Abstract
A conjecture connected with quantum physics led N. Katz to discover some amazing mixed character sum identities over a field of q elements, where q is a power of a prime \(p >3\). His proof required deep algebro-geometric techniques, and he expressed interest in finding a more straightforward direct proof. The first author recently gave such a proof of his identities when \(q \equiv 1 \pmod 4\), and this paper provides such a proof for the remaining case \(q \equiv 3 \pmod 4\). Our proofs are valid for all characteristics \(p>2\). Along the way we prove some elegant new character sum identities.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Background
Let \(\mathbb {F}_q\) be a field of q elements, where q is a power of an odd prime p. Throughout this paper, A, B, C, D, \(\chi \), \(\lambda \), \(\nu \), \(\mu \), \(\varepsilon \), \(\phi \) denote complex multiplicative characters on \(\mathbb {F}_q^*\), extended to map 0 to 0. Here \(\varepsilon \) and \(\phi \) always denote the trivial and quadratic characters, respectively. Define \(\delta (A)\) to be 1 or 0 according as A is trivial or not, and let \(\delta (j,k)\) denote the Kronecker delta for \(j,k \in \mathbb {F}_q\).
Much of this paper deals with the extension field \(\mathbb {F}_{q^2}\) of \(\mathbb {F}_q\). Let \(M_4\) denote a fixed quartic character on \(\mathbb {F}_{q^2}\) and let \(M_8\) denote a fixed octic character on \(\mathbb {F}_{q^2}\) such that \(M_8^2 = M_4\).
Define the additive character \(\psi \) on \(\mathbb {F}_q\) by
The corresponding additive character on \(\mathbb {F}_{q^2}\) will be denoted by \(\psi _2\).
Recall the definitions of the Gauss and Jacobi sums over \(\mathbb {F}_q\):
These sums have the familiar properties
and for nontrivial A,
Gauss and Jacobi sums are related by [5, p. 59]
and
The Hasse–Davenport product relation [5, p. 351] yields
As in [12, p. 82], define the hypergeometric \({}_2F_1\) function over \(\mathbb {F}_q\) by
For \(j, k \in \mathbb {F}_q\) and \(a \in \mathbb {F}_q^*\), Katz [13, p. 224] defined the mixed exponential sums
Note that
Katz proved an equidistribution conjecture of Wootters [13, p. 226], [1] connected with quantum physics by constructing explicit character sums V(j) [13, pp. 226–229] for which the identities
hold for all \(j,k \in \mathbb {F}_q\). (The q-dimensional vector \((V(j))_{j \in \mathbb {F}_q} \) is a minimum uncertainty state, as described by Sussman and Wootters [17].) Katz’s proof [13, Theorem 10.2] of the identities (1.5) required the characteristic p to exceed 3, in order to guarantee that various sheaves of ranks 2, 3, and 4 have geometric and arithmetic monodromy groups which are SL(2), SO(3), and SO(4), respectively.
As Katz indicated in [13, p. 223], his proof of (1.5) is quite complex, invoking the theory of Kloosterman sheaves and their rigidity properties, as well as results of Deligne [6] and Beilinson, Bernstein, Deligne [4]. Katz [13, p. 223] wrote, “It would be interesting to find direct proofs of these identities.”
The goal of this paper is to respond to Katz’s challenge by giving a direct proof of (1.5) (a “character sum proof” not involving algebraic geometry). This has the benefit of making the demonstration of his useful identities accessible to a wider audience of mathematicians and physicists. Since a direct proof for \(q \equiv 1 \pmod 4\) has been given in [8], we will assume from here on that \(q \equiv 3 \pmod 4\).
A big advantage of our proof is that it works for all odd characteristics p, including \(p=3\). As a bonus, we obtain some elegant new double character sum evaluations in (5.11)–(5.14).
Our method of proof is to show (in Sect. 6) that the double Mellin transforms of both sides of (1.5) are equal. The Mellin transforms of the left and right sides of (1.5) are given in Theorems 3.1 and 5.1, respectively. A key feature of our proof is a formula (Theorem 4.1) relating a norm-restricted Jacobi sum over \(\mathbb {F}_{q^2}\) to a hypergeometric \({}_2F_1\) character sum over \(\mathbb {F}_q\). Theorem 4.1 will be applied to prove Theorem 5.3, an identity for a weighted sum of hypergeometric \({}_2F_1\) character sums. Theorem 5.3 is crucial for our proof of (1.5) in Sect. 6.
Hypergeometric character sums over finite fields have had a variety of applications in number theory. For some recent examples, see [2, 3, 7, 11, 14,15,16].
Since \(q \equiv 3 \pmod 4\), we have \(\phi (-1)=-1\), and every element \(z \in \mathbb {F}_{q^2}\) has the form
where i is a fixed primitive fourth root of unity in \(\mathbb {F}_{q^2}\). Write \(\overline{z}= x - iy\) and note that \(\overline{z}= z^q\). The restriction of \(M_8\) to \(\mathbb {F}_q\) equals \(\varepsilon \) or \(\phi \) according as q is congruent to 7 or 3 mod 8. In particular,
For a character C on \(\mathbb {F}_q\), we let CN denote the character on \(\mathbb {F}_{q^2}\) obtained by composing C with the norm map N on \(\mathbb {F}_{q^2}\) defined by
Given a character B on \(\mathbb {F}_q\), BCN is to be interpreted as the character (BC)N, i.e., BNCN.
For the same a as in (1.3), define
where the choice of square root is fixed. Katz defined the sums V(j) to be the following norm-restricted Gauss sums:
Note that
2 Mellin transform of the sums V(j)
This section begins with some results related to Gauss sums over \(\mathbb {F}_{q^2}\) that will be used in this paper. We use the notation \(G_2\) and \(J_2\) for Gauss and Jacobi sums over \(\mathbb {F}_{q^2}\), in order to distinguish them from the Gauss and Jacobi sums G and J over \(\mathbb {F}_q\). For any character \(\beta \) on \(\mathbb {F}_{q^2}\), we have
for example, for a character C on \(\mathbb {F}_q\), \(G_2(CN M_8)\) equals \(G_2(CN \overline{M}_8)\) or \(G_2(CN M_8^3)\) according as q is congruent to 7 or 3 mod 8. The Hasse-Davenport theorem on lifted Gauss sums [5, Theorem 11.5.2] gives
From [9, (4.10)],
For any character \(\beta \) on \(\mathbb {F}_{q^2}\), define
It is easily seen that
where \(E_2(\beta )\) is the Eisenstein sum
Let \(\beta ^*\) denote the restriction of \(\beta \) to \(\mathbb {F}_q\). Applying [5, Theorem 12.1.1] with q in place of p, we can express \(E_2(\beta )\) in terms of Gauss sums when \(\beta \) is nontrivial, as follows:
For any character \(\chi \) on \(\mathbb {F}_q\), define the Mellin transform
In the case that \(\chi \) is odd, we may write \(\chi = \phi \lambda ^2\) for some character \(\lambda \) on \(\mathbb {F}_q\). In that case, we may assume without loss of generality that \(\lambda \) is even, otherwise replace \(\lambda \) by \(\phi \lambda \). In summary, when \(\chi \) is odd,
for some character \(\nu \) on \(\mathbb {F}_q\).
The next theorem gives an evaluation of \(S(\chi )\) in terms of Gauss sums.
Theorem 2.1
If \(\chi \) is even, then \(S(\chi )=0\). If \(\chi \) is odd [so that (2.9) holds], then
Proof
If \(\chi \) is even, then \(S(\chi )\) vanishes by (1.8) and (2.8). Now assume that \(\chi \) is odd, so that \(\chi = \phi \nu ^4\). Then
Replace z by \(z/j^2\) to get
The sum on j on the right equals \(q-1\) when \(C \in \{\nu , \phi \nu \}\) and it equals 0 otherwise. Since \(\phi N = M_8^4\), the result now follows from the definition of \(G_2\). \(\square \)
3 Double Mellin transform of V(j)V(k)
For characters \(\chi _1, \chi _2\), define the double Mellin transform
As in (2.9), when \(\chi _1\) and \(\chi _2\) are both odd,
for some characters \(\nu _1\), \(\nu _2\) on \(\mathbb {F}_q\). In this case, write
The following theorem evaluates S in terms of Gauss and Jacobi sums.
Theorem 3.1
If \(\chi _1\) or \(\chi _2\) is even, then \(S=0\). If \(\chi _1\) and \(\chi _2\) are both odd [so that (3.2) and (3.3) hold], then
Proof
By (3.1), \(S=S(\chi _1)S(\chi _2)\). By Theorem 2.1, \(S=0\) when \(\chi _1\) or \(\chi _2\) is even. Thus assume that \(\chi _1\) and \(\chi _2\) are both odd. Then Theorem 2.1 yields
A straightforward computation with the aid of (2.1) shows that (3.5) is equivalent to (3.4). The computation is facilitated by noting that \(M_8(-a)\) equals 1 or \(-\phi (a)\) according as q is congruent to 7 or 3 mod 8, so that the bracketed expression for \(i=0\) in (3.4) is to be compared to that for \(i=1\) in (3.5) when q is congruent to 3 mod 8. \(\square \)
4 Identity for a norm-restricted Jacobi sum in terms of a \({}_2F_1\)
Let D be a character on \(\mathbb {F}_q\). Define the norm-restricted Jacobi sums
The next theorem provides a formula expressing R(D, j) in terms of a \({}_2F_1\) hypergeometric character sum.
Theorem 4.1
For \(j = \pm 1\),
For all other \(j \in \mathbb {F}_q^*\),
Proof
Replace z in (4.1) by \(-zj^2\). By (1.6), we obtain
Each z in the sum must be a square, since N(z) is a square in \(\mathbb {F}_q\). Thus
Writing \(z=x+iy\), we have
where it is understood that the sum is over all \(x,y \in \mathbb {F}_q\) for which \(x^2 + y^2 =1\). Thus, since \(M_4(\pm i)=M_8(-1)=\phi (2)\),
where
Replacing y by yx, we have
Since \(\overline{M}_4= \phi N M_4\), this yields
First consider the case where \(j = \pm 1\). By (4.4) and (4.5),
The restriction of \(DN M_4\) to \(\mathbb {F}_q\) is \(D^2\). Thus by (2.7),
if \(D^2\) is nontrivial, and
if \(D^2\) is trivial. By (2.3),
Consequently,
for every D, which completes the proof when \(j = \pm 1\). Thus assume for the remainder of this proof that \(j^2 \ne 1\).
By (4.5), Q(D, j) equals
By the “binomial theorem” [12, (2.10)], the rightmost factor above equals
where the “binomial coefficient” over \(\mathbb {F}_q\) is defined by [12, p. 80]
Replacing \(\chi \) with \(\overline{\chi }\) and observing that [12, p. 80]
we see that
where
Comparing (4.6) and (4.7), we see that
where the last equality follows from the Hasse-Davenport relation (1.1). Consequently,
Replace \(\chi \) by \(D \chi \) to get
By [12, (2.15)] with \(A=D \chi \), \(B=\chi \), and \(C=D^2 \phi \chi ^2\),
since by [12, (2.6)],
Thus
By [12, Theorem 4.16] with \(A=D\), \(B=\phi D^2\), and \(x = -(j+1)^2/(j-1)^2\), we have
Multiply by \(-\phi (j)q\overline{D}^4(j-1)\) to get
Thus by (4.8),
Combining (4.4) and (4.9), we arrive at the desired result (4.3). \(\square \)
5 Double Mellin transform of P(j, k)
For characters \(\chi _1, \chi _2\), define the double Mellin transform
Note that \(T(\chi _1, \chi _2)\) is symmetric in \(\chi _1\), \(\chi _2\).
The following theorem evaluates T.
Theorem 5.1
If \(\chi _1\) or \(\chi _2\) is even, then \(T=0\). If \(\chi _1\) and \(\chi _2\) are both odd [so that (3.2) and (3.3) hold], then
where for a character D on \(\mathbb {F}_q\) and \(j \in \mathbb {F}_q^*\), we define
Proof
By (1.4), \(P(j, k)=-P(j,-k)\), so \(T=0\) if \(\chi _1\) or \(\chi _2\) is even. Thus assume that (3.2) and (3.3) hold. Replacing j by jk in (1.3), we obtain
Since \(\delta (\mu ^4)=\delta (\mu ^2)\), this becomes
There is no contribution from the 1 in the rightmost factor \((1+\phi (k))\); to see this, replace k and x by their negatives. Therefore,
It follows that
After replacing x by ax and employing (5.3), the desired result (5.2) readily follows. \(\square \)
We proceed to analyze h(D, j).
Lemma 5.2
We have
and for \(j \ne \pm 1\) and nontrivial D, we have
Finally, if \(j \ne \pm 1\) and D is trivial, then \(h(D,j)=0\).
Proof
The evaluation in (5.4) follows directly from the definition of h(D, j) in (5.3). The evaluation in (5.5) is the same as that in [8, (5.21)], the proof of which is valid for q congruent to either 1 or 3 mod 4. Finally, let \(j \ne \pm 1\). Then since
replacement of x by \(1-x(2j^2+2)/(j+1)^2\) shows that \(h(\varepsilon ,j)= -1 + 1 = 0\). \(\square \)
Theorem 5.3
For a character D on \(\mathbb {F}_q\), define
Then \(W(\varepsilon )=2\), and for nontrivial D,
Proof
It follows directly from Lemma 5.2 that \(W(\varepsilon )=2\). Let D be nontrivial. By Lemma 5.2,
This simplifies to
For brevity, let Y(D) denote this sum on j. It remains to prove that
Since the fourth powers in \(\mathbb {F}_q\) are precisely the squares, it follows from definition (4.1) that
Thus
The sum on j on the right vanishes unless \(\chi \in \{\nu _1, \nu _1\phi \}\), and so we obtain the desired result (5.10). \(\square \)
As interesting consequences of Theorem 5.3, we record the elegant double character sum evaluations (5.11)–(5.14) below.
Theorem 5.4
For any character \(\nu \) on \(\mathbb {F}_q\),
Proof
This follows by putting \(D=\phi \) in (5.7). \(\square \)
Theorem 5.5
When \(q \equiv 7 \pmod 8\), we have
When \(q \equiv 3 \pmod 8\), we have
where |u|, |v| is the unique pair of positive integers with \(p \not \mid u\) for which \(q^2=u^2 + 2v^2\), and where the sign of u is determined by the congruence \(u \equiv -1 \pmod 8\). In particular, when \(q=p \equiv 3 \pmod 8\), we have
where \(p = a_8^2 + 2b_8^2\).
Proof
By (5.11) with \(\nu = \varepsilon \), the sum in (5.12) equals
First suppose that \(q \equiv 7 \pmod 8\). Then
by [5, Theorem 11.6.1]. Thus each Jacobi sum above equals q, which proves (5.12).
Now suppose that \(q \equiv 3 \pmod 8\). An application of (2.1) shows that \(J_2(M_8^5, \phi N)\) is the complex conjugate of \(J_2(M_8, \phi N)\), so that the sum in (5.13) equals \(2 \mathfrak {R}J_2(M_8, \phi N)\).
First consider the case where q is prime, i.e., \(q=p\). Then \(G_2(\phi N) = p\) and by [5, Theorems 12.1.1 and 12.7.1(b)],
where \(\pi =a_8 + i b_8 \sqrt{2}\) is a prime in \(\mathbb {Q}(i\sqrt{2})\) of norm \(p=\pi \overline{\pi }=a_8^2 + 2b_8^2\). Note that \(\pi ^2 = u_1 + iv_1\sqrt{2}\), where
so that
In the general case where say \(q = p^t\), the Hasse–Davenport lifting theorem [5, Theorem 11.5.2] yields
for integers u, v such that \(q^2 = u^2 + 2v^2\). Since \(u_1 \equiv -1 \pmod 8\), it is easily seen using the binomial theorem that \(u \equiv -1 \pmod 8\). If \(p=\pi \overline{\pi }\) divided u, then p would divide v, so that the prime \(\overline{\pi }\) would divide \(\pi ^{2t}\), which is impossible. Thus \(p \not \mid u\). For an elementary proof of the uniqueness of |u|, |v|, see [5, Lemma 3.0.1]. \(\square \)
Remark
The sum in Theorem 5.5, namely
can be evaluated when \(q \equiv 1 \pmod 4\) as well. We have \(Z=0\) when \(q \equiv 5 \pmod 8\), which can be seen by applying [8, Lemma 5.1] with \(\phi \) in place of D, and then replacing j by jI, where I is a primitive fourth root of unity in \(\mathbb {F}_q\). More work is needed to evaluate Z in the remaining case where \(q \equiv 1 \pmod 8\). In this case Z is equal to the sum \(R_2\) in [8, (5.44)] with \(\nu _1= B_8\) and \(A_4 = B_8^2\) for an octic character \(B_8\) on \(\mathbb {F}_q\). The proof of [5, Theorem 3.3.1] shows that
Using this equality to evaluate the sum \(R_2\), we have
We will use (5.16) to show that
and
where c and d are the unique pair of integers up to sign for which
First suppose that \(p \equiv 7 \pmod 8\). Then \(q = p^{2t}\) for some \(t \ge 1\). If \(t=1\), then \(J(B_8, \phi )=p\) by [5, Theorem 11.6.1]. For general t, the Hasse-Davenport lifting theorem thus yields \(J(B_8, \phi )=(-1)^{t-1}p^t\), so that \(J(B_8, \phi )^2=q\). Thus \(Z=4q\) by (5.16).
Now suppose that \(p \equiv 5 \pmod 8\). Since \(G(B_8)=G(B_8^p)=G(B_8^5)\) by [5, Theorem 1.1.4(d)], \(J(B_8, \phi ) = G(\phi )\). Thus \(J(B_8, \phi )^2=q\), so again \(Z=4q\). This completes the proof of (5.17).
Next suppose that \(p \equiv 3 \pmod 8\). Then \(q = p^{2t}\) for some \(t \ge 1\). Since \(-2\) is a square \(\pmod p\), we have the prime splitting \(p = \pi \overline{\pi }\) in \(\mathbb {Q}(i\sqrt{2})\). Assume first that \(t=1\). Then
We cannot have \(J(B_8, \phi )= \pm p\), otherwise the prime ideal factorization of \(J(B_8, \phi )\) in [5, Theorems 11.2.3, 11.2.9] would yield the contradiction that p ramifies in the cyclotomic field \(\mathbb {Q}(\exp (2\pi i/8))\). In view of (5.20) and unique factorization in \(\mathbb {Q}(i\sqrt{2})\), we may suppose without loss of generality that \(J(B_8, \phi ) = \pi ^2\) when \(t=1\). For general t,
for some integers c and d such \(c^2 + 2d^2 = q\). Note that p cannot divide c, for otherwise p also divides d (since \(c^2 + 2d^2 = q\)), so that p divides \(\pi ^{2t}\), yielding the contradiction that the prime \(\overline{\pi }\) divides \(\pi \). By (5.21),
so that by (5.16), \(Z = 4c^2\).
Finally, suppose that \(p \equiv 1 \pmod 8\), and write \(q=p^t\) for some \(t \ge 1\). Since \(-2\) is a square \(\pmod p\), we have the prime splitting \(p = \pi \overline{\pi }\) in \(\mathbb {Q}(i\sqrt{2})\). If \(t=1\), then without loss of generality, \(J(B_8, \phi ) = \pi \). For general t,
for some integers c and d such \(c^2 + 2d^2 = q\). Arguing as in the case \(p \equiv 3 \pmod 8\), we again obtain \(Z=4c^2\) for c as in (5.19). This completes the proof of (5.18).
6 Proof of Katz’s identities (1.5)
When \(jk=0\), both sides of (1.5) vanish, by (1.4) and (1.8). We thus assume that \(jk \ne 0\). It suffices to show that the Mellin transforms of the left and right sides of (1.5) are the same for all characters, for then (1.5) follows by taking inverse Mellin transforms. Thus it remains to show that \(S=T\), where S and T are given in Theorems 3.1 and 5.1, respectively. These theorems show that S and T both vanish when \(\chi _1\) or \(\chi _2\) is even, so we may assume that (3.2) and (3.3) hold. For brevity, write \(D = \mu \phi ^i\), where \(i \in \{0,1\}\). Then the equality \(S=T\) is equivalent to
Noting that \(G_2(\overline{D}N)= -G(\overline{D})^2\) by (2.2), and using the formula for W(D) in Theorem 5.3, we easily see that (6.1) holds. This completes the proof that \(S=T\).
References
Amburg, I., Sharma, R., Sussman, D.M., Wootters, W.K.: States that “look the same” with respect to every basis in a mutually unbiased set. J. Math. Phys. 55(12), 122206 (2014)
Barman, R., Kalita, G.: Hyperelliptic curves over \(\mathbb{F}_q\) and Gaussian hypergeometric series. J. Ramanujan Math. Soc. 30, 331–348 (2015)
Barman, R., Saikia, N.: On the polynomials \(x^d+ax^i+b\) and \(x^d+ax^{d-i}+b\) over \(\mathbb{F}_q\) and Gaussian hypergeometric series. Ramanujan J. 35, 427–441 (2014)
Beilinson, A., Bernstein, J., Deligne, P.: Faisceaux pervers, Analyse et topologie sur les éspaces singuliers, I (Conférence de Luminy, 1981), Astérisque 100. Soc. Math. France, Paris (1982)
Berndt, B.C., Evans, R.J., Williams, K.S.: Gauss and Jacobi sums. Wiley-Interscience, New York (1998)
Deligne, P.: La conjecture de Weil II. Publ. Math. IHES 52, 313–428 (1981)
El-Guindy, A., Ono, K.: Hasse invariants for the Clausen elliptic curves. Ramanujan J. 31, 3–13 (2013)
Evans, R.J.: Some mixed character sum identities of Katz, arXiv:1607.05889 (2016)
Evans, R.J., Greene, J.R.: Evaluations of hypergeometric functions over finite fields. Hiroshima Math. J. 39, 217–235 (2009)
Evans, R.J., Greene, J.R.: A quadratic hypergeometric \({}_2F_1\) transformation over finite fields, Proc. Amer. Math. Soc. 145, 1071–1076 (2017)
Fuselier, J., Long, L., Ramakrishna, R., Swisher, H., Tu, F.-T.: Hypergeometric functions over finite fields, arXiv:1510.02575
Greene, J.R.: Hypergeometric functions over finite fields. Trans. Amer. Math. Soc. 301, 77–101 (1987)
Katz, N.M.: Rigid local systems and a question of Wootters. Commun. Number Theory Phys. 6(2), 223–278 (2012)
Lin, Y.-H., Tu, F.-T.: Twisted Kloosterman sums. J. Number Theory 147, 666–690 (2015)
McCarthy, D., Papanikolas, M.: A finite field hypergeometric function associated to eigenvalues of a Siegel eigenform. Int. J. Number Theory 11, 2431–2450 (2015)
Salerno, A.: Counting points over finite fields and hypergeometric functions. Funct. Approx. Comment. Math. 49, 137–157 (2013)
Sussman, D.M., Wootters, W.K.: Discrete phase space and minimum-uncertainty states. In: Hirota, O., Shapiro, J.H., Sasaki, M. (eds.) Proceedings of the Eighth International Conference on Quantum Communication, Measurement and Computing, NICT Press (2007). arXiv:0704.1277
Open Access
This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0), which permits use, duplication, adaptation, distribution, and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Evans, R., Greene, J. Some mixed character sum identities of Katz II. Res. number theory 3, 8 (2017). https://doi.org/10.1007/s40993-016-0071-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s40993-016-0071-5
Keywords
- Hypergeometric \({}_2F_1\) character sums over finite fields
- Gauss and Jacobi sums
- Norm-restricted Gauss and Jacobi sums
- Eisenstein sums
- Hasse–Davenport theorems
- Quantum physics