--- title: 'Real-Variable Kloosterman Sums: Theory & Applications' url: https://www.emergentmind.com/topics/real-variable-kloosterman-sums type: topic --- # Real-Variable Kloosterman Sums: Theory & Applications Real-variable Kloosterman sums arise in several closely connected senses within modern analytic number theory. In the classical prime-modulus setting, the family \(S(a,1;p)\) is real-valued and admits the normalization \[ \frac{S(a,1;p)}{\sqrt p}=2\cos \theta_p(a),\qquad 0\le \theta_p(a)\le \pi, \] so questions about moments, signs, and limiting distributions become questions about a bounded real variable indexed by \(a\bmod p\) [1108.0746]. In a more literal sense, recent work defines Kloosterman-type sums \(K(r,n,c,\nu)\) in which one arithmetic index is replaced by a real parameter \(r\), and uses them to construct radial Fourier interpolation formulas in dimensions \(3\) and \(4\) [2510.04873]. Between these two poles lies a broad literature on short sums, sparse averages over primes, half-integral-weight analogues, and sign-change problems. ## 1. Classical real-valued Kloosterman sums For an odd prime \(p\) and integers \(a,b\) with \((ab,p)=1\), the classical Kloosterman sum is \[ S(a,b;p)=\sum_{x \bmod p}^{*} e\!\left(\frac{ax+b\overline{x}}{p}\right),\qquad e(z)=e^{2\pi i z}, \] where \(\sum^*\) runs over a reduced residue system modulo \(p\) and \(\overline{x}\) is the inverse of \(x\bmod p\). Weil’s bound gives \[ |S(a,b;p)|\le 2p^{1/2}, \] and the one-parameter family \(S(a,1;p)\) is real-valued with size \(\ll \sqrt p\) [1108.0746]. The standard normalization \[ \frac{S(a,1;p)}{\sqrt p}=2\cos \theta_p(a) \] packages the real-valued behavior into angles \(\theta_p(a)\in[0,\pi]\). For integers \(k\ge1\), the associated moments are \[ V_k(p):=\sum_{a\bmod p}^{*} S(a,1;p)^k, \qquad \widetilde V_k(p):=\sum_{a\bmod p}^{*}|S(a,1;p)|^k. \] For small \(k\), explicit identities are known: \[ V_1(p)=1,\quad V_2(p)=p^2-p-1,\quad V_3(p)=\Big(\tfrac{2}{p}\Big)p^2+2p+1,\quad V_4(p)=2p^3-3p^2-3p-1. \] This real normalization is the basic template for much of the subject. It makes Kloosterman sums amenable to moment methods, orthogonal polynomial expansions, and probabilistic language, while retaining the arithmetic rigidity coming from the modulus \(p\). ## 2. Moments, equidistribution, and sign distribution For fixed \(k\in\mathbb N\) and sufficiently large prime \(p\), the even moments satisfy \[ V_{2k}(p)=\frac{1}{k+1}\binom{2k}{k}p^{k+1}+O_k\!\bigl(p^{k+1/2}\bigr), \] the odd moments satisfy \[ V_{2k+1}(p)\ll_k p^{k+1}, \] and the odd absolute moments satisfy \[ \widetilde V_{2k+1}(p) = \frac{4^{k+1}}{\pi}\frac{(2k+1)!!}{(2k+3)!!}\binom{2k}{k}p^{k+3/2} +O_k\!\bigl(p^{k+1}\bigr) \] [1108.0746]. The odd raw moments are thus smaller by a factor \(p^{1/2}\) than the trivial estimate coming from Weil’s bound, while the odd absolute moments retain the full size \(p^{k+3/2}\). The proofs combine the parametrization \(S(a,1;p)=2\sqrt p\cos\theta_p(a)\), expansions of powers and absolute powers in Chebyshev polynomials \(U_k\), and Katz’s estimate \[ \left|\sum_{a\bmod p}^{*} U_k(\cos\theta_p(a))\right| \le 5(k+1)p^{1/2}. \] The relevant orthogonality is with respect to the measure \[ \frac{2}{\pi}\sqrt{1-x^2}\,dx\qquad (x\in[-1,1]), \] equivalently \[ d\mu(\theta)=\frac{2}{\pi}\sin^2\theta\,d\theta\qquad (\theta\in[0,\pi]). \] The moment asymptotics therefore match the moments of \(2\cos\Theta\) for \(\Theta\sim d\mu\), and the normalized values \(S(a,1;p)/\sqrt p\) behave asymptotically like a compactly supported symmetric random variable on \([-2,2]\). The same method yields quantitative sign information. If \[ N_+(p):=\#\{a\bmod p:S(a,1;p)>0\},\qquad N_-(p):=\#\{a\bmod p:S(a,1;p)<0\}, \] then for sufficiently large \(p\), \[ N_+(p)\ge \frac49 p+O(p^{1/2}),\qquad N_-(p)\ge \frac49 p+O(p^{1/2}). \] In particular, both signs occur for a positive proportion of residue classes, and there are infinitely many \(a\bmod p\) with \(S(a,1;p)>0\) and infinitely many with \(S(a,1;p)<0\) as \(p\to\infty\). ## 3. Short sums, sparse averages, and oscillatory analysis A second major theme studies Kloosterman sums after averaging over thin arithmetic sets. For fixed \(n\ge2\), normalized classical Kloosterman sums modulo \(p^n\), \[ \mathsf{Kl}_{p^n}(a)=\frac{1}{p^{n/2}}S(a,1;p^n), \] satisfy \(|\mathsf{Kl}_{p^n}(a)|\le2\). If \(I_{p^n}\subset(\mathbb Z/p^n\mathbb Z)^\times\) satisfies the separation condition \(x\neq y\Rightarrow p\nmid x-y\) and \[ \log |I_{p^n}|=o(\log p), \] then the translated short sums \[ \frac{1}{\sqrt{|I_{p^n}|}}\sum_{t\in I_{p^n}}\mathsf{Kl}_{p^n}(t+x) \] converge in law, as \(p\to\infty\), to a real-valued standard Gaussian random variable [1907.01834]. This places short sums modulo prime powers within the general probabilistic framework of trace functions, while retaining an explicitly real-valued limit law. A complementary line of work studies Kloosterman phases over primes. For \[ S_q(a;x)=\sum_{\substack{p\sim x\\ (p,q)=1}} e\!\left(\frac{a\overline p}{q}\right), \] Irving proved average bounds over moduli \(q\sim Q\). In the balanced range \(x\asymp Q\), \[ \sum_{q\sim Q}\max_{(a,q)=1}|S_q(a;x)|\ll_\epsilon Q^{19/10+\epsilon}, \qquad \sum_{q\sim Q}|S_q(a;x)|\ll_\epsilon Q^{15/8+\epsilon}, \] improving earlier exponents \(23/12\) and \(95/48\) in the corresponding settings [1301.6372]. The method uses Vaughan’s identity, bilinear decompositions, high moments, and congruence counting for reciprocal variables. For normalized Kloosterman sums over primes in an arithmetic progression, \(\mathrm{Kl}_2(p;q)\), where \(q\) is prime and \(1\le v\le q^{1/100}\), one has \[ \sum_{\substack{p\le X\\ p\equiv u \!\!\!\!\pmod v\\ p\ \mathrm{prime}}}\mathrm{Kl}_2(p;q) \ll_\varepsilon X^{1-1/192+\varepsilon} \qquad (1\le X\le q) \] [1801.05880]. Here the real-variable content is explicit: additive detection of the congruence condition produces oscillatory weights, tempered Voronoi summation converts bilinear forms in \(\mathrm{Kl}_2\) into Fourier transforms of two-variable weights, and stationary phase yields main oscillatory factors of the form \(e(-mn/(cq^2))\). The resulting cancellation occurs not only inside individual oscillatory integrals but across families of such integrals. ## 4. Half-integral weight, plus-space, and twisted spectral variants Real-variable Kloosterman phenomena persist in half-integral weight. For an even lattice \(L\) of odd rank and Weil representation \(\rho_L\), the generalized Kloosterman sums \[ S_{\alpha,\beta}(m,n,c) \] are defined by summing matrix coefficients \(\rho_{\alpha\beta}(\tilde\gamma)\) against the phase \(e_{2\Delta c}(ma+nd)\). If \(m=m_0v^2\) with \((-1)^{(g-1)/2}m_0\) a fundamental discriminant and \((v,\Delta)=1\), then \[ S_{\alpha,\beta}(m,n,c)\ll_L 2^{\omega(c)}\,\tau((v,c))\,(m_0n,c)^{1/2}\,c^{1/2} \] [2309.08528]. The proof proceeds through an identity expressing a divisor sum of generalized Kloosterman sums as a sparse exponential sum in an auxiliary variable \(\ell\), which generalizes Kohnen’s identity. For the half-integral-weight Kohnen plus space on \(\Gamma_0(4)\), the plus-space Kloosterman sums \(S_k^+(m,n,c)\), \(k=\pm\frac12\), are real-valued and satisfy \[ S_k^+(m,n,c)=S_{-k}^+(-m,-n,c). \] If \[ (-1)^\lambda m=v^2d',\qquad (-1)^\lambda n=w^2d, \] with \(d,d'\) fundamental discriminants, then \[ \sum_{4\mid c\le x}\frac{S_k^+(m,n,c)}{c} \ll \left(x^{1/6}+(dd')^{2/9}(vw)^{1/3}\right)(mnx)^\varepsilon \] [1801.08174]. This is obtained from a Kuznetsov formula restricted to the plus space, together with Young’s hybrid subconvexity estimates for twisted modular \(L\)-functions. A further extension twists Kloosterman sums by modular symbols. For \(m,n\neq0\), \[ Z^*_{m,n}(s)=\sum_{N\mid c>0}\frac{S^*(m,n;c)}{c^{2s}} \] is a Kloosterman zeta function formed with a modular symbol. It continues meromorphically to \(\Re(s)>1/2\), and if \[ \beta=\overline{\lim}_{\substack{c\to\infty\\ N\mid c}} \frac{\log|S^*(m,n;c)|}{\log c}, \] then \[ \sum_{\substack{00\), the real-variable Kloosterman sum is \[ K(r,n,c,\nu_\Theta^{2k}) = \sum_{\substack{-c0}\frac{K(r^2,n,c,\nu_\Theta^4)}{c} J_1\!\Bigl(\frac{2\pi |r|\sqrt n}{c}\Bigr), \] together with an explicit divisor-sum correction. In dimension \(3\), \[ B_{3,n}(r) = \frac{(\tfrac18)}{|r|} \sum_{2\mid c>0} \frac{K(r^2,n,c,\nu_\Theta^3)}{\sqrt c}\, \sin\!\Bigl(\frac{2\pi |r|\sqrt n}{c}\Bigr), \] together with explicit terms involving the Hurwitz class number \(H(n)\) and the three-squares function \(r_3(n)\). The real-variable representation sharpens the \(n\)-aspect of the interpolation basis. For fixed \(r>0\), the earlier general bounds \[ a_{4,n}(r)=O_r(n^{1+o(1)}),\qquad a_{3,n}(r)=O_r(n^{3/4+o(1)}) \] are improved to \[ B_{4,n}(r),\widetilde B_{4,n}(r)=O_r(n^{3/4+\varepsilon}), \qquad B_{3,n}(r),\widetilde B_{3,n}(r)=O_r(n^{1/2+\varepsilon}). \] The analytic input is an interpolation identity reducing \(K(r,n,c,\nu_\Theta^{2k})\) to finite linear combinations of classical half-integral-weight Kloosterman sums, followed by spectral bounds for sums over the modulus. ## 6. Distribution, sign changes, and non-real generalizations The classical real-valued theory does not exhaust the subject. For a subgroup \(\Lambda\le (\mathbb Z/m\mathbb Z)^\times\), the generalized Kloosterman sum \[ K(a,b,m,\Lambda)=\sum_{u\in\Lambda} e\!\left(\frac{au+bu^{-1}}{m}\right) \] is typically complex-valued; unlike classical Kloosterman sums, generalized Kloosterman sums display image sets in the complex plane with highly structured geometry [1505.00018]. When \(d\) is an odd prime and \(p\equiv1\pmod d\), the values \(K(a,b,q,d)\) lie in the filled \(d\)-cusped hypocycloid \(H_d\) and become dense in \(H_d\) as \(q=p^\alpha\to\infty\). For \(d=9\), the image is contained in the threefold Minkowski sum \(H_3+H_3+H_3\), and powers of \(2\) in \((p-1)\) produce the “boxes in boxes” phenomena described via Salié sums and projections onto real parts. A different analogue replaces the additive character by a multiplicative character. For an even nontrivial \(\chi\bmod p\), \[ K(\chi)=\frac{1}{\sqrt p}\sideset{}{^*}\sum_{a\bmod p}\chi(a+\bar a) \] satisfies \(|K(\chi)|\le2\), and the twisted variable \(\varepsilon_p\overline{\chi}(2)K(\chi)\) is real-valued in \([-2,2]\). Its moments satisfy \[ M_\kappa(n,p) = \frac{\Gamma(2\kappa+1)}{2\,\Gamma(\kappa+1)^2}\,\Delta_n\,p +O\bigl(\kappa\,4^\kappa\,p^{1/2}\log p\bigr), \] leading by the method of moments to an arcsine law on \([-2,2]\) rather than the Sato–Tate-type law of classical normalized Kloosterman sums [2105.15051]. Sign-change problems remain central. A major open problem is to show that \(\Kl(1;p)\) changes sign infinitely often over primes \(p\), but recent work proves that on square-free moduli \(q\in(X,2X]\) having at most five prime factors, each sign of the normalized Kloosterman sum \(\Kl(1;q)\) occurs \(\gg X/\log X\) times [2607.10862]. Earlier work showed that the existence of exceptional real zeroes of Dirichlet \(L\)-functions would imply cancellation in \[ \sum_{p\le x}\Kl(1,p) \] and sign changes of \(\Kl(1,n)\) for \(n\) with exactly two prime factors [1802.10278]. These results place the real-variable study of Kloosterman sums at the intersection of equidistribution, sieve theory, trace formulas, and the arithmetic of exceptional characters. Source: https://www.emergentmind.com/topics/real-variable-kloosterman-sums