavior and moment estimates, have been extensively studied; see, for example, [8 , 9 , 25 , 26 , 38 , 43 ] and the references therein. Random Euler products have also played an important role in the study of value distributions and moments of various L L L -functions; see, for example, [15 , 36 , 37 , 40 ] and the references therein. In the context of the Riemann zeta function, Arguin, Belius, and Harper [2 ] studied the extreme values of a randomized model based on truncated Euler products; see also the subsequent work of Arguin, Bourgade, and Radziwiłł [3 , 4 ] on the Riemann zeta function itself. These developments illustrate the importance of random Euler products in the probabilistic study of the Riemann zeta function and other L L L -functions.
Main results
In the present paper, we investigate the complex moments of the derivative ζ ′ ( σ , X ) \zeta'(\sigma, X) ζ ′ ( σ , X ) of the random Euler product and their asymptotic behaviour as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . For comparison, the corresponding moments without differentiation admit a direct evaluation using the independence of the prime-indexed random variables. Indeed, the Euler product representation implies that, for every s ∈ C s \in\mathbb{C} s ∈ C ,
lim σ ↓ 1 / 2 ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ] = a s , (5) \lim_{\sigma\downarrow1/2} (2\sigma- 1)^{s^2} \mathbb{E}\left[|\zeta(\sigma, X)|^{2s}\right] = a_s, \tag*{(5)} σ ↓ 1/2 lim ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ] = a s , (5)
where a s a_s a s is the arithmetic factor defined in (3) . Here and throughout the paper, E \mathbb{E} E denotes expectation with respect to the probability measure P \mathbb{P} P defined in (24) .
In contrast, differentiation introduces a factor of log n \log n log n into the Dirichlet coefficients, which are no longer multiplicative. Consequently, the moments of ζ ′ ( σ , X ) \zeta'(\sigma, X) ζ ′ ( σ , X ) require a different analysis. Our first main result, proved in Subsection 4.9, determines the limiting behaviour of the normalized complex moments of ζ ′ ( σ , X ) \zeta'(\sigma, X) ζ ′ ( σ , X ) as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
Theorem 1.2. For every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , one has
lim σ ↓ 1 / 2 ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (6) \lim_{\sigma\downarrow1/2} (2\sigma- 1)^{s^2+2s}\mathbb{E}\left[|\zeta'(\sigma, X)|^{2s}\right] = a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2), \tag*{(6)} σ ↓ 1/2 lim ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (6) where a s a_s a s is the arithmetic factor defined in (3) , and 1 F 1 ( a ; b ; z ) {}_1F_1(a;b;z) 1 F 1 ( a ; b ; z ) denotes the confluent hypergeometric function of the first kind. Moreover, the convergence is locally uniform in s s s throughout the half-plane Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 .
We next establish a connection between the complex moments of the derivative of the random Euler product and the corresponding mean values of the derivative of the Riemann zeta function. More precisely, we show that, under the Lindelöf hypothesis, these moments coincide for every fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 . We also establish this identity unconditionally in a restricted range of the complex parameter s s s . The following result is proved in Section 2.1.
Theorem 1.3. Let σ > 1 / 2 \sigma> 1/2 σ > 1/2 be fixed, and let ζ ( w , X ) \zeta(w, X) ζ ( w , X ) be defined by (50) for Re ( w ) > 1 / 2 \operatorname{Re}(w) > 1/2 Re ( w ) > 1/2 . Assume the Lindelöf hypothesis. Then, for every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , one has
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] . (7) \lim_{T \to\infty} \frac{1}{T}\int_1^T |\zeta'(\sigma+ it)|^{2s}\,\mathrm{d}t = \mathbb{E}\left[|\zeta'(\sigma, X)|^{2s}\right]. \tag*{(7)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] . (7) Moreover, (7) holds unconditionally for every s ∈ C s \in\mathbb{C} s ∈ C with 0 < Re ( s ) < 2 0 < \operatorname{Re}(s) < 2 0 < Re ( s ) < 2 .
Combining Theorems 1.2 and 1.3, we obtain the following consequence concerning Conjecture 1.1.
Corollary 1.4. Assume the Lindelöf hypothesis. Then, for every s > 0 s > 0 s > 0 , Conjecture 1.1 holds. Moreover, Conjecture 1.1 holds unconditionally for every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) ∈ ( 0 , 2 ) \operatorname{Re}(s) \in(0,2) Re ( s ) ∈ ( 0 , 2 ) and s = 2 s = 2 s = 2 .
We now outline the main steps in the proof of Theorem 1.2. We first consider real parameters s > 0 s > 0 s > 0 , for which a change-of-measure argument allows us to analyze the moments of the logarithmic derivative of the random Euler product under a tilted probability measure. For complex parameters s s s with Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , we employ Fourier–Riesz methods to establish the corresponding asymptotic formula.
Real moments via a tilted probability
The value distribution of the logarithmic derivative of the Riemann zeta function has been investigated by several authors. In particular, Lester [39 ] established a central limit theorem for ζ ′ / ζ ( σ + i t ) \zeta'/\zeta(\sigma+it) ζ ′ / ζ ( σ + i t ) in a range where σ \sigma σ approaches the critical line from the right. In a related direction, Fazzari [17 , 18 ] investigated weighted central limit theorems for the logarithm of the Riemann zeta function on the critical line, under the Riemann hypothesis.
The logarithmic derivative L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) of the random Euler product provides a probabilistic counterpart to ( ζ ′ / ζ ) ( σ + i t ) (\zeta'/\zeta)(\sigma+it) ( ζ ′ / ζ ) ( σ + i t ) . For every fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 , the limiting value distribution of the latter as T → ∞ T\to\infty T → ∞ is described by the distribution of L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) . In the present work, for real s > 0 s>0 s > 0 , we study L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) under a tilted probability measure obtained by weighting the original probability measure with ∣ ζ ( σ , X ) ∣ 2 s |\zeta(\sigma,X)|^{2s} ∣ ζ ( σ , X ) ∣ 2 s . Under this new measure, the prime-indexed random variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p remain independent, although their marginal distributions are modified. This allows us to apply the multivariate Lindeberg–Feller central limit theorem to establish a Gaussian limit with a nonzero mean arising from the change of measure. More precisely, we obtain the following result.
Proposition 1.5. Let s > 0 s>0 s > 0 be fixed. As σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 , under the tilted probability measure P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s ,
− ( 2 σ − 1 ) L ′ ( σ , X ) ⟶ d s + G , (8) -(2\sigma-1)L'(\sigma,X)\mathrel{\mathop{\longrightarrow}\limits^{\mathrm{d}}}s+G, \tag*{(8)} − ( 2 σ − 1 ) L ′ ( σ , X ) ⟶ d s + G , (8) where L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) is defined by the prime series in (60) , which converges in L 2 ( Ω , P ~ σ , s ) L^2(\Omega,\widetilde{\mathbb{P}}_{\sigma,s}) L 2 ( Ω , P σ , s ) , and G G G is a standard circularly symmetric complex Gaussian random variable satisfying
E [ G ] = 0 , E [ ∣ G ∣ 2 ] = 1 , E [ G 2 ] = 0. \mathbb{E}[G]=0,\qquad\mathbb{E}[|G|^2]=1,\qquad\mathbb{E}[G^2]=0. E [ G ] = 0 , E [ ∣ G ∣ 2 ] = 1 , E [ G 2 ] = 0. Equivalently, G = G 1 + i G 2 G=G_1+iG_2 G = G 1 + i G 2 , where G 1 G_1 G 1 and G 2 G_2 G 2 are independent N ( 0 , 1 / 2 ) N(0,1/2) N ( 0 , 1/2 ) random variables.
Proof of Proposition 1.5 is given in Subsection 3.3. To pass from convergence in distribution to convergence of moments, we establish uniform integrability using Rosenthal’s inequality for sums of independent, centered random variables indexed by the primes. More precisely, for some q > 2 s q>2s q > 2 s and σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 , we prove that
sup 1 / 2 < σ ≤ σ 0 E ~ σ , s [ ∣ ( 2 σ − 1 ) L ′ ( σ , X ) ∣ q ] < ∞ . (9) \sup_{1/2<\sigma\leq\sigma_0}\widetilde{\mathbb{E}}_{\sigma,s}\left[|(2\sigma-1)L'(\sigma,X)|^q\right]<\infty. \tag*{(9)} 1/2 < σ ≤ σ 0 sup E σ , s [ ∣ ( 2 σ − 1 ) L ′ ( σ , X ) ∣ q ] < ∞. (9)
Consequently, the Gaussian limit also holds at the level of moments:
lim σ ↓ 1 / 2 ( 2 σ − 1 ) 2 s E ~ σ , s [ ∣ L ′ ( σ , X ) ∣ 2 s ] = E [ ∣ s + G ∣ 2 s ] . \lim_{\sigma\downarrow1/2}(2\sigma-1)^{2s}\widetilde{\mathbb{E}}_{\sigma,s}\left[|L'(\sigma,X)|^{2s}\right]=\mathbb{E}\left[|s+G|^{2s}\right]. σ ↓ 1/2 lim ( 2 σ − 1 ) 2 s E σ , s [ ∣ L ′ ( σ , X ) ∣ 2 s ] = E [ ∣ s + G ∣ 2 s ] .
Combining this convergence with the asymptotic formula for E [ ∣ ζ ( σ , X ) ∣ 2 s ] \mathbb{E}[|\zeta(\sigma,X)|^{2s}] E [ ∣ ζ ( σ , X ) ∣ 2 s ] , obtained from the Euler product, yields the desired asymptotics for the real moments of ζ ′ ( σ , X ) \zeta'(\sigma,X) ζ ′ ( σ , X ) . The resulting Gaussian factor can be evaluated explicitly in terms of a confluent hypergeometric function:
E [ ∣ s + G ∣ 2 s ] = e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (10) \mathbb{E}\left[|s+G|^{2s}\right]=e^{-s^2}\Gamma(s+1)\,{}_1F_1(s+1;1;s^2), \tag*{(10)} E [ ∣ s + G ∣ 2 s ] = e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (10)
where 1 F 1 {}_1F_1 1 F 1 denotes the confluent hypergeometric function of the first kind. This agrees precisely with the formula in (4) . Both (9) and (10) are established in Proposition 3.2 .
Complex moments via Fourier–Riesz analysis
The preceding argument for real s > 0 s > 0 s > 0 in Subsection 1.2 relies essentially on the positivity of the weight ∣ ζ ( σ , X ) ∣ 2 s |\zeta(\sigma, X)|^{2s} ∣ ζ ( σ , X ) ∣ 2 s , which allows us to introduce a tilted probability measure and apply probabilistic limit theorems. For non-real s s s with Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , the same weighting procedure gives rise to a complex measure rather than a probability measure. Although the probabilistic argument does not extend directly to this setting, the resulting complex measure provides a natural framework for studying complex moments through Fourier analysis and the theory of Riesz distributions.
Weighted complex measures and Fourier asymptotics
Write
T σ = ( 2 σ − 1 ) L ′ ( σ , X ) . (11) T_\sigma= (2\sigma- 1)L'(\sigma, X). \tag*{(11)} T σ = ( 2 σ − 1 ) L ′ ( σ , X ) . (11)
For Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , we introduce the weighted complex Borel measure
ν σ , s ( B ) : = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s 1 { T σ ∈ B } ] , B ∈ B ( R 2 ) . (12) \nu_{\sigma,s}(B) := (2\sigma- 1)^{s^2}\mathbb{E}\left[|\zeta(\sigma, X)|^{2s}\mathbf{1}_{\{T_\sigma\in B\}}\right], \qquad B \in\mathcal{B}(\mathbb{R}^2). \tag*{(12)} ν σ , s ( B ) := ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s 1 { T σ ∈ B } ] , B ∈ B ( R 2 ) . (12)
By the fixed-σ \sigma σ moment estimates established in Lemma 4.2 , this is a finite complex Borel measure. For real s > 0 s > 0 s > 0 , the normalized measure
ν σ , s ν σ , s ( R 2 ) \frac{\nu_{\sigma,s}}{\nu_{\sigma,s}(\mathbb{R}^2)} ν σ , s ( R 2 ) ν σ , s
coincides with the distribution of T σ T_\sigma T σ under the tilted probability measure P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s . For non-real s s s , however, ν σ , s \nu_{\sigma,s} ν σ , s is generally a complex measure and does not admit such a probabilistic interpretation. Instead, the complex measure ν σ , s \nu_{\sigma,s} ν σ , s allows us to express the weighted complex moments as integrals with respect to a finite complex Borel measure. By studying its associated Fourier transform H σ , s H_{\sigma,s} H σ , s , defined in (13) below, we establish the asymptotic behaviour of the weighted joint Fourier transform, together with suitable decay and regularity estimates. These properties allow us to apply the Fourier transform identity for the homogeneous distribution ∣ x ∣ 2 s |x|^{2s} ∣ x ∣ 2 s established in (22) , together with the meromorphic continuation of Riesz distributions, to derive a Fourier–Riesz representation of the complex moments. More precisely, we consider the associated weighted joint Fourier transform
H σ , s ( λ , u ) : = ∫ R 2 e i ( λ x 1 + u x 2 ) d ν σ , s ( x ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s e i λ Re T σ + i u Im T σ ] . (13) \begin{aligned}
H_{\sigma,s}(\lambda,u) &:= \int_{\mathbb{R}^2} e^{\mathrm{i}(\lambda x_1+u x_2)}\,d\nu_{\sigma,s}(x) \\
&= (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}e^{\mathrm{i}\lambda\operatorname{Re}T_\sigma+\mathrm{i}u\operatorname{Im}T_\sigma}\right]. \tag*{(13)}
\end{aligned} H σ , s ( λ , u ) := ∫ R 2 e i ( λ x 1 + u x 2 ) d ν σ , s ( x ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s e i λ Re T σ + i u Im T σ ] . (13)
Our first key result describes the limiting behaviour of this transform as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
Proposition 1.6. As σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 ,
H σ , s ( λ , u ) ⟶ a s exp ( − i s λ − λ 2 + u 2 4 ) (14) H_{\sigma,s}(\lambda,u) \longrightarrow a_s\exp\left(-\mathrm{i}s\lambda-\frac{\lambda^2+u^2}{4}\right) \tag*{(14)} H σ , s ( λ , u ) ⟶ a s exp ( − i s λ − 4 λ 2 + u 2 ) (14) locally uniformly for
Re ( s ) > 0 , ( λ , u ) ∈ R 2 . \operatorname{Re}(s)>0,\qquad(\lambda,u)\in\mathbb{R}^2. Re ( s ) > 0 , ( λ , u ) ∈ R 2 . Here H σ , s ( λ , μ ) H_{\sigma,s}(\lambda,\mu) H σ , s ( λ , μ ) and a s a_s a s are defined in (13) and (3) , respectively. Moreover, on every bounded complex polydisc in ( λ , u ) (\lambda,u) ( λ , u ) , the Euler-product expression for H σ , s H_{\sigma,s} H σ , s admits a holomorphic continuation, and the same convergence holds locally uniformly.
Proof of Proposition 1.6 is given in Subsection 4.3. The limiting function has a Gaussian-type structure. Unlike the real case, however, for non-real s s s it is not the characteristic function of a probability distribution. To use this convergence in the analysis of complex moments, we also require uniform estimates for H σ , s H_{\sigma,s} H σ , s over the entire Fourier space. The following proposition provides the necessary global decay.
Proposition 1.7. Let K K K be a compact subset of the open right half-plane { s ∈ C : Re ( s ) > 0 } \{s \in\mathbb{C} : \operatorname{Re}(s) > 0\} { s ∈ C : Re ( s ) > 0 } . For every N > 0 N > 0 N > 0 , there exist σ 0 > 1 / 2 \sigma_0 > 1/2 σ 0 > 1/2 and C K , N > 0 C_{K,N} > 0 C K , N > 0 such that
∣ H σ , s ( λ , u ) ∣ ≤ C K , N ( 1 + λ 2 + u 2 ) − N (15) \lvert H_{\sigma,s}(\lambda,u)\rvert\le C_{K,N}(1+\lambda^2+u^2)^{-N} \tag*{(15)} ∣ H σ , s ( λ , u )∣ ≤ C K , N ( 1 + λ 2 + u 2 ) − N (15) for all
1 2 < σ ≤ σ 0 , s ∈ K , ( λ , u ) ∈ R 2 . \frac{1}{2}<\sigma\le\sigma_0,\qquad s\in K,\qquad(\lambda,u)\in\mathbb{R}^2. 2 1 < σ ≤ σ 0 , s ∈ K , ( λ , u ) ∈ R 2 . Here H σ , s ( λ , u ) H_{\sigma,s}(\lambda,u) H σ , s ( λ , u ) is defined by (13) .
Proof of Proposition 1.7 is given in Subsection 4.4.1. Together with the local derivative estimates established in Lemma 4.10, these results provide the analytic control needed to establish a Fourier–Riesz representation of the complex moments.
Removal of Gaussian regularization via Riesz distributions
To recover the complex moments from H σ , s H_{\sigma,s} H σ , s , we first introduce, for t > 0 t>0 t > 0 , the Gaussian-regularized moment
M σ , t ( s ) : = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] . (16) \mathcal{M}_{\sigma,t}(s):=(2\sigma-1)^{s^2}\mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{2s}\lvert T_\sigma\rvert^{2s}e^{-t\lvert T_\sigma\rvert^2}\right]. \tag*{(16)} M σ , t ( s ) := ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] . (16)
By Lemma 4.12, this quantity admits a Fourier representation in terms of H σ , s H_{\sigma,s} H σ , s . For each fixed t > 0 t>0 t > 0 , Lemma 4.13 establishes the convergence of M σ , t ( s ) \mathcal{M}_{\sigma,t}(s) M σ , t ( s ) as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . Furthermore, Lemma 4.14 gives the explicit formula
M t ( s ) : = lim σ ↓ 1 / 2 M σ , t ( s ) = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 1 F 1 ( s + 1 ; 1 ; s 2 1 + t ) . (17) \mathcal{M}_t(s):=\lim_{\sigma\downarrow1/2}\mathcal{M}_{\sigma,t}(s)=a_s e^{-s^2}\Gamma(s+1)(1+t)^{-s-1}{}_1F_1\left(s+1;1;\frac{s^2}{1+t}\right). \tag*{(17)} M t ( s ) := σ ↓ 1/2 lim M σ , t ( s ) = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 1 F 1 ( s + 1 ; 1 ; 1 + t s 2 ) . (17)
We now explain how the Gaussian-regularized moments are related to the complex moments in Theorem 1.2. Since ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) \zeta'(\sigma,X)=\zeta(\sigma,X)L'(\sigma,X) ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) and T σ = ( 2 σ − 1 ) L ′ ( σ , X ) T_\sigma=(2\sigma-1)L'(\sigma,X) T σ = ( 2 σ − 1 ) L ′ ( σ , X ) , we have
( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] = : M σ , 0 ( s ) . (18) (2\sigma-1)^{s^2+2s}\mathbb{E}\left[\lvert\zeta'(\sigma,X)\rvert^{2s}\right]=(2\sigma-1)^{s^2}\mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{2s}\lvert T_\sigma\rvert^{2s}\right]=:\mathcal{M}_{\sigma,0}(s). \tag*{(18)} ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] =: M σ , 0 ( s ) . (18)
Thus, the quantity appearing on the left-hand side of our main theorem is precisely the unregularized moment M σ , 0 ( s ) \mathcal{M}_{\sigma,0}(s) M σ , 0 ( s ) . On the other hand, letting t ↓ 0 t\downarrow0 t ↓ 0 in (17) gives
lim t ↓ 0 lim σ ↓ 1 / 2 M σ , t ( s ) = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (19) \lim_{t\downarrow0}\lim_{\sigma\downarrow1/2}\mathcal{M}_{\sigma,t}(s)=a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2), \tag*{(19)} t ↓ 0 lim σ ↓ 1/2 lim M σ , t ( s ) = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (19)
which agrees with the right-hand side of Theorem 1.2. To establish Theorem 1.2, however, we need to prove that
lim σ ↓ 1 / 2 M σ , 0 ( s ) = lim σ ↓ 1 / 2 lim t ↓ 0 M σ , t ( s ) \lim_{\sigma\downarrow1/2}\mathcal{M}_{\sigma,0}(s)=\lim_{\sigma\downarrow1/2}\lim_{t\downarrow0}\mathcal{M}_{\sigma,t}(s) σ ↓ 1/2 lim M σ , 0 ( s ) = σ ↓ 1/2 lim t ↓ 0 lim M σ , t ( s )
exists and equals the expression in (19) . The main difficulty is therefore to remove the Gaussian regularization while controlling the limit as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
For real s > 0 s>0 s > 0 , the uniform integrability established in (9) provides the necessary control and permits the interchange of the limits t ↓ 0 t\downarrow0 t ↓ 0 and σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . For non-real s s s , however, the same argument does not apply directly. Writing s = a + i b s=a+ib s = a + ib with a > 0 a>0 a > 0 , we have
∣ ( 2 σ − 1 ) s 2 ∣ = ( 2 σ − 1 ) a 2 − b 2 . \left|(2\sigma-1)^{s^2}\right|=(2\sigma-1)^{a^2-b^2}. ( 2 σ − 1 ) s 2 = ( 2 σ − 1 ) a 2 − b 2 .
Taking absolute values in the difference between the regularized and unregularized moments introduces an additional factor ( 2 σ − 1 ) − b 2 (2\sigma-1)^{-b^2} ( 2 σ − 1 ) − b 2 compared with the corresponding estimates for real parameters. Consequently, for b ≠ 0 b\ne0 b = 0 , the available uniform moment bounds do not directly justify the interchange of these limits.
To overcome this difficulty, we use a Fourier–Riesz approach that avoids interchanging the two limits. We first fix σ > 1 / 2 \sigma>1/2 σ > 1/2 and remove the Gaussian regularization by letting t ↓ 0 t\downarrow0 t ↓ 0 . We then represent the resulting complex moment in terms of a regularized Riesz functional of H σ , s H_{\sigma,s} H σ , s . Finally, the Fourier asymptotics and decay estimates established above allow us to pass to the limit σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 in this representation. More precisely, we introduce the circular average and a regularized Riesz functional arising from the classical theory of homogeneous distributions.
The circular average is the two-dimensional specialization of the spherical mean used by Gel’fand and Shilov in their study of the homogeneous distribution r λ r^\lambda r λ ; see [20 Chapter I, Section 3.9, Eq. (2), p. 71] . For a Schwartz function H ∈ S ( R 2 ) H\in\mathcal{S}(\mathbb{R}^2) H ∈ S ( R 2 ) , define its circular average by
A H ( r ) : = 1 2 π ∫ 0 2 π H ( r cos θ , r sin θ ) d θ , r ≥ 0. (20) AH(r):=\frac{1}{2\pi}\int_{0}^{2\pi}H(r\cos\theta,r\sin\theta)\,\mathrm{d}\theta,\qquad r\ge0. \tag*{(20)} A H ( r ) := 2 π 1 ∫ 0 2 π H ( r cos θ , r sin θ ) d θ , r ≥ 0. (20)
To handle singular radial integrals at the origin, we use the analytic continuation method based on Taylor subtraction, developed in [20 Chapter I, Section 3.8, Eq. (2), pp. 66–67] , together with the expansion of spherical means given by Pizzetti’s formula; see [20 Chapter I, Section 3.9, Eq. (7), p. 74] . Let
Δ : = ∂ 2 ∂ λ 2 + ∂ 2 ∂ u 2 \Delta:=\frac{\partial^2}{\partial\lambda^2}+\frac{\partial^2}{\partial u^2} Δ := ∂ λ 2 ∂ 2 + ∂ u 2 ∂ 2
denote the Laplacian in the Fourier variables ( λ , u ) (\lambda,u) ( λ , u ) . For H ∈ S ( R 2 ) H\in\mathcal{S}(\mathbb{R}^2) H ∈ S ( R 2 ) , M ∈ Z ≥ 0 M\in\mathbb{Z}_{\ge0} M ∈ Z ≥ 0 , and
− 1 < Re ( w ) < M + 1 , w ∉ { 0 , 1 , … , M } , -1<\operatorname{Re}(w)<M+1,\qquad w\notin\{0,1,\ldots,M\}, − 1 < Re ( w ) < M + 1 , w ∈ / { 0 , 1 , … , M } ,
we define the regularized Riesz functional
R M , w [ H ] : = ∫ 0 1 [ A H ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H ( 0 ) ] r − 2 w − 1 d r + ∫ 1 ∞ A H ( r ) r − 2 w − 1 d r + ∑ j = 0 M Δ j H ( 0 ) 4 j ( j ! ) 2 ( 2 j − 2 w ) . (21) \mathcal{R}_{M,w}[H]:=\int_{0}^{1}\left[AH(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^j(j!)^2}\Delta^jH(0)\right]r^{-2w-1}\,\mathrm{d}r+\int_{1}^{\infty}AH(r)r^{-2w-1}\,\mathrm{d}r+\sum_{j=0}^{M}\frac{\Delta^jH(0)}{4^j(j!)^2(2j-2w)}. \tag*{(21)} R M , w [ H ] := ∫ 0 1 [ A H ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H ( 0 ) ] r − 2 w − 1 d r + ∫ 1 ∞ A H ( r ) r − 2 w − 1 d r + j = 0 ∑ M 4 j ( j ! ) 2 ( 2 j − 2 w ) Δ j H ( 0 ) . (21)
Here 0 0 0 denotes the origin in Fourier space R 2 \mathbb{R}^2 R 2 . The Taylor subtraction removes the singularity at the origin, while the finite correction terms provide the meromorphic continuation of the radial integral. Thus, this functional gives an explicit two-dimensional realization of the regularization procedure for homogeneous distributions described in [20 Chapter I, Sections 3.8–3.9] .
The Fourier–Riesz representation of the complex moments is based on the classical Fourier transform formula for the homogeneous distribution ∣ x ∣ 2 w |x|^{2w} ∣ x ∣ 2 w ; see [20 Chapter II, Section 3.3] . Under the Fourier transform convention
f ^ ( ξ ) : = ∫ R 2 e − i ⟨ x , ξ ⟩ f ( x ) d x , \widehat{f}(\xi):=\int_{\mathbb{R}^2}e^{-i\langle x,\xi\rangle}f(x)\,\mathrm{d}x, f ( ξ ) := ∫ R 2 e − i ⟨ x , ξ ⟩ f ( x ) d x ,
this formula takes the form
∣ x ∣ 2 w ^ ( ξ ) = 2 2 w + 2 π Γ ( w + 1 ) Γ ( − w ) Reg ∣ ξ ∣ − 2 w − 2 , (22) \widehat{|x|^{2w}}(\xi)=2^{2w+2}\pi\frac{\Gamma(w+1)}{\Gamma(-w)}\operatorname{Reg}|\xi|^{-2w-2}, \tag*{(22)} ∣ x ∣ 2 w ( ξ ) = 2 2 w + 2 π Γ ( − w ) Γ ( w + 1 ) Reg ∣ ξ ∣ − 2 w − 2 , (22)
in the sense of tempered distributions, where the right-hand side is understood through meromorphic continuation. For Schwartz functions H H H , the action of the regularized distribution is given by
⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ = 2 π R M , w [ H ] . \left\langle\operatorname{Reg}|\xi|^{-2w-2},H\right\rangle=2\pi\mathcal{R}_{M,w}[H]. ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ = 2 π R M , w [ H ] .
The Fourier–Riesz identity (22) provides the basis for expressing the original complex moments in terms of the regularized Riesz functional R M , s [ H σ , s ] \mathcal{R}_{M,s}[H_{\sigma,s}] R M , s [ H σ , s ] . Although the Fourier–Riesz identity is initially formulated for Schwartz functions, we will extend the corresponding representation to a broader class of smooth and integrable functions, allowing us to apply it to the weighted Fourier transform H σ , s H_{\sigma,s} H σ , s .
The following lemma establishes an exact Fourier–Riesz representation of the complex moments after removing the Gaussian regularization for fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 .
Lemma 1.8 (Removal of Gaussian regularization). Let M ∈ Z ≥ 0 M\in\mathbb{Z}_{\geq0} M ∈ Z ≥ 0 , and let K K K be a compact subset of the open right half-plane { s ∈ C : Re ( s ) > 0 } \{s\in\mathbb{C}:\operatorname{Re}(s)>0\} { s ∈ C : Re ( s ) > 0 } . There exists σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 such that, for every
1 2 < σ ≤ σ 0 , s ∈ K ∩ Ω M , \frac{1}{2}<\sigma\leq\sigma_0,\qquad s\in K\cap\Omega_M, 2 1 < σ ≤ σ 0 , s ∈ K ∩ Ω M , where
Ω M : = { w ∈ C : − 1 < Re ( w ) < M + 1 } ∖ { 0 , 1 , … , M } , \Omega_M:=\{w\in\mathbb{C}:-1<\operatorname{Re}(w)<M+1\}\setminus\{0,1,\ldots,M\}, Ω M := { w ∈ C : − 1 < Re ( w ) < M + 1 } ∖ { 0 , 1 , … , M } , one has
lim t ↓ 0 M σ , t ( s ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] = 2 2 s + 1 Γ ( s + 1 ) Γ ( − s ) R M , s [ H σ , s ] . (23) \lim_{t\downarrow0}\mathcal{M}_{\sigma,t}(s)=(2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2s}\right]=2^{2s+1}\frac{\Gamma(s+1)}{\Gamma(-s)}\mathcal{R}_{M,s}[H_{\sigma,s}]. \tag*{(23)} t ↓ 0 lim M σ , t ( s ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] = 2 2 s + 1 Γ ( − s ) Γ ( s + 1 ) R M , s [ H σ , s ] . (23)
Here M σ , t ( s ) \mathcal{M}_{\sigma,t}(s) M σ , t ( s ) , T σ T_\sigma T σ , H σ , s H_{\sigma,s} H σ , s , and R M , s [ H σ , s ] \mathcal{R}_{M,s}[H_{\sigma,s}] R M , s [ H σ , s ] are defined in (16) , (11) , (13) , and (21) , respectively.
Lemma 1.8 is proved in Subsection 4.7.1. The first equality follows from dominated convergence for fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 , while the second is obtained using the Fourier transform identity for homogeneous distributions and the analytic continuation of Riesz distributions. In particular, Lemma 1.8 expresses the original complex moments exactly in terms of the regularized Riesz functional R M , s [ H σ , s ] \mathcal{R}_{M,s}[H_{\sigma,s}] R M , s [ H σ , s ] for each fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 . This allows us to study their asymptotic behaviour as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 through the convergence of the corresponding Riesz functionals, without having to justify the interchange of the limits t ↓ 0 t\downarrow0 t ↓ 0 and σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . We then study the limit σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 through this Fourier–Riesz representation. Lemma 4.20 establishes that
R M , s [ H σ , s ] ⟶ R M , s [ H s ] , \mathcal{R}_{M,s}[H_{\sigma,s}]\longrightarrow\mathcal{R}_{M,s}[H_s], R M , s [ H σ , s ] ⟶ R M , s [ H s ] ,
locally uniformly away from the positive integers. Here H s H_s H s denotes the limiting function of H σ , s H_{\sigma,s} H σ , s as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 , given explicitly in (14) . Finally, Lemma 4.21 gives the explicit evaluation
2 2 s + 1 Γ ( s + 1 ) Γ ( − s ) R M , s [ H s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . 2^{2s+1}\frac{\Gamma(s+1)}{\Gamma(-s)}\mathcal{R}_{M,s}[H_s]=a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2). 2 2 s + 1 Γ ( − s ) Γ ( s + 1 ) R M , s [ H s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) .
The positive integer values of s s s are then treated by holomorphicity and the maximum modulus principle, completing the proof of Theorem 1.2 .
Use of AI. ChatGPT was used to assist in improving the clarity and presentation of the manuscript, including reviewing the presentation of mathematical arguments, checking references, and identifying typographical errors.
A probabilistic model for the derivative of the Riemann zeta function
In this section, we construct a probabilistic model for the derivative of the Riemann zeta function using independent random variables indexed by the primes. We then establish the corresponding limiting distribution and moment identities, and identify the random model with the derivative of a random Euler product.
Construction of the random model
The classical Bohr–Jessen theorem describes the limiting value distribution of log ζ ( σ + i t ) \log\zeta(\sigma+ it) log ζ ( σ + i t ) for fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 . A natural probabilistic model for this limiting distribution is obtained by replacing the phases p − i t p^{-it} p − i t associated with distinct primes by independent random variables on the unit circle. More precisely, let
Ω : = ∏ p T , T : = { z ∈ C : ∣ z ∣ = 1 } , \Omega:= \prod_{p} \mathbb{T}, \qquad\mathbb{T} := \{z \in\mathbb{C} : |z| = 1\}, Ω := p ∏ T , T := { z ∈ C : ∣ z ∣ = 1 } ,
and equip Ω \Omega Ω with the product topology and the corresponding Borel σ \sigma σ -algebra
F : = ⨂ p B ( T ) , \mathcal{F} := \bigotimes_{p} \mathcal{B}(\mathbb{T}), F := p ⨂ B ( T ) ,
where B ( T ) \mathcal{B}(\mathbb{T}) B ( T ) denotes the Borel σ \sigma σ -algebra on T \mathbb{T} T . We endow ( Ω , F ) (\Omega,\mathcal{F}) ( Ω , F ) with the probability measure
P : = ⨂ p m T , (24) \mathbb{P} := \bigotimes_{p} m_{\mathbb{T}}, \tag*{(24)} P := p ⨂ m T , (24)
where m T m_{\mathbb{T}} m T denotes the normalized Haar measure on T \mathbb{T} T . Then Ω \Omega Ω is a compact abelian group and P \mathbb{P} P is its normalized Haar probability measure. Recall that E \mathbb{E} E denotes expectation with respect to P \mathbb{P} P . For each prime p p p , define
X ( p ) ( ω ) : = ω p , ω = ( ω p ) p ∈ Ω . X(p)(\omega) := \omega_p, \qquad\omega= (\omega_p)_p \in\Omega. X ( p ) ( ω ) := ω p , ω = ( ω p ) p ∈ Ω.
Then the random variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p are independent and Haar-uniform on T \mathbb{T} T . We extend X X X completely multiplicatively to N \mathbb{N} N by
X ( n ) : = ∏ p ν ∥ n X ( p ) ν . (25) X(n) := \prod_{p^\nu\parallel n} X(p)^\nu. \tag*{(25)} X ( n ) := p ν ∥ n ∏ X ( p ) ν . (25)
For fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 , we define
Z ( σ , X ) = lim N → ∞ ( − ∑ n ≤ N log n n σ X ( n ) ) in L 2 ( Ω , P ) . (26) Z(\sigma, X) = \lim_{N \to\infty} \left(-\sum_{n \leq N} \frac{\log n}{n^\sigma}X(n)\right) \quad\text{in } L^2(\Omega,\mathbb{P}). \tag*{(26)} Z ( σ , X ) = N → ∞ lim ( − n ≤ N ∑ n σ log n X ( n ) ) in L 2 ( Ω , P ) . (26)
This limit is well-defined since
E [ ∣ ∑ M < n ≤ N log n n σ X ( n ) ∣ 2 ] = ∑ M < n ≤ N ( log n ) 2 n 2 σ ≤ ∑ n > M ( log n ) 2 n 2 σ ⟶ 0 \mathbb{E}\left[\left|\sum_{M<n\leq N}\frac{\log n}{n^\sigma}X(n)\right|^2\right]
= \sum_{M<n\leq N}\frac{(\log n)^2}{n^{2\sigma}}
\leq\sum_{n>M}\frac{(\log n)^2}{n^{2\sigma}} \longrightarrow0 E M < n ≤ N ∑ n σ log n X ( n ) 2 = M < n ≤ N ∑ n 2 σ ( log n ) 2 ≤ n > M ∑ n 2 σ ( log n ) 2 ⟶ 0
as M → ∞ M \to\infty M → ∞ , uniformly for $N > M.
Limiting value distribution and moment identity at fixed σ \sigma σ
We shall show below that Z ( σ , X ) Z(\sigma, X) Z ( σ , X ) is the limiting random variable associated with ζ ′ ( σ + i t ) \zeta'(\sigma+ it) ζ ′ ( σ + i t ) , and that, under the Lindelöf hypothesis, the corresponding complex moments converge.
Proposition 2.1. Assume the Lindelöf hypothesis and let σ > 1 / 2 \sigma> 1/2 σ > 1/2 be fixed. Let Z ( σ , X ) Z(\sigma, X) Z ( σ , X ) be defined as in (26) . Then, for every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 ,
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ Z ( σ , X ) ∣ 2 s ] . (27) \lim_{T \to\infty} \frac{1}{T}\int_{1}^{T} \left|\zeta'(\sigma+ it)\right|^{2s}\,dt = \mathbb{E}\left[\left|Z(\sigma, X)\right|^{2s}\right]. \tag*{(27)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ Z ( σ , X ) ∣ 2 s ] . (27) In particular, the limit on the left-hand side exists and is finite. Here, for z ≠ 0 z \ne0 z = 0 ,
∣ z ∣ 2 s : = exp ( 2 s log ∣ z ∣ ) , |z|^{2s} := \exp(2s\log|z|), ∣ z ∣ 2 s := exp ( 2 s log ∣ z ∣ ) , and we set 0 2 s : = 0 0^{2s} := 0 0 2 s := 0 .
Proof. We first establish the limiting distribution of ζ ′ ( σ + i t ) \zeta'(\sigma+ it) ζ ′ ( σ + i t ) . For every fixed integer n ≥ 1 n \ge1 n ≥ 1 , we claim that
lim T → ∞ 1 T ∫ 1 T ζ ′ ( σ + i t ) n i t d t = 1 n σ . (28) \lim_{T \to\infty} \frac{1}{T}\int_{1}^{T} \zeta'(\sigma+ it)n^{it}\,dt = \frac{1}{n^\sigma}. \tag*{(28)} T → ∞ lim T 1 ∫ 1 T ζ ′ ( σ + i t ) n i t d t = n σ 1 . (28)
If σ > 1 \sigma> 1 σ > 1 , this follows immediately from the absolutely convergent Dirichlet series
ζ ( σ + i t ) = ∑ m = 1 ∞ m − σ − i t . \zeta(\sigma+ it) = \sum_{m=1}^{\infty} m^{-\sigma-it}. ζ ( σ + i t ) = m = 1 ∑ ∞ m − σ − i t .
We therefore only need to consider 1 / 2 < σ ≤ 1 1/2 < \sigma\le1 1/2 < σ ≤ 1 . Applying Cauchy’s theorem to
F ( w ) : = ζ ( w ) n w F(w) := \zeta(w)n^w F ( w ) := ζ ( w ) n w
on the rectangle with vertices
σ + i , 2 + i , 2 + i T , σ + i T , \sigma+ i,\quad2 + i,\quad2 + iT,\quad\sigma+ iT, σ + i , 2 + i , 2 + i T , σ + i T ,
and using the standard estimate (see e.g., [45 Chapter V, section 5.1] ),
sup σ ≤ u ≤ 2 ∣ ζ ( u + i T ) ∣ = o ( T ) , T → ∞ , (29) \sup_{\sigma\le u \le2} |\zeta(u + iT)| = o(T),\qquad T \to\infty, \tag*{(29)} σ ≤ u ≤ 2 sup ∣ ζ ( u + i T ) ∣ = o ( T ) , T → ∞ , (29)
we may shift the vertical line of integration from Re ( w ) = σ \operatorname{Re}(w) = \sigma Re ( w ) = σ to Re ( w ) = 2 \operatorname{Re}(w) = 2 Re ( w ) = 2 . The two horizontal integrals, after division by T T T , tend to zero, and hence
1 T ∫ 1 T ζ ′ ( σ + i t ) n i t d t − n 2 − σ 1 T ∫ 1 T ζ ′ ( 2 + i t ) n i t d t ⟶ 0. (30) \frac{1}{T}\int_{1}^{T} \zeta'(\sigma+ it)n^{it}\,dt - n^{2-\sigma}\frac{1}{T}\int_{1}^{T} \zeta'(2 + it)n^{it}\,dt \longrightarrow0. \tag*{(30)} T 1 ∫ 1 T ζ ′ ( σ + i t ) n i t d t − n 2 − σ T 1 ∫ 1 T ζ ′ ( 2 + i t ) n i t d t ⟶ 0. (30)
The argument above on the line Re ( w ) = 2 \operatorname{Re}(w) = 2 Re ( w ) = 2 gives
lim T → ∞ n 2 − σ ∫ 1 T ζ ′ ( 2 + i t ) n i t d t = n − σ . \lim_{T \to\infty} n^{2-\sigma}\int_{1}^{T} \zeta'(2 + it)n^{it}\,dt = n^{-\sigma}. T → ∞ lim n 2 − σ ∫ 1 T ζ ′ ( 2 + i t ) n i t d t = n − σ .
and therefore (28) holds for every σ > 1 / 2 \sigma> 1/2 σ > 1/2 . An integration by parts now gives
lim T → ∞ 1 T ∫ 1 T ζ ′ ( σ + i t ) n i t d t = − log n n σ . (31) \lim_{T \to\infty} \frac{1}{T}\int_{1}^{T} \zeta'(\sigma+ it)n^{it}\,dt = -\frac{\log n}{n^\sigma}. \tag*{(31)} T → ∞ lim T 1 ∫ 1 T ζ ′ ( σ + i t ) n i t d t = − n σ log n . (31)
Indeed, if g ( t ) : = ζ ( σ + i t ) g(t) := \zeta(\sigma+ it) g ( t ) := ζ ( σ + i t ) and λ : = log n \lambda:= \log n λ := log n , then
g ′ ( t ) = i ζ ′ ( σ + i t ) , g'(t) = i\zeta'(\sigma+ it), g ′ ( t ) = i ζ ′ ( σ + i t ) ,
and hence
∫ 1 T ζ ′ ( σ + i t ) e i λ t d t = − i [ g ( t ) e i λ t ] 1 T − λ ∫ 1 T g ( t ) e i λ t d t . \int_1^T \zeta'(\sigma+ it)e^{i\lambda t}\,dt = -i\left[g(t)e^{i\lambda t}\right]_1^T - \lambda\int_1^T g(t)e^{i\lambda t}\,dt. ∫ 1 T ζ ′ ( σ + i t ) e iλ t d t = − i [ g ( t ) e iλ t ] 1 T − λ ∫ 1 T g ( t ) e iλ t d t .
The boundary term divided by T T T tends to zero by (29) , and (31) follows from (28) . Set
a n : = log n n σ , P N ( t ) : = − ∑ n ≤ N a n n − i t . a_n := \frac{\log n}{n^\sigma}, \qquad P_N(t) := -\sum_{n\leq N} a_n n^{-it}. a n := n σ log n , P N ( t ) := − n ≤ N ∑ a n n − i t .
The case k = 1 k=1 k = 1 of the integer moment formula follows from the mean-value theorem for Dirichlet series, e.g., see [30 Theorem A and Equation (2)] ,
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 d t = ∑ n = 1 ∞ a n 2 . (32) \lim_{T\to\infty}\frac{1}{T}\int_1^T \left|\zeta'(\sigma+ it)\right|^2dt = \sum_{n=1}^{\infty}a_n^2. \tag*{(32)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 d t = n = 1 ∑ ∞ a n 2 . (32)
Notice that
∑ n = 1 ∞ a n 2 = ∑ n = 1 ∞ ( log n ) 2 n 2 σ < ∞ , \sum_{n=1}^{\infty}a_n^2 = \sum_{n=1}^{\infty}\frac{(\log n)^2}{n^{2\sigma}} < \infty, n = 1 ∑ ∞ a n 2 = n = 1 ∑ ∞ n 2 σ ( log n ) 2 < ∞ ,
since σ > 1 / 2 \sigma> 1/2 σ > 1/2 . Since P N P_N P N is a finite Dirichlet polynomial,
lim T → ∞ 1 T ∫ 1 T ∣ P N ( t ) ∣ 2 d t = ∑ n ≤ N a n 2 . (33) \lim_{T\to\infty}\frac{1}{T}\int_1^T |P_N(t)|^2dt = \sum_{n\leq N}a_n^2. \tag*{(33)} T → ∞ lim T 1 ∫ 1 T ∣ P N ( t ) ∣ 2 d t = n ≤ N ∑ a n 2 . (33)
By (31) ,
lim T → ∞ 1 T ∫ 1 T ζ ′ ( σ + i t ) P N ( t ) ‾ d t = ∑ n ≤ N a n 2 . (34) \lim_{T\to\infty}\frac{1}{T}\int_1^T \zeta'(\sigma+ it)\overline{P_N(t)}\,dt = \sum_{n\leq N}a_n^2. \tag*{(34)} T → ∞ lim T 1 ∫ 1 T ζ ′ ( σ + i t ) P N ( t ) d t = n ≤ N ∑ a n 2 . (34)
Expanding the square and using (32) –(34) , we obtain
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) − P N ( t ) ∣ 2 d t = ∑ n > N a n 2 . (35) \lim_{T\to\infty}\frac{1}{T}\int_1^T \left|\zeta'(\sigma+ it)-P_N(t)\right|^2dt = \sum_{n>N}a_n^2. \tag*{(35)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) − P N ( t ) ∣ 2 d t = n > N ∑ a n 2 . (35)
In particular,
lim N → ∞ lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) − P N ( t ) ∣ 2 d t = 0. (36) \lim_{N\to\infty}\lim_{T\to\infty}\frac{1}{T}\int_1^T \left|\zeta'(\sigma+ it)-P_N(t)\right|^2dt = 0. \tag*{(36)} N → ∞ lim T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) − P N ( t ) ∣ 2 d t = 0. (36)
Define
Y σ , N : = − ∑ n ≤ N a n X ( n ) , Y_{\sigma,N} := -\sum_{n\leq N}a_nX(n), Y σ , N := − n ≤ N ∑ a n X ( n ) ,
with X ( n ) X(n) X ( n ) defined as in (25) . Note that, for M < N M<N M < N ,
E ∣ Y σ , N − Y σ , M ∣ 2 = ∑ M < n ≤ N a n 2 . \mathbb{E}|Y_{\sigma,N}-Y_{\sigma,M}|^2 = \sum_{M<n\leq N}a_n^2. E ∣ Y σ , N − Y σ , M ∣ 2 = M < n ≤ N ∑ a n 2 .
Since ∑ n a n 2 < ∞ \sum_n a_n^2<\infty ∑ n a n 2 < ∞ , the sequence { Y σ , N } N ≥ 1 \{Y_{\sigma,N}\}_{N\geq1} { Y σ , N } N ≥ 1 is Cauchy in L 2 L^2 L 2 . By the definition of Z ( σ , X ) Z(\sigma,X) Z ( σ , X ) , so
Y σ , N ⟶ Z ( σ , X ) in L 2 . (37) Y_{\sigma,N}\longrightarrow Z(\sigma,X) \quad\text{in } L^2. \tag*{(37)} Y σ , N ⟶ Z ( σ , X ) in L 2 . (37)
We next identify Z ( σ , X ) Z(\sigma, X) Z ( σ , X ) as the limiting distribution of ζ ′ ( σ + i t ) \zeta'(\sigma+ it) ζ ′ ( σ + i t ) . Fix N N N , and let p 1 , … , p r p_1,\ldots,p_r p 1 , … , p r denote all primes not exceeding N N N . Every n ≤ N n \le N n ≤ N can be written uniquely in the form
n = p 1 ν 1 ( n ) ⋯ p r ν r ( n ) . n = p_1^{\nu_1(n)} \cdots p_r^{\nu_r(n)}. n = p 1 ν 1 ( n ) ⋯ p r ν r ( n ) .
The numbers
log p 1 , … , log p r \log p_1,\ldots,\log p_r log p 1 , … , log p r
are linearly independent over Q \mathbb{Q} Q . Indeed, if
m 1 log p 1 + ⋯ + m r log p r = 0 , m j ∈ Z , m_1 \log p_1+\cdots+m_r\log p_r=0,\qquad m_j\in\mathbb{Z}, m 1 log p 1 + ⋯ + m r log p r = 0 , m j ∈ Z ,
then
p 1 m 1 ⋯ p r m r = 1 , p_1^{m_1}\cdots p_r^{m_r}=1, p 1 m 1 ⋯ p r m r = 1 ,
and unique factorization implies m 1 = ⋯ = m r = 0 m_1=\cdots=m_r=0 m 1 = ⋯ = m r = 0 . It follows from the continuous Kronecker-Weyl theorem that
( e − i t log p 1 , … , e − i t log p r ) \left(e^{-it\log p_1},\ldots,e^{-it\log p_r}\right) ( e − i t l o g p 1 , … , e − i t l o g p r )
is equidistributed on T r \mathbb{T}^r T r with respect to Haar measure. Hence, for every bounded continuous function Φ : C → C \Phi:\mathbb{C}\to\mathbb{C} Φ : C → C ,
lim T → ∞ 1 T ∫ 1 T Φ ( P N ( t ) ) d t = E [ Φ ( Y σ , N ) ] . (38) \lim_{T\to\infty}\frac{1}{T}\int_1^T\Phi(P_N(t))\,dt=\mathbb{E}[\Phi(Y_{\sigma,N})]. \tag*{(38)} T → ∞ lim T 1 ∫ 1 T Φ ( P N ( t )) d t = E [ Φ ( Y σ , N )] . (38)
We now pass from P N P_N P N to ζ ′ \zeta' ζ ′ . Let Φ \Phi Φ be bounded and Lipschitz, with Lipschitz constant Lip ( Φ ) \operatorname{Lip}(\Phi) Lip ( Φ ) . By Cauchy–Schwarz and (35) ,
lim sup T → ∞ ∣ 1 T ∫ 1 T Φ ( ζ ′ ( σ + i t ) ) d t − 1 T ∫ 1 T Φ ( P N ( t ) ) d t ∣ ≤ Lip ( Φ ) ( ∑ n > N a n 2 ) 1 / 2 . (39) \limsup_{T\to\infty}\left|\frac{1}{T}\int_1^T\Phi(\zeta'(\sigma+it))\,dt-\frac{1}{T}\int_1^T\Phi(P_N(t))\,dt\right|
\le\operatorname{Lip}(\Phi)\left(\sum_{n>N}a_n^2\right)^{1/2}. \tag*{(39)} T → ∞ lim sup T 1 ∫ 1 T Φ ( ζ ′ ( σ + i t )) d t − T 1 ∫ 1 T Φ ( P N ( t )) d t ≤ Lip ( Φ ) ( n > N ∑ a n 2 ) 1/2 . (39)
On the other hand, by (37) ,
∣ E [ Φ ( Y σ , N ) ] − E [ Φ ( Z ( σ , X ) ) ] ∣ ≤ Lip ( Φ ) ∥ Y σ , N − Y σ ∥ L 2 = Lip ( Φ ) ( ∑ n > N a n 2 ) 1 / 2 . (40) \left|\mathbb{E}[\Phi(Y_{\sigma,N})]-\mathbb{E}[\Phi(Z(\sigma,X))]\right|
\le\operatorname{Lip}(\Phi)\|Y_{\sigma,N}-Y_\sigma\|_{L^2}
=\operatorname{Lip}(\Phi)\left(\sum_{n>N}a_n^2\right)^{1/2}. \tag*{(40)} ∣ E [ Φ ( Y σ , N )] − E [ Φ ( Z ( σ , X ))] ∣ ≤ Lip ( Φ ) ∥ Y σ , N − Y σ ∥ L 2 = Lip ( Φ ) ( n > N ∑ a n 2 ) 1/2 . (40)
Combining (38) –(40) , first letting T → ∞ T\to\infty T → ∞ and then N → ∞ N\to\infty N → ∞ , gives
lim T → ∞ 1 T ∫ 1 T Φ ( ζ ′ ( σ + i t ) ) d t = E [ Φ ( Z ( σ , X ) ) ] (41) \lim_{T\to\infty}\frac{1}{T}\int_1^T\Phi(\zeta'(\sigma+it))\,dt=\mathbb{E}[\Phi(Z(\sigma,X))] \tag*{(41)} T → ∞ lim T 1 ∫ 1 T Φ ( ζ ′ ( σ + i t )) d t = E [ Φ ( Z ( σ , X ))] (41)
for every bounded Lipschitz function Φ \Phi Φ . Hence
ζ ′ ( σ + i U T ) → T → ∞ d Z ( σ , X ) (42) \zeta'(\sigma+iU_T)\xrightarrow[T\to\infty]{d} Z(\sigma,X) \tag*{(42)} ζ ′ ( σ + i U T ) d T → ∞ Z ( σ , X ) (42)
where U T U_T U T is uniformly distributed on [ 1 , T ] [1,T] [ 1 , T ] . We now pass from bounded test functions to the complex moment. Write
a : = Re ( s ) > 0 a:=\operatorname{Re}(s)>0 a := Re ( s ) > 0
and choose an integer
k > max { a , 2 } . k>\max\{a,2\}. k > max { a , 2 } .
Under the Lindelöf hypothesis, the integer moment estimate [42 Lemma 3.2] gives
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 k d t = A k , σ < ∞ . (43) \lim_{T\to\infty}\frac{1}{T}\int_{1}^{T}|\zeta'(\sigma+\mathrm{i}t)|^{2k}\,\mathrm{d}t=A_{k,\sigma}<\infty. \tag*{(43)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 k d t = A k , σ < ∞. (43)
In particular, for all sufficiently large T T T ,
1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 k d t ≤ C \frac{1}{T}\int_{1}^{T}|\zeta'(\sigma+\mathrm{i}t)|^{2k}\,\mathrm{d}t\le C T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 k d t ≤ C
for some C = C ( k , σ ) C=C(k,\sigma) C = C ( k , σ ) . Since
∣ ∣ z ∣ 2 s ∣ = ∣ z ∣ 2 a , \left||z|^{2s}\right|=|z|^{2a}, ∣ z ∣ 2 s = ∣ z ∣ 2 a ,
we have, on ∣ z ∣ > R |z|>R ∣ z ∣ > R ,
∣ ∣ z ∣ 2 s ∣ ≤ R − 2 ( k − a ) ∣ z ∣ 2 k . \left||z|^{2s}\right|\le R^{-2(k-a)}|z|^{2k}. ∣ z ∣ 2 s ≤ R − 2 ( k − a ) ∣ z ∣ 2 k .
Hence, for all sufficiently large T T T ,
1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s 1 { ∣ ζ ′ ( σ + i t ) ∣ > R } d t ≤ C R − 2 ( k − a ) . (44) \frac{1}{T}\int_{1}^{T}|\zeta'(\sigma+\mathrm{i}t)|^{2s}\mathbf{1}_{\{|\zeta'(\sigma+\mathrm{i}t)|>R\}}\,\mathrm{d}t\le CR^{-2(k-a)}. \tag*{(44)} T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s 1 { ∣ ζ ′ ( σ + i t ) ∣ > R } d t ≤ C R − 2 ( k − a ) . (44)
We also need the corresponding bound for the limiting random variable. For M > 0 M>0 M > 0 , set
G M ( z ) : = min { ∣ z ∣ 2 k , M } . G_M(z):=\min\{|z|^{2k},M\}. G M ( z ) := min { ∣ z ∣ 2 k , M } .
By (42) ,
E [ G M ( Z ( σ , X ) ) ] = lim T → ∞ 1 T ∫ 1 T G M ( ζ ′ ( σ + i t ) ) d t ≤ A k , σ . \mathbb{E}[G_M(Z(\sigma,X))]=\lim_{T\to\infty}\frac{1}{T}\int_{1}^{T}G_M(\zeta'(\sigma+\mathrm{i}t))\,\mathrm{d}t\le A_{k,\sigma}. E [ G M ( Z ( σ , X ))] = T → ∞ lim T 1 ∫ 1 T G M ( ζ ′ ( σ + i t )) d t ≤ A k , σ .
Letting M → ∞ M\to\infty M → ∞ and using monotone convergence, we obtain
E [ ∣ Z ( σ , X ) ∣ 2 k ] ≤ A k , σ < ∞ . (45) \mathbb{E}\left[|Z(\sigma,X)|^{2k}\right]\le A_{k,\sigma}<\infty. \tag*{(45)} E [ ∣ Z ( σ , X ) ∣ 2 k ] ≤ A k , σ < ∞. (45)
Consequently,
E [ ∣ Z ( σ , X ) ∣ 2 s 1 { ∣ Z ( σ , X ) ∣ > R } ] ≤ A k , σ R − 2 ( k − a ) . (46) \mathbb{E}\left[|Z(\sigma,X)|^{2s}\mathbf{1}_{\{|Z(\sigma,X)|>R\}}\right]\le A_{k,\sigma}R^{-2(k-a)}. \tag*{(46)} E [ ∣ Z ( σ , X ) ∣ 2 s 1 { ∣ Z ( σ , X ) ∣ > R } ] ≤ A k , σ R − 2 ( k − a ) . (46)
Choose a continuous function χ : [ 0 , ∞ ) → [ 0 , 1 ] \chi:[0,\infty)\to[0,1] χ : [ 0 , ∞ ) → [ 0 , 1 ] such that
χ ( r ) = 1 ( 0 ≤ r ≤ 1 ) , χ ( r ) = 0 ( r ≥ 2 ) , \chi(r)=1\quad(0\le r\le1),\qquad\chi(r)=0\quad(r\ge2), χ ( r ) = 1 ( 0 ≤ r ≤ 1 ) , χ ( r ) = 0 ( r ≥ 2 ) ,
and define
F R ( z ) : = χ ( ∣ z ∣ / R ) ∣ z ∣ 2 s . F_R(z):=\chi(|z|/R)|z|^{2s}. F R ( z ) := χ ( ∣ z ∣/ R ) ∣ z ∣ 2 s .
Since Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 , the function z ↦ ∣ z ∣ 2 s z\mapsto|z|^{2s} z ↦ ∣ z ∣ 2 s , with the value 0 0 0 assigned at z = 0 z=0 z = 0 , is continuous on C \mathbb{C} C . So F R F_R F R is bounded and continuous. Hence, by (42) ,
lim T → ∞ 1 T ∫ 1 T F R ( ζ ′ ( σ + i t ) ) d t = E [ F R ( Z ( σ , X ) ) ] . (47) \lim_{T\to\infty}\frac{1}{T}\int_{1}^{T}F_R(\zeta'(\sigma+\mathrm{i}t))\,\mathrm{d}t=\mathbb{E}[F_R(Z(\sigma,X))]. \tag*{(47)} T → ∞ lim T 1 ∫ 1 T F R ( ζ ′ ( σ + i t )) d t = E [ F R ( Z ( σ , X ))] . (47)
It follows from (44) that
lim sup T → ∞ ∣ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t − 1 T ∫ 1 T F R ( ζ ′ ( σ + i t ) ) d t ∣ ≤ C R − 2 ( k − a ) . (48) \limsup_{T\to\infty}\left|\frac{1}{T}\int_{1}^{T}|\zeta'(\sigma+\mathrm{i}t)|^{2s}\,\mathrm{d}t-\frac{1}{T}\int_{1}^{T}F_R(\zeta'(\sigma+\mathrm{i}t))\,\mathrm{d}t\right|\le CR^{-2(k-a)}. \tag*{(48)} T → ∞ lim sup T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t − T 1 ∫ 1 T F R ( ζ ′ ( σ + i t )) d t ≤ C R − 2 ( k − a ) . (48)
Similarly, by (45) ,
E [ ∣ Z ( σ , X ) ∣ 2 s ] − E [ F R ( Z ( σ , X ) ) ] ≤ A k , σ R − 2 ( k − a ) . (49) \mathbb{E}\left[|Z(\sigma,X)|^{2s}\right]-\mathbb{E}\left[F_R(Z(\sigma,X))\right]\le A_{k,\sigma}R^{-2(k-a)}. \tag*{(49)} E [ ∣ Z ( σ , X ) ∣ 2 s ] − E [ F R ( Z ( σ , X )) ] ≤ A k , σ R − 2 ( k − a ) . (49)
Using (47) –(49) , we obtain
lim sup T → ∞ ∣ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t − E [ ∣ Z ( σ , X ) ∣ 2 s ] ∣ ≤ ( C + A k , σ ) R − 2 ( k − a ) . \limsup_{T\to\infty}\left|\frac{1}{T}\int_1^T|\zeta'(\sigma+it)|^{2s}\,dt-\mathbb{E}\left[|Z(\sigma,X)|^{2s}\right]\right|\le(C+A_{k,\sigma})R^{-2(k-a)}. T → ∞ lim sup T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t − E [ ∣ Z ( σ , X ) ∣ 2 s ] ≤ ( C + A k , σ ) R − 2 ( k − a ) .
Since k > a k>a k > a , letting R → ∞ R\to\infty R → ∞ gives
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ Z ( σ , X ) ∣ 2 s ] . \lim_{T\to\infty}\frac{1}{T}\int_1^T|\zeta'(\sigma+it)|^{2s}\,dt=\mathbb{E}\left[|Z(\sigma,X)|^{2s}\right]. T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ Z ( σ , X ) ∣ 2 s ] .
By (45) , the right-hand side is finite.
Identification with the derivative of the random Euler product
Proposition 2.1 expresses the limiting complex moments of ζ ′ ( σ + i t ) \zeta'(\sigma+it) ζ ′ ( σ + i t ) in terms of the random variable Z ( σ , X ) Z(\sigma,X) Z ( σ , X ) defined in (26) . We now identify this random variable with the derivative of the random Euler product ζ ( σ , X ) \zeta(\sigma,X) ζ ( σ , X ) . More precisely, in the following lemma 2.2, we prove that
Z ( σ , X ) = ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) Z(\sigma,X)=\zeta'(\sigma,X)=\zeta(\sigma,X)L'(\sigma,X) Z ( σ , X ) = ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X )
almost surely.
This identification allows us to express the limiting complex moments in terms of the derivative of the random Euler product, as stated in Theorem 1.3 .
Lemma 2.2. Let { X ( p ) } p \{X(p)\}_p { X ( p ) } p be independent random variables uniformly distributed on T \mathbb{T} T , and extend X X X completely multiplicatively to the positive integers. For Re ( w ) > 1 / 2 \operatorname{Re}(w)>1/2 Re ( w ) > 1/2 , let
ζ ( w , X ) : = ∏ p ( 1 − X ( p ) p w ) − 1 (50) \zeta(w,X):=\prod_p\left(1-\frac{X(p)}{p^w}\right)^{-1} \tag*{(50)} ζ ( w , X ) := p ∏ ( 1 − p w X ( p ) ) − 1 (50) denote the corresponding random Euler product, and define
L ( w , X ) : = ∑ p log ( 1 − X ( p ) p w ) − 1 (51) L(w,X):=\sum_p\log\left(1-\frac{X(p)}{p^w}\right)^{-1} \tag*{(51)} L ( w , X ) := p ∑ log ( 1 − p w X ( p ) ) − 1 (51) where the logarithm is defined by
log ( 1 − z ) − 1 = ∑ m = 1 ∞ z m m , ∣ z ∣ < 1. \log(1-z)^{-1}=\sum_{m=1}^{\infty}\frac{z^m}{m},\qquad|z|<1. log ( 1 − z ) − 1 = m = 1 ∑ ∞ m z m , ∣ z ∣ < 1. Then, almost surely, L ( ⋅ , X ) L(\mathord{\cdot},X) L ( ⋅ , X ) and ζ ( ⋅ , X ) \zeta(\mathord{\cdot},X) ζ ( ⋅ , X ) are holomorphic on the half-plane Re ( w ) > 1 / 2 \operatorname{Re}(w)>1/2 Re ( w ) > 1/2 and
ζ ( w , X ) = e L ( w , X ) . \zeta(w,X)=e^{L(w,X)}. ζ ( w , X ) = e L ( w , X ) . Moreover, for every fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 , let Z ( σ , X ) Z(\sigma,X) Z ( σ , X ) be defined as in (26) . Then
Z ( σ , X ) = ζ ′ ( σ , X ) = ζ ( σ , X ) ∂ ∂ σ L ( σ , X ) (52) Z(\sigma,X)=\zeta'(\sigma,X)=\zeta(\sigma,X)\frac{\partial}{\partial\sigma}L(\sigma,X) \tag*{(52)} Z ( σ , X ) = ζ ′ ( σ , X ) = ζ ( σ , X ) ∂ σ ∂ L ( σ , X ) (52) almost surely.
Proof. Fix σ > 1 / 2 \sigma>1/2 σ > 1/2 , and choose τ \tau τ such that
1 2 < τ < σ . \frac{1}{2}<\tau<\sigma. 2 1 < τ < σ .
By [35 Proposition 3.2.9] , almost surely, both
ζ ( w , X ) = ∏ p ( 1 − X ( p ) p w ) − 1 \zeta(w,X)=\prod_p\left(1-\frac{X(p)}{p^w}\right)^{-1} ζ ( w , X ) = p ∏ ( 1 − p w X ( p ) ) − 1
and
ζ ~ ( w , X ) : = ∑ n = 1 ∞ X ( n ) n w \widetilde{\zeta}(w,X):=\sum_{n=1}^{\infty}\frac{X(n)}{n^w} ζ ( w , X ) := n = 1 ∑ ∞ n w X ( n )
converge on the half-plane
U τ : = { w ∈ C : Re ( w ) > τ } . U_\tau:=\{w\in\mathbb{C}:\operatorname{Re}(w)>\tau\}. U τ := { w ∈ C : Re ( w ) > τ } .
locally uniformly on compact subsets, define holomorphic functions there, and satisfy
ζ ( w , X ) = ζ ~ ( w , X ) ( w ∈ U τ ) (53) \zeta(w,X)=\widetilde{\zeta}(w,X) \qquad(w\in U_{\tau}) \tag*{(53)} ζ ( w , X ) = ζ ( w , X ) ( w ∈ U τ ) (53)
almost surely. More precisely, if
L N ( w , X ) : = ∑ p ≤ N log ( 1 − X ( p ) p w ) − 1 , L_N(w,X):=\sum_{p\leq N}\log\left(1-\frac{X(p)}{p^w}\right)^{-1}, L N ( w , X ) := p ≤ N ∑ log ( 1 − p w X ( p ) ) − 1 ,
then L N ( ⋅ , X ) L_N(\cdot,X) L N ( ⋅ , X ) converges almost surely, locally uniformly on U τ U_{\tau} U τ , to a holomorphic random function L ( ⋅ , X ) L(\cdot,X) L ( ⋅ , X ) . Consequently,
exp ( L N ( w , X ) ) = ∏ p ≤ N ( 1 − X ( p ) p w ) − 1 \exp(L_N(w,X))=\prod_{p\leq N}\left(1-\frac{X(p)}{p^w}\right)^{-1} exp ( L N ( w , X )) = p ≤ N ∏ ( 1 − p w X ( p ) ) − 1
converges locally uniformly to e L ( w , X ) e^{L(w,X)} e L ( w , X ) . Hence
ζ ( w , X ) = e L ( w , X ) (54) \zeta(w,X)=e^{L(w,X)} \tag*{(54)} ζ ( w , X ) = e L ( w , X ) (54)
almost surely on U τ U_{\tau} U τ . Since both sides of (54) are holomorphic, the ordinary chain rule gives
ζ ′ ( w , X ) = e L ( w , X ) L ′ ( w , X ) (55) \zeta'(w,X)=e^{L(w,X)}L'(w,X) \tag*{(55)} ζ ′ ( w , X ) = e L ( w , X ) L ′ ( w , X ) (55)
almost surely for every w ∈ U τ w\in U_{\tau} w ∈ U τ .
It remains to identify ζ ′ ( σ , X ) \zeta'(\sigma,X) ζ ′ ( σ , X ) with the random variable Z ( σ , X ) Z(\sigma,X) Z ( σ , X ) defined through the L 2 L^2 L 2 limit. For N ≥ 1 N\geq1 N ≥ 1 , set
ζ N ( w , X ) : = ∑ n ≤ N X ( n ) n w . \zeta_N(w,X):=\sum_{n\leq N}\frac{X(n)}{n^w}. ζ N ( w , X ) := n ≤ N ∑ n w X ( n ) .
By the almost sure local uniform convergence of the random Dirichlet series,
ζ N ( ⋅ , X ) ⟶ ζ ( ⋅ , X ) \zeta_N(\cdot,X)\longrightarrow\zeta(\cdot,X) ζ N ( ⋅ , X ) ⟶ ζ ( ⋅ , X )
locally uniformly on U τ U_{\tau} U τ almost surely. Since the ζ N \zeta_N ζ N are holomorphic, the Weierstrass theorem, or equivalently Cauchy’s integral formula, implies that their derivatives converge locally uniformly to the derivative of the limit. Note that σ > τ \sigma>\tau σ > τ . Thus
ζ N ′ ( σ , X ) ⟶ ζ ′ ( σ , X ) \zeta_N'(\sigma,X)\longrightarrow\zeta'(\sigma,X) ζ N ′ ( σ , X ) ⟶ ζ ′ ( σ , X )
almost surely.
But
ζ N ′ ( σ , X ) = − ∑ n ≤ N log n n σ X ( n ) . \zeta_N'(\sigma,X)=-\sum_{n\leq N}\frac{\log n}{n^\sigma}X(n). ζ N ′ ( σ , X ) = − n ≤ N ∑ n σ log n X ( n ) .
Therefore
− ∑ n ≤ N log n n σ X ( n ) ⟶ ζ ′ ( σ , X ) (56) -\sum_{n\leq N}\frac{\log n}{n^\sigma}X(n)\longrightarrow\zeta'(\sigma,X) \tag*{(56)} − n ≤ N ∑ n σ log n X ( n ) ⟶ ζ ′ ( σ , X ) (56)
almost surely.
On the other hand, using
E [ X ( n ) X ( m ) ‾ ] = δ n , m , \mathbb{E}\left[X(n)\overline{X(m)}\right]=\delta_{n,m}, E [ X ( n ) X ( m ) ] = δ n , m ,
we obtain, for M < N M<N M < N ,
E [ ∣ ζ N ′ ( σ , X ) − ζ M ′ ( σ , X ) ∣ 2 ] = ∑ M < n ≤ N ( log n ) 2 n 2 σ . \mathbb{E}\left[\left|\zeta_N'(\sigma,X)-\zeta_M'(\sigma,X)\right|^2\right]
=\sum_{M<n\leq N}\frac{(\log n)^2}{n^{2\sigma}}. E [ ∣ ζ N ′ ( σ , X ) − ζ M ′ ( σ , X ) ∣ 2 ] = M < n ≤ N ∑ n 2 σ ( log n ) 2 .
Since σ > 1 / 2 \sigma> 1/2 σ > 1/2 ,
∑ n = 1 ∞ ( log n ) 2 n 2 σ < ∞ . \sum_{n=1}^{\infty} \frac{(\log n)^2}{n^{2\sigma}} < \infty. n = 1 ∑ ∞ n 2 σ ( log n ) 2 < ∞.
Hence { ζ N ′ ( σ , X ) } \{\zeta'_N(\sigma, X)\} { ζ N ′ ( σ , X )} is Cauchy in L 2 L^2 L 2 , and by definition of Z ( σ , X ) Z(\sigma, X) Z ( σ , X ) ,
ζ N ′ ( σ , X ) ⟶ Z ( σ , X ) in L 2 . \zeta'_N(\sigma, X) \longrightarrow Z(\sigma, X) \quad\text{in } L^2. ζ N ′ ( σ , X ) ⟶ Z ( σ , X ) in L 2 .
In particular,
ζ N ′ ( σ , X ) ⟶ Z ( σ , X ) in probability . (57) \zeta'_N(\sigma, X) \longrightarrow Z(\sigma, X) \quad\text{in probability}. \tag*{(57)} ζ N ′ ( σ , X ) ⟶ Z ( σ , X ) in probability . (57)
By (56) ,
ζ N ′ ( σ , X ) ⟶ ζ ′ ( σ , X ) almost surely , \zeta'_N(\sigma, X) \longrightarrow\zeta'(\sigma, X) \quad\text{almost surely}, ζ N ′ ( σ , X ) ⟶ ζ ′ ( σ , X ) almost surely ,
and hence also in probability. Since limits in probability are unique up to almost sure equality, we conclude that
ζ ′ ( σ , X ) = Z ( σ , X ) almost surely . (58) \zeta'(\sigma, X) = Z(\sigma, X) \quad\text{almost surely}. \tag*{(58)} ζ ′ ( σ , X ) = Z ( σ , X ) almost surely . (58)
Combining (58) with (55) , evaluated at w = σ w=\sigma w = σ , gives
Z ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) almost surely . Z(\sigma, X) = \zeta(\sigma, X)L'(\sigma, X) \quad\text{almost surely}. Z ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) almost surely .
Finally, since L ( ⋅ , X ) L(\mathord{\cdot}, X) L ( ⋅ , X ) is holomorphic, its complex derivative at the real point σ \sigma σ agrees with its derivative along the real axis:
L ′ ( σ , X ) = ∂ ∂ σ L ( σ , X ) . L'(\sigma, X) = \frac{\partial}{\partial\sigma}L(\sigma, X). L ′ ( σ , X ) = ∂ σ ∂ L ( σ , X ) .
We are now ready to prove Theorem 1.3 .
Proof of Theorem 1.3. The assertion under the Lindelöf hypothesis follows directly from Proposition 2.1 and Lemma 2.2. The unconditional assertion for 0 < Re ( s ) < 2 0 < \operatorname{Re}(s) < 2 0 < Re ( s ) < 2 then follows from Remark 2.1.
For the subsequent analysis of the moments of ζ ′ ( σ , X ) \zeta'(\sigma, X) ζ ′ ( σ , X ) , we shall use the prime-sum representation of the logarithmic derivative L ′ ( σ , X ) L'(\sigma, X) L ′ ( σ , X ) established in Lemma 2.3 below.
Lemma 2.3. Let σ > 1 / 2 \sigma> 1/2 σ > 1/2 be fixed. Then, almost surely,
L ′ ( σ , X ) = − lim N → ∞ ∑ p ≤ N ( log p ) X ( p ) p − σ 1 − X ( p ) p − σ . (59) L'(\sigma, X) = -\lim_{N\to\infty}\sum_{p\leq N}\frac{(\log p)X(p)p^{-\sigma}}{1-X(p)p^{-\sigma}}. \tag*{(59)} L ′ ( σ , X ) = − N → ∞ lim p ≤ N ∑ 1 − X ( p ) p − σ ( log p ) X ( p ) p − σ . (59) Equivalently,
L ′ ( σ , X ) = − ∑ p ( log p ) X ( p ) p σ − X ( p ) almost surely , (60) L'(\sigma, X) = -\sum_p\frac{(\log p)X(p)}{p^\sigma-X(p)} \quad\text{almost surely}, \tag*{(60)} L ′ ( σ , X ) = − p ∑ p σ − X ( p ) ( log p ) X ( p ) almost surely , (60) where the series on the right-hand side is understood as the almost surely convergent limit of its partial sums over the primes.
Proof. Fix τ \tau τ such that
1 2 < τ < σ , \frac{1}{2}<\tau<\sigma, 2 1 < τ < σ ,
and set
U τ : = { w ∈ C : Re ( w ) > τ } . U_\tau:=\{w\in\mathbb{C}:\operatorname{Re}(w)>\tau\}. U τ := { w ∈ C : Re ( w ) > τ } .
For N ≥ 2 N\ge2 N ≥ 2 , define
L N ( w , X ) : = ∑ p ≤ N log ( 1 − X ( p ) p w ) − 1 . L_N(w,X):=\sum_{p\le N}\log\left(1-\frac{X(p)}{p^w}\right)^{-1}. L N ( w , X ) := p ≤ N ∑ log ( 1 − p w X ( p ) ) − 1 .
By the proof of [35 Proposition 3.2.9] , there exists an event Ω 0 ⊂ Ω \Omega_0\subset\Omega Ω 0 ⊂ Ω with P ( Ω 0 ) = 1 \mathbb{P}(\Omega_0)=1 P ( Ω 0 ) = 1 such that, for every ω ∈ Ω 0 \omega\in\Omega_0 ω ∈ Ω 0 ,
L N ( ⋅ , X ( ω ) ) ⟶ L ( ⋅ , X ( ω ) ) as N → ∞ , L_N(\cdot,X(\omega))\longrightarrow L(\cdot,X(\omega))\qquad\text{as }N\to\infty, L N ( ⋅ , X ( ω )) ⟶ L ( ⋅ , X ( ω )) as N → ∞ ,
locally uniformly on U τ U_\tau U τ . Fix ω ∈ Ω 0 \omega\in\Omega_0 ω ∈ Ω 0 . Since each L N ( ⋅ , X ( ω ) ) L_N(\cdot,X(\omega)) L N ( ⋅ , X ( ω )) is holomorphic on U τ U_\tau U τ and the convergence is locally uniform, the standard theorem on locally uniform limits of holomorphic functions implies that L ( ⋅ , X ( ω ) ) L(\cdot,X(\omega)) L ( ⋅ , X ( ω )) is holomorphic and
L N ′ ( ⋅ , X ( ω ) ) ⟶ L ′ ( ⋅ , X ( ω ) ) as N → ∞ , L_N'(\cdot,X(\omega))\longrightarrow L'(\cdot,X(\omega))\qquad\text{as }N\to\infty, L N ′ ( ⋅ , X ( ω )) ⟶ L ′ ( ⋅ , X ( ω )) as N → ∞ ,
locally uniformly on U τ U_\tau U τ . In particular, since σ ∈ U τ \sigma\in U_\tau σ ∈ U τ ,
L N ′ ( σ , X ( ω ) ) ⟶ L ′ ( σ , X ( ω ) ) , as N → ∞ . (61) L_N'(\sigma,X(\omega))\longrightarrow L'(\sigma,X(\omega)),\qquad\text{as }N\to\infty. \tag*{(61)} L N ′ ( σ , X ( ω )) ⟶ L ′ ( σ , X ( ω )) , as N → ∞. (61)
For every fixed N N N , the sum defining L N L_N L N is finite, and hence it may be differentiated term by term. For each prime p p p , since
∣ X ( p ) p w ∣ < 1 , w ∈ U τ , \left|\frac{X(p)}{p^w}\right|<1,\qquad w\in U_\tau, p w X ( p ) < 1 , w ∈ U τ ,
we have
log ( 1 − X ( p ) p w ) − 1 = ∑ m = 1 ∞ X ( p ) m m p m w . \log\left(1-\frac{X(p)}{p^w}\right)^{-1}
=\sum_{m=1}^{\infty}\frac{X(p)^m}{mp^{mw}}. log ( 1 − p w X ( p ) ) − 1 = m = 1 ∑ ∞ m p m w X ( p ) m .
Therefore,
d d w log ( 1 − X ( p ) p w ) − 1 = − ∑ m = 1 ∞ ( log p ) X ( p ) m p m w = − ( log p ) X ( p ) p − w 1 − X ( p ) p − w . \frac{\mathrm{d}}{\mathrm{d}w}\log\left(1-\frac{X(p)}{p^w}\right)^{-1}
=-\sum_{m=1}^{\infty}\frac{(\log p)X(p)^m}{p^{mw}}
=-(\log p)\frac{X(p)p^{-w}}{1-X(p)p^{-w}}. d w d log ( 1 − p w X ( p ) ) − 1 = − m = 1 ∑ ∞ p m w ( log p ) X ( p ) m = − ( log p ) 1 − X ( p ) p − w X ( p ) p − w .
Consequently,
L N ′ ( σ , X ( ω ) ) = − ∑ p ≤ N ( log p ) X ( p ) p − σ 1 − X ( p ) p − σ . L_N'(\sigma,X(\omega))=-\sum_{p\le N}\frac{(\log p)X(p)p^{-\sigma}}{1-X(p)p^{-\sigma}}. L N ′ ( σ , X ( ω )) = − p ≤ N ∑ 1 − X ( p ) p − σ ( log p ) X ( p ) p − σ .
Combining this identity with (61) yields
L ′ ( σ , X ( ω ) ) = − lim N → ∞ ∑ p ≤ N ( log p ) X ( p ) p − σ 1 − X ( p ) p − σ . L'(\sigma,X(\omega))=-\lim_{N\to\infty}\sum_{p\le N}\frac{(\log p)X(p)p^{-\sigma}}{1-X(p)p^{-\sigma}}. L ′ ( σ , X ( ω )) = − N → ∞ lim p ≤ N ∑ 1 − X ( p ) p − σ ( log p ) X ( p ) p − σ .
Since ω ∈ Ω 0 \omega\in\Omega_0 ω ∈ Ω 0 was arbitrary and P ( Ω 0 ) = 1 \mathbb{P}(\Omega_0)=1 P ( Ω 0 ) = 1 , (59) holds almost surely.
Real moments via tilted measures
In § 2.1 , we established Theorem 1.3 , which identifies, under the Lindelöf hypothesis, the complex moments of the derivative of the random Euler product with the corresponding limiting moments of the derivative of the Riemann zeta function for fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 . We now investigate the asymptotic behaviour of the former as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . We begin with real parameters s > 0 s > 0 s > 0 , for which the positivity of the weight ∣ ζ ( σ , X ) ∣ 2 s |\zeta(\sigma, X)|^{2s} ∣ ζ ( σ , X ) ∣ 2 s allows us to introduce a tilted probability measure. .
Construction of the tilted probability measure
In this subsection, we introduce a tilted probability measure associated with the random Euler product. Throughout this section, let
σ > 1 2 , s > 0 , \sigma> \frac{1}{2}, \qquad s > 0, σ > 2 1 , s > 0 ,
and recall that, almost surely,
ζ ( σ , X ) = e L ( σ , X ) = ∏ p ( 1 − X ( p ) p σ ) − 1 . \zeta(\sigma, X) = e^{L(\sigma,X)} = \prod_p \left(1-\frac{X(p)}{p^\sigma}\right)^{-1}. ζ ( σ , X ) = e L ( σ , X ) = p ∏ ( 1 − p σ X ( p ) ) − 1 .
Since L ( σ , X ) L(\sigma, X) L ( σ , X ) is finite almost surely, it follows that
0 < ∣ ζ ( σ , X ) ∣ < ∞ almost surely . (62) 0 < |\zeta(\sigma, X)| < \infty\qquad\text{almost surely}. \tag*{(62)} 0 < ∣ ζ ( σ , X ) ∣ < ∞ almost surely . (62)
We first verify that the moments required for the change of measure are finite. For N ≥ 2 N \ge2 N ≥ 2 , let
ζ N ( σ , X ) : = ∏ p ≤ N ( 1 − X ( p ) p σ ) − 1 . \zeta_N(\sigma, X) := \prod_{p \le N} \left(1-\frac{X(p)}{p^\sigma}\right)^{-1}. ζ N ( σ , X ) := p ≤ N ∏ ( 1 − p σ X ( p ) ) − 1 .
For 0 < r < 1 0 < r < 1 0 < r < 1 , set
m s ( r ) : = E [ ∣ 1 − r X ( p ) ∣ − 2 s ] . (63) m_s(r) := \mathbb{E}\left[|1-rX(p)|^{-2s}\right]. \tag*{(63)} m s ( r ) := E [ ∣1 − r X ( p ) ∣ − 2 s ] . (63)
Since X ( p ) X(p) X ( p ) is Haar-uniform on T \mathbb{T} T , we have
m s ( r ) = 1 2 π ∫ 0 2 π ∣ 1 − r e i θ ∣ − 2 s d θ . m_s(r) = \frac{1}{2\pi}\int_0^{2\pi}|1-re^{i\theta}|^{-2s}\,\mathrm{d}\theta. m s ( r ) = 2 π 1 ∫ 0 2 π ∣1 − r e i θ ∣ − 2 s d θ .
Using
( 1 − z ) − s = ∑ k = 0 ∞ ( s ) k k ! z k , ∣ z ∣ < 1 , (1-z)^{-s}=\sum_{k=0}^{\infty}\frac{(s)_k}{k!}z^k,\qquad|z|<1, ( 1 − z ) − s = k = 0 ∑ ∞ k ! ( s ) k z k , ∣ z ∣ < 1 ,
together with the orthogonality of the characters on T \mathbb{T} T , we obtain
m s ( r ) = ∑ k = 0 ∞ ( ( s ) k k ! ) 2 r 2 k . (64) m_s(r)=\sum_{k=0}^{\infty}\left(\frac{(s)_k}{k!}\right)^2r^{2k}. \tag*{(64)} m s ( r ) = k = 0 ∑ ∞ ( k ! ( s ) k ) 2 r 2 k . (64)
In particular, as p → ∞ p\to\infty p → ∞ ,
m s ( p − σ ) = 1 + s 2 p − 2 σ + O s ( p − 4 σ ) . (65) m_s(p^{-\sigma})=1+s^2p^{-2\sigma}+O_s(p^{-4\sigma}). \tag*{(65)} m s ( p − σ ) = 1 + s 2 p − 2 σ + O s ( p − 4 σ ) . (65)
Here, recall that ( r ) k (r)_k ( r ) k denotes the Pochhammer symbol,
( s ) 0 : = 1 , ( s ) k : = s ( s + 1 ) ⋯ ( s + k − 1 ) , k ≥ 1. (s)_0:=1,\qquad(s)_k:=s(s+1)\cdots(s+k-1),\qquad k\ge1. ( s ) 0 := 1 , ( s ) k := s ( s + 1 ) ⋯ ( s + k − 1 ) , k ≥ 1.
Since 2 σ > 1 2\sigma> 1 2 σ > 1 , it follows that
∑ p ∣ m s ( p − σ ) − 1 ∣ < ∞ , \sum_p \left|m_s(p^{-\sigma})-1\right| < \infty, p ∑ m s ( p − σ ) − 1 < ∞ ,
and therefore
0 < ∏ p m s ( p − σ ) < ∞ . (66) 0 < \prod_p m_s(p^{-\sigma}) < \infty. \tag*{(66)} 0 < p ∏ m s ( p − σ ) < ∞. (66)
By the independence of the random variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p ,
E [ ∣ ζ N ( σ , X ) ∣ 2 s ] = ∏ p ≤ N m s ( p − σ ) . (67) \mathbb{E}\left[\left|\zeta_N(\sigma,X)\right|^{2s}\right]=\prod_{p\leq N}m_s(p^{-\sigma}). \tag*{(67)} E [ ∣ ζ N ( σ , X ) ∣ 2 s ] = p ≤ N ∏ m s ( p − σ ) . (67)
Moreover, ζ N ( σ , X ) → ζ ( σ , X ) \zeta_N(\sigma,X)\to\zeta(\sigma,X) ζ N ( σ , X ) → ζ ( σ , X ) almost surely as N → ∞ N\to\infty N → ∞ . To pass to the limit in (67) , fix q > 1 q>1 q > 1 . By (66) ,
sup N ≥ 2 E [ ∣ ζ N ( σ , X ) ∣ 2 s q ] = sup N ≥ 2 ∏ p ≤ N m s q ( p − σ ) < ∞ . \sup_{N\geq2}\mathbb{E}\left[\left|\zeta_N(\sigma,X)\right|^{2sq}\right]
=\sup_{N\geq2}\prod_{p\leq N}m_{sq}(p^{-\sigma})<\infty. N ≥ 2 sup E [ ∣ ζ N ( σ , X ) ∣ 2 s q ] = N ≥ 2 sup p ≤ N ∏ m s q ( p − σ ) < ∞.
Hence the family
{ ∣ ζ N ( σ , X ) ∣ 2 s } N ≥ 2 (68) \left\{\left|\zeta_N(\sigma,X)\right|^{2s}\right\}_{N\geq2} \tag*{(68)} { ∣ ζ N ( σ , X ) ∣ 2 s } N ≥ 2 (68)
is uniformly integrable. Since
∣ ζ N ( σ , X ) ∣ 2 s ⟶ ∣ ζ ( σ , X ) ∣ 2 s almost surely , \left|\zeta_N(\sigma,X)\right|^{2s}\longrightarrow\left|\zeta(\sigma,X)\right|^{2s}
\qquad\text{almost surely}, ∣ ζ N ( σ , X ) ∣ 2 s ⟶ ∣ ζ ( σ , X ) ∣ 2 s almost surely ,
Vitali’s convergence theorem yields
E [ ∣ ζ ( σ , X ) ∣ 2 s ] = ∏ p E [ ∣ 1 − X ( p ) p σ ∣ − 2 s ] ∈ ( 0 , ∞ ) . (69) \mathbb{E}\left[\left|\zeta(\sigma,X)\right|^{2s}\right]
=\prod_p\mathbb{E}\left[\left|1-\frac{X(p)}{p^\sigma}\right|^{-2s}\right]\in(0,\infty). \tag*{(69)} E [ ∣ ζ ( σ , X ) ∣ 2 s ] = p ∏ E [ 1 − p σ X ( p ) − 2 s ] ∈ ( 0 , ∞ ) . (69)
By (69) , the denominator below is finite and strictly positive. We may therefore introduce the following well-defined tilted probability measure and the corresponding tilted expectation.
Definition 3.1. Let σ > 1 / 2 \sigma>1/2 σ > 1/2 and s > 0 s>0 s > 0 . We define the tilted probability measure P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s on ( Ω , F ) (\Omega,\mathcal{F}) ( Ω , F ) by
d P ~ σ , s d P : = ∣ ζ ( σ , X ) ∣ 2 s E [ ∣ ζ ( σ , X ) ∣ 2 s ] . (70) \frac{d\widetilde{\mathbb{P}}_{\sigma,s}}{d\mathbb{P}}
:=\frac{|\zeta(\sigma,X)|^{2s}}{\mathbb{E}[|\zeta(\sigma,X)|^{2s}]}. \tag*{(70)} d P d P σ , s := E [ ∣ ζ ( σ , X ) ∣ 2 s ] ∣ ζ ( σ , X ) ∣ 2 s . (70) Equivalently, for every A ∈ F A\in\mathcal{F} A ∈ F ,
P ~ σ , s ( A ) = E [ ∣ ζ ( σ , X ) ∣ 2 s 1 A ] E [ ∣ ζ ( σ , X ) ∣ 2 s ] . \widetilde{\mathbb{P}}_{\sigma,s}(A)
=\frac{\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}\mathbf{1}_A\right]}{\mathbb{E}[|\zeta(\sigma,X)|^{2s}]}. P σ , s ( A ) = E [ ∣ ζ ( σ , X ) ∣ 2 s ] E [ ∣ ζ ( σ , X ) ∣ 2 s 1 A ] . For every complex-valued random variable F F F satisfying
E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ F ∣ ] < ∞ , \mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|F|\right]<\infty, E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ F ∣ ] < ∞ , we define the corresponding tilted expectation by
E ~ σ , s [ F ] : = ∫ Ω F d P ~ σ , s = E [ ∣ ζ ( σ , X ) ∣ 2 s F ] E [ ∣ ζ ( σ , X ) ∣ 2 s ] . (71) \widetilde{\mathbb{E}}_{\sigma,s}[F]
:=\int_\Omega F\,d\widetilde{\mathbb{P}}_{\sigma,s}
=\frac{\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}F\right]}{\mathbb{E}[|\zeta(\sigma,X)|^{2s}]}. \tag*{(71)} E σ , s [ F ] := ∫ Ω F d P σ , s = E [ ∣ ζ ( σ , X ) ∣ 2 s ] E [ ∣ ζ ( σ , X ) ∣ 2 s F ] . (71)
In fact, the Radon-Nikodym derivative in (70) is nonnegative and satisfies
∫ Ω ∣ ζ ( σ , X ) ∣ 2 s E [ ∣ ζ ( σ , X ) ∣ 2 s ] d P = 1. \int_{\Omega} \frac{|\zeta(\sigma,X)|^{2s}}{\mathbb{E}[|\zeta(\sigma,X)|^{2s}]} \, \mathrm{d}\mathbb{P}=1. ∫ Ω E [ ∣ ζ ( σ , X ) ∣ 2 s ] ∣ ζ ( σ , X ) ∣ 2 s d P = 1.
Hence P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s is a probability measure. Moreover, by (62) , its Radon-Nikodym derivative is strictly positive almost surely, so P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s and P \mathbb{P} P are equivalent probability measures. Finally, (71) is precisely the condition that
F ∈ L 1 ( Ω , P ~ σ , s ) . F \in L^1(\Omega,\widetilde{\mathbb{P}}_{\sigma,s}). F ∈ L 1 ( Ω , P σ , s ) .
We finally record that the logarithmic derivative L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) is integrable with respect to the tilted measure. Recall from (60) that
L ′ ( σ , X ) = − ∑ p ( log p ) X ( p ) p σ − X ( p ) almost surely, (72) L'(\sigma,X)=-\sum_p\frac{(\log p)X(p)}{p^\sigma-X(p)}
\quad\text{almost surely,} \tag*{(72)} L ′ ( σ , X ) = − p ∑ p σ − X ( p ) ( log p ) X ( p ) almost surely, (72)
where the series is understood as the limit of its partial sums over the primes. Set
Y p : = − ( log p ) X ( p ) p σ − X ( p ) = − ( log p ) ∑ m = 1 ∞ X ( p ) m p m σ . Y_p:=-\frac{(\log p)X(p)}{p^\sigma-X(p)}
=-(\log p)\sum_{m=1}^{\infty}\frac{X(p)^m}{p^{m\sigma}}. Y p := − p σ − X ( p ) ( log p ) X ( p ) = − ( log p ) m = 1 ∑ ∞ p mσ X ( p ) m .
Since
E [ X ( p ) m ] = 0 , m ≥ 1 , \mathbb{E}[X(p)^m]=0,\qquad m\ge1, E [ X ( p ) m ] = 0 , m ≥ 1 ,
we have E [ Y p ] = 0 \mathbb{E}[Y_p]=0 E [ Y p ] = 0 . Furthermore, the orthogonality of the characters on T \mathbb{T} T gives
E [ ∣ Y p ∣ 2 ] = ( log p ) 2 ∑ m = 1 ∞ p − 2 m σ = ( log p ) 2 p 2 σ − 1 . (73) \mathbb{E}[|Y_p|^2]=(\log p)^2\sum_{m=1}^{\infty}p^{-2m\sigma}
=\frac{(\log p)^2}{p^{2\sigma}-1}. \tag*{(73)} E [ ∣ Y p ∣ 2 ] = ( log p ) 2 m = 1 ∑ ∞ p − 2 mσ = p 2 σ − 1 ( log p ) 2 . (73)
Since the random variables { Y p } p \{Y_p\}_p { Y p } p are independent and centered, and
∑ p ( log p ) 2 p 2 σ − 1 < ∞ , \sum_p\frac{(\log p)^2}{p^{2\sigma}-1}<\infty, p ∑ p 2 σ − 1 ( log p ) 2 < ∞ ,
the partial sums in (72) converge in L 2 ( Ω , P ) L^2(\Omega,\mathbb{P}) L 2 ( Ω , P ) . Consequently,
L ′ ( σ , X ) ∈ L 2 ( Ω , P ) , (74) L'(\sigma,X)\in L^2(\Omega,\mathbb{P}), \tag*{(74)} L ′ ( σ , X ) ∈ L 2 ( Ω , P ) , (74)
and in fact
E [ ∣ L ′ ( σ , X ) ∣ 2 ] = ∑ p ( log p ) 2 p 2 σ − 1 < ∞ . \mathbb{E}\left[|L'(\sigma,X)|^2\right]
=\sum_p\frac{(\log p)^2}{p^{2\sigma}-1}<\infty. E [ ∣ L ′ ( σ , X ) ∣ 2 ] = p ∑ p 2 σ − 1 ( log p ) 2 < ∞.
Combining (69) , with r = 2 s r=2s r = 2 s , and (74) , the Cauchy–Schwarz inequality gives
E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ L ′ ( σ , X ) ∣ ] ≤ ( E [ ∣ ζ ( σ , X ) ∣ 4 s ] ) 1 / 2 ( E [ ∣ L ′ ( σ , X ) ∣ 2 ] ) 1 / 2 < ∞ . (75) \mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|L'(\sigma,X)|\right]
\leq\left(\mathbb{E}\left[|\zeta(\sigma,X)|^{4s}\right]\right)^{1/2}
\left(\mathbb{E}\left[|L'(\sigma,X)|^2\right]\right)^{1/2}<\infty. \tag*{(75)} E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ L ′ ( σ , X ) ∣ ] ≤ ( E [ ∣ ζ ( σ , X ) ∣ 4 s ] ) 1/2 ( E [ ∣ L ′ ( σ , X ) ∣ 2 ] ) 1/2 < ∞. (75)
Hence
L ′ ( σ , X ) ∈ L 1 ( Ω , P ~ σ , s ) , L'(\sigma,X)\in L^1(\Omega,\widetilde{\mathbb{P}}_{\sigma,s}), L ′ ( σ , X ) ∈ L 1 ( Ω , P σ , s ) ,
and therefore
E ~ σ , s [ L ′ ( σ , X ) ] = E [ ∣ ζ ( σ , X ) ∣ 2 s L ′ ( σ , X ) ] E [ ∣ ζ ( σ , X ) ∣ 2 s ] (76) \widetilde{\mathbb{E}}_{\sigma,s}\left[L'(\sigma,X)\right]
=\frac{\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}L'(\sigma,X)\right]}
{\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}\right]} \tag*{(76)} E σ , s [ L ′ ( σ , X ) ] = E [ ∣ ζ ( σ , X ) ∣ 2 s ] E [ ∣ ζ ( σ , X ) ∣ 2 s L ′ ( σ , X ) ] (76)
is a well-defined finite complex number.
We remark that all the statements in this section hold unconditionally for every σ > 1 / 2 \sigma>1/2 σ > 1/2 and s > 0 s>0 s > 0 .
Independence of prime-indexed random variables under the tilted measure
Under the original probability measure P \mathbb{P} P , the random variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p are independent and uniformly distributed on the unit circle. We now investigate whether this independence is preserved under the tilted probability measure P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s defined in (70) . Lemma 3.1 below shows that the prime-indexed random variables remain independent under the tilted measure, although their marginal distributions are modified by the corresponding Euler factors.
Lemma 3.1. Let σ > 1 / 2 \sigma> 1/2 σ > 1/2 and s > 0 s > 0 s > 0 . Let m s ( p − σ ) m_s(p^{-\sigma}) m s ( p − σ ) be defined in (63) . Under the tilted probability measure P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s defined in (3.1), the random variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p remain independent. More precisely, for each prime p p p , the distribution of X ( p ) X(p) X ( p ) under P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s is the probability measure
d μ ~ p , σ , s ( z ) : = ∣ 1 − p − σ z ∣ − 2 s m s ( p − σ ) d m T ( z ) , z ∈ T . (77) \mathrm{d}\widetilde{\mu}_{p,\sigma,s}(z):=\frac{|1-p^{-\sigma}z|^{-2s}}{m_s(p^{-\sigma})}\,\mathrm{d}m_{\mathbb{T}}(z),\qquad z\in\mathbb{T}. \tag*{(77)} d μ p , σ , s ( z ) := m s ( p − σ ) ∣1 − p − σ z ∣ − 2 s d m T ( z ) , z ∈ T . (77) Consequently,
P ~ σ , s = ⨂ p μ ~ p , σ , s . (78) \widetilde{\mathbb{P}}_{\sigma,s}=\bigotimes_p\widetilde{\mu}_{p,\sigma,s}. \tag*{(78)} P σ , s = p ⨂ μ p , σ , s . (78)
Proof. Since p − σ < 1 p^{-\sigma}<1 p − σ < 1 ,
0 < m s ( p − σ ) < ∞ , 0<m_s(p^{-\sigma})<\infty, 0 < m s ( p − σ ) < ∞ ,
and therefore (77) indeed defines a probability measure on T \mathbb{T} T . For N ≥ 2 N\geq2 N ≥ 2 , let
ζ N ( σ , X ) : = ∏ p ≤ N ( 1 − X ( p ) p σ ) − 1 \zeta_N(\sigma,X):=\prod_{p\leq N}\left(1-\frac{X(p)}{p^\sigma}\right)^{-1} ζ N ( σ , X ) := p ≤ N ∏ ( 1 − p σ X ( p ) ) − 1
and define
R N : = ∣ ζ N ( σ , X ) ∣ 2 s E [ ∣ ζ N ( σ , X ) ∣ 2 s ] . R_N:=\frac{|\zeta_N(\sigma,X)|^{2s}}{\mathbb{E}[|\zeta_N(\sigma,X)|^{2s}]}. R N := E [ ∣ ζ N ( σ , X ) ∣ 2 s ] ∣ ζ N ( σ , X ) ∣ 2 s .
By the independence of the random variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p under P \mathbb{P} P ,
E [ ∣ ζ N ( σ , X ) ∣ 2 s ] = ∏ p ≤ N m s ( p − σ ) , \mathbb{E}[|\zeta_N(\sigma,X)|^{2s}]=\prod_{p\leq N}m_s(p^{-\sigma}), E [ ∣ ζ N ( σ , X ) ∣ 2 s ] = p ≤ N ∏ m s ( p − σ ) ,
and hence
R N = ∏ p ≤ N ∣ 1 − p − σ X ( p ) ∣ − 2 s m s ( p − σ ) . (79) R_N=\prod_{p\leq N}\frac{|1-p^{-\sigma}X(p)|^{-2s}}{m_s(p^{-\sigma})}. \tag*{(79)} R N = p ≤ N ∏ m s ( p − σ ) ∣1 − p − σ X ( p ) ∣ − 2 s . (79)
We first show that R N R_N R N converges in L 1 ( Ω , P ) L^1(\Omega,\mathbb{P}) L 1 ( Ω , P ) to the Radon–Nikodym derivative defining P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s . Set
Y N : = ∣ ζ N ( σ , X ) ∣ 2 s , Y : = ∣ ζ ( σ , X ) ∣ 2 s . Y_N:=|\zeta_N(\sigma,X)|^{2s},\qquad Y:=|\zeta(\sigma,X)|^{2s}. Y N := ∣ ζ N ( σ , X ) ∣ 2 s , Y := ∣ ζ ( σ , X ) ∣ 2 s .
By (68) ,
Y N ⟶ Y in L 1 ( Ω , P ) , Y_N\longrightarrow Y\qquad\text{in }L^1(\Omega,\mathbb{P}), Y N ⟶ Y in L 1 ( Ω , P ) ,
and therefore
E [ Y N ] ⟶ E [ Y ] . \mathbb{E}[Y_N]\longrightarrow\mathbb{E}[Y]. E [ Y N ] ⟶ E [ Y ] .
Moreover, by (69) ,
0 < E [ Y ] < ∞ . 0<\mathbb{E}[Y]<\infty. 0 < E [ Y ] < ∞.
Thus, for all sufficiently large N N N , M N M_N M N is bounded away from zero, and
E [ ∣ Y N E [ Y N ] − Y E [ Y ] ∣ ] ≤ 1 E [ Y N ] E [ ∣ Y N − Y ∣ ] + ∣ 1 E [ Y N ] − 1 E [ Y ] ∣ E [ ∣ Y ∣ ] ⟶ 0. \mathbb{E}\left[\left|\frac{Y_N}{\mathbb{E}[Y_N]}-\frac{Y}{\mathbb{E}[Y]}\right|\right]\leq\frac{1}{\mathbb{E}[Y_N]}\mathbb{E}[|Y_N-Y|]+\left|\frac{1}{\mathbb{E}[Y_N]}-\frac{1}{\mathbb{E}[Y]}\right|\mathbb{E}[|Y|]\longrightarrow0. E [ E [ Y N ] Y N − E [ Y ] Y ] ≤ E [ Y N ] 1 E [ ∣ Y N − Y ∣ ] + E [ Y N ] 1 − E [ Y ] 1 E [ ∣ Y ∣ ] ⟶ 0.
Consequently,
R N ⟶ ∣ ζ ( σ , X ) ∣ 2 s E [ ∣ ζ ( σ , X ) ∣ 2 s ] in L 1 ( Ω , P ) . (80) R_N\longrightarrow\frac{|\zeta(\sigma,X)|^{2s}}{\mathbb{E}[|\zeta(\sigma,X)|^{2s}]}\quad\text{in }L^1(\Omega,\mathbb{P}). \tag*{(80)} R N ⟶ E [ ∣ ζ ( σ , X ) ∣ 2 s ] ∣ ζ ( σ , X ) ∣ 2 s in L 1 ( Ω , P ) . (80)
Now let p 1 , … , p k p_1,\ldots,p_k p 1 , … , p k be distinct primes and let f 1 , … , f k f_1,\ldots,f_k f 1 , … , f k be bounded Borel measurable functions on T \mathbb{T} T . Choose N N N sufficiently large so that p j ≤ N p_j\leq N p j ≤ N for all 1 ≤ j ≤ k 1\leq j\leq k 1 ≤ j ≤ k . Using (79) and the independence of the variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p under P \mathbb{P} P , we obtain
E [ R N ∏ j = 1 k f j ( X ( p j ) ) ] = ∏ j = 1 k E [ ∣ 1 − p j − σ X ( p j ) ∣ − 2 s f j ( X ( p j ) ) ] m s ( p j − σ ) ∏ p ≤ N p ∉ { p 1 , … , p k } E [ ∣ 1 − p − σ X ( p ) ∣ − 2 s ] m s ( p − σ ) . \mathbb{E}\left[R_N\prod_{j=1}^{k}f_j(X(p_j))\right]=\prod_{j=1}^{k}\frac{\mathbb{E}\left[|1-p_j^{-\sigma}X(p_j)|^{-2s}f_j(X(p_j))\right]}{m_s(p_j^{-\sigma})}\prod_{\substack{p\leq N\\p\notin\{p_1,\ldots,p_k\}}}\frac{\mathbb{E}\left[|1-p^{-\sigma}X(p)|^{-2s}\right]}{m_s(p^{-\sigma})}. E [ R N j = 1 ∏ k f j ( X ( p j )) ] = j = 1 ∏ k m s ( p j − σ ) E [ ∣1 − p j − σ X ( p j ) ∣ − 2 s f j ( X ( p j )) ] p ≤ N p ∈ / { p 1 , … , p k } ∏ m s ( p − σ ) E [ ∣1 − p − σ X ( p ) ∣ − 2 s ] .
The second product on the right-hand side is equal to 1. Hence
E [ R N ∏ j = 1 k f j ( X ( p j ) ) ] = ∏ j = 1 k E [ ∣ 1 − p j − σ X ( p j ) ∣ − 2 s f j ( X ( p j ) ) ] m s ( p j − σ ) . \mathbb{E}\left[R_N\prod_{j=1}^{k}f_j(X(p_j))\right]=\prod_{j=1}^{k}\frac{\mathbb{E}\left[|1-p_j^{-\sigma}X(p_j)|^{-2s}f_j(X(p_j))\right]}{m_s(p_j^{-\sigma})}. E [ R N j = 1 ∏ k f j ( X ( p j )) ] = j = 1 ∏ k m s ( p j − σ ) E [ ∣1 − p j − σ X ( p j ) ∣ − 2 s f j ( X ( p j )) ] .
Since
∏ j = 1 k f j ( X ( p j ) ) \prod_{j=1}^{k}f_j(X(p_j)) j = 1 ∏ k f j ( X ( p j ))
is bounded, the L 1 L^1 L 1 convergence in (80) implies that
E [ R N ∏ j = 1 k f j ( X ( p j ) ) ] ⟶ E [ ∣ ζ ( σ , X ) ∣ 2 s E [ ∣ ζ ( σ , X ) ∣ 2 s ] ∏ j = 1 k f j ( X ( p j ) ) ] . \mathbb{E}\left[R_N\prod_{j=1}^{k}f_j(X(p_j))\right]\longrightarrow\mathbb{E}\left[\frac{|\zeta(\sigma,X)|^{2s}}{\mathbb{E}[|\zeta(\sigma,X)|^{2s}]}\prod_{j=1}^{k}f_j(X(p_j))\right]. E [ R N j = 1 ∏ k f j ( X ( p j )) ] ⟶ E [ E [ ∣ ζ ( σ , X ) ∣ 2 s ] ∣ ζ ( σ , X ) ∣ 2 s j = 1 ∏ k f j ( X ( p j )) ] .
By the definition of P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s , we therefore obtain
E ~ σ , s [ ∏ j = 1 k f j ( X ( p j ) ) ] = ∏ j = 1 k E [ ∣ 1 − p j − σ X ( p j ) ∣ − 2 s f j ( X ( p j ) ) ] m s ( p j − σ ) = ∏ j = 1 k ∫ T f j ( z ) d μ ~ p j , σ , s ( z ) . (81) \widetilde{\mathbb{E}}_{\sigma,s}\left[\prod_{j=1}^{k}f_j(X(p_j))\right]=\prod_{j=1}^{k}\frac{\mathbb{E}\left[|1-p_j^{-\sigma}X(p_j)|^{-2s}f_j(X(p_j))\right]}{m_s(p_j^{-\sigma})}
=\prod_{j=1}^{k}\int_{\mathbb{T}}f_j(z)\,d\widetilde{\mu}_{p_j,\sigma,s}(z). \tag*{(81)} E σ , s [ j = 1 ∏ k f j ( X ( p j )) ] = j = 1 ∏ k m s ( p j − σ ) E [ ∣1 − p j − σ X ( p j ) ∣ − 2 s f j ( X ( p j )) ] = j = 1 ∏ k ∫ T f j ( z ) d μ p j , σ , s ( z ) . (81)
Taking
f j = 1 A j , A j ∈ B ( T ) , f_j=\mathbf{1}_{A_j},\qquad A_j\in\mathcal{B}(\mathbb{T}), f j = 1 A j , A j ∈ B ( T ) ,
in (81) , we obtain
P ~ σ , s ( X ( p 1 ) ∈ A 1 , … , X ( p k ) ∈ A k ) = ∏ j = 1 k μ ~ p j , σ , s ( A j ) . (82) \widetilde{\mathbb{P}}_{\sigma,s}\left(X(p_1)\in A_1,\ldots,X(p_k)\in A_k\right)=\prod_{j=1}^{k}\widetilde{\mu}_{p_j,\sigma,s}(A_j). \tag*{(82)} P σ , s ( X ( p 1 ) ∈ A 1 , … , X ( p k ) ∈ A k ) = j = 1 ∏ k μ p j , σ , s ( A j ) . (82)
Taking k = 1 k=1 k = 1 shows that the marginal distribution of X ( p ) X(p) X ( p ) under P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s is μ ~ p , σ , s \widetilde{\mu}_{p,\sigma,s} μ p , σ , s . Since (82) holds for every finite collection of distinct primes and every collection of Borel sets A 1 , … , A k ⊂ T A_1,\ldots,A_k\subset\mathbb{T} A 1 , … , A k ⊂ T , it follows directly from the definition of independence that the random variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p are independent under P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s . Finally, let
μ ∗ : = ⨂ p μ ~ p , σ , s . \mu^* := \bigotimes_p \widetilde{\mu}_{p,\sigma,s}. μ ∗ := p ⨂ μ p , σ , s .
For every cylinder set of the form
C = { ω ∈ Ω : X ( p j ) ( ω ) ∈ A j , 1 ≤ j ≤ k } , C = \{\omega\in\Omega: X(p_j)(\omega) \in A_j,\quad1 \le j \le k\}, C = { ω ∈ Ω : X ( p j ) ( ω ) ∈ A j , 1 ≤ j ≤ k } ,
with A j ∈ B ( T ) A_j \in\mathcal{B}(\mathbb{T}) A j ∈ B ( T ) , we have by (82)
P ~ σ , s ( C ) = ∏ j = 1 k μ ~ p j , σ , s ( A j ) = μ ∗ ( C ) . \widetilde{\mathbb{P}}_{\sigma,s}(C) = \prod_{j=1}^{k} \widetilde{\mu}_{p_j,\sigma,s}(A_j) = \mu^*(C). P σ , s ( C ) = j = 1 ∏ k μ p j , σ , s ( A j ) = μ ∗ ( C ) .
Since such cylinder sets form a π \pi π -system generating
F = ⨂ p B ( T ) , \mathcal{F} = \bigotimes_p \mathcal{B}(\mathbb{T}), F = p ⨂ B ( T ) ,
the uniqueness theorem for probability measures yields
P ~ σ , s = μ ∗ = ⨂ p μ ~ p , σ , s . \widetilde{\mathbb{P}}_{\sigma,s} = \mu^* = \bigotimes_p \widetilde{\mu}_{p,\sigma,s}. P σ , s = μ ∗ = p ⨂ μ p , σ , s .
Gaussian limit for the logarithmic derivative under the tilted measure
The independence established in Lemma 3.1 allows us to analyze the normalized logarithmic derivative ( 2 σ − 1 ) L ′ ( σ , X ) (2\sigma-1)L'(\sigma,X) ( 2 σ − 1 ) L ′ ( σ , X ) as a sum of independent random variables and establish its Gaussian limit as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . We now prove Proposition 1.5 .
Proof of Proposition 1.5 . It suffices to prove the assertion along an arbitrary sequence σ n ↓ 1 / 2 \sigma_n \downarrow1/2 σ n ↓ 1/2 . Set
δ n : = 2 σ n − 1 , \delta_n := 2\sigma_n - 1, δ n := 2 σ n − 1 ,
so that δ n ↓ 0 \delta_n \downarrow0 δ n ↓ 0 , and define
W p , n : = X ( p ) p σ n − X ( p ) = p − σ n X ( p ) 1 − p − σ n X ( p ) . W_{p,n} := \frac{X(p)}{p^{\sigma_n}-X(p)} = \frac{p^{-\sigma_n}X(p)}{1-p^{-\sigma_n}X(p)}. W p , n := p σ n − X ( p ) X ( p ) = 1 − p − σ n X ( p ) p − σ n X ( p ) .
By Lemma 3.1, under P ~ σ n , s \widetilde{\mathbb{P}}_{\sigma_n,s} P σ n , s the variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p are independent, and the law of X ( p ) X(p) X ( p ) is
d μ ~ p , σ n , s ( z ) = ∣ 1 − p − σ n z ∣ − 2 s m s ( p − σ n ) d m T ( z ) , \mathrm{d}\widetilde{\mu}_{p,\sigma_n,s}(z) = \frac{\lvert1-p^{-\sigma_n}z\rvert^{-2s}}{m_s(p^{-\sigma_n})}\,\mathrm{d}m_{\mathbb{T}}(z), d μ p , σ n , s ( z ) = m s ( p − σ n ) ∣ 1 − p − σ n z ∣ − 2 s d m T ( z ) ,
where m s ( r ) m_s(r) m s ( r ) is defined in (63) . We first establish the local moment estimates. By (64) ,
m s ( p − σ n ) = 1 + s 2 p − 2 σ n + s 2 ( s + 1 ) 2 4 p − 4 σ n + O s ( p − 6 σ n ) . (83) m_s(p^{-\sigma_n}) = 1+s^2p^{-2\sigma_n}+\frac{s^2(s+1)^2}{4}p^{-4\sigma_n}+O_s(p^{-6\sigma_n}). \tag*{(83)} m s ( p − σ n ) = 1 + s 2 p − 2 σ n + 4 s 2 ( s + 1 ) 2 p − 4 σ n + O s ( p − 6 σ n ) . (83)
Since
0 ≤ p − σ n ≤ 2 − 1 / 2 < 1 , 0 \le p^{-\sigma_n} \le2^{-1/2} < 1, 0 ≤ p − σ n ≤ 2 − 1/2 < 1 ,
the above estimates may be taken uniformly in p p p and n n n . Moreover, for 0 ≤ r < 1 0 \le r < 1 0 ≤ r < 1 ,
d d r ∫ T ∣ 1 − r z ∣ − 2 s d m T ( z ) = s ∫ T ∣ 1 − r z ∣ − 2 s ( z 1 − r z + z ‾ 1 − r z ‾ ) d m T ( z ) . \frac{\mathrm{d}}{\mathrm{d}r}\int_{\mathbb{T}} |1-rz|^{-2s}\,\mathrm{d}m_{\mathbb{T}}(z)
= s\int_{\mathbb{T}} |1-rz|^{-2s}\left(\frac{z}{1-rz}+\frac{\overline{z}}{1-r\overline{z}}\right)\,\mathrm{d}m_{\mathbb{T}}(z). d r d ∫ T ∣1 − r z ∣ − 2 s d m T ( z ) = s ∫ T ∣1 − r z ∣ − 2 s ( 1 − r z z + 1 − r z z ) d m T ( z ) .
The local tilted measure is invariant under complex conjugation, and hence
E ~ σ n , s [ W p , n ] ∈ R . \widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}] \in\mathbb{R}. E σ n , s [ W p , n ] ∈ R .
It follows that
E ~ σ n , s [ W p , n ] = p − σ n 2 s 1 m s ( p − σ n ) d d r ∫ T ∣ 1 − r z ∣ − 2 s d m T ( z ) ∣ r = p − σ n . \widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}]
= \frac{p^{-\sigma_n}}{2s}\frac{1}{m_s(p^{-\sigma_n})}
\left.\frac{\mathrm{d}}{\mathrm{d}r}\int_{\mathbb{T}} |1-rz|^{-2s}\,\mathrm{d}m_{\mathbb{T}}(z)\right|_{r=p^{-\sigma_n}}. E σ n , s [ W p , n ] = 2 s p − σ n m s ( p − σ n ) 1 d r d ∫ T ∣1 − r z ∣ − 2 s d m T ( z ) r = p − σ n .
By (64) , we obtain
E ~ σ n , s [ W p , n ] = s p − 2 σ n + O s ( p − 4 σ n ) . (84) \widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}]
= sp^{-2\sigma_n}+O_s(p^{-4\sigma_n}). \tag*{(84)} E σ n , s [ W p , n ] = s p − 2 σ n + O s ( p − 4 σ n ) . (84)
Next, note that
∣ W p , n ∣ 2 = p − 2 σ n ∣ 1 − p − σ n X ( p ) ∣ 2 , |W_{p,n}|^2=\frac{p^{-2\sigma_n}}{|1-p^{-\sigma_n}X(p)|^2}, ∣ W p , n ∣ 2 = ∣1 − p − σ n X ( p ) ∣ 2 p − 2 σ n ,
so that
E ~ σ n , s [ ∣ W p , n ∣ 2 ] = p − 2 σ n m s + 1 ( p − σ n ) m s ( p − σ n ) . \widetilde{\mathbb{E}}_{\sigma_n,s}[|W_{p,n}|^2]
= p^{-2\sigma_n}\frac{m_{s+1}(p^{-\sigma_n})}{m_s(p^{-\sigma_n})}. E σ n , s [ ∣ W p , n ∣ 2 ] = p − 2 σ n m s ( p − σ n ) m s + 1 ( p − σ n ) .
Hence
E ~ σ n , s [ ∣ W p , n ∣ 2 ] = p − 2 σ n + O s ( p − 4 σ n ) . (85) \widetilde{\mathbb{E}}_{\sigma_n,s}[|W_{p,n}|^2]
= p^{-2\sigma_n}+O_s(p^{-4\sigma_n}). \tag*{(85)} E σ n , s [ ∣ W p , n ∣ 2 ] = p − 2 σ n + O s ( p − 4 σ n ) . (85)
For the second complex moment, expanding the factors into power series gives
E ~ σ n , s [ W p , n 2 ] = p − 4 σ n m s ( p − σ n ) ∑ k = 0 ∞ ( s + 2 ) k ( s ) k + 2 k ! ( k + 2 ) ! r p , n 2 k , \widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}^2]
= \frac{p^{-4\sigma_n}}{m_s(p^{-\sigma_n})}
\sum_{k=0}^{\infty}\frac{(s+2)_k(s)_{k+2}}{k!(k+2)!}r_{p,n}^{2k}, E σ n , s [ W p , n 2 ] = m s ( p − σ n ) p − 4 σ n k = 0 ∑ ∞ k ! ( k + 2 )! ( s + 2 ) k ( s ) k + 2 r p , n 2 k ,
and consequently
E ~ σ n , s [ W p , n 2 ] = s ( s + 1 ) 2 p − 4 σ n + O s ( p − 6 σ n ) . (86) \widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}^2]
= \frac{s(s+1)}{2}p^{-4\sigma_n}+O_s(p^{-6\sigma_n}). \tag*{(86)} E σ n , s [ W p , n 2 ] = 2 s ( s + 1 ) p − 4 σ n + O s ( p − 6 σ n ) . (86)
By (84) –(86) ,
E ~ σ n , s [ ∣ W p , n − E ~ σ n , s [ W p , n ] ∣ 2 ] = p − 2 σ n + O s ( p − 4 σ n ) , (87) \widetilde{\mathbb{E}}_{\sigma_n,s}\left[\left|W_{p,n}-\widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}]\right|^2\right]
= p^{-2\sigma_n}+O_s(p^{-4\sigma_n}), \tag*{(87)} E σ n , s [ W p , n − E σ n , s [ W p , n ] 2 ] = p − 2 σ n + O s ( p − 4 σ n ) , (87)
and
E ~ σ n , s [ ( W p , n − E ~ σ n , s [ W p , n ] ) 2 ] = O s ( p − 4 σ n ) . (88) \widetilde{\mathbb{E}}_{\sigma_n,s}\left[(W_{p,n}-\widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}])^2\right]
= O_s(p^{-4\sigma_n}). \tag*{(88)} E σ n , s [ ( W p , n − E σ n , s [ W p , n ] ) 2 ] = O s ( p − 4 σ n ) . (88)
We next verify that the infinite prime sum is well defined. For every fixed n n n , (84) gives
∑ p ( log p ) ∣ E ~ σ n , s [ W p , n ] ∣ < ∞ , \sum_p(\log p)\left|\widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}]\right|<\infty, p ∑ ( log p ) E σ n , s [ W p , n ] < ∞ ,
since 2 σ n > 1 2\sigma_n>1 2 σ n > 1 . and define
Z p , n : = ( log p ) ( W p , n − E ~ σ n , s [ W p , n ] ) . Z_{p,n}:=(\log p)(W_{p,n}-\widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}]). Z p , n := ( log p ) ( W p , n − E σ n , s [ W p , n ]) .
Likewise, (87) gives
∑ p E ~ σ n , s [ ∣ Z p , n ∣ 2 ] < ∞ . \sum_p\widetilde{\mathbb{E}}_{\sigma_n,s}[|Z_{p,n}|^2]<\infty. p ∑ E σ n , s [ ∣ Z p , n ∣ 2 ] < ∞.
The variables { Z p , n } p \{Z_{p,n}\}_p { Z p , n } p are independent and centered under P ~ σ n , s \widetilde{\mathbb{P}}_{\sigma_n,s} P σ n , s , and therefore their partial sums form a Cauchy sequence in L 2 ( Ω , P ~ σ n , s ) L^2(\Omega,\widetilde{\mathbb{P}}_{\sigma_n,s}) L 2 ( Ω , P σ n , s ) . We may thus define
S n : = ∑ p ( log p ) W p , n = ∑ p ( log p ) E ~ σ n , s [ W p , n ] + ∑ p Z p , n , (89) S_n := \sum_p (\log p)W_{p,n}=\sum_p(\log p)\widetilde{\mathbb{E}}_{\sigma_n,s}[W_{p,n}]+\sum_p Z_{p,n}, \tag*{(89)} S n := p ∑ ( log p ) W p , n = p ∑ ( log p ) E σ n , s [ W p , n ] + p ∑ Z p , n , (89)
where the second sum is understood in L 2 ( Ω , P ~ σ n , s ) L^2(\Omega,\widetilde{\mathbb{P}}_{\sigma_n,s}) L 2 ( Ω , P σ n , s ) . This definition agrees with the almost sure prime-series representation of the logarithmic derivative. Indeed, by (60) ,
− ∑ p ( log p ) X ( p ) p σ n − X ( p ) = L ′ ( σ n , X ) P -almost surely . -\sum_p \frac{(\log p)X(p)}{p^{\sigma_n}-X(p)}=L'(\sigma_n,X)\qquad\mathbb{P}\text{-almost surely}. − p ∑ p σ n − X ( p ) ( log p ) X ( p ) = L ′ ( σ n , X ) P -almost surely .
Since P ~ σ n , s \widetilde{\mathbb{P}}_{\sigma_n,s} P σ n , s is equivalent to P \mathbb{P} P , every P \mathbb{P} P -null set is also P ~ σ n , s \widetilde{\mathbb{P}}_{\sigma_n,s} P σ n , s -null. Hence the same identity holds P ~ σ n , s \widetilde{\mathbb{P}}_{\sigma_n,s} P σ n , s -almost surely. On the other hand, the partial sums in (89) converge in L 2 ( P ~ σ n , s ) L^2(\widetilde{\mathbb{P}}_{\sigma_n,s}) L 2 ( P σ n , s ) , hence in probability, to S n S_n S n . Uniqueness of limits in probability therefore gives
S n = − L ′ ( σ n , X ) P ~ σ n , s -almost surely . (90) S_n=-L'(\sigma_n,X)\qquad\widetilde{\mathbb{P}}_{\sigma_n,s}\text{-almost surely}. \tag*{(90)} S n = − L ′ ( σ n , X ) P σ n , s -almost surely . (90)
We now determine the asymptotic mean. From (84) ,
E ~ σ n , s [ S n ] = s ∑ p log p p 2 σ n + O s ( ∑ p ( log p ) 2 p 4 σ n ) = s ∑ p log p p 1 + δ n + O s ( 1 ) , (91) \widetilde{\mathbb{E}}_{\sigma_n,s}[S_n]=s\sum_p\frac{\log p}{p^{2\sigma_n}}+O_s\left(\sum_p\frac{(\log p)^2}{p^{4\sigma_n}}\right)=s\sum_p\frac{\log p}{p^{1+\delta_n}}+O_s(1), \tag*{(91)} E σ n , s [ S n ] = s p ∑ p 2 σ n log p + O s ( p ∑ p 4 σ n ( log p ) 2 ) = s p ∑ p 1 + δ n log p + O s ( 1 ) , (91)
because 4 σ n ≥ 2 4\sigma_n\ge2 4 σ n ≥ 2 . It is known that
∑ p log p p 1 + δ = 1 δ + O ( 1 ) . (92) \sum_p\frac{\log p}{p^{1+\delta}}=\frac{1}{\delta}+O(1). \tag*{(92)} p ∑ p 1 + δ log p = δ 1 + O ( 1 ) . (92)
Then, it follows from (91) that
δ n E ~ σ n , s [ S n ] ⟶ s . (93) \delta_n\widetilde{\mathbb{E}}_{\sigma_n,s}[S_n]\longrightarrow s. \tag*{(93)} δ n E σ n , s [ S n ] ⟶ s . (93)
We next determine the covariance of the centered sum. By (87) ,
∑ p E ~ σ n , s [ ∣ Z p , n ∣ 2 ] = ∑ p ( log p ) 2 p 2 σ n + O s ( ∑ p ( log p ) 2 p 4 σ n ) = ∑ p ( log p ) 2 p 1 + δ n + O s ( 1 ) . (94) \sum_p\widetilde{\mathbb{E}}_{\sigma_n,s}[|Z_{p,n}|^2]=\sum_p\frac{(\log p)^2}{p^{2\sigma_n}}+O_s\left(\sum_p\frac{(\log p)^2}{p^{4\sigma_n}}\right)=\sum_p\frac{(\log p)^2}{p^{1+\delta_n}}+O_s(1). \tag*{(94)} p ∑ E σ n , s [ ∣ Z p , n ∣ 2 ] = p ∑ p 2 σ n ( log p ) 2 + O s ( p ∑ p 4 σ n ( log p ) 2 ) = p ∑ p 1 + δ n ( log p ) 2 + O s ( 1 ) . (94)
By (88) ,
∑ p E ~ σ n , s [ Z p , n 2 ] = O s ( 1 ) . \sum_p\widetilde{\mathbb{E}}_{\sigma_n,s}[Z_{p,n}^2]=O_s(1). p ∑ E σ n , s [ Z p , n 2 ] = O s ( 1 ) .
It is known that
∑ p ( log p ) 2 p 1 + δ = 1 δ 2 + O ( 1 ) . \sum_p\frac{(\log p)^2}{p^{1+\delta}}=\frac{1}{\delta^2}+O(1). p ∑ p 1 + δ ( log p ) 2 = δ 2 1 + O ( 1 ) .
Consequently,
δ n 2 ( ∑ p E ~ σ n , s [ ∣ Z p , n ∣ 2 ] ) ⟶ 1 , as n → ∞ , (95) \delta_n^2\left(\sum_p\widetilde{\mathbb{E}}_{\sigma_n,s}[|Z_{p,n}|^2]\right)\longrightarrow1,\qquad\text{as }n\to\infty, \tag*{(95)} δ n 2 ( p ∑ E σ n , s [ ∣ Z p , n ∣ 2 ] ) ⟶ 1 , as n → ∞ , (95)
while
δ n 2 ( ∑ p E ~ σ n , s [ Z p , n 2 ] ) ⟶ 0 , as n → ∞ . (96) \delta_n^2\left(\sum_p\widetilde{\mathbb{E}}_{\sigma_n,s}[Z_{p,n}^2]\right)\longrightarrow0,\qquad\text{as }n\to\infty. \tag*{(96)} δ n 2 ( p ∑ E σ n , s [ Z p , n 2 ] ) ⟶ 0 , as n → ∞. (96)
We now reduce the centered infinite sum to a finite triangular array. For each n n n , since
∑ p E ~ σ n , s [ ∣ Z p , n ∣ 2 ] < ∞ , \sum_{p}\widetilde{\mathbb{E}}_{\sigma_n,s}[|Z_{p,n}|^2]<\infty, p ∑ E σ n , s [ ∣ Z p , n ∣ 2 ] < ∞ ,
we may choose P n ≥ δ n − 1 P_n\geq\delta_n^{-1} P n ≥ δ n − 1 sufficiently large so that
∑ p > P n E ~ σ n , s [ ∣ Z p , n ∣ 2 ] ≤ 1. \sum_{p>P_n}\widetilde{\mathbb{E}}_{\sigma_n,s}[|Z_{p,n}|^2]\leq1. p > P n ∑ E σ n , s [ ∣ Z p , n ∣ 2 ] ≤ 1.
In particular, P n → ∞ P_n\to\infty P n → ∞ , and
δ n 2 ∑ p > P n E ~ σ n , s [ ∣ Z p , n ∣ 2 ] ⟶ 0 , n → ∞ . (97) \delta_n^2\sum_{p>P_n}\widetilde{\mathbb{E}}_{\sigma_n,s}[|Z_{p,n}|^2]\longrightarrow0,\qquad n\to\infty. \tag*{(97)} δ n 2 p > P n ∑ E σ n , s [ ∣ Z p , n ∣ 2 ] ⟶ 0 , n → ∞. (97)
Furthermore,
δ n 2 ∣ ∑ p > P n E ~ σ n , s [ Z p , n 2 ] ∣ ⟶ 0. (98) \delta_n^2\left|\sum_{p>P_n}\widetilde{\mathbb{E}}_{\sigma_n,s}[Z_{p,n}^2]\right|\longrightarrow0. \tag*{(98)} δ n 2 p > P n ∑ E σ n , s [ Z p , n 2 ] ⟶ 0. (98)
Consider the finite triangular array
Y p , n : = δ n ( Re Z p , n Im Z p , n ) , p ≤ P n . Y_{p,n}:=\delta_n\begin{pmatrix}\operatorname{Re}Z_{p,n}\\ \operatorname{Im}Z_{p,n}\end{pmatrix},\qquad p\leq P_n. Y p , n := δ n ( Re Z p , n Im Z p , n ) , p ≤ P n .
Its members are independent and centered. By (95) , (96) , (97) , and (98) , the covariance matrix of
∑ p < P n Y p , n \sum_{p<P_n}Y_{p,n} p < P n ∑ Y p , n
converges to 1 2 I 2 \frac{1}{2}I_2 2 1 I 2 , as n → ∞ n\to\infty n → ∞ . Indeed, for any centered complex random variable Z Z Z ,
Var ( Re Z ) = 1 2 ( E ∣ Z ∣ 2 + Re E [ Z 2 ] ) , Var ( Im Z ) = 1 2 ( E ∣ Z ∣ 2 − Re E [ Z 2 ] ) , \operatorname{Var}(\operatorname{Re}Z)=\frac{1}{2}\left(\mathbb{E}|Z|^2+\operatorname{Re}\mathbb{E}[Z^2]\right),\qquad\operatorname{Var}(\operatorname{Im}Z)=\frac{1}{2}\left(\mathbb{E}|Z|^2-\operatorname{Re}\mathbb{E}[Z^2]\right), Var ( Re Z ) = 2 1 ( E ∣ Z ∣ 2 + Re E [ Z 2 ] ) , Var ( Im Z ) = 2 1 ( E ∣ Z ∣ 2 − Re E [ Z 2 ] ) ,
and
Cov ( Re Z , Im Z ) = 1 2 Im E [ Z 2 ] . \operatorname{Cov}(\operatorname{Re}Z,\operatorname{Im}Z)=\frac{1}{2}\operatorname{Im}\mathbb{E}[Z^2]. Cov ( Re Z , Im Z ) = 2 1 Im E [ Z 2 ] .
It remains to verify the Lindeberg condition. Since
∣ W p , n ∣ ≤ 1 p σ n − 1 ≤ 1 p − 1 , |W_{p,n}|\leq\frac{1}{p^{\sigma_n}-1}\leq\frac{1}{\sqrt{p}-1}, ∣ W p , n ∣ ≤ p σ n − 1 1 ≤ p − 1 1 ,
we have
( log p ) ∣ W p , n ∣ ≤ log p p − 1 . (\log p)|W_{p,n}|\leq\frac{\log p}{\sqrt{p}-1}. ( log p ) ∣ W p , n ∣ ≤ p − 1 log p .
The function on the right is bounded on the set of primes. Thus there exists an absolute constant C > 0 C>0 C > 0 such that
( log p ) ∣ W p , n ∣ ≤ C (\log p)|W_{p,n}|\leq C ( log p ) ∣ W p , n ∣ ≤ C
for every p p p and n n n . Therefore
∣ Z p , n ∣ ≤ ( log p ) ∣ W p , n ∣ + E ~ σ n , s [ ( log p ) ∣ W p , n ∣ ] ≤ 2 C . |Z_{p,n}|\leq(\log p)|W_{p,n}|+\widetilde{\mathbb{E}}_{\sigma_n,s}[(\log p)|W_{p,n}|]\leq2C. ∣ Z p , n ∣ ≤ ( log p ) ∣ W p , n ∣ + E σ n , s [( log p ) ∣ W p , n ∣ ] ≤ 2 C .
Consequently,
sup p ≤ P n ∣ Y p , n ∣ ≤ 2 C δ n ⟶ 0 , n → ∞ . \sup_{p\le P_n} |Y_{p,n}| \le2C\delta_n \longrightarrow0,\qquad n\to\infty. p ≤ P n sup ∣ Y p , n ∣ ≤ 2 C δ n ⟶ 0 , n → ∞.
Hence, for every ε > 0 \varepsilon>0 ε > 0 , for all sufficiently large n n n , ∣ Y p , n ∣ < ε |Y_{p,n}|<\varepsilon ∣ Y p , n ∣ < ε for every p ≤ P n p\le P_n p ≤ P n , and therefore the Lindeberg condition is satisfied. By the multivariate Lindeberg–Feller central limit theorem,
δ n ∑ p ≤ P n ( Re Z p , n Im Z p , n ) → d N ( ( 0 0 ) , 1 2 I 2 ) . (99) \delta_n\sum_{p\le P_n}
\begin{pmatrix}
\operatorname{Re} Z_{p,n}\\
\operatorname{Im} Z_{p,n}
\end{pmatrix}
\xrightarrow{\mathrm{d}}
N\left(
\begin{pmatrix}
0\\
0
\end{pmatrix},
\frac{1}{2}I_2
\right). \tag*{(99)} δ n p ≤ P n ∑ ( Re Z p , n Im Z p , n ) d N ( ( 0 0 ) , 2 1 I 2 ) . (99)
Finally, we shall remove the truncation. By independence and centering,
E ~ σ n , s [ ∣ δ n ∑ p > P n Z p , n ∣ 2 ] = δ n 2 ∑ p > P n E ~ σ n , s [ ∣ Z p , n ∣ 2 ] ⟶ 0 , n → ∞ \widetilde{\mathbb{E}}_{\sigma_n,s}\left[\left|\delta_n\sum_{p>P_n} Z_{p,n}\right|^2\right]
=\delta_n^2\sum_{p>P_n}\widetilde{\mathbb{E}}_{\sigma_n,s}\left[|Z_{p,n}|^2\right]
\longrightarrow0,\qquad n\to\infty E σ n , s δ n p > P n ∑ Z p , n 2 = δ n 2 p > P n ∑ E σ n , s [ ∣ Z p , n ∣ 2 ] ⟶ 0 , n → ∞
by (97) . Hence
δ n ∑ p > P n Z p , n ⟶ 0 , n → ∞ \delta_n\sum_{p>P_n} Z_{p,n}\longrightarrow0,\qquad n\to\infty δ n p > P n ∑ Z p , n ⟶ 0 , n → ∞
in L 2 ( P ~ σ n , s ) L^2(\widetilde{\mathbb{P}}_{\sigma_n,s}) L 2 ( P σ n , s ) and therefore in probability. By (99) and Slutsky’s theorem,
δ n ∑ p Z p , n → d G , n → ∞ (100) \delta_n\sum_p Z_{p,n}\xrightarrow{\mathrm{d}}G,\qquad n\to\infty \tag*{(100)} δ n p ∑ Z p , n d G , n → ∞ (100)
where
G = G 1 + i G 2 , G 1 , G 2 ∼ i i d N ( 0 , 1 / 2 ) . G=G_1+iG_2,\qquad G_1,G_2\overset{\mathrm{iid}}{\sim}N(0,1/2). G = G 1 + i G 2 , G 1 , G 2 ∼ iid N ( 0 , 1/2 ) .
Therefore, by (89) ,
δ n S n = δ n E ~ σ n , s [ S n ] + δ n ∑ p Z p , n . \delta_nS_n=\delta_n\widetilde{\mathbb{E}}_{\sigma_n,s}[S_n]+\delta_n\sum_p Z_{p,n}. δ n S n = δ n E σ n , s [ S n ] + δ n p ∑ Z p , n .
Combining (90) , (93) and (100) , and applying Slutsky’s theorem once more, yields
− ( 2 σ n − 1 ) L ′ ( σ n , X ) → d s + G , n → ∞ . -(2\sigma_n-1)L'(\sigma_n,X)\xrightarrow{\mathrm{d}}s+G,\qquad n\to\infty. − ( 2 σ n − 1 ) L ′ ( σ n , X ) d s + G , n → ∞.
Since the sequence σ n ↓ 1 / 2 \sigma_n\downarrow1/2 σ n ↓ 1/2 was arbitrary, the stated convergence holds as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
Uniform integrability and asymptotics of real moments
Proposition 1.5 establishes the convergence in distribution of the normalized logarithmic derivative T σ : = ( 2 σ − 1 ) L ′ ( σ , X ) T_\sigma:=(2\sigma-1)L'(\sigma,X) T σ := ( 2 σ − 1 ) L ′ ( σ , X ) under the tilted probability measure as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . However, convergence in distribution alone does not imply convergence of the corresponding moments. To obtain the asymptotic formula for the real moments of ζ ′ ( σ , X ) \zeta'(\sigma,X) ζ ′ ( σ , X ) , we establish uniform integrability of the relevant powers of T σ T_\sigma T σ under the tilted measure. Combining this with the Gaussian limit then yields the following result.
Proposition 3.2. Let s > 0 s>0 s > 0 , and let L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) be defined in (60) . Set
T σ = ( 2 σ − 1 ) L ′ ( σ , X ) . T_\sigma=(2\sigma-1)L'(\sigma,X). T σ = ( 2 σ − 1 ) L ′ ( σ , X ) . There exist q > 2 s q>2s q > 2 s and σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 such that
sup 1 / 2 < σ ≤ σ 0 E ~ σ , s [ ∣ T σ ∣ q ] < ∞ . (101) \sup_{1/2<\sigma\le\sigma_0}\widetilde{\mathbb{E}}_{\sigma,s}[|T_\sigma|^q]<\infty. \tag*{(101)} 1/2 < σ ≤ σ 0 sup E σ , s [ ∣ T σ ∣ q ] < ∞. (101)
where E ~ σ , s \widetilde{\mathbb{E}}_{\sigma,s} E σ , s is defined in (71) . Consequently, as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 ,
E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] ∼ a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) ( 2 σ − 1 ) − ( s 2 + 2 s ) (102) \mathbb{E}\left[\left|\zeta'(\sigma,X)\right|^{2s}\right] \sim a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2)(2\sigma-1)^{-(s^2+2s)} \tag*{(102)} E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] ∼ a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) ( 2 σ − 1 ) − ( s 2 + 2 s ) (102)
where a s a_s a s is defined in (3) .
Proof. Set
δ : = 2 σ − 1. \delta:= 2\sigma-1. δ := 2 σ − 1.
By (52) ,
ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) \zeta'(\sigma,X)=\zeta(\sigma,X)L'(\sigma,X) ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X )
almost surely, the definition of the tilted expectation gives
δ s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = ( δ s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ] ) E ~ σ , s [ ∣ δ L ′ ( σ , X ) ∣ 2 s ] , (103) \delta^{s^2+2s}\mathbb{E}\left[\left|\zeta'(\sigma,X)\right|^{2s}\right]
=
\left(\delta^{s^2}\mathbb{E}\left[\left|\zeta(\sigma,X)\right|^{2s}\right]\right)
\widetilde{\mathbb{E}}_{\sigma,s}\left[\left|\delta L'(\sigma,X)\right|^{2s}\right], \tag*{(103)} δ s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = ( δ s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ] ) E σ , s [ ∣ δ L ′ ( σ , X ) ∣ 2 s ] , (103)
where the identity is initially understood in the extended nonnegative sense. We first show that
δ s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ] ⟶ a s ( σ ↓ 1 / 2 ) . (104) \delta^{s^2}\mathbb{E}\left[\left|\zeta(\sigma,X)\right|^{2s}\right]\longrightarrow a_s
\qquad(\sigma\downarrow1/2). \tag*{(104)} δ s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ] ⟶ a s ( σ ↓ 1/2 ) . (104)
By (69) , (64) , and the definition of m s ( p − σ ) m_s(p^{-\sigma}) m s ( p − σ ) in (63) , we have
E [ ∣ ζ ( σ , X ) ∣ 2 s ] = ∏ p m s ( p − σ ) , \mathbb{E}\left[\left|\zeta(\sigma,X)\right|^{2s}\right]
=
\prod_p m_s(p^{-\sigma}), E [ ∣ ζ ( σ , X ) ∣ 2 s ] = p ∏ m s ( p − σ ) ,
where
m s ( p − σ ) = ∑ k = 0 ∞ ( ( s ) k k ! ) 2 p − 2 k σ . m_s(p^{-\sigma})=\sum_{k=0}^{\infty}\left(\frac{(s)_k}{k!}\right)^2p^{-2k\sigma}. m s ( p − σ ) = k = 0 ∑ ∞ ( k ! ( s ) k ) 2 p − 2 k σ .
The local expansion
m s ( p − σ ) = 1 + s 2 p − 2 σ + O s ( p − 4 σ ) m_s(p^{-\sigma})=1+s^2p^{-2\sigma}+O_s(p^{-4\sigma}) m s ( p − σ ) = 1 + s 2 p − 2 σ + O s ( p − 4 σ )
implies
( 1 − p − 2 σ ) s 2 m s ( p − σ ) = 1 + O s ( p − 4 σ ) , (1-p^{-2\sigma})^{s^2}m_s(p^{-\sigma})=1+O_s(p^{-4\sigma}), ( 1 − p − 2 σ ) s 2 m s ( p − σ ) = 1 + O s ( p − 4 σ ) ,
uniformly for σ ≥ 1 / 2 \sigma\ge1/2 σ ≥ 1/2 . Since these factors are strictly positive, their logarithms satisfy
log [ ( 1 − p − 2 σ ) s 2 m s ( p − σ ) ] = O s ( p − 4 σ ) . \log\left[(1-p^{-2\sigma})^{s^2}m_s(p^{-\sigma})\right]=O_s(p^{-4\sigma}). log [ ( 1 − p − 2 σ ) s 2 m s ( p − σ ) ] = O s ( p − 4 σ ) .
As
∑ p p − 4 σ ≤ ∑ p p − 2 < ∞ , \sum_p p^{-4\sigma}\le\sum_p p^{-2}<\infty, p ∑ p − 4 σ ≤ p ∑ p − 2 < ∞ ,
the product
A σ ( s ) : = ∏ p ( 1 − p − 2 σ ) s 2 m s ( p − σ ) A_\sigma(s):=\prod_p(1-p^{-2\sigma})^{s^2}m_s(p^{-\sigma}) A σ ( s ) := p ∏ ( 1 − p − 2 σ ) s 2 m s ( p − σ )
converges, and the convergence is uniform for σ ≥ 1 / 2 \sigma\ge1/2 σ ≥ 1/2 . Hence
A σ ( s ) ⟶ ∏ p ( 1 − p − 1 ) s 2 ∑ k = 0 ∞ ( ( s ) k k ! ) 2 p − k = a s . A_\sigma(s)\longrightarrow
\prod_p(1-p^{-1})^{s^2}\sum_{k=0}^{\infty}\left(\frac{(s)_k}{k!}\right)^2p^{-k}
=a_s. A σ ( s ) ⟶ p ∏ ( 1 − p − 1 ) s 2 k = 0 ∑ ∞ ( k ! ( s ) k ) 2 p − k = a s .
Moreover,
E [ ∣ ζ ( σ , X ) ∣ 2 s ] = ζ ( 2 σ ) s 2 A σ ( s ) . \mathbb{E}\left[\left|\zeta(\sigma,X)\right|^{2s}\right]=\zeta(2\sigma)^{s^2}A_\sigma(s). E [ ∣ ζ ( σ , X ) ∣ 2 s ] = ζ ( 2 σ ) s 2 A σ ( s ) .
Since
δ ζ ( 1 + δ ) ⟶ 1 , \delta\zeta(1+\delta)\longrightarrow1, δ ζ ( 1 + δ ) ⟶ 1 ,
we obtain (104) . We next establish the uniform moment bound (101) . Choose
q > max { 2 s , 2 } . q>\max\{2s,2\}. q > max { 2 s , 2 } .
By Lemma 3.1, the variables { X ( p ) } p \{X(p)\}_p { X ( p ) } p remain independent under P ~ σ , s \widetilde{\mathbb{P}}_{\sigma,s} P σ , s . Put
W p , σ : = X ( p ) p σ − X ( p ) W_{p,\sigma}:=\frac{X(p)}{p^\sigma-X(p)} W p , σ := p σ − X ( p ) X ( p )
and
Z p , σ : = ( log p ) ( W p , σ − E ~ σ , s [ W p , σ ] ) . Z_{p,\sigma}:=(\log p)\left(W_{p,\sigma}-\widetilde{\mathbb{E}}_{\sigma,s}[W_{p,\sigma}]\right). Z p , σ := ( log p ) ( W p , σ − E σ , s [ W p , σ ] ) .
The centered variables Z p , σ Z_{p,\sigma} Z p , σ are independent, and the prime-series representation of the logarithmic derivative gives
− L ′ ( σ , X ) = − E ~ σ , s [ L ′ ( σ , X ) ] + ∑ p Z p , σ , (105) -L'(\sigma,X)=-\widetilde{\mathbb{E}}_{\sigma,s}[L'(\sigma,X)]+\sum_p Z_{p,\sigma}, \tag*{(105)} − L ′ ( σ , X ) = − E σ , s [ L ′ ( σ , X )] + p ∑ Z p , σ , (105)
where the centered series converges in L 2 ( P ~ σ , s ) L^2(\widetilde{\mathbb{P}}_{\sigma,s}) L 2 ( P σ , s ) . By (93) and (94) ,
− δ E ~ σ , s [ L ′ ( σ , X ) ] ⟶ s (106) -\delta\widetilde{\mathbb{E}}_{\sigma,s}[L'(\sigma,X)]\longrightarrow s \tag*{(106)} − δ E σ , s [ L ′ ( σ , X )] ⟶ s (106)
and
∑ p E ~ σ , s [ ∣ Z p , σ ∣ 2 ] = O s ( δ − 2 ) . (107) \sum_p\widetilde{\mathbb{E}}_{\sigma,s}[|Z_{p,\sigma}|^2]=O_s(\delta^{-2}). \tag*{(107)} p ∑ E σ , s [ ∣ Z p , σ ∣ 2 ] = O s ( δ − 2 ) . (107)
Furthermore, since σ > 1 / 2 \sigma>1/2 σ > 1/2 ,
( log p ) ∣ W p , σ ∣ ≤ log p p − 1 . (\log p)|W_{p,\sigma}|\leq\frac{\log p}{\sqrt{p}-1}. ( log p ) ∣ W p , σ ∣ ≤ p − 1 log p .
The expression on the right is bounded uniformly over all primes. Consequently,
∣ Z p , σ ∣ ≤ C s (108) |Z_{p,\sigma}|\leq C_s \tag*{(108)} ∣ Z p , σ ∣ ≤ C s (108)
for every prime p p p and every σ > 1 / 2 \sigma>1/2 σ > 1/2 . Applying Rosenthal’s inequality to the real and imaginary parts of the centered sum, we obtain
E ~ σ , s [ ∣ ∑ p ≤ P Z p , σ ∣ q ] ≪ q ( ∑ p ≤ P E ~ σ , s [ ∣ Z p , σ ∣ 2 ] ) q / 2 + ∑ p ≤ P E ~ σ , s [ ∣ Z p , σ ∣ q ] . (109) \widetilde{\mathbb{E}}_{\sigma,s}\left[\left|\sum_{p\leq P}Z_{p,\sigma}\right|^q\right]\ll_q\left(\sum_{p\leq P}\widetilde{\mathbb{E}}_{\sigma,s}[|Z_{p,\sigma}|^2]\right)^{q/2}+\sum_{p\leq P}\widetilde{\mathbb{E}}_{\sigma,s}[|Z_{p,\sigma}|^q]. \tag*{(109)} E σ , s p ≤ P ∑ Z p , σ q ≪ q p ≤ P ∑ E σ , s [ ∣ Z p , σ ∣ 2 ] q /2 + p ≤ P ∑ E σ , s [ ∣ Z p , σ ∣ q ] . (109)
By (108) ,
∣ Z p , σ ∣ q ≤ C s q − 2 ∣ Z p , σ ∣ 2 . |Z_{p,\sigma}|^q\leq C_s^{q-2}|Z_{p,\sigma}|^2. ∣ Z p , σ ∣ q ≤ C s q − 2 ∣ Z p , σ ∣ 2 .
Thus (107) gives
∑ p E ~ σ , s [ ∣ Z p , σ ∣ q ] = O s , q ( δ − 2 ) . \sum_p\widetilde{\mathbb{E}}_{\sigma,s}[|Z_{p,\sigma}|^q]=O_{s,q}(\delta^{-2}). p ∑ E σ , s [ ∣ Z p , σ ∣ q ] = O s , q ( δ − 2 ) .
Since q > 2 q>2 q > 2 , it follows from (109) that
E ~ σ , s [ ∣ ∑ p ≤ P Z p , σ ∣ q ] ≪ s , q δ − q , (110) \widetilde{\mathbb{E}}_{\sigma,s}\left[\left|\sum_{p\leq P}Z_{p,\sigma}\right|^q\right]\ll_{s,q}\delta^{-q}, \tag*{(110)} E σ , s p ≤ P ∑ Z p , σ q ≪ s , q δ − q , (110)
uniformly in P P P , for σ \sigma σ sufficiently close to 1 / 2 1/2 1/2 . For each fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 , the partial sums converge in L 2 ( P ~ σ , s ) L^2(\widetilde{\mathbb P}_{\sigma,s}) L 2 ( P σ , s ) and hence in probability. Passing to an almost surely convergent subsequence and applying Fatou’s lemma to (110) , we obtain
E ~ σ , s [ ∣ ∑ p Z p , σ ∣ q ] ≪ s , q δ − q . (111) \widetilde{\mathbb E}_{\sigma,s}\left[\left|\sum_p Z_{p,\sigma}\right|^q\right]\ll_{s,q}\delta^{-q}. \tag*{(111)} E σ , s [ p ∑ Z p , σ q ] ≪ s , q δ − q . (111)
Combining (105) , (106) , and (111) , we conclude that there exists σ 0 > 1 / 2 \sigma_0 > 1/2 σ 0 > 1/2 such that
sup 1 / 2 < σ ≤ σ 0 E ~ σ , s [ ∣ δ L ′ ( σ , X ) ∣ q ] < ∞ . \sup_{1/2<\sigma\leq\sigma_0}\widetilde{\mathbb E}_{\sigma,s}\left[\left|\delta L'(\sigma,X)\right|^q\right]<\infty. 1/2 < σ ≤ σ 0 sup E σ , s [ ∣ δ L ′ ( σ , X ) ∣ q ] < ∞.
This proves (101) . Since q > 2 s q > 2s q > 2 s , the preceding bound implies that the family
{ ∣ δ L ′ ( σ , X ) ∣ 2 s : 1 2 < σ ≤ σ 0 } \left\{\left|\delta L'(\sigma,X)\right|^{2s}: \frac{1}{2}<\sigma\leq\sigma_0\right\} { ∣ δ L ′ ( σ , X ) ∣ 2 s : 2 1 < σ ≤ σ 0 }
is uniformly integrable under the corresponding tilted probability measures. On the other hand, Proposition 1.5 shows that
− δ L ′ ( σ , X ) → d s + G ( σ ↓ 1 / 2 ) -\delta L'(\sigma,X)\xrightarrow{\mathrm d}s+G\qquad(\sigma\downarrow1/2) − δ L ′ ( σ , X ) d s + G ( σ ↓ 1/2 )
under P ~ σ , s \widetilde{\mathbb P}_{\sigma,s} P σ , s . By the continuous mapping theorem and uniform integrability, we therefore obtain
E ~ σ , s [ ∣ δ L ′ ( σ , X ) ∣ 2 s ] ⟶ E [ ∣ s + G ∣ 2 s ] . (112) \widetilde{\mathbb E}_{\sigma,s}\left[\left|\delta L'(\sigma,X)\right|^{2s}\right]\longrightarrow\mathbb E\left[|s+G|^{2s}\right]. \tag*{(112)} E σ , s [ ∣ δ L ′ ( σ , X ) ∣ 2 s ] ⟶ E [ ∣ s + G ∣ 2 s ] . (112)
Combining (103) , (104) , and (112) , we obtain
δ s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] ⟶ a s E [ ∣ s + G ∣ 2 s ] . (113) \delta^{s^2+2s}\mathbb E\left[|\zeta'(\sigma,X)|^{2s}\right]\longrightarrow a_s\mathbb E\left[|s+G|^{2s}\right]. \tag*{(113)} δ s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] ⟶ a s E [ ∣ s + G ∣ 2 s ] . (113)
It remains to evaluate the Gaussian moment. Since G G G has density
1 π e − ∣ z ∣ 2 d 2 z \frac{1}{\pi}e^{-|z|^2}\,d^2z π 1 e − ∣ z ∣ 2 d 2 z
on C \mathbb C C , translation and polar coordinates give
E [ ∣ s + G ∣ 2 s ] = 2 e − s 2 ∫ 0 ∞ r 2 s + 1 e − r 2 I 0 ( 2 s r ) d r , \mathbb E\left[|s+G|^{2s}\right]=2e^{-s^2}\int_0^\infty r^{2s+1}e^{-r^2}I_0(2sr)\,dr, E [ ∣ s + G ∣ 2 s ] = 2 e − s 2 ∫ 0 ∞ r 2 s + 1 e − r 2 I 0 ( 2 sr ) d r ,
where I 0 I_0 I 0 is the modified Bessel function of the first kind of order zero. Using its series expansion
I 0 ( 2 s r ) = ∑ k = 0 ∞ s 2 k r 2 k ( k ! ) 2 I_0(2sr)=\sum_{k=0}^{\infty}\frac{s^{2k}r^{2k}}{(k!)^2} I 0 ( 2 sr ) = k = 0 ∑ ∞ ( k ! ) 2 s 2 k r 2 k
and applying Tonelli’s theorem, we obtain
E [ ∣ s + G ∣ 2 s ] = 2 e − s 2 ∑ k = 0 ∞ s 2 k ( k ! ) 2 ∫ 0 ∞ r 2 s + 2 k + 1 e − r 2 d r = e − s 2 ∑ k = 0 ∞ s 2 k Γ ( s + k + 1 ) ( k ! ) 2 = e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . \begin{aligned}
\mathbb E\left[|s+G|^{2s}\right]=2e^{-s^2}\sum_{k=0}^{\infty}\frac{s^{2k}}{(k!)^2}\int_0^\infty r^{2s+2k+1}e^{-r^2}\,dr \\
&=e^{-s^2}\sum_{k=0}^{\infty}\frac{s^{2k}\Gamma(s+k+1)}{(k!)^2} \\
&=e^{-s^2}\Gamma(s+1)\,{}_1F_1(s+1;1;s^2).
\end{aligned} E [ ∣ s + G ∣ 2 s ] = 2 e − s 2 k = 0 ∑ ∞ ( k ! ) 2 s 2 k ∫ 0 ∞ r 2 s + 2 k + 1 e − r 2 d r = e − s 2 k = 0 ∑ ∞ ( k ! ) 2 s 2 k Γ ( s + k + 1 ) = e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) .
Substituting this identity into (113) proves (102) and completes the proof. □ \square □
Complex moments via Fourier–Riesz analysis
In this section we extend the asymptotic formula for the moments of the derivative of the random zeta function to complex moment parameters. Throughout this section, let σ > 1 2 \sigma> \frac{1}{2} σ > 2 1 . Recall that
L ′ ( σ , X ) = − ∑ p ( log p ) X ( p ) p σ − X ( p ) . L'(\sigma, X) = -\sum_p \frac{(\log p)X(p)}{p^\sigma- X(p)}. L ′ ( σ , X ) = − p ∑ p σ − X ( p ) ( log p ) X ( p ) .
so that
ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) \zeta'(\sigma, X) = \zeta(\sigma, X)L'(\sigma, X) ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X )
almost surely. For x > 0 x > 0 x > 0 and s ∈ C s \in\mathbb{C} s ∈ C , we use the convention
x s : = exp ( s log x ) , x^s := \exp(s \log x), x s := exp ( s log x ) ,
where log x \log x log x denotes the real logarithm. If Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , we set 0 s : = 0 0^s := 0 0 s := 0 . Since 2 σ − 1 > 0 2\sigma- 1 > 0 2 σ − 1 > 0 , we also define
( 2 σ − 1 ) s : = exp ( s log ( 2 σ − 1 ) ) . (2\sigma- 1)^s := \exp\left(s\log(2\sigma- 1)\right). ( 2 σ − 1 ) s := exp ( s log ( 2 σ − 1 ) ) .
In particular,
( 2 σ − 1 ) s 2 + 2 s = exp ( ( s 2 + 2 s ) log ( 2 σ − 1 ) ) . (2\sigma- 1)^{s^2+2s} = \exp\left((s^2+2s)\log(2\sigma- 1)\right). ( 2 σ − 1 ) s 2 + 2 s = exp ( ( s 2 + 2 s ) log ( 2 σ − 1 ) ) .
Recall the arithmetic factor
a s : = ∏ p ( 1 − p − 1 ) s 2 ∑ m = 0 ∞ ( ( s ) m m ! ) 2 p − m . a_s := \prod_p (1-p^{-1})^{s^2}\sum_{m=0}^{\infty}\left(\frac{(s)_m}{m!}\right)^2p^{-m}. a s := p ∏ ( 1 − p − 1 ) s 2 m = 0 ∑ ∞ ( m ! ( s ) m ) 2 p − m .
Our main result in this section is the following asymptotic formula for complex moments of the derivative of the random zeta function.
Theorem 4.1 (Restatement of Theorem 1.2 ). For every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , one has
lim σ ↓ 1 / 2 ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 , 1 ; s 2 ) , (114) \lim_{\sigma\downarrow1/2}(2\sigma-1)^{s^2+2s}\mathbb{E}\left[|\zeta'(\sigma,X)|^{2s}\right]
=a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1,1;s^2), \tag*{(114)} σ ↓ 1/2 lim ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 , 1 ; s 2 ) , (114) where a s a_s a s is the arithmetic factor defined in (3) , and 1 F 1 ( a , b ; z ) {}_1F_1(a,b;z) 1 F 1 ( a , b ; z ) denotes the confluent hypergeometric function of the first kind. Moreover, the convergence is locally uniform in the half-plane Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 .
Combining Theorem 1.2 with Theorem 1.3 , which identifies the mean values of complex moments of ζ ′ ( σ + i t ) \zeta'(\sigma+it) ζ ′ ( σ + i t ) with the corresponding moments of the random model for each fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 , we obtain Theorem 1.4 .
Proof of Corollary 1.4 assuming Theorem 1.2
Deduction of Corollary 1.4 from Theorem 1.2. Assume the Lindelöf hypothesis. By Theorem 1.3 , for every fixed σ > 1 / 2 \sigma> 1/2 σ > 1/2 and every s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s) > 0 Re ( s ) > 0 , we have
lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] . (115) \lim_{T\to\infty}\frac{1}{T}\int_1^T|\zeta'(\sigma+it)|^{2s}\,dt
=\mathbb{E}\left[|\zeta'(\sigma,X)|^{2s}\right]. \tag*{(115)} T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t = E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] . (115)
Consequently, Theorem 1.2 gives
lim σ ↓ 1 / 2 ( 2 σ − 1 ) s 2 + 2 s ( lim T → ∞ 1 T ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t ) = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . \begin{aligned}
\lim_{\sigma\downarrow1/2}(2\sigma-1)^{s^2+2s}
\left(\lim_{T\to\infty}\frac{1}{T}\int_1^T|\zeta'(\sigma+it)|^{2s}\,dt\right) \\
&=a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2).
\end{aligned} σ ↓ 1/2 lim ( 2 σ − 1 ) s 2 + 2 s ( T → ∞ lim T 1 ∫ 1 T ∣ ζ ′ ( σ + i t ) ∣ 2 s d t ) = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) .
Moreover, since (115) holds unconditionally for 0 < Re ( s ) < 2 0<\operatorname{Re}(s)<2 0 < Re ( s ) < 2 , the same asymptotic formula holds unconditionally in this range. For s = 2 s=2 s = 2 , the result follows directly from [42 Theorem 1.7] . This completes the proof. □
It therefore remains to prove Theorem 1.2 , whose proof is given in Subsection 4.9. We begin by establishing several preliminary estimates that will be used in the proof.
Preliminary estimates
We first show the convergence and integrability properties of the random Euler product which will be used below.
Lemma 4.2. Let σ > 1 / 2 \sigma>1/2 σ > 1/2 . Define
ζ P ( σ , X ) = ∏ p ≤ P ( 1 − p − σ X ( p ) ) − 1 (116) \zeta_P(\sigma,X)=\prod_{p\leq P}(1-p^{-\sigma}X(p))^{-1} \tag*{(116)} ζ P ( σ , X ) = p ≤ P ∏ ( 1 − p − σ X ( p ) ) − 1 (116) and
L P ′ ( σ , X ) = − ∑ p ≤ P ( log p ) X ( p ) p σ − X ( p ) . (117) L'_P(\sigma,X)=-\sum_{p\leq P}\frac{(\log p)X(p)}{p^\sigma-X(p)}. \tag*{(117)} L P ′ ( σ , X ) = − p ≤ P ∑ p σ − X ( p ) ( log p ) X ( p ) . (117) Then
ζ P ( σ , X ) ⟶ ζ ( σ , X ) \zeta_P(\sigma,X)\longrightarrow\zeta(\sigma,X) ζ P ( σ , X ) ⟶ ζ ( σ , X ) and
L P ′ ( σ , X ) ⟶ L ′ ( σ , X ) L'_P(\sigma,X)\longrightarrow L'(\sigma,X) L P ′ ( σ , X ) ⟶ L ′ ( σ , X ) almost surely as P → ∞ P\to\infty P → ∞ . Moreover, for every q > 0 q>0 q > 0 ,
E [ ∣ ζ ( σ , X ) ∣ q ] < ∞ , E [ ∣ L ′ ( σ , X ) ∣ q ] < ∞ , E [ ∣ ζ ′ ( σ , X ) ∣ q ] < ∞ . (118) \mathbb{E}\left[|\zeta(\sigma,X)|^q\right]<\infty,\qquad
\mathbb{E}\left[|L'(\sigma,X)|^q\right]<\infty,\qquad
\mathbb{E}\left[|\zeta'(\sigma,X)|^q\right]<\infty. \tag*{(118)} E [ ∣ ζ ( σ , X ) ∣ q ] < ∞ , E [ ∣ L ′ ( σ , X ) ∣ q ] < ∞ , E [ ∣ ζ ′ ( σ , X ) ∣ q ] < ∞. (118)
Proof. Write
log ζ P ( σ , X ) = ∑ p ≤ P ∑ k ≥ 1 X ( p ) k k p k σ . \log\zeta_P(\sigma,X)=\sum_{p\leq P}\sum_{k\geq1}\frac{X(p)^k}{kp^{k\sigma}}. log ζ P ( σ , X ) = p ≤ P ∑ k ≥ 1 ∑ k p k σ X ( p ) k .
The contribution from k ≥ 2 k\geq2 k ≥ 2 converges absolutely, since
∑ p ∑ k ≥ 2 1 k p k σ < ∞ . \sum_p\sum_{k\geq2}\frac{1}{kp^{k\sigma}}<\infty. p ∑ k ≥ 2 ∑ k p k σ 1 < ∞.
The random variables X ( p ) p − σ X(p)p^{-\sigma} X ( p ) p − σ are independent and centered, and
∑ p E [ ∣ X ( p ) p σ ∣ 2 ] = ∑ p p − 2 σ < ∞ . \sum_p\mathbb{E}\left[\left|\frac{X(p)}{p^\sigma}\right|^2\right]
=\sum_p p^{-2\sigma}<\infty. p ∑ E [ p σ X ( p ) 2 ] = p ∑ p − 2 σ < ∞.
Hence Kolmogorov’s convergence theorem gives almost sure convergence of
∑ p X ( p ) p σ . \sum_p\frac{X(p)}{p^\sigma}. p ∑ p σ X ( p ) .
Thus log ζ P ( σ , X ) \log\zeta_P(\sigma, X) log ζ P ( σ , X ) converges almost surely, and therefore so does ζ P ( σ , X ) \zeta_P(\sigma, X) ζ P ( σ , X ) . For the logarithmic derivative, put
Z p , σ : = ( log p ) X ( p ) p σ − X ( p ) = ( log p ) ∑ k ≥ 1 X ( p ) k p k σ . Z_{p,\sigma} := \frac{(\log p)X(p)}{p^\sigma-X(p)} = (\log p)\sum_{k\ge1}\frac{X(p)^k}{p^{k\sigma}}. Z p , σ := p σ − X ( p ) ( log p ) X ( p ) = ( log p ) k ≥ 1 ∑ p k σ X ( p ) k .
Then
E [ Z p , σ ] = 0 \mathbb{E}\left[Z_{p,\sigma}\right] = 0 E [ Z p , σ ] = 0
and
E [ ∣ Z p , σ ∣ 2 ] ≪ σ ( log p ) 2 p 2 σ . \mathbb{E}\left[\lvert Z_{p,\sigma}\rvert^2\right] \ll_\sigma\frac{(\log p)^2}{p^{2\sigma}}. E [ ∣ Z p , σ ∣ 2 ] ≪ σ p 2 σ ( log p ) 2 .
Hence
∑ p E [ ∣ Z p , σ ∣ 2 ] < ∞ , \sum_p \mathbb{E}\left[\lvert Z_{p,\sigma}\rvert^2\right] < \infty, p ∑ E [ ∣ Z p , σ ∣ 2 ] < ∞ ,
and Kolmogorov’s convergence theorem again gives almost sure convergence of ∑ p Z p , σ \sum_p Z_{p,\sigma} ∑ p Z p , σ . For the moment estimates, independence gives, for the partial Euler products,
E [ ∣ ζ P ( σ , X ) ∣ q ] = ∏ p ≤ P ∫ T ∣ 1 − p − σ z ∣ − q d m T ( z ) . \mathbb{E}\left[\lvert\zeta_P(\sigma, X)\rvert^q\right] = \prod_{p\le P}\int_{\mathbb{T}}\lvert1-p^{-\sigma}z\rvert^{-q}\,{\rm d}m_{\mathbb{T}}(z). E [ ∣ ζ P ( σ , X ) ∣ q ] = p ≤ P ∏ ∫ T ∣ 1 − p − σ z ∣ − q d m T ( z ) .
For any q > 0 q>0 q > 0 , the local factors satisfy
∫ T ∣ 1 − p − σ z ∣ − q d m T ( z ) = 1 + O q , σ ( p − 2 σ ) , \int_{\mathbb{T}}\lvert1-p^{-\sigma}z\rvert^{-q}\,{\rm d}m_{\mathbb{T}}(z) = 1+O_{q,\sigma}(p^{-2\sigma}), ∫ T ∣ 1 − p − σ z ∣ − q d m T ( z ) = 1 + O q , σ ( p − 2 σ ) ,
and hence
sup P E [ ∣ ζ P ( σ , X ) ∣ q ] < ∞ . \sup_P\mathbb{E}\left[\lvert\zeta_P(\sigma, X)\rvert^q\right] < \infty. P sup E [ ∣ ζ P ( σ , X ) ∣ q ] < ∞.
Fatou’s lemma therefore gives
E [ ∣ ζ ( σ , X ) ∣ q ] < ∞ . \mathbb{E}\left[\lvert\zeta(\sigma, X)\rvert^q\right] < \infty. E [ ∣ ζ ( σ , X ) ∣ q ] < ∞.
For q ≥ 2 q\ge2 q ≥ 2 ,
∑ p E [ ∣ Z p , σ ∣ q ] ≪ q , σ ∑ p ( log p ) q p q σ < ∞ . \sum_p\mathbb{E}\left[\lvert Z_{p,\sigma}\rvert^q\right] \ll_{q,\sigma}\sum_p\frac{(\log p)^q}{p^{q\sigma}} < \infty. p ∑ E [ ∣ Z p , σ ∣ q ] ≪ q , σ p ∑ p q σ ( log p ) q < ∞.
Rosenthal’s inequality, first applied to finite partial sums and then combined with Fatou’s lemma, gives
E [ ∣ L ′ ( σ , X ) ∣ q ] < ∞ . \mathbb{E}\left[\lvert L'(\sigma, X)\rvert^q\right] < \infty. E [ ∣ L ′ ( σ , X ) ∣ q ] < ∞.
The case 0 < q < 2 0<q<2 0 < q < 2 follows from the L 2 L^2 L 2 bound. Hence, we have for any q > 0 q>0 q > 0 , E [ ∣ L ′ ( σ , X ) ∣ q ] < ∞ \mathbb{E}\left[\lvert L'(\sigma, X)\rvert^q\right]<\infty E [ ∣ L ′ ( σ , X ) ∣ q ] < ∞ . Finally, Using
∣ ζ ′ ( σ , X ) ∣ = ∣ ζ ( σ , X ) ∣ ∣ L ′ ( σ , X ) ∣ , \lvert\zeta'(\sigma, X)\rvert= \lvert\zeta(\sigma, X)\rvert\lvert L'(\sigma, X)\rvert, ∣ ζ ′ ( σ , X )∣ = ∣ ζ ( σ , X )∣ ∣ L ′ ( σ , X )∣ ,
and Hölder’s inequality, we have E [ ∣ ζ ′ ( σ , X ) ∣ q ] < ∞ \mathbb{E}\left[\lvert\zeta'(\sigma, X)\rvert^q\right]<\infty E [ ∣ ζ ′ ( σ , X ) ∣ q ] < ∞ for any q > 0 q>0 q > 0 .
We now introduce a local moment function that extends the quantity m s ( r ) m_s(r) m s ( r ) defined in (63) to complex values of s s s .
Lemma 4.3. For 0 ≤ r < 1 0 \le r < 1 0 ≤ r < 1 and s ∈ C s \in\mathbb{C} s ∈ C , define
m s ( r ) : = ∫ T ∣ 1 − r z ∣ − 2 s d m T ( z ) . (119) m_s(r) := \int_{\mathbb{T}} |1-rz|^{-2s}\,\mathrm{d}m_{\mathbb{T}}(z). \tag*{(119)} m s ( r ) := ∫ T ∣1 − r z ∣ − 2 s d m T ( z ) . (119) Then, for 0 ≤ r < 1 0 \le r < 1 0 ≤ r < 1 ,
m s ( r ) = ∑ m = 0 ∞ ( ( s ) m m ! ) 2 r 2 m . (120) m_s(r) = \sum_{m=0}^{\infty} \left(\frac{(s)_m}{m!}\right)^2 r^{2m}. \tag*{(120)} m s ( r ) = m = 0 ∑ ∞ ( m ! ( s ) m ) 2 r 2 m . (120) Moreover, for every compact set K ⊂ C K \subset\mathbb{C} K ⊂ C ,
m s ( r ) = 1 + s 2 r 2 + O K ( r 4 ) (121) m_s(r) = 1+s^2r^2+O_K(r^4) \tag*{(121)} m s ( r ) = 1 + s 2 r 2 + O K ( r 4 ) (121) uniformly for s ∈ K s \in K s ∈ K and 0 ≤ r ≤ 2 − 1 / 2 0 \le r \le2^{-1/2} 0 ≤ r ≤ 2 − 1/2 .
Proof. Since r < 1 r<1 r < 1 , the number 1 − r z 1-rz 1 − r z lies in the open right half-plane for z ∈ T z \in\mathbb{T} z ∈ T . Hence the principal logarithm may be used and
∣ 1 − r z ∣ − 2 s = ( 1 − r z ) − s ( 1 − r z ‾ ) − s . |1-rz|^{-2s}=(1-rz)^{-s}(1-r\overline{z})^{-s}. ∣1 − r z ∣ − 2 s = ( 1 − r z ) − s ( 1 − r z ) − s .
Using
( 1 − r z ) − s = ∑ m ≥ 0 ( s ) m m ! r m z m (1-rz)^{-s}=\sum_{m\ge0}\frac{(s)_m}{m!}r^m z^m ( 1 − r z ) − s = m ≥ 0 ∑ m ! ( s ) m r m z m
and Haar orthogonality gives (120) . The expansion (121) follows immediately. □
We shall use the following standard estimates for sums over primes, which follow from the classical properties of the Riemann zeta function near its pole at 1.
Lemma 4.4. As δ ↓ 0 \delta\downarrow0 δ ↓ 0 , the following asymptotic estimates hold, where all sums are taken over primes:
∑ p 1 p 1 + δ = log 1 δ + O ( 1 ) , (122) \sum_p \frac{1}{p^{1+\delta}}=\log\frac{1}{\delta}+O(1), \tag*{(122)} p ∑ p 1 + δ 1 = log δ 1 + O ( 1 ) , (122) ∑ p log p p 1 + δ = 1 δ + O ( 1 ) , (123) \sum_p \frac{\log p}{p^{1+\delta}}=\frac{1}{\delta}+O(1), \tag*{(123)} p ∑ p 1 + δ log p = δ 1 + O ( 1 ) , (123) ∑ p ( log p ) 2 p 1 + δ = 1 δ 2 + O ( 1 ) . (124) \sum_p \frac{(\log p)^2}{p^{1+\delta}}=\frac{1}{\delta^2}+O(1). \tag*{(124)} p ∑ p 1 + δ ( log p ) 2 = δ 2 1 + O ( 1 ) . (124) These asymptotic estimates remain valid if any fixed finite set of primes is omitted from the sums.
Proof. The first estimate follows from
log ζ ( 1 + δ ) = ∑ p ∑ k ≥ 1 1 k p k ( 1 + δ ) = log 1 δ + O ( 1 ) , \log\zeta(1+\delta)=\sum_p\sum_{k\ge1}\frac{1}{kp^{k(1+\delta)}}=\log\frac{1}{\delta}+O(1), log ζ ( 1 + δ ) = p ∑ k ≥ 1 ∑ k p k ( 1 + δ ) 1 = log δ 1 + O ( 1 ) ,
since the contribution from k ≥ 2 k\ge2 k ≥ 2 is uniformly bounded. Similarly,
− ζ ′ ζ ( 1 + δ ) = ∑ p ∑ k ≥ 1 log p p k ( 1 + δ ) = 1 δ + O ( 1 ) , -\frac{\zeta'}{\zeta}(1+\delta)=\sum_p\sum_{k\ge1}\frac{\log p}{p^{k(1+\delta)}}=\frac{1}{\delta}+O(1), − ζ ζ ′ ( 1 + δ ) = p ∑ k ≥ 1 ∑ p k ( 1 + δ ) log p = δ 1 + O ( 1 ) ,
which gives (123) . Differentiating with respect to δ \delta δ yields
∑ p ∑ k ≥ 1 k ( log p ) 2 p k ( 1 + δ ) = 1 δ 2 + O ( 1 ) , \sum_p\sum_{k\ge1}\frac{k(\log p)^2}{p^{k(1+\delta)}}=\frac{1}{\delta^2}+O(1), p ∑ k ≥ 1 ∑ p k ( 1 + δ ) k ( log p ) 2 = δ 2 1 + O ( 1 ) ,
and the k ≥ 2 k\ge2 k ≥ 2 terms are again uniformly bounded. □ \square □
We also require the following uniform tail estimate.
Lemma 4.5. There exist P 0 ≥ 2 P_0\ge2 P 0 ≥ 2 and c > 0 c>0 c > 0 such that, uniformly for P ≥ P 0 P\ge P_0 P ≥ P 0 and 0 < δ ≤ 1 / 2 0<\delta\le1/2 0 < δ ≤ 1/2 ,
∑ p ≥ P ( log p ) 2 p 1 + δ ≥ c P − δ ( log P δ + 1 δ 2 ) . (125) \sum_{p\ge P}\frac{(\log p)^2}{p^{1+\delta}}\ge cP^{-\delta}\left(\frac{\log P}{\delta}+\frac{1}{\delta^2}\right). \tag*{(125)} p ≥ P ∑ p 1 + δ ( log p ) 2 ≥ c P − δ ( δ log P + δ 2 1 ) . (125)
Proof. Let
ϑ ( x ) : = ∑ p ≤ x log p . \vartheta(x):=\sum_{p\le x}\log p. ϑ ( x ) := p ≤ x ∑ log p .
By the prime number theorem (see [1 Chapter 4, Theorem 4.4] ), after increasing P 0 P_0 P 0 if necessary, there exists c 0 > 0 c_0>0 c 0 > 0 such that
ϑ ( 2 x ) − ϑ ( x ) ≥ c 0 x , x ≥ P 0 . \vartheta(2x)-\vartheta(x)\ge c_0x,\qquad x\ge P_0. ϑ ( 2 x ) − ϑ ( x ) ≥ c 0 x , x ≥ P 0 .
Choosing a P P P with P ≥ P 0 P\ge P_0 P ≥ P 0 . Put P k = 2 k P P_k=2^kP P k = 2 k P . Then
∑ P k < p ≤ 2 P k ( log p ) 2 p 1 + δ ≥ ( log P k ) ( 2 P k ) − 1 − δ ( ϑ ( 2 P k ) − ϑ ( P k ) ) ≫ ( log P k ) P k − δ . \begin{aligned}
\sum_{P_k<p\le2P_k}\frac{(\log p)^2}{p^{1+\delta}}
\ge(\log P_k)(2P_k)^{-1-\delta}\bigl(\vartheta(2P_k)-\vartheta(P_k)\bigr) \\
&\gg(\log P_k)P_k^{-\delta}.
\end{aligned} P k < p ≤ 2 P k ∑ p 1 + δ ( log p ) 2 ≥ ( log P k ) ( 2 P k ) − 1 − δ ( ϑ ( 2 P k ) − ϑ ( P k ) ) ≫ ( log P k ) P k − δ .
Here and below, for nonnegative quantities f f f and g g g , we use the notation f ≪ g f\ll g f ≪ g (equivalently, g ≫ f g\gg f g ≫ f ) to mean that f ≤ c g f\le cg f ≤ c g for some positive constant c c c , and f ≍ g f\asymp g f ≍ g to mean that both f ≪ g f\ll g f ≪ g and g ≪ f g\ll f g ≪ f . Summing over k ≥ 0 k\ge0 k ≥ 0 gives
∑ p ≥ P ( log p ) 2 p 1 + δ ≫ P − δ ∑ k = 0 ∞ ( log P + k log 2 ) e − k δ log 2 . \sum_{p\ge P}\frac{(\log p)^2}{p^{1+\delta}}
\gg P^{-\delta}\sum_{k=0}^{\infty}(\log P+k\log2)e^{-k\delta\log2}. p ≥ P ∑ p 1 + δ ( log p ) 2 ≫ P − δ k = 0 ∑ ∞ ( log P + k log 2 ) e − k δ l o g 2 .
Since, uniformly for 0 < δ ≤ 1 / 2 0<\delta\le1/2 0 < δ ≤ 1/2 ,
∑ k ≥ 0 e − k δ log 2 ≍ δ − 1 , ∑ k ≥ 0 k e − k δ log 2 ≍ δ − 2 , \sum_{k\ge0}e^{-k\delta\log2}\asymp\delta^{-1},\qquad
\sum_{k\ge0}ke^{-k\delta\log2}\asymp\delta^{-2}, k ≥ 0 ∑ e − k δ l o g 2 ≍ δ − 1 , k ≥ 0 ∑ k e − k δ l o g 2 ≍ δ − 2 ,
the assertion follows. □ \square □
Finally, we record the real Euler-product bound required later.
Lemma 4.6. Let 0 < a 0 ≤ a ≤ a 1 0<a_0\le a\le a_1 0 < a 0 ≤ a ≤ a 1 . Uniformly for a ∈ [ a 0 , a 1 ] a\in[a_0,a_1] a ∈ [ a 0 , a 1 ] and σ \sigma σ sufficiently close to 1 / 2 1/2 1/2 ,
( 2 σ − 1 ) a 2 ∏ p m a ( p − σ ) ≪ a 0 , a 1 1. (126) (2\sigma-1)^{a^2}\prod_p m_a(p^{-\sigma})\ll_{a_0,a_1}1. \tag*{(126)} ( 2 σ − 1 ) a 2 p ∏ m a ( p − σ ) ≪ a 0 , a 1 1. (126)
Here, m a ( r ) m_a(r) m a ( r ) is defined by (119) . Consequently,
( 2 σ − 1 ) a 2 E [ ∣ ζ ( σ , X ) ∣ 2 a ] ≪ a 0 , a 1 1. (127) (2\sigma-1)^{a^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2a}\right]\ll_{a_0,a_1}1. \tag*{(127)} ( 2 σ − 1 ) a 2 E [ ∣ ζ ( σ , X ) ∣ 2 a ] ≪ a 0 , a 1 1. (127)
Proof. By (121) ,
( 1 − p − 2 σ ) a 2 m a ( p − σ ) = 1 + O a 0 , a 1 ( p − 4 σ ) . (1-p^{-2\sigma})a^2m_a(p^{-\sigma})=1+O_{a_0,a_1}(p^{-4\sigma}). ( 1 − p − 2 σ ) a 2 m a ( p − σ ) = 1 + O a 0 , a 1 ( p − 4 σ ) .
Therefore
A σ ( a ) : = ∏ p ( 1 − p − 2 σ ) a 2 m a ( p − σ ) A_\sigma(a):=\prod_p(1-p^{-2\sigma})^{a^2}m_a(p^{-\sigma}) A σ ( a ) := p ∏ ( 1 − p − 2 σ ) a 2 m a ( p − σ )
is uniformly bounded for σ \sigma σ sufficiently close to 1 / 2 1/2 1/2 . Since
∏ p ( 1 − p − 2 σ ) − a 2 = ζ ( 2 σ ) a 2 , \prod_p(1-p^{-2\sigma})^{-a^2}=\zeta(2\sigma)^{a^2}, p ∏ ( 1 − p − 2 σ ) − a 2 = ζ ( 2 σ ) a 2 ,
we obtain
( 2 σ − 1 ) a 2 ∏ p m a ( p − σ ) = ( ( 2 σ − 1 ) ζ ( 2 σ ) ) a 2 A σ ( a ) ≪ 1. (2\sigma-1)^{a^2}\prod_p m_a(p^{-\sigma})=((2\sigma-1)\zeta(2\sigma))^{a^2}A_\sigma(a)\ll1. ( 2 σ − 1 ) a 2 p ∏ m a ( p − σ ) = (( 2 σ − 1 ) ζ ( 2 σ ) ) a 2 A σ ( a ) ≪ 1.
This proves (126) . For the partial Euler products,
E [ ∣ ζ P ( σ , X ) ∣ 2 a ] = ∏ p ≤ P m a ( p − σ ) . \mathbb{E}\left[|\zeta_P(\sigma,X)|^{2a}\right]=\prod_{p\le P}m_a(p^{-\sigma}). E [ ∣ ζ P ( σ , X ) ∣ 2 a ] = p ≤ P ∏ m a ( p − σ ) .
Letting P → ∞ P\to\infty P → ∞ and using Fatou’s lemma gives
E [ ∣ ζ ( σ , X ) ∣ 2 a ] ≤ ∏ p m a ( p − σ ) , \mathbb{E}\left[|\zeta(\sigma,X)|^{2a}\right]\le\prod_pm_a(p^{-\sigma}), E [ ∣ ζ ( σ , X ) ∣ 2 a ] ≤ p ∏ m a ( p − σ ) ,
which proves (127) .
The weighted joint Fourier transform and Proof of Proposition 1.6
We begin by establishing an Euler product representation for the weighted joint Fourier transform.
Lemma 4.7. Let σ > 1 / 2 \sigma>1/2 σ > 1/2 , s ∈ C s\in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 , and ( λ , u ) ∈ R 2 (\lambda,u)\in\mathbb{R}^2 ( λ , u ) ∈ R 2 . Recall the weighted joint Fourier transform H σ , s ( λ , u ) H_{\sigma,s}(\lambda,u) H σ , s ( λ , u ) defined in (13) . Then H σ , s ( λ , u ) H_{\sigma,s}(\lambda,u) H σ , s ( λ , u ) admits the following Euler-product factorization,
H σ , s ( λ , u ) = ( 2 σ − 1 ) s 2 ∏ p p r i m e Φ p , σ ( s ; λ , u ) , (128) H_{\sigma,s}(\lambda,u)=(2\sigma-1)^{s^2}\prod_{p\ \mathrm{prime}}\Phi_{p,\sigma}(s;\lambda,u), \tag*{(128)} H σ , s ( λ , u ) = ( 2 σ − 1 ) s 2 p prime ∏ Φ p , σ ( s ; λ , u ) , (128) where, for each prime p p p ,
Φ p , σ ( s ; λ , u ) : = ∫ T ∣ 1 − p − σ z ∣ − 2 s exp { − i ( 2 σ − 1 ) log p [ λ Re p − σ z 1 − p − σ z + u Im p − σ z 1 − p − σ z ] } d m T ( z ) . (129) \Phi_{p,\sigma}(s;\lambda,u):=\int_{\mathbb{T}}|1-p^{-\sigma}z|^{-2s}\exp\left\{-i(2\sigma-1)\log p\left[\lambda\operatorname{Re}\frac{p^{-\sigma}z}{1-p^{-\sigma}z}+u\operatorname{Im}\frac{p^{-\sigma}z}{1-p^{-\sigma}z}\right]\right\}\,dm_{\mathbb{T}}(z). \tag*{(129)} Φ p , σ ( s ; λ , u ) := ∫ T ∣1 − p − σ z ∣ − 2 s exp { − i ( 2 σ − 1 ) log p [ λ Re 1 − p − σ z p − σ z + u Im 1 − p − σ z p − σ z ] } d m T ( z ) . (129)
Proof. Let ζ P \zeta_P ζ P and L P ′ L'_P L P ′ be defined in (116) and (117) , respectively. By independence,
E [ ∣ ζ P ( σ , X ) ∣ 2 s e i λ ( 2 σ − 1 ) Re L P ′ ( σ , X ) + i u ( 2 σ − 1 ) Im L P ′ ( σ , X ) ] = ∏ p ≤ P Φ p , σ ( s ; λ , u ) . \mathbb{E}\left[|\zeta_P(\sigma,X)|^{2s}e^{i\lambda(2\sigma-1)\operatorname{Re}L'_P(\sigma,X)+iu(2\sigma-1)\operatorname{Im}L'_P(\sigma,X)}\right]
=\prod_{p\le P}\Phi_{p,\sigma}(s;\lambda,u). E [ ∣ ζ P ( σ , X ) ∣ 2 s e iλ ( 2 σ − 1 ) Re L P ′ ( σ , X ) + i u ( 2 σ − 1 ) Im L P ′ ( σ , X ) ] = p ≤ P ∏ Φ p , σ ( s ; λ , u ) .
By Lemma 4.2, the integrands converge almost surely to the corresponding infinite-product integrand. Since λ , u \lambda,u λ , u are real, the modulus of the finite-product integrand is
∣ ζ P ( σ , X ) ∣ 2 Re ( s ) . |\zeta_P(\sigma,X)|^{2\operatorname{Re}(s)}. ∣ ζ P ( σ , X ) ∣ 2 Re ( s ) .
Choose q > 1 q>1 q > 1 . The proof of Lemma 4.2 gives
sup P E [ ∣ ζ P ( σ , X ) ∣ 2 Re ( s ) q ] < ∞ . \sup_P \mathbb{E}\left[|\zeta_P(\sigma,X)|^{2\operatorname{Re}(s)q}\right]<\infty. P sup E [ ∣ ζ P ( σ , X ) ∣ 2 Re ( s ) q ] < ∞.
Hence the integrands are uniformly integrable. Vitali’s convergence theorem therefore allows us to pass to the limit in the expectation, which proves (128) .
We next obtain a uniform expansion of the local factors.
Lemma 4.8. Let K K K be a compact subset of the open right half-plane { s ∈ C : Re ( s ) > 0 } \{s\in\mathbb{C}:\operatorname{Re}(s)>0\} { s ∈ C : Re ( s ) > 0 } and let B 0 ⊂ C B_0\subset\mathbb{C} B 0 ⊂ C be bounded. For each prime p p p , set t p = ( 2 σ − 1 ) log p t_p=(2\sigma-1)\log p t p = ( 2 σ − 1 ) log p . Uniformly for s ∈ K s\in K s ∈ K , ( λ , u ) ∈ B 0 (\lambda,u)\in B_0 ( λ , u ) ∈ B 0 , and σ \sigma σ sufficiently close to 1 / 2 1/2 1/2 ,
Φ p , σ ( s ; λ , u ) = m s ( p − σ ) − i s λ t p p − 2 σ − λ 2 + u 2 4 t p 2 p − 2 σ + R p , σ , (130) \Phi_{p,\sigma}(s;\lambda,u)=m_s(p^{-\sigma})-is\lambda t_p p^{-2\sigma}-\frac{\lambda^2+u^2}{4}t_p^2p^{-2\sigma}+R_{p,\sigma}, \tag*{(130)} Φ p , σ ( s ; λ , u ) = m s ( p − σ ) − i s λ t p p − 2 σ − 4 λ 2 + u 2 t p 2 p − 2 σ + R p , σ , (130) where
∣ R p , σ ∣ ≪ K , B 0 t p p − 4 σ + t p 2 p − 3 σ + t p 3 p − 3 σ . (131) |R_{p,\sigma}|\ll_{K,B_0}t_p p^{-4\sigma}+t_p^2p^{-3\sigma}+t_p^3p^{-3\sigma}. \tag*{(131)} ∣ R p , σ ∣ ≪ K , B 0 t p p − 4 σ + t p 2 p − 3 σ + t p 3 p − 3 σ . (131) Here, Φ p , σ ( s ; λ , u ) \Phi_{p,\sigma}(s;\lambda,u) Φ p , σ ( s ; λ , u ) is defined in (129) . Consequently,
∑ p sup s ∈ K ( λ , u ) ∈ B 0 ∣ R p , σ ∣ = O K , B 0 ( 2 σ − 1 ) . (132) \sum_p\sup_{\substack{s\in K\\(\lambda,u)\in B_0}}|R_{p,\sigma}|=O_{K,B_0}(2\sigma-1). \tag*{(132)} p ∑ s ∈ K ( λ , u ) ∈ B 0 sup ∣ R p , σ ∣ = O K , B 0 ( 2 σ − 1 ) . (132)
Proof. Write
t = t p , W r ( z ) = r z 1 − r z , t=t_p,\qquad W_r(z)=\frac{rz}{1-rz}, t = t p , W r ( z ) = 1 − r z r z ,
and
Q λ , u ( z ) = λ Re W r ( z ) + u Im W r ( z ) . Q_{\lambda,u}(z)=\lambda\operatorname{Re}W_r(z)+u\operatorname{Im}W_r(z). Q λ , u ( z ) = λ Re W r ( z ) + u Im W r ( z ) .
So, we have
t p Q λ , u ( z ) = O B 0 ( 2 σ − 1 ) t_pQ_{\lambda,u}(z)=O_{B_0}(2\sigma-1) t p Q λ , u ( z ) = O B 0 ( 2 σ − 1 )
uniformly in primes p p p and z z z with ∣ z ∣ = 1 |z|=1 ∣ z ∣ = 1 . We note that
W r ( z ‾ ) = W r ( z ) ‾ , W_r(\overline{z})=\overline{W_r(z)}, W r ( z ) = W r ( z ) ,
we have
Im W r ( z ‾ ) = − Im W r ( z ) . \operatorname{Im}W_r(\overline{z})=-\operatorname{Im}W_r(z). Im W r ( z ) = − Im W r ( z ) .
By the change of variables z ↦ z ‾ z\mapsto\overline{z} z ↦ z and the invariance of Haar measure on T \mathbb{T} T under complex conjugation,
∫ T ∣ 1 − r z ∣ − 2 s Im W r ( z ) d m T ( z ) = 0. \int_{\mathbb{T}}|1-rz|^{-2s}\operatorname{Im}W_r(z)\,d m_{\mathbb{T}}(z)=0. ∫ T ∣1 − r z ∣ − 2 s Im W r ( z ) d m T ( z ) = 0.
Differentiating m s ( r ) m_s(r) m s ( r ) gives
m s ′ ( r ) = 2 s r ∫ T ∣ 1 − r z ∣ − 2 s Re W r ( z ) d m T ( z ‾ ) . m_s'(r)=\frac{2s}{r}\int_{\mathbb{T}}|1-rz|^{-2s}\operatorname{Re}W_r(z)\,d m_{\mathbb{T}}(\overline{z}). m s ′ ( r ) = r 2 s ∫ T ∣1 − r z ∣ − 2 s Re W r ( z ) d m T ( z ) .
Since K ⊂ { Re ( s ) > 0 } K\subset\{\operatorname{Re}(s)>0\} K ⊂ { Re ( s ) > 0 } , we have inf s ∈ K ∣ s ∣ > 0 \inf_{s\in K}|s|>0 inf s ∈ K ∣ s ∣ > 0 , so division by s s s is uniform on K K K . From (121) ,
m s ′ ( r ) = 2 s 2 r + O K ( r 3 ) , m_s'(r)=2s^2r+O_K(r^3), m s ′ ( r ) = 2 s 2 r + O K ( r 3 ) ,
and hence
∫ T ∣ 1 − r z ∣ − 2 s Q λ , u ( z ) d m T ( z ) = s λ r 2 + O K , B 0 ( r 4 ) . (133) \int_{\mathbb{T}} |1-rz|^{-2s}Q_{\lambda,u}(z)\,{\rm d}m_{\mathbb{T}}(z)=s\lambda r^2+O_{K,B_0}(r^4). \tag*{(133)} ∫ T ∣1 − r z ∣ − 2 s Q λ , u ( z ) d m T ( z ) = s λ r 2 + O K , B 0 ( r 4 ) . (133)
Moreover,
W r ( z ) = r z + O ( r 2 ) , ∣ 1 − r z ∣ − 2 s = 1 + O K ( r ) , W_r(z)=rz+O(r^2),\qquad|1-rz|^{-2s}=1+O_K(r), W r ( z ) = r z + O ( r 2 ) , ∣1 − r z ∣ − 2 s = 1 + O K ( r ) ,
so Haar orthogonality gives
∫ T ∣ 1 − r z ∣ − 2 s Q λ , u ( z ) 2 d m T ( z ) = λ 2 + u 2 2 r 2 + O K , B 0 ( r 3 ) . (134) \int_{\mathbb{T}} |1-rz|^{-2s}Q_{\lambda,u}(z)^2\,{\rm d}m_{\mathbb{T}}(z)=\frac{\lambda^2+u^2}{2}r^2+O_{K,B_0}(r^3). \tag*{(134)} ∫ T ∣1 − r z ∣ − 2 s Q λ , u ( z ) 2 d m T ( z ) = 2 λ 2 + u 2 r 2 + O K , B 0 ( r 3 ) . (134)
Taylor’s formula yields
e − i t Q = 1 − i t Q − t 2 Q 2 2 + O B 0 ( t 3 r 3 ) . e^{-itQ}=1-itQ-\frac{t^2Q^2}{2}+O_{B_0}(t^3r^3). e − i tQ = 1 − i tQ − 2 t 2 Q 2 + O B 0 ( t 3 r 3 ) .
Substitution of (133) and (134) gives (130) . Finally,
∑ p t p p − 4 σ ≪ ( 2 σ − 1 ) ∑ n ≥ 2 log n n 2 ≪ ( 2 σ − 1 ) , ∑ p t p 2 p − 3 σ ≪ ( 2 σ − 1 ) 2 ∑ n ≥ 2 ( log n ) 2 n 3 / 2 ≪ ( 2 σ − 1 ) 2 , ∑ p t p 3 p − 3 σ ≪ ( 2 σ − 1 ) 3 ∑ n ≥ 2 ( log n ) 3 n 3 / 2 ≪ ( 2 σ − 1 ) 3 . \begin{aligned}
\sum_p t_p p^{-4\sigma}&\ll(2\sigma-1)\sum_{n\geq2}\frac{\log n}{n^2}\ll(2\sigma-1),\\
\sum_p t_p^2p^{-3\sigma}&\ll(2\sigma-1)^2\sum_{n\geq2}\frac{(\log n)^2}{n^{3/2}}\ll(2\sigma-1)^2,\\
\sum_p t_p^3p^{-3\sigma}&\ll(2\sigma-1)^3\sum_{n\geq2}\frac{(\log n)^3}{n^{3/2}}\ll(2\sigma-1)^3.
\end{aligned} p ∑ t p p − 4 σ p ∑ t p 2 p − 3 σ p ∑ t p 3 p − 3 σ ≪ ( 2 σ − 1 ) n ≥ 2 ∑ n 2 log n ≪ ( 2 σ − 1 ) , ≪ ( 2 σ − 1 ) 2 n ≥ 2 ∑ n 3/2 ( log n ) 2 ≪ ( 2 σ − 1 ) 2 , ≪ ( 2 σ − 1 ) 3 n ≥ 2 ∑ n 3/2 ( log n ) 3 ≪ ( 2 σ − 1 ) 3 .
This proves (132) .
Combining the local expansion in Lemma 4.8 with the prime-sum estimates in Lemma 4.4 , we now prove Proposition 1.6 , which establishes the asymptotic behaviour of the weighted joint Fourier transform as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
Proof of Proposition 1.6 . Fix a compact subset
K ⊂ { Re ( s ) > 0 } K\subset\{\operatorname{Re}(s)>0\} K ⊂ { Re ( s ) > 0 }
and a bounded closed polydisc B 0 ⊂ C 2 B_0\subset\mathbb{C}^2 B 0 ⊂ C 2 . By (121) , there exists P 0 = P 0 ( K ) P_0=P_0(K) P 0 = P 0 ( K ) such that, for all p > P 0 p>P_0 p > P 0 , s ∈ K s\in K s ∈ K , and σ ≥ 1 / 2 \sigma\geq1/2 σ ≥ 1/2 ,
∣ m s ( p − σ ) ∣ ≥ 3 4 . |m_s(p^{-\sigma})|\geq\frac{3}{4}. ∣ m s ( p − σ ) ∣ ≥ 4 3 .
Here m s ( r ) m_s(r) m s ( r ) is defined by (119) . Thus we divide only by the large-prime factors and write
∏ p Φ p , σ = ( ∏ p ≤ P 0 Φ p , σ ) ( ∏ p > P 0 m s ( p − σ ) ) ∏ p > P 0 Φ p , σ m s ( p − σ ) . (135) \prod_p\Phi_{p,\sigma}=\left(\prod_{p\leq P_0}\Phi_{p,\sigma}\right)\left(\prod_{p>P_0}m_s(p^{-\sigma})\right)\prod_{p>P_0}\frac{\Phi_{p,\sigma}}{m_s(p^{-\sigma})}. \tag*{(135)} p ∏ Φ p , σ = p ≤ P 0 ∏ Φ p , σ p > P 0 ∏ m s ( p − σ ) p > P 0 ∏ m s ( p − σ ) Φ p , σ . (135)
where Φ p , σ ( s ; λ , u ) \Phi_{p,\sigma}(s;\lambda,u) Φ p , σ ( s ; λ , u ) is defined by (129) . For the finitely many primes p ≤ P 0 p\leq P_0 p ≤ P 0 , as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 ,
Φ p , σ ( s ; λ , u ) ⟶ m s ( p − 1 / 2 ) (136) \Phi_{p,\sigma}(s;\lambda,u)\longrightarrow m_s(p^{-1/2}) \tag*{(136)} Φ p , σ ( s ; λ , u ) ⟶ m s ( p − 1/2 ) (136)
uniformly for s ∈ K s\in K s ∈ K and ( λ , u ) ∈ B 0 (\lambda,u)\in B_0 ( λ , u ) ∈ B 0 . For the unperturbed large-prime factors, (121) gives
( 1 − p − 2 σ ) s 2 m s ( p − σ ) = 1 + O K ( p − 4 σ ) (1-p^{-2\sigma})^{s^2}m_s(p^{-\sigma})=1+O_K(p^{-4\sigma}) ( 1 − p − 2 σ ) s 2 m s ( p − σ ) = 1 + O K ( p − 4 σ )
uniformly for s ∈ K s \in K s ∈ K and σ ≥ 1 / 2 \sigma\ge1/2 σ ≥ 1/2 . Since p − 4 σ ≤ p − 2 p^{-4\sigma} \le p^{-2} p − 4 σ ≤ p − 2 , the corresponding infinite product converges locally uniformly, and as σ ↓ 1 2 \sigma\downarrow\frac{1}{2} σ ↓ 2 1 ,
∏ p > P 0 ( 1 − p − 2 σ ) s 2 m s ( p − σ ) ⟶ ∏ p > P 0 ( 1 − p − 1 ) s 2 m s ( p − 1 / 2 ) \prod_{p>P_0} (1-p^{-2\sigma})^{s^2}m_s(p^{-\sigma}) \longrightarrow\prod_{p>P_0}(1-p^{-1})^{s^2}m_s(p^{-1/2}) p > P 0 ∏ ( 1 − p − 2 σ ) s 2 m s ( p − σ ) ⟶ p > P 0 ∏ ( 1 − p − 1 ) s 2 m s ( p − 1/2 )
uniformly for s ∈ K s \in K s ∈ K . Moreover,
( 2 σ − 1 ) s 2 ∏ p > P 0 m s ( p − σ ) = ( ( 2 σ − 1 ) ζ ( 2 σ ) ) s 2 ∏ p ≤ P 0 ( 1 − p − 2 σ ) s 2 ∏ p > P 0 ( 1 − p − 2 σ ) s 2 m s ( p − σ ) . (137) (2\sigma-1)^{s^2}\prod_{p>P_0}m_s(p^{-\sigma})=\left((2\sigma-1)\zeta(2\sigma)\right)^{s^2}\prod_{p\le P_0}(1-p^{-2\sigma})^{s^2}\prod_{p>P_0}(1-p^{-2\sigma})^{s^2}m_s(p^{-\sigma}). \tag*{(137)} ( 2 σ − 1 ) s 2 p > P 0 ∏ m s ( p − σ ) = ( ( 2 σ − 1 ) ζ ( 2 σ ) ) s 2 p ≤ P 0 ∏ ( 1 − p − 2 σ ) s 2 p > P 0 ∏ ( 1 − p − 2 σ ) s 2 m s ( p − σ ) . (137)
It follows that, as σ ↓ 1 2 \sigma\downarrow\frac{1}{2} σ ↓ 2 1 ,
( 2 σ − 1 ) s 2 ∏ p > P 0 m s ( p − σ ) ⟶ ∏ p ≤ P 0 ( 1 − p − 1 ) s 2 ∏ p > P 0 ( 1 − p − 1 ) s 2 m s ( p − 1 / 2 ) (138) (2\sigma-1)^{s^2}\prod_{p>P_0}m_s(p^{-\sigma})\longrightarrow\prod_{p\le P_0}(1-p^{-1})^{s^2}\prod_{p>P_0}(1-p^{-1})^{s^2}m_s(p^{-1/2}) \tag*{(138)} ( 2 σ − 1 ) s 2 p > P 0 ∏ m s ( p − σ ) ⟶ p ≤ P 0 ∏ ( 1 − p − 1 ) s 2 p > P 0 ∏ ( 1 − p − 1 ) s 2 m s ( p − 1/2 ) (138)
uniformly for s ∈ K s \in K s ∈ K . Observe that multiplying the limit in (138) by the finite-prime factor
∏ p ≤ P 0 m s ( p − 1 / 2 ) \prod_{p\le P_0}m_s(p^{-1/2}) p ≤ P 0 ∏ m s ( p − 1/2 )
and using (120) gives exactly the arithmetic factor a s a_s a s defined in (3) . Combining this with (138) and (136) , we have
( 2 σ − 1 ) s 2 ( ∏ p ≤ P 0 Φ p , σ ) ( ∏ p > P 0 m s ( p − σ ) ) ⟶ a s , σ ↓ 1 2 (139) (2\sigma-1)^{s^2}\left(\prod_{p\le P_0}\Phi_{p,\sigma}\right)\left(\prod_{p>P_0}m_s(p^{-\sigma})\right)\longrightarrow a_s,\qquad\sigma\downarrow\frac{1}{2} \tag*{(139)} ( 2 σ − 1 ) s 2 p ≤ P 0 ∏ Φ p , σ p > P 0 ∏ m s ( p − σ ) ⟶ a s , σ ↓ 2 1 (139)
uniformly on K × B 0 K \times B_0 K × B 0 . For p > P 0 p>P_0 p > P 0 , set
D p , σ = D p , σ ( s ; λ , μ ) : = Φ p , σ m s ( p − σ ) − 1. (140) D_{p,\sigma}=D_{p,\sigma}(s;\lambda,\mu):=\frac{\Phi_{p,\sigma}}{m_s(p^{-\sigma})}-1. \tag*{(140)} D p , σ = D p , σ ( s ; λ , μ ) := m s ( p − σ ) Φ p , σ − 1. (140)
Lemma Lemma 4.8 , together with
m s ( p − σ ) − 1 = 1 + O K ( p − 2 σ ) , m_s(p^{-\sigma})^{-1}=1+O_K(p^{-2\sigma}), m s ( p − σ ) − 1 = 1 + O K ( p − 2 σ ) ,
gives
D p , σ = − i s λ ( 2 σ − 1 ) log p p 1 + δ − λ 2 + u 2 4 ( 2 σ − 1 ) 2 ( log p ) 2 p 1 + δ + E p , σ , (141) D_{p,\sigma}=-is\lambda(2\sigma-1)\frac{\log p}{p^{1+\delta}}-\frac{\lambda^2+u^2}{4}(2\sigma-1)^2\frac{(\log p)^2}{p^{1+\delta}}+E_{p,\sigma}, \tag*{(141)} D p , σ = − i s λ ( 2 σ − 1 ) p 1 + δ log p − 4 λ 2 + u 2 ( 2 σ − 1 ) 2 p 1 + δ ( log p ) 2 + E p , σ , (141)
where
∑ p > P 0 sup s ∈ K ( λ , u ) ∈ B 0 ∣ E p , σ ∣ = o ( 1 ) , σ ↓ 1 / 2. (142) \sum_{p>P_0}\sup_{\substack{s\in K\\(\lambda,u)\in B_0}}|E_{p,\sigma}|=o(1),\qquad\sigma\downarrow1/2. \tag*{(142)} p > P 0 ∑ s ∈ K ( λ , u ) ∈ B 0 sup ∣ E p , σ ∣ = o ( 1 ) , σ ↓ 1/2. (142)
The same estimates also give
sup p > P 0 , s ∈ K ( λ , u ) ∈ B 0 ∣ D p , σ ∣ = o ( 1 ) , σ ↓ 1 / 2 , (143) \sup_{\substack{p>P_0,\ s\in K\\(\lambda,u)\in B_0}}|D_{p,\sigma}|=o(1),\qquad\sigma\downarrow1/2, \tag*{(143)} p > P 0 , s ∈ K ( λ , u ) ∈ B 0 sup ∣ D p , σ ∣ = o ( 1 ) , σ ↓ 1/2 , (143)
and
∑ p > P 0 sup s ∈ K ( λ , u ) ∈ B 0 ∣ D p , σ ∣ = O K , B 0 ( 1 ) . (144) \sum_{p>P_0}\sup_{\substack{s\in K\\(\lambda,u)\in B_0}}|D_{p,\sigma}|=O_{K,B_0}(1). \tag*{(144)} p > P 0 ∑ s ∈ K ( λ , u ) ∈ B 0 sup ∣ D p , σ ∣ = O K , B 0 ( 1 ) . (144)
by Lemma 4.4. Hence the branch of log ( 1 + D p , σ ) \log(1+D_{p,\sigma}) log ( 1 + D p , σ ) near 1 is well defined uniformly on K × B 0 K \times B_0 K × B 0 . Furthermore,
∑ p > P 0 sup s ∈ K ( λ , u ) ∈ B 0 ∣ D p , σ ∣ 2 ≤ ( sup p > P 0 , s ∈ K ( λ , u ) ∈ B 0 ∣ D p , σ ∣ ) ∑ p > P 0 sup s ∈ K ( λ , u ) ∈ B 0 ∣ D p , σ ∣ = o ( 1 ) . \sum_{p>P_0}\sup_{\substack{s\in K\\(\lambda,u)\in B_0}}|D_{p,\sigma}|^2\leq\left(\sup_{\substack{p>P_0,\ s\in K\\(\lambda,u)\in B_0}}|D_{p,\sigma}|\right)\sum_{p>P_0}\sup_{\substack{s\in K\\(\lambda,u)\in B_0}}|D_{p,\sigma}|=o(1). p > P 0 ∑ s ∈ K ( λ , u ) ∈ B 0 sup ∣ D p , σ ∣ 2 ≤ p > P 0 , s ∈ K ( λ , u ) ∈ B 0 sup ∣ D p , σ ∣ p > P 0 ∑ s ∈ K ( λ , u ) ∈ B 0 sup ∣ D p , σ ∣ = o ( 1 ) .
Therefore
∑ p > P 0 log ( 1 + D p , σ ) = ∑ p > P 0 D p , σ + o ( 1 ) (145) \sum_{p>P_0}\log(1+D_{p,\sigma})=\sum_{p>P_0}D_{p,\sigma}+o(1) \tag*{(145)} p > P 0 ∑ log ( 1 + D p , σ ) = p > P 0 ∑ D p , σ + o ( 1 ) (145)
uniformly on K × B 0 K \times B_0 K × B 0 . Using (140) , (145) , (141) , and Lemma 4.4, we obtain we obtain
∑ p > P 0 log Φ p , σ m s ( p − σ ) ⟶ − i s λ − λ 2 + u 2 4 , σ ↓ 1 2 , (146) \sum_{p>P_0}\log\frac{\Phi_{p,\sigma}}{m_s(p^{-\sigma})}\longrightarrow-is\lambda-\frac{\lambda^2+u^2}{4},\qquad\sigma\downarrow\frac{1}{2}, \tag*{(146)} p > P 0 ∑ log m s ( p − σ ) Φ p , σ ⟶ − i s λ − 4 λ 2 + u 2 , σ ↓ 2 1 , (146)
uniformly on K × B 0 K \times B_0 K × B 0 . Combining this with (135) and (139) yields
( 2 σ − 1 ) s 2 ∏ p Φ p , σ ( s ; λ , u ) ⟶ a s e − i s λ − ( λ 2 + u 2 ) / 4 (2\sigma-1)^{s^2}\prod_p\Phi_{p,\sigma}(s;\lambda,u)\longrightarrow a_s e^{-is\lambda-(\lambda^2+u^2)/4} ( 2 σ − 1 ) s 2 p ∏ Φ p , σ ( s ; λ , u ) ⟶ a s e − i s λ − ( λ 2 + u 2 ) /4
uniformly on K × B 0 K \times B_0 K × B 0 .
For complex ( λ , u ) ∈ B ~ 0 (\lambda,u)\in\widetilde{B}_0 ( λ , u ) ∈ B 0 , where B ~ 0 ⊂ C 2 \widetilde{B}_0\subset\mathbb{C}^2 B 0 ⊂ C 2 is a bounded open polydisc, the local factors Φ p , σ ( s ; λ , u ) \Phi_{p,\sigma}(s;\lambda,u) Φ p , σ ( s ; λ , u ) are entire in ( λ , u ) (\lambda,u) ( λ , u ) . Moreover, the estimates (141) –(144) and (145) remain valid locally uniformly on K × B ~ 0 K\times\widetilde{B}_0 K × B 0 . Hence the same Euler-product argument yields a holomorphic continuation to B ~ 0 \widetilde{B}_0 B 0 , with the same convergence locally uniformly on K × B ~ 0 K\times\widetilde{B}_0 K × B 0 . On the real plane, this continuation agrees with H σ , s H_{\sigma,s} H σ , s by Lemma 4.7. This proves (14) , together with the asserted holomorphic continuation and locally uniform convergence.
Having established the locally uniform convergence of the weighted joint Fourier transform in Proposition 1.6, we next derive uniform Fourier estimates that will be used to justify the convergence of the Gaussian-regularized moments and the subsequent limiting arguments involving the regularized Fourier–Riesz representation. We begin with a global decay estimate on the real Fourier plane.
Uniform Fourier estimates
Global decay on the real Fourier axis and proof of Proposition 1.7
Write
For each prime p p p , define the probability measure
d μ p , a ( z ) = ∣ 1 − p − σ z ∣ − 2 a m a ( p − σ ) d m T ( z ) , \mathrm{d}\mu_{p,a}(z)=\frac{|1-p^{-\sigma}z|^{-2a}}{m_a(p^{-\sigma})}\,\mathrm{d}m_{\mathbb{T}}(z), d μ p , a ( z ) = m a ( p − σ ) ∣1 − p − σ z ∣ − 2 a d m T ( z ) ,
where m s ( r ) m_s(r) m s ( r ) is defined in (119) . Let Φ p , σ ( s ; λ , u ) \Phi_{p,\sigma}(s;\lambda,u) Φ p , σ ( s ; λ , u ) be defined in (129) . Then
Φ p , σ ( s ; λ , u ) = m a ( p − σ ) χ p , σ , \Phi_{p,\sigma}(s;\lambda,u)=m_a(p^{-\sigma})\chi_{p,\sigma}, Φ p , σ ( s ; λ , u ) = m a ( p − σ ) χ p , σ ,
where
χ p , σ = ∫ T e i Ψ p , σ ( z ) d μ p , a ( z ) (147) \chi_{p,\sigma}=\int_{\mathbb{T}}e^{i\Psi_{p,\sigma}(z)}\,d\mu_{p,a}(z) \tag*{(147)} χ p , σ = ∫ T e i Ψ p , σ ( z ) d μ p , a ( z ) (147)
and
Ψ p , σ ( z ) = − 2 b log ∣ 1 − p − σ z ∣ − ( 2 σ − 1 ) log p ( λ Re p − σ z 1 − p − σ z + u Im p − σ z 1 − p − σ z ) . (148) \Psi_{p,\sigma}(z)=-2b\log|1-p^{-\sigma}z|-(2\sigma-1)\log p\left(\lambda\operatorname{Re}\frac{p^{-\sigma}z}{1-p^{-\sigma}z}+u\operatorname{Im}\frac{p^{-\sigma}z}{1-p^{-\sigma}z}\right). \tag*{(148)} Ψ p , σ ( z ) = − 2 b log ∣1 − p − σ z ∣ − ( 2 σ − 1 ) log p ( λ Re 1 − p − σ z p − σ z + u Im 1 − p − σ z p − σ z ) . (148)
For real λ \lambda λ , u u u , the phase is real, and therefore ∣ χ p , σ ∣ ≤ 1 |\chi_{p,\sigma}|\le1 ∣ χ p , σ ∣ ≤ 1 .
Lemma 4.9. Let s = a + i b ∈ C s=a+ib\in\mathbb{C} s = a + ib ∈ C with Re s = a > 0 \operatorname{Re}s=a>0 Re s = a > 0 , and suppose that
0 < a 0 ≤ a ≤ a 1 , ∣ b ∣ ≤ B . 0<a_0\le a\le a_1,\qquad|b|\le B. 0 < a 0 ≤ a ≤ a 1 , ∣ b ∣ ≤ B . Let σ > 1 / 2 \sigma>1/2 σ > 1/2 and let p p p be a prime. Let χ p , σ \chi_{p,\sigma} χ p , σ be defined as in (147) , and set
t p : = ( 2 σ − 1 ) log p . t_p:=(2\sigma-1)\log p. t p := ( 2 σ − 1 ) log p . Then there exist constants
r 0 = r 0 ( a 0 , a 1 , B ) > 0 , η = η ( a 0 , a 1 , B ) > 0 , r_0=r_0(a_0,a_1,B)>0,\qquad\eta=\eta(a_0,a_1,B)>0, r 0 = r 0 ( a 0 , a 1 , B ) > 0 , η = η ( a 0 , a 1 , B ) > 0 , such that, whenever
0 < p − σ ≤ r 0 0<p^{-\sigma}\le r_0 0 < p − σ ≤ r 0 and
p − σ ( 2 b − t p λ ) 2 + t p 2 μ 2 ≤ η , p^{-\sigma}\sqrt{(2b-t_p\lambda)^2+t_p^2\mu^2}\le\eta, p − σ ( 2 b − t p λ ) 2 + t p 2 μ 2 ≤ η , one has
log ∣ χ p , σ ( λ , μ ) ∣ = − p − 2 σ 4 [ ( 2 b − t p λ ) 2 + t p 2 μ 2 ] + E p , σ ( λ , μ ) , (149) \log|\chi_{p,\sigma}(\lambda,\mu)|=-\frac{p^{-2\sigma}}{4}\left[(2b-t_p\lambda)^2+t_p^2\mu^2\right]+E_{p,\sigma}(\lambda,\mu), \tag*{(149)} log ∣ χ p , σ ( λ , μ ) ∣ = − 4 p − 2 σ [ ( 2 b − t p λ ) 2 + t p 2 μ 2 ] + E p , σ ( λ , μ ) , (149) where
∣ E p , σ ( λ , μ ) ∣ ≪ a 0 , a 1 , B B 2 p − 3 σ + t p 2 ( λ 2 + μ 2 ) p − 3 σ + B 4 p − 4 σ + t p 4 ( λ 2 + μ 2 ) 2 p − 4 σ . (150) |E_{p,\sigma}(\lambda,\mu)|\ll_{a_0,a_1,B}B^2p^{-3\sigma}+t_p^2(\lambda^2+\mu^2)p^{-3\sigma}+B^4p^{-4\sigma}+t_p^4(\lambda^2+\mu^2)^2p^{-4\sigma}. \tag*{(150)} ∣ E p , σ ( λ , μ ) ∣ ≪ a 0 , a 1 , B B 2 p − 3 σ + t p 2 ( λ 2 + μ 2 ) p − 3 σ + B 4 p − 4 σ + t p 4 ( λ 2 + μ 2 ) 2 p − 4 σ . (150) The implied constant is uniform in σ \sigma σ , p p p , λ \lambda λ , μ \mu μ , and s = a + i b s=a+ib s = a + ib subject to the above conditions.
Proof. Note that uniformly for z ∈ T z\in\mathbb{T} z ∈ T , as p − σ → 0 p^{-\sigma}\to0 p − σ → 0 ,
− log ∣ 1 − p − σ z ∣ = p − σ Re z + O ( p − 2 σ ) -\log|1-p^{-\sigma}z|=p^{-\sigma}\operatorname{Re}z+O(p^{-2\sigma}) − log ∣1 − p − σ z ∣ = p − σ Re z + O ( p − 2 σ )
and
p − σ z 1 − p − σ z = p − σ z + O ( p − 2 σ ) . \frac{p^{-\sigma}z}{1-p^{-\sigma}z}=p^{-\sigma}z+O(p^{-2\sigma}). 1 − p − σ z p − σ z = p − σ z + O ( p − 2 σ ) .
Hence, by the definition of Ψ p , σ \Psi_{p,\sigma} Ψ p , σ in (148) ,
Ψ p , σ ( z ) = p − σ [ ( 2 b − t p λ ) Re z − t p μ Im z ] + O ( p − 2 σ ( B + t p λ 2 + μ 2 ) ) , (151) \Psi_{p,\sigma}(z)=p^{-\sigma}\left[(2b-t_p\lambda)\operatorname{Re}z-t_p\mu\operatorname{Im}z\right]+O\left(p^{-2\sigma}\left(B+t_p\sqrt{\lambda^2+\mu^2}\right)\right), \tag*{(151)} Ψ p , σ ( z ) = p − σ [ ( 2 b − t p λ ) Re z − t p μ Im z ] + O ( p − 2 σ ( B + t p λ 2 + μ 2 ) ) , (151)
uniformly for z ∈ T z\in\mathbb{T} z ∈ T . Moreover, by the triangle inequality,
t p λ 2 + μ 2 ≤ ( 2 b − t p λ ) 2 + t p 2 μ 2 + 2 ∣ b ∣ , t_p\sqrt{\lambda^2+\mu^2}\le\sqrt{(2b-t_p\lambda)^2+t_p^2\mu^2+2|b|}, t p λ 2 + μ 2 ≤ ( 2 b − t p λ ) 2 + t p 2 μ 2 + 2∣ b ∣ ,
and therefore
p − σ t p λ 2 + μ 2 ≤ η + 2 B p − σ . p^{-\sigma}t_p\sqrt{\lambda^2+\mu^2}\leq\eta+2Bp^{-\sigma}. p − σ t p λ 2 + μ 2 ≤ η + 2 B p − σ .
It follows that
sup z ∈ T ∣ Ψ p , σ ( z ) ∣ ≪ a 0 , a 1 , B η + B p − σ . \sup_{z\in\mathbb{T}}|\Psi_{p,\sigma}(z)|\ll_{a_0,a_1,B}\eta+Bp^{-\sigma}. z ∈ T sup ∣ Ψ p , σ ( z ) ∣ ≪ a 0 , a 1 , B η + B p − σ .
Thus, after choosing η > 0 \eta>0 η > 0 and r 0 > 0 r_0>0 r 0 > 0 sufficiently small,
∣ χ p , σ ( λ , μ ) − 1 ∣ < 1 2 . (152) |\chi_{p,\sigma}(\lambda,\mu)-1|<\frac{1}{2}. \tag*{(152)} ∣ χ p , σ ( λ , μ ) − 1∣ < 2 1 . (152)
In particular, χ p , σ ( λ , μ ) \chi_{p,\sigma}(\lambda,\mu) χ p , σ ( λ , μ ) stays uniformly away from zero. Next, from
d μ p , a d m T ( z ) = ∣ 1 − p − σ z ∣ − 2 a m a ( p − σ ) , \frac{d\mu_{p,a}}{dm_{\mathbb{T}}}(z)=\frac{|1-p^{-\sigma}z|^{-2a}}{m_a(p^{-\sigma})}, d m T d μ p , a ( z ) = m a ( p − σ ) ∣1 − p − σ z ∣ − 2 a ,
together with a 0 ≤ a ≤ a 1 a_0\leq a\leq a_1 a 0 ≤ a ≤ a 1 , we have uniformly for z ∈ T z\in\mathbb{T} z ∈ T ,
d μ p , a d m T ( z ) = 1 + 2 a p − σ Re z + O a 0 , a 1 ( p − 2 σ ) . (153) \frac{d\mu_{p,a}}{dm_{\mathbb{T}}}(z)=1+2ap^{-\sigma}\operatorname{Re}z+O_{a_0,a_1}(p^{-2\sigma}). \tag*{(153)} d m T d μ p , a ( z ) = 1 + 2 a p − σ Re z + O a 0 , a 1 ( p − 2 σ ) . (153)
Using the expansion of Ψ p , σ \Psi_{p,\sigma} Ψ p , σ in (151) and (153) , a direct calculation gives
Var μ p , a ( Ψ p , σ ) : = ∫ T Ψ p , σ ( z ) 2 d μ p , a ( z ) − ( ∫ T Ψ p , σ ( z ) d μ p , a ( z ) ) 2 = p − 2 σ 2 [ ( 2 b − t p λ ) 2 + t p 2 μ 2 ] + O a 0 , a 1 , B ( B 2 p − 3 σ + t p 2 ( λ 2 + μ 2 ) p − 3 σ ) . \begin{aligned}
\operatorname{Var}_{\mu_{p,a}}(\Psi_{p,\sigma}):=\int_{\mathbb{T}}\Psi_{p,\sigma}(z)^2\,d\mu_{p,a}(z)-\left(\int_{\mathbb{T}}\Psi_{p,\sigma}(z)\,d\mu_{p,a}(z)\right)^2 \\
&=\frac{p^{-2\sigma}}{2}\left[(2b-t_p\lambda)^2+t_p^2\mu^2\right]+O_{a_0,a_1,B}\left(B^2p^{-3\sigma}+t_p^2(\lambda^2+\mu^2)p^{-3\sigma}\right).
\end{aligned} Var μ p , a ( Ψ p , σ ) := ∫ T Ψ p , σ ( z ) 2 d μ p , a ( z ) − ( ∫ T Ψ p , σ ( z ) d μ p , a ( z ) ) 2 = 2 p − 2 σ [ ( 2 b − t p λ ) 2 + t p 2 μ 2 ] + O a 0 , a 1 , B ( B 2 p − 3 σ + t p 2 ( λ 2 + μ 2 ) p − 3 σ ) .
Let Ψ p , σ ′ \Psi'_{p,\sigma} Ψ p , σ ′ be an independent copy of Ψ p , σ \Psi_{p,\sigma} Ψ p , σ under μ p , a \mu_{p,a} μ p , a . Then
∣ χ p , σ ( λ , μ ) ∣ 2 = E μ p , a ⊗ μ p , a [ e i ( Ψ p , σ − Ψ p , σ ′ ) ] = E μ p , a ⊗ μ p , a [ cos ( Ψ p , σ − Ψ p , σ ′ ) ] . |\chi_{p,\sigma}(\lambda,\mu)|^2=\mathbb{E}_{\mu_{p,a}\otimes\mu_{p,a}}\left[e^{i(\Psi_{p,\sigma}-\Psi'_{p,\sigma})}\right]=\mathbb{E}_{\mu_{p,a}\otimes\mu_{p,a}}\left[\cos(\Psi_{p,\sigma}-\Psi'_{p,\sigma})\right]. ∣ χ p , σ ( λ , μ ) ∣ 2 = E μ p , a ⊗ μ p , a [ e i ( Ψ p , σ − Ψ p , σ ′ ) ] = E μ p , a ⊗ μ p , a [ cos ( Ψ p , σ − Ψ p , σ ′ ) ] .
Using
cos x = 1 − x 2 2 + O ( x 4 ) \cos x=1-\frac{x^2}{2}+O(x^4) cos x = 1 − 2 x 2 + O ( x 4 )
and
1 2 E μ p , a ⊗ μ p , a [ ( Ψ p , σ − Ψ p , σ ′ ) 2 ] = Var μ p , a ( Ψ p , σ ) , \frac{1}{2}\mathbb{E}_{\mu_{p,a}\otimes\mu_{p,a}}\left[(\Psi_{p,\sigma}-\Psi'_{p,\sigma})^2\right]=\operatorname{Var}_{\mu_{p,a}}(\Psi_{p,\sigma}), 2 1 E μ p , a ⊗ μ p , a [ ( Ψ p , σ − Ψ p , σ ′ ) 2 ] = Var μ p , a ( Ψ p , σ ) ,
we obtain
∣ χ p , σ ( λ , μ ) ∣ 2 = 1 − Var μ p , a ( Ψ p , σ ) + O ( E μ p , a ⊗ μ p , a [ ∣ Ψ p , σ − Ψ p , σ ′ ∣ 4 ] ) . |\chi_{p,\sigma}(\lambda,\mu)|^2=1-\operatorname{Var}_{\mu_{p,a}}(\Psi_{p,\sigma})+O\left(\mathbb{E}_{\mu_{p,a}\otimes\mu_{p,a}}\left[|\Psi_{p,\sigma}-\Psi'_{p,\sigma}|^4\right]\right). ∣ χ p , σ ( λ , μ ) ∣ 2 = 1 − Var μ p , a ( Ψ p , σ ) + O ( E μ p , a ⊗ μ p , a [ ∣ Ψ p , σ − Ψ p , σ ′ ∣ 4 ] ) .
The bounds on Ψ p , σ \Psi_{p,\sigma} Ψ p , σ in (151) imply
E μ p , a ⊗ μ p , a [ ∣ Ψ p , σ − Ψ p , σ ′ ∣ 4 ] ≪ a 0 , a 1 , B B 4 p − 4 σ + t p 4 ( λ 2 + μ 2 ) 2 p − 4 σ . \mathbb{E}_{\mu_{p,a}\otimes\mu_{p,a}}\left[|\Psi_{p,\sigma}-\Psi'_{p,\sigma}|^4\right]\ll_{a_0,a_1,B}B^4p^{-4\sigma}+t_p^4(\lambda^2+\mu^2)^2p^{-4\sigma}. E μ p , a ⊗ μ p , a [ ∣ Ψ p , σ − Ψ p , σ ′ ∣ 4 ] ≪ a 0 , a 1 , B B 4 p − 4 σ + t p 4 ( λ 2 + μ 2 ) 2 p − 4 σ .
Consequently,
∣ χ p , σ ( λ , μ ) ∣ 2 = 1 − p − 2 σ 2 [ ( 2 b − t p λ ) 2 + t p 2 μ 2 ] + O a 0 , a 1 , B ( B 2 p − 3 σ + t p 2 ( λ 2 + μ 2 ) p − 3 σ + B 4 p − 4 σ + t p 4 ( λ 2 + μ 2 ) 2 p − 4 σ ) . |\chi_{p,\sigma}(\lambda,\mu)|^2=1-\frac{p^{-2\sigma}}{2}\left[(2b-t_p\lambda)^2+t_p^2\mu^2\right]
+O_{a_0,a_1,B}\left(B^2p^{-3\sigma}+t_p^2(\lambda^2+\mu^2)p^{-3\sigma}+B^4p^{-4\sigma}+t_p^4(\lambda^2+\mu^2)^2p^{-4\sigma}\right). ∣ χ p , σ ( λ , μ ) ∣ 2 = 1 − 2 p − 2 σ [ ( 2 b − t p λ ) 2 + t p 2 μ 2 ] + O a 0 , a 1 , B ( B 2 p − 3 σ + t p 2 ( λ 2 + μ 2 ) p − 3 σ + B 4 p − 4 σ + t p 4 ( λ 2 + μ 2 ) 2 p − 4 σ ) .
By (152) , ∣ χ p , σ ( λ , μ ) − 1 ∣ < 1 / 2 |\chi_{p,\sigma}(\lambda,\mu)-1|<1/2 ∣ χ p , σ ( λ , μ ) − 1∣ < 1/2 , the quantity ∣ χ p , σ ( λ , μ ) ∣ 2 |\chi_{p,\sigma}(\lambda,\mu)|^2 ∣ χ p , σ ( λ , μ ) ∣ 2 is uniformly bounded away from zero. Hence we may take logarithms and use
log ( 1 + x ) = x + O ( x 2 ) \log(1+x)=x+O(x^2) log ( 1 + x ) = x + O ( x 2 )
for x x x sufficiently small. Hence, we obtain (149) –(150) .
We now establish a uniform global decay estimate for the weighted joint Fourier transform H σ , s H_{\sigma,s} H σ , s on R 2 \mathbb{R}^{2} R 2 in Proposition 1.7 .
Proof of Proposition 1.7 . Write
s = a + i b , δ = 2 σ − 1 , ρ = ( λ 2 + u 2 ) 1 / 2 . s = a + ib,\qquad\delta= 2\sigma- 1,\qquad\rho= (\lambda^2 + u^2)^{1/2}. s = a + ib , δ = 2 σ − 1 , ρ = ( λ 2 + u 2 ) 1/2 .
Since K K K is a compact subset of the open right half-plane, there exist constants 0 < a 0 ≤ a 1 0 < a_0 \le a_1 0 < a 0 ≤ a 1 and B > 0 B > 0 B > 0 , depending only on K K K , such that
a 0 ≤ a ≤ a 1 , ∣ b ∣ ≤ B , s ∈ K . a_0 \le a \le a_1,\qquad|b| \le B,\qquad s \in K. a 0 ≤ a ≤ a 1 , ∣ b ∣ ≤ B , s ∈ K .
Fix N > 0 N > 0 N > 0 , and choose A = A K , N > 0 A = A_{K,N} > 0 A = A K , N > 0 sufficiently large. The precise lower bound required for A A A will be specified at the end of the proof. Fix σ 0 ∈ ( 1 / 2 , 1 ) \sigma_0 \in(1/2,1) σ 0 ∈ ( 1/2 , 1 ) . Let r 0 , η 0 > 0 r_0,\eta_0 > 0 r 0 , η 0 > 0 be as in Lemma 4.9 , and let C E = C E ( K ) C_E = C_E(K) C E = C E ( K ) denote the constant occurring in its error estimate. Choose 0 < η ≤ η 0 0 < \eta\le\eta_0 0 < η ≤ η 0 sufficiently small so that
C E η 2 16 ≤ 1 32 . (154) \frac{C_E\eta^2}{16} \le\frac{1}{32}. \tag*{(154)} 16 C E η 2 ≤ 32 1 . (154)
Let c ∗ > 0 c_* > 0 c ∗ > 0 and P ∗ ≥ 2 P_* \ge2 P ∗ ≥ 2 be such that Lemma 4.4 gives
∑ p ≥ P ( log p ) 2 p 1 + δ ≥ c ∗ P − δ δ 2 ( 1 + δ log P ) , P ≥ P ∗ , \sum_{p\ge P}\frac{(\log p)^2}{p^{1+\delta}} \ge c_*\frac{P^{-\delta}}{\delta^2}(1+\delta\log P),\qquad P\ge P_*, p ≥ P ∑ p 1 + δ ( log p ) 2 ≥ c ∗ δ 2 P − δ ( 1 + δ log P ) , P ≥ P ∗ ,
uniformly for 0 < δ ≤ 2 σ 0 − 1 0 < \delta\le2\sigma_0 - 1 0 < δ ≤ 2 σ 0 − 1 . Set
c 1 : = c ∗ η 2 16 > 0. (155) c_1 := \frac{c_*\eta^2}{16} > 0. \tag*{(155)} c 1 := 16 c ∗ η 2 > 0. (155)
We now choose Y 0 > 3 Y_0 > 3 Y 0 > 3 sufficiently large so that
e − Y 0 / 2 ≤ r 0 , 2 B e − Y 0 / 2 ≤ η 4 , C E e − Y 0 / 2 ≤ 1 32 , (156) e^{-Y_0/2} \le r_0,\qquad2Be^{-Y_0/2} \le\frac{\eta}{4},\qquad C_Ee^{-Y_0/2} \le\frac{1}{32}, \tag*{(156)} e − Y 0 /2 ≤ r 0 , 2 B e − Y 0 /2 ≤ 4 η , C E e − Y 0 /2 ≤ 32 1 , (156)
and
e Y 0 ≥ P ∗ . (157) e^{Y_0} \ge P_*. \tag*{(157)} e Y 0 ≥ P ∗ . (157)
We also require
c 1 e y y 2 ≥ 4 A y , y ≥ Y 0 . (158) c_1\frac{e^y}{y^2} \ge4Ay,\qquad y \ge Y_0. \tag*{(158)} c 1 y 2 e y ≥ 4 A y , y ≥ Y 0 . (158)
Having fixed η \eta η and Y 0 Y_0 Y 0 , choose
c 0 : = η e σ 0 Y 0 2 Y 0 . (159) c_0 := \frac{\eta e^{\sigma_0Y_0}}{2Y_0}. \tag*{(159)} c 0 := 2 Y 0 η e σ 0 Y 0 . (159)
Finally, choose P 0 P_0 P 0 sufficiently large so that, for every p > P 0 p > P_0 p > P 0 ,
p − 1 / 2 ≤ r 0 p^{-1/2} \le r_0 p − 1/2 ≤ r 0
and
2 B p − 1 / 2 + c 0 log p p < η . 2Bp^{-1/2} + c_0\frac{\log p}{\sqrt{p}} < \eta. 2 B p − 1/2 + c 0 p log p < η .
We now distinguish the two cases
δ ρ ≤ c 0 and δ ρ > c 0 . \delta\rho\le c_0 \qquad\text{and}\qquad\delta\rho> c_0. δ ρ ≤ c 0 and δ ρ > c 0 .
Case 1: δ ρ ≤ c 0 \delta\rho\le c_0 δ ρ ≤ c 0 . By Lemma 4.7, and the independence of { X ( p ) } p \{X(p)\}_p { X ( p ) } p , we have
H σ , s ( λ , u ) = δ s 2 ∏ p ∫ T ∣ 1 − p − σ z ∣ − 2 a e i Ψ p , σ ( z ) d m T ( z ) = δ s 2 ∏ p m a ( p − σ ) χ p , σ ( λ , u ) , \begin{aligned}
H_{\sigma,s}(\lambda,u)=\delta^{s^2}\prod_p\int_{\mathbb{T}}|1-p^{-\sigma}z|^{-2a}e^{i\Psi_{p,\sigma}(z)}\,\mathrm{d}m_{\mathbb{T}}(z) \\
&=\delta^{s^2}\prod_p m_a(p^{-\sigma})\chi_{p,\sigma}(\lambda,u),
\end{aligned} H σ , s ( λ , u ) = δ s 2 p ∏ ∫ T ∣1 − p − σ z ∣ − 2 a e i Ψ p , σ ( z ) d m T ( z ) = δ s 2 p ∏ m a ( p − σ ) χ p , σ ( λ , u ) ,
where m a ( r ) m_a(r) m a ( r ) , Ψ p , σ ( z ) \Psi_{p,\sigma}(z) Ψ p , σ ( z ) , and χ p , σ \chi_{p,\sigma} χ p , σ are defined in (119) , (148) , and (147) , respectively. Since ∣ δ s 2 ∣ = δ a 2 − b 2 |\delta^{s^2}|=\delta^{a^2-b^2} ∣ δ s 2 ∣ = δ a 2 − b 2 , combining Lemma 4.6 with the above prime factorization of H σ , s H_{\sigma,s} H σ , s gives,
∣ H σ , s ( λ , u ) ∣ ≪ K δ − b 2 ∏ p ∣ χ p , σ ( λ , u ) ∣ . (160) |H_{\sigma,s}(\lambda,u)|\ll_K\delta^{-b^2}\prod_p|\chi_{p,\sigma}(\lambda,u)|. \tag*{(160)} ∣ H σ , s ( λ , u ) ∣ ≪ K δ − b 2 p ∏ ∣ χ p , σ ( λ , u ) ∣. (160)
If χ p , σ ( λ , u ) = 0 \chi_{p,\sigma}(\lambda,u)=0 χ p , σ ( λ , u ) = 0 for some p ≤ P 0 p\le P_0 p ≤ P 0 , then H σ , s ( λ , u ) = 0 H_{\sigma,s}(\lambda,u)=0 H σ , s ( λ , u ) = 0 , and there is nothing to prove. Otherwise, since ∣ χ p , σ ( λ , u ) ∣ ≤ 1 |\chi_{p,\sigma}(\lambda,u)|\le1 ∣ χ p , σ ( λ , u ) ∣ ≤ 1 by (147) , the finitely many primes p ≤ P 0 p\le P_0 p ≤ P 0 may be discarded when deriving an upper bound. Note that for p > P 0 p>P_0 p > P 0 ,
p − σ ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ≤ 2 B p − σ + c 0 ( log p ) p − σ , p^{-\sigma}\sqrt{(2b-\delta\lambda\log p)^2+\delta^2u^2(\log p)^2}\le2Bp^{-\sigma}+c_0(\log p)p^{-\sigma}, p − σ ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ≤ 2 B p − σ + c 0 ( log p ) p − σ ,
which is smaller than the constant η \eta η in Lemma 4.9. So applying Lemma 4.9,
∑ p log ∣ χ p , σ ( λ , u ) ∣ ≤ − 1 4 ∑ p > P 0 ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 p 1 + δ + ∑ p > P 0 ∣ E p , σ ∣ . \sum_p\log|\chi_{p,\sigma}(\lambda,u)|\le-\frac{1}{4}\sum_{p>P_0}\frac{(2b-\delta\lambda\log p)^2+\delta^2u^2(\log p)^2}{p^{1+\delta}}+\sum_{p>P_0}|E_{p,\sigma}|. p ∑ log ∣ χ p , σ ( λ , u ) ∣ ≤ − 4 1 p > P 0 ∑ p 1 + δ ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 + p > P 0 ∑ ∣ E p , σ ∣.
Under the assumption δ ρ ≤ c 0 \delta\rho\le c_0 δ ρ ≤ c 0 , (150) gives
∑ p > P 0 ∣ E p , σ ∣ = O K ( 1 ) . \sum_{p>P_0}|E_{p,\sigma}|=O_K(1). p > P 0 ∑ ∣ E p , σ ∣ = O K ( 1 ) .
Expanding the quadratic term and applying Lemma 4.5, we obtain
1 4 ∑ p > P 0 ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 p 1 + δ = b 2 log 1 δ − b λ + ρ 2 4 + O K ( 1 ) . (161) \frac{1}{4}\sum_{p>P_0}\frac{(2b-\delta\lambda\log p)^2+\delta^2u^2(\log p)^2}{p^{1+\delta}}=b^2\log\frac{1}{\delta}-b\lambda+\frac{\rho^2}{4}+O_K(1). \tag*{(161)} 4 1 p > P 0 ∑ p 1 + δ ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 = b 2 log δ 1 − bλ + 4 ρ 2 + O K ( 1 ) . (161)
Since ∣ b ∣ ≤ B |b|\le B ∣ b ∣ ≤ B ,
b λ − ρ 2 4 ≤ − ρ 2 8 + 2 B 2 . b\lambda-\frac{\rho^2}{4}\le-\frac{\rho^2}{8}+2B^2. bλ − 4 ρ 2 ≤ − 8 ρ 2 + 2 B 2 .
Consequently,
∏ p ∣ χ p , σ ( λ , u ) ∣ ≪ K δ b 2 e − ρ 2 / 8 . \prod_p|\chi_{p,\sigma}(\lambda,u)|\ll_K\delta^{b^2}e^{-\rho^2/8}. p ∏ ∣ χ p , σ ( λ , u ) ∣ ≪ K δ b 2 e − ρ 2 /8 .
Substituting this into (160) gives
∣ H σ , s ( λ , u ) ∣ ≪ K e − ρ 2 / 8 . |H_{\sigma,s}(\lambda,u)|\ll_K e^{-\rho^2/8}. ∣ H σ , s ( λ , u ) ∣ ≪ K e − ρ 2 /8 .
In particular, for every N > 0 N>0 N > 0 ,
∣ H σ , s ( λ , u ) ∣ ≪ K , N ( 1 + ρ 2 ) − N . |H_{\sigma,s}(\lambda,u)|\ll_{K,N}(1+\rho^2)^{-N}. ∣ H σ , s ( λ , u ) ∣ ≪ K , N ( 1 + ρ 2 ) − N .
Case 2: δ ρ > c 0 \delta_\rho> c_0 δ ρ > c 0 . By the definition of c 0 c_0 c 0 in (159) ,
η 4 c 0 = 1 2 Y 0 e − σ 0 Y 0 . \frac{\eta}{4c_0} = \frac{1}{2}Y_0e^{-\sigma_0Y_0}. 4 c 0 η = 2 1 Y 0 e − σ 0 Y 0 .
Hence, since δ ρ > c 0 \delta_\rho> c_0 δ ρ > c 0 and σ ≤ σ 0 \sigma\le\sigma_0 σ ≤ σ 0 ,
η 4 δ ρ < η 4 c 0 = 1 2 Y 0 e − σ 0 Y 0 < Y 0 e − σ Y 0 . \frac{\eta}{4\delta_\rho} < \frac{\eta}{4c_0} = \frac{1}{2}Y_0e^{-\sigma_0Y_0} < Y_0e^{-\sigma Y_0}. 4 δ ρ η < 4 c 0 η = 2 1 Y 0 e − σ 0 Y 0 < Y 0 e − σ Y 0 .
Since Y 0 > 3 Y_0 > 3 Y 0 > 3 , the function x ↦ x e − σ x x \mapsto xe^{-\sigma x} x ↦ x e − σ x is strictly decreasing on [ Y 0 , ∞ ) [Y_0,\infty) [ Y 0 , ∞ ) and tends to zero as x → ∞ x \to\infty x → ∞ . Therefore there exists a unique y > Y 0 y > Y_0 y > Y 0 such that
δ ρ y e − σ y = η 4 . (162) \delta_\rho ye^{-\sigma y} = \frac{\eta}{4}. \tag*{(162)} δ ρ y e − σ y = 4 η . (162)
Set
P : = e y . (163) P := e^y. \tag*{(163)} P := e y . (163)
We first verify that Lemma 4.9 applies for every prime p ≥ P p \ge P p ≥ P . Since p ≥ P = e y p \ge P = e^y p ≥ P = e y , we have log p ≥ y > Y 0 \log p \ge y > Y_0 log p ≥ y > Y 0 . By (162) ,
δ ρ ( log p ) p − σ ≤ η 4 . (164) \delta_\rho(\log p)p^{-\sigma} \le\frac{\eta}{4}. \tag*{(164)} δ ρ ( log p ) p − σ ≤ 4 η . (164)
Moreover, by our choice of Y 0 Y_0 Y 0 in (156) ,
p − σ ≤ e − σ y < e − Y 0 / 2 ≤ r 0 (165) p^{-\sigma} \le e^{-\sigma y} < e^{-Y_0/2} \le r_0 \tag*{(165)} p − σ ≤ e − σ y < e − Y 0 /2 ≤ r 0 (165)
and
2 B p − σ ≤ 2 B e − Y 0 / 2 ≤ η 4 . (166) 2Bp^{-\sigma} \le2Be^{-Y_0/2} \le\frac{\eta}{4}. \tag*{(166)} 2 B p − σ ≤ 2 B e − Y 0 /2 ≤ 4 η . (166)
Therefore
p − σ ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ≤ 2 B p − σ + δ ρ ( log p ) p − σ ≤ η 2 < η . p^{-\sigma}\sqrt{(2b-\delta\lambda\log p)^2+\delta^2u^2(\log p)^2}
\le2Bp^{-\sigma}+\delta_\rho(\log p)p^{-\sigma}
\le\frac{\eta}{2}<\eta. p − σ ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ≤ 2 B p − σ + δ ρ ( log p ) p − σ ≤ 2 η < η .
Thus the local damping estimate in Lemma 4.9 applies for every p ≥ P p \ge P p ≥ P . If χ p , σ ( λ , u ) = 0 \chi_{p,\sigma}(\lambda,u)=0 χ p , σ ( λ , u ) = 0 for some prime p p p , then H σ , s ( λ , u ) = 0 H_{\sigma,s}(\lambda,u)=0 H σ , s ( λ , u ) = 0 , and there is nothing to prove. Otherwise, since ∣ χ p , σ ∣ ≤ 1 |\chi_{p,\sigma}|\le1 ∣ χ p , σ ∣ ≤ 1 ,
∑ p log ∣ χ p , σ ∣ ≤ ∑ p ≥ P log ∣ χ p , σ ∣ . \sum_p\log|\chi_{p,\sigma}| \le\sum_{p\ge P}\log|\chi_{p,\sigma}|. p ∑ log ∣ χ p , σ ∣ ≤ p ≥ P ∑ log ∣ χ p , σ ∣.
Lemma 4.9 gives
log ∣ χ p , σ ( λ , u ) ∣ = − 1 4 p 1 + δ ( ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ) + E p , σ , \log|\chi_{p,\sigma}(\lambda,u)|=-\frac{1}{4p^{1+\delta}}\left((2b-\delta\lambda\log p)^2+\delta^2u^2(\log p)^2\right)+E_{p,\sigma}, log ∣ χ p , σ ( λ , u ) ∣ = − 4 p 1 + δ 1 ( ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ) + E p , σ ,
where
∣ E p , σ ∣ ≤ C E ( B 2 p − 3 σ + δ 2 ρ 2 ( log p ) 2 p − 3 σ + B 4 p − 4 σ + δ 4 ρ 4 ( log p ) 4 p − 4 σ ) . |E_{p,\sigma}|\le C_E\left(B^2p^{-3\sigma}+\delta^2\rho^2(\log p)^2p^{-3\sigma}+B^4p^{-4\sigma}+\delta^4\rho^4(\log p)^4p^{-4\sigma}\right). ∣ E p , σ ∣ ≤ C E ( B 2 p − 3 σ + δ 2 ρ 2 ( log p ) 2 p − 3 σ + B 4 p − 4 σ + δ 4 ρ 4 ( log p ) 4 p − 4 σ ) .
Put
Q σ ( ρ ) : = ρ 2 δ 2 ∑ p ≥ P ( log p ) 2 p 1 + δ . Q_\sigma(\rho):=\rho^2\delta^2\sum_{p\ge P}\frac{(\log p)^2}{p^{1+\delta}}. Q σ ( ρ ) := ρ 2 δ 2 p ≥ P ∑ p 1 + δ ( log p ) 2 .
Note that 1 + δ = 2 σ 1+\delta=2\sigma 1 + δ = 2 σ . By (165) ,
∑ p ≥ P δ 2 ρ 2 ( log p ) 2 p − 3 σ ≤ P − σ Q σ ( ρ ) < e − Y 0 / 2 Q σ ( ρ ) . \sum_{p\ge P}\delta^2\rho^2(\log p)^2p^{-3\sigma}\le P^{-\sigma}Q_\sigma(\rho)<e^{-Y_0/2}Q_\sigma(\rho). p ≥ P ∑ δ 2 ρ 2 ( log p ) 2 p − 3 σ ≤ P − σ Q σ ( ρ ) < e − Y 0 /2 Q σ ( ρ ) .
Furthermore, by (164) ,
∑ p ≥ P δ 4 ρ 4 ( log p ) 4 p − 4 σ ≤ η 2 16 Q σ ( ρ ) . \sum_{p\ge P}\delta^4\rho^4(\log p)^4p^{-4\sigma}\le\frac{\eta^2}{16}Q_\sigma(\rho). p ≥ P ∑ δ 4 ρ 4 ( log p ) 4 p − 4 σ ≤ 16 η 2 Q σ ( ρ ) .
The remaining terms satisfy
∑ p ≥ P ( B 2 p − 3 σ + B 4 p − 4 σ ) ≪ K 1. \sum_{p\ge P}\left(B^2p^{-3\sigma}+B^4p^{-4\sigma}\right)\ll_K 1. p ≥ P ∑ ( B 2 p − 3 σ + B 4 p − 4 σ ) ≪ K 1.
By the choices of Y 0 Y_0 Y 0 and η \eta η in (156) and (154) , respectively,
C E e − Y 0 / 2 ≤ 1 32 , C E η 2 16 ≤ 1 32 . C_Ee^{-Y_0/2}\le\frac{1}{32},\qquad\frac{C_E\eta^2}{16}\le\frac{1}{32}. C E e − Y 0 /2 ≤ 32 1 , 16 C E η 2 ≤ 32 1 .
It follows that
∑ p ≥ P ∣ E p , σ ∣ ≤ 1 16 Q σ ( ρ ) + C K . (167) \sum_{p\ge P}|E_{p,\sigma}|\le\frac{1}{16}Q_\sigma(\rho)+C_K. \tag*{(167)} p ≥ P ∑ ∣ E p , σ ∣ ≤ 16 1 Q σ ( ρ ) + C K . (167)
Next,
( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ≥ 1 2 δ 2 ρ 2 ( log p ) 2 − 4 B 2 . (2b-\delta\lambda\log p)^2+\delta^2u^2(\log p)^2\ge\frac{1}{2}\delta^2\rho^2(\log p)^2-4B^2. ( 2 b − δ λ log p ) 2 + δ 2 u 2 ( log p ) 2 ≥ 2 1 δ 2 ρ 2 ( log p ) 2 − 4 B 2 .
Therefore
∑ p log ∣ χ p , σ ∣ ≤ − 1 8 Q σ ( ρ ) + B 2 ∑ p ≥ P 1 p 1 + δ + ∑ p ≥ P ∣ E p , σ ∣ ≤ − 1 16 Q σ ( ρ ) + B 2 ∑ p ≥ P 1 p 1 + δ + C K . \begin{aligned}
\sum_p\log|\chi_{p,\sigma}|&\le-\frac{1}{8}Q_\sigma(\rho)+B^2\sum_{p\ge P}\frac{1}{p^{1+\delta}}+\sum_{p\ge P}|E_{p,\sigma}|\\
&\le-\frac{1}{16}Q_\sigma(\rho)+B^2\sum_{p\ge P}\frac{1}{p^{1+\delta}}+C_K.
\end{aligned} p ∑ log ∣ χ p , σ ∣ ≤ − 8 1 Q σ ( ρ ) + B 2 p ≥ P ∑ p 1 + δ 1 + p ≥ P ∑ ∣ E p , σ ∣ ≤ − 16 1 Q σ ( ρ ) + B 2 p ≥ P ∑ p 1 + δ 1 + C K .
By (122) , we obtain
∑ p log ∣ χ p , σ ∣ ≤ − 1 16 Q σ ( ρ ) + C K log 1 δ + C K . (168) \sum_p\log|\chi_{p,\sigma}|\le-\frac{1}{16}Q_\sigma(\rho)+C_K\log\frac{1}{\delta}+C_K. \tag*{(168)} p ∑ log ∣ χ p , σ ∣ ≤ − 16 1 Q σ ( ρ ) + C K log δ 1 + C K . (168)
Here, throughout the proof, C K > 0 C_K>0 C K > 0 denotes a constant depending only on K K K , whose value may change from line to line. By (163) , (157) , and the fact that y ≥ Y 0 y\ge Y_0 y ≥ Y 0 , we have
P = e y ≥ e Y 0 ≥ P ∗ . P=e^y\ge e^{Y_0}\ge P_*. P = e y ≥ e Y 0 ≥ P ∗ .
Hence Lemma 4.5 gives
∑ p ≥ P ( log p ) 2 p 1 + δ ≥ c ∗ P − δ δ 2 ( 1 + δ log P ) . \sum_{p\ge P}\frac{(\log p)^2}{p^{1+\delta}}\ge c_*\frac{P^{-\delta}}{\delta^2}(1+\delta\log P). p ≥ P ∑ p 1 + δ ( log p ) 2 ≥ c ∗ δ 2 P − δ ( 1 + δ log P ) .
Combining the above estimate with (162) , we obtain
Q σ ( ρ ) ≥ c 1 e y ( 1 + δ y ) δ 2 y 2 . (169) Q_\sigma(\rho)\ge c_1\frac{e^y(1+\delta y)}{\delta^2y^2}. \tag*{(169)} Q σ ( ρ ) ≥ c 1 δ 2 y 2 e y ( 1 + δ y ) . (169)
with c 1 c_1 c 1 as in (155) . By our initial choice of Y 0 Y_0 Y 0 as in (158) ,
c 1 e y y 2 ≥ 4 A y , y ≥ Y 0 > 1. c_1\frac{e^y}{y^2}\ge4Ay,\qquad y\ge Y_0>1. c 1 y 2 e y ≥ 4 A y , y ≥ Y 0 > 1.
Hence
Q σ ( ρ ) ≥ 4 A y δ 2 . Q_\sigma(\rho) \ge\frac{4Ay}{\delta^2}. Q σ ( ρ ) ≥ δ 2 4 A y .
Since 0 < δ < 1 0 < \delta< 1 0 < δ < 1 , we have
δ − 2 ≥ 1 and δ − 2 ≥ log 1 δ . \delta^{-2} \ge1 \qquad\text{and}\qquad\delta^{-2} \ge\log\frac{1}{\delta}. δ − 2 ≥ 1 and δ − 2 ≥ log δ 1 .
It follows that
Q σ ( ρ ) ≥ 2 A ( y + log 1 δ ) . Q_\sigma(\rho) \ge2A\left(y+\log\frac{1}{\delta}\right). Q σ ( ρ ) ≥ 2 A ( y + log δ 1 ) .
Moreover, from (162) ,
log ( 2 + ρ ) ≤ y + log 1 δ + C K . \log(2+\rho) \le y+\log\frac{1}{\delta}+C_K. log ( 2 + ρ ) ≤ y + log δ 1 + C K .
Consequently,
Q σ ( ρ ) ≥ A ( log 1 δ + log ( 2 + ρ ) ) − C K , A . (170) Q_\sigma(\rho) \ge A\left(\log\frac{1}{\delta}+\log(2+\rho)\right)-C_{K,A}. \tag*{(170)} Q σ ( ρ ) ≥ A ( log δ 1 + log ( 2 + ρ ) ) − C K , A . (170)
Substituting (170) into (168) , we obtain
∑ p log ∣ χ p , σ ∣ ≤ − A 16 log ( 2 + ρ ) − ( A 16 − C K ) log 1 δ + C ~ K , A . \sum_p \log|\chi_{p,\sigma}| \le-\frac{A}{16}\log(2+\rho)-\left(\frac{A}{16}-C_K\right)\log\frac{1}{\delta}+\widetilde C_{K,A}. p ∑ log ∣ χ p , σ ∣ ≤ − 16 A log ( 2 + ρ ) − ( 16 A − C K ) log δ 1 + C K , A .
Therefore
∏ p ∣ χ p , σ ∣ ≪ K , A δ A / 16 − C K ( 2 + ρ ) − A / 16 . \prod_p|\chi_{p,\sigma}| \ll_{K,A}\delta^{A/16-C_K}(2+\rho)^{-A/16}. p ∏ ∣ χ p , σ ∣ ≪ K , A δ A /16 − C K ( 2 + ρ ) − A /16 .
Combining this with (160) and using ∣ b ∣ ≤ B |b|\le B ∣ b ∣ ≤ B , we find
∣ H σ , s ( λ , u ) ∣ ≪ K , A δ A / 16 − C K − B 2 ( 2 + ρ ) − A / 16 . |H_{\sigma,s}(\lambda,u)|\ll_{K,A}\delta^{A/16-C_K-B^2}(2+\rho)^{-A/16}. ∣ H σ , s ( λ , u ) ∣ ≪ K , A δ A /16 − C K − B 2 ( 2 + ρ ) − A /16 .
We now specify the lower bound required for A A A . Choose A = A K , N A=A_{K,N} A = A K , N sufficiently large so that
A 16 ≥ C K + B 2 and A 16 ≥ 2 N . \frac{A}{16}\ge C_K+B^2 \qquad\text{and}\qquad\frac{A}{16}\ge2N. 16 A ≥ C K + B 2 and 16 A ≥ 2 N .
Since 0 < δ < 1 0<\delta<1 0 < δ < 1 ,
∣ H σ , s ( λ , u ) ∣ ≪ K , N ( 2 + ρ ) − 2 N ≪ K , N ( 1 + ρ 2 ) − N . |H_{\sigma,s}(\lambda,u)|\ll_{K,N}(2+\rho)^{-2N}\ll_{K,N}(1+\rho^2)^{-N}. ∣ H σ , s ( λ , u ) ∣ ≪ K , N ( 2 + ρ ) − 2 N ≪ K , N ( 1 + ρ 2 ) − N .
This completes Case 2, and hence the proof of the proposition.
In addition to the global decay estimate established in Proposition 1.7 , we require local bounds for the derivatives of H σ , s H_{\sigma,s} H σ , s to control the Taylor remainders arising in the subsequent regularization of Riesz distributions.
Local Fourier derivative bounds
Lemma 4.10. Let K K K be a compact subset of the open right half-plane { s ∈ C : Re ( s ) > 0 } \{s\in\mathbb{C}:\operatorname{Re}(s)>0\} { s ∈ C : Re ( s ) > 0 } and let R > 0 R>0 R > 0 . For every pair of nonnegative integers α , β \alpha,\beta α , β , there exist σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 and C K , R , α , β > 0 C_{K,R,\alpha,\beta}>0 C K , R , α , β > 0 such that
sup 1 / 2 < σ ≤ σ 0 s ∈ K ∣ λ ∣ ≤ R , ∣ u ∣ ≤ R ∣ ∂ λ α ∂ u β H σ , s ( λ , u ) ∣ ≤ C K , R , α , β . (171) \sup_{\substack{1/2<\sigma\le\sigma_0\\ s\in K\\ |\lambda|\le R,\ |u|\le R}}
\left|\partial_\lambda^\alpha\partial_u^\beta H_{\sigma,s}(\lambda,u)\right|
\le C_{K,R,\alpha,\beta}. \tag*{(171)} 1/2 < σ ≤ σ 0 s ∈ K ∣ λ ∣ ≤ R , ∣ u ∣ ≤ R sup ∂ λ α ∂ u β H σ , s ( λ , u ) ≤ C K , R , α , β . (171)
Here H σ , s ( λ , u ) H_{\sigma,s}(\lambda,u) H σ , s ( λ , u ) is defined by (13) , and ∂ λ \partial_\lambda ∂ λ and ∂ u \partial_u ∂ u denote partial differentiation with respect to λ \lambda λ and u u u , respectively. Moreover, as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 ,
∂ λ α ∂ u β H σ , s ( λ , u ) ⟶ ∂ λ α ∂ u β [ a s e − i s λ − ( λ 2 + u 2 ) / 4 ] (172) \partial_\lambda^\alpha\partial_u^\beta H_{\sigma,s}(\lambda,u) \longrightarrow\partial_\lambda^\alpha\partial_u^\beta\left[a_s e^{-is\lambda-(\lambda^2+u^2)/4}\right] \tag*{(172)} ∂ λ α ∂ u β H σ , s ( λ , u ) ⟶ ∂ λ α ∂ u β [ a s e − i s λ − ( λ 2 + u 2 ) /4 ] (172)
locally uniformly for bounded real ( λ , u ) (\lambda,u) ( λ , u ) and locally uniformly on Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 . Here a s a_s a s is defined in (3) .
Proof. Choose
R < R 2 < R 1 . R < R_2 < R_1. R < R 2 < R 1 .
By Proposition 1.6, for σ > 1 / 2 \sigma>1/2 σ > 1/2 sufficiently close to 1 / 2 1/2 1/2 , the function H σ , s H_{\sigma,s} H σ , s admits a holomorphic continuation H σ , s ( λ , u ) H_{\sigma,s}(\lambda,u) H σ , s ( λ , u ) to the polydisc ∣ λ ∣ , ∣ u ∣ < R 1 |\lambda|,|u|<R_1 ∣ λ ∣ , ∣ u ∣ < R 1 , and
H σ , s ( λ , u ) ⟶ a s e − i s λ − ( λ 2 + u 2 ) / 4 H_{\sigma,s}(\lambda,u) \longrightarrow a_s e^{-is\lambda-(\lambda^2+u^2)/4} H σ , s ( λ , u ) ⟶ a s e − i s λ − ( λ 2 + u 2 ) /4
uniformly for s ∈ K s\in K s ∈ K and
∣ λ ∣ ≤ R 2 , ∣ u ∣ ≤ R 2 . |\lambda|\le R_2,\qquad|u|\le R_2. ∣ λ ∣ ≤ R 2 , ∣ u ∣ ≤ R 2 .
Hence we may choose σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 such that the family H σ , s H_{\sigma,s} H σ , s is uniformly bounded on this closed polydisc for
1 / 2 < σ ≤ σ 0 , s ∈ K . 1/2<\sigma\le\sigma_0,\qquad s\in K. 1/2 < σ ≤ σ 0 , s ∈ K .
Set
r : = R 2 − R 2 > 0. r:=\frac{R_2-R}{2}>0. r := 2 R 2 − R > 0.
For ∣ λ ∣ , ∣ u ∣ ≤ R |\lambda|,|u|\le R ∣ λ ∣ , ∣ u ∣ ≤ R , the circles
∣ z − λ ∣ = r , ∣ w − u ∣ = r |z-\lambda|=r,\qquad|w-u|=r ∣ z − λ ∣ = r , ∣ w − u ∣ = r
are contained in the polydisc
∣ z ∣ < R 2 , ∣ w ∣ < R 2 . |z|<R_2,\qquad|w|<R_2. ∣ z ∣ < R 2 , ∣ w ∣ < R 2 .
The two-variable Cauchy integral formula therefore gives
∣ ∂ λ α ∂ u β H σ , s ( λ , u ) ∣ ≤ α ! β ! r α + β sup ∣ z ∣ ≤ R 2 ∣ w ∣ ≤ R 2 ∣ H σ , s ( ζ , ω ) ∣ . \left|\partial_\lambda^\alpha\partial_u^\beta H_{\sigma,s}(\lambda,u)\right|
\le\frac{\alpha!\beta!}{r^{\alpha+\beta}}
\sup_{\substack{|z|\le R_2\\|w|\le R_2}}|H_{\sigma,s}(\zeta,\omega)|. ∂ λ α ∂ u β H σ , s ( λ , u ) ≤ r α + β α ! β ! ∣ z ∣ ≤ R 2 ∣ w ∣ ≤ R 2 sup ∣ H σ , s ( ζ , ω ) ∣.
Since H σ , s = H σ , s H_{\sigma,s}=H_{\sigma,s} H σ , s = H σ , s for real ( λ , u ) (\lambda,u) ( λ , u ) , this proves (171) . Applying the same Cauchy estimate to
H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) / 4 H_{\sigma,s}(\lambda,u)-a_s e^{-is\lambda-(\lambda^2+u^2)/4} H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) /4
and using its uniform convergence to zero on K × { ∣ λ ∣ , ∣ u ∣ ≤ R 2 } K\times\{|\lambda|,|u|\le R_2\} K × { ∣ λ ∣ , ∣ u ∣ ≤ R 2 } , we obtain
sup ∣ λ ∣ ≤ R , ∣ u ∣ ≤ R s ∈ K ∣ ∂ λ α ∂ u β H σ , s ( λ , u ) − ∂ λ α ∂ u β [ a s e − i s λ − ( λ 2 + u 2 ) / 4 ] ∣ ⟶ 0 \sup_{\substack{|\lambda|\le R,\ |u|\le R\\s\in K}}
\left|\partial_\lambda^\alpha\partial_u^\beta H_{\sigma,s}(\lambda,u)-\partial_\lambda^\alpha\partial_u^\beta\left[a_s e^{-is\lambda-(\lambda^2+u^2)/4}\right]\right|
\longrightarrow0 ∣ λ ∣ ≤ R , ∣ u ∣ ≤ R s ∈ K sup ∂ λ α ∂ u β H σ , s ( λ , u ) − ∂ λ α ∂ u β [ a s e − i s λ − ( λ 2 + u 2 ) /4 ] ⟶ 0
as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . This proves (172) .
In view of the global Fourier decay and local derivative bounds established above for H σ , s H_{\sigma,s} H σ , s , we now turn to the Fourier representation of Gaussian-regularized complex moments. We first derive a Fourier inversion formula for ∣ x ∣ 2 s e − t ∣ x ∣ 2 |x|^{2s}e^{-t|x|^2} ∣ x ∣ 2 s e − t ∣ x ∣ 2 , which will be used to express the Gaussian-regularized moments in terms of H σ , s H_{\sigma,s} H σ , s .
Fourier analysis for the Gaussian-regularized moment
Lemma 4.11. Let s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 . Then, for every ( x , y ) ∈ R 2 (x,y)\in\mathbb{R}^{2} ( x , y ) ∈ R 2 and t ∈ R > 0 t\in\mathbb{R}_{>0} t ∈ R > 0 ,
( x 2 + y 2 ) s e − t ( x 2 + y 2 ) = Γ ( s + 1 ) 4 π t s + 1 ∫ R 2 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) e i ( λ x + u y ) d λ d u . (173) (x^{2}+y^{2})^{s}e^{-t(x^{2}+y^{2})}=\frac{\Gamma(s+1)}{4\pi t^{s+1}}\int_{\mathbb{R}^{2}} {}_{1}F_{1}\left(s+1;1;-\frac{\lambda^{2}+u^{2}}{4t}\right)e^{\mathrm{i}(\lambda x+uy)}\,\mathrm{d}\lambda\,\mathrm{d}u. \tag*{(173)} ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) = 4 π t s + 1 Γ ( s + 1 ) ∫ R 2 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) e i ( λ x + u y ) d λ d u . (173) The integral on the right-hand side is absolutely convergent. Let T σ T_{\sigma} T σ be as in (11) , that is
T σ = ( 2 σ − 1 ) L ′ ( σ , X ) . T_{\sigma}=(2\sigma-1)L'(\sigma,X). T σ = ( 2 σ − 1 ) L ′ ( σ , X ) . Consequently, taking
x = Re ( T σ ) , y = Im ( T σ ) , x=\operatorname{Re}(T_{\sigma}),\qquad y=\operatorname{Im}(T_{\sigma}), x = Re ( T σ ) , y = Im ( T σ ) , we obtain, pathwise,
∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 = Γ ( s + 1 ) 4 π t s + 1 ∫ R 2 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) e i λ Re T σ + i u Im T σ d λ d u . (174) |T_{\sigma}|^{2s}e^{-t|T_{\sigma}|^{2}}=\frac{\Gamma(s+1)}{4\pi t^{s+1}}\int_{\mathbb{R}^{2}} {}_{1}F_{1}\left(s+1;1;-\frac{\lambda^{2}+u^{2}}{4t}\right)e^{\mathrm{i}\lambda\operatorname{Re}T_{\sigma}+\mathrm{i}u\operatorname{Im}T_{\sigma}\,}\,\mathrm{d}\lambda\,\mathrm{d}u. \tag*{(174)} ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 = 4 π t s + 1 Γ ( s + 1 ) ∫ R 2 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) e i λ Re T σ + i u Im T σ d λ d u . (174) Here powers of positive real numbers are defined using the real logarithm.
Proof. Let
f s , t ( x , y ) : = ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) . f_{s,t}(x,y):=(x^{2}+y^{2})^{s}e^{-t(x^{2}+y^{2})}. f s , t ( x , y ) := ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) .
Note that f s , t f_{s,t} f s , t is radial. Using polar coordinate, its Fourier transform is given by
f ^ s , t ( λ , u ) = 2 π ∫ 0 ∞ r 2 s + 1 e − t r 2 J 0 ( ρ r ) d r , ρ = λ 2 + u 2 . \widehat{f}_{s,t}(\lambda,u)=2\pi\int_{0}^{\infty}r^{2s+1}e^{-tr^{2}}J_{0}(\rho r)\,\mathrm{d}r,\qquad\rho=\sqrt{\lambda^{2}+u^{2}}. f s , t ( λ , u ) = 2 π ∫ 0 ∞ r 2 s + 1 e − t r 2 J 0 ( ρ r ) d r , ρ = λ 2 + u 2 .
where J 0 J_{0} J 0 is the Bessel function of the first kind of order zero. By the standard Hankel transform formula (see [41 Eq. (13.4.3)] ),
f ^ s , t ( λ , u ) = π Γ ( s + 1 ) t s + 1 1 F 1 ( s + 1 ; 1 ; − ρ 2 4 t ) . (175) \widehat{f}_{s,t}(\lambda,u)=\frac{\pi\Gamma(s+1)}{t^{s+1}}{}_{1}F_{1}\left(s+1;1;-\frac{\rho^{2}}{4t}\right). \tag*{(175)} f s , t ( λ , u ) = t s + 1 π Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; − 4 t ρ 2 ) . (175)
It remains to verify that
f ^ s , t ∈ L 1 ( R 2 ) , \widehat{f}_{s,t}\in L^{1}(\mathbb{R}^{2}), f s , t ∈ L 1 ( R 2 ) ,
so that the ordinary Fourier inversion theorem is applicable. For bounded ρ \rho ρ , the elementary estimate
∣ f ^ s , t ( λ , u ) ∣ ≤ ∥ f s , t ∥ L 1 (176) |\widehat{f}_{s,t}(\lambda,u)|\leq\lVert f_{s,t}\rVert_{L^{1}} \tag*{(176)} ∣ f s , t ( λ , u ) ∣ ≤ ∥ f s , t ∥ L 1 (176)
is sufficient. For large ρ \rho ρ , we use the standard Kummer asymptotic on the negative real axis. By the large-argument expansion for Kummer’s function [41 Eq. (13.7.2)] , with a = s + 1 a=s+1 a = s + 1 , b = 1 b=1 b = 1 , and z = − x z=-x z = − x , we obtain, as x → + ∞ x\to+\infty x → + ∞ ,
1 F 1 ( s + 1 ; 1 ; − x ) = x − s − 1 Γ ( − s ) ( 1 + O s ( x − 1 ) ) + O s ( e − x / 2 ( 1 + x ) C s ) , (177) {}_{1}F_{1}(s+1;1;-x)=\frac{x^{-s-1}}{\Gamma(-s)}\left(1+O_{s}(x^{-1})\right)+O_{s}\left(e^{-x/2}(1+x)^{C_{s}}\right), \tag*{(177)} 1 F 1 ( s + 1 ; 1 ; − x ) = Γ ( − s ) x − s − 1 ( 1 + O s ( x − 1 ) ) + O s ( e − x /2 ( 1 + x ) C s ) , (177)
with the algebraic term absent when s ∈ Z ≥ 0 s\in\mathbb{Z}_{\geq0} s ∈ Z ≥ 0 , since then 1 / Γ ( − s ) = 0 1/\Gamma(-s)=0 1/Γ ( − s ) = 0 . In the latter case, Kummer’s transformation gives
1 F 1 ( s + 1 ; 1 ; − x ) = e − x 1 F 1 ( − s ; 1 ; x ) . {}_{1}F_{1}(s+1;1;-x)=e^{-x}{}_{1}F_{1}(-s;1;x). 1 F 1 ( s + 1 ; 1 ; − x ) = e − x 1 F 1 ( − s ; 1 ; x ) .
and the second factor is a polynomial. It follows from (175) that, for ρ ≥ 1 \rho\ge1 ρ ≥ 1 ,
∣ f ^ s , t ( λ , u ) ∣ ≤ C s , t [ ρ − 2 Re ( s ) − 2 + e − c t ρ 2 ( 1 + ρ ) C s ] . (178) \left|\widehat{f}_{s,t}(\lambda,u)\right| \le C_{s,t}\left[\rho^{-2\operatorname{Re}(s)-2} + e^{-c_t\rho^2}(1+\rho)^{C_s}\right]. \tag*{(178)} f s , t ( λ , u ) ≤ C s , t [ ρ − 2 Re ( s ) − 2 + e − c t ρ 2 ( 1 + ρ ) C s ] . (178)
Therefore, using polar coordinates,
∫ ρ ≥ 1 ∣ f ^ s , t ( λ , u ) ∣ d λ d u ≪ s , t ∫ 1 ∞ ρ − 2 Re ( s ) − 2 ρ d ρ + ∫ 1 ∞ e − c t ρ 2 ( 1 + ρ ) C s ρ d ρ . \int_{\rho\ge1}\left|\widehat{f}_{s,t}(\lambda,u)\right|\mathrm{d}\lambda\,\mathrm{d}u \ll_{s,t} \int_1^\infty\rho^{-2\operatorname{Re}(s)-2}\rho\,\mathrm{d}\rho+\int_1^\infty e^{-c_t\rho^2}(1+\rho)^{C_s}\rho\,\mathrm{d}\rho. ∫ ρ ≥ 1 f s , t ( λ , u ) d λ d u ≪ s , t ∫ 1 ∞ ρ − 2 Re ( s ) − 2 ρ d ρ + ∫ 1 ∞ e − c t ρ 2 ( 1 + ρ ) C s ρ d ρ .
The second integral is finite, and the first is finite since Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 . Hence
f ^ s , t ∈ L 1 ( R 2 ) . (179) \widehat{f}_{s,t}\in L^1(\mathbb{R}^2). \tag*{(179)} f s , t ∈ L 1 ( R 2 ) . (179)
We have therefore proved that f s , t , f ^ s , t ∈ L 1 ( R 2 ) f_{s,t},\widehat{f}_{s,t}\in L^1(\mathbb{R}^2) f s , t , f s , t ∈ L 1 ( R 2 ) , and f s , t f_{s,t} f s , t is continuous. Hence, the Fourier inversion theorem gives (173) . Finally, substituting
x = Re T σ , y = Im T σ x=\operatorname{Re}T_\sigma,\qquad y=\operatorname{Im}T_\sigma x = Re T σ , y = Im T σ
yields (174) pathwise.
Applying the Fourier inversion formula established in Lemma 4.11 and taking expectations, we obtain the following Fourier representation of the Gaussian-regularized complex moments in terms of H σ , s H_{\sigma,s} H σ , s .
Lemma 4.12 (Fourier representation of the Gaussian-regularized moment). Let s ∈ C s\in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 . Let T σ T_\sigma T σ be as in (11) . Then for any σ \sigma σ with σ > 1 / 2 \sigma>1/2 σ > 1/2 and any t t t with t > 0 t>0 t > 0 ,
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] = Γ ( s + 1 ) 4 π t s + 1 ∫ R 2 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) H σ , s ( λ , u ) d λ d u . (180) (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2s}e^{-t|T_\sigma|^2}\right]
=\frac{\Gamma(s+1)}{4\pi t^{s+1}}\int_{\mathbb{R}^2}{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)H_{\sigma,s}(\lambda,u)\,\mathrm{d}\lambda\,\mathrm{d}u. \tag*{(180)} ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] = 4 π t s + 1 Γ ( s + 1 ) ∫ R 2 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) H σ , s ( λ , u ) d λ d u . (180) Moreover, the integral on the right-hand side is absolutely convergent.
Proof. We first multiply (174) by ( 2 σ − 1 ) s 2 ∣ ζ ( σ , X ) ∣ 2 s (2\sigma-1)^{s^2}|\zeta(\sigma,X)|^{2s} ( 2 σ − 1 ) s 2 ∣ ζ ( σ , X ) ∣ 2 s and take expectations. We now verify that Fubini’s theorem is applicable. This is because that
∫ R 2 E [ ∣ ( 2 σ − 1 ) s 2 ∣ ζ ( σ , X ) ∣ 2 s 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) e i λ Re T σ + i u Im T σ ∣ ] d λ d u = ∣ ( 2 σ − 1 ) s 2 ∣ E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] ∫ R 2 ∣ 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) ∣ d λ d u < ∞ . (181) \begin{aligned}
&\int_{\mathbb{R}^2}\mathbb{E}\left[\left|(2\sigma-1)^{s^2}|\zeta(\sigma,X)|^{2s}{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)e^{i\lambda\operatorname{Re}T_\sigma+iu\operatorname{Im}T_\sigma}\right|\right]\,\mathrm{d}\lambda\,\mathrm{d}u\\
&=\left|(2\sigma-1)^{s^2}\right|\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}\right]\int_{\mathbb{R}^2}\left|{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)\right|\,\mathrm{d}\lambda\,\mathrm{d}u<\infty. \tag*{(181)}
\end{aligned} ∫ R 2 E [ ( 2 σ − 1 ) s 2 ∣ ζ ( σ , X ) ∣ 2 s 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) e iλ Re T σ + i u Im T σ ] d λ d u = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] ∫ R 2 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) d λ d u < ∞. (181)
In fact, the first factor is finite by (69) , that is
E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞ , \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}\right]<\infty, E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞ ,
and the second one is finite by (179) . Hence Fubini’s theorem allows us to interchange expectation and the Fourier integral. We obtain
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] = Γ ( s + 1 ) 4 π t s + 1 ∫ R 2 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s e i λ Re T σ + i u Im T σ ] d λ d u . \begin{aligned}
&(2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2s}e^{-t|T_\sigma|^2}\right]\\
&=\frac{\Gamma(s+1)}{4\pi t^{s+1}}\int_{\mathbb{R}^2}{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)(2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}e^{i\lambda\operatorname{Re}T_\sigma+iu\operatorname{Im}T_\sigma}\right]\,\mathrm{d}\lambda\,\mathrm{d}u.
\end{aligned} ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] = 4 π t s + 1 Γ ( s + 1 ) ∫ R 2 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s e iλ Re T σ + i u Im T σ ] d λ d u .
By the definition of H σ , s H_{\sigma,s} H σ , s in (13) , this is exactly (180) . Finally, the same estimate (181) shows that the Fourier integral in (180) is absolutely convergent.
Using the Fourier representation established in the preceding lemma, together with the locally uniform convergence in Proposition 1.6 and the global decay estimate in Proposition 1.7 , we obtain the following limit of the Gaussian-regularized complex moments as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
Lemma 4.13 (Limit of the Gaussian-regularized moments). *Let t > 0 t>0 t > 0 be fixed. Let a s a_s a s be the arithmetic factor as in (3) . Then, as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 ,
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2s}e^{-t|T_\sigma|^2}\right] ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] ⟶ M t ( s ) : = a s Γ ( s + 1 ) 4 π t s + 1 ∫ R 2 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) e − i s λ − ( λ 2 + u 2 ) / 4 d λ d u , (182) \longrightarrow\mathcal{M}_t(s):=\frac{a_s\Gamma(s+1)}{4\pi t^{s+1}}\int_{\mathbb{R}^2}{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)e^{-is\lambda-(\lambda^2+u^2)/4}\,d\lambda\,du, \tag*{(182)} ⟶ M t ( s ) := 4 π t s + 1 a s Γ ( s + 1 ) ∫ R 2 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) e − i s λ − ( λ 2 + u 2 ) /4 d λ d u , (182) locally uniformly for
Re ( s ) > 0. \operatorname{Re}(s)>0. Re ( s ) > 0. Here T σ T_\sigma T σ is as in (11) .
Proof. Let
K ⊂ { s ∈ C : Re ( s ) > 0 } . K\subset\{s\in\mathbb{C}:\operatorname{Re}(s)>0\}. K ⊂ { s ∈ C : Re ( s ) > 0 } .
We shall prove that the convergence in (182) is uniform for s ∈ K s\in K s ∈ K . By Lemma 4.12 , it suffices to prove that as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 ,
sup s ∈ K ∣ ∫ R 2 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) ( H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) / 4 ) d λ d u ∣ ⟶ 0. (183) \sup_{s\in K}\left|\int_{\mathbb{R}^2}{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)\left(H_{\sigma,s}(\lambda,u)-a_se^{-is\lambda-(\lambda^2+u^2)/4}\right)\,d\lambda\,du\right|\longrightarrow0. \tag*{(183)} s ∈ K sup ∫ R 2 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) ( H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) /4 ) d λ d u ⟶ 0. (183)
We first obtain a bound for 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) {}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right) 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) uniformly bounded for Fourier variables. By equation (175) and (176) ,
∣ 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) ∣ ≤ ∥ f ∥ L 1 ( R 2 ) π ∣ Γ ( s + 1 ) ∣ t a − 1 , (184) \left|{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)\right|\leq\frac{\|f\|_{L^1(\mathbb{R}^2)}}{\pi|\Gamma(s+1)|t^{a-1}}, \tag*{(184)} 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) ≤ π ∣Γ ( s + 1 ) ∣ t a − 1 ∥ f ∥ L 1 ( R 2 ) , (184)
where
f ( x , y ) = ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) . f(x,y)=(x^2+y^2)^se^{-t(x^2+y^2)}. f ( x , y ) = ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) .
Using polar coordinates,
∥ f ∥ L 1 ( R 2 ) = π t − a − 1 Γ ( a + 1 ) . \|f\|_{L^1(\mathbb{R}^2)}=\pi t^{-a-1}\Gamma(a+1). ∥ f ∥ L 1 ( R 2 ) = π t − a − 1 Γ ( a + 1 ) .
Then, by (184) ,
∣ 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) ∣ ≤ Γ ( Re ( s ) + 1 ) ∣ Γ ( s + 1 ) ∣ . \left|{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)\right|\leq\frac{\Gamma(\operatorname{Re}(s)+1)}{|\Gamma(s+1)|}. 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) ≤ ∣Γ ( s + 1 ) ∣ Γ ( Re ( s ) + 1 ) .
Since K K K is compact and the Gamma function has no zeros,
sup s ∈ K ( λ , u ) ∈ R 2 ∣ 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) ∣ ≤ C K . (185) \sup_{\substack{s\in K\\(\lambda,u)\in\mathbb{R}^2}}\left|{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)\right|\leq C_K. \tag*{(185)} s ∈ K ( λ , u ) ∈ R 2 sup 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) ≤ C K . (185)
Then for any R > 0 R>0 R > 0 , by (185) , on the disk
B R = { ( λ , u ) ∈ R 2 : λ 2 + u 2 ≤ R 2 } , B_R=\{(\lambda,u)\in\mathbb{R}^2:\lambda^2+u^2\leq R^2\}, B R = {( λ , u ) ∈ R 2 : λ 2 + u 2 ≤ R 2 } ,
we have
sup s ∈ K ∣ ∫ B R 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) ( H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) / 4 ) d λ d u ∣ ≤ C K ∣ B R ∣ sup s ∈ K ( λ , u ) ∈ B R ∣ ( H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) / 4 ) ∣ → 0 , (186) \sup_{s\in K}\left|\int_{B_R} {}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right)\left(H_{\sigma,s}(\lambda,u)-a_s e^{-is\lambda-(\lambda^2+u^2)/4}\right)\,d\lambda\,du\right|
\le C_K|B_R|\sup_{\substack{s\in K\\(\lambda,u)\in B_R}}\left|\left(H_{\sigma,s}(\lambda,u)-a_s e^{-is\lambda-(\lambda^2+u^2)/4}\right)\right|\to0, \tag*{(186)} s ∈ K sup ∫ B R 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) ( H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) /4 ) d λ d u ≤ C K ∣ B R ∣ s ∈ K ( λ , u ) ∈ B R sup ( H σ , s ( λ , u ) − a s e − i s λ − ( λ 2 + u 2 ) /4 ) → 0 , (186)
the last expression tends to zero as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 Proposition 1.6 .
On the other hand, applying the global Fourier decay estimate with the fixed exponent N = 2 N=2 N = 2 , there exist σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 and C K > 0 C_K>0 C K > 0 such that
∣ H σ , s ( λ , u ) ∣ ≤ C K ( 1 + λ 2 + u 2 ) − 2 (187) |H_{\sigma,s}(\lambda,u)|\le C_K(1+\lambda^2+u^2)^{-2} \tag*{(187)} ∣ H σ , s ( λ , u ) ∣ ≤ C K ( 1 + λ 2 + u 2 ) − 2 (187)
for
1 2 < σ ≤ σ 0 , s ∈ K , ( λ , u ) ∈ R 2 . \frac{1}{2}<\sigma\le\sigma_0,\qquad s\in K,\qquad(\lambda,u)\in\mathbb{R}^2. 2 1 < σ ≤ σ 0 , s ∈ K , ( λ , u ) ∈ R 2 .
It is not to check that the limiting function satisfies a bound of the same form, that is, there is C K ′ > 0 C'_K>0 C K ′ > 0 such that
∣ a s e − i s λ − ( λ 2 + u 2 ) / 4 ∣ ≤ C K ′ ( 1 + λ 2 + u 2 ) − 2 , s ∈ K . (188) \left|a_s e^{-is\lambda-(\lambda^2+u^2)/4}\right|\le C'_K(1+\lambda^2+u^2)^{-2},\qquad s\in K. \tag*{(188)} a s e − i s λ − ( λ 2 + u 2 ) /4 ≤ C K ′ ( 1 + λ 2 + u 2 ) − 2 , s ∈ K . (188)
Then by (185) , (187) , and (188) , for any ϵ > 0 \epsilon>0 ϵ > 0 , we may choose R > 0 R>0 R > 0 so large that
sup 1 / 2 < σ ≤ σ 0 s ∈ K ∫ ∣ ( λ , u ) ∣ > R ∣ K s , t ( λ , u ) ∣ ( ∣ H σ , s ( λ , u ) ∣ + ∣ H s ( λ , u ) ∣ ) d λ d u < ϵ . (189) \sup_{\substack{1/2<\sigma\le\sigma_0\\s\in K}}\int_{|(\lambda,u)|>R}|K_{s,t}(\lambda,u)|\left(|H_{\sigma,s}(\lambda,u)|+|H_s(\lambda,u)|\right)\,d\lambda\,du<\epsilon. \tag*{(189)} 1/2 < σ ≤ σ 0 s ∈ K sup ∫ ∣ ( λ , u ) ∣ > R ∣ K s , t ( λ , u ) ∣ ( ∣ H σ , s ( λ , u ) ∣ + ∣ H s ( λ , u ) ∣ ) d λ d u < ϵ . (189)
Combining (186) with (189) , we obtain (183) .
We now evaluate the limiting Gaussian-regularized moment obtained in the preceding lemma. By expressing the corresponding Gaussian integral in polar coordinates, we obtain an explicit formula in terms of the confluent hypergeometric function.
Lemma 4.14 (Evaluation of the limiting Gaussian-regularized moment). Let s ∈ C s\in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 . Let t > 0 t>0 t > 0 . Let M t ( s ) M_t(s) M t ( s ) be defined in (182) . Then
M t ( s ) = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 1 F 1 ( s + 1 ; 1 ; s 2 1 + t ) . (190) M_t(s)=a_s e^{-s^2}\Gamma(s+1)(1+t)^{-s-1}{}_1F_1\left(s+1;1;\frac{s^2}{1+t}\right). \tag*{(190)} M t ( s ) = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 1 F 1 ( s + 1 ; 1 ; 1 + t s 2 ) . (190) Here a s a_s a s is the arithmetic factor as in (3) . Moreover, both sides are holomorphic in Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 , and the identity holds locally uniformly in this half-plane.
Proof. Using the fact that for any u ∈ R u\in\mathbb{R} u ∈ R ,
∫ R e i u y e − y 2 d y = π e − u 2 / 4 , \int_{\mathbb{R}}e^{iuy}e^{-y^2}\,dy=\sqrt{\pi}e^{-u^2/4}, ∫ R e i u y e − y 2 d y = π e − u 2 /4 ,
it is not hard to check that
1 π ∫ R 2 e i ( λ x + u y ) e − ( x + s ) 2 − y 2 d x d y = e − i s λ − ( λ 2 + u 2 ) / 4 . (191) \frac{1}{\pi}\int_{\mathbb{R}^2}e^{i(\lambda x+uy)}e^{-(x+s)^2-y^2}\,dx\,dy=e^{-is\lambda-(\lambda^2+u^2)/4}. \tag*{(191)} π 1 ∫ R 2 e i ( λ x + u y ) e − ( x + s ) 2 − y 2 d x d y = e − i s λ − ( λ 2 + u 2 ) /4 . (191)
By Lemma 4.11 that the Fourier transform of ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) (x^2+y^2)^s e^{-t(x^2+y^2)} ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) is
π Γ ( s + 1 ) t − s − 1 1 F 1 ( s + 1 ; 1 ; − λ 2 + u 2 4 t ) . (192) \pi\Gamma(s+1)t^{-s-1}{}_1F_1\left(s+1;1;-\frac{\lambda^2+u^2}{4t}\right). \tag*{(192)} π Γ ( s + 1 ) t − s − 1 1 F 1 ( s + 1 ; 1 ; − 4 t λ 2 + u 2 ) . (192)
In the following, we denote
f s , t ( x , y ) : = ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) , g s ( x , y ) = 1 π e − ( x + s ) 2 − y 2 , f_{s,t}(x,y):=(x^2+y^2)^s e^{-t(x^2+y^2)},\qquad g_s(x,y)=\frac{1}{\pi}e^{-(x+s)^2-y^2}, f s , t ( x , y ) := ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) , g s ( x , y ) = π 1 e − ( x + s ) 2 − y 2 ,
and use the convention
f ^ ( λ , u ) = ∫ R 2 e − i ( λ x + u y ) f ( x , y ) d x d y . \widehat{f}(\lambda,u)=\int_{\mathbb{R}^2}e^{-i(\lambda x+uy)}f(x,y)\,\mathrm{d}x\,\mathrm{d}y. f ( λ , u ) = ∫ R 2 e − i ( λ x + u y ) f ( x , y ) d x d y .
Hence, by (191) and (192) , together with the definition of M t ( s ) \mathcal{M}_t(s) M t ( s ) in (182) ,
M t ( s ) = a s ( 2 π ) 2 ∫ R 2 f ^ s , t ( λ , u ) ( ∫ R 2 e i ( λ x + u y ) g s ( x , y ) d x d y ) d λ d u . \mathcal{M}_t(s)=\frac{a_s}{(2\pi)^2}\int_{\mathbb{R}^2}\widehat{f}_{s,t}(\lambda,u)\left(\int_{\mathbb{R}^2}e^{i(\lambda x+uy)}g_s(x,y)\,\mathrm{d}x\,\mathrm{d}y\right)\,\mathrm{d}\lambda\,\mathrm{d}u. M t ( s ) = ( 2 π ) 2 a s ∫ R 2 f s , t ( λ , u ) ( ∫ R 2 e i ( λ x + u y ) g s ( x , y ) d x d y ) d λ d u .
Since
f s , t , f ^ s , t ∈ L 1 ( R 2 ) f_{s,t},\widehat{f}_{s,t}\in L^1(\mathbb{R}^2) f s , t , f s , t ∈ L 1 ( R 2 )
for Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 , while g s ∈ L 1 ( R 2 ) g_s\in L^1(\mathbb{R}^2) g s ∈ L 1 ( R 2 ) , Fubini’s theorem and Fourier inversion yield
M t ( s ) = ∫ R 2 f s , t ( x , y ) g s ( x , y ) d x d y = a s π ∫ R 2 ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) e − ( x + s ) 2 − y 2 d x d y . (193) \mathcal{M}_t(s)=\int_{\mathbb{R}^2}f_{s,t}(x,y)g_s(x,y)\,\mathrm{d}x\,\mathrm{d}y
=\frac{a_s}{\pi}\int_{\mathbb{R}^2}(x^2+y^2)^s e^{-t(x^2+y^2)}e^{-(x+s)^2-y^2}\,\mathrm{d}x\,\mathrm{d}y. \tag*{(193)} M t ( s ) = ∫ R 2 f s , t ( x , y ) g s ( x , y ) d x d y = π a s ∫ R 2 ( x 2 + y 2 ) s e − t ( x 2 + y 2 ) e − ( x + s ) 2 − y 2 d x d y . (193)
The integral in (193) is absolutely convergent. Indeed,
∣ ( x 2 + y 2 ) s ∣ = ( x 2 + y 2 ) Re ( s ) \left|(x^2+y^2)^s\right|=(x^2+y^2)^{\operatorname{Re}(s)} ( x 2 + y 2 ) s = ( x 2 + y 2 ) Re ( s )
and
∣ e − ( x + s ) 2 − y 2 ∣ = e ( Im ( s ) ) 2 e − ( x + Re ( s ) ) 2 − y 2 . \left|e^{-(x+s)^2-y^2}\right|=e^{(\operatorname{Im}(s))^2}e^{-(x+\operatorname{Re}(s))^2-y^2}. e − ( x + s ) 2 − y 2 = e ( Im ( s ) ) 2 e − ( x + Re ( s ) ) 2 − y 2 .
Hence the absolute value of the integrand is bounded by a polynomial times a Gaussian. We may therefore pass to polar coordinates,
x = r cos θ , y = r sin θ , d x d y = r d r d θ . x=r\cos\theta,\qquad y=r\sin\theta,\qquad\mathrm{d}x\,\mathrm{d}y=r\,\mathrm{d}r\,\mathrm{d}\theta. x = r cos θ , y = r sin θ , d x d y = r d r d θ .
Since
( x + s ) 2 + y 2 = r 2 + 2 s r cos θ + s 2 , (x+s)^2+y^2=r^2+2sr\cos\theta+s^2, ( x + s ) 2 + y 2 = r 2 + 2 sr cos θ + s 2 ,
equation (193) becomes
M t ( s ) = a s e − s 2 π ∫ 0 ∞ r 2 s + 1 e − ( 1 + t ) r 2 ( ∫ 0 2 π e − 2 s r cos θ d θ ) d r . (194) \mathcal{M}_t(s)=\frac{a_s e^{-s^2}}{\pi}\int_0^\infty r^{2s+1}e^{-(1+t)r^2}\left(\int_0^{2\pi}e^{-2sr\cos\theta}\,\mathrm{d}\theta\right)\,\mathrm{d}r. \tag*{(194)} M t ( s ) = π a s e − s 2 ∫ 0 ∞ r 2 s + 1 e − ( 1 + t ) r 2 ( ∫ 0 2 π e − 2 sr c o s θ d θ ) d r . (194)
Recall that the modified Bessel function of the first kind of order zero satisfies (see, e.g., [41 Eq. (10.32.1)] )
I 0 ( z ) = 1 2 π ∫ 0 2 π e z cos θ d θ = ∑ k = 0 ∞ 1 ( k ! ) 2 ( z 2 ) 2 k , z ∈ C . I_0(z)=\frac{1}{2\pi}\int_0^{2\pi}e^{z\cos\theta}\,\mathrm{d}\theta=\sum_{k=0}^{\infty}\frac{1}{(k!)^2}\left(\frac{z}{2}\right)^{2k},\qquad z\in\mathbb{C}. I 0 ( z ) = 2 π 1 ∫ 0 2 π e z c o s θ d θ = k = 0 ∑ ∞ ( k ! ) 2 1 ( 2 z ) 2 k , z ∈ C .
In particular, I 0 I_0 I 0 is an even function. Hence
M t ( s ) = 2 a s e − s 2 ∫ 0 ∞ r 2 s + 1 e − ( 1 + t ) r 2 I 0 ( 2 s r ) d r . (195) \mathcal{M}_t(s)=2a_s e^{-s^2}\int_0^\infty r^{2s+1}e^{-(1+t)r^2}I_0(2sr)\,\mathrm{d}r. \tag*{(195)} M t ( s ) = 2 a s e − s 2 ∫ 0 ∞ r 2 s + 1 e − ( 1 + t ) r 2 I 0 ( 2 sr ) d r . (195)
Note that the series
∑ k = 0 ∞ ∣ s ∣ 2 k ( k ! ) 2 ∫ 0 ∞ r 2 Re ( s ) + 2 k + 1 e − ( 1 + t ) r 2 d r = 1 2 ∑ k = 0 ∞ ∣ s ∣ 2 k ( k ! ) 2 ( 1 + t ) − Re ( s ) − k − 1 Γ ( Re ( s ) + k + 1 ) . \sum_{k=0}^{\infty} \frac{|s|^{2k}}{(k!)^2} \int_{0}^{\infty} r^{2\operatorname{Re}(s)+2k+1} e^{-(1+t)r^2}\,\mathrm{d}r
= \frac{1}{2}\sum_{k=0}^{\infty} \frac{|s|^{2k}}{(k!)^2}(1+t)^{-\operatorname{Re}(s)-k-1}\Gamma(\operatorname{Re}(s)+k+1). k = 0 ∑ ∞ ( k ! ) 2 ∣ s ∣ 2 k ∫ 0 ∞ r 2 Re ( s ) + 2 k + 1 e − ( 1 + t ) r 2 d r = 2 1 k = 0 ∑ ∞ ( k ! ) 2 ∣ s ∣ 2 k ( 1 + t ) − Re ( s ) − k − 1 Γ ( Re ( s ) + k + 1 ) .
is convergent. This justifies the termwise integration obtained by substituting the power series expansion of I 0 ( 2 s r ) I_0(2sr) I 0 ( 2 sr ) into (195) . Consequently,
M t ( s ) = 2 a s e − s 2 ∑ k = 0 ∞ s 2 k ( k ! ) 2 ∫ 0 ∞ r 2 s + 2 k + 1 e − ( 1 + t ) r 2 d r = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 ∑ k = 0 ∞ ( s + 1 ) k ( k ! ) 2 ( s 2 1 + t ) k . \begin{aligned}
\mathcal{M}_t(s)=2a_s e^{-s^2}\sum_{k=0}^{\infty}\frac{s^{2k}}{(k!)^2}\int_0^\infty r^{2s+2k+1}e^{-(1+t)r^2}\,\mathrm{d}r \\
&= a_s e^{-s^2}\Gamma(s+1)(1+t)^{-s-1}\sum_{k=0}^{\infty}\frac{(s+1)_k}{(k!)^2}\left(\frac{s^2}{1+t}\right)^k.
\end{aligned} M t ( s ) = 2 a s e − s 2 k = 0 ∑ ∞ ( k ! ) 2 s 2 k ∫ 0 ∞ r 2 s + 2 k + 1 e − ( 1 + t ) r 2 d r = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 k = 0 ∑ ∞ ( k ! ) 2 ( s + 1 ) k ( 1 + t s 2 ) k .
Finally, by the series expansion of the confluent hypergeometric function, see, e.g., [41 Eq. (13.2.2)] ,
1 F 1 ( a ; 1 ; z ) = ∑ k = 0 ∞ ( a ) k ( k ! ) 2 z k , {}_1F_1(a;1;z)=\sum_{k=0}^{\infty}\frac{(a)_k}{(k!)^2}z^k, 1 F 1 ( a ; 1 ; z ) = k = 0 ∑ ∞ ( k ! ) 2 ( a ) k z k ,
we conclude that (190) . Since all the preceding interchanges and identities are locally uniform in s s s , (190) holds locally uniformly in the half-plane Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 .
It is important to note that the limit t ↓ 0 t\downarrow0 t ↓ 0 cannot be interchanged directly with the limit σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . Indeed, for each fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 , dominated convergence gives
M σ , t ( s ) : = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] ⟶ M σ , 0 ( s ) , t ↓ 0 , \mathcal{M}_{\sigma,t}(s):=(2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2s}e^{-t|T_\sigma|^2}\right]\longrightarrow\mathcal{M}_{\sigma,0}(s),\qquad t\downarrow0, M σ , t ( s ) := ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s e − t ∣ T σ ∣ 2 ] ⟶ M σ , 0 ( s ) , t ↓ 0 ,
where
T σ = ( 2 σ − 1 ) L ′ ( σ , X ) , M σ , 0 ( s ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] . T_\sigma=(2\sigma-1)L'(\sigma,X),\qquad\mathcal{M}_{\sigma,0}(s)=(2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2s}\right]. T σ = ( 2 σ − 1 ) L ′ ( σ , X ) , M σ , 0 ( s ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] .
However, to interchange the two limits one would need a bound of the form
sup σ ↓ 1 / 2 ∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ⟶ 0 , t ↓ 0. \sup_{\sigma\downarrow1/2}|\mathcal{M}_{\sigma,0}(s)-\mathcal{M}_{\sigma,t}(s)|\longrightarrow0,\qquad t\downarrow0. σ ↓ 1/2 sup ∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ⟶ 0 , t ↓ 0.
Note that
∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ≤ ( 2 σ − 1 ) ( Re ( s ) ) 2 − ( Im ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( s ) ( 1 − e − t ∣ T σ ∣ 2 ) ] . |\mathcal{M}_{\sigma,0}(s)-\mathcal{M}_{\sigma,t}(s)|\le(2\sigma-1)^{(\operatorname{Re}(s))^2-(\operatorname{Im}(s))^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(s)}\left(1-e^{-t|T_\sigma|^2}\right)\right]. ∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ≤ ( 2 σ − 1 ) ( Re ( s ) ) 2 − ( Im ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( s ) ( 1 − e − t ∣ T σ ∣ 2 ) ] .
Taking absolute values replaces the complex weight ∣ ζ ∣ 2 s ∣ T σ ∣ 2 s |\zeta|^{2s}|T_\sigma|^{2s} ∣ ζ ∣ 2 s ∣ T σ ∣ 2 s by the positive weight ∣ ζ ∣ 2 a ∣ T σ ∣ 2 a |\zeta|^{2a}|T_\sigma|^{2a} ∣ ζ ∣ 2 a ∣ T σ ∣ 2 a and therefore destroys the cancellation present in the original complex expectation. One obtains only
∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ≤ t ( 2 σ − 1 ) ( Re ( s ) ) 2 − ( Im ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( s ) + 2 ] . |\mathcal{M}_{\sigma,0}(s)-\mathcal{M}_{\sigma,t}(s)|\le t(2\sigma-1)^{(\operatorname{Re}(s))^2-(\operatorname{Im}(s))^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(s)+2}\right]. ∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ≤ t ( 2 σ − 1 ) ( Re ( s ) ) 2 − ( Im ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( s ) + 2 ] .
Even if the corresponding real-weighted moment satisfies the natural bound
( 2 σ − 1 ) a 2 E [ ∣ ζ ( σ , X ) ∣ 2 a ∣ T σ ∣ 2 a + 2 ] ≪ 1 , (2\sigma-1)^{a^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2a}|T_\sigma|^{2a+2}\right]\ll1, ( 2 σ − 1 ) a 2 E [ ∣ ζ ( σ , X ) ∣ 2 a ∣ T σ ∣ 2 a + 2 ] ≪ 1 ,
the preceding estimate yields only
∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ≪ t ( 2 σ − 1 ) − b 2 , |\mathcal{M}_{\sigma,0}(s)-\mathcal{M}_{\sigma,t}(s)|\ll t(2\sigma-1)^{-b^2}, ∣ M σ , 0 ( s ) − M σ , t ( s ) ∣ ≪ t ( 2 σ − 1 ) − b 2 ,
which is not uniform as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 when b ≠ 0 b \ne0 b = 0 . Thus a direct absolute-value argument does not provide the uniform control required to interchange the limits. This is the reason for using the regularized Fourier–Riesz representation, which preserves the analytic cancellation in the complex parameter s s s .
For complex s s s , rather than interchanging the limits t ↓ 0 t \downarrow0 t ↓ 0 and σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 , we first fix σ > 1 / 2 \sigma> 1/2 σ > 1/2 and remove the Gaussian cutoff by letting t ↓ 0 t \downarrow0 t ↓ 0 . Using the theory of Riesz distributions and their meromorphic continuation, we then obtain a Fourier representation of the unregularized complex moments. This representation allows us to subsequently take the limit σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . We begin by removing the Gaussian cutoff in the space of tempered distributions.
Riesz distributions and Fourier–Riesz identities
We first observe that, for every w ∈ C w \in\mathbb{C} w ∈ C with Re ( w ) > − 1 \operatorname{Re}(w) > -1 Re ( w ) > − 1 ,
∣ x ∣ 2 w e − t ∣ x ∣ 2 ⟶ ∣ x ∣ 2 w in S ′ ( R 2 ) , t ↓ 0. (196) |x|^{2w}e^{-t|x|^2} \longrightarrow|x|^{2w} \quad\text{in } \mathcal{S}'(\mathbb{R}^2), \qquad t \downarrow0. \tag*{(196)} ∣ x ∣ 2 w e − t ∣ x ∣ 2 ⟶ ∣ x ∣ 2 w in S ′ ( R 2 ) , t ↓ 0. (196)
Indeed, for every ϕ ∈ S ( R 2 ) \phi\in\mathcal{S}(\mathbb{R}^2) ϕ ∈ S ( R 2 ) ,
∣ ∣ x ∣ 2 w ( e − t ∣ x ∣ 2 − 1 ) ϕ ( x ) ∣ ≤ 2 ∣ x ∣ 2 Re ( w ) ∣ ϕ ( x ) ∣ . \left||x|^{2w}\left(e^{-t|x|^2}-1\right)\phi(x)\right| \le2|x|^{2\operatorname{Re}(w)}|\phi(x)|. ∣ x ∣ 2 w ( e − t ∣ x ∣ 2 − 1 ) ϕ ( x ) ≤ 2∣ x ∣ 2 Re ( w ) ∣ ϕ ( x ) ∣.
Since the right-hand side is integrable over R 2 \mathbb{R}^2 R 2 when Re ( w ) > − 1 \operatorname{Re}(w) > -1 Re ( w ) > − 1 , the convergence follows from the dominated convergence theorem.
Fourier transform of homogeneous distributions
Since the Fourier transform is a continuous automorphism of S ′ ( R 2 ) \mathcal{S}'(\mathbb{R}^2) S ′ ( R 2 ) , (196) implies
∣ x ∣ 2 w e − t ∣ x ∣ 2 ^ ⟶ ∣ x ∣ 2 w ^ in S ′ ( R 2 ) . \widehat{|x|^{2w}e^{-t|x|^2}} \longrightarrow\widehat{|x|^{2w}} \quad\text{in } \mathcal{S}'(\mathbb{R}^2). ∣ x ∣ 2 w e − t ∣ x ∣ 2 ⟶ ∣ x ∣ 2 w in S ′ ( R 2 ) .
The theory of homogeneous distributions and Riesz distributions is classical; see [27 Chapter III, Section 3.2] and [44 Chapter V, Section 1] . In particular, the meromorphic continuation of radial powers r λ r^\lambda r λ is treated in [20 Chapter I, §3.9, Eq. (3) and the discussion immediately following it] , while their Fourier transforms are computed in [20 Chapter II, §3.3, Eq. (2)] . Thus, with n = 2 n=2 n = 2 and λ = 2 w \lambda=2w λ = 2 w , we obtain the Fourier–Riesz identity (22) , that is
∣ x ∣ 2 w ^ ( ξ ) = 2 2 w + 2 π Γ ( w + 1 ) Γ ( − w ) Reg ∣ ξ ∣ − 2 w − 2 , (197) \widehat{|x|^{2w}}(\xi) = 2^{2w+2}\pi\frac{\Gamma(w+1)}{\Gamma(-w)} \operatorname{Reg}|\xi|^{-2w-2}, \tag*{(197)} ∣ x ∣ 2 w ( ξ ) = 2 2 w + 2 π Γ ( − w ) Γ ( w + 1 ) Reg ∣ ξ ∣ − 2 w − 2 , (197)
where Reg ∣ ξ ∣ − 2 w − 2 \operatorname{Reg}|\xi|^{-2w-2} Reg ∣ ξ ∣ − 2 w − 2 denotes the meromorphic continuation of the homogeneous distribution ∣ ξ ∣ − 2 w − 2 |\xi|^{-2w-2} ∣ ξ ∣ − 2 w − 2 , which is locally integrable for Re w < 0 \operatorname{Re} w < 0 Re w < 0 . The strip
− 1 < Re w < 0 -1 < \operatorname{Re} w < 0 − 1 < Re w < 0
is the common range in which both ∣ x ∣ 2 w |x|^{2w} ∣ x ∣ 2 w and ∣ ξ ∣ − 2 w − 2 |\xi|^{-2w-2} ∣ ξ ∣ − 2 w − 2 are represented by locally integrable functions.
To make the identity (197) explicit for the subsequent analysis of complex moments, we now construct the meromorphic continuation of the Riesz distribution through Taylor subtraction at the origin. This yields a regularized functional that provides an explicit representation of the pairing between the Riesz distribution and the weighted joint Fourier transform H σ , s H_{\sigma,s} H σ , s .
Taylor-subtraction realization of Riesz distributions
The following construction is the two-dimensional specialization of the meromorphic continuation of the radial homogeneous distribution r λ r^\lambda r λ developed by Gelfand and Shilov; see [20 Chapter I, §3.9, pp. 71–74] . In particular, the spherical mean appearing there becomes the circular mean introduced below, while the corresponding Pizzetti formula gives the coefficients involving the iterated Laplacian. For H ∈ S ( R 2 ) H \in\mathcal{S}(\mathbb{R}^2) H ∈ S ( R 2 ) , recall that A H ( r ) \mathcal{A}H(r) A H ( r ) denotes the circular average of H H H , as defined in (20) . For − 1 < Re w < 0 -1 < \operatorname{Re} w < 0 − 1 < Re w < 0 , whenever the pairing is absolutely convergent, polar coordinates give
⟨ ∣ ξ ∣ − 2 w − 2 , H ⟩ = 2 π ∫ 0 ∞ A H ( r ) r − 2 w − 1 d r . \langle|\xi|^{-2w-2},H\rangle=2\pi\int_{0}^{\infty}\mathcal{A}H(r)r^{-2w-1}\,dr. ⟨ ∣ ξ ∣ − 2 w − 2 , H ⟩ = 2 π ∫ 0 ∞ A H ( r ) r − 2 w − 1 d r .
This is the two-dimensional specialization of the spherical-mean representation of r λ r^\lambda r λ in [20 Chapter I, §3.9, Eq. (2)] , with
λ = − 2 w − 2. \lambda=-2w-2. λ = − 2 w − 2.
In what follows, we use the notation
∂ λ : = ∂ ∂ λ , ∂ u : = ∂ ∂ u . \partial_\lambda:=\frac{\partial}{\partial\lambda},\qquad\partial_u:=\frac{\partial}{\partial u}. ∂ λ := ∂ λ ∂ , ∂ u := ∂ u ∂ .
Thus,
cos θ ∂ λ + sin θ ∂ u \cos\theta\,\partial_\lambda+\sin\theta\,\partial_u cos θ ∂ λ + sin θ ∂ u
denotes the directional derivative in the direction ( cos θ , sin θ ) (\cos\theta,\sin\theta) ( cos θ , sin θ ) . Let M ≥ 0 M \ge0 M ≥ 0 . Taylor’s theorem at the origin gives
H ( r cos θ , r sin θ ) = ∑ k = 0 2 M + 1 r k k ! ( cos θ ∂ λ + sin θ ∂ u ) k H ( 0 ) + O ( r 2 M + 2 ) , r ↓ 0 , H(r\cos\theta,r\sin\theta)=\sum_{k=0}^{2M+1}\frac{r^k}{k!}(\cos\theta\,\partial_\lambda+\sin\theta\,\partial_u)^kH(0)+O(r^{2M+2}),\qquad r\downarrow0, H ( r cos θ , r sin θ ) = k = 0 ∑ 2 M + 1 k ! r k ( cos θ ∂ λ + sin θ ∂ u ) k H ( 0 ) + O ( r 2 M + 2 ) , r ↓ 0 ,
where the implied constant depends only on M M M and on a C 2 M + 2 C^{2M+2} C 2 M + 2 -bound for H H H in a fixed neighbourhood of the origin, and is uniform in θ ∈ [ 0 , 2 π ] \theta\in[0,2\pi] θ ∈ [ 0 , 2 π ] . After averaging over θ \theta θ , all the odd-order terms vanish. Moreover,
1 2 π ∫ 0 2 π ( cos θ ∂ λ + sin θ ∂ u ) 2 j d θ = ( 2 j ) ! 4 j ( j ! ) 2 Δ j . \frac{1}{2\pi}\int_{0}^{2\pi}(\cos\theta\,\partial_\lambda+\sin\theta\,\partial_u)^{2j}\,d\theta=\frac{(2j)!}{4^j(j!)^2}\Delta^j. 2 π 1 ∫ 0 2 π ( cos θ ∂ λ + sin θ ∂ u ) 2 j d θ = 4 j ( j ! ) 2 ( 2 j )! Δ j .
Here recall that Δ = ∂ λ 2 + ∂ u 2 \Delta=\partial_\lambda^2+\partial_u^2 Δ = ∂ λ 2 + ∂ u 2 denotes the Laplacian in the variables ( λ , u ) (\lambda,u) ( λ , u ) . Consequently,
A H ( r ) = ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H ( 0 ) + O ( r 2 M + 2 ) , r ↓ 0. (198) \mathcal{A}H(r)=\sum_{j=0}^{M}\frac{r^{2j}}{4^j(j!)^2}\Delta^jH(0)+O(r^{2M+2}),\qquad r\downarrow0. \tag*{(198)} A H ( r ) = j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H ( 0 ) + O ( r 2 M + 2 ) , r ↓ 0. (198)
This is precisely the two-dimensional form of the spherical-mean expansion in [20 Chapter I, §3.9, Eq. (7)] . The original radial Riesz integral
∫ 0 ∞ A H ( r ) r − 2 w − 1 d r \int_{0}^{\infty}\mathcal{A}H(r)r^{-2w-1}\,dr ∫ 0 ∞ A H ( r ) r − 2 w − 1 d r
is absolutely convergent near the origin only in the initial strip − 1 < Re w < 0 -1 < \operatorname{Re} w < 0 − 1 < Re w < 0 . To continue it to larger values of Re w \operatorname{Re} w Re w , we subtract sufficiently many terms of the Taylor expansion of the circular average at the origin and add back their explicitly integrated contributions. This leads to the following regularized Riesz functional. This is the two-dimensional specialization of the Taylor-subtraction procedure for algebraic singularities developed by Gelfand and Shilov [20, Chapter I, § 3.2 , Eq. (3) , combined with their spherical-mean representation of the radial homogeneous distribution r λ r^{\lambda} r λ in [20 ] , Chapter I, §3.9 (here we set λ = − 2 w − 1 \lambda=-2w-1 λ = − 2 w − 1 ). For
− 1 < Re w < M + 1 , w ∉ { 0 , 1 , … , M } , -1<\operatorname{Re}w<M+1,\qquad w\notin\{0,1,\ldots,M\}, − 1 < Re w < M + 1 , w ∈ / { 0 , 1 , … , M } ,
let the regularized Riesz functional R M , w [ H ] \mathcal{R}_{M,w}[H] R M , w [ H ] be defined in (21) . In particular, for
− 1 < Re w < 0 , -1<\operatorname{Re}w<0, − 1 < Re w < 0 ,
no regularization is needed at the origin. Indeed,
∫ 0 1 r 2 j − 2 w − 1 d r = 1 2 ( j − w ) , \int_{0}^{1}r^{2j-2w-1}\,\mathrm{d}r=\frac{1}{2(j-w)}, ∫ 0 1 r 2 j − 2 w − 1 d r = 2 ( j − w ) 1 ,
and therefore
R M , w [ H ] = ∫ 0 ∞ A H ( r ) r − 2 w − 1 d r . (199) \mathcal{R}_{M,w}[H]=\int_{0}^{\infty}\mathcal{A}H(r)r^{-2w-1}\,\mathrm{d}r. \tag*{(199)} R M , w [ H ] = ∫ 0 ∞ A H ( r ) r − 2 w − 1 d r . (199)
Thus, using polar coordinates,
⟨ ∣ ξ ∣ − 2 w − 2 , H ⟩ = ∫ R 2 ∣ ξ ∣ − 2 w − 2 H ( ξ ) d ξ = 2 π ∫ 0 ∞ A H ( r ) r − 2 w − 1 d r = 2 π R M , w [ H ] , \langle|\xi|^{-2w-2},H\rangle=\int_{\mathbb{R}^{2}}|\xi|^{-2w-2}H(\xi)\,\mathrm{d}\xi=2\pi\int_{0}^{\infty}\mathcal{A}H(r)r^{-2w-1}\,\mathrm{d}r=2\pi\mathcal{R}_{M,w}[H], ⟨ ∣ ξ ∣ − 2 w − 2 , H ⟩ = ∫ R 2 ∣ ξ ∣ − 2 w − 2 H ( ξ ) d ξ = 2 π ∫ 0 ∞ A H ( r ) r − 2 w − 1 d r = 2 π R M , w [ H ] ,
in the initial strip − 1 < Re w < 0 -1<\operatorname{Re}w<0 − 1 < Re w < 0 .
The Taylor subtraction in (21) is the explicit two-dimensional realization of the analytic-continuation procedure for r − 2 w − 2 r^{-2w-2} r − 2 w − 2 in [20 ] , Chapter I, §3.9. Indeed, by (198) ,
A H ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H ( 0 ) = O ( r 2 M + 2 ) , \mathcal{A}H(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^{j}(j!)^{2}}\Delta^{j}H(0)=O(r^{2M+2}), A H ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H ( 0 ) = O ( r 2 M + 2 ) ,
so that
[ A H ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H ( 0 ) ] r − 2 w − 1 = O ( r 2 M + 1 − 2 Re w ) . \left[\mathcal{A}H(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^{j}(j!)^{2}}\Delta^{j}H(0)\right]r^{-2w-1}=O\left(r^{2M+1-2\operatorname{Re}w}\right). [ A H ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H ( 0 ) ] r − 2 w − 1 = O ( r 2 M + 1 − 2 Re w ) .
Hence the first integral in (21) is locally uniformly absolutely convergent, and therefore holomorphic in w w w , for
Re w < M + 1. \operatorname{Re}w<M+1. Re w < M + 1.
The correction terms
Δ j H ( 0 ) 2 4 j ( j ! ) 2 ( j − w ) \frac{\Delta^{j}H(0)}{2\,4^{j}(j!)^{2}(j-w)} 2 4 j ( j ! ) 2 ( j − w ) Δ j H ( 0 )
give the meromorphic continuation of the subtracted Taylor monomials and exhibit the possible simple poles at
w = 0 , 1 , … , M . w=0,1,\ldots,M. w = 0 , 1 , … , M .
This agrees with the pole structure of the meromorphic continuation of r − 2 w − 2 r^{-2w-2} r − 2 w − 2 described in [20 ] , Chapter I, §3.9, p. 72. Provided that H H H has sufficient decay at infinity, the second integral in (21) is also locally uniformly convergent in the range under consideration.
By the Gelfand-Shilov meromorphic continuation of radial homogeneous distributions [20 ] , Chapter I, §3.9, the map
w ⟼ ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ w\longmapsto\left\langle\operatorname{Reg}|\xi|^{-2w-2},H\right\rangle w ⟼ ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩
is meromorphic in the strip under consideration. Here H ∈ S ( R 2 ) H \in\mathcal{S}(\mathbb{R}^{2}) H ∈ S ( R 2 ) . On the other hand, the first two integrals defining R M , w [ H ] \mathcal{R}_{M,w}[H] R M , w [ H ] are holomorphic there by locally uniform absolute convergence, while the finite correction sum is meromorphic with possible simple poles at w = 0 , 1 , … , M w=0,1,\ldots,M w = 0 , 1 , … , M . Hence both sides are meromorphic functions of w w w . Since they agree for − 1 < Re w < 0 -1<\operatorname{Re}w<0 − 1 < Re w < 0 , uniqueness of meromorphic continuation yields
⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ = 2 π R M , w [ H ] . (200) \left\langle\operatorname{Reg}|\xi|^{-2w-2},H\right\rangle=2\pi\mathcal{R}_{M,w}[H]. \tag*{(200)} ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ = 2 π R M , w [ H ] . (200)
Recall the circular average A H ( r ) \mathcal{A}H(r) A H ( r ) introduced for Schwartz functions in (20) . This definition extends naturally to functions H ∈ C ( R 2 ) H\in C(\mathbb{R}^{2}) H ∈ C ( R 2 ) by the same formula:
A H ( r ) : = 1 2 π ∫ 0 2 π H ( r cos θ , r sin θ ) d θ , r ≥ 0. (201) \mathcal{A}H(r):=\frac{1}{2\pi}\int_{0}^{2\pi}H(r\cos\theta,r\sin\theta)\,d\theta,\qquad r\ge0. \tag*{(201)} A H ( r ) := 2 π 1 ∫ 0 2 π H ( r cos θ , r sin θ ) d θ , r ≥ 0. (201)
We also recall the regularized Riesz functional R M , w [ H ] \mathcal{R}_{M,w}[H] R M , w [ H ] defined in (21) . Note that it involves derivatives of H H H at the origin. Hence, for M ∈ Z ≥ 0 M\in\mathbb{Z}_{\ge0} M ∈ Z ≥ 0 , formula (21) is well-defined whenever
H ∈ C 2 M + 2 ( R 2 ) ∩ L 1 ( R 2 ) , H\in C^{2M+2}(\mathbb{R}^{2})\cap L^{1}(\mathbb{R}^{2}), H ∈ C 2 M + 2 ( R 2 ) ∩ L 1 ( R 2 ) ,
and
− 1 < Re ( w ) < M + 1 , w ∉ { 0 , 1 , … , M } . -1<\operatorname{Re}(w)<M+1,\qquad w\notin\{0,1,\ldots,M\}. − 1 < Re ( w ) < M + 1 , w ∈ / { 0 , 1 , … , M } .
Indeed, Taylor’s theorem gives
A H ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H ( 0 ) = O ( r 2 M + 2 ) , r ↓ 0 , \mathcal{A}H(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^{j}(j!)^{2}}\Delta^{j}H(0)=O(r^{2M+2}),\qquad r\downarrow0, A H ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H ( 0 ) = O ( r 2 M + 2 ) , r ↓ 0 ,
which ensures convergence of the regularized integral at the origin whenever Re ( w ) < M + 1 \operatorname{Re}(w)<M+1 Re ( w ) < M + 1 . The L 1 L^{1} L 1 -assumption, together with Re ( w ) > − 1 \operatorname{Re}(w)>-1 Re ( w ) > − 1 , guarantees absolute convergence of the integral at infinity.
The following proposition shows that the regularized functional R M , w [ H ] \mathcal{R}_{M,w}[H] R M , w [ H ] agrees with an extension of the Riesz distributional pairing to functions in C 2 M + 2 ( R 2 ) ∩ L 1 ( R 2 ) C^{2M+2}(\mathbb{R}^{2})\cap L^{1}(\mathbb{R}^{2}) C 2 M + 2 ( R 2 ) ∩ L 1 ( R 2 ) . In particular, this extension allows us to express the original complex moments, without Gaussian regularization, in terms of the Riesz functional R M , w [ H σ , s ] \mathcal{R}_{M,w}[H_{\sigma,s}] R M , w [ H σ , s ] .
Proposition 4.15. Let M ∈ Z ≥ 0 M\in\mathbb{Z}_{\ge0} M ∈ Z ≥ 0 . Suppose that
H ∈ C ∞ ( R 2 ) ∩ L 1 ( R 2 ) . H\in C^{\infty}(\mathbb{R}^{2})\cap L^{1}(\mathbb{R}^{2}). H ∈ C ∞ ( R 2 ) ∩ L 1 ( R 2 ) . Define
Ω M : = { w ∈ C : − 1 < Re w < M + 1 } ∖ { 0 , 1 , … , M } . \Omega_{M}:=\{w\in\mathbb{C}:-1<\operatorname{Re}w<M+1\}\setminus\{0,1,\ldots,M\}. Ω M := { w ∈ C : − 1 < Re w < M + 1 } ∖ { 0 , 1 , … , M } . Let R M , w [ H ] \mathcal{R}_{M,w}[H] R M , w [ H ] be defined in (21) for w ∈ Ω M w\in\Omega_{M} w ∈ Ω M . Then the Gelfand–Shilov pairing with the regularized homogeneous distribution admits an extension to H H H on Ω M \Omega_{M} Ω M , given by
⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ e x t = 2 π R M , w [ H ] , w ∈ Ω M . (202) \left\langle\operatorname{Reg}|\xi|^{-2w-2},H\right\rangle_{\mathrm{ext}}=2\pi\mathcal{R}_{M,w}[H],\qquad w\in\Omega_{M}. \tag*{(202)} ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ ext = 2 π R M , w [ H ] , w ∈ Ω M . (202) The extended pairing is independent of the cutoff used in its definition and is holomorphic in w w w on Ω M \Omega_{M} Ω M .
Proof. Choose a radial function χ ∈ C ∞ ( R 2 ) \chi\in C^\infty(\mathbb{R}^2) χ ∈ C ∞ ( R 2 ) such that
0 ≤ χ ≤ 1 , χ ( ξ ) = 1 for ∣ ξ ∣ ≤ 1. 0 \leq\chi\leq1,\qquad\chi(\xi)=1\quad\text{for }|\xi|\leq1. 0 ≤ χ ≤ 1 , χ ( ξ ) = 1 for ∣ ξ ∣ ≤ 1.
For R ≥ 1 R \geq1 R ≥ 1 , set
χ R ( ξ ) : = χ ( ξ / R ) , H R ( ξ ) : = χ R ( ξ ) H ( ξ ) . \chi_R(\xi):=\chi(\xi/R),\qquad H_R(\xi):=\chi_R(\xi)H(\xi). χ R ( ξ ) := χ ( ξ / R ) , H R ( ξ ) := χ R ( ξ ) H ( ξ ) .
Then
H R ∈ C ∞ ( R 2 ) ⊂ S ( R 2 ) , H_R\in C^\infty(\mathbb{R}^2)\subset\mathcal{S}(\mathbb{R}^2), H R ∈ C ∞ ( R 2 ) ⊂ S ( R 2 ) ,
and hence the Gelfand–Shilov distribution Reg ∣ ξ ∣ − 2 w − 2 \operatorname{Reg}|\xi|^{-2w-2} Reg ∣ ξ ∣ − 2 w − 2 acts on H R H_R H R in the usual distributional sense. By (200) ,
⟨ Reg ∣ ξ ∣ − 2 w − 2 , H R ⟩ = 2 π R M , w [ H R ] , w ∈ Ω M . (203) \left\langle\operatorname{Reg}|\xi|^{-2w-2},H_R\right\rangle=2\pi\mathcal{R}_{M,w}[H_R],\qquad w\in\Omega_M. \tag*{(203)} ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H R ⟩ = 2 π R M , w [ H R ] , w ∈ Ω M . (203)
and the later one is holomorphic in w w w on Ω M \Omega_M Ω M . We next pass to the limit R → ∞ R\to\infty R → ∞ . Since χ R = 1 \chi_R=1 χ R = 1 on B ( 0 , 1 ) B(0,1) B ( 0 , 1 ) for every R ≥ 1 R\geq1 R ≥ 1 ,
H R ( ξ ) = H ( ξ ) , ∣ ξ ∣ ≤ 1. H_R(\xi)=H(\xi),\qquad|\xi|\leq1. H R ( ξ ) = H ( ξ ) , ∣ ξ ∣ ≤ 1.
Consequently,
A H R ( r ) = A H ( r ) , 0 ≤ r ≤ 1 , \mathcal{A}H_R(r)=\mathcal{A}H(r),\qquad0\leq r\leq1, A H R ( r ) = A H ( r ) , 0 ≤ r ≤ 1 ,
and
Δ j H R ( 0 ) = Δ j H ( 0 ) , 0 ≤ j ≤ M . \Delta^jH_R(0)=\Delta^jH(0),\qquad0\leq j\leq M. Δ j H R ( 0 ) = Δ j H ( 0 ) , 0 ≤ j ≤ M .
Thus the first and third terms in R M , w [ H R ] \mathcal{R}_{M,w}[H_R] R M , w [ H R ] coincide exactly with the corresponding terms in R M , w [ H ] \mathcal{R}_{M,w}[H] R M , w [ H ] . It therefore remains only to consider the tail integral. Let K K K be a compact set contained in Ω M \Omega_M Ω M . For r ≥ 1 r\geq1 r ≥ 1 ,
r − 2 Re w − 2 ≤ r − 2 inf w ∈ K Re w − 2 ≤ 1. r^{-2\operatorname{Re}w-2}\leq r^{-2\inf_{w\in K}\operatorname{Re}w-2}\leq1. r − 2 Re w − 2 ≤ r − 2 i n f w ∈ K Re w − 2 ≤ 1.
Hence
sup w ∈ K ∣ ∫ 1 ∞ A ( H R − H ) ( r ) r − 2 w − 1 d r ∣ ≤ ∫ 1 ∞ ∣ A ( H R − H ) ( r ) ∣ r d r ≤ 1 2 π ∫ ∣ ξ ∣ ≥ 1 ∣ H R ( ξ ) − H ( ξ ) ∣ d ξ . (204) \sup_{w\in K}\left|\int_1^\infty\mathcal{A}(H_R-H)(r)r^{-2w-1}\,dr\right|
\leq\int_1^\infty|\mathcal{A}(H_R-H)(r)|r\,dr
\leq\frac{1}{2\pi}\int_{|\xi|\geq1}|H_R(\xi)-H(\xi)|\,d\xi. \tag*{(204)} w ∈ K sup ∫ 1 ∞ A ( H R − H ) ( r ) r − 2 w − 1 d r ≤ ∫ 1 ∞ ∣ A ( H R − H ) ( r ) ∣ r d r ≤ 2 π 1 ∫ ∣ ξ ∣ ≥ 1 ∣ H R ( ξ ) − H ( ξ ) ∣ d ξ . (204)
Moreover,
H R ( ξ ) ⟶ H ( ξ ) ( R → ∞ ) H_R(\xi)\longrightarrow H(\xi)\qquad(R\to\infty) H R ( ξ ) ⟶ H ( ξ ) ( R → ∞ )
for every ξ \xi ξ , and, since 0 ≤ χ R ≤ 1 0\leq\chi_R\leq1 0 ≤ χ R ≤ 1 ,
∣ H R ( ξ ) − H ( ξ ) ∣ ≤ ∣ H ( ξ ) ∣ . |H_R(\xi)-H(\xi)|\leq|H(\xi)|. ∣ H R ( ξ ) − H ( ξ ) ∣ ≤ ∣ H ( ξ ) ∣.
Because H ∈ L 1 ( R 2 ) H\in L^1(\mathbb{R}^2) H ∈ L 1 ( R 2 ) , dominated convergence gives
∥ H R − H ∥ L 1 ( R 2 ) ⟶ 0. \|H_R-H\|_{L^1(\mathbb{R}^2)}\longrightarrow0. ∥ H R − H ∥ L 1 ( R 2 ) ⟶ 0.
It follows from (204) that
R M , w [ H R ] ⟶ R M , w [ H ] (205) \mathcal{R}_{M,w}[H_R]\longrightarrow\mathcal{R}_{M,w}[H] \tag*{(205)} R M , w [ H R ] ⟶ R M , w [ H ] (205)
locally uniformly for w ∈ Ω M w\in\Omega_M w ∈ Ω M . This motivates the following extension of the Gelfand–Shilov pairing to H H H :
⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ e x t : = lim R → ∞ ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H R ⟩ . (206) \left\langle\operatorname{Reg}|\xi|^{-2w-2},H\right\rangle_{\mathrm{ext}}
:=\lim_{R\to\infty}\left\langle\operatorname{Reg}|\xi|^{-2w-2},H_R\right\rangle. \tag*{(206)} ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ ext := R → ∞ lim ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H R ⟩ . (206)
The limit exists locally uniformly in w ∈ Ω M w \in\Omega_M w ∈ Ω M . Combining (203) with (205) yields
⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ e x t = 2 π R M , w [ H ] , \left\langle\operatorname{Reg}|\xi|^{-2w-2},H\right\rangle_{\mathrm{ext}}=2\pi\mathcal{R}_{M,w}[H], ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H ⟩ ext = 2 π R M , w [ H ] ,
which proves (202) . The right-hand side depends only on H H H , and not on the particular cutoff function χ \chi χ . Hence the extended pairing is independent of the choice of cutoff. Finally, since for every R R R the function
w ⟼ ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H R ⟩ w\longmapsto\left\langle\operatorname{Reg}|\xi|^{-2w-2},H_R\right\rangle w ⟼ ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H R ⟩
is holomorphic on Ω M \Omega_M Ω M , and the convergence in (206) is locally uniform there, the Weierstrass theorem implies that the extended pairing is holomorphic on Ω M \Omega_M Ω M .
The extended Riesz distributional pairing established in Proposition 4.15 will allow us to express the weighted mixed moments of ζ ( σ , X ) \zeta(\sigma,X) ζ ( σ , X ) and L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) in terms of the regularized Riesz functional R M , w [ H σ , s ] \mathcal{R}_{M,w}[H_{\sigma,s}] R M , w [ H σ , s ] . To establish this representation, we first prove the integrability of the relevant mixed moments and then introduce the weighted measure ν σ , s \nu_{\sigma,s} ν σ , s defined in (12) . We subsequently derive the corresponding Fourier–Riesz identity. We first establish the integrability of the corresponding mixed moments.
The Fourier–Riesz identity for the weighted measure
Lemma 4.16. Let σ > 1 / 2 \sigma>1/2 σ > 1/2 , s ∈ C s\in\mathbb{C} s ∈ C with Re s > 0 \operatorname{Re}s>0 Re s > 0 , and w ∈ C w\in\mathbb{C} w ∈ C with Re w > − 1 \operatorname{Re}w>-1 Re w > − 1 . Let T σ T_\sigma T σ be as in (11) . Then
E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ 2 Re w ] < ∞ . (207) \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}s}|T_\sigma|^{2\operatorname{Re}w}\right]<\infty. \tag*{(207)} E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ 2 Re w ] < ∞. (207) Consequently, for every t > 0 t>0 t > 0 ,
E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ 2 Re w e − t ∣ T σ ∣ 2 ] < ∞ . (208) \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}s}|T_\sigma|^{2\operatorname{Re}w}e^{-t|T_\sigma|^2}\right]<\infty. \tag*{(208)} E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ 2 Re w e − t ∣ T σ ∣ 2 ] < ∞. (208) Proof. We distinguish two cases. Suppose first that − 1 < Re ( w ) < 0 -1<\operatorname{Re}(w)<0 − 1 < Re ( w ) < 0 . Since H σ , Re ( s ) ∈ L 1 ( R 2 ) H_{\sigma,\operatorname{Re}(s)}\in L^1(\mathbb{R}^2) H σ , Re ( s ) ∈ L 1 ( R 2 ) by (15) , Fourier inversion yields a bounded continuous density
h σ , Re ( s ) ( x ) = 1 ( 2 π ) 2 ∫ R 2 e i ⟨ x , ξ ⟩ H σ , Re ( s ) ( ξ ) d ξ h_{\sigma,\operatorname{Re}(s)}(x)=\frac{1}{(2\pi)^2}\int_{\mathbb{R}^2}e^{i\langle x,\xi\rangle}H_{\sigma,\operatorname{Re}(s)}(\xi)\,d\xi h σ , Re ( s ) ( x ) = ( 2 π ) 2 1 ∫ R 2 e i ⟨ x , ξ ⟩ H σ , Re ( s ) ( ξ ) d ξ such that, for every nonnegative Borel function G G G ,
( 2 σ − 1 ) ( Re ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) G ( T σ ) ] = ∫ R 2 G ( x ) h σ , Re ( s ) ( x ) d x , (2\sigma-1)(\operatorname{Re}(s))^2\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}G(T_\sigma)\right]=\int_{\mathbb{R}^2}G(x)h_{\sigma,\operatorname{Re}(s)}(x)\,dx, ( 2 σ − 1 ) ( Re ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) G ( T σ ) ] = ∫ R 2 G ( x ) h σ , Re ( s ) ( x ) d x , whenever either side is finite, with the identity for general nonnegative G G G obtained by monotone convergence. Taking
G ( x ) = ∣ x ∣ 2 Re ( w ) 1 { ∣ x ∣ < 1 } , G(x)=|x|^{2\operatorname{Re}(w)}1_{\{|x|<1\}}, G ( x ) = ∣ x ∣ 2 Re ( w ) 1 { ∣ x ∣ < 1 } , we obtain
( 2 σ − 1 ) ( Re ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) 1 { ∣ T σ ∣ < 1 } ] = ∫ ∣ x ∣ < 1 ∣ x ∣ 2 Re ( w ) h σ , Re ( s ) ( x ) d x ≤ 2 π ∥ h σ , Re ( s ) ∥ ∞ ∫ 0 1 r 2 Re ( w ) + 1 d r < ∞ . \begin{aligned}
&(2\sigma-1)(\operatorname{Re}(s))^2\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(w)}1_{\{|T_\sigma|<1\}}\right]\\
&=\int_{|x|<1}|x|^{2\operatorname{Re}(w)}h_{\sigma,\operatorname{Re}(s)}(x)\,dx\\
&\le2\pi\|h_{\sigma,\operatorname{Re}(s)}\|_\infty\int_0^1r^{2\operatorname{Re}(w)+1}\,dr<\infty.
\end{aligned} ( 2 σ − 1 ) ( Re ( s ) ) 2 E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) 1 { ∣ T σ ∣ < 1 } ] = ∫ ∣ x ∣ < 1 ∣ x ∣ 2 Re ( w ) h σ , Re ( s ) ( x ) d x ≤ 2 π ∥ h σ , Re ( s ) ∥ ∞ ∫ 0 1 r 2 Re ( w ) + 1 d r < ∞.
since Re ( w ) > − 1 \operatorname{Re}(w)>-1 Re ( w ) > − 1 . On the other hand, since Re ( w ) < 0 \operatorname{Re}(w)<0 Re ( w ) < 0 ,
∣ T σ ∣ 2 Re ( w ) ≤ 1 on { ∣ T σ ∣ ≥ 1 } , |T_\sigma|^{2\operatorname{Re}(w)}\le1 \qquad\text{on }\{|T_\sigma|\ge1\}, ∣ T σ ∣ 2 Re ( w ) ≤ 1 on { ∣ T σ ∣ ≥ 1 } ,
and therefore by (118) ,
E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) 1 { ∣ T σ ∣ ≥ 1 } ] ≤ E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞ . \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(w)}\mathbf{1}_{\{|T_\sigma|\ge1\}}\right]\le\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}\right]<\infty. E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) 1 { ∣ T σ ∣ ≥ 1 } ] ≤ E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞.
Thus (207) holds when − 1 < Re ( w ) < 0 -1<\operatorname{Re}(w)<0 − 1 < Re ( w ) < 0 . Suppose now that Re ( w ) ≥ 0 \operatorname{Re}(w)\ge0 Re ( w ) ≥ 0 . If Re ( w ) = 0 \operatorname{Re}(w)=0 Re ( w ) = 0 , then
E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) ] = E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞ . \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(w)}\right]=\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}\right]<\infty. E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) ] = E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞.
If Re ( w ) > 0 \operatorname{Re}(w)>0 Re ( w ) > 0 , using
T σ = ( 2 σ − 1 ) L ′ ( σ , X ) T_\sigma=(2\sigma-1)L'(\sigma,X) T σ = ( 2 σ − 1 ) L ′ ( σ , X )
and (118) , together with the Cauchy–Schwarz inequality gives
E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) ] ≤ ( 2 σ − 1 ) 2 Re ( w ) ( E [ ∣ ζ ( σ , X ) ∣ 4 Re ( s ) ] ) 1 / 2 ( E [ ∣ L ′ ( σ , X ) ∣ 4 Re ( w ) ] ) 1 / 2 < ∞ , \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(w)}\right]
\le(2\sigma-1)^{2\operatorname{Re}(w)}
\left(\mathbb{E}\left[|\zeta(\sigma,X)|^{4\operatorname{Re}(s)}\right]\right)^{1/2}
\left(\mathbb{E}\left[|L'(\sigma,X)|^{4\operatorname{Re}(w)}\right]\right)^{1/2}<\infty, E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) ] ≤ ( 2 σ − 1 ) 2 Re ( w ) ( E [ ∣ ζ ( σ , X ) ∣ 4 Re ( s ) ] ) 1/2 ( E [ ∣ L ′ ( σ , X ) ∣ 4 Re ( w ) ] ) 1/2 < ∞ ,
This proves (207) for all Re ( w ) > − 1 \operatorname{Re}(w)>-1 Re ( w ) > − 1 . Finally, since
0 < e − t ∣ T σ ∣ 2 ≤ 1 , 0<e^{-t|T_\sigma|^2}\le1, 0 < e − t ∣ T σ ∣ 2 ≤ 1 ,
we have
∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) e − t ∣ T σ ∣ 2 ≤ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) , |\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(w)}e^{-t|T_\sigma|^2}
\le|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}|T_\sigma|^{2\operatorname{Re}(w)}, ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) e − t ∣ T σ ∣ 2 ≤ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ∣ T σ ∣ 2 Re ( w ) ,
and hence (208) follows immediately from (207) .
The integrability established in Lemma 4.16 allows us to study the weighted mixed moments through a complex measure on R 2 \mathbb{R}^2 R 2 . To this end, we introduce the following weighted measure, whose Fourier transform is given by H σ , s H_{\sigma,s} H σ , s .
Recall the weighted complex Borel measure ν σ , s \nu_{\sigma,s} ν σ , s introduced in (12) . We first verify that this measure is finite. Fix σ > 1 / 2 \sigma>1/2 σ > 1/2 and s ∈ C s\in\mathbb{C} s ∈ C with Re ( s ) > 0 \operatorname{Re}(s)>0 Re ( s ) > 0 . By the moment estimate (118) ,
E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞ . \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}\right]<\infty. E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞.
Moreover, for every Borel set B ⊂ R 2 B\subset\mathbb{R}^2 B ⊂ R 2 ,
∣ ν σ , s ( B ) ∣ ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) 1 { T σ ∈ B } ] . |\nu_{\sigma,s}(B)|\le(2\sigma-1)^{\operatorname{Re}(s^2)}\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}\mathbf{1}_{\{T_\sigma\in B\}}\right]. ∣ ν σ , s ( B ) ∣ ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) 1 { T σ ∈ B } ] .
Consequently, the total variation measure satisfies
∣ ν σ , s ∣ ( R 2 ) ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞ . |\nu_{\sigma,s}|(\mathbb{R}^2)\le(2\sigma-1)^{\operatorname{Re}(s^2)}\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}(s)}\right]<\infty. ∣ ν σ , s ∣ ( R 2 ) ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re ( s ) ] < ∞.
Hence ν σ , s \nu_{\sigma,s} ν σ , s is a finite complex Borel measure.
The measure ν σ , s \nu_{\sigma,s} ν σ , s thus encodes the weighted distribution of T σ T_\sigma T σ . Recall that T σ T_\sigma T σ denotes ( 2 σ − 1 ) L ′ ( σ , X ) (2\sigma-1)L'(\sigma,X) ( 2 σ − 1 ) L ′ ( σ , X ) . The following lemma establishes the corresponding integration formula, allowing weighted expectations of Borel functions of T σ T_\sigma T σ to be expressed as integrals with respect to ν σ , s \nu_{\sigma,s} ν σ , s .
Lemma 4.17. Fix σ > 1 / 2 \sigma> 1/2 σ > 1/2 and s ∈ C s \in\mathbb{C} s ∈ C with Re s > 0 \operatorname{Re}s > 0 Re s > 0 . Let T σ T_\sigma T σ be as in (11) . Let ν σ , s \nu_{\sigma,s} ν σ , s be the measure defined in (12) . Let G : R 2 → C G : \mathbb{R}^{2} \to\mathbb{C} G : R 2 → C be a Borel measurable function such that
E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ G ( T σ ) ∣ ] < ∞ . (209) \mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}s}|G(T_\sigma)|\right] < \infty. \tag*{(209)} E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ G ( T σ ) ∣ ] < ∞. (209) Then G ∈ L 1 ( R 2 , d ∣ ν σ , s ∣ ) G \in L^{1}(\mathbb{R}^{2},d|\nu_{\sigma,s}|) G ∈ L 1 ( R 2 , d ∣ ν σ , s ∣ ) , and
∫ R 2 G ( x ) d ν σ , s ( x ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s G ( T σ ) ] . (210) \int_{\mathbb{R}^{2}} G(x)\,d\nu_{\sigma,s}(x)=(2\sigma-1)^{s^{2}}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}G(T_\sigma)\right]. \tag*{(210)} ∫ R 2 G ( x ) d ν σ , s ( x ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s G ( T σ ) ] . (210)
Proof. We first consider a Borel simple function
G ( x ) = ∑ j = 1 N c j 1 B j ( x ) , G(x)=\sum_{j=1}^{N}c_j1_{B_j}(x), G ( x ) = j = 1 ∑ N c j 1 B j ( x ) ,
where the B j B_j B j 's are Borel sets. By the definition of ν σ , s \nu_{\sigma,s} ν σ , s ,
∫ R 2 G ( x ) d ν σ , s ( x ) = ∑ j = 1 N c j ν σ , s ( B j ) = ( 2 σ − 1 ) s 2 ∑ j = 1 N c j E [ ∣ ζ ( σ , X ) ∣ 2 s 1 { T σ ∈ B j } ] = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s G ( T σ ) ] . \begin{aligned}
\int_{\mathbb{R}^{2}}G(x)\,d\nu_{\sigma,s}(x)=\sum_{j=1}^{N}c_j\nu_{\sigma,s}(B_j) \\
&=(2\sigma-1)^{s^{2}}\sum_{j=1}^{N}c_j\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}1_{\{T_\sigma\in B_j\}}\right] \\
&=(2\sigma-1)^{s^{2}}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}G(T_\sigma)\right].
\end{aligned} ∫ R 2 G ( x ) d ν σ , s ( x ) = j = 1 ∑ N c j ν σ , s ( B j ) = ( 2 σ − 1 ) s 2 j = 1 ∑ N c j E [ ∣ ζ ( σ , X ) ∣ 2 s 1 { T σ ∈ B j } ] = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s G ( T σ ) ] .
Thus (210) holds for Borel simple functions. Moreover, for every nonnegative Borel simple function Φ \Phi Φ , the definition of the total variation measure gives
∫ R 2 Φ ( x ) d ∣ ν σ , s ∣ ( x ) ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re s Φ ( T σ ) ] . \int_{\mathbb{R}^{2}}\Phi(x)\,d|\nu_{\sigma,s}|(x)\leq(2\sigma-1)^{\operatorname{Re}(s^{2})}\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}s}\Phi(T_\sigma)\right]. ∫ R 2 Φ ( x ) d ∣ ν σ , s ∣ ( x ) ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re s Φ ( T σ ) ] .
By monotone convergence, the same inequality holds for every nonnegative Borel function Φ \Phi Φ . Taking Φ = ∣ G ∣ \Phi=|G| Φ = ∣ G ∣ and using (209) , we obtain
∫ R 2 ∣ G ( x ) ∣ d ∣ ν σ , s ∣ ( x ) ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ G ( T σ ) ∣ ] < ∞ . \int_{\mathbb{R}^{2}}|G(x)|\,d|\nu_{\sigma,s}|(x)\leq(2\sigma-1)^{\operatorname{Re}(s^{2})}\mathbb{E}\left[|\zeta(\sigma,X)|^{2\operatorname{Re}s}|G(T_\sigma)|\right]<\infty. ∫ R 2 ∣ G ( x ) ∣ d ∣ ν σ , s ∣ ( x ) ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ G ( T σ ) ∣ ] < ∞.
Hence
G ∈ L 1 ( R 2 , d ∣ ν σ , s ∣ ) . G\in L^{1}(\mathbb{R}^{2},d|\nu_{\sigma,s}|). G ∈ L 1 ( R 2 , d ∣ ν σ , s ∣ ) .
Now choose Borel simple functions G n G_n G n such that
G n ( x ) ⟶ G ( x ) G_n(x)\longrightarrow G(x) G n ( x ) ⟶ G ( x )
pointwise and
∣ G n ( x ) ∣ ≤ ∣ G ( x ) ∣ , x ∈ R 2 . |G_n(x)|\leq|G(x)|,\qquad x\in\mathbb{R}^{2}. ∣ G n ( x ) ∣ ≤ ∣ G ( x ) ∣ , x ∈ R 2 .
For each n n n , the simple-function case gives
∫ R 2 G n ( x ) d ν σ , s ( x ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s G n ( T σ ) ] . \int_{\mathbb{R}^{2}}G_n(x)\,d\nu_{\sigma,s}(x)=(2\sigma-1)^{s^{2}}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}G_n(T_\sigma)\right]. ∫ R 2 G n ( x ) d ν σ , s ( x ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s G n ( T σ ) ] .
Since
∣ G n ( x ) ∣ ≤ ∣ G ( x ) ∣ ∈ L 1 ( d ∣ ν σ , s ∣ ) , |G_n(x)|\leq|G(x)|\in L^{1}(d|\nu_{\sigma,s}|), ∣ G n ( x ) ∣ ≤ ∣ G ( x ) ∣ ∈ L 1 ( d ∣ ν σ , s ∣ ) ,
dominated convergence on the left yields
∫ R 2 G n ( x ) d ν σ , s ( x ) ⟶ ∫ R 2 G ( x ) d ν σ , s ( x ) . \int_{\mathbb{R}^{2}} G_n(x)\,\mathrm{d}\nu_{\sigma,s}(x) \longrightarrow\int_{\mathbb{R}^{2}} G(x)\,\mathrm{d}\nu_{\sigma,s}(x). ∫ R 2 G n ( x ) d ν σ , s ( x ) ⟶ ∫ R 2 G ( x ) d ν σ , s ( x ) .
On the other hand,
∣ ∣ ζ ( σ , X ) ∣ 2 s G n ( T σ ) ∣ ≤ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ G ( T σ ) ∣ , \left|\lvert\zeta(\sigma,X)\rvert^{2s}G_n(T_\sigma)\right| \leq\lvert\zeta(\sigma,X)\rvert^{2\operatorname{Re}s}\lvert G(T_\sigma)\rvert, ∣ ζ ( σ , X ) ∣ 2 s G n ( T σ ) ≤ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ G ( T σ )∣ ,
which is integrable by (209) . Hence another application of the dominated convergence theorem gives
E [ ∣ ζ ( σ , X ) ∣ 2 s G n ( T σ ) ] ⟶ E [ ∣ ζ ( σ , X ) ∣ 2 s G ( T σ ) ] . \mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{2s}G_n(T_\sigma)\right] \longrightarrow\mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{2s}G(T_\sigma)\right]. E [ ∣ ζ ( σ , X ) ∣ 2 s G n ( T σ ) ] ⟶ E [ ∣ ζ ( σ , X ) ∣ 2 s G ( T σ ) ] .
Passing to the limit proves (210) .
We next establish the smoothness and integrability of H σ , s H_{\sigma,s} H σ , s , which are needed to apply the extended Riesz distributional pairing from Proposition 4.15. The proof relies on the moment estimates established earlier and the global Fourier decay in Proposition 1.7.
Lemma 4.18. Let K K K be a compact subset of the open right half-plane { s ∈ C : Re s > 0 } \{s \in\mathbb{C} : \operatorname{Re}s > 0\} { s ∈ C : Re s > 0 } . There exists σ 0 > 1 / 2 \sigma_0 > 1/2 σ 0 > 1/2 such that, for every
1 2 < σ ≤ σ 0 , s ∈ K , \frac{1}{2} < \sigma\leq\sigma_0,\qquad s \in K, 2 1 < σ ≤ σ 0 , s ∈ K , one has
H σ , s ∈ C b ∞ ( R 2 ) ∩ L 1 ( R 2 ) H_{\sigma,s} \in C_b^\infty(\mathbb{R}^{2}) \cap L^1(\mathbb{R}^{2}) H σ , s ∈ C b ∞ ( R 2 ) ∩ L 1 ( R 2 ) with H σ , s H_{\sigma,s} H σ , s defined in (13) . Moreover,
sup 1 / 2 < σ ≤ σ 0 s ∈ K ∥ H σ , s ∥ L 1 ( R 2 ) < ∞ . \sup_{\substack{1/2<\sigma\leq\sigma_0\\s\in K}}\lVert H_{\sigma,s}\rVert_{L^1(\mathbb{R}^{2})}<\infty. 1/2 < σ ≤ σ 0 s ∈ K sup ∥ H σ , s ∥ L 1 ( R 2 ) < ∞.
Here C b ∞ ( R 2 ) C_b^\infty(\mathbb{R}^{2}) C b ∞ ( R 2 ) denotes the space of infinitely differentiable functions on R 2 \mathbb{R}^{2} R 2 whose derivatives of all orders are bounded.
Proof. Recall that T σ T_\sigma T σ is given in (11) , that is
T σ = ( 2 σ − 1 ) L ′ ( σ , X ) . T_\sigma=(2\sigma-1)L'(\sigma,X). T σ = ( 2 σ − 1 ) L ′ ( σ , X ) .
Write
T σ = ( T σ , 1 , T σ , 2 ) T_\sigma=(T_{\sigma,1},T_{\sigma,2}) T σ = ( T σ , 1 , T σ , 2 )
and recall that in (13) ,
H σ , s ( ξ ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s e − i ⟨ ξ , T σ ⟩ ] . H_{\sigma,s}(\xi)=(2\sigma-1)^{s^2}\mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{2s}e^{-i\langle\xi,T_\sigma\rangle}\right]. H σ , s ( ξ ) = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s e − i ⟨ ξ , T σ ⟩ ] .
Let α = ( α 1 , α 2 ) ∈ Z ≥ 0 2 \alpha=(\alpha_1,\alpha_2)\in\mathbb{Z}_{\geq0}^{2} α = ( α 1 , α 2 ) ∈ Z ≥ 0 2 be a multi-index. By Lemma 4.2 and the Cauchy-Schwarz inequality,
E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ ∣ α ∣ ] ≤ ( 2 σ − 1 ) ∣ α ∣ ( E [ ∣ ζ ( σ , X ) ∣ 4 Re s ] ) 1 / 2 ( E [ ∣ L ′ ( σ , X ) ∣ 2 ∣ α ∣ ] ) 1 / 2 < ∞ . \mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{2\operatorname{Re}s}\lvert T_\sigma\rvert^{\lvert\alpha\rvert}\right]
\leq(2\sigma-1)^{\lvert\alpha\rvert}
\left(\mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{4\operatorname{Re}s}\right]\right)^{1/2}
\left(\mathbb{E}\left[\lvert L'(\sigma,X)\rvert^{2\lvert\alpha\rvert}\right]\right)^{1/2}<\infty. E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ ∣ α ∣ ] ≤ ( 2 σ − 1 ) ∣ α ∣ ( E [ ∣ ζ ( σ , X ) ∣ 4 Re s ] ) 1/2 ( E [ ∣ L ′ ( σ , X ) ∣ 2 ∣ α ∣ ] ) 1/2 < ∞.
Hence differentiation under the expectation is justified and yields
∂ α H σ , s ( ξ ) = ( − i ) ∣ α ∣ ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s T σ , 1 α 1 T σ , 2 α 2 e − i ⟨ ξ , T σ ⟩ ] . \partial^\alpha H_{\sigma,s}(\xi)=(-i)^{\lvert\alpha\rvert}(2\sigma-1)^{s^2}\mathbb{E}\left[\lvert\zeta(\sigma,X)\rvert^{2s}T_{\sigma,1}^{\alpha_1}T_{\sigma,2}^{\alpha_2}e^{-i\langle\xi,T_\sigma\rangle}\right]. ∂ α H σ , s ( ξ ) = ( − i ) ∣ α ∣ ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s T σ , 1 α 1 T σ , 2 α 2 e − i ⟨ ξ , T σ ⟩ ] .
Dominated convergence shows that these derivatives are continuous, and
sup ξ ∈ R 2 ∣ ∂ α H σ , s ( ξ ) ∣ ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ ∣ α ∣ ] < ∞ . \sup_{\xi\in\mathbb{R}^{2}} \left|\partial^\alpha H_{\sigma,s}(\xi)\right| \le(2\sigma-1)^{\operatorname{Re}(s^{2})}\mathbb{E}\left[\left|\zeta(\sigma,X)\right|^{2\operatorname{Re}s}\left|T_\sigma\right|^{|\alpha|}\right] < \infty. ξ ∈ R 2 sup ∣ ∂ α H σ , s ( ξ ) ∣ ≤ ( 2 σ − 1 ) Re ( s 2 ) E [ ∣ ζ ( σ , X ) ∣ 2 Re s ∣ T σ ∣ ∣ α ∣ ] < ∞.
Since α \alpha α is arbitrary, H σ , s ∈ C b ∞ ( R 2 ) H_{\sigma,s}\in C_b^\infty(\mathbb{R}^{2}) H σ , s ∈ C b ∞ ( R 2 ) . It remains to prove integrability. Applying Proposition 1.7 with, for instance, N = 2 N=2 N = 2 , and decreasing σ 0 \sigma_0 σ 0 if necessary, gives
∣ H σ , s ( λ , u ) ∣ ≤ C K , 2 ( 1 + λ 2 + u 2 ) − 2 \left|H_{\sigma,s}(\lambda,u)\right| \le C_{K,2}(1+\lambda^2+u^2)^{-2} ∣ H σ , s ( λ , u ) ∣ ≤ C K , 2 ( 1 + λ 2 + u 2 ) − 2
uniformly for
1 2 < σ ≤ σ 0 , s ∈ K . \frac{1}{2}<\sigma\le\sigma_0,\qquad s\in K. 2 1 < σ ≤ σ 0 , s ∈ K .
Therefore
∥ H σ , s ∥ L 1 ( R 2 ) ≤ C K , 2 ∫ R 2 ( 1 + λ 2 + u 2 ) − 2 d λ d u = 2 π C K , 2 ∫ 0 ∞ r ( 1 + r 2 ) 2 d r = π C K , 2 < ∞ . \begin{aligned}
\left\|H_{\sigma,s}\right\|_{L^1(\mathbb{R}^{2})}\le C_{K,2}\int_{\mathbb{R}^{2}}(1+\lambda^2+u^2)^{-2}\,d\lambda\,du \\
&=2\pi C_{K,2}\int_{0}^{\infty}\frac{r}{(1+r^2)^2}\,dr \\
&=\pi C_{K,2}<\infty.
\end{aligned} ∥ H σ , s ∥ L 1 ( R 2 ) ≤ C K , 2 ∫ R 2 ( 1 + λ 2 + u 2 ) − 2 d λ d u = 2 π C K , 2 ∫ 0 ∞ ( 1 + r 2 ) 2 r d r = π C K , 2 < ∞.
Hence H σ , s ∈ L 1 ( R 2 ) H_{\sigma,s}\in L^1(\mathbb{R}^{2}) H σ , s ∈ L 1 ( R 2 ) , and the same estimate also gives
sup 1 / 2 < σ ≤ σ 0 s ∈ K ∥ H σ , s ∥ L 1 ( R 2 ) ≤ π C K , 2 . \sup_{\substack{1/2<\sigma\le\sigma_0\\s\in K}}\left\|H_{\sigma,s}\right\|_{L^1(\mathbb{R}^{2})}\le\pi C_{K,2}. 1/2 < σ ≤ σ 0 s ∈ K sup ∥ H σ , s ∥ L 1 ( R 2 ) ≤ π C K , 2 .
This completes the proof.
By Lemmas 4.16 and 4.17 , the weighted mixed moments of ζ ( σ , X ) \zeta(\sigma,X) ζ ( σ , X ) and L ′ ( σ , X ) L'(\sigma,X) L ′ ( σ , X ) are well defined and can be expressed as integrals with respect to the weighted measure ν σ , s \nu_{\sigma,s} ν σ , s . Moreover, Lemma 4.18 ensures that H σ , s H_{\sigma,s} H σ , s satisfies the conditions required to apply the extended Riesz distributional pairing established in Proposition 4.15 . Combining these results with the Fourier transform identity for the homogeneous distribution ∣ x ∣ 2 s |x|^{2s} ∣ x ∣ 2 s established in (22) , we first remove the Gaussian regularization for each fixed σ > 1 / 2 \sigma>1/2 σ > 1/2 sufficiently close to 1 / 2 1/2 1/2 and derive a Fourier–Riesz representation of the original complex moments as in Lemma 1.8 . We then use this representation to study the limit as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 as in Lemma 4.20 , without interchanging the two limits t ↓ 0 t\downarrow0 t ↓ 0 and σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 .
Removal of Gaussian regularization and convergence
Removal of Gaussian regularization and proof of Lemma 1.8
Proof of Lemma 1.8. Let K K K be a compact subset of the open right half-plane { s ∈ C : Re s > 0 } \{s\in\mathbb{C}:\operatorname{Re}s>0\} { s ∈ C : Re s > 0 } , and let σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 be as in Lemma 4.18 . Fix
1 / 2 < σ ≤ σ 0 , s ∈ K ∩ Ω M . 1/2<\sigma\le\sigma_0,\qquad s\in K\cap\Omega_M. 1/2 < σ ≤ σ 0 , s ∈ K ∩ Ω M .
Throughout the proof, s s s is fixed and w w w is regarded as an auxiliary complex variable. By the definition of ν σ , s \nu_{\sigma,s} ν σ , s and the moment estimates (118), ν \nu ν is a finite complex Borel measure. Applying Lemma 4.17 with G ( x ) = e − i ⟨ ξ , x ⟩ G(x)=e^{-i\langle\xi,x\rangle} G ( x ) = e − i ⟨ ξ , x ⟩ , we obtain, with the Fourier transform convention,
H σ , s ( ξ ) = ν ^ σ , s ( ξ ) = ∫ R 2 e − i ⟨ ξ , x ⟩ d ν σ , s ( x ) . H_{\sigma,s}(\xi)=\widehat{\nu}_{\sigma,s}(\xi)=\int_{\mathbb{R}^{2}}e^{-i\langle\xi,x\rangle}\,d\nu_{\sigma,s}(x). H σ , s ( ξ ) = ν σ , s ( ξ ) = ∫ R 2 e − i ⟨ ξ , x ⟩ d ν σ , s ( x ) .
By Lemma 4.18 ,
H σ , s ∈ C b ∞ ( R 2 ) ∩ L 1 ( R 2 ) . H_{\sigma,s} \in C_b^\infty(\mathbb{R}^2) \cap L^1(\mathbb{R}^2). H σ , s ∈ C b ∞ ( R 2 ) ∩ L 1 ( R 2 ) .
Therefore, Fourier inversion for finite measures implies that ν \nu ν is absolutely continuous with respect to Lebesgue measure:
d ν σ , s ( x ) = h ( x ) d x , (211) \mathrm{d}\nu_{\sigma,s}(x) = h(x)\,\mathrm{d}x, \tag*{(211)} d ν σ , s ( x ) = h ( x ) d x , (211)
where
h ( x ) = 1 ( 2 π ) 2 ∫ R 2 e i ⟨ x , ξ ⟩ H σ , s ( ξ ) d ξ . (212) h(x) = \frac{1}{(2\pi)^2} \int_{\mathbb{R}^2} \mathrm{e}^{\mathrm{i}\langle x,\xi\rangle} H_{\sigma,s}(\xi)\,\mathrm{d}\xi. \tag*{(212)} h ( x ) = ( 2 π ) 2 1 ∫ R 2 e i ⟨ x , ξ ⟩ H σ , s ( ξ ) d ξ . (212)
In particular,
h ∈ C 0 ( R 2 ) ∩ L ∞ ( R 2 ) , (213) h \in C_0(\mathbb{R}^2) \cap L^\infty(\mathbb{R}^2), \tag*{(213)} h ∈ C 0 ( R 2 ) ∩ L ∞ ( R 2 ) , (213)
where C 0 ( R 2 ) C_0(\mathbb{R}^2) C 0 ( R 2 ) denotes the space of continuous functions vanishing at infinity. Hence, by (213) ,
x ⟼ ∣ x ∣ 2 Re ( w ) ∈ L 2 ( d ∣ ν σ , s ∣ ) (214) x \longmapsto|x|^{2\operatorname{Re}(w)} \in L^2(\mathrm{d}|\nu_{\sigma,s}|) \tag*{(214)} x ⟼ ∣ x ∣ 2 Re ( w ) ∈ L 2 ( d ∣ ν σ , s ∣ ) (214)
when − 1 < Re ( w ) < 0 -1 < \operatorname{Re}(w) < 0 − 1 < Re ( w ) < 0 . In fact, the integrability near the origin follows from the bounded density, while for ∣ x ∣ ≥ 1 |x| \ge1 ∣ x ∣ ≥ 1 one has ∣ x ∣ 2 Re ( w ) ≤ 1 |x|^{2\operatorname{Re}(w)} \le1 ∣ x ∣ 2 Re ( w ) ≤ 1 . Applying Lemmas 4.16 and 4.17 , for any w ∈ C w \in\mathbb{C} w ∈ C with Re ( w ) > − 1 \operatorname{Re}(w) > -1 Re ( w ) > − 1 and t > 0 t > 0 t > 0 ,
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w e − t ∣ T σ ∣ 2 ] = ∫ R 2 ∣ x ∣ 2 w e − t ∣ x ∣ 2 d ν σ , s ( x ) . (215) (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2w}\mathrm{e}^{-t|T_\sigma|^2}\right]
= \int_{\mathbb{R}^2}|x|^{2w}\mathrm{e}^{-t|x|^2}\,\mathrm{d}\nu_{\sigma,s}(x). \tag*{(215)} ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w e − t ∣ T σ ∣ 2 ] = ∫ R 2 ∣ x ∣ 2 w e − t ∣ x ∣ 2 d ν σ , s ( x ) . (215)
We first let t ↓ 0 t \downarrow0 t ↓ 0 in the strip
− 1 < Re ( w ) < 0. -1 < \operatorname{Re}(w) < 0. − 1 < Re ( w ) < 0.
By (214) and the dominated convergence give
lim t ↓ 0 ∫ R 2 ∣ x ∣ 2 w e − t ∣ x ∣ 2 d ν σ , s ( x ) = ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) . (216) \lim_{t\downarrow0}\int_{\mathbb{R}^2}|x|^{2w}\mathrm{e}^{-t|x|^2}\,\mathrm{d}\nu_{\sigma,s}(x)
= \int_{\mathbb{R}^2}|x|^{2w}\,\mathrm{d}\nu_{\sigma,s}(x). \tag*{(216)} t ↓ 0 lim ∫ R 2 ∣ x ∣ 2 w e − t ∣ x ∣ 2 d ν σ , s ( x ) = ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) . (216)
We next consider the left-hand side of (223) as t ↓ 0 t \downarrow0 t ↓ 0 . By the Fourier representation proved in Lemma 4.12 gives
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w e − t ∣ T σ ∣ 2 ] = Γ ( w + 1 ) 4 π t w + 1 ∫ R 2 1 F 1 ( w + 1 ; 1 ; − ∣ ξ ∣ 2 4 t ) H σ , s ( ξ ) d ξ . (217) (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2w}\mathrm{e}^{-t|T_\sigma|^2}\right]
= \frac{\Gamma(w+1)}{4\pi t^{w+1}}\int_{\mathbb{R}^2}{}_1F_1\left(w+1;1;-\frac{|\xi|^2}{4t}\right)H_{\sigma,s}(\xi)\,\mathrm{d}\xi. \tag*{(217)} ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w e − t ∣ T σ ∣ 2 ] = 4 π t w + 1 Γ ( w + 1 ) ∫ R 2 1 F 1 ( w + 1 ; 1 ; − 4 t ∣ ξ ∣ 2 ) H σ , s ( ξ ) d ξ . (217)
For − 1 < Re ( w ) < 0 -1 < \operatorname{Re}(w) < 0 − 1 < Re ( w ) < 0 , the Kummer asymptotic proved in (177) gives, for every ξ ≠ 0 \xi\ne0 ξ = 0 ,
t w − 1 1 F 1 ( w + 1 ; 1 ; − ∣ ξ ∣ 2 4 t ) ⟶ 4 w + 1 Γ ( − w ) ∣ ξ ∣ − 2 w − 2 , t ↓ 0. (218) t^{w-1}{}_1F_1\left(w+1;1;-\frac{|\xi|^2}{4t}\right)
\longrightarrow\frac{4^{w+1}}{\Gamma(-w)}|\xi|^{-2w-2}, \qquad t \downarrow0. \tag*{(218)} t w − 1 1 F 1 ( w + 1 ; 1 ; − 4 t ∣ ξ ∣ 2 ) ⟶ Γ ( − w ) 4 w + 1 ∣ ξ ∣ − 2 w − 2 , t ↓ 0. (218)
Moreover, on compact subsets of
− 1 < Re ( w ) < 0 , -1 < \operatorname{Re}(w) < 0, − 1 < Re ( w ) < 0 ,
one has, uniformly for 0 < t ≤ 1 0 < t \le1 0 < t ≤ 1 ,
∣ t w − 1 1 F 1 ( w + 1 ; 1 ; − ∣ ξ ∣ 2 4 t ) ∣ ≤ C w ∣ ξ ∣ − 2 Re ( w ) − 2 . (219) \left|t^{w-1}{}_1F_1\left(w+1;1;-\frac{|\xi|^2}{4t}\right)\right|
\le C_w|\xi|^{-2\operatorname{Re}(w)-2}. \tag*{(219)} t w − 1 1 F 1 ( w + 1 ; 1 ; − 4 t ∣ ξ ∣ 2 ) ≤ C w ∣ ξ ∣ − 2 Re ( w ) − 2 . (219)
Since Re ( w ) < 0 \operatorname{Re}(w) < 0 Re ( w ) < 0 ,
∣ ξ ∣ − 2 Re ( w ) − 2 |\xi|^{-2\operatorname{Re}(w)-2} ∣ ξ ∣ − 2 Re ( w ) − 2
is locally integrable at ξ = 0 \xi= 0 ξ = 0 in two dimensions. At infinity, the decay condition (187) implies that
∣ ξ ∣ − 2 Re ( w ) − 2 ∣ H σ , s ( ξ ) ∣ |\xi|^{-2\operatorname{Re}(w)-2}|H_{\sigma,s}(\xi)| ∣ ξ ∣ − 2 Re ( w ) − 2 ∣ H σ , s ( ξ ) ∣
is integrable. Hence, by (207) and (216) , together with the dominated convergence on the right-hand side of (217) yields
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] = ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) = 2 2 w Γ ( w + 1 ) π Γ ( − w ) ∫ R 2 ∣ ξ ∣ − 2 w − 2 H σ , s ( ξ ) d ξ . (220) (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2w}\right]=\int_{\mathbb{R}^2}|x|^{2w}\,d\nu_{\sigma,s}(x)=\frac{2^{2w}\Gamma(w+1)}{\pi\Gamma(-w)}\int_{\mathbb{R}^2}|\xi|^{-2w-2}H_{\sigma,s}(\xi)\,d\xi. \tag*{(220)} ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] = ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) = π Γ ( − w ) 2 2 w Γ ( w + 1 ) ∫ R 2 ∣ ξ ∣ − 2 w − 2 H σ , s ( ξ ) d ξ . (220)
In the strip − 1 < Re w < 0 -1<\operatorname{Re}w<0 − 1 < Re w < 0 , the integral is absolutely convergent, and the ordinary pairing agrees with the extended Gelfand–Shilov pairing:
∫ R 2 ∣ ξ ∣ − 2 w − 2 H σ , s ( ξ ) d ξ = ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H σ , s ⟩ e x t . \int_{\mathbb{R}^2}|\xi|^{-2w-2}H_{\sigma,s}(\xi)\,d\xi=\left\langle\operatorname{Reg}|\xi|^{-2w-2},H_{\sigma,s}\right\rangle_{\mathrm{ext}}. ∫ R 2 ∣ ξ ∣ − 2 w − 2 H σ , s ( ξ ) d ξ = ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H σ , s ⟩ ext .
Applying Proposition 4.15 , we obtain
⟨ Reg ∣ ξ ∣ − 2 w − 2 , H σ , s ⟩ e x t = 2 π R M , w [ H σ , s ] , \left\langle\operatorname{Reg}|\xi|^{-2w-2},H_{\sigma,s}\right\rangle_{\mathrm{ext}}=2\pi\mathcal{R}_{M,w}[H_{\sigma,s}], ⟨ Reg ∣ ξ ∣ − 2 w − 2 , H σ , s ⟩ ext = 2 π R M , w [ H σ , s ] ,
and hence
∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) = 2 2 w + 1 Γ ( w + 1 ) Γ ( − w ) R M , w [ H σ , s ] , − 1 < Re w < 0. (221) \int_{\mathbb{R}^2}|x|^{2w}\,d\nu_{\sigma,s}(x)=2^{2w+1}\frac{\Gamma(w+1)}{\Gamma(-w)}\mathcal{R}_{M,w}[H_{\sigma,s}],\qquad-1<\operatorname{Re}w<0. \tag*{(221)} ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) = 2 2 w + 1 Γ ( − w ) Γ ( w + 1 ) R M , w [ H σ , s ] , − 1 < Re w < 0. (221)
By Proposition 4.15 ,
w ⟼ R M , w [ H σ , s ] w\longmapsto\mathcal{R}_{M,w}[H_{\sigma,s}] w ⟼ R M , w [ H σ , s ]
is holomorphic on Ω M \Omega_M Ω M . Therefore
w ⟼ 2 2 w + 1 Γ ( w + 1 ) Γ ( − w ) R M , w [ H σ , s ] (222) w\longmapsto2^{2w+1}\frac{\Gamma(w+1)}{\Gamma(-w)}\mathcal{R}_{M,w}[H_{\sigma,s}] \tag*{(222)} w ⟼ 2 2 w + 1 Γ ( − w ) Γ ( w + 1 ) R M , w [ H σ , s ] (222)
is holomorphic on Ω M \Omega_M Ω M . Hence the function (222) gives an analytic continuation of w ⟼ ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] w\longmapsto(2\sigma-1)^{s^2}\mathbb{E}[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2w}] w ⟼ ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] from the initial strip { w ∈ C : − 1 < Re ( w ) < 0 } \{w\in\mathbb{C}:-1<\operatorname{Re}(w)<0\} { w ∈ C : − 1 < Re ( w ) < 0 } to Ω M \Omega_M Ω M . Note that, for any w ∈ C w\in\mathbb{C} w ∈ C with Re ( w ) > − 1 \operatorname{Re}(w)>-1 Re ( w ) > − 1 ,
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] = ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) (223) (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2w}\right]=\int_{\mathbb{R}^2}|x|^{2w}\,d\nu_{\sigma,s}(x) \tag*{(223)} ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] = ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) (223)
by Lemmas 4.16 and 4.17 . To prove the claim of this lemma, it remains only to show that
F σ , s ( w ) : = ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) (224) F_{\sigma,s}(w):=\int_{\mathbb{R}^2}|x|^{2w}\,d\nu_{\sigma,s}(x) \tag*{(224)} F σ , s ( w ) := ∫ R 2 ∣ x ∣ 2 w d ν σ , s ( x ) (224)
is holomorphic on { Re w > − 1 } \{\operatorname{Re}w>-1\} { Re w > − 1 } . Indeed, once this is established, the connectedness of Ω M \Omega_M Ω M , together with the identity theorem applied to (221) , implies, in view of (223) , that
( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] = 2 2 w + 1 Γ ( w + 1 ) Γ ( − w ) R M , w [ H σ , s ] , w ∈ Ω M . (225) (2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2w}\right]=2^{2w+1}\frac{\Gamma(w+1)}{\Gamma(-w)}\mathcal{R}_{M,w}[H_{\sigma,s}],\qquad w\in\Omega_M. \tag*{(225)} ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 w ] = 2 2 w + 1 Γ ( − w ) Γ ( w + 1 ) R M , w [ H σ , s ] , w ∈ Ω M . (225)
Equation (23) then follows by setting w = s w=s w = s in (225) .
It therefore remains to prove that F σ , s ( w ) F_{\sigma,s}(w) F σ , s ( w ) is holomorphic on { w ∈ C : Re w > − 1 } \{w\in\mathbb{C}:\operatorname{Re}w>-1\} { w ∈ C : Re w > − 1 } . Let L ⊂ { w ∈ C : Re w > − 1 } L\subset\{w\in\mathbb{C}:\operatorname{Re}w>-1\} L ⊂ { w ∈ C : Re w > − 1 } be compact, and set
a : = inf w ∈ L Re w > − 1 , b : = sup w ∈ L Re w . a:=\inf_{w\in L}\operatorname{Re}w>-1,\qquad b:=\sup_{w\in L}\operatorname{Re}w. a := w ∈ L inf Re w > − 1 , b := w ∈ L sup Re w .
For x ≠ 0 x\ne0 x = 0 , the function
w ⟼ ∣ x ∣ 2 w = exp ( 2 w log ∣ x ∣ ) w\longmapsto|x|^{2w}=\exp(2w\log|x|) w ⟼ ∣ x ∣ 2 w = exp ( 2 w log ∣ x ∣ )
is entire, and for every k ∈ Z ≥ 0 k\in\mathbb{Z}_{\ge0} k ∈ Z ≥ 0 ,
∂ k ∂ w k ∣ x ∣ 2 w = ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w . \frac{\partial^k}{\partial w^k}|x|^{2w}=(2\log|x|)^k|x|^{2w}. ∂ w k ∂ k ∣ x ∣ 2 w = ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w .
We now establish an integrable majorant, uniformly for w ∈ L w\in L w ∈ L . For ∣ x ∣ ≤ 1 |x|\le1 ∣ x ∣ ≤ 1 ,
∣ ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w ∣ ≤ 2 k ∣ log ∣ x ∣ ∣ k ∣ x ∣ 2 a . |(2\log|x|)^k|x|^{2w}|\le2^k|\log|x||^k|x|^{2a}. ∣ ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w ∣ ≤ 2 k ∣ log ∣ x ∣ ∣ k ∣ x ∣ 2 a .
By (211) , d ν σ , s = h d x d\nu_{\sigma,s}=h\,dx d ν σ , s = h d x with h ∈ L ∞ ( R 2 ) h\in L^\infty(\mathbb{R}^2) h ∈ L ∞ ( R 2 ) , we have
∫ ∣ x ∣ ≤ 1 ∣ log ∣ x ∣ ∣ k ∣ x ∣ 2 a d ∣ ν σ , s ∣ ( x ) = ∫ ∣ x ∣ ≤ 1 ∣ log ∣ x ∣ ∣ k ∣ x ∣ 2 a ∣ h ( x ) ∣ d x \int_{|x|\le1}|\log|x||^k|x|^{2a}\,d|\nu_{\sigma,s}|(x)=\int_{|x|\le1}|\log|x||^k|x|^{2a}|h(x)|\,dx ∫ ∣ x ∣ ≤ 1 ∣ log ∣ x ∣ ∣ k ∣ x ∣ 2 a d ∣ ν σ , s ∣ ( x ) = ∫ ∣ x ∣ ≤ 1 ∣ log ∣ x ∣ ∣ k ∣ x ∣ 2 a ∣ h ( x ) ∣ d x
≤ 2 π ∥ h ∥ ∞ ∫ 0 1 ∣ log r ∣ k r 2 a + 1 d r < ∞ , \le2\pi\|h\|_\infty\int_0^1|\log r|^k r^{2a+1}\,dr<\infty, ≤ 2 π ∥ h ∥ ∞ ∫ 0 1 ∣ log r ∣ k r 2 a + 1 d r < ∞ ,
because a > − 1 a>-1 a > − 1 . For ∣ x ∣ ≥ 1 |x|\ge1 ∣ x ∣ ≥ 1 , choose q > max { 2 b , 0 } q>\max\{2b,0\} q > max { 2 b , 0 } . Then
( log r ) k r 2 b ≪ L , k , q r q , r ≥ 1 , (\log r)^k r^{2b}\ll_{L,k,q}r^q,\qquad r\ge1, ( log r ) k r 2 b ≪ L , k , q r q , r ≥ 1 ,
and therefore
∣ ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w ∣ ≪ L , k , q ∣ x ∣ q , ∣ x ∣ ≥ 1. |(2\log|x|)^k|x|^{2w}|\ll_{L,k,q}|x|^q,\qquad|x|\ge1. ∣ ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w ∣ ≪ L , k , q ∣ x ∣ q , ∣ x ∣ ≥ 1.
Moreover,
∫ R 2 ∣ x ∣ q d ∣ ν σ , s ∣ ( x ) < ∞ \int_{\mathbb{R}^2}|x|^q\,d|\nu_{\sigma,s}|(x)<\infty ∫ R 2 ∣ x ∣ q d ∣ ν σ , s ∣ ( x ) < ∞
by Lemma 4.17 and (207) . Thus, for every k ≥ 0 k\ge0 k ≥ 0 , the family
{ ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w : w ∈ L } \left\{(2\log|x|)^k|x|^{2w}:w\in L\right\} { ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w : w ∈ L }
is dominated by a d ∣ ν σ , s ∣ d|\nu_{\sigma,s}| d ∣ ν σ , s ∣ -integrable function. Hence differentiation under the integral sign is justified, and
F σ , s ( k ) ( w ) = ∫ R 2 ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w d ν σ , s ( x ) , w ∈ { w ∈ C : Re w > − 1 } . F_{\sigma,s}^{(k)}(w)=\int_{\mathbb{R}^2}(2\log|x|)^k|x|^{2w}\,d\nu_{\sigma,s}(x),\qquad w\in\{w\in\mathbb{C}:\operatorname{Re}w>-1\}. F σ , s ( k ) ( w ) = ∫ R 2 ( 2 log ∣ x ∣ ) k ∣ x ∣ 2 w d ν σ , s ( x ) , w ∈ { w ∈ C : Re w > − 1 } .
In particular, F σ , s F_{\sigma,s} F σ , s is holomorphic on { w ∈ C : Re w > − 1 } \{w\in\mathbb{C}:\operatorname{Re}w>-1\} { w ∈ C : Re w > − 1 } .
□ \square □
Convergence of the regularized Fourier–Riesz functional
We now turn to the limit σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 in the regularized Fourier–Riesz functional
R M , s [ H σ , s ] . \mathcal{R}_{M,s}[H_{\sigma,s}]. R M , s [ H σ , s ] .
We treat separately the contributions from a neighbourhood of the origin in Fourier space, the Fourier tail, and the finite correction terms arising from Taylor subtraction. Near the origin, the factor r − 2 s − 1 r^{-2s-1} r − 2 s − 1 is singular, so local uniform convergence of H σ , s H_{\sigma,s} H σ , s alone is not sufficient. Instead, we use the local Fourier derivative bounds to subtract the circular Taylor polynomial uniformly and obtain a remainder of order r 2 M + 2 r^{2M+2} r 2 M + 2 . This produces an integrable majorant provided sup s ∈ K Re ( s ) < M + 1 \sup_{s\in K}\operatorname{Re}(s)<M+1 sup s ∈ K Re ( s ) < M + 1 . The precise estimate is given in the following lemma.
Lemma 4.19 (Uniform control near the Fourier origin). Let K K K be a compact subset of the open right half-plane
{ s ∈ C : Re s > 0 } . \{s \in\mathbb{C} : \operatorname{Re} s > 0\}. { s ∈ C : Re s > 0 } . and let M ∈ Z ≥ 0 M \in\mathbb{Z}_{\ge0} M ∈ Z ≥ 0 satisfy
sup s ∈ K Re s < M + 1. (226) \sup_{s\in K} \operatorname{Re} s < M+1. \tag*{(226)} s ∈ K sup Re s < M + 1. (226) Let H σ , s H_{\sigma,s} H σ , s be defined in (13) . Then there exist σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 and C K , M > 0 C_{K,M}>0 C K , M > 0 such that
∣ A H σ , s ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H σ , s ( 0 ) ∣ ≤ C K , M r 2 M + 2 \left|\mathcal{A}H_{\sigma,s}(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^j(j!)^2}\Delta^jH_{\sigma,s}(0)\right|\le C_{K,M}r^{2M+2} A H σ , s ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H σ , s ( 0 ) ≤ C K , M r 2 M + 2 uniformly for
1 2 < σ ≤ σ 0 , s ∈ K , 0 ≤ r ≤ 1. \frac{1}{2}<\sigma\le\sigma_0,\qquad s\in K,\qquad0\le r\le1. 2 1 < σ ≤ σ 0 , s ∈ K , 0 ≤ r ≤ 1. Moreover, as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 ,
∫ 0 1 [ A H σ , s ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H σ , s ( 0 ) ] r − 2 Re s − 1 d r ⟶ ∫ 0 1 [ A H s ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H s ( 0 ) ] r − 2 s − 1 d r \int_0^1\left[\mathcal{A}H_{\sigma,s}(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^j(j!)^2}\Delta^jH_{\sigma,s}(0)\right]r^{-2\operatorname{Re}s-1}\,dr
\longrightarrow
\int_0^1\left[\mathcal{A}H_s(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^j(j!)^2}\Delta^jH_s(0)\right]r^{-2s-1}\,dr ∫ 0 1 [ A H σ , s ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H σ , s ( 0 ) ] r − 2 Re s − 1 d r ⟶ ∫ 0 1 [ A H s ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H s ( 0 ) ] r − 2 s − 1 d r uniformly for s ∈ K s\in K s ∈ K . Here,
H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) / 4 (227) H_s(\lambda,u)=a_se^{-is\lambda-(\lambda^2+u^2)/4} \tag*{(227)} H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) /4 (227) and A H \mathcal{A}H A H is defined by (201) .
Proof. Taking R = 1 R=1 R = 1 in Lemma 4.10 and choosing a common σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 for the finitely many pairs
( α , β ) = ( j , 2 M + 2 − j ) , 0 ≤ j ≤ 2 M + 2 , (\alpha,\beta)=(j,2M+2-j),\qquad0\le j\le2M+2, ( α , β ) = ( j , 2 M + 2 − j ) , 0 ≤ j ≤ 2 M + 2 ,
the chain rule gives
sup 1 / 2 < σ ≤ σ 0 s ∈ K , 0 ≤ r ≤ 1 θ ∈ [ 0 , 2 π ] ∣ d 2 M + 2 d r 2 M + 2 H σ , s ( r cos θ , r sin θ ) ∣ ≤ C K , M < ∞ . \sup_{\substack{1/2<\sigma\le\sigma_0\\s\in K,\ 0\le r\le1\\\theta\in[0,2\pi]}}
\left|\frac{d^{2M+2}}{dr^{2M+2}}H_{\sigma,s}(r\cos\theta,r\sin\theta)\right|\le C_{K,M}<\infty. 1/2 < σ ≤ σ 0 s ∈ K , 0 ≤ r ≤ 1 θ ∈ [ 0 , 2 π ] sup d r 2 M + 2 d 2 M + 2 H σ , s ( r cos θ , r sin θ ) ≤ C K , M < ∞.
By Taylor’s theorem at r = 0 r=0 r = 0 , with the remainder in integral form, we obtain
H σ , s ( r cos θ , r sin θ ) = ∑ k = 0 2 M + 1 r k k ! ( cos θ ∂ λ + sin θ ∂ u ) k H σ , s ( 0 ) + O K , M ( r 2 M + 2 ) , (228) H_{\sigma,s}(r\cos\theta,r\sin\theta)=\sum_{k=0}^{2M+1}\frac{r^k}{k!}(\cos\theta\,\partial_\lambda+\sin\theta\,\partial_u)^kH_{\sigma,s}(0)+O_{K,M}(r^{2M+2}), \tag*{(228)} H σ , s ( r cos θ , r sin θ ) = k = 0 ∑ 2 M + 1 k ! r k ( cos θ ∂ λ + sin θ ∂ u ) k H σ , s ( 0 ) + O K , M ( r 2 M + 2 ) , (228)
where the error term is uniform for
1 2 < σ ≤ σ 0 , s ∈ K , 0 ≤ r ≤ 1 , θ ∈ [ 0 , 2 π ] . \frac{1}{2}<\sigma\le\sigma_0,\qquad s\in K,\qquad0\le r\le1,\qquad\theta\in[0,2\pi]. 2 1 < σ ≤ σ 0 , s ∈ K , 0 ≤ r ≤ 1 , θ ∈ [ 0 , 2 π ] .
Averaging (228) in θ \theta θ , all odd terms vanish. For the even terms,
1 2 π ∫ 0 2 π ( cos θ ∂ λ + sin θ ∂ u ) 2 j d θ = ( 2 j ) ! 4 j ( j ! ) 2 Δ j , (229) \frac{1}{2\pi}\int_0^{2\pi}(\cos\theta\,\partial_\lambda+\sin\theta\,\partial_u)^{2j}\,d\theta=\frac{(2j)!}{4^j(j!)^2}\Delta^j, \tag*{(229)} 2 π 1 ∫ 0 2 π ( cos θ ∂ λ + sin θ ∂ u ) 2 j d θ = 4 j ( j ! ) 2 ( 2 j )! Δ j , (229)
where
Δ = ∂ λ 2 + ∂ u 2 . \Delta= \partial_{\lambda}^{2}+\partial_{u}^{2}. Δ = ∂ λ 2 + ∂ u 2 .
Indeed, note that the differential operators ∂ λ \partial_{\lambda} ∂ λ and ∂ u \partial_{u} ∂ u commute on sufficiently smooth functions, the operator on the left-hand side is a rotationally invariant homogeneous polynomial of degree 2 j 2j 2 j in ∂ λ , ∂ u \partial_{\lambda},\partial_{u} ∂ λ , ∂ u , and is therefore a constant multiple of Δ j \Delta^{j} Δ j . The constant is determined by the coefficient of ∂ λ 2 j \partial_{\lambda}^{2j} ∂ λ 2 j :
1 2 π ∫ 0 2 π cos 2 j θ d θ = ( 2 j ) ! 4 j ( j ! ) 2 . \frac{1}{2\pi}\int_{0}^{2\pi}\cos^{2j}\theta\,\mathrm{d}\theta=\frac{(2j)!}{4^{j}(j!)^{2}}. 2 π 1 ∫ 0 2 π cos 2 j θ d θ = 4 j ( j ! ) 2 ( 2 j )! .
It follows that
A H σ , s ( r ) = ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H σ , s ( 0 ) + O K , M ( r 2 M + 2 ) , (230) \mathcal{A}H_{\sigma,s}(r)=\sum_{j=0}^{M}\frac{r^{2j}}{4^{j}(j!)^{2}}\Delta^{j}H_{\sigma,s}(0)+O_{K,M}(r^{2M+2}), \tag*{(230)} A H σ , s ( r ) = j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H σ , s ( 0 ) + O K , M ( r 2 M + 2 ) , (230)
uniformly for
1 2 < σ ≤ σ 0 , s ∈ K , 0 ≤ r ≤ 1. \frac{1}{2}<\sigma\leq\sigma_{0},\qquad s\in K,\qquad0\leq r\leq1. 2 1 < σ ≤ σ 0 , s ∈ K , 0 ≤ r ≤ 1.
The locally uniform convergence in Lemma 4.10 also implies the corresponding uniform derivative bounds for the limiting function H s H_s H s . Hence the same argument gives
A H s ( r ) = ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H s ( 0 ) + O K , M ( r 2 M + 2 ) , (231) \mathcal{A}H_s(r)=\sum_{j=0}^{M}\frac{r^{2j}}{4^{j}(j!)^{2}}\Delta^{j}H_s(0)+O_{K,M}(r^{2M+2}), \tag*{(231)} A H s ( r ) = j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H s ( 0 ) + O K , M ( r 2 M + 2 ) , (231)
uniformly for
s ∈ K , 0 ≤ r ≤ 1. s\in K,\qquad0\leq r\leq1. s ∈ K , 0 ≤ r ≤ 1.
Define
R σ , s ( r ) : = A H σ , s ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H σ , s ( 0 ) (232) R_{\sigma,s}(r):=\mathcal{A}H_{\sigma,s}(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^{j}(j!)^{2}}\Delta^{j}H_{\sigma,s}(0) \tag*{(232)} R σ , s ( r ) := A H σ , s ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H σ , s ( 0 ) (232)
and
R s ( r ) : = A H s ( r ) − ∑ j = 0 M r 2 j 4 j ( j ! ) 2 Δ j H s ( 0 ) . (233) R_s(r):=\mathcal{A}H_s(r)-\sum_{j=0}^{M}\frac{r^{2j}}{4^{j}(j!)^{2}}\Delta^{j}H_s(0). \tag*{(233)} R s ( r ) := A H s ( r ) − j = 0 ∑ M 4 j ( j ! ) 2 r 2 j Δ j H s ( 0 ) . (233)
Equations (230) and (231) imply
∣ R σ , s ( r ) ∣ + ∣ R s ( r ) ∣ ≤ C K , M r 2 M + 2 , 0 ≤ r ≤ 1. (234) |R_{\sigma,s}(r)|+|R_s(r)|\leq C_{K,M}r^{2M+2},\qquad0\leq r\leq1. \tag*{(234)} ∣ R σ , s ( r ) ∣ + ∣ R s ( r ) ∣ ≤ C K , M r 2 M + 2 , 0 ≤ r ≤ 1. (234)
Moreover, Lemma 4.10 gives
sup s ∈ K 0 ≤ r ≤ 1 ∣ A H σ , s ( r ) − A H s ( r ) ∣ ⟶ 0 \sup_{\substack{s\in K\\0\leq r\leq1}}|\mathcal{A}H_{\sigma,s}(r)-\mathcal{A}H_s(r)|\longrightarrow0 s ∈ K 0 ≤ r ≤ 1 sup ∣ A H σ , s ( r ) − A H s ( r ) ∣ ⟶ 0
and, for every $0\leq j\leq M,
sup s ∈ K ∣ Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) ∣ ⟶ 0. \sup_{s\in K}|\Delta^{j}H_{\sigma,s}(0)-\Delta^{j}H_s(0)|\longrightarrow0. s ∈ K sup ∣ Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) ∣ ⟶ 0.
Therefore
sup s ∈ K 0 ≤ r ≤ 1 ∣ R σ , s ( r ) − R s ( r ) ∣ ⟶ 0 (235) \sup_{\substack{s\in K\\0\leq r\leq1}}|R_{\sigma,s}(r)-R_s(r)|\longrightarrow0 \tag*{(235)} s ∈ K 0 ≤ r ≤ 1 sup ∣ R σ , s ( r ) − R s ( r ) ∣ ⟶ 0 (235)
as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . It remains to prove the uniform convergence of the integrals. By (226) and (234) ,
∣ ( R σ , s ( r ) − R s ( r ) ) r − 2 s − 1 ∣ ≤ C K , M r 2 M + 1 − 2 sup s ∈ K Re s ∈ L 1 ( 0 , 1 ] . |(R_{\sigma,s}(r)-R_s(r))r^{-2s-1}|\leq C_{K,M}r^{2M+1-2\sup_{s\in K}\operatorname{Re}s}\in L^{1}(0,1]. ∣ ( R σ , s ( r ) − R s ( r )) r − 2 s − 1 ∣ ≤ C K , M r 2 M + 1 − 2 s u p s ∈ K Re s ∈ L 1 ( 0 , 1 ] .
Hence, for ϵ > 0 \epsilon> 0 ϵ > 0 , choose δ ∈ ( 0 , 1 ) \delta\in(0,1) δ ∈ ( 0 , 1 ) such that
∫ 0 δ ∣ R σ , s ( r ) − R s ( r ) ∣ r − 2 Re s − 1 d r < ϵ 2 . \int_{0}^{\delta} \lvert R_{\sigma,s}(r)-R_s(r)\rvert r^{-2\operatorname{Re}s-1}\,\mathrm{d}r < \frac{\epsilon}{2}. ∫ 0 δ ∣ R σ , s ( r ) − R s ( r )∣ r − 2 Re s − 1 d r < 2 ϵ .
uniformly for s ∈ K s \in K s ∈ K . On the interval [ δ , 1 ] [\delta,1] [ δ , 1 ] , one has
sup s ∈ K ∣ ∫ δ 1 ( R σ , s ( r ) − R s ( r ) ) r − 2 s − 1 d r ∣ ≤ δ − 2 sup s ∈ K Re s − 1 sup s ∈ K δ ≤ r ≤ 1 ∣ R σ , s ( r ) − R s ( r ) ∣ . \sup_{s\in K}\left\lvert\int_{\delta}^{1}\left(R_{\sigma,s}(r)-R_s(r)\right)r^{-2s-1}\,\mathrm{d}r\right\rvert
\leq\delta^{-2\sup_{s\in K}\operatorname{Re}s-1}\sup_{\substack{s\in K\\\delta\leq r\leq1}}\lvert R_{\sigma,s}(r)-R_s(r)\rvert. s ∈ K sup ∫ δ 1 ( R σ , s ( r ) − R s ( r ) ) r − 2 s − 1 d r ≤ δ − 2 s u p s ∈ K Re s − 1 s ∈ K δ ≤ r ≤ 1 sup ∣ R σ , s ( r ) − R s ( r )∣ .
By (235) , the right-hand side tends to zero as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . Thus it is smaller than ϵ / 2 \epsilon/2 ϵ /2 for σ \sigma σ sufficiently close to 1 / 2 1/2 1/2 . Combining the estimates on ( 0 , δ ) (0,\delta) ( 0 , δ ) and [ δ , 1 ] [\delta,1] [ δ , 1 ] , we obtain
sup s ∈ K ∣ ∫ 0 1 R σ , s ( r ) r − 2 s − 1 d r − ∫ 0 1 R s ( r ) r − 2 s − 1 d r ∣ ⟶ 0 , \sup_{s\in K}\left\lvert\int_{0}^{1}R_{\sigma,s}(r)r^{-2s-1}\,\mathrm{d}r-\int_{0}^{1}R_s(r)r^{-2s-1}\,\mathrm{d}r\right\rvert\longrightarrow0, s ∈ K sup ∫ 0 1 R σ , s ( r ) r − 2 s − 1 d r − ∫ 0 1 R s ( r ) r − 2 s − 1 d r ⟶ 0 ,
which proves the desired uniform convergence.
The contribution near the Fourier origin is controlled by the local Fourier derivative estimates and the resulting uniform Taylor remainder, while the contribution from large Fourier variables is controlled by the global decay of H σ , s H_{\sigma,s} H σ , s . The finite correction terms are handled by the local convergence of the derivatives at the origin. The following lemma makes these three steps precise.
Lemma 4.20 (Convergence of the regularized Fourier–Riesz functional). Let
s ∈ { s ∈ C : Re s > 0 } ∖ Z > 0 . s\in\{s\in\mathbb{C}:\operatorname{Re}s>0\}\setminus\mathbb{Z}_{>0}. s ∈ { s ∈ C : Re s > 0 } ∖ Z > 0 . Let H σ , s H_{\sigma,s} H σ , s and H s H_s H s be defined by (13) and (244) , respectively. Then, for every M ∈ Z ≥ 0 M\in\mathbb{Z}_{\geq0} M ∈ Z ≥ 0 satisfying
Re s < M + 1 , \operatorname{Re}s<M+1, Re s < M + 1 , one has
R M , s [ H σ , s ] ⟶ R M , s [ H s ] , as σ ↓ 1 2 , (236) \mathcal{R}_{M,s}[H_{\sigma,s}]\longrightarrow\mathcal{R}_{M,s}[H_s],\qquad\text{as }\sigma\downarrow\frac{1}{2}, \tag*{(236)} R M , s [ H σ , s ] ⟶ R M , s [ H s ] , as σ ↓ 2 1 , (236) where R M , s [ ⋅ ] \mathcal{R}_{M,s}[\cdot] R M , s [ ⋅ ] is defined by (21) . Moreover, if K K K is a compact subset of { s ∈ C : Re s > 0 } ∖ Z > 0 \{s\in\mathbb{C}:\operatorname{Re}s>0\}\setminus\mathbb{Z}_{>0} { s ∈ C : Re s > 0 } ∖ Z > 0 and M ∈ Z ≥ 0 M\in\mathbb{Z}_{\geq0} M ∈ Z ≥ 0 satisfies
sup s ∈ K Re s < M + 1 , \sup_{s\in K}\operatorname{Re}s<M+1, s ∈ K sup Re s < M + 1 , then
sup s ∈ K ∣ R M , s [ H σ , s ] − R M , s [ H s ] ∣ ⟶ 0 , as σ ↓ 1 2 . \sup_{s\in K}\lvert\mathcal{R}_{M,s}[H_{\sigma,s}]-\mathcal{R}_{M,s}[H_s]\rvert\longrightarrow0,\qquad\text{as }\sigma\downarrow\frac{1}{2}. s ∈ K sup ∣ R M , s [ H σ , s ] − R M , s [ H s ]∣ ⟶ 0 , as σ ↓ 2 1 . Proof. Let
D : = { s ∈ C : Re s > 0 } ∖ Z > 0 . D:=\{s\in\mathbb{C}:\operatorname{Re}s>0\}\setminus\mathbb{Z}_{>0}. D := { s ∈ C : Re s > 0 } ∖ Z > 0 . We first prove the locally uniform statement. Let K K K be a compact subset of D D D , and let M ∈ Z ≥ 0 M\in\mathbb{Z}_{\geq0} M ∈ Z ≥ 0 satisfy
sup s ∈ K Re s < M + 1. \sup_{s\in K}\operatorname{Re}s<M+1. s ∈ K sup Re s < M + 1. We shall prove that
sup s ∈ K ∣ R M , s [ H σ , s ] − R M , s [ H s ] ∣ ⟶ 0 as σ ↓ 1 2 . \sup_{s\in K}\lvert\mathcal{R}_{M,s}[H_{\sigma,s}]-\mathcal{R}_{M,s}[H_s]\rvert\longrightarrow0\qquad\text{as }\sigma\downarrow\frac{1}{2}. s ∈ K sup ∣ R M , s [ H σ , s ] − R M , s [ H s ]∣ ⟶ 0 as σ ↓ 2 1 .
By the definition of R M , s \mathcal{R}_{M,s} R M , s , we have
R M , s [ H σ , s ] − R M , s [ H s ] = ∫ 0 1 [ R σ , s ( r ) − R s ( r ) ] r − 2 s − 1 d r + ∫ 1 ∞ A ( H σ , s − H s ) ( r ) r − 2 s − 1 d r + ∑ j = 0 M Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) 4 j ( j ! ) 2 ( 2 j − 2 s ) , (237) \mathcal{R}_{M,s}[H_{\sigma,s}]-\mathcal{R}_{M,s}[H_s]
=
\int_0^1 [R_{\sigma,s}(r)-R_s(r)]r^{-2s-1}\,dr+\int_1^\infty\mathcal{A}(H_{\sigma,s}-H_s)(r)r^{-2s-1}\,dr
+\sum_{j=0}^{M}\frac{\Delta^j H_{\sigma,s}(0)-\Delta^j H_s(0)}{4^j(j!)^2(2j-2s)}, \tag*{(237)} R M , s [ H σ , s ] − R M , s [ H s ] = ∫ 0 1 [ R σ , s ( r ) − R s ( r )] r − 2 s − 1 d r + ∫ 1 ∞ A ( H σ , s − H s ) ( r ) r − 2 s − 1 d r + j = 0 ∑ M 4 j ( j ! ) 2 ( 2 j − 2 s ) Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) , (237)
where R σ , s ( r ) R_{\sigma,s}(r) R σ , s ( r ) and R s ( r ) R_s(r) R s ( r ) are defined by (232) and (233) , respectively. We treat the three terms on the right-hand side of (237) separately. First, Lemma 4.19 gives
sup s ∈ K ∣ ∫ 0 1 [ R σ , s ( r ) − R s ( r ) ] r − 2 s − 1 d r ∣ ⟶ 0 (238) \sup_{s\in K}\left|\int_0^1 [R_{\sigma,s}(r)-R_s(r)]r^{-2s-1}\,dr\right|\longrightarrow0 \tag*{(238)} s ∈ K sup ∫ 0 1 [ R σ , s ( r ) − R s ( r )] r − 2 s − 1 d r ⟶ 0 (238)
as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . We next consider the finite sum. Since K K K is compact and
K ∩ { 0 , 1 , … , M } = ∅ , K\cap\{0,1,\ldots,M\}=\varnothing, K ∩ { 0 , 1 , … , M } = ∅ ,
we have
d K , M : = min s ∈ K 0 ≤ j ≤ M ∣ j − s ∣ > 0. d_{K,M}:=\min_{\substack{s\in K\\0\leq j\leq M}}|j-s|>0. d K , M := s ∈ K 0 ≤ j ≤ M min ∣ j − s ∣ > 0.
By Lemma 4.10, for every 0 ≤ j ≤ M 0\leq j\leq M 0 ≤ j ≤ M ,
Δ j H σ , s ( 0 ) ⟶ Δ j H s ( 0 ) \Delta^j H_{\sigma,s}(0)\longrightarrow\Delta^j H_s(0) Δ j H σ , s ( 0 ) ⟶ Δ j H s ( 0 )
uniformly for s ∈ K s\in K s ∈ K . Therefore
sup s ∈ K ∣ ∑ j = 0 M Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) 2 4 j ( j ! ) 2 ( j − s ) ∣ ≤ ∑ j = 0 M 1 2 4 j ( j ! ) 2 d K , M sup s ∈ K ∣ Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) ∣ ⟶ 0. (239) \sup_{s\in K}\left|\sum_{j=0}^{M}\frac{\Delta^j H_{\sigma,s}(0)-\Delta^j H_s(0)}{2\,4^j(j!)^2(j-s)}\right|
\leq\sum_{j=0}^{M}\frac{1}{2\,4^j(j!)^2d_{K,M}}\sup_{s\in K}|\Delta^j H_{\sigma,s}(0)-\Delta^j H_s(0)|\longrightarrow0. \tag*{(239)} s ∈ K sup j = 0 ∑ M 2 4 j ( j ! ) 2 ( j − s ) Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) ≤ j = 0 ∑ M 2 4 j ( j ! ) 2 d K , M 1 s ∈ K sup ∣ Δ j H σ , s ( 0 ) − Δ j H s ( 0 ) ∣ ⟶ 0. (239)
It remains to treat the integral over r ≥ 1 r\geq1 r ≥ 1 . By Proposition 1.7, after decreasing σ 0 > 1 / 2 \sigma_0>1/2 σ 0 > 1/2 if necessary, for every N > 0 N>0 N > 0 there exists C K , N > 0 C_{K,N}>0 C K , N > 0 such that
∣ H σ , s ( λ , u ) ∣ ≤ C K , N ( 1 + λ 2 + u 2 ) − N |H_{\sigma,s}(\lambda,u)|\leq C_{K,N}(1+\lambda^2+u^2)^{-N} ∣ H σ , s ( λ , u ) ∣ ≤ C K , N ( 1 + λ 2 + u 2 ) − N
for 1 2 < σ ≤ σ 0 \frac{1}{2}<\sigma\leq\sigma_0 2 1 < σ ≤ σ 0 , s ∈ K s\in K s ∈ K , ( λ , u ) ∈ R 2 (\lambda,u)\in\mathbb{R}^2 ( λ , u ) ∈ R 2 .
The explicit formula
H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) / 4 H_s(\lambda,u)=a_s e^{-is\lambda-(\lambda^2+u^2)/4} H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) /4
implies the same type of rapid-decay bound for H s H_s H s , uniformly for s ∈ K s\in K s ∈ K . Consequently, combining this with the definition of A H \mathcal{A}H A H in (201) ,
sup 1 / 2 < σ ≤ σ 0 s ∈ K ∣ A ( H σ , s − H s ) ( r ) ∣ ≤ C K , N ( 1 + r 2 ) − N , r ≥ 1. (240) \sup_{\substack{1/2<\sigma\leq\sigma_0\\s\in K}}|\mathcal{A}(H_{\sigma,s}-H_s)(r)|\leq C_{K,N}(1+r^2)^{-N},\qquad r\geq1. \tag*{(240)} 1/2 < σ ≤ σ 0 s ∈ K sup ∣ A ( H σ , s − H s ) ( r ) ∣ ≤ C K , N ( 1 + r 2 ) − N , r ≥ 1. (240)
Fix R > 1 R>1 R > 1 . It follows that
sup s ∈ K ∣ ∫ 1 ∞ A ( H σ , s − H s ) ( r ) r − 2 s − 1 d r ∣ ≤ ∫ 1 R sup s ∈ K ∣ A ( H σ , s − H s ) ( r ) ∣ r − 2 inf s ∈ K Re s − 1 d r + C K , N ∫ R ∞ ( 1 + r 2 ) − N r − 2 inf s ∈ K Re s − 1 d r . (241) \sup_{s\in K}\left|\int_{1}^{\infty}\mathcal{A}(H_{\sigma,s}-H_s)(r)r^{-2s-1}\,dr\right|
\leq\int_{1}^{R}\sup_{s\in K}|\mathcal{A}(H_{\sigma,s}-H_s)(r)|r^{-2\inf_{s\in K}\operatorname{Re}s-1}\,dr
+C_{K,N}\int_{R}^{\infty}(1+r^2)^{-N}r^{-2\inf_{s\in K}\operatorname{Re}s-1}\,dr. \tag*{(241)} s ∈ K sup ∫ 1 ∞ A ( H σ , s − H s ) ( r ) r − 2 s − 1 d r ≤ ∫ 1 R s ∈ K sup ∣ A ( H σ , s − H s ) ( r ) ∣ r − 2 i n f s ∈ K Re s − 1 d r + C K , N ∫ R ∞ ( 1 + r 2 ) − N r − 2 i n f s ∈ K Re s − 1 d r . (241)
For fixed R R R , Lemma 4.10, with α = β = 0 \alpha=\beta=0 α = β = 0 , gives
H σ , s ( λ , u ) ⟶ H s ( λ , u ) H_{\sigma,s}(\lambda,u)\longrightarrow H_s(\lambda,u) H σ , s ( λ , u ) ⟶ H s ( λ , u )
uniformly for
s ∈ K , λ 2 + u 2 ≤ R 2 . s\in K,\qquad\lambda^2+u^2\leq R^2. s ∈ K , λ 2 + u 2 ≤ R 2 .
Hence
sup s ∈ K 1 ≤ r ≤ R ∣ A ( H σ , s − H s ) ( r ) ∣ ⟶ 0. \sup_{\substack{s\in K\\1\leq r\leq R}}|\mathcal{A}(H_{\sigma,s}-H_s)(r)|\longrightarrow0. s ∈ K 1 ≤ r ≤ R sup ∣ A ( H σ , s − H s ) ( r ) ∣ ⟶ 0.
Thus the first integral on the right-hand side of (241) tends to zero as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . On the other hand, the second integral in (241) tends to zero as R → ∞ R\to\infty R → ∞ , uniformly in σ \sigma σ and s ∈ K s\in K s ∈ K . Therefore,
sup s ∈ K ∣ ∫ 1 ∞ A ( H σ , s − H s ) ( r ) r − 2 s − 1 d r ∣ ⟶ 0 (242) \sup_{s\in K}\left|\int_{1}^{\infty}\mathcal{A}(H_{\sigma,s}-H_s)(r)r^{-2s-1}\,dr\right|\longrightarrow0 \tag*{(242)} s ∈ K sup ∫ 1 ∞ A ( H σ , s − H s ) ( r ) r − 2 s − 1 d r ⟶ 0 (242)
as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 . Combining (238) , (239) , and (242) in (237) , we obtain
sup s ∈ K ∣ R M , s [ H σ , s ] − R M , s [ H s ] ∣ ⟶ 0 as σ ↓ 1 2 . \sup_{s\in K}|\mathcal{R}_{M,s}[H_{\sigma,s}]-\mathcal{R}_{M,s}[H_s]|\longrightarrow0
\qquad\text{as }\sigma\downarrow\frac{1}{2}. s ∈ K sup ∣ R M , s [ H σ , s ] − R M , s [ H s ] ∣ ⟶ 0 as σ ↓ 2 1 .
This proves the stated local uniform convergence. The pointwise convergence follows immediately from the compact-uniform convergence established above. This completes the proof.
We have now shown that the regularized Fourier–Riesz functional associated with H σ , s H_{\sigma,s} H σ , s converges, as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 , to the corresponding functional associated with the limiting Fourier transform
H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) / 4 . H_s(\lambda,u)=a_s e^{-is\lambda-(\lambda^2+u^2)/4}. H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) /4 .
It therefore remains to evaluate
R M , s [ H s ] \mathcal{R}_{M,s}[H_s] R M , s [ H s ]
explicitly. Since H s H_s H s is the Fourier transform of a shifted complex Gaussian density, the same Fourier–Riesz argument as above reduces this evaluation to an explicit Gaussian integral. The result is given in the following lemma.
Evaluation of the limiting Fourier–Riesz functional
Lemma 4.21. Let
s ∈ C ∖ Z > 0 , Re s > 0 , s\in\mathbb{C}\setminus\mathbb{Z}_{>0},\qquad\operatorname{Re}s>0, s ∈ C ∖ Z > 0 , Re s > 0 ,
and let a s a_s a s be the arithmetic factor defined in (3) . Then, for any M ∈ Z ≥ 0 M \in\mathbb{Z}_{\ge0} M ∈ Z ≥ 0 satisfying
Re s < M + 1 , \operatorname{Re} s < M + 1, Re s < M + 1 ,
one has
2 2 s + 1 Γ ( s + 1 ) Γ ( − s ) R M , s [ H s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (243) 2^{2s+1}\frac{\Gamma(s+1)}{\Gamma(-s)}\mathcal{R}_{M,s}[H_s]=a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2), \tag*{(243)} 2 2 s + 1 Γ ( − s ) Γ ( s + 1 ) R M , s [ H s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) , (243)
where
H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) / 4 . (244) H_s(\lambda,u)=a_s e^{-is\lambda-(\lambda^2+u^2)/4}. \tag*{(244)} H s ( λ , u ) = a s e − i s λ − ( λ 2 + u 2 ) /4 . (244)
Proof. Fix s ∈ C ∖ Z > 0 s \in\mathbb{C} \setminus\mathbb{Z}_{>0} s ∈ C ∖ Z > 0 , Re s > 0 \operatorname{Re}s>0 Re s > 0 , and let M ∈ Z ≥ 0 M \in\mathbb{Z}_{\ge0} M ∈ Z ≥ 0 satisfy
Re s < M + 1. \operatorname{Re}s<M+1. Re s < M + 1.
Then s ∈ Ω M s \in\Omega_M s ∈ Ω M . For t > 0 t>0 t > 0 , set
f s , t ( x ) : = ∣ x ∣ 2 s e − t ∣ x ∣ 2 . f_{s,t}(x):=|x|^{2s}e^{-t|x|^2}. f s , t ( x ) := ∣ x ∣ 2 s e − t ∣ x ∣ 2 .
By the definition of M t ( s ) \mathcal{M}_t(s) M t ( s ) in (182) , together with (191) , we have
M t ( s ) = 1 ( 2 π ) 2 ⟨ f ^ s , t , H s ⟩ , (245) \mathcal{M}_t(s)=\frac{1}{(2\pi)^2}\left\langle\widehat{f}_{s,t},H_s\right\rangle, \tag*{(245)} M t ( s ) = ( 2 π ) 2 1 ⟨ f s , t , H s ⟩ , (245)
where H s ( λ , u ) H_s(\lambda,u) H s ( λ , u ) is defined in (244) . In particular, H s ∈ S ( R 2 ) H_s\in\mathcal{S}(\mathbb{R}^2) H s ∈ S ( R 2 ) . By (196) ,
f s , t ⟶ ∣ x ∣ 2 s in S ′ ( R 2 ) f_{s,t}\longrightarrow|x|^{2s}\quad\text{in }\mathcal{S}'(\mathbb{R}^2) f s , t ⟶ ∣ x ∣ 2 s in S ′ ( R 2 )
as t ↓ 0 t\downarrow0 t ↓ 0 . Since the Fourier transform is a continuous automorphism of S ′ ( R 2 ) \mathcal{S}'(\mathbb{R}^2) S ′ ( R 2 ) , it follows that
f ^ s , t ⟶ ∣ x ∣ 2 s ^ in S ′ ( R 2 ) . \widehat{f}_{s,t}\longrightarrow\widehat{|x|^{2s}}\quad\text{in }\mathcal{S}'(\mathbb{R}^2). f s , t ⟶ ∣ x ∣ 2 s in S ′ ( R 2 ) .
Since H s ∈ S ( R 2 ) H_s\in\mathcal{S}(\mathbb{R}^2) H s ∈ S ( R 2 ) , passing to the limit in (245) gives
lim t ↓ 0 M t ( s ) = 1 ( 2 π ) 2 ⟨ ∣ x ∣ 2 s ^ , H s ⟩ . (246) \lim_{t\downarrow0}\mathcal{M}_t(s)=\frac{1}{(2\pi)^2}\left\langle\widehat{|x|^{2s}},H_s\right\rangle. \tag*{(246)} t ↓ 0 lim M t ( s ) = ( 2 π ) 2 1 ⟨ ∣ x ∣ 2 s , H s ⟩ . (246)
By (22) , with w = s w=s w = s ,
∣ x ∣ 2 s ^ ( ξ ) = 2 2 s + 2 π Γ ( s + 1 ) Γ ( − s ) Reg ∣ ξ ∣ − 2 s − 2 . \widehat{|x|^{2s}}(\xi)=2^{2s+2}\pi\frac{\Gamma(s+1)}{\Gamma(-s)}\operatorname{Reg}|\xi|^{-2s-2}. ∣ x ∣ 2 s ( ξ ) = 2 2 s + 2 π Γ ( − s ) Γ ( s + 1 ) Reg ∣ ξ ∣ − 2 s − 2 .
Hence (246) becomes
lim t ↓ 0 M t ( s ) = 2 2 s + 2 π ( 2 π ) 2 Γ ( s + 1 ) Γ ( − s ) ⟨ Reg ∣ ξ ∣ − 2 s − 2 , H s ⟩ . \lim_{t\downarrow0}\mathcal{M}_t(s)=\frac{2^{2s+2}\pi}{(2\pi)^2}\frac{\Gamma(s+1)}{\Gamma(-s)}
\left\langle\operatorname{Reg}|\xi|^{-2s-2},H_s\right\rangle. t ↓ 0 lim M t ( s ) = ( 2 π ) 2 2 2 s + 2 π Γ ( − s ) Γ ( s + 1 ) ⟨ Reg ∣ ξ ∣ − 2 s − 2 , H s ⟩ .
Since H s ∈ S ( R 2 ) H_s\in\mathcal{S}(\mathbb{R}^2) H s ∈ S ( R 2 ) , the usual S ′ − S \mathcal{S}'-\mathcal{S} S ′ − S pairing agrees with the extended pairing of Proposition 4.15. Since s ∈ Ω M s\in\Omega_M s ∈ Ω M , that proposition therefore gives
⟨ Reg ∣ ξ ∣ − 2 s − 2 , H s ⟩ = 2 π R M , s [ H s ] . \left\langle\operatorname{Reg}|\xi|^{-2s-2},H_s\right\rangle=2\pi\mathcal{R}_{M,s}[H_s]. ⟨ Reg ∣ ξ ∣ − 2 s − 2 , H s ⟩ = 2 π R M , s [ H s ] .
Consequently,
lim t ↓ 0 M t ( s ) = 2 2 s + 1 Γ ( s + 1 ) Γ ( − s ) R M , s [ H s ] . (247) \lim_{t\downarrow0}\mathcal{M}_t(s)=2^{2s+1}\frac{\Gamma(s+1)}{\Gamma(-s)}\mathcal{R}_{M,s}[H_s]. \tag*{(247)} t ↓ 0 lim M t ( s ) = 2 2 s + 1 Γ ( − s ) Γ ( s + 1 ) R M , s [ H s ] . (247)
On the other hand, Lemma 4.14 gives
M t ( s ) = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 1 F 1 ( s + 1 ; 1 ; s 2 1 + t ) . \mathcal{M}_t(s)=a_s e^{-s^2}\Gamma(s+1)(1+t)^{-s-1}{}_1F_1\left(s+1;1;\frac{s^2}{1+t}\right). M t ( s ) = a s e − s 2 Γ ( s + 1 ) ( 1 + t ) − s − 1 1 F 1 ( s + 1 ; 1 ; 1 + t s 2 ) .
Letting t ↓ 0 t\downarrow0 t ↓ 0 , we obtain
lim t ↓ 0 M t ( s ) = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . (248) \lim_{t\downarrow0}\mathcal{M}_t(s)=a_s e^{-s^2}\Gamma(s+1){}_1F_1\left(s+1;1;s^2\right). \tag*{(248)} t ↓ 0 lim M t ( s ) = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . (248)
Comparing (247) and (248) yields (243) .
Proof of Theorem 1.2
We now combine the preceding results to prove Theorem 1.2 .
Proof of Theorem 1.2. We first consider
s ∈ C ∖ Z > 0 , Re s > 0. s\in\mathbb{C}\setminus\mathbb{Z}_{>0},\qquad\operatorname{Re}s>0. s ∈ C ∖ Z > 0 , Re s > 0.
Choose M ∈ Z ≥ 0 M\in\mathbb{Z}_{\geq0} M ∈ Z ≥ 0 such that
Re s < M + 1. \operatorname{Re}s<M+1. Re s < M + 1.
Recall that
T σ = ( 2 σ − 1 ) L ′ ( σ , X ) T_\sigma=(2\sigma-1)L'(\sigma,X) T σ = ( 2 σ − 1 ) L ′ ( σ , X )
and
ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) . \zeta'(\sigma,X)=\zeta(\sigma,X)L'(\sigma,X). ζ ′ ( σ , X ) = ζ ( σ , X ) L ′ ( σ , X ) .
Consequently,
( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] . (2\sigma-1)^{s^2+2s}\mathbb{E}\left[|\zeta'(\sigma,X)|^{2s}\right]=(2\sigma-1)^{s^2}\mathbb{E}\left[|\zeta(\sigma,X)|^{2s}|T_\sigma|^{2s}\right]. ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = ( 2 σ − 1 ) s 2 E [ ∣ ζ ( σ , X ) ∣ 2 s ∣ T σ ∣ 2 s ] .
By Lemma 1.8 , for all σ > 1 / 2 \sigma>1/2 σ > 1/2 sufficiently close to 1 / 2 1/2 1/2 ,
( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = 2 2 s + 1 Γ ( s + 1 ) Γ ( − s ) R M , s [ H σ , s ] . (249) (2\sigma-1)^{s^2+2s}\mathbb{E}\left[|\zeta'(\sigma,X)|^{2s}\right]=2^{2s+1}\frac{\Gamma(s+1)}{\Gamma(-s)}\mathcal{R}_{M,s}[H_{\sigma,s}]. \tag*{(249)} ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = 2 2 s + 1 Γ ( − s ) Γ ( s + 1 ) R M , s [ H σ , s ] . (249)
By Lemma 4.20 ,
R M , s [ H σ , s ] ⟶ R M , s [ H s ] as σ ↓ 1 2 . \mathcal{R}_{M,s}[H_{\sigma,s}]\longrightarrow\mathcal{R}_{M,s}[H_s]\qquad\text{as }\sigma\downarrow\frac{1}{2}. R M , s [ H σ , s ] ⟶ R M , s [ H s ] as σ ↓ 2 1 .
Combining this convergence with Lemma 4.21 , we obtain
lim σ ↓ 1 / 2 ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . (250) \lim_{\sigma\downarrow1/2}(2\sigma-1)^{s^2+2s}\mathbb{E}\left[|\zeta'(\sigma,X)|^{2s}\right]=a_s e^{-s^2}\Gamma(s+1){}_1F_1\left(s+1;1;s^2\right). \tag*{(250)} σ ↓ 1/2 lim ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . (250)
Moreover, the convergence is locally uniform for
s ∈ D : = { s ∈ C : Re s > 0 } ∖ Z > 0 , s\in D:=\left\{s\in\mathbb{C}:\operatorname{Re}s>0\right\}\setminus\mathbb{Z}_{>0}, s ∈ D := { s ∈ C : Re s > 0 } ∖ Z > 0 ,
by the locally uniform convergence established in Lemma 4.20 . It remains to extend the result to s ∈ Z > 0 s\in\mathbb{Z}_{>0} s ∈ Z > 0 . For σ > 1 / 2 \sigma>1/2 σ > 1/2 , define
F σ ( s ) : = ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] , F_\sigma(s):=(2\sigma-1)^{s^2+2s}\mathbb{E}\left[|\zeta'(\sigma,X)|^{2s}\right], F σ ( s ) := ( 2 σ − 1 ) s 2 + 2 s E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] ,
and let
F ( s ) : = a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) . F(s) := a_s e^{-s^2}\Gamma(s+1){}_1F_1(s+1;1;s^2). F ( s ) := a s e − s 2 Γ ( s + 1 ) 1 F 1 ( s + 1 ; 1 ; s 2 ) .
By (118), for every compact set K ⊂ { s : Re s > 0 } K \subset\{s : \operatorname{Re} s > 0\} K ⊂ { s : Re s > 0 } , there exists A > 0 A > 0 A > 0 such that
sup s ∈ K ∣ ζ ′ ( σ , X ) ∣ 2 s ≤ 1 + ∣ ζ ′ ( σ , X ) ∣ 2 A . \sup_{s\in K} |\zeta'( \sigma,X)|^{2s} \le1 + |\zeta'(\sigma,X)|^{2A}. s ∈ K sup ∣ ζ ′ ( σ , X ) ∣ 2 s ≤ 1 + ∣ ζ ′ ( σ , X ) ∣ 2 A .
whose expectation is finite. Since ∣ ζ ′ ( σ , X ) ∣ 2 s |\zeta'(\sigma,X)|^{2s} ∣ ζ ′ ( σ , X ) ∣ 2 s is holomorphic in s s s for Re s > 0 \operatorname{Re} s > 0 Re s > 0 , dominated convergence and Morera’s theorem imply that E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] \mathbb{E}[|\zeta'(\sigma,X)|^{2s}] E [ ∣ ζ ′ ( σ , X ) ∣ 2 s ] is holomorphic in this half-plane. Consequently, F σ F_\sigma F σ is holomorphic there. Consequently, F σ ( s ) F_\sigma(s) F σ ( s ) is holomorphic in this half-plane. The function F F F is also holomorphic in the half-plane Re s > 0 \operatorname{Re} s > 0 Re s > 0 . Indeed, the Euler product defining a s a_s a s converges locally uniformly in s s s , since each Euler factor equals 1 + O K ( p − 2 ) 1 + O_K(p^{-2}) 1 + O K ( p − 2 ) uniformly on every compact set K ⊂ C K \subset\mathbb{C} K ⊂ C . Moreover, Γ ( s + 1 ) \Gamma(s+1) Γ ( s + 1 ) is holomorphic for Re s > 0 \operatorname{Re} s > 0 Re s > 0 , while e − s 2 e^{-s^2} e − s 2 and 1 F 1 ( s + 1 ; 1 ; s 2 ) {}_1F_1(s+1;1;s^2) 1 F 1 ( s + 1 ; 1 ; s 2 ) are entire functions of s s s . Fix m ∈ Z > 0 m \in\mathbb{Z}_{>0} m ∈ Z > 0 and choose 0 < r < 1 / 2 0 < r < 1/2 0 < r < 1/2 . The closed disc
B ( m , r ) ‾ = { s ∈ C : ∣ s − m ∣ ≤ r } \overline{B(m,r)} = \{s \in\mathbb{C} : |s-m| \le r\} B ( m , r ) = { s ∈ C : ∣ s − m ∣ ≤ r }
lies in the half-plane Re s > 0 \operatorname{Re} s > 0 Re s > 0 , and its boundary contains no positive integers. Since F σ − F F_\sigma-F F σ − F is holomorphic on a neighbourhood of B ( m , r ) ‾ \overline{B(m,r)} B ( m , r ) , the maximum modulus principle gives
sup ∣ s − m ∣ ≤ r ∣ F σ ( s ) − F ( s ) ∣ ≤ sup ∣ s − m ∣ = r ∣ F σ ( s ) − F ( s ) ∣ . \sup_{|s-m|\le r} |F_\sigma(s)-F(s)| \le\sup_{|s-m|=r} |F_\sigma(s)-F(s)|. ∣ s − m ∣ ≤ r sup ∣ F σ ( s ) − F ( s ) ∣ ≤ ∣ s − m ∣ = r sup ∣ F σ ( s ) − F ( s ) ∣.
The right-hand side tends to zero as σ ↓ 1 / 2 \sigma\downarrow1/2 σ ↓ 1/2 , by the locally uniform convergence already established on D D D . Hence
sup ∣ s − m ∣ ≤ r ∣ F σ ( s ) − F ( s ) ∣ ⟶ 0. \sup_{|s-m|\le r} |F_\sigma(s)-F(s)| \longrightarrow0. ∣ s − m ∣ ≤ r sup ∣ F σ ( s ) − F ( s ) ∣ ⟶ 0.
In particular,
lim σ ↓ 1 / 2 F σ ( m ) = F ( m ) . \lim_{\sigma\downarrow1/2} F_\sigma(m) = F(m). σ ↓ 1/2 lim F σ ( m ) = F ( m ) .
Since m ∈ Z > 0 m \in\mathbb{Z}_{>0} m ∈ Z > 0 was arbitrary, the asymptotic formula holds for every s ∈ C s \in\mathbb{C} s ∈ C with Re s > 0 \operatorname{Re} s > 0 Re s > 0 . The locally uniform convergence on D D D , together with the uniform convergence on the discs around positive integers, also establishes locally uniform convergence throughout the half-plane Re s > 0 \operatorname{Re} s > 0 Re s > 0 . This completes the proof.