Ekedahl-Oort Types and Newton Polygons of Abelian Covers of 𝐏 1 \mathbf{P}^{1} Branched at Three Points Darren Schmidt Abstract In this paper, we study the Newton polygons and Ekedahl-Oort types of reductions of abelian covers of the projective line branched at three points modulo a prime. We study the natural density of primes where these covers give supersingular and superspecial curves and show they appear much more often than expected. We also show that unlikely Newton polygons and Ekedahl-Oort types in the moduli space of curves appear frequently. Finally, we prove a theorem that provides evidence of Oort’s Conjecture about Newton polygons in certain cases and gives new constructions of supersingular curves. † † footnotetext: We thank Jeremy Booher for helpful comments. 1 Introduction Let p be a prime and k be an algebraically closed field of characteristic p . If X is a smooth projective curve of genus g over k , then the Jacobian of X , Jac (X) , is an abelian variety over k . Abelian varieties in positive characteristic can be studied through their invariants, such as their Newton polygon [ 16 , Section 4.2] or their Ekedahl-Oort type [ 16 , Section 4.4] . If g is 1,2, or 3 then we know that every symmetric Newton polygon of height 2g with endpoints (0,0) and (2g,g) and integer breakpoints occurs for the Jacobian of a smooth projective curve of genus g . In general, we do not know exactly which symmetric Newton polygons or Ekedahl-Oort types occur for Jacobians of smooth projective curves. In this paper, we study which Newton polygons and Ekedahl-Oort types occur for Jac( X ) when X is an abelian cover of the projective line \mathbf{P}^{1} branched at exactly three points. We have that X is the quotient of a Fermat curve, and therefore Jac (X) has complex multiplication. This makes the actions of Frobenius and Verschiebung on the cohomology of the curve easier to analyze, which lets us determine the Newton polygon and Ekedahl-Oort type of X . We will see that they are often “unlikely” in a sense we now make precise. Let \mathcal{M}_{g} be the moduli space of smooth projective curves of genus g in characteristic p and \mathcal{A}_{g} be the moduli space of principally polarized abelian varieties of dimension g in characteristic p . We know that if g\geq 2 , \mathcal{M}_{g} has dimension 3g-3 and \mathcal{A}_{g} has dimension g(g+1)/2 [ 15 , Section 5.3] . Definition 1.1 . If \xi is a Newton polygon (or Ekedahl-Oort type) in dimension g , let \mathcal{M}_{g}[\xi] be the moduli space of curves of genus g in characteristic p with Newton polygon (respectively, Ekedahl-Oort type) \xi and let \mathcal{A}_{g}[\xi] be the moduli space of principally polarized abelian varieties of dimension g with Newton polygon (respectively, Ekedeahl-Oort type) \xi . We say that \xi is an unlikely Newton polygon (Ekedahl-Oort Type) if the codimension of \mathcal{A}_{g}[\xi] in \mathcal{A}_{g} exceeds the dimension of \mathcal{M}_{g} . We say \xi is realized in characterstic p if \mathcal{M}_{g}[\xi] is non-empty. If the Torelli locus and the Newton polygon strata (or Ekedahl-Oort strata) were dimensionally transverse, unlikely Newton polygons (or Ekedahl-Oort types) would not occur. This naïve heuristic is known to be incorrect. (See for example the existence of supersingular curves of any genus in characteristic two [ 20 ] , Pieper’s analysis of the hyperelliptic locus in genus three [ 14 ] , and the examples in [ 2 ] .) We provide additional illustrations that this heuristic is far from correct using abelian covers of the projective line \mathbf{P}^{1} branched at exactly three points. Some of our results are computational — they then inspired some theoretical results — see 1.1 and 1.2 . Supersingular and superspecial curves are commonly studied types of curves, and we pay particular attention to them. If E is an elliptic curve over k , we say that E is supersingular if E(k)[p]=0 . We call X supersingular if Jac (X) is isogenous to a product of g supersingular elliptic curves. In a similar vein, we call X superspecial if Jac (X) is isomorphic to a product of g supersingular elliptic curves. We can determine if a curve is supersingular by examining its Newton polygon, and similarly we can deterime if a curve is superspecial by examining its Ekedahl-Oort type. If the genus is large enough, then supersingular and superspecial curves have an unlikely Newton polygon and an unlikely Ekedahl-Oort type, respectively. Remark 1.2 . We note that if \xi is a supersingular Newton polygon of dimension g then \mathcal{A}_{g}[\xi] has dimension \lfloor g^{2}/4\rfloor [ 15 , Section 2.4] . Therefore, if g\geq 9 , then a supersingular curve of genus g , if it exists, has an unlikely Newton polygon. Also, if \xi is a superspecial Ekedahl-Oort type of dimension g , then \mathcal{A}_{g}[\xi] has dimension 0 [ 13 , Page 1] . We then see that if g\geq 4 , then a superspecial curve has an unlikely Ekedahl-Oort type. Supersingular and superspecial curves arising from abelian covers of \mathbf{P}^{1} branched at three points have been studied before. In [ 6 ] , Ekedahl determined for which primes a superspecial curve of this type occurs when g=4,5 . 1.1 Computational Results In [ 2 ] , for a fixed genus g , Booher and Pries determined the natural density of primes p such that a supersingular curve that is an abelian cover of \mathbf{P}^{1} branched at three points and defined over \overline{\mathbf{F}_{p}} occurs for 5\leq g\leq 10 . In particular, they wrote code in SageMath that can compute the Newton polygon when the curve is a cyclic cover. In the case where the curve is a non-cyclic cover, they decomposed the Jacobian of the curve by hand into a product of cyclic cases to determine the Newton polygon. In this paper, we extend their work by creating an algorithm in SageMath that can handle the non-cyclic case. In doing so, for genus up to 50 we are able to utilize SageMath to compute the natural density of primes where a supersingular curve occurs. We also write a program to compute the Ekedahl-Oort type and determine the natural density of primes such that there exist a cover that gives a superspecial curve, which we use to fix an error in [ 6 ] for genus 5 — see Proposition 3.3 . Furthermore, we also use this program to find additional curves with unlikely Newton polygons or Ekedahl-Oort types and compute the natural density of primes where they occur. From our computations, we determined that supersingular curves, superspecial curves, unlikely Newton polygons, and unlikely Ekedahl-Oort types occur much more frequently than expected for such a specific family of curves. Let \delta_{ss}(g) , \delta_{ssp}(g) , and \delta_{nu}(g) be the densities of primes such that there exists an abelian cover of \mathbf{P}^{1} branched at three points of genus g that is supersingular, superspecial, and that has an unlikely Newton polygon, respectively. Then, if 9\leq g\leq 50 , \begin{array}[]{ccc}\delta_{ss}(g)>0.7,&\delta_{ssp}(g)>0.2,&\delta_{nu}(g)>0.875.\end{array} The data we collected led us to make the following conjecture: Conjecture 1.3 . Let g\geq 18 , and let m be the least common multiple of the exponents of all of the Galois groups of abelian covers of \mathbf{P}^{1} branched at three points of genus g . If p is a prime such that p\not\equiv 1 modulo m , then there exist abelian covers X_{1} and X_{2} of \mathbf{P}^{1} branched at three points defined over \overline{\mathbf{F}_{p}} such that X_{1} has an unlikely Newton polygon and X_{2} has an unlikely Ekedahl-Oort type. 1.2 Theoretical Results After noticing a pattern in the results of these computations, we are also able to prove several results about the following conjecture and expectation of Oort: Conjecture 1.4 ( [ 13 , Conjecture 8.5.7] ) . If p is a prime and \xi_{1} and \xi_{2} are Newton polygons that are realized in characteristic p , then \xi_{1}\oplus\xi_{2} is realized in characteristic p . Expectation 1.5 ( [ 13 , Expectation 8.5.4] ) . Given a symmetric Newton polygon \xi of height 2g such that \xi is unlikely and has slopes with denominators (say, greater than 11), then \xi is not realized in characteristic p for all primes p . Theorem 1.6 . Let \ell>3 be a prime and n\geq 1 be coprime to \ell . Define g:=(\ell-1)/2 and suppose \alpha is the number of quadratic residues modulo \ell contained in the intervals (0,\ell/4) , (\ell/3,\ell/2) , and (2\ell/3,3\ell/4) . If p is a prime with order g in \left(\mathbf{Z}/\ell\mathbf{Z}\right)^{\times} such that gcd(p,n\ell)=1 , and the order of p in \left(\mathbf{Z}/n\mathbf{Z}\right)^{\times} is coprime to g , then there exists a cyclic cover \phi:X\to\mathbf{P}^{1} branched at three points defined over \overline{\mathbf{F}_{p}} such that X has genus ng and the only slopes of the the Newton polygon of \textrm{Jac}(X) are \alpha/g and (g-\alpha)/g . Theorem 1.6 is a generalization of [ 10 , Corollary 3.10] , which focuses on the case when g is a Sophie Germain prime. For each Sophie Germain prime, [ 10 , Corollary 3.10] constructs a finite number of supersingular curves; however, Theorem 1.6 gives us infinitely many examples of supersingular curves without having to assume that there are infinitely many Sophie Germain primes. It also gives us examples of Newton polygons that satisfy Conjecture 1.4 , and it also can be used to construct counterexamples to Expectation 1.5 . The following corollary is immediate and gives infinitely many Newton polygons that satisfy Conjecture 1.4 Corollary 1.7 . Let \ell>3 be a prime. If p>2 is a prime with p\neq\ell such that p is congruent to a non-zero quadratic residue modulo \ell , then there is a Newton polygon \xi such that \xi is realized in characteristic p and \xi\oplus\xi is realized in characteristic p . The next two corollaries show how to use Theorem 1.6 to construct supersingular curves. Corollary 1.8 . Let \ell>3 be a prime with \ell\equiv 1 modulo 4 and n\geq 1 be coprime to \ell . Define g:=\frac{\ell-1}{2} . If p is a prime with order g in \left(\mathbf{Z}/\ell\mathbf{Z}\right)^{\times} such that gcd(p,n\ell)=1 , and the order of p in (\mathbf{Z}/n\mathbf{Z})^{\times} is coprime to g , then there exists a supersingular curve of genus ng defined over \overline{\mathbf{F}_{p}} . Corollary 1.9 . Let \ell>3 be a prime and n\geq 1 be coprime to \ell . Define g:=\frac{\ell-1}{2} . If p is a prime with order 2g in \left(\mathbf{Z}/\ell\mathbf{Z}\right)^{\times} such that gcd(p,n\ell)=1 , and the order of p in (\mathbf{Z}/n\mathbf{Z})^{\times} is coprime to g , then there exists a supersingular curve of genus ng defined over \overline{\mathbf{F}_{p}} . In particular, if g>2 is a Sophie Germain prime, then \ell=2g+1 is 3 modulo 4 , so Corollary 1.9 with n=1 gives the same supersingular curves as [ 10 , Corollary 3.10] . We are also able to use Theorem 1.6 to give explicit counterexamples to Expectation 1.5 . Proposition 1.10 . If g>363 is a Sophie Germain prime, then for every n\geq 1 such that gcd(n,2g+1)=1 , there exists a curve of genus ng whose Newton polygon is unlikely and has slopes with denominator g . We note that [ 10 , Corollary 3.10] gives the same Newton polygons as Proposition 1.10 for n=1 , but Proposition 1.10 shows we have infinitely many unlikely Newton polygons with large denominators even though we don’t know if there are infinitely many Sophie Germain primes. Finally, we are able to use Theorem 1.6 to prove the following results about the natural density \delta_{ss}(g) : Proposition 1.11 . We have that \limsup_{g\to\infty}\delta_{ss}(g)>0.9999 . Furthermore, if we assume Dickson’s Conjecture — stated in Conjecture 5.4 — then we can compute the value of \limsup_{g\to\infty}\delta_{ss}(g) . Theorem 1.12 . If Dickson’s Conjecture is true, then \limsup_{g\to\infty}\delta_{ss}(g)=1 . Section 2 of this paper gives a background on abelian covers of \mathbf{P}^{1} branched at three points and explains how to utilize the fact that the Jacobian has complex multiplication to compute Newton polygons and Ekedahl-Oort types. Section 3 gives computational results on the natural density of primes that give supersingular and superspecial curves for each genus, and also provides data on how often unlikely Newton polygons and Ekedahl-Oort types occur. Section 4 gives the proof of Theorem 1.6 and its corollaries by utilizing the Shimura-Taniyama formula to analyze the orbits of a specific cover, and Section 5 utilizes Theorem 1.6 to prove results about natural densities and Newton polygons which has slopes with large denominators. 2 Abelian Covers Let X be a smooth projective connected curve over k , where char (k)=p . Let \phi:X\to\mathbf{P}^{1} be a Galois cover branched at exactly B:=\{0,1,\infty\} . Let g be the genus of X . Assume that G is abelian and that p does not divide the order of G . 2.1 Ramification and Inertia Types If b\in B , let I_{b} be the inertia group above b and define c_{b}=|I_{b}| . As p does not divide the order of G , \phi is tamely ramified; therefore, I_{b} is cyclic. We assume without loss of generality that c_{0}\geq c_{1}\geq c_{\infty} . Definition 2.1 . The ramification type of the cover \phi is the tuple z=[c_{0},c_{1},c_{\infty},s] where s is the index of I_{0} in G . As p does not divide the order of G , \phi is tamely ramified. Therefore, the ramification type determines the genus of X by the Riemann-Hurwitz formula as in [ 2 , Lemma 2.3] . Definition 2.2 . An inertia type associated to a ramification type z=[c_{0},c_{1},c_{\infty},s] is the tuple a=(a_{0},a_{1},a_{\infty}) of elements of G such that 1. |a_{b}|=c_{b} for b\in B 2. a_{0}+a_{1}+a_{\infty}=0 in G 3. a_{0} , a_{1} , and a_{\infty} generate G For a given ramification type z , we can relabel the branch points if their inertia groups have the same order. Furthermore, we can modify the inertia type by applying automorphisms of G . If two inertia types are in the same orbit under these two actions, we say they are equivalent. 2.2 Cyclic Covers Assume that \phi is a cyclic cover of degree m . Then, G\simeq\mathbf{Z}/m\mathbf{Z} . The abelian variety Jac (X) has complex multiplication by \prod_{d}\mathbf{Q}(\zeta_{d}) where the product is taken over all integers d such that 10.7 . Definition 3.2 . Let SSP_{g} be the set of primes p such that there exists a superspecial curve X of genus g such that X is an abelian cover of \mathbf{P}^{1} branched at three points defined over \overline{\mathbf{F}_{p}} . Let \delta_{ssp}(g) denote the natural density of the primes in SSP_{g} . The superspecial primes for g=4,5 were computed in [ 6 ] . We give the value (or an underestimate of the value) of \delta_{ssp}(g) for 5\leq g\leq 50 in Table 1. The natural densities in Definitions 3.1 and 3.2 exist as the set of primes is given by congruence conditions. For a given cover, if e is the exponent of the Galois group, we get a list of congruence classes in \mathbf{Z}/e\mathbf{Z} that categorize which primes give us the type of curve that we’re interested in. To compute the density of those primes, we compute the least common multiple of all the exponents that occur \ell . We then use the Chinese Remainder Theorem to determine which congruence classes modulo \ell that are coprime to \ell satisfy the congruence conditions for any of the covers found. The bottleneck is that as the genus grows larger, there are generally more covers with larger Galois groups. This causes \ell to grow very large very fast, so it becomes infeasible to keep track of all congruence classes modulo \ell . We have denoted which computations for which this happens in the table so that the reader is aware they are only estimates of the true value. From our computation, we noticed an omission in Ekedahl’s list of primes for which there exists a superspecial curve of genus 5 that is an abelian cover of \mathbf{P}^{1} branched at three points. We correct it below. Proposition 3.3 . There exists an abelian cover \phi:X\to\mathbf{P}^{1} of genus 5 branched at three points in characteristic p such that X is superspecial if and only if p\equiv-1 modulo 8,11,12,15,20 , p\equiv 11 modulo 15 , or p\equiv 11 modulo 20 . Proof. The only case not proven in [ 6 , Page 173] is when p\equiv 11 modulo 20 . Consider the cover with ramification type [20,20,2,1] and inertia type (1,9,10) . This cover has genus 5 by [ 2 , Lemma 2.3] . The orbits of the action of multiplication by p on \{1,2,\ldots,19\}\subseteq\mathbf{Z}/20\mathbf{Z} are as follows: \{\{1,11\},\{2\},\{3,13\},\{4\},\{5,15\},\{6\},\{7,17\},\{8\},\{9,19\},\{10\},\{12\},\{14\},\{16\},\{18\}\}. The signature type f(j)=1 if and only if j\in\{1,3,5,7,9\} . Therefore, by Theorem 2.9 F(e_{11})=V(e_{11})=e_{1} , F(e_{13})=V(e_{13})=e_{3} , F(e_{15})=V(e_{15})=e_{5} , F(e_{17})=V(e_{17})=e_{7} , and F(e_{19})=V(e_{19})=e_{9} . We see that H^{1}_{dR}(X) is the direct sum of 5 copies of H^{1}_{dR} of a supersingular elliptic curve. Therefore, Jac (X) is isomorphic to a product of 5 supersingular elliptic curves, and X is superspecial. ∎ 3.2 Unlikely Newton Polygons and Ekedahl-Oort Types We also investigates the prime density for covers that have unlikely Newton polygons and Ekedahl-Oort types. We give some theorems on the dimensions of some relevant moduli spaces first. Theorem 3.4 ( [ 16 , Theorem 6.10] ) . Let \xi be a symmetric Newton polygon of height 2g . The dimension of \mathcal{A}_{g}[\xi] is \#\{(x,y)\in\mathbf{Z}\times\mathbf{Z}|y0.9999 19 3/4 0.7500 697/1296 0.5378 6911/6912 0.9999 20 1021/1024 0.9971 3437/8000 0.4296 127999/128000 >0.9999 21 353/384 0.9193 121151/282240 0.4292 423359/423360 >0.9999 22 61/64 0.9531 7751/17600 0.4404 3839/3840 0.9997 23 7/8 0.8750 3907/16192 0.2413 4047/4048 0.9998 24 32763/32768 0.9998 >102447/229376 0.4466 7741439/7741440 >0.9999 25 17529/20480 0.8559 10083/20480 0.4923 921599/921600 >0.9999 26 497/512 0.9707 7303/19968 0.3657 99839/99840 >0.9999 27 121/128 0.9453 87493/207360 0.4219 933119/933120 >0.9999 28 4061/4096 0.9915 1937/4096 0.4729 290303/290304 >0.9999 29 405/512 0.7910 28335/103936 0.2726 38975/38976 >0.9999 30 >2549/2560 0.9957 >1365281/3686400 0.3704 >15551999/15552000 >0.9999 31 25/32 0.7812 4073/9600 0.4243 345599/345600 >0.9999 32 4095/4096 0.9998 2927/8192 0.3573 8847359/8847360 >0.9999 33 >1973/2048 0.9634 >45389/139392 0.3256 >8363519/8363520 >0.9999 34 1497/2048 0.7310 25565/67584 0.3783 168959/168960 >0.9999 35 439/512 0.8574 80319/224000 0.3586 1007999/1008000 >0.9999 36 >1023/1024 0.9990 >47497/93312 0.5090 >13436927/13436928 >0.9999 37 >183/256 0.7148 >102941/311040 0.3310 >5598719/5598720 >0.9999 38 239/256 0.9336 26683/103680 0.2574 207359/207360 >0.9999 39 2023/2304 0.8780 236209/808704 0.2921 4852223/4852224 >0.9999 40 >81871/81920 0.9994 >4201/10240 0.4103 >6911999/6912000 >0.9999 41 >197/256 0.7695 >42833/131200 0.3265 >1574399/1574400 >0.9999 42 >287/288 0.9965 >67/256 0.2617 >5419007/5419008 >0.9999 43 3/4 0.7500 117/448 0.2612 564479/564480 >0.9999 44 >1021/1024 0.9971 >4923/17600 0.2797 >4646399/4646400 >0.9999 45 >1117/1152 0.9696 >788639/2073600 0.3803 >13996799/13996800 >0.9999 46 >15/16 0.9375 >35659/105600 0.3377 >6071999/6072000 >0.9999 47 >51/64 0.7969 >59071/264960 0.2229 >794879/794880 >0.9999 48 >262141/262144 >0.9999 >565847/2064384 0.2741 >15482879/15482880 >0.9999 49 >25/32 0.7812 >110717/286720 0.3862 >7741439/7741440 >0.9999 50 >63/64 0.9844 >3911/14400 0.2716 >5759999/5760000 >0.9999 Table 1: Supersingular Covers, Superspecial Covers, and Covers with Unlikely Newton Polygons {adjustbox} width= g \delta_{2nu}(g) \approx\delta_{2nu}(g) \delta_{eu}(g) \approx\delta_{eu}(g) \delta_{2eu}(g) \approx\delta_{2eu}(g) 5 0 0.0000 139/160 0.8688 0 0.0000 6 0 0.0000 507/512 0.9902 0 0.0000 7 0 0.0000 7/8 0.8750 0 0.0000 8 0 0.0000 2045/2048 0.9985 0 0.0000 9 0 0.0000 3887/3888 0.9997 0 0.0000 10 0 0.0000 5759/5760 0.9998 69/80 0.8625 11 0 0.0000 879/880 0.9989 41/55 0.7455 12 0 0.0000 34559/34560 >0.9999 2291/2304 0.9944 13 0 0.0000 1727/1728 0.9994 559/768 0.7279 14 0 0.0000 2687/2688 0.9996 2651/2688 0.9862 15 0 0.0000 172799/172800 >0.9999 85861/86400 0.9938 16 0 0.0000 61439/61440 >0.9999 20443/20480 0.9982 17 0 0.0000 575/576 0.9983 89/96 0.9271 18 0 0.0000 13436927/13436928 >0.9999 1679615/1679616 >0.9999 19 0 0.0000 20735/20736 >0.9999 431/432 0.9977 20 0 0.0000 1151999/1152000 >0.9999 63999/64000 >0.9999 21 353/384 0.9193 1693439/1693440 >0.9999 21167/21168 >0.9999 22 61/64 0.9531 316799/316800 >0.9999 6329/6336 0.9989 23 7/8 0.8750 48575/48576 >0.9999 97111/97152 0.9996 24 >229357/229376 0.9999 >15482879/15482880 >0.9999 >4423679/4423680 >0.9999 25 97757/102400 0.9547 16588799/16588800 >0.9999 921599/921600 >0.9999 26 19931/19968 0.9981 10782719/10782720 >0.9999 59903/59904 >0.9999 27 1909/1920 0.9943 1866239/1866240 >0.9999 466559/466560 >0.9999 28 773989/774144 0.9998 1161215/1161216 >0.9999 290303/290304 >0.9999 29 51867/51968 0.9981 467711/467712 >0.9999 38975/38976 >0.9999 30 >2879797/2880000 0.9999 >15551999/15552000 >0.9999 >6911999/6912000 >0.9999 31 28681/28800 0.9959 2764799/2764800 >0.9999 345599/345600 >0.9999 32 61439/61440 >0.9999 8847359/8847360 >0.9999 8847359/8847360 >0.9999 33 >2230271/2230272 >0.9999 >16727039/16727040 >0.9999 >8363519/8363520 >0.9999 34 202619/202752 0.9993 3041279/3041280 >0.9999 168959/168960 >0.9999 35 671917/672000 0.9999 2015999/2016000 >0.9999 335999/336000 >0.9999 36 >3359231/3359232 >0.9999 >20155391/20155392 >0.9999 >6718463/6718464 >0.9999 37 >34549/34560 0.9997 >5598719/5598720 >0.9999 >2799359/2799360 >0.9999 38 69107/69120 0.9998 1244159/1244160 >0.9999 207359/207360 >0.9999 39 269563/269568 >0.9999 9704447/9704448 >0.9999 4852223/4852224 >0.9999 40 >2303993/2304000 >0.9999 >16588799/16588800 >0.9999 >6911999/6912000 >0.9999 41 >6296393/6297600 0.9998 >6297599/6297600 >0.9999 >1574399/1574400 >0.9999 42 >5418991/5419008 >0.9999 >10838015/10838016 >0.9999 >5419007/5419008 >0.9999 43 3919/3920 0.9997 1128959/1128960 >0.9999 564479/564480 >0.9999 44 619517/619520 >0.9999 >18585599/18585600 >0.9999 >4646399/4646400 >0.9999 45 >5598719/5598720 >0.9999 >27993599/27993600 >0.9999 >13996799/13996800 >0.9999 46 >1011999/1012000 >0.9999 >12143999/12144000 >0.9999 >6071999/6072000 >0.9999 47 >26495/26496 >0.9999 >1589759/1589760 >0.9999 >794879/794880 >0.9999 48 >1935359/1935360 >0.9999 >46448639/46448640 >0.9999 >7741439/7741440 >0.9999 49 >806399/806400 >0.9999 >7741439/7741440 >0.9999 >7741439/7741440 >0.9999 50 >1919999/1920000 >0.9999 >5759999/5760000 >0.9999 >5759999/5760000 >0.9999 Table 2: Covers with Unlikely Newton polygons and Unlikely Ekedahl-Oort Types 4 Unlikely Newton Polygons Using the algorithm to compute Newton polygons of covers, we found repeated occurences of certain kinds of Newton polygons, including supersingular Newton polygons and unlikely Newton polygons with large denominators. In this section, we prove Theorem 1.6 , which formalizes the patterns we saw in the data. We then prove the corollaries that follow from the theorem. Before proving Theorem 1.6 , we state a specific counterexample to Expectation 1.5 that does not arise from Theorem 1.6 . 4.1 A New Newton Polygon Pries notes in [ 15 , Section 5.3] that a curve of genus 12 with Newton polygon that only has slopes 5/12 and 7/12 is only known to exist in characteristic 2 . Expectation 1.5 says that no curve with this Newton polygon exists. Through the Shimura-Taniyama method we find that this Newton polygon also occurs for a curve in some odd characteristics. Proposition 4.1 . There is a curve with Newton polygon in characteristic p with slopes 5/12 and 7/12 , contrary to Expectation 1.5 , when p\equiv 3,12,17,33 modulo 35 . We demonstrate this when p\equiv 3 modulo 35 . Proof. The genus 12 curve has the equation y^{35}=x(1-x)^{20}. The ramification type and inertia type of this curve is [35,7,5,1] and (1,20,14) , respectively. If p\equiv 3 modulo 35 , then the orbits of the action of multiplication by p on \{1,2,3,\ldots,34\} are \displaystyle\{ \displaystyle\{1,3,9,27,11,33,29,17,16,13,4,12\},\{2,6,18,19,22,31,23,34,32,26,8,24\}, \displaystyle\{ \displaystyle 5,15,10,30,20,25\},\{7,21,28,14\}\} As 5 and 7 have orders 7 and 5 , respectively, they aren’t relevant in the Shimura-Taniyama method. We then see that the signature type f(j)=1 if and only if j\in\{1,2,3,4,6,8,9,11,13,16,18,23\} . Therefore, the first orbit has 7 elements in it with signature type 1 and the second orbit has 5 elements in it with signature type 1, so the Shimura-Taniyama formula tells us that the Newton polygon only has slopes 5/12 and 7/12 . ∎ 4.2 Quadratic Excesses and Newton Polygons of Covers In order to prove Theorem 1.6 , we define the quadratic excess of an interval. Definition 4.2 . Let \ell be a prime and a,b\in\mathbf{R} with a3 be a prime. If n and r are positive integers such that n\geq r , then \displaystyle\left(\frac{-1}{\ell}\right)q\left(\frac{\ell r}{4n},\frac{4\ell r}{4n}\right)=\sum_{i=1}^{3}q\left(\frac{\ell(in-r)}{4n},\frac{i\ell n}{4n}\right). (4.1) Proof. First, we note that \displaystyle q\left(\frac{\ell(n+r-1)}{4n},\frac{\ell(n+r)}{4n}\right)=\left(\frac{-1}{\ell}\right)q\left(\frac{\ell(3n-r)}{4n},\frac{\ell(3n-r+1)}{4n}\right). (4.2) We proceed by induction. If r=1 , then by [ 8 , Main Theorem] \displaystyle\begin{split}\left(\frac{4}{\ell}\right)q\left(0,\frac{\ell}{n}\right)&=q\left(0,\frac{\ell}{n}\right)=q\left(0,\frac{\ell}{4n}\right)+\left(\frac{-1}{\ell}\right)q\left(\frac{\ell(n-1)}{4n},\frac{\ell n}{4n}\right)\\ &+q\left(\frac{\ell n}{4n},\frac{\ell(n+1)}{4n}\right)+\left(\frac{-1}{\ell}\right)q\left(\frac{\ell(2n-1)}{4n},\frac{2\ell n}{4n}\right).\\ \end{split} (4.3) Substitute equation ( 4.2 ) with r=1 into equation ( 4.3 ), and after rearranging we get \displaystyle\left(\frac{-1}{\ell}\right)q\left(\frac{\ell}{4n},\frac{\ell}{n}\right) \displaystyle=q\left(\frac{\ell(n-1)}{4n},\frac{\ell n}{4n}\right)+q\left(\frac{\ell(2n-1)}{4n},\frac{2\ell n}{4n}\right)+q\left(\frac{\ell(3n-1)}{4n},\frac{3\ell n}{4n}\right), which is equation ( 4.1 ) when r=1 . Now, assume that for r=a3 be a prime and n\geq 1 such that gcd (n,\ell)=1 . Pick 0\leq k\leq n-1 such that k\equiv-3\ell^{-1} modulo n . Consider the cover \phi:X\to\mathbf{P}^{1} branched at three points with ramification type [n\ell,n\ell,\ell,1] and inertia type (1,(n-k)\ell-4,k\ell+3) . The equation for X is y^{n\ell}=x(1-x)^{(n-k)\ell-4} . Definition 4.7 . If \ell is a prime and n\geq 1 , we define a function I:\{0,1,2,\ldots,6n-1\}\to\mathcal{P}(\mathbf{\mathbf{Z}}) by I(i)=\begin{cases}\left[\left\lfloor\frac{\ell\left(\frac{i}{2}\right)}{3}\right\rfloor+1,\left\lfloor\frac{\ell\left(\frac{i}{2}+\left\lceil\frac{i+1}{6}\right\rceil\right)}{4}\right\rfloor\right]&i\textrm{ even}\\ \\ \left[\left\lfloor\frac{\ell\left(\frac{i-1}{2}+\left\lceil\frac{i}{6}\right\rceil\right)}{4}\right\rfloor+1,\left\lfloor\frac{\ell\left(\frac{i+1}{2}\right)}{3}\right\rfloor\right]&i\textrm{ odd}\end{cases} Note that \{I(i)\mid 0\leq i\leq 6n-1\} is a partition of the set of integers in the interval [1,\ell n-1] . Lemma 4.8 . Let \phi be the cover in Construction 4.6 , and let f be the signature type of \phi . If j\in\{1,\ldots,\ell n-1\} and j\not\equiv 0 modulo \ell , pick i such that 0\leq i\leq 6n-1 and j\in I(i) . Then, \displaystyle f(j)=\left\lceil\frac{jk+\left\lfloor\frac{i+2}{2}\right\rfloor}{n}\right\rceil-\left\lfloor\frac{jk+\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n}\right\rfloor Proof. Section 2.2 tells us that \displaystyle f(j)=-1+\left\langle\frac{j}{n\ell}\right\rangle \displaystyle+\left\langle\frac{j(k\ell+3)}{n\ell}\right\rangle+\left\langle\frac{j((n-k)\ell-4)}{n\ell}\right\rangle (4.6) \displaystyle=\left\lceil\frac{j(k\ell+3)}{n\ell}\right\rceil-\left\lfloor\frac{j(k\ell+4)}{n\ell}\right\rfloor (4.7) with equation ( 4.7 ) following from the definition of the fractional part of a real number. Given j\in\{1,2,\ldots,\ell n-1\} with j\not\equiv 0 modulo \ell and the i that determines the interval j lies in, we will prove the following: \displaystyle\left\lceil\frac{j(k\ell+3)}{n\ell}\right\rceil \displaystyle=\left\lceil\frac{jk+\lfloor\frac{i+2}{2}\rfloor}{n}\right\rceil (4.8) \displaystyle\left\lfloor\frac{j(k\ell+4)}{n\ell}\right\rfloor \displaystyle=\left\lfloor\frac{jk+\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n}\right\rfloor (4.9) To prove equation ( 4.8 ), we note that if i is even, then \displaystyle\frac{3j}{n\ell}\leq\frac{3\ell\left(\frac{i}{2}+\left\lceil\frac{i+1}{6}\right\rceil\right)}{4n\ell}\leq\frac{3\ell\left(\frac{i}{2}+\frac{i+5}{6}\right)}{4n\ell}=\frac{\frac{i+2}{2}}{n}-\frac{\ell}{8n\ell}<\frac{\left\lfloor\frac{i+2}{2}\right\rfloor}{n} and \displaystyle\frac{3j}{n\ell}\geq\frac{\ell\left(\frac{i}{2}\right)-2}{n\ell}+\frac{3}{n\ell}=\frac{\frac{i+2}{2}}{n}+\frac{1-\ell}{n\ell}>\frac{\left\lfloor\frac{i+2}{2}\right\rfloor}{n}-\frac{1}{n}. If i is odd, then \displaystyle\frac{3j}{n\ell}\leq\frac{\ell\left(\frac{i+1}{2}\right)}{n\ell}=\frac{\frac{i+1}{2}}{n}=\frac{\left\lfloor\frac{i+2}{2}\right\rfloor}{n} and \displaystyle\frac{3j}{n\ell}\geq\frac{3\ell\left(\frac{i-1}{2}+\frac{i}{6}\right)-9}{4n\ell}+\frac{3}{n\ell}=\frac{\frac{i+1}{2}}{n}+\frac{\frac{-21\ell}{6}+3}{4n\ell}>\frac{\left\lfloor\frac{i+2}{2}\right\rfloor}{n}-\frac{1}{n} For equation ( 4.9 ), if i is even, note that \left\lfloor\frac{i}{6}\right\rfloor+1=\left\lceil\frac{i+1}{6}\right\rceil and \frac{i}{2}+\left\lceil\frac{i+1}{6}\right\rceil is never 0 modulo 4 . Then, \displaystyle\frac{4j}{n\ell}\leq\frac{\ell\left(\frac{i}{2}+\left\lceil\frac{i+1}{6}\right\rceil\right)-1}{n\ell}=\frac{\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n}+\frac{\ell-1}{n\ell}<\frac{\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n}+\frac{1}{n} and \displaystyle\frac{4j}{n\ell}\geq\frac{4\ell\left(\frac{i}{2}\right)-8}{3n\ell}+\frac{4}{n\ell}=\frac{\frac{i}{2}+\frac{i}{6}}{n}+\frac{4}{3n\ell}>\frac{\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n} Observe that j is a multiple of \ell if and only if j is the rightmost endpoint of I(i) for some i\equiv 5 modulo 6 . Since we are excluding the case where j\equiv 0 modulo \ell , this gives us that if i is odd, \frac{i+1}{2} is never a multiple of 3 . Also, when i is odd and not 5 modulo 6 , we have that \frac{i+1}{6}\leq\left\lfloor\frac{i}{6}\right\rfloor+1 . We now show that if i is odd, \displaystyle\frac{4j}{n\ell}\leq\frac{4\ell\left(\frac{i+1}{2}\right)-1}{3n\ell}=\frac{\frac{i+1}{2}+\frac{i+1}{6}}{n}-\frac{1}{3n\ell}\leq\frac{\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor+1}{n}-\frac{1}{3n\ell}<\frac{\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n}+\frac{1}{n} and as i\not\equiv 0 modulo 6 in this case, \left\lceil\frac{i}{6}\right\rceil-1=\left\lfloor\frac{i}{6}\right\rfloor , so \displaystyle\frac{4j}{n\ell}\geq\frac{\ell\left(\frac{i-1}{2}+\left\lceil\frac{i}{6}\right\rceil\right)-3}{n\ell}+\frac{4}{n\ell}=\frac{\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n}+\frac{\ell+1}{n\ell}>\frac{\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor}{n}. ∎ Lemma 4.9 . Let \phi be the cover in Construction 4.8 , and let f be the signature type of \phi . Let j\in\{1,2,\ldots,\ell n-1\} such that j\equiv a\ell modulo n for some 0\leq a\leq n-1 and j\not\equiv 0 modulo \ell . We know that f(j)=1 if and only if • 0\leq 3a\leq n-1 and j\in\left(0,\frac{3a\ell}{4}\right)\cup\left(a\ell,\frac{\ell(n+3a)}{4}\right)\cup\left(\frac{\ell(n+3a)}{3},\frac{\ell(2n+3a)}{4}\right)\cup\left(\frac{\ell(2n+3a)}{3},\frac{\ell(3n+3a)}{4}\right), • or n\leq 3a\leq 2n-1 and j\in\left(0,\frac{\ell(3a-n)}{4}\right)\cup\left(\frac{\ell(3a-n)}{3},\frac{3\ell a}{4}\right)\cup\left(a\ell,\frac{\ell(n+3a)}{4}\right)\cup\left(\frac{\ell(n+3a)}{3},\frac{\ell(2n+3a)}{4}\right), • or 2n\leq 3a\leq 3n-1 and j\in\left(0,\frac{\ell(3a-2n)}{4}\right)\cup\left(\frac{\ell(3a-2n)}{3},\frac{\ell(3a-n)}{4}\right)\cup\left(\frac{\ell(3a-n)}{3},\frac{3\ell a}{4}\right)\cup\left(a\ell,\frac{\ell(n+3a)}{4}\right). . If j\equiv 0 modulo \ell , then by equation ( 4.7 ), f(j)=0 . The signature type f(j)=0 if and only if there is a multiple of n between jk+\left\lfloor\frac{i+2}{2}\right\rfloor+1 and jk+\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor . This means that if j\equiv a\ell modulo n for 0\leq a\leq n-1 , then f(j)=1 if and only if for all m such that \left\lfloor\frac{i+2}{2}\right\rfloor+1\leq m\leq\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor , m\not\equiv 3a modulo n . Note that jk\equiv-3a modulo n , so if m\equiv 3a modulo n , then jk+m\equiv 0 modulo n . We see that if i=0 , \left\lfloor\frac{i+2}{2}\right\rfloor>\left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor . Also, since 0\leq i\leq 6n-1 , \left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor-\left\lfloor\frac{i+2}{2}\right\rfloor\leq n-1 . We will prove the first case, where 0\leq 3a\leq n-1 . We want to find the largest i such that \left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor=3a-1 , which occurs when i is even. Note that when i is even, \left\lfloor\frac{i+1}{2}\right\rfloor+\left\lfloor\frac{i}{6}\right\rfloor=\frac{i}{2}+\left\lceil\frac{i+1}{6}\right\rceil-1 , so we have i even and \frac{i}{2}+\left\lceil\frac{i+1}{6}\right\rceil=3a , which appears in the rightmost bound of I(i) when i is even. Therefore, if j\equiv a\ell modulo n and 00 and converges to 1/2 as g approaches infinity, by doing a quick computation we have that if g>341 , then f(g)>0 and f(g) increases as g increases. Let \xi_{1} be the symmetric Newton polygon of height 2g with two slopes, where the smaller slope is \lfloor gf(g)\rfloor/g dim (\mathcal{A}_{g}[\xi_{2}]) and the codimension of \mathcal{A}_{g}[\xi_{2}] is bigger than the codimension of \mathcal{A}_{g}[\xi_{2}] in \mathcal{A}_{g} . We use Theorem 3.4 to compute the dimension of \mathcal{A}_{g}[\xi_{1}] , and we show \xi_{1} is unlikely and therefore \xi_{2} is unlikely. By applying Pick’s Theorem we can show that \textrm{dim}(\mathcal{A}_{g}[\xi_{1}])=\frac{g^{2}-g\lfloor gf(g)\rfloor-\lfloor gf(g)\rfloor+1}{2} and the codimension of \mathcal{A}_{g}[\xi_{1}] in \mathcal{A}_{g} is \frac{g+g\lfloor gf(g)\rfloor+\lfloor gf(g)\rfloor-1}{2}. The codimension is increasing as g increases if g>341 , and a computation shows that the codimension exceeds 3g-3 if g>363 , and therefore if g>363 is a Sophie Germain prime, \xi_{1} is unlikely and has large denominators. ∎ Example 5.1 . Let g=419 , which is a Sophie Germain prime. If n is a power of 2 , then Proposition 1.10 tells us that there exists a curve of genus ng defined over \overline{\mathbf{F}_{3}} whose Newton polygon has slopes 193/419 and 226/419 and is unlikely. Remark 5.2 . If Dickson’s conjecture is true, then if we consider the density of primes such that there exists a curve of genus g that has an unlikely Newton polygon with slopes that have large denominators, then Proposition 1.10 shows that the limit superior as g goes to infinity is 1 . Even without Dickson’s conjecture, there are very large Sophie Germain primes known. For example, for g=183027\times 2^{265440}-1 and any prime p\not\equiv-1 modulo (2g+1) , there exists a curve of arbitrarily large genus defined over \overline{\mathbf{F}_{p}} with an unlikely Newton polygon which has slopes that all have denominator g . 5.2 Results on Density Let g_{1},g_{2},\ldots,g_{k} be distinct odd Sophie Germain primes such that for 1\leq i,j\leq k , g_{i}\neq 2g_{j}+1 and g_{i}\nmid(g_{j}-1) . Let a\geq 0 and for 1\leq i\leq k define n_{i}:=2^{a}\prod_{b\neq i}g_{b} . Then, n_{i} is coprime to 2g_{i}+1 , g_{i}n_{i}=g_{j}n_{j} , and for i\neq j , gcd (g_{i},\varphi(n_{j}))=1 where \varphi is Euler’s totient function. Therefore, if p is any prime with order 2g_{i} in \mathbf{Z}/(2g_{i}+1)\mathbf{Z} for any i , then there is a supersingular curve of genus n_{i}g_{i} defined over \overline{\mathbf{F}_{p}} by Corollary 1.9 . We can use this construction to give a lower bound for the limit superior of \delta_{ss}(g) . We know that p having order 2g_{i} is equivalent to \ell being a quadratic nonresidue modulo 2g_{i}+1 such that p\not\equiv-1 modulo 2g_{i}+1 . There are exactly g_{i}-1 congruence classes that satisfy this, so the prime density in the case of one Sophie Germain prime is (g_{i}-1)/(2g_{i}) . We use the Inclusion-Exclusion Principle to compute the density of such primes given k Sophie Germain primes as above: \displaystyle\sum_{i=1}^{k}\frac{g_{i}-1}{2g_{i}}-\sum_{1\leq i2^{2m-2}-1 be a Sophie Germain prime. Assume there exists a Cunningham chain of the first kind of length 2m with \ell_{1}=\ell . Then, if we choose the Sophie Germain primes \ell_{1},\ell_{3},\ell_{5},\ldots,\ell_{2m-1} and relabel them as g_{1},g_{2},\ldots,g_{m} , then for 1\leq i,j\leq m , g_{i}\neq 2g_{j}+1 . Also, as g_{m}=2^{2m-2}g_{1}+2^{2m-2}-1 and \ell>2^{2m-2}-1 we have that g_{i}\nmid(g_{j}-1) . Conjecture 5.4 (Dickson) . [ 4 , Conjecture 1.3] If m\geq 1 , a_{1},a_{2},\ldots,a_{m} are integers, b_{1},b_{2},\ldots,b_{m} are all integers greater than or equal to 1 , and there is no prime that divides all of a_{1}+b_{1}n,a_{2}+b_{2}n,\ldots,a_{m}+b_{m}n for every n , then there are infinitely many n such that all of a_{1}+b_{1}n,a_{2}+b_{2}n,\ldots,a_{m}+b_{m}n are prime. Proof of Theorem 1.12 . Let a_{i}=2^{i-1}-1 and b_{i}=2^{i-1} for 1\leq i\leq 2m , then a_{1}+b_{1}n=n and a_{2}+b_{2}n=2n+1 . If p divides n , then p does not divide 2n+1 . Therefore, Dickson’s conjecture implies there are infinitely many Cunningham chains of the first kind of length 2m . Since there are infinitely many chains of length 2k , we can always choose one where the first prime is bigger than 2^{2k-2}-1 . Therefore, if Dickson’s conjecture is true, \limsup_{g\to\infty}\delta_{ss}(g)=1 . ∎ Remark 5.5 . As supersingular curves have an unlikely Newton polygon when g\geq 9 , Proposition 4.6 and Theorem 4.8 also show that \limsup_{g\to\infty}\delta_{nu}(g)>0.999 , and if Dickson’s Conjecture is true, \limsup_{g\to\infty}\delta_{nu}(g)=1 . 5.3 Proof of a Known Result As a final remark, we show that Theorem 1.6 gives another proof of a known but significant result. Theorem 5.6 ( [ 19 , Theorem 13.7] , [ 1 , Corollary 3.7(ii)] , [ 3 , Proposition 1.8.5] ) . Let p be an odd prime, then for arbitrarily large g there exists a supersingular curve of genus g defined over \overline{\mathbf{F}_{p}} . Proof. We need to choose a prime \ell such that p\not\equiv 1 modulo \ell and p is a quadratic residue modulo \ell . Suppose \ell\equiv 1 modulo 4 , then by quadratic reciprocity \Biggl(\frac{\ell}{p}\Biggr)\Biggl(\frac{p}{\ell}\Biggr)=1. This shows that \ell must be a quadratic residue modulo p , so choose \ell such that \ell\equiv 1 modulo 4 and \ell\equiv 1 modulo p . As long as we select \ell so that \ell does not divide p-1 , p\not\equiv 1 modulo \ell . Therefore, by Corollary 1.8 , there exists a supersingular curve of genus \frac{\ell-1}{2} defined over \overline{\mathbf{F}_{p}} . ∎