Introduction
A billiard ball in a polygon whose angles are rational multiples of π \pi π moves, after unfolding, along straight lines on a compact translation surface: the surface glued from the copies of the polygon under the finite group generated by the linear parts of the reflections in its sides [32 ] . Veech [28 ] singled out the translation surfaces whose affine automorphisms have derivatives forming a lattice in S L 2 ( R ) \mathrm{SL}_2(\mathbb{R}) SL 2 ( R ) . On such a Veech surface the straight-line flow in each direction is either completely periodic or uniquely ergodic, and the number of periodic families of length at most a given bound grows like a constant times the square of the bound. A rational polygon is a lattice polygon when its unfolding is a Veech surface. Veech showed that the obtuse isosceles triangles with angles ( π / n , π / n , ( n − 2 ) π / n ) (\pi/n,\pi/n,(n-2)\pi/n) ( π / n , π / n , ( n − 2 ) π / n ) , n ≥ 5 n \ge5 n ≥ 5 , are lattice polygons [28 Theorem 1.1 and §8] , and the question of which rational triangles are lattice polygons has been studied since.
We write ( p , q , r ) / n (p,q,r)/n ( p , q , r ) / n for the triangle with angles p π / n p\pi/n p π / n , q π / n q\pi/n q π / n , r π / n r\pi/n r π / n , where p ≤ q ≤ r p \le q \le r p ≤ q ≤ r are positive integers with gcd ( p , q , r ) = 1 \gcd(p,q,r)=1 g cd( p , q , r ) = 1 and n = p + q + r n=p+q+r n = p + q + r ; the integer n n n is the denominator . The triangle is obtuse exactly when r > n / 2 r>n/2 r > n /2 . It is a lattice triangle if the Veech group of its unfolding is a lattice, the unfolding being taken as a translation surface whose only marked points are its cone points of angle above 2 π 2\pi 2 π ; precise definitions are in § 2 .
Theorem 1.1. An obtuse rational triangle is a lattice triangle if and only if it is ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n with n ≥ 5 n \ge5 n ≥ 5 , ( 1 , 2 , 2 k − 3 ) / 2 k (1,2,2k-3)/2k ( 1 , 2 , 2 k − 3 ) /2 k with k ≥ 4 k \ge4 k ≥ 4 , or ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 .
The obtuse lattice triangles are those of Veech, of Vorobets and Ward, and Hooper’s triangle. This answers a question of Hooper [11 Open Question 2] . McMullen lists the known lattice triangles and asks whether the list is complete [21 Theorem 3.9 and Question 3.11] . The obtuse triangles of his list are all the obtuse lattice triangles, which answers his question for obtuse triangles. Theorem 1.1 also proves the conjecture, as stated by Angdinata, Chen, Ono, Thorner, Zhang and Zhang, that there are no obtuse lattice triangles other than the known ones [1 p. 2] . By Theorem 1.1 there are at most two obtuse lattice triangles of each denominator other than 12, and three of denominator 12. The proof is assembled in § 7 .
The idea of the proof is a first-order test on relative periods. By the criterion of Mirzakhani and Wright in the form of Larsen, Norton and Zykoski (Proposition 2.4), an obtuse triangle with a witness is not a lattice triangle; a witness is a unit a a a modulo n n n with 2 a ≠ 2 ( m o d n ) 2a \ne2 \pmod n 2 a = 2 ( mod n ) satisfying at least two of three inequalities, defined in § 2.3 . The criterion compares two eigenforms of the cyclic symmetry of the unfolding through the variation of the period matrix. We say that the criterion is blind on a triangle with no witness. Where it is blind we use the relative periods instead. On a Veech surface the difference of two zeros of the holomorphic form is a torsion point of the factor of the Jacobian with real multiplication by the trace field, of order dividing a fixed integer at every surface of the Teichmüller curve, by Möller’s theorem (Proposition 3.2). On the unfolding of a lattice triangle the eigenforms of the cyclic symmetry whose character is a unit modulo n n n lie in the cohomology of that factor, so the relative period of such an eigenform between two zeros, over a suitable rational relative class, vanishes along the whole Teichmüller disc. Its first derivative is given by integrals over the plane, which can be computed in closed form. For an integer x x x let [ x ] n [x]_n [ x ] n be its residue in [ 0 , n ) [0,n) [ 0 , n ) .
Theorem 1.2. Let Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n be a rational triangle. (a) If Δ \Delta Δ is obtuse, n n n is odd, and p ∤ n p \nmid n p ∤ n or q ∤ n q \nmid n q ∤ n , then Δ \Delta Δ is not a lattice triangle. (b) Let x ∈ { p , q , r } x \in\{p,q,r\} x ∈ { p , q , r } with x ∤ n x \nmid n x ∤ n and g = gcd ( x , n ) > 1 g=\gcd(x,n)>1 g = g cd( x , n ) > 1 , and let b b b be a unit modulo n n n with b ≡ 2 ( m o d n / g ) b \equiv2 \pmod{n/g} b ≡ 2 ( mod n / g ) , b ≠ 2 ( m o d n ) b \ne2 \pmod n b = 2 ( mod n ) and [ b p ] n + [ b q ] n + [ b r ] n = n [bp]_n+[bq]_n+[br]_n=n [ b p ] n + [ b q ] n + [ b r ] n = n . Then Δ \Delta Δ is not a lattice triangle.
The condition [ b p ] n + [ b q ] n + [ b r ] n = n [bp]_n+[bq]_n+[br]_n=n [ b p ] n + [ b q ] n + [ b r ] n = n in (b) says that the eigenspace of character b b b contains a holomorphic form. Part (a) places no conditions on the size of the obtuse angle, and part (b) none on the angles or on the parity of n n n .
The proof of Theorem 1.1 has three cases, by the parity of n n n and the size r π / n r\pi/n r π / n of the obtuse angle. The criterion rules out every triangle with a witness. When the obtuse angle is at most 2 π / 3 2\pi/3 2 π /3 , the window of § 4 , it is blind exactly on three families and a list (Theorem 4.1 ): A \mathcal{A} A and B \mathcal{B} B , in which s = 2 r − n s=2r-n s = 2 r − n divides n n n with odd quotient, with p = 1 p=1 p = 1 in A \mathcal{A} A , and 2 ≤ p ≤ s / 2 2\leq p\leq s/2 2 ≤ p ≤ s /2 and a condition on units in B \mathcal{B} B ; C \mathcal{C} C , in which s = 2 s=2 s = 2 , 4 ∣ n 4\mid n 4 ∣ n and p p p is 1 1 1 , 2 2 2 or, for n ≡ 4 ( m o d 8 ) n\equiv4\pmod{8} n ≡ 4 ( mod 8 ) , 4 4 4 ; and the list E \mathcal{E} E of thirty triangles with n ≤ 48 n\leq48 n ≤ 48 . Above 2 π / 3 2\pi/3 2 π /3 it is blind exactly on the six one-parameter families (1) to (6) and the seven triples (7) of Theorem 2.5 . For each case the table names the result that says where it is blind, the lattice triangles, and the result that rules out each of the other triangles; cylinders means two parallel cylinders whose moduli have an irrational ratio, or one whose share of the area has a Galois conjugate outside ( 0 , 1 ] (0,1] ( 0 , 1 ] , which cannot occur on a Veech surface (Lemma 2.3 ).
Table 1.
Theorem 4.1 , for every denominator, is the counterpart in the window of the theorem of Larsen, Norton and Zykoski for obtuse angle above 2 π / 3 2\pi/3 2 π /3 (Theorem 2.5). Theorem 4.1 (a), for finitely many denominators, and Theorem 6.1 are computer-assisted; the computations are exact and are described in Appendix A.
1.1. Earlier work. For right and acute isosceles triangles the answer is known. Kenyon and Smillie proved that a right triangle is a lattice triangle exactly when its smallest angle is π / n \pi/n π / n with n ≥ 4 n\geq4 n ≥ 4 , and an acute isosceles triangle exactly when its apex angle is π / n \pi/n π / n with $n\geq3 {citep}[15 Theorem 2] (the ‘if’ parts were known before {citep}[15 p. 67] ). For acute scalene triangles they reduced the question to a number-theoretic statement, which they verified for denominators up to 10,000 {citep}[15 Theorem 1] . {citet}Puchta [26 ] proved the statement under a coprimality assumption, and the general case is treated in a later preprint {citep}[27 ] . {citet}Apisa determined the orbit closure of the unfolding of every rational isosceles triangle {citep}[2 Theorems 1.4 and 7.7] ; in particular the obtuse isosceles lattice triangles are Veech’s. What has remained open is the case of obtuse scalene triangles (see {citep}[21 Theorem 3.9 and Corollary 3.10] `).
Two kinds of obtuse scalene lattice triangles are known. The triangles with angles ( π / 2 k , π / k , ( 2 k − 3 ) π / 2 k ) (\pi/2k,\pi/k,(2k-3)\pi/2k) ( π /2 k , π / k , ( 2 k − 3 ) π /2 k ) , k ≥ 4 k\geq4 k ≥ 4 , were suggested by computations of Boshernitzan reported by Veech [28 Remark on p. 582] , announced by Vorobets [29 Theorem 5] , proved to be lattice triangles in [30 Proposition 4.2 and Theorem 4.4] , and studied by Ward, who determined their Veech groups [31 Theorem A] . Bouw and Möller constructed, for every m < n m<n m < n , a Teichmüller curve uniformalized by a group commensurable to the triangle group ( m , n , ∞ ) (m,n,\infty) ( m , n , ∞ ) , and equal to it when m m m or n n n is odd [4 ] , with those of Veech and of Ward as special cases. The triangle with angles ( π / 12 , π / 3 , 7 π / 12 ) (\pi/12,\pi/3,7\pi/12) ( π /12 , π /3 , 7 π /12 ) was found by Hooper [11 Theorem 1] (a second proof is McMullen’s [19 end of §1] ). Four of the obtuse lattice triangles named so far have obtuse angle at most 2 π / 3 2\pi/3 2 π /3 : those with angles ( π / 5 , π / 5 , 3 π / 5 ) (\pi/5,\pi/5,3\pi/5) ( π /5 , π /5 , 3 π /5 ) , ( π / 6 , π / 6 , 2 π / 3 ) (\pi/6,\pi/6,2\pi/3) ( π /6 , π /6 , 2 π /3 ) , ( π / 8 , π / 4 , 5 π / 8 ) (\pi/8,\pi/4,5\pi/8) ( π /8 , π /4 , 5 π /8 ) and ( π / 12 , π / 3 , 7 π / 12 ) (\pi/12,\pi/3,7\pi/12) ( π /12 , π /3 , 7 π /12 ) . In the other direction, Vorobets [30 Proposition 4.2 and Theorem 4.5] showed that the triangles with angles ( π / 6 , π / 4 , 7 π / 12 ) (\pi/6,\pi/4,7\pi/12) ( π /6 , π /4 , 7 π /12 ) and ( π / 12 , π / 4 , 2 π / 3 ) (\pi/12,\pi/4,2\pi/3) ( π /12 , π /4 , 2 π /3 ) are not lattice triangles, and Ward [31 Theorem B] that those with angles ( π / n , q π / n , ( n − q − 1 ) π / n ) (\pi/n,q\pi/n,(n-q-1)\pi/n) ( π / n , q π / n , ( n − q − 1 ) π / n ) with q ≥ 2 q \ge2 q ≥ 2 , 4 q ≤ n 4q \le n 4 q ≤ n , and q q q or n n n odd are not, both by exhibiting two parallel cylinders whose moduli have an irrational ratio.
Larsen, Norton and Zykoski [17 ] answered Hooper’s question when the obtuse angle is at least 3 π / 4 3\pi/4 3 π /4 . Their tool is a criterion of Mirzakhani and Wright [22 ] , which shows that the orbit closure of the unfolding has rank at least two. Building on an observation of Wright, they showed that it applies to almost all obtuse triangles [17 Proposition 7.5] . They also showed that no lower bound on the smallest numerator p p p makes it apply to all obtuse triangles [17 Remark 3.6 and Proposition 3.7] , and determined exactly where it applies when the obtuse angle exceeds 2 π / 3 2\pi/3 2 π /3 : everywhere except on six one-parameter families and seven single triangles [17 Theorem 4.1] . Angdinata, Chen, Ono, Thorner, Zhang and Zhang [1 ] proved with the same criterion that the obtuse lattice triangles of denominator n n n form at most a proportion c ( ε ) n − 1 + ε c(\varepsilon)n^{-1+\varepsilon} c ( ε ) n − 1 + ε of all obtuse triangles of denominator n n n , and remark that the exponent − 1 -1 − 1 is the limit of their method [1 Remark 2] . Delecroix and Rüth showed with flatsurf that the unfoldings of the scalene triangles with p + q + r ≤ 58 p+q+r \le58 p + q + r ≤ 58 have dense orbit closures in their ambient strata (for even n n n , strata of quadratic differentials, as the unfolding has the rotation by π \pi π ), apart from an explicit list [6 Theorem 4.1] .
Further results
1.2. Further results. Theorem 1.3 extends the classification of Larsen, Norton and Zykoski [17 Theorem 1.1] from obtuse angle at least 3 π / 4 3\pi/4 3 π /4 to obtuse angle above 2 π / 3 2\pi/3 2 π /3 , and Theorem 1.4 is the case of odd denominators of Theorem 1.1.
Theorem 1.3. A rational triangle whose largest angle is greater than 2 π / 3 2\pi/3 2 π /3 is a lattice triangle if and only if it is ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n with n ≥ 7 n \ge7 n ≥ 7 or ( 1 , 2 , 2 k − 3 ) / 2 k (1,2,2k-3)/2k ( 1 , 2 , 2 k − 3 ) /2 k with k ≥ 5 k \ge5 k ≥ 5 .
Theorem 1.4. Let n n n be odd. An obtuse rational triangle of denominator n n n is a lattice triangle if and only if it is ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n with n ≥ 5 n \ge5 n ≥ 5 .
The proof of Theorem 1.4 uses neither Theorem 4.1 of Larsen, Norton and Zykoski nor any search over denominators.
Corollary 1.5. For every prime n ≥ 5 n \ge5 n ≥ 5 , the only obtuse lattice triangle of denominator n n n is ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n .
Corollary 1.5 follows from Theorem 1.2(a) and Veech’s theorem, since for prime n n n each of p p p , q q q divides n n n only if it is 1.
1.3. Plan. Section 2 sets up the unfolding, its cylinders and the criterion. Section 3 proves Theorem 1.2. Section 4 determines where the criterion is blind and proves Theorems 4.1 and 4.2; up to its last subsection, which applies Theorem 1.2 to two of the families, it is pure number theory and can be read on its own. Section 5 proves the cylinder theorems, and Section 6 treats the single triangles that no general theorem reaches. Section 7 proves Theorem 1.4, Corollary 1.5, Theorem 1.3 and Theorem 1.1. Appendix A describes the computations, Appendix B proves lemmas on residues for every denominator and Theorem 4.14, and Appendix C proves Theorem 4.1(a) for odd n n n .
Unfoldings, cylinders and the criterion
Triangles and their unfoldings
Throughout, Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n is a rational triangle as in § 1 , and A A A , B B B , C C C are its vertices with angles p π / n p\pi/n p π / n , q π / n q\pi/n q π / n , r π / n r\pi/n r π / n . When Δ \Delta Δ is obtuse, p ≤ q < r p\leq q<r p ≤ q < r and r r r does not divide n n n . For a real number t t t we write { t } \{t\} { t } for its fractional part.
Let D D D be the dihedral group of order 2 n 2n 2 n generated by the linear parts of the reflections in the sides of Δ \Delta Δ . The unfolding X X X of Δ \Delta Δ is the translation surface made of the 2 n 2n 2 n copies δ ( Δ ) \delta(\Delta) δ ( Δ ) , δ ∈ D \delta\in D δ ∈ D , where the copy δ ( Δ ) \delta(\Delta) δ ( Δ ) is glued along its side δ ( e ) \delta(e) δ ( e ) to the copy δ κ e ( Δ ) \delta\kappa_e(\Delta) δ κ e ( Δ ) , κ e \kappa_e κ e the linear part of the reflection in the side e e e , by the translation that makes the two copies of the side coincide [32 ] . A vertex of X X X is a class of corners of copies identified by these gluings.
Above the vertex of Δ \Delta Δ with angle x π / n x\pi/n x π / n there are gcd ( x , n ) \gcd(x,n) g cd( x , n ) vertices of X X X , each of cone angle 2 π x / gcd ( x , n ) 2\pi x/\gcd(x,n) 2 π x / g cd( x , n ) . A vertex of cone angle 2 π 2\pi 2 π , that is, a vertex above an angle x π / n x\pi/n x π / n with x ∣ n x\mid n x ∣ n , is an ordinary point of the flat surface. We write Σ \Sigma Σ for the set of vertices of cone angle greater than 2 π 2\pi 2 π , and regard X X X as a translation surface with singular set Σ \Sigma Σ . If g p g_p g p , g q g_q g q , g r g_r g r denote gcd ( p , n ) \gcd(p,n) g cd( p , n ) , gcd ( q , n ) \gcd(q,n) g cd( q , n ) , gcd ( r , n ) \gcd(r,n) g cd( r , n ) , the Euler characteristic gives
2 genus ( X ) − 2 = n − g p − g q − g r . (1) 2\operatorname{genus}(X)-2=n-g_p-g_q-g_r. \tag*{(1)} 2 genus ( X ) − 2 = n − g p − g q − g r . (1)
So X X X has genus at least two if and only if Σ \Sigma Σ is non-empty; this is the case for every obtuse triangle, as r ∤ n r\nmid n r ∤ n .
The Veech group Γ ( X ) \Gamma(X) Γ ( X ) is the group of derivatives of the orientation-preserving homeomorphisms of X X X that are affine in the translation charts; such a homeomorphism permutes Σ \Sigma Σ . The triangle Δ \Delta Δ is a lattice triangle if Γ ( X ) \Gamma(X) Γ ( X ) is a lattice in S L 2 ( R ) \mathrm{SL}_2(\mathbb{R}) SL 2 ( R ) .
Remark 2.1. If the vertices of cone angle 2 π 2\pi 2 π are added to the singular set, the Veech group can only become smaller, and a discrete subgroup of S L 2 ( R ) \mathrm{SL}_2(\mathbb{R}) SL 2 ( R ) containing a lattice is a lattice. So every statement below of the form “Δ \Delta Δ is not a lattice triangle” holds in that convention as well. The converse statements do not: for odd n ≥ 5 n\geq5 n ≥ 5 the two vertices of cone angle 2 π 2\pi 2 π of the unfolding of ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n are the centres of the two regular n n n -gons of Veech’s double n n n -gon, and with them marked the Veech group is not a lattice. This is due to Hubert and Schmidt [12 ] (see [18 §2.5.1] ), and for every odd n ≥ 5 n\geq5 n ≥ 5 it follows from the theorem of Apisa, Saavedra and Zhang that the periodic points of the double n n n -gon are exactly the Weierstrass points that are not singularities [3 Theorem 1.3] , since the centres of the two n n n -gons are not Weierstrass points for odd n n n .
Algebraically, X X X is the normalization of the curve y n = z p ( z − 1 ) q y^n=z^p(z-1)^q y n = z p ( z − 1 ) q , and the translation structure is given by
ω = y d z z ( z − 1 ) = z p / n − 1 ( z − 1 ) q / n − 1 d z , (2) \omega=\frac{y\,dz}{z(z-1)}=z^{p/n-1}(z-1)^{q/n-1}\,dz, \tag*{(2)} ω = z ( z − 1 ) y d z = z p / n − 1 ( z − 1 ) q / n − 1 d z , (2)
since the Schwarz–Christoffel map ∫ z p / n − 1 ( z − 1 ) q / n − 1 d z \int z^{p/n-1}(z-1)^{q/n-1}\,dz ∫ z p / n − 1 ( z − 1 ) q / n − 1 d z takes the upper half-plane onto a triangle with angles p π / n p\pi/n p π / n , q π / n q\pi/n q π / n , r π / n r\pi/n r π / n at the images of 0 0 0 , 1 1 1 , ∞ \infty ∞ (see [17 §2] and [22 §6] ). The points of X X X above 0 0 0 , 1 1 1 , ∞ \infty ∞ are the vertices above A A A , B B B , C C C , and at a point above the vertex with numerator x x x the form ω \omega ω has a zero of order x / gcd ( x , n ) − 1 x/\gcd(x,n)-1 x / g cd( x , n ) − 1 . The automorphism T ( z , y ) = ( z , ζ y ) T(z,y)=(z,\zeta y) T ( z , y ) = ( z , ζ y ) , ζ = e 2 π i / n \zeta=e^{2\pi i/n} ζ = e 2 π i / n , satisfies T ∗ ω = ζ ω T^*\omega=\zeta\omega T ∗ ω = ζ ω ; its linear part on the flat surface is the rotation by 2 π / n 2\pi/n 2 π / n .
Cylinders
Fix a direction d ∈ C ∖ { 0 } d\in\mathbb{C}\setminus\{0\} d ∈ C ∖ { 0 } . A closed regular leaf in direction d d d is a closed geodesic of X X X in direction d d d that does not meet Σ \Sigma Σ . The union of the closed regular leaves in direction d d d is open in X X X ; a cylinder in direction d d d is one of its connected components. When X X X has genus at least two, a cylinder is an open annulus foliated by closed leaves of a common holonomy c c c , its two ends are unions of saddle connections, and it is a maximal cylinder of closed leaves in Veech’s sense. Its circumference is ∣ c ∣ |c| ∣ c ∣ , its height H H H is its width measured perpendicular to d d d , and its modulus is H / ∣ c ∣ H/|c| H /∣ c ∣ . A cylinder ends only at points of Σ \Sigma Σ : a vertex of cone angle 2 π 2\pi 2 π on a boundary leaf does not stop it.
The unfolding X X X with its singular set Σ \Sigma Σ is a positive F F F -structure in the sense of Veech [28 §2] , with punctures Σ \Sigma Σ , and we use the following proposition of his.
Proposition 2.2 (Veech [28 Proposition 2.10] ). Let X X X be a translation surface with finite singular set Σ \Sigma Σ , regarded as a positive F F F -structure with punctures Σ \Sigma Σ , whose Veech group is a lattice, and let d ∈ C ∖ { 0 } d \in\mathbb{C} \setminus\{0\} d ∈ C ∖ { 0 } be a direction in which X X X has a saddle connection. Then X X X has an affine automorphism whose derivative is not the identity and which maps the foliation of X X X in direction d d d to itself. In particular, every leaf of that foliation is closed or a saddle connection.
Let ϑ = 2 π / n \vartheta= 2\pi/n ϑ = 2 π / n , so that ζ = e i ϑ \zeta= e^{i\vartheta} ζ = e i ϑ , and Q ( ζ ) + = Q ( ζ ) ∩ R = Q ( cos ϑ ) \mathbb{Q}(\zeta)^{+} = \mathbb{Q}(\zeta) \cap\mathbb{R} = \mathbb{Q}(\cos\vartheta) Q ( ζ ) + = Q ( ζ ) ∩ R = Q ( cos ϑ ) , a totally real field; for a unit ℓ \ell ℓ modulo n n n let σ ℓ \sigma_{\ell} σ ℓ be the automorphism of Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) with σ ℓ ( e i ϑ ) = e i ℓ ϑ \sigma_{\ell}(e^{i\vartheta}) = e^{i\ell\vartheta} σ ℓ ( e i ϑ ) = e i ℓ ϑ . It commutes with complex conjugation and acts on Q ( ζ ) + \mathbb{Q}(\zeta)^{+} Q ( ζ ) + by cos j ϑ ↦ cos j ℓ ϑ \cos j\vartheta\mapsto\cos j\ell\vartheta cos j ϑ ↦ cos j ℓ ϑ , and every embedding of Q ( ζ ) + \mathbb{Q}(\zeta)^{+} Q ( ζ ) + in R \mathbb{R} R is the restriction of some σ ℓ \sigma_{\ell} σ ℓ . Place Δ \Delta Δ with vertices 0 0 0 , 1 1 1 , v v v , Im v > 0 \operatorname{Im}v > 0 Im v > 0 , and angles p π / n p\pi/n p π / n at 0 0 0 , q π / n q\pi/n q π / n at 1 1 1 . By the law of sines v = sin ( q π / n ) e i p π / n / sin ( r π / n ) v = \sin(q\pi/n)e^{ip\pi/n}/\sin(r\pi/n) v = sin ( q π / n ) e i p π / n / sin ( r π / n ) ; multiplying numerator and denominator of
e i ( p + q ) π / n − e i ( p − q ) π / n e i r π / n − e − i r π / n \frac{e^{i(p+q)\pi/n}-e^{i(p-q)\pi/n}}{e^{ir\pi/n}-e^{-ir\pi/n}} e i r π / n − e − i r π / n e i ( p + q ) π / n − e i ( p − q ) π / n
by e i r π / n e^{ir\pi/n} e i r π / n and using e i ( p + q + r ) π / n = − 1 e^{i(p+q+r)\pi/n} = -1 e i ( p + q + r ) π / n = − 1 gives v = ( e − i q ϑ − 1 ) / ( e i r ϑ − 1 ) ∈ Q ( ζ ) v = (e^{-iq\vartheta}-1)/(e^{ir\vartheta}-1) \in\mathbb{Q}(\zeta) v = ( e − i q ϑ − 1 ) / ( e i r ϑ − 1 ) ∈ Q ( ζ ) . The 2 n 2n 2 n copies of Δ \Delta Δ in X X X are e i j ϑ Δ e^{ij\vartheta}\Delta e ij ϑ Δ and e i j ϑ Δ ‾ e^{ij\vartheta}\overline{\Delta} e ij ϑ Δ , j ∈ Z / n j \in\mathbb{Z}/n j ∈ Z / n , so their vertices lie in Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) , and the holonomy of every saddle connection and of every closed geodesic of X X X lies in Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) , being the holonomy of a path along sides of the copies. This is the observation of Kenyon and Smillie [15 §5] . For a cylinder Z Z Z of X X X let ρ ( Z ) = area ( Z ) / area ( X ) \rho(Z) = \operatorname{area}(Z)/\operatorname{area}(X) ρ ( Z ) = area ( Z ) / area ( X ) be its share of the area. If c , w ∈ Q ( ζ ) c,w \in\mathbb{Q}(\zeta) c , w ∈ Q ( ζ ) are the holonomies of the core curve of Z Z Z and of a segment crossing Z Z Z between singular points on its two boundary components, then
ρ ( Z ) = c ‾ w − c w ‾ n ( v − v ‾ ) (up to sign) , \rho(Z) = \frac{\overline{c}w-c\overline{w}}{n(v-\overline{v})} \qquad\text{(up to sign)}, ρ ( Z ) = n ( v − v ) c w − c w (up to sign) ,
since area ( Z ) = ∣ Im ( c w ‾ ) ∣ \operatorname{area}(Z) = |\operatorname{Im}(c\overline{w})| area ( Z ) = ∣ Im ( c w ) ∣ and area ( X ) = 2 n ⋅ 1 2 Im v \operatorname{area}(X) = 2n \cdot\frac{1}{2}\operatorname{Im}v area ( X ) = 2 n ⋅ 2 1 Im v . So ρ ( Z ) \rho(Z) ρ ( Z ) is a real element of Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) : ρ ( Z ) ∈ Q ( ζ ) + \rho(Z) \in\mathbb{Q}(\zeta)^{+} ρ ( Z ) ∈ Q ( ζ ) + .
Part (ii) of the next lemma follows in two lines from the proof of Theorem 1 of Fenves with his Lemma 3 [7 ] ; Kenyon and Smillie used a positivity condition of the same kind [15 Theorem 21 and Proposition 23] .
Lemma 2.3. Let Δ \Delta Δ be a rational triangle whose unfolding X X X has genus at least two, and let Z Z Z be a maximal cylinder of X X X , in direction d d d . If Δ \Delta Δ is a lattice triangle, then:
any two maximal cylinders of X X X in direction d d d have moduli whose ratio is rational;
0 < σ ( ρ ( Z ) ) ≤ 1 0 < \sigma(\rho(Z)) \le1 0 < σ ( ρ ( Z )) ≤ 1 for every embedding σ \sigma σ of Q ( ζ ) + \mathbb{Q}(\zeta)^{+} Q ( ζ ) + in R \mathbb{R} R .
Proof. A boundary component of Z Z Z is a union of saddle connections in direction d d d , so by Proposition 2.2 every leaf in direction d d d is closed or a saddle connection. Then X X X is the union of the closures of finitely many maximal cylinders Z j Z_j Z j in direction d d d [28 Proposition 2.4] , and their moduli μ j \mu_j μ j satisfy μ j = λ j μ 1 \mu_j = \lambda_j\mu_1 μ j = λ j μ 1 with λ j ∈ Q > 0 \lambda_j \in\mathbb{Q}_{>0} λ j ∈ Q > 0 [28 Remark on p. 582] . A closed leaf of Z Z Z meets no saddle connection, so it lies in one of the Z j Z_j Z j ; both Z Z Z and Z j Z_j Z j are connected components of the union of the closed regular leaves in direction $d containing that leaf, so Z = Z j Z = Z_j Z = Z j . The same holds for every maximal cylinder in direction d d d , so each of them is one of the Z j Z_j Z j ; this proves (i). Let c j ∈ Q ( ζ ) c_j \in\mathbb{Q}(\zeta) c j ∈ Q ( ζ ) be the holonomy of the core curve of Z j Z_j Z j . Then area ( Z j ) = μ j ∣ c j ∣ 2 \operatorname{area}(Z_j)=\mu_j|c_j|^2 area ( Z j ) = μ j ∣ c j ∣ 2 , and the numbers x j = c j / c 1 x_j=c_j/c_1 x j = c j / c 1 lie in Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) and are real, since the c j c_j c j are parallel; so x j ∈ Q ( ζ ) + ∖ { 0 } x_j\in\mathbb{Q}(\zeta)^+\setminus\{0\} x j ∈ Q ( ζ ) + ∖ { 0 } , and
ρ ( Z i ) = μ i ∣ c i ∣ 2 ∑ j μ j ∣ c j ∣ 2 = λ i x i 2 ∑ j λ j x j 2 . \rho(Z_i)=\frac{\mu_i|c_i|^2}{\sum_j\mu_j|c_j|^2}=\frac{\lambda_i x_i^2}{\sum_j\lambda_jx_j^2}. ρ ( Z i ) = ∑ j μ j ∣ c j ∣ 2 μ i ∣ c i ∣ 2 = ∑ j λ j x j 2 λ i x i 2 .
Both sides are elements of Q ( ζ ) + \mathbb{Q}(\zeta)^+ Q ( ζ ) + . For an embedding σ \sigma σ of Q ( ζ ) + \mathbb{Q}(\zeta)^+ Q ( ζ ) + , each σ ( x j ) \sigma(x_j) σ ( x j ) is a non-zero real number, so σ ( ρ ( Z i ) ) \sigma(\rho(Z_i)) σ ( ρ ( Z i )) is a quotient of positive numbers whose numerator is one of the terms of the denominator, and 0 < σ ( ρ ( Z i ) ) ≤ 1 0<\sigma(\rho(Z_i))\leq1 0 < σ ( ρ ( Z i )) ≤ 1 . □
Conversely, a translation surface on which every direction containing a cylinder decomposes into cylinders with pairwise commensurable moduli is a Veech surface [7 Theorem 1] ; so every rational triangle that is not a lattice triangle has a direction containing a cylinder in which X X X does not decompose into cylinders with pairwise commensurable moduli.
The criterion of Mirzakhani and Wright
Mirzakhani and Wright proved that the orbit closure of the unfolding has rank at least two (the rank is half the complex dimension of the image of its tangent space in absolute cohomology; a closed orbit has rank one) when the eigenspaces of T T T of characters a a a and 2 − a 2-a 2 − a both contain holomorphic forms, for a unit a a a with 2 a ≠ 2 ( m o d n ) 2a\ne2\pmod n 2 a = 2 ( mod n ) [22 Theorem 7.5] . Larsen, Norton and Zykoski put this in the following form.
Proposition 2.4 (Larsen, Norton and Zykoski [17 Proposition 2.1] ). Let Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n be obtuse. If some unit a a a modulo n n n with 2 a ≠ 2 ( m o d n ) 2a\ne2\pmod n 2 a = 2 ( mod n ) satisfies two of the three inequalities
[ a p ] n < [ 2 p ] n , [ a q ] n < [ 2 q ] n , [ a r ] n < [ 2 r ] n , [ap]_n<[2p]_n,\qquad[aq]_n<[2q]_n,\qquad[ar]_n<[2r]_n, [ a p ] n < [ 2 p ] n , [ a q ] n < [ 2 q ] n , [ a r ] n < [ 2 r ] n , then Δ \Delta Δ is not a lattice triangle.
Following Larsen, Norton and Zykoski, for an integer x x x set
S x = { a ∈ Z / n Z : [ a x ] n < [ 2 x ] n } , S_x=\{a\in\mathbb{Z}/n\mathbb{Z}:[ax]_n<[2x]_n\}, S x = { a ∈ Z / n Z : [ a x ] n < [ 2 x ] n } ,
and call a unit a a a modulo n n n usable if 2 a ≠ 2 ( m o d n ) 2a\ne2\pmod n 2 a = 2 ( mod n ) . A witness for an obtuse triangle is a usable unit a a a for which at least two of the three inequalities [ a p ] n < [ 2 p ] n [ap]_n<[2p]_n [ a p ] n < [ 2 p ] n , [ a q ] n < [ 2 q ] n [aq]_n<[2q]_n [ a q ] n < [ 2 q ] n , [ a r ] n < [ 2 r ] n [ar]_n<[2r]_n [ a r ] n < [ 2 r ] n hold, that is, which lies in at least two of S p , S q , S r S_p,S_q,S_r S p , S q , S r , where for p = q p=q p = q the first two inequalities are one inequality and count as two, as in Proposition 2.4.
For an obtuse triangle write s = 2 r − n s=2r-n s = 2 r − n , so that [ 2 p ] n = 2 p [2p]_n=2p [ 2 p ] n = 2 p , [ 2 q ] n = 2 q [2q]_n=2q [ 2 q ] n = 2 q and [ 2 r ] n = s [2r]_n=s [ 2 r ] n = s .
Theorem 2.5 ([17 Theorem 4.1] ). An obtuse triangle ( p , q , r ) / n (p,q,r)/n ( p , q , r ) / n with r > 2 n / 3 r>2n/3 r > 2 n /3 has a witness if and only if none of the following is the case:
p = q = 1 p=q=1 p = q = 1
p = 1 p=1 p = 1 , q = 2 q=2 q = 2 , n n n is even
p = 1 p=1 p = 1 , q = 4 q=4 q = 4 , r ≡ 7 ( m o d 8 ) r\equiv7\pmod8 r ≡ 7 ( mod 8 )
p = 1 p=1 p = 1 , r = 3 q + 1 r=3q+1 r = 3 q + 1
p = 2 p=2 p = 2 , r = 3 q + 2 r=3q+2 r = 3 q + 2
p = 4 p=4 p = 4 , r = 3 q + 4 r=3q+4 r = 3 q + 4
( p , q , r ) (p,q,r) ( p , q , r ) is one of the following triples: ( 1 , 4 , 11 ) (1,4,11) ( 1 , 4 , 11 ) , ( 1 , 3 , 16 ) (1,3,16) ( 1 , 3 , 16 ) , ( 2 , 3 , 17 ) (2,3,17) ( 2 , 3 , 17 ) , ( 1 , 4 , 21 ) (1,4,21) ( 1 , 4 , 21 ) , ( 1 , 8 , 19 ) (1,8,19) ( 1 , 8 , 19 ) , ( 3 , 8 , 29 ) (3,8,29) ( 3 , 8 , 29 ) , ( 2 , 11 , 29 ) (2,11,29) ( 2 , 11 , 29 ) .
The members of family (3) with n > 12 n>12 n > 12 are not lattice triangles: Larsen, Norton and Zykoski find two horizontal cylinders whose moduli have the irrational ratio 4 cos 2 ( 2 π / n ) 4\cos^2(2\pi/n) 4 cos 2 ( 2 π / n ) [17 §10] .
Relative periods
This section proves Theorem 1.2 , which is applied in § 7 to odd denominators and to the single triangles of Lemma 7.1, and in Section 4.4 to the families A \mathcal{A} A and B \mathcal{B} B . On the unfolding of a lattice triangle, the relative period of an eigenform with unit character between two zeros of ω \omega ω , over a suitable rational relative class, vanishes along the whole Teichmüller disc (Lemmas 3.3 , 3.4 and 3.5 ). Its first derivative is the difference of the values at these zeros of a solution G G G of a ∂ ˉ \bar{\partial} ∂ ˉ -equation (Lemma 3.6 ), and on a triangle unfolding these values are integrals over the plane that can be computed. The obstruction is thus of first order, as is the one behind the criterion of Mirzakhani and Wright, whose proof, as they remark, can be viewed as showing that the S L 2 ( R ) \mathrm{SL}_{2}(\mathbb{R}) SL 2 ( R ) -orbit of the unfolding is not even tangent to the locus of the surfaces with the endomorphisms that an orbit closure of rank one would force [22 after the proof of Theorem 7.5] .
Forms on the unfolding
In this section we allow any placement of the vertices. Let ( x 0 , x 1 , x ∞ ) (x_0,x_1,x_\infty) ( x 0 , x 1 , x ∞ ) be an ordering of ( p , q , r ) (p,q,r) ( p , q , r ) , set θ i = x i / n \theta_i=x_i/n θ i = x i / n and g i = gcd ( x i , n ) g_i=\gcd(x_i,n) g i = g cd( x i , n ) for i ∈ { 0 , 1 , ∞ } i\in\{0,1,\infty\} i ∈ { 0 , 1 , ∞ } , and let X X X be the normalization of the curve y n = z x 0 ( z − 1 ) x 1 y^n=z^{x_0}(z-1)^{x_1} y n = z x 0 ( z − 1 ) x 1 , with ω = y d z / ( z ( z − 1 ) ) \omega=y\,\mathrm{d}z/(z(z-1)) ω = y d z / ( z ( z − 1 )) and T ( z , y ) = ( z , ζ y ) T(z,y)=(z,\zeta y) T ( z , y ) = ( z , ζ y ) , ζ = e 2 π i / n \zeta=e^{2\pi i/n} ζ = e 2 π i / n . The substitutions z ↦ 1 − z z\mapsto1-z z ↦ 1 − z and z ↦ 1 / z z\mapsto1/z z ↦ 1/ z (with y y y multiplied by a suitable constant, respectively by a suitable constant times 1 / z 1/z 1/ z ) permute the numerators and multiply ω \omega ω by a non-zero constant, so ( X , ω ) (X,\omega) ( X , ω ) is the unfolding of Δ \Delta Δ of § 2.1 up to a constant factor, which changes neither the zeros of ω \omega ω nor whether Δ \Delta Δ is a lattice triangle. Above z = 0 , 1 , ∞ z=0,1,\infty z = 0 , 1 , ∞ there are g 0 g_0 g 0 , g 1 g_1 g 1 , g ∞ g_\infty g ∞ points; at a point above z = 0 z=0 z = 0 there is a local coordinate u u u with z = u n / g 0 z=u^{n/g_0} z = u n / g 0 and y / u x 0 / g 0 y/u^{x_0/g_0} y / u x 0 / g 0 a unit (as ( y / u x 0 / g 0 ) n = ( z − 1 ) x 1 (y/u^{x_0/g_0})^n=(z-1)^{x_1} ( y / u x 0 / g 0 ) n = ( z − 1 ) x 1 ), and similarly above 1 1 1 and above ∞ \infty ∞ (with z = u − n / g ∞ z=u^{-n/g_\infty} z = u − n / g ∞ ). The form ω \omega ω has a zero of order x i / g i − 1 x_i/g_i-1 x i / g i − 1 at each point above the vertex i i i , so these points are zeros of ω \omega ω exactly when x i ∤ n x_i\nmid n x i ∤ n ; the set of zeros is the set Σ \Sigma Σ of § 2.1 .
A meromorphic 1 1 1 -form η \eta η on X X X has character b ∈ Z / n Z b\in\mathbb{Z}/n\mathbb{Z} b ∈ Z / n Z if T ∗ η = ζ b η T^*\eta=\zeta^b\eta T ∗ η = ζ b η ; for instance ω \omega ω has character 1 1 1 . Let H b 1 ⊂ H 1 ( X , C ) H_b^1\subset H^1(X,\mathbb{C}) H b 1 ⊂ H 1 ( X , C ) be the ζ b \zeta^b ζ b -eigenspace of T ∗ T^* T ∗ and H b 1 , 0 H_b^{1,0} H b 1 , 0 the space of holomorphic forms of character b b b . Since T ∗ T^* T ∗ is real, complex conjugation maps H b 1 , 0 H_b^{1,0} H b 1 , 0 onto the ζ − b \zeta^{-b} ζ − b -eigenspace of T ∗ T^* T ∗ on antiholomorphic forms, so H b 1 = H b 1 , 0 ⊕ H − b 0 , 1 H_b^1=H_b^{1,0}\oplus H_{-b}^{0,1} H b 1 = H b 1 , 0 ⊕ H − b 0 , 1 .
Lemma 3.1. Let b b b be an integer.
(a) Every non-zero meromorphic form of character b b b is y b R ( z ) d z y^bR(z)\,\mathrm{d}z y b R ( z ) d z with R R R rational. It is holomorphic if and only if R = z i ( z − 1 ) j h ( z ) R=z^i(z-1)^jh(z) R = z i ( z − 1 ) j h ( z ) with i , j i,j i , j and a polynomial h h h with h ( 0 ) h ( 1 ) ≠ 0 h(0)h(1)\ne0 h ( 0 ) h ( 1 ) = 0 such that, with e 0 = b θ 0 + i e_0=b\theta_0+i e 0 = b θ 0 + i and e 1 = b θ 1 + j e_1=b\theta_1+j e 1 = b θ 1 + j ,
e 0 > − 1 , e 1 > − 1 , e 0 + e 1 + deg h < − 1. e_0>-1,\qquad e_1>-1,\qquad e_0+e_1+\deg h<-1. e 0 > − 1 , e 1 > − 1 , e 0 + e 1 + deg h < − 1. For any R = z i ( z − 1 ) j h R=z^i(z-1)^jh R = z i ( z − 1 ) j h its order at a point above 0 0 0 is ( n / g 0 ) ( e 0 + 1 ) − 1 (n/g_0)(e_0+1)-1 ( n / g 0 ) ( e 0 + 1 ) − 1 , above 1 1 1 it is ( n / g 1 ) ( e 1 + 1 ) − 1 (n/g_1)(e_1+1)-1 ( n / g 1 ) ( e 1 + 1 ) − 1 , and above ∞ \infty ∞ it is − ( n / g ∞ ) ( e 0 + e 1 + deg h + 1 ) − 1 -(n/g_\infty)(e_0+e_1+\deg h+1)-1 − ( n / g ∞ ) ( e 0 + e 1 + deg h + 1 ) − 1 .
(b) If b x i ≠ 0 ( m o d n ) bx_i\ne0\pmod n b x i = 0 ( mod n ) for i = 0 , 1 , ∞ i=0,1,\infty i = 0 , 1 , ∞ , then dim H b 1 , 0 = 2 − ( { b θ 0 } + { b θ 1 } + { b θ ∞ } ) \dim H_b^{1,0}=2-\left(\{b\theta_0\}+\{b\theta_1\}+\{b\theta_\infty\}\right) dim H b 1 , 0 = 2 − ( { b θ 0 } + { b θ 1 } + { b θ ∞ } ) . So H b 1 , 0 H_b^{1,0} H b 1 , 0 has dimension 1 1 1 if [ b x 0 ] n + [ b x 1 ] n + [ b x ∞ ] n = n [bx_0]_n+[bx_1]_n+[bx_\infty]_n=n [ b x 0 ] n + [ b x 1 ] n + [ b x ∞ ] n = n , spanned by y b z i ( z − 1 ) j d z y^bz^i(z-1)^j\mathrm{d}z y b z i ( z − 1 ) j d z with b θ 0 + i = { b θ 0 } − 1 b\theta_0+i=\{b\theta_0\}-1 b θ 0 + i = { b θ 0 } − 1 and b θ 1 + j = { b θ 1 } − 1 b\theta_1+j=\{b\theta_1\}-1 b θ 1 + j = { b θ 1 } − 1 , and it is zero otherwise. Moreover dim H b 1 = 1 \dim H_b^1=1 dim H b 1 = 1 .
(c) If b ≠ 0 b\ne0 b = 0 but b x i ≡ 0 ( m o d n ) bx_i\equiv0\pmod n b x i ≡ 0 ( mod n ) for some i i i , then H b 1 = 0 H_b^1=0 H b 1 = 0 . Also H 0 1 , 0 = 0 H_0^{1,0}=0 H 0 1 , 0 = 0 .
Proof. (a) If η \eta η has character b b b , then η / ( y b d z ) \eta/(y^b\,\mathrm{d}z) η / ( y b d z ) is a T T T -invariant meromorphic function, hence a rational function of z z z . Above a point of C ∖ { 0 , 1 } \mathbb{C}\setminus\{0,1\} C ∖ { 0 , 1 } the map z z z is unramified and u u u is a unit, so a holomorphic η \eta η has R R R without poles off 0 0 0 , 1 1 1 , ∞ \infty ∞ , that is R = z i ( z − 1 ) j h ( z ) R=z^i(z-1)^jh(z) R = z i ( z − 1 ) j h ( z ) as stated. In the local coordinate above 0 0 0 the form is a unit times u b x 0 / g 0 + i n / g 0 d ( u n / g 0 ) u^{bx_0/g_0+in/g_0}\,d(u^{n/g_0}) u b x 0 / g 0 + in / g 0 d ( u n / g 0 ) , of order ( n / g 0 ) ( b θ 0 + i + 1 ) − 1 (n/g_0)(b\theta_0+i+1)-1 ( n / g 0 ) ( b θ 0 + i + 1 ) − 1 , an integer; it is non-negative if and only if e 0 > − 1 e_0>-1 e 0 > − 1 . The same computation applies above 1 1 1 , and above ∞ \infty ∞ , where y y y is a unit times u − ( x 0 + x 1 ) / g ∞ u^{-(x_0+x_1)/g_\infty} u − ( x 0 + x 1 ) / g ∞ and z i ( z − 1 ) j h ( z ) d z z^i(z-1)^jh(z)\,dz z i ( z − 1 ) j h ( z ) d z is a unit times u − ( n / g ∞ ) ( i + j + deg h + 1 ) − 1 d u u^{-(n/g_\infty)(i+j+\deg h+1)-1}\,du u − ( n / g ∞ ) ( i + j + d e g h + 1 ) − 1 d u .
(b), (c) For b ≠ 0 b\ne0 b = 0 these are [22 Lemma 6.1 and Corollary 6.2] , with k = n k=n k = n , q i = x i q_i=x_i q i = x i and t ( b ) = ∑ i { b θ i } t(b)=\sum_i\{b\theta_i\} t ( b ) = ∑ i { b θ i } in their notation, after the change of variable that puts the three vertices at 0 0 0 , 1 1 1 , ∞ \infty ∞ : a Möbius transformation sending distinct points z 0 , z 1 , z ∞ ∈ C z_0,z_1,z_\infty\in\mathbb{C} z 0 , z 1 , z ∞ ∈ C to 0 , 1 , ∞ 0,1,\infty 0 , 1 , ∞ , with y y y replaced by a constant times y / ( z − z ∞ ) y/(z-z_\infty) y / ( z − z ∞ ) , takes their curve y n = ∏ i ( z − z i ) x i y^n=\prod_i(z-z_i)^{x_i} y n = ∏ i ( z − z i ) x i to ours, as x 0 + x 1 + x ∞ = n x_0+x_1+x_\infty=n x 0 + x 1 + x ∞ = n , and commutes with T T T . The form named in (b) has character b b b and is holomorphic by (a): for it h = 1 h=1 h = 1 , e 0 = { b θ 0 } − 1 > − 1 e_0=\{b\theta_0\}-1>-1 e 0 = { b θ 0 } − 1 > − 1 , e 1 = { b θ 1 } − 1 > − 1 e_1=\{b\theta_1\}-1>-1 e 1 = { b θ 1 } − 1 > − 1 and e 0 + e 1 = − 1 − { b θ ∞ } < − 1 e_0+e_1=-1-\{b\theta_\infty\}<-1 e 0 + e 1 = − 1 − { b θ ∞ } < − 1 . In (c) exactly one b x i bx_i b x i is divisible by n n n (if two were, all three would be, and then b ≡ 0 b\equiv0 b ≡ 0 , as gcd ( x 0 , x 1 , x ∞ ) = 1 \gcd(x_0,x_1,x_\infty)=1 g cd( x 0 , x 1 , x ∞ ) = 1 ), so dim H b 1 = t ( b ) + t ( − b ) − 2 = 0 \dim H^1_b=t(b)+t(-b)-2=0 dim H b 1 = t ( b ) + t ( − b ) − 2 = 0 . Finally H 0 1 , 0 = 0 H^{1,0}_0=0 H 0 1 , 0 = 0 , because a T T T -invariant holomorphic form descends to X / ⟨ T ⟩ = P 1 X/\langle T\rangle=\mathbb{P}^1 X / ⟨ T ⟩ = P 1 .
Summing (b) over b b b gives the genus (1) . A holomorphic form of character b b b is therefore determined up to a constant when it exists, and we write η b \eta_b η b for it.
The Teichmüller disc and Möller’s theorem
The Teichmüller disc. Let ( X , ω ) (X,\omega) ( X , ω ) be a translation surface with zero set Σ \Sigma Σ , and let D \mathbb{D} D be the unit disc. For t ∈ D t\in\mathbb{D} t ∈ D let X t X_t X t be the Riemann surface whose underlying topological surface is that of X X X , marked by the identity, and whose holomorphic charts are w + t w ˉ w+t\bar{w} w + t w ˉ off Σ \Sigma Σ , where w w w is a local primitive of ω \omega ω , and u t = u ( 1 + t u ˉ k + 1 / u k + 1 ) 1 / ( k + 1 ) u_t=u(1+t\bar{u}^{k+1}/u^{k+1})^{1/(k+1)} u t = u ( 1 + t u ˉ k + 1 / u k + 1 ) 1/ ( k + 1 ) near a zero of order k k k , where u u u is a coordinate with u k + 1 = ( k + 1 ) w u^{k+1}=(k+1)w u k + 1 = ( k + 1 ) w . Then u t k + 1 = ( k + 1 ) ( w + t w ˉ ) u_t^{k+1}=(k+1)(w+t\bar{w}) u t k + 1 = ( k + 1 ) ( w + t w ˉ ) , so the charts are compatible, and ω t = d ( w + t w ˉ ) \omega_t=d(w+t\bar{w}) ω t = d ( w + t w ˉ ) is a holomorphic form on X t X_t X t with the same zeros. The pair ( X t , ω t ) (X_t,\omega_t) ( X t , ω t ) is the image of ( X , ω ) (X,\omega) ( X , ω ) under the real-linear map v ↦ v + t v ˉ v\mapsto v+t\bar{v} v ↦ v + t v ˉ of C = R 2 \mathbb{C}=\mathbb{R}^2 C = R 2 , which has positive determinant 1 − ∣ t ∣ 2 1-|t|^2 1 − ∣ t ∣ 2 ; every element of G L 2 + ( R ) \mathrm{GL}^+_2(\mathbb{R}) GL 2 + ( R ) is a complex multiple of exactly one such map, so t ↦ X t t\mapsto X_t t ↦ X t is the Teichmüller disc of ( X , ω ) (X,\omega) ( X , ω ) , and its image in the moduli space is the Teichmüller curve C = H / Γ C=\mathbb{H}/\Gamma C = H /Γ when ( X , ω ) (X,\omega) ( X , ω ) is a Veech surface with Veech group Γ \Gamma Γ (we identify D \mathbb{D} D with the upper half-plane H \mathbb{H} H ). The identity map X → X t X\to X_t X → X t is a homeomorphism, not in general smooth at the zeros, and it identifies H 1 ( X t , Q ) H^1(X_t,\mathbb{Q}) H 1 ( X t , Q ) with H 1 ( X , Q ) H^1(X,\mathbb{Q}) H 1 ( X , Q ) for every t t t . We write X 0 = X X_0=X X 0 = X .
The complex surface D × X \mathbb{D}\times X D × X . The product D × X \mathbb{D}\times X D × X carries the complex structure with charts ( t , w + t w ˉ ) (t,w+t\bar{w}) ( t , w + t w ˉ ) away from the zeros of ω \omega ω and ( t , u t ) (t,u_t) ( t , u t ) near a zero. Both charts are holomorphic in t t t at a fixed point of X X X , so t ↦ ( t , P ) t\mapsto(t,P) t ↦ ( t , P ) is a holomorphic section for every point P P P of X X X , and the fibre over t t t is X t X_t X t . A holomorphic family η t \eta_t η t of forms, η t \eta_t η t holomorphic on X t X_t X t , is a holomorphic relative 1 1 1 -form on this complex surface; a local primitive Φ ( t , x ) \Phi(t,x) Φ ( t , x ) along the fibres is a holomorphic function, determined up to adding a holomorphic function of t t t .
The family over the Teichmüller curve. Let ( X , ω ) (X,\omega) ( X , ω ) be a Veech surface of genus at least two whose singular set is its zero set Σ \Sigma Σ , with Veech group Γ \Gamma Γ , trace field K = Q ( Tr Γ ) K=\mathbb{Q}(\operatorname{Tr}\Gamma) K = Q ( Tr Γ ) . For every subgroup Γ 1 ⊂ Γ \Gamma_1\subset\Gamma Γ 1 ⊂ Γ of finite index that is small enough there is a family of curves f : X 1 → C 1 = H / Γ 1 f:X_1\to C_1=\mathbb{H}/\Gamma_1 f : X 1 → C 1 = H / Γ 1 , the pull-back of the universal family over a finite cover of the moduli space of curves, whose fibre over the image of a point of H = D \mathbb{H}=\mathbb{D} H = D is the corresponding surface of the Teichmüller disc [23 §1] . Let V = R 1 f ∗ Z \mathbb{V}=R^1f_*\mathbb{Z} V = R 1 f ∗ Z , with its variation of Hodge structure. By [24 Proposition 2.4] it splits as V Q = W Q ⊕ M Q \mathbb{V}_{\mathbb{Q}}=\mathbb{W}_{\mathbb{Q}}\oplus\mathbb{M}_{\mathbb{Q}} V Q = W Q ⊕ M Q , a direct sum of rational variations of Hodge structure of weight one, where W Q \mathbb{W}_{\mathbb{Q}} W Q has rank 2 [ K : Q ] 2[K:\mathbb{Q}] 2 [ K : Q ] and the fibre of W Q ⊗ R \mathbb{W}_{\mathbb{Q}}\otimes\mathbb{R} W Q ⊗ R at X X X contains R ω \mathbb{R}\omega R ω and I ω \mathbb{I}\omega I ω . Möller’s factor A 1 A_1 A 1 of Jac ( X 1 / C 1 ) \operatorname{Jac}(X_1/C_1) Jac ( X 1 / C 1 ) with real multiplication by K K K is the one given by this splitting: there is a homomorphism π : Jac ( X 1 / C 1 ) → A 1 \pi:\operatorname{Jac}(X_1/C_1)\to A_1 π : Jac ( X 1 / C 1 ) → A 1 of abelian schemes over C 1 C_1 C 1 whose pull-back maps the rational first cohomology of the fibres of A 1 A_1 A 1 isomorphically onto W Q \mathbb{W}_{\mathbb{Q}} W Q (see [23 (2), and the paragraph after Theorem 3.1] and [24 before Proposition 2.6] ). By [23 Theorem 3.1] the group of sections of A 1 → C 1 A_1 \to C_1 A 1 → C 1 is finite.
Proposition 3.2. Let ( X , ω ) (X,\omega) ( X , ω ) be a Veech surface of genus at least two whose singular set is its zero set Σ \Sigma Σ , with trace field K K K . There is a subspace W ⊂ H 1 ( X , Q ) W \subset H^1(X,\mathbb{Q}) W ⊂ H 1 ( X , Q ) of dimension 2 [ K : Q ] 2[K:\mathbb{Q}] 2 [ K : Q ] such that:
(a) [ ω ] ∈ W ⊗ C [\omega] \in W \otimes\mathbb{C} [ ω ] ∈ W ⊗ C ; for every t ∈ D t \in\mathbb{D} t ∈ D , W ⊗ C = W t 1 , 0 ⊕ W t 1 , 0 ‾ W \otimes\mathbb{C} = W_t^{1,0} \oplus\overline{W_t^{1,0}} W ⊗ C = W t 1 , 0 ⊕ W t 1 , 0 with W t 1 , 0 = ( W ⊗ C ) ∩ H 1 , 0 ( X t ) W_t^{1,0} = (W \otimes\mathbb{C}) \cap H^{1,0}(X_t) W t 1 , 0 = ( W ⊗ C ) ∩ H 1 , 0 ( X t ) ; and the spaces W t 1 , 0 W_t^{1,0} W t 1 , 0 form a holomorphic subbundle of the Hodge bundle over D \mathbb{D} D ;
(b) for every divisor D D D of degree 0 0 0 supported on Σ \Sigma Σ there is an integer N ≥ 1 N \ge1 N ≥ 1 with the following property. Let γ \gamma γ be a 1 1 1 -chain on X X X with ∂ γ = D \partial\gamma= D ∂ γ = D . For every t ∈ D t \in\mathbb{D} t ∈ D there is λ t ∈ H 1 ( X , Z ) \lambda_t \in H_1(X,\mathbb{Z}) λ t ∈ H 1 ( X , Z ) with
N ∫ γ η = ∫ λ t η for every holomorphic form η on X t with [ η ] ∈ W ⊗ C . N\int_\gamma\eta= \int_{\lambda_t}\eta\qquad\text{for every holomorphic form }\eta\text{ on }X_t\text{ with }[\eta]\in W\otimes\mathbb{C}. N ∫ γ η = ∫ λ t η for every holomorphic form η on X t with [ η ] ∈ W ⊗ C .
Proof. Take Γ 1 ⊂ Γ \Gamma_1 \subset\Gamma Γ 1 ⊂ Γ of finite index, acting freely on H \mathbb{H} H (Selberg’s lemma; the lattice Γ \Gamma Γ is finitely generated), and so small that f f f exists and that every P ∈ Σ P \in\Sigma P ∈ Σ extends to a section P ~ \widetilde{P} P of f f f whose value on each fibre is a zero of the form of that fibre: the orders of the zeros are constant along the G L 2 + ( R ) \mathrm{GL}_2^+(\mathbb{R}) GL 2 + ( R ) -orbit, so the zeros form a curve in X 1 X_1 X 1 that is finite and unramified over C 1 C_1 C 1 , and it becomes trivial over a finite unramified covering of C 1 C_1 C 1 (as in the proof of [23 Lemma 1.2] ). Then D ≅ H → C 1 \mathbb{D} \cong\mathbb{H} \to C_1 D ≅ H → C 1 is a universal covering, and C 1 → C C_1 \to C C 1 → C is unramified as a map of orbifolds, though not in general as a map of the underlying curves: the surface X X X itself is an orbifold point of C C C when its Veech group contains a rotation, as it does for a triangle unfolding. Every derivative below is taken in the parameter t t t of the disc, which is a local coordinate of C 1 C_1 C 1 everywhere. The pull-back of f f f along D → C 1 \mathbb{D} \to C_1 D → C 1 is a family of curves of genus at least two over the simply connected disc, marked at X X X ; a marked family over D \mathbb{D} D is the pull-back of the universal family over Teichmüller space by its classifying map, which here is the Teichmüller disc t ↦ X t t \mapsto X_t t ↦ X t . So the pull-back is the family t ↦ X t t \mapsto X_t t ↦ X t (as a complex surface it is D × X \mathbb{D} \times X D × X , described at the start of this subsection). The sections P ~ \widetilde{P} P pull back to the constant sections t ↦ P t \mapsto P t ↦ P : the pull-back of P ~ \widetilde{P} P is continuous, its values lie in the finite set Σ \Sigma Σ of zeros of ω t \omega_t ω t , and its value at t = 0 t = 0 t = 0 is P P P .
Let W W W be the fibre of W Q \mathbb{W}_{\mathbb{Q}} W Q at X X X , a constant subspace of H 1 ( X , Q ) H^1(X,\mathbb{Q}) H 1 ( X , Q ) along the disc, of dimension 2 [ K : Q ] 2[K:\mathbb{Q}] 2 [ K : Q ] . As W Q \mathbb{W}_{\mathbb{Q}} W Q is a sub-variation of Hodge structure of weight one and W ⊗ R W \otimes\mathbb{R} W ⊗ R contains Re ω \operatorname{Re}\omega Re ω and Im ω \operatorname{Im}\omega Im ω , part (a) holds, and W t 1 , 0 W_t^{1,0} W t 1 , 0 has dimension [ K : Q ] [K:\mathbb{Q}] [ K : Q ] .
Let D = ∑ P D ( P ) P D = \sum_P D(P)P D = ∑ P D ( P ) P be a divisor of degree 0 0 0 on Σ \Sigma Σ , and σ D = ∑ P D ( P ) π ( P ~ − P ~ 0 ) \sigma_D = \sum_P D(P)\,\pi(\widetilde{P} - \widetilde{P}_0) σ D = ∑ P D ( P ) π ( P − P 0 ) for a fixed P 0 ∈ Σ P_0 \in\Sigma P 0 ∈ Σ , a section of A 1 → C 1 A_1 \to C_1 A 1 → C 1 ; it has finite order N 1 N_1 N 1 . Let t ∈ D t \in\mathbb{D} t ∈ D , with image c ∈ C 1 c \in C_1 c ∈ C 1 , and write Jac ( X t ) = H 1 ( X t ) / H 1 ( X , Z ) \operatorname{Jac}(X_t) = H^1(X_t)/H_1(X,\mathbb{Z}) Jac ( X t ) = H 1 ( X t ) / H 1 ( X , Z ) and A 1 , c = H 1 ( A 1 , c ) / H 1 ( A 1 , c , Z ) A_{1,c} = H^1(A_{1,c})/H_1(A_{1,c},\mathbb{Z}) A 1 , c = H 1 ( A 1 , c ) / H 1 ( A 1 , c , Z ) . The differential d π c d\pi_c d π c is the transpose of π c ∗ : H 1 ( A 1 , c ) → H 1 ( X t ) \pi_c^*: H^1(A_{1,c}) \to H^1(X_t) π c ∗ : H 1 ( A 1 , c ) → H 1 ( X t ) ; as π c ∗ \pi_c^* π c ∗ is an injective morphism of Hodge structures with image W W W , it maps H 1 , 0 ( A 1 , c ) H^{1,0}(A_{1,c}) H 1 , 0 ( A 1 , c ) onto W t 1 , 0 W_t^{1,0} W t 1 , 0 (both have dimension [ K : Q ] [K:\mathbb{Q}] [ K : Q ] ), so the kernel of d π c d\pi_c d π c consists of the functionals that vanish on W t 1 , 0 W_t^{1,0} W t 1 , 0 . The cokernels of ( π c ) ∗ : H 1 ( X , Z ) → H 1 ( A 1 , c , Z ) (\pi_c)_*: H_1(X,\mathbb{Z}) \to H_1(A_{1,c},\mathbb{Z}) ( π c ) ∗ : H 1 ( X , Z ) → H 1 ( A 1 , c , Z ) are finite and form a local system on C 1 C_1 C 1 ; let m m m be the exponent of its fibres. As the sections P ~ \widetilde{P} P are constant along the disc, σ D ( c ) \sigma_D(c) σ D ( c ) is the image under π c \pi_c π c of the class of η ↦ ∫ γ η \eta\mapsto\int_\gamma\eta η ↦ ∫ γ η , for any chain γ \gamma γ with ∂ γ = D \partial\gamma= D ∂ γ = D . So N 1 σ D ( c ) = 0 N_1\sigma_D(c) = 0 N 1 σ D ( c ) = 0 gives λ t ∈ H 1 ( X , Z ) \lambda_t \in H_1(X,\mathbb{Z}) λ t ∈ H 1 ( X , Z ) with d π c ( m N 1 ∫ γ ) = d π c ( ∫ λ t ) d\pi_c(mN_1\int_\gamma) = d\pi_c(\int_{\lambda_t}) d π c ( m N 1 ∫ γ ) = d π c ( ∫ λ t ) , that is, m N 1 ∫ γ η = ∫ λ t η mN_1\int_\gamma\eta= \int_{\lambda_t}\eta m N 1 ∫ γ η = ∫ λ t η for every η ∈ W t 1 , 0 \eta\in W_t^{1,0} η ∈ W t 1 , 0 . This is (b) with N = m N 1 N = mN_1 N = m N 1 .
For t ∈ D t \in\mathbb{D} t ∈ D let A t = ( W t 1 , 0 ) ∨ / H 1 ( X , Z ) A_t = (W_t^{1,0})^\vee/H_1(X,\mathbb{Z}) A t = ( W t 1 , 0 ) ∨ / H 1 ( X , Z ) , where a class λ \lambda λ acts by η ↦ ∫ λ η \eta\mapsto\int_\lambda\eta η ↦ ∫ λ η ; by (a) it is a complex torus of dimension [ K : Q ] [K:\mathbb{Q}] [ K : Q ] . Part (b) says that the image of D D D in A t A_t A t , the class of functional η ↦ ∫ γ η \eta\mapsto\int_{\gamma}\eta η ↦ ∫ γ η on W t 1 , 0 W_t^{1,0} W t 1 , 0 , has order dividing N N N for every t t t . As the kernel of d π C d\pi_{\mathcal C} d π C consists of the functionals that vanish on W t 1 , 0 W_t^{1,0} W t 1 , 0 and the cokernel of ( π C ) ∗ (\pi_{\mathcal C})_* ( π C ) ∗ is finite, π C \pi_{\mathcal C} π C factors through the map Jac ( X t ) → A t \operatorname{Jac}(X_t) \to A_t Jac ( X t ) → A t that restricts functionals to W t 1 , 0 W_t^{1,0} W t 1 , 0 , and the induced map from A t A_t A t to the fibre of A 1 A_1 A 1 at the image of t t t in C 1 C_1 C 1 is an isogeny. This is the factor A A A of the Jacobian, of dimension [ K : Q ] [K:\mathbb{Q}] [ K : Q ] , in which the differences of the zeros stay torsion. When [ K : Q ] [K:\mathbb{Q}] [ K : Q ] is the genus, W = H 1 ( X , Q ) W=H^1(X,\mathbb{Q}) W = H 1 ( X , Q ) and A t = Jac ( X t ) A_t=\operatorname{Jac}(X_t) A t = Jac ( X t ) , so they stay torsion in the whole Jacobian. On the unfolding of a triangle the difference of two zeros of ω \omega ω is torsion in the Jacobian, as Mirzakhani and Wright remark [22 §1] (it also follows from Lemma 3.4 below); what a lattice triangle adds is that it stays torsion, in A A A , along the whole Teichmüller disc.
In Filip’s extension of Möller’s theorem to affine invariant manifolds the torsion statement holds for the combinations of zeros annihilated by the relative directions of the manifold, and for Teichmüller curves, which have none, for all of them [8 Theorem 1.3 and Remark 1.4 (ii), (iii)] . So the argument below shows that a triangle is not a lattice triangle, not that its orbit closure has rank at least two, and indeed Theorem 1.2 (b) applies to ( 1 , 3 , 6 ) / 10 (1,3,6)/10 ( 1 , 3 , 6 ) /10 and ( 1 , 3 , 8 ) / 12 (1,3,8)/12 ( 1 , 3 , 8 ) /12 (Example 3.11). Delecroix and Rüth list these two as rank-one orbit closures that are not Teichmüller curves, with a reference to work in preparation [6 Theorem 4.1] . A criterion that proves rank at least two cannot apply to a triangle of rank one. Möller’s torsion condition has been used to exclude Teichmüller curves on the stable curves over the cusps, by McMullen [20 ] and, for a factor of the Jacobian, by Lanneau and Möller [16 §3] ; here it is read at the surface of the triangle itself.
Relative periods that vanish along the disc.
Lemma 3.3. Let Δ \Delta Δ be a lattice triangle whose unfolding ( X , ω ) (X,\omega) ( X , ω ) has genus at least two, and let W W W be as in Proposition 3.2. Then H b 1 ⊂ W ⊗ C H_b^1 \subset W \otimes\mathbb{C} H b 1 ⊂ W ⊗ C for every unit b b b modulo n n n . In particular H b 1 , 0 ⊂ W 0 1 , 0 H_b^{1,0} \subset W_0^{1,0} H b 1 , 0 ⊂ W 0 1 , 0 for every unit b b b , and b = 2 b=2 b = 2 is a unit exactly when n n n is odd.
Proof. The automorphism T ∗ T^* T ∗ of H 1 ( X , Q ) H^1(X,\mathbb{Q}) H 1 ( X , Q ) is defined over Q \mathbb{Q} Q , so its ζ \zeta ζ -eigenspace H 1 1 H_1^1 H 1 1 is defined over Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) ; it contains [ ω ] [\omega] [ ω ] and has dimension 1 by Lemma 3.1 (b), as 1 is a unit and so is non-trivial at the three vertices. Let v ∈ H 1 ( X , Q ( ζ ) ) v \in H^1(X,\mathbb{Q}(\zeta)) v ∈ H 1 ( X , Q ( ζ )) span it. Then v ∈ C [ ω ] ⊂ W ⊗ C v \in\mathbb{C}[\omega] \subset W \otimes\mathbb{C} v ∈ C [ ω ] ⊂ W ⊗ C by Proposition 3.2(a), so v ∈ ( W ⊗ C ) ∩ H 1 ( X , Q ( ζ ) ) = W ⊗ Q ( ζ ) v \in(W \otimes\mathbb{C}) \cap H^1(X,\mathbb{Q}(\zeta)) = W \otimes\mathbb{Q}(\zeta) v ∈ ( W ⊗ C ) ∩ H 1 ( X , Q ( ζ )) = W ⊗ Q ( ζ ) , as W W W is a rational subspace. For a unit b b b let σ b \sigma_b σ b be the automorphism of Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) with σ b ( ζ ) = ζ b \sigma_b(\zeta)=\zeta^b σ b ( ζ ) = ζ b , acting on the coefficients. Then σ b ( v ) ∈ W ⊗ Q ( ζ ) \sigma_b(v) \in W \otimes\mathbb{Q}(\zeta) σ b ( v ) ∈ W ⊗ Q ( ζ ) , since W W W is rational, and T ∗ σ b ( v ) = σ b ( T ∗ v ) = ζ b σ b ( v ) T^*\sigma_b(v)=\sigma_b(T^*v)=\zeta^b\sigma_b(v) T ∗ σ b ( v ) = σ b ( T ∗ v ) = ζ b σ b ( v ) , since T ∗ T^* T ∗ is rational. So σ b ( v ) \sigma_b(v) σ b ( v ) is a non-zero vector of H b 1 H_b^1 H b 1 , which has dimension 1 by Lemma 3.1 (b); hence H b 1 ⊂ W ⊗ C H_b^1 \subset W \otimes\mathbb{C} H b 1 ⊂ W ⊗ C . □
Lemma 3.4. Let X X X be the curve of § 3.1 for any rational triangle, and let Σ 0 \Sigma_0 Σ 0 , Σ 1 \Sigma_1 Σ 1 , Σ ∞ \Sigma_\infty Σ ∞ be the sums of the points of X X X above z = 0 z=0 z = 0 , 1 1 1 , ∞ \infty ∞ .
(a) div ( z ) = ( n / g 0 ) Σ 0 − ( n / g ∞ ) Σ ∞ \operatorname{div}(z)=(n/g_0)\Sigma_0-(n/g_\infty)\Sigma_\infty div ( z ) = ( n / g 0 ) Σ 0 − ( n / g ∞ ) Σ ∞ and div ( z − 1 ) = ( n / g 1 ) Σ 1 − ( n / g ∞ ) Σ ∞ \operatorname{div}(z-1)=(n/g_1)\Sigma_1-(n/g_\infty)\Sigma_\infty div ( z − 1 ) = ( n / g 1 ) Σ 1 − ( n / g ∞ ) Σ ∞ .
(b) Let P P P , P ′ P' P ′ be two points above the same vertex, say z = 0 z=0 z = 0 , and g = g 0 g=g_0 g = g 0 . Then ( n / g ) ( P − P ′ ) (n/g)(P-P') ( n / g ) ( P − P ′ ) is the divisor of a meromorphic function on X X X .
Consequently every divisor D D D of degree 0 supported above { 0 , 1 , ∞ } \{0,1,\infty\} { 0 , 1 , ∞ } is torsion in Jac ( X ) \operatorname{Jac}(X) Jac ( X ) , and there is a class γ ~ ∈ H 1 ( X , supp D ; Q ) \widetilde{\gamma}\in H_1(X,\operatorname{supp}D;\mathbb{Q}) γ ∈ H 1 ( X , supp D ; Q ) with ∂ γ ~ = D \partial\widetilde{\gamma}=D ∂ γ = D such that ∫ γ ~ η = 0 \int_{\widetilde{\gamma}}\eta=0 ∫ γ η = 0 and ∫ γ ~ η ‾ = 0 \int_{\widetilde{\gamma}}\overline{\eta}=0 ∫ γ η = 0 for every holomorphic form η \eta η on X X X .
Proof. (a) follows from the local coordinates of § 3.1 . (b) The substitutions of § 3.1 move any vertex to z = 0 z=0 z = 0 , so we may assume the points lie above 0. The quotient X / ⟨ T g ⟩ X/\langle T^g\rangle X / ⟨ T g ⟩ is the normalization of y ′ ′ g = z x 0 ( z − 1 ) x 1 y''{}^g=z^{x_0}(z-1)^{x_1} y ′′ g = z x 0 ( z − 1 ) x 1 with y ′ ′ = y n / g y''=y^{n/g} y ′′ = y n / g , which is invariant under T g T^g T g ; with w = y ′ ′ / z x 0 / g w=y''/z^{x_0/g} w = y ′′ / z x 0 / g it becomes w g = ( z − 1 ) x 1 w^g=(z-1)^{x_1} w g = ( z − 1 ) x 1 . Here gcd ( x 1 , g ) = 1 \gcd(x_1,g)=1 g cd( x 1 , g ) = 1 , since a a a prime dividing g g g and x 1 x_1 x 1 divides x 0 x_0 x 0 , x 1 x_1 x 1 and n n n , hence all three numerators. Choosing integers α \alpha α , β \beta β with α x 1 + β g = 1 \alpha x_1+\beta g=1 α x 1 + β g = 1 , the function v = w α ( z − 1 ) β v=w^\alpha(z-1)^\beta v = w α ( z − 1 ) β satisfies v g = z − 1 v^g=z-1 v g = z − 1 and w = v x 1 w=v^{x_1} w = v x 1 , so the quotient has genus 0. The stabilizer of a point above z = 0 z=0 z = 0 in ⟨ T ⟩ \langle T\rangle ⟨ T ⟩ has order n / g n/g n / g , since the g g g points above 0 form one orbit; it is ⟨ T g ⟩ \langle T^g\rangle ⟨ T g ⟩ . So the quotient map is totally ramified, of index n / g n/g n / g , at P P P and at P ′ P' P ′ , and the pull-back of a function on the quotient with divisor (image of P P P ) − - − (image of P ′ P' P ′ ) has divisor ( n / g ) ( P − P ′ ) (n/g)(P-P') ( n / g ) ( P − P ′ ) .
The differences of two points above the same vertex, and the divisors g ∞ Σ 0 − g 0 Σ ∞ g_\infty\Sigma_0-g_0\Sigma_\infty g ∞ Σ 0 − g 0 Σ ∞ , g ∞ Σ 1 − g 1 Σ ∞ g_\infty\Sigma_1-g_1\Sigma_\infty g ∞ Σ 1 − g 1 Σ ∞ (of which n / ( g 0 g ∞ ) n/(g_0g_\infty) n / ( g 0 g ∞ ) and n / ( g 1 g ∞ ) n/(g_1g_\infty) n / ( g 1 g ∞ ) times are principal; g 0 g_0 g 0 , g 1 g_1 g 1 , g ∞ g_\infty g ∞ are pairwise coprime because gcd ( p , q , r ) = 1 \gcd(p,q,r)=1 g cd( p , q , r ) = 1 ), generate the divisors of degree 0 supported above { 0 , 1 , ∞ } \{0,1,\infty\} { 0 , 1 , ∞ } up to finite index, so every such D D D has a multiple N D = div ( F ) ND=\operatorname{div}(F) N D = div ( F ) . Let γ \gamma γ be a 1-chain with ∂ γ = D \partial\gamma=D ∂ γ = D . By Abel’s theorem η ↦ N ∫ γ η \eta\mapsto N\int_\gamma\eta η ↦ N ∫ γ η is the period map of some λ ∈ H 1 ( X , Z ) \lambda\in H_1(X,\mathbb{Z}) λ ∈ H 1 ( X , Z ) on holomorphic forms, and γ ^ = γ − λ / N \hat{\gamma}=\gamma-\lambda/N γ ^ = γ − λ / N has the stated property; since γ ^ \hat{\gamma} γ ^ is a real class, ∫ γ ^ η ˉ = ∫ γ ^ η = 0 \int_{\hat{\gamma}}\bar{\eta}=\int_{\hat{\gamma}}\eta=0 ∫ γ ^ η ˉ = ∫ γ ^ η = 0 . □
Lemma 3.5. Let ( X , ω ) (X,\omega) ( X , ω ) be a Veech surface as in Proposition 3.2, D D D a divisor of degree 0 supported on Σ \Sigma Σ , and γ ^ ∈ H 1 ( X , Σ ; Q ) \hat{\gamma}\in H_1(X,\Sigma;\mathbb{Q}) γ ^ ∈ H 1 ( X , Σ ; Q ) a class with ∂ γ ^ = D \partial\hat{\gamma}=D ∂ γ ^ = D such that ∫ γ ^ η = 0 \int_{\hat{\gamma}}\eta=0 ∫ γ ^ η = 0 for every holomorphic form η \eta η on X X X . Then ∫ γ ^ η = 0 \int_{\hat{\gamma}}\eta=0 ∫ γ ^ η = 0 for every t ∈ D t\in\mathbb{D} t ∈ D and every holomorphic form η \eta η on X t X_t X t with [ η ] ∈ W ⊗ C [\eta]\in W\otimes\mathbb{C} [ η ] ∈ W ⊗ C .
The relative homology H 1 ( X , Σ ; Z ) H_1(X,\Sigma;\mathbb{Z}) H 1 ( X , Σ ; Z ) and the homology H 1 ( X , Z ) H_1(X,\mathbb{Z}) H 1 ( X , Z ) do not change along the disc, because the underlying surface and the points of Σ \Sigma Σ do not move; only the complex structure does.
Proof. Let Λ ⊂ Hom ( W , R ) \Lambda\subset\operatorname{Hom}(W,\mathbb{R}) Λ ⊂ Hom ( W , R ) be the image of H 1 ( X , Z ) H_1(X,\mathbb{Z}) H 1 ( X , Z ) , a lattice, and for t ∈ D t\in\mathbb{D} t ∈ D let ι t : Hom ( W , R ) → ( W t 1 , 0 ) ∨ \iota_t:\operatorname{Hom}(W,\mathbb{R})\to(W_t^{1,0})^\vee ι t : Hom ( W , R ) → ( W t 1 , 0 ) ∨ send a functional to the restriction of its complex-linear extension; by Proposition 3.2(a), ι t \iota_t ι t is an R \mathbb{R} R -linear isomorphism, and it depends continuously on t t t . Fix a chain γ \gamma γ with ∂ γ = D \partial\gamma=D ∂ γ = D and let N N N be as in Proposition 3.2(b). The functional η ↦ N ∫ γ η \eta\mapsto N\int_\gamma\eta η ↦ N ∫ γ η on W t 1 , 0 W_t^{1,0} W t 1 , 0 depends continuously on t t t (see the complex surface D × X \mathbb{D}\times X D × X in § 3.2 ), and by Proposition 3.2(b) it equals ι t ( λ ( t ) ) \iota_t(\lambda(t)) ι t ( λ ( t )) with λ ( t ) ∈ Λ \lambda(t)\in\Lambda λ ( t ) ∈ Λ . So λ ( t ) = ι t − 1 ( N ∫ γ ∣ W t 1 , 0 ) \lambda(t)=\iota_t^{-1}(N\int_\gamma|_{W_t^{1,0}}) λ ( t ) = ι t − 1 ( N ∫ γ ∣ W t 1 , 0 ) is a continuous function on the disc with values in the discrete set Λ \Lambda Λ , hence constant, λ ( t ) = λ \lambda(t)=\lambda λ ( t ) = λ . Let λ ~ ∈ H 1 ( X , Q ) \tilde{\lambda}\in H_1(X,\mathbb{Q}) λ ~ ∈ H 1 ( X , Q ) map to λ \lambda λ and let γ ′ = γ − λ ~ / N \gamma'=\gamma-\tilde{\lambda}/N γ ′ = γ − λ ~ / N , a rational relative class with ∂ γ ′ = D \partial\gamma'=D ∂ γ ′ = D and ∫ γ ′ η = 0 \int_{\gamma'}\eta=0 ∫ γ ′ η = 0 for every η ∈ W t 1 , 0 \eta\in W_t^{1,0} η ∈ W t 1 , 0 and every t t t . The difference γ ′ − γ ^ \gamma'-\hat{\gamma} γ ′ − γ ^ lies in H 1 ( X , Q ) H_1(X,\mathbb{Q}) H 1 ( X , Q ) and vanishes on W 1 , 0 W^{1,0} W 1 , 0 ; being real, it vanishes on W 0 , 1 W^{0,1} W 0 , 1 , hence on W ⊗ C W\otimes\mathbb{C} W ⊗ C . As W W W and γ ′ − γ ^ \gamma'-\hat{\gamma} γ ′ − γ ^ do not depend on t t t , γ ′ − γ ^ \gamma'-\hat{\gamma} γ ′ − γ ^ vanishes on W t 1 , 0 W_t^{1,0} W t 1 , 0 for every t t t , and so does γ ^ \hat{\gamma} γ ^ . □
In this proof the rational relative class γ ′ = γ − λ ~ / N \gamma'=\gamma-\tilde{\lambda}/N γ ′ = γ − λ ~ / N vanishes on W t 1 , 0 W_t^{1,0} W t 1 , 0 for every t t t ; for affine invariant manifolds the corresponding flatness is shown in the proof of [8 ] , Theorem 5.2.
The first-order formula.
Lemma 3.6. Let ( X , ω ) (X,\omega) ( X , ω ) be a translation surface, X t X_t X t (t ∈ D t\in\mathbb{D} t ∈ D ) as in § 3.2 , S ⊂ X S\subset X S ⊂ X a finite set, and γ ^ ∈ H 1 ( X , S ; R ) \hat{\gamma}\in H_1(X,S;\mathbb{R}) γ ^ ∈ H 1 ( X , S ; R ) a class with ∫ γ ^ η = 0 \int_{\hat{\gamma}}\eta=0 ∫ γ ^ η = 0 for every holomorphic form η \eta η on X 0 X_0 X 0 ; let D = ∂ γ ^ D=\partial\hat{\gamma} D = ∂ γ ^ . Let η t \eta_t η t be a holomorphic family of forms, holomorphic on X t X_t X t . On X 0 X_0 X 0 write
η 0 ω ω ˉ = ∂ ˉ G + H \frac{\eta_0}{\omega}\,\bar{\omega}=\bar{\partial}G+H ω η 0 ω ˉ = ∂ ˉ G + H with H H H antiholomorphic and G G G a continuous function. Then
d d t ∣ t = 0 ∫ γ ^ η t = G ( D ) : = ∑ P D ( P ) G ( P ) . \left.\frac{d}{dt}\right|_{t=0}\int_{\hat{\gamma}}\eta_t=G(D):=\sum_P D(P)G(P). d t d t = 0 ∫ γ ^ η t = G ( D ) := P ∑ D ( P ) G ( P ) .
Such G G G and H H H exist: the left side is a bounded ( 0 , 1 ) (0,1) ( 0 , 1 ) -form, smooth off the zeros of ω \omega ω (at a zero of order k k k , η 0 / ω \eta_0/\omega η 0 / ω has a pole of order at most k k k and ω ˉ \bar{\omega} ω ˉ vanishes to order k k k ), H H H is its antiholomorphic part in the Hodge decomposition, and G G G is smooth where the left side is and Hölder continuous everywhere, by elliptic regularity; G G G is unique up to an additive constant, which does not change G ( D ) G(D) G ( D ) . The derivative depends on the family only through η 0 \eta_0 η 0 , since two families with the same η 0 \eta_0 η 0 differ by t t t times a holomorphic family plus O ( t 2 ) O(t^2) O ( t 2 ) , and γ ^ \hat{\gamma} γ ^ vanishes on holomorphic forms on X 0 X_0 X 0 . Kahn and Wright [14 Theorem 1.1] give a formula of the same kind for the derivative of the map from a stratum to the moduli space of curves, as an integral against a quadratic differential of the ( 0 , 1 ) (0,1) ( 0 , 1 ) -part of a closed form, divided by ω \omega ω , plus terms at the zeros.
Proof. On the complex surface D × X \mathbb{D}\times X D × X of § 3.2 the family η t \eta_t η t is a holomorphic relative 1 1 1 -form; let Φ ( t , x ) \Phi(t,x) Φ ( t , x ) be a local primitive along the fibres, a holomorphic function determined up to adding a holomorphic function of t t t . Let Φ ˙ = ∂ t Φ ∣ t = 0 \dot{\Phi}=\partial_t\Phi|_{t=0} Φ ˙ = ∂ t Φ ∣ t = 0 at fixed x ∈ X x\in X x ∈ X : it is defined locally up to a constant, d Φ ˙ d\dot{\Phi} d Φ ˙ is a closed 1 1 1 -form on X X X off the zeros, and for a path γ \gamma γ in X X X ,
d d t ∣ 0 ∫ γ η t = ∫ γ d Φ ˙ . \left.\frac{d}{dt}\right|_{0}\int_{\gamma}\eta_t=\int_{\gamma}d\dot{\Phi}. d t d 0 ∫ γ η t = ∫ γ d Φ ˙ .
Away from the zeros, Φ = F ( t , w + t w ˉ ) \Phi=F(t,w+t\bar{w}) Φ = F ( t , w + t w ˉ ) with F F F holomorphic in two variables, so Φ ˙ = F t ( 0 , w ) + F v ( 0 , w ) w ˉ \dot{\Phi}=F_t(0,w)+F_v(0,w)\bar{w} Φ ˙ = F t ( 0 , w ) + F v ( 0 , w ) w ˉ and ∂ ˉ Φ ˙ = F v ( 0 , w ) d w ˉ = ( η 0 / ω ) ω ˉ \bar{\partial}\dot{\Phi}=F_v(0,w)d\bar{w}=(\eta_0/\omega)\bar{\omega} ∂ ˉ Φ ˙ = F v ( 0 , w ) d w ˉ = ( η 0 / ω ) ω ˉ , as η 0 = F v ( 0 , w ) d w \eta_0=F_v(0,w)dw η 0 = F v ( 0 , w ) d w and ω = d w \omega=dw ω = d w . Near a zero of order k k k , Φ = F ~ ( t , u t ) \Phi=\widetilde{F}(t,u_t) Φ = F ( t , u t ) and Φ ˙ = F ~ t ( 0 , u ) + F ~ u ( 0 , u ) u ˉ k + 1 ( k + 1 ) u k \dot{\Phi}=\widetilde{F}_t(0,u)+\widetilde{F}_u(0,u)\frac{\bar{u}^{k+1}}{(k+1)u^k} Φ ˙ = F t ( 0 , u ) + F u ( 0 , u ) ( k + 1 ) u k u ˉ k + 1 , which is continuous at u = 0 u=0 u = 0 . Now η ~ = d Φ ˙ − d G − H \widetilde{\eta}=d\dot{\Phi}-dG-H η = d Φ ˙ − d G − H is a form of type ( 1 , 0 ) (1,0) ( 1 , 0 ) off a finite set, since its ( 0 , 1 ) (0,1) ( 0 , 1 ) -part is ∂ ˉ Φ ˙ − ∂ ˉ G − H = 0 \bar{\partial}\dot{\Phi}-\bar{\partial}G-H=0 ∂ ˉ Φ ˙ − ∂ ˉ G − H = 0 , and it is closed, since d H = 0 dH=0 d H = 0 ; so it is holomorphic there. Near each point of the finite set it is the differential of the continuous function Φ ˙ − G − \dot{\Phi}-G- Φ ˙ − G − (a local primitive of H H H ), which is holomorphic off the point and hence across it; so η ~ \widetilde{\eta} η is a holomorphic form on X 0 X_0 X 0 . Hence d Φ ˙ = d G + H + η ~ d\dot{\Phi}=dG+H+\widetilde{\eta} d Φ ˙ = d G + H + η , and
d d t ∣ 0 ∫ γ ^ η t = ∫ γ ^ d Φ ˙ = G ( D ) + ∫ γ ^ H + ∫ γ ^ η ~ = G ( D ) , \left.\frac{d}{dt}\right|_{0}\int_{\hat{\gamma}}\eta_t=\int_{\hat{\gamma}}d\dot{\Phi}=G(D)+\int_{\hat{\gamma}}H+\int_{\hat{\gamma}}\widetilde{\eta}=G(D), d t d 0 ∫ γ ^ η t = ∫ γ ^ d Φ ˙ = G ( D ) + ∫ γ ^ H + ∫ γ ^ η = G ( D ) ,
the last two integrals vanishing because γ ^ \hat{\gamma} γ ^ is real and vanishes on holomorphic forms on X 0 X_0 X 0 .
Corollary 3.7. Let Δ \Delta Δ be a lattice triangle whose unfolding ( X , ω ) (X,\omega) ( X , ω ) has genus at least two, b b b a unit modulo n n n with H b 1 , 0 ≠ 0 H_b^{1,0}\ne0 H b 1 , 0 = 0 , and write ( η b / ω ) ω ˉ = ∂ ˉ G b + H b (\eta_b/\omega)\bar{\omega}=\bar{\partial}G_b+H_b ( η b / ω ) ω ˉ = ∂ ˉ G b + H b as in Lemma 3.6 . Then G b ( D ) = 0 G_b(D)=0 G b ( D ) = 0 for every divisor D D D of degree 0 0 0 supported on the zeros of ω \omega ω .
Proof. By Lemma 3.4 there is γ ^ ∈ H 1 ( X , supp D ; Q ) \hat{\gamma}\in H_1(X,\operatorname{supp}D;\mathbb{Q}) γ ^ ∈ H 1 ( X , supp D ; Q ) with ∂ γ ^ = D \partial\hat{\gamma}=D ∂ γ ^ = D vanishing on holomorphic forms on X 0 X_0 X 0 . By Lemma 3.3 , η b ∈ W 0 1 , 0 \eta_b\in W_0^{1,0} η b ∈ W 0 1 , 0 , and as W t 1 , 0 W_t^{1,0} W t 1 , 0 is a holomorphic subbundle (Proposition 3.2(a)) there is a holomorphic family η t ∈ W t 1 , 0 \eta_t\in W_t^{1,0} η t ∈ W t 1 , 0 with η 0 = η b \eta_0=\eta_b η 0 = η b . By Lemma 3.5 , ∫ γ ^ η t = 0 \int_{\hat{\gamma}}\eta_t=0 ∫ γ ^ η t = 0 for all t ∈ D t\in\mathbb{D} t ∈ D . Differentiating at t = 0 t=0 t = 0 (Lemma 3.6 ) gives G b ( D ) = 0 G_b(D)=0 G b ( D ) = 0 .
Lemma 3.8. Let ϖ \varpi ϖ be a bounded ( 0 , 1 ) (0,1) ( 0 , 1 ) -form on a compact Riemann surface X X X , smooth off a finite set, with ϖ = ∂ ˉ G \varpi=\bar{\partial}G ϖ = ∂ ˉ G for a continuous function G G G , and let ξ \xi ξ be a meromorphic 1 1 1 -form on X X X with at most simple poles. Then
∫ X ϖ ∧ ξ = − 2 π i ∑ Q G ( Q ) Res Q ξ . \int_X\varpi\wedge\xi=-2\pi i\sum_Q G(Q)\operatorname{Res}_Q\xi. ∫ X ϖ ∧ ξ = − 2 π i Q ∑ G ( Q ) Res Q ξ . Proof. Remove a coordinate disc of radius ε \varepsilon ε about each pole of ξ \xi ξ and each point where ϖ \varpi ϖ is not smooth. On the rest d ( G ξ ) = ∂ ˉ G ∧ ξ = ϖ ∧ ξ d(G\xi)=\bar{\partial}G\wedge\xi=\varpi\wedge\xi d ( G ξ ) = ∂ ˉ G ∧ ξ = ϖ ∧ ξ , so by Stokes’s theorem ∫ ϖ ∧ ξ = − ∑ ∮ G ξ \int\varpi\wedge\xi=-\sum\oint G\xi ∫ ϖ ∧ ξ = − ∑ ∮ G ξ over the circles, oriented counterclockwise. As ε → 0 \varepsilon\to0 ε → 0 the left side tends to ∫ X ϖ ∧ ξ \int_X\varpi\wedge\xi ∫ X ϖ ∧ ξ , since ϖ \varpi ϖ is bounded and ξ \xi ξ has at most simple poles, and ∮ G ξ → 2 π i G ( Q ) Res Q ξ \oint G\xi\to2\pi iG(Q)\operatorname{Res}_Q\xi ∮ G ξ → 2 π i G ( Q ) Res Q ξ by continuity of G G G .
Both proofs below rest on one computation. As T ∗ ω = ζ ω T^*\omega=\zeta\omega T ∗ ω = ζ ω and T ∗ ω ˉ = ζ − 1 ω ˉ T^*\bar{\omega}=\zeta^{-1}\bar{\omega} T ∗ ω ˉ = ζ − 1 ω ˉ , a form η b \eta_b η b of character b b b gives
T ∗ ( η b ω ω ˉ ) = ζ b − 2 η b ω ω ˉ . T^*\left(\frac{\eta_b}{\omega}\bar{\omega}\right)=\zeta^{b-2}\frac{\eta_b}{\omega}\bar{\omega}. T ∗ ( ω η b ω ˉ ) = ζ b − 2 ω η b ω ˉ .
So the antiholomorphic part H b H_b H b is the conjugate of a holomorphic form of the complementary character b ′ = 2 − b b'=2-b b ′ = 2 − b , G b G_b G b can be taken with G b ∘ T = ζ − b ′ G b G_b\circ T=\zeta^{-b'}G_b G b ∘ T = ζ − b ′ G b , and for a meromorphic form ξ \xi ξ of character b ′ b' b ′ with at most simple poles the form ( η b / ω ) ω ˉ ∧ ξ (\eta_b/\omega)\bar{\omega}\wedge\xi ( η b / ω ) ω ˉ ∧ ξ is T T T -invariant, so that its integral over X X X is n n n times an integral over the plane. When H b = 0 H_b=0 H b = 0 , Lemma 3.8 expresses the same integral through the values of G b G_b G b at the poles of ξ \xi ξ . In part (a), b = 2 b=2 b = 2 , a unit as n n n is odd; then b ′ = 0 b'=0 b ′ = 0 , H 2 = 0 H_2=0 H 2 = 0 as H 0 1 = 0 H^1_0=0 H 0 1 = 0 , and ξ \xi ξ is d z / z dz/z d z / z or d z / ( z − 1 ) dz/(z-1) d z / ( z − 1 ) . In part (b), b ≡ 2 ( m o d n / g ) b\equiv2\pmod{n/g} b ≡ 2 ( mod n / g ) and b ≢ 2 ( m o d n ) b\not\equiv2\pmod n b ≡ 2 ( mod n ) say that b ′ x ≡ 0 b'x\equiv0 b ′ x ≡ 0 and b ′ ≢ 0 ( m o d n ) b'\not\equiv0\pmod n b ′ ≡ 0 ( mod n ) . So H b = 0 H_b=0 H b = 0 by Lemma 3.1 (c); ξ \xi ξ can be given a simple pole at each of the g g g points above x x x and no other pole; and ζ − b ′ ≠ 1 \zeta^{-b'}\ne1 ζ − b ′ = 1 , so that, if Δ \Delta Δ is a lattice triangle, Corollary 3.7 applied to D = T Q − Q D=TQ-Q D = T Q − Q for a point Q Q Q above x x x forces G b G_b G b to vanish at all the points above x x x . Corollary 3.7 concerns only divisors supported on the zeros of ω \omega ω , and the points above a vertex x x x are zeros exactly when x ∤ n x\nmid n x ∤ n (§ 3.1 ): hence the hypothesis x ∤ n x\nmid n x ∤ n in part (b).
Proof of Theorem 1.2(a)
The two integrals of the next lemma are the cases ( ϵ 1 , ϵ 2 ) = ( 0 , 1 ) (\epsilon_1,\epsilon_2)=(0,1) ( ϵ 1 , ϵ 2 ) = ( 0 , 1 ) and ( 1 , 0 ) (1,0) ( 1 , 0 ) , with q 1 / k = θ 0 q_1/k=\theta_0 q 1 / k = θ 0 and q 2 / k = θ 1 q_2/k=\theta_1 q 2 / k = θ 1 , of the integral of [22 Lemma 7.15] , which arises there from the pairing of holomorphic forms of complementary characters in the proof of their Proposition 7.3. It is proved non-zero there, and the argument, a comparison of the integrand at symmetric points, also gives the signs.
Lemma 3.9. Let θ 0 , θ 1 > 0 \theta_0,\theta_1>0 θ 0 , θ 1 > 0 with θ 0 + θ 1 < 1 / 2 \theta_0+\theta_1<1/2 θ 0 + θ 1 < 1/2 , and K ( z ) = ∣ z ∣ 2 θ 0 − 2 ∣ z − 1 ∣ 2 θ 1 − 2 K(z)=|z|^{2\theta_0-2}|z-1|^{2\theta_1-2} K ( z ) = ∣ z ∣ 2 θ 0 − 2 ∣ z − 1 ∣ 2 θ 1 − 2 . The integrals of K ⋅ ( z − 1 ) K\cdot(z-1) K ⋅ ( z − 1 ) and K ⋅ z K\cdot z K ⋅ z over C \mathbb{C} C with respect to the area form d 2 z d^2z d 2 z converge absolutely, and
∫ C K ⋅ ( z − 1 ) d 2 z < 0 , ∫ C K ⋅ z d 2 z > 0. \int_{\mathbb{C}} K\cdot(z-1)\,d^2z<0,\qquad\int_{\mathbb{C}}K\cdot z\,d^2z>0. ∫ C K ⋅ ( z − 1 ) d 2 z < 0 , ∫ C K ⋅ z d 2 z > 0.
Proof. Near 0 and 1 the exponents 2 θ i − 2 2\theta_i-2 2 θ i − 2 exceed − 2 -2 − 2 , and at infinity the integrands are O ( ∣ z ∣ 2 θ 0 + 2 θ 1 − 3 ) O(|z|^{2\theta_0+2\theta_1-3}) O ( ∣ z ∣ 2 θ 0 + 2 θ 1 − 3 ) with 2 θ 0 + 2 θ 1 − 3 < − 2 2\theta_0+2\theta_1-3<-2 2 θ 0 + 2 θ 1 − 3 < − 2 . Write I 1 I_1 I 1 and I 2 I_2 I 2 for the integrals of K ⋅ ( z − 1 ) K\cdot(z-1) K ⋅ ( z − 1 ) and K ⋅ z K\cdot z K ⋅ z . As z − ( z − 1 ) = 1 z-(z-1)=1 z − ( z − 1 ) = 1 , I 2 − I 1 = ∫ C K d 2 z > 0 I_2-I_1=\int_{\mathbb{C}}K\,d^2z>0 I 2 − I 1 = ∫ C K d 2 z > 0 . Let f = ∣ z ∣ 2 θ 0 ∣ z − 1 ∣ 2 θ 1 / ( z ˉ ( z ˉ − 1 ) ) f=|z|^{2\theta_0}|z-1|^{2\theta_1}/(\bar{z}(\bar{z}-1)) f = ∣ z ∣ 2 θ 0 ∣ z − 1 ∣ 2 θ 1 / ( z ˉ ( z ˉ − 1 )) , the function that appears in the proof of Theorem 1.2 (a) below. Off { 0 , 1 } \{0,1\} { 0 , 1 } ,
∂ f ∂ z = f ⋅ ( θ 0 z + θ 1 z − 1 ) = K ⋅ ( θ 0 ( z − 1 ) + θ 1 z ) . \frac{\partial f}{\partial z}=f\cdot\left(\frac{\theta_0}{z}+\frac{\theta_1}{z-1}\right)=K\cdot\left(\theta_0(z-1)+\theta_1z\right). ∂ z ∂ f = f ⋅ ( z θ 0 + z − 1 θ 1 ) = K ⋅ ( θ 0 ( z − 1 ) + θ 1 z ) .
and d ( f d z ˉ ) = − 2 i ( ∂ f / ∂ z ) d 2 z d(f\,d\bar{z})=-2i(\partial f/\partial z)d^2z d ( f d z ˉ ) = − 2 i ( ∂ f / ∂ z ) d 2 z . As ∣ f ∣ = ∣ z ∣ 2 θ 0 − 1 ∣ z − 1 ∣ 2 θ 1 − 1 |f|=|z|^{2\theta_0-1}|z-1|^{2\theta_1-1} ∣ f ∣ = ∣ z ∣ 2 θ 0 − 1 ∣ z − 1 ∣ 2 θ 1 − 1 , the integrals of f d z ˉ f\,d\bar{z} f d z ˉ over the circles ∣ z ∣ = ϵ |z|=\epsilon ∣ z ∣ = ϵ , ∣ z − 1 ∣ = ϵ |z-1|=\epsilon ∣ z − 1∣ = ϵ and ∣ z ∣ = R |z|=R ∣ z ∣ = R are O ( ϵ 2 θ 0 ) O(\epsilon^{2\theta_0}) O ( ϵ 2 θ 0 ) , O ( ϵ 2 θ 1 ) O(\epsilon^{2\theta_1}) O ( ϵ 2 θ 1 ) and O ( R 2 θ 0 + 2 θ 1 − 1 ) O(R^{2\theta_0+2\theta_1-1}) O ( R 2 θ 0 + 2 θ 1 − 1 ) . So Stokes’s theorem on { ϵ < ∣ z ∣ < R , ∣ z − 1 ∣ > ϵ } \{\epsilon<|z|<R,\ |z-1|>\epsilon\} { ϵ < ∣ z ∣ < R , ∣ z − 1∣ > ϵ } and absolute convergence give, as ϵ → 0 \epsilon\to0 ϵ → 0 and R → ∞ R\to\infty R → ∞ ,
θ 0 I 1 + θ 1 I 2 = 0. \theta_0I_1+\theta_1I_2=0. θ 0 I 1 + θ 1 I 2 = 0.
Hence
I 1 = − θ 1 θ 0 + θ 1 ( I 2 − I 1 ) < 0 , I 2 = θ 0 θ 0 + θ 1 ( I 2 − I 1 ) > 0. I_1=-\frac{\theta_1}{\theta_0+\theta_1}(I_2-I_1)<0,\qquad I_2=\frac{\theta_0}{\theta_0+\theta_1}(I_2-I_1)>0. I 1 = − θ 0 + θ 1 θ 1 ( I 2 − I 1 ) < 0 , I 2 = θ 0 + θ 1 θ 0 ( I 2 − I 1 ) > 0.
Proof of Theorem 1.2(a). Suppose that Δ \Delta Δ is a lattice triangle. Place the obtuse vertex at ∞ \infty ∞ : ( x 0 , x 1 , x ∞ ) = ( p , q , r ) (x_0,x_1,x_\infty)=(p,q,r) ( x 0 , x 1 , x ∞ ) = ( p , q , r ) , so that θ 0 + θ 1 < 1 / 2 \theta_0+\theta_1<1/2 θ 0 + θ 1 < 1/2 .
The form. Let η 2 = y q d z / ( z ( z − 1 ) ) \eta_2=y^q\,dz/(z(z-1)) η 2 = y q d z / ( z ( z − 1 )) , of character 2. In Lemma 3.1 (a), it has i = j = − 1 i=j=-1 i = j = − 1 , e 0 = 2 θ 0 − 1 > − 1 e_0=2\theta_0-1>-1 e 0 = 2 θ 0 − 1 > − 1 , e 1 = 2 θ 1 − 1 > − 1 e_1=2\theta_1-1>-1 e 1 = 2 θ 1 − 1 > − 1 and e 0 + e 1 = 2 ( θ 0 + θ 1 ) − 2 < − 1 e_0+e_1=2(\theta_0+\theta_1)-2<-1 e 0 + e 1 = 2 ( θ 0 + θ 1 ) − 2 < − 1 , the last exactly because r > n / 2 r>n/2 r > n /2 ; so η 2 \eta_2 η 2 is holomorphic. As n n n is odd, 2 is a unit and 2 x ≠ 0 ( m o d n ) 2x\ne0\pmod n 2 x = 0 ( mod n ) for x = p , q , r x=p,q,r x = p , q , r , so H 2 1 , 0 = C η 2 H^{1,0}_2=\mathbb{C}\eta_2 H 2 1 , 0 = C η 2 by Lemma 3.1 (b). The function. On X X X ,
η 2 ω ω ˉ = y y ˉ d z ˉ z ˉ ( z ˉ − 1 ) = f d z ˉ , f = ∣ z ∣ 2 θ 0 ∣ z − 1 ∣ 2 θ 1 z ˉ ( z ˉ − 1 ) , \frac{\eta_2}{\omega}\,\bar{\omega}=y\bar{y}\frac{d\bar{z}}{\bar{z}(\bar{z}-1)}=f\,d\bar{z},\qquad f=\frac{|z|^{2\theta_0}|z-1|^{2\theta_1}}{\bar{z}(\bar{z}-1)}, ω η 2 ω ˉ = y y ˉ z ˉ ( z ˉ − 1 ) d z ˉ = f d z ˉ , f = z ˉ ( z ˉ − 1 ) ∣ z ∣ 2 θ 0 ∣ z − 1 ∣ 2 θ 1 ,
which is T T T -invariant. In its decomposition f d z ˉ = ∂ G + H f\,d\bar{z}=\partial G+H f d z ˉ = ∂ G + H the antiholomorphic form H H H is unique, hence T T T -invariant, and H ˉ ∈ H 0 1 , 0 = 0 \bar{H}\in H^{1,0}_0=0 H ˉ ∈ H 0 1 , 0 = 0 (Lemma 3.1 (c)); averaging G G G over ⟨ T ⟩ \langle T\rangle ⟨ T ⟩ we may take G G G to be T T T -invariant, that is, a continuous function of z ∈ P 1 z\in\mathbb{P}^1 z ∈ P 1 . Write G ( 0 ) G(0) G ( 0 ) , G ( 1 ) G(1) G ( 1 ) , G ( ∞ ) G(\infty) G ( ∞ ) for its values above the three vertices.
Its values. Apply Lemma 3.8 with ϖ = f d z ˉ \varpi=f\,d\bar{z} ϖ = f d z ˉ and ξ = d z / z \xi=dz/z ξ = d z / z . The form ξ \xi ξ has simple poles above 0 0 0 , with residue n / g 0 n/g_0 n / g 0 at each of the g 0 g_0 g 0 points, and above ∞ \infty ∞ , with residue − n / g ∞ -n/g_\infty − n / g ∞ at each of the g ∞ g_\infty g ∞ points, and it is holomorphic above 1 1 1 . The integrand ϖ ∧ ξ = ( f / z ) d z ˉ ∧ d z \varpi\wedge\xi=(f/z)d\bar{z}\wedge dz ϖ ∧ ξ = ( f / z ) d z ˉ ∧ d z is T T T -invariant, f / z = K ⋅ ( z − 1 ) f/z=\mathcal{K}\cdot(z-1) f / z = K ⋅ ( z − 1 ) with K \mathcal{K} K as in Lemma 3.9 , and d z ˉ ∧ d z = 2 i d 2 z d\bar{z}\wedge dz=2i\,d^2z d z ˉ ∧ d z = 2 i d 2 z ; so
2 i n ∫ C K ⋅ ( z − 1 ) d 2 z = − 2 π i n ( G ( 0 ) − G ( ∞ ) ) , 2in\int_C\mathcal{K}\cdot(z-1)d^2z=-2\pi in\bigl(G(0)-G(\infty)\bigr), 2 in ∫ C K ⋅ ( z − 1 ) d 2 z = − 2 π in ( G ( 0 ) − G ( ∞ ) ) ,
that is,
G ( 0 ) − G ( ∞ ) = − 1 π ∫ C K ⋅ ( z − 1 ) d 2 z > 0. G(0)-G(\infty)=-\frac{1}{\pi}\int_C\mathcal{K}\cdot(z-1)d^2z>0. G ( 0 ) − G ( ∞ ) = − π 1 ∫ C K ⋅ ( z − 1 ) d 2 z > 0.
With ξ = d z / ( z − 1 ) \xi=dz/(z-1) ξ = d z / ( z − 1 ) and f / ( z − 1 ) = K ⋅ z f/(z-1)=\mathcal{K}\cdot z f / ( z − 1 ) = K ⋅ z the same computation gives G ( 1 ) − G ( ∞ ) = − 1 π ∫ C K ⋅ z d 2 z < 0 G(1)-G(\infty)=-\frac{1}{\pi}\int_C\mathcal{K}\cdot z\,d^2z<0 G ( 1 ) − G ( ∞ ) = − π 1 ∫ C K ⋅ z d 2 z < 0 (Lemma 3.9 ).
The contradiction. The points above ∞ \infty ∞ are zeros of ω \omega ω , since r ∤ n r\nmid n r ∤ n . By hypothesis the points above 0 0 0 , or those above 1 1 1 , are zeros too, say those above i ∈ { 0 , 1 } i\in\{0,1\} i ∈ { 0 , 1 } . The divisor D = g ∞ Σ i − g i Σ ∞ D=g_\infty\Sigma_i-g_i\Sigma_\infty D = g ∞ Σ i − g i Σ ∞ has degree $0 and is supported on zeros of ω \omega ω ; as ω \omega ω has zeros, X X X has genus at least two, and Corollary 3.7 gives 0 = G ( D ) = g i g ∞ ( G ( i ) − G ( ∞ ) ) 0=G(D)=g_ig_\infty\bigl(G(i)-G(\infty)\bigr) 0 = G ( D ) = g i g ∞ ( G ( i ) − G ( ∞ ) ) , which contradicts the inequalities above. □
That the triangle is obtuse is needed for the statement and not only for the method: the acute triangles ( 2 , 3 , 4 ) / 9 (2,3,4)/9 ( 2 , 3 , 4 ) /9 and ( 1 , 2 , 2 ) / 5 (1,2,2)/5 ( 1 , 2 , 2 ) /5 are lattice triangles [15 Theorems 1 and 2] , with n n n odd and a numerator not dividing n n n .
Example 3.10. The smallest triangle to which Theorem 1.2 (a) applies and on which the criterion of Mirzakhani and Wright is blind is ( 1 , 2 , 4 ) / 7 (1,2,4)/7 ( 1 , 2 , 4 ) /7 , the member s = 1 s=1 s = 1 , n = 7 n=7 n = 7 of the family A A A of § 4 . Here ω \omega ω has a simple zero above 1 1 1 and a zero of order 3 3 3 above ∞ \infty ∞ , X X X has genus 3 3 3 , and with D = P 1 − P ∞ D=P_1-P_\infty D = P 1 − P ∞ the derivative of ∫ γ t η t \int_{\gamma_t}\eta_t ∫ γ t η t at t = 0 t=0 t = 0 is
G ( 1 ) − G ( ∞ ) = − 1 π ∫ C K ⋅ z d 2 z ≠ 0 G(1)-G(\infty)=-\frac{1}{\pi}\int_C\mathcal{K}\cdot z\,d^2z\ne0 G ( 1 ) − G ( ∞ ) = − π 1 ∫ C K ⋅ z d 2 z = 0
(Lemma 3.9 ).
Proof of Theorem 1.2(b)
We use the complex beta integral in the form given by Neretin [25 (1.4)] , who attributes it to Gel’fand, Graev and Vilenkin [10 Section II.3.7] and Gel’fand, Graev and Retakh [9 Section 1.6] . With z u ∣ u ′ : = z u z ˉ u ′ z^{u|u'}:=z^u\bar{z}^{u'} z u ∣ u ′ := z u z ˉ u ′ , which is single-valued when u − u ′ ∈ Z u-u'\in\mathbb{Z} u − u ′ ∈ Z , it reads
B C ( α ∣ α ′ , β ∣ β ′ ) : = 1 π ∫ C z α − 1 ∣ α ′ − 1 ( 1 − z ) β − 1 ∣ β ′ − 1 d 2 z = Γ ( α ) Γ ( β ) Γ ( α + β ) ⋅ Γ ( 1 − α ′ − β ′ ) Γ ( 1 − α ′ ) Γ ( 1 − β ′ ) . (3) B^{\mathbb{C}}(\alpha|\alpha',\beta|\beta'):=\frac{1}{\pi}\int_C z^{\alpha-1|\alpha'-1}(1-z)^{\beta-1|\beta'-1}\,d^2z
=\frac{\Gamma(\alpha)\Gamma(\beta)}{\Gamma(\alpha+\beta)}\cdot\frac{\Gamma(1-\alpha'-\beta')}{\Gamma(1-\alpha')\Gamma(1-\beta')}. \tag*{(3)} B C ( α ∣ α ′ , β ∣ β ′ ) := π 1 ∫ C z α − 1∣ α ′ − 1 ( 1 − z ) β − 1∣ β ′ − 1 d 2 z = Γ ( α + β ) Γ ( α ) Γ ( β ) ⋅ Γ ( 1 − α ′ ) Γ ( 1 − β ′ ) Γ ( 1 − α ′ − β ′ ) . (3)
Here z u ∣ u ′ = ∣ z ∣ 2 u ′ z u − u ′ z^{u|u'}=|z|^{2u'}z^{u-u'} z u ∣ u ′ = ∣ z ∣ 2 u ′ z u − u ′ , and the integral converges absolutely exactly when Re ( α + α ′ ) > 0 \operatorname{Re}(\alpha+\alpha')>0 Re ( α + α ′ ) > 0 , Re ( β + β ′ ) > 0 \operatorname{Re}(\beta+\beta')>0 Re ( β + β ′ ) > 0 and Re ( α + α ′ + β + β ′ ) < 2 \operatorname{Re}(\alpha+\alpha'+\beta+\beta')<2 Re ( α + α ′ + β + β ′ ) < 2 ; we use (3) only in this region.
Proof of Theorem 1.2(b). Suppose that Δ \Delta Δ is a lattice triangle. Place the vertex x x x at 0 0 0 : x 0 = x x_0=x x 0 = x , g 0 = g g_0=g g 0 = g , and x 1 , x ∞ x_1,x_\infty x 1 , x ∞ are the other two numerators in either order. The g 0 g_0 g 0 points Q j = T j Q 0 Q_j=T^jQ_0 Q j = T j Q 0 (j ∈ Z / g 0 Z j\in\mathbb{Z}/g_0\mathbb{Z} j ∈ Z / g 0 Z ) above 0 0 0 are zeros of ω \omega ω , as x ∤ n x\nmid n x ∤ n . Fix an integer representing b b b and let b ′ = 2 − b b'=2-b b ′ = 2 − b . Then n ∣ b ′ n\mid b' n ∣ b ′ and ( n / g 0 ) ∣ b ′ (n/g_0)\mid b' ( n / g 0 ) ∣ b ′ , so b ′ θ 0 ∈ Z b'\theta_0\in\mathbb{Z} b ′ θ 0 ∈ Z ; and b ′ θ 1 ∈ Z b'\theta_1\in\mathbb{Z} b ′ θ 1 ∈ Z , since otherwise b ′ θ ∞ ∈ Z b'\theta_\infty\in\mathbb{Z} b ′ θ ∞ ∈ Z too, and then n n n divides b ′ x 0 b'x_0 b ′ x 0 , b ′ x 1 b'x_1 b ′ x 1 , b ′ x ∞ b'x_\infty b ′ x ∞ , hence b ′ b' b ′ , as gcd ( p , q , r ) = 1 \gcd(p,q,r)=1 g cd( p , q , r ) = 1 . The form and the function. As b b b is a unit, b x i ≢ 0 ( m o d n ) b x_i \not\equiv0 \pmod{n} b x i ≡ 0 ( mod n ) for each i i i , and with [ b p ] n + [ b q ] n + [ b r ] n = n [bp]_n+[bq]_n+[br]_n=n [ b p ] n + [ b q ] n + [ b r ] n = n , Lemma 3.1(b) gives H b 1 , 0 = C η b H_b^{1,0}=C\eta_b H b 1 , 0 = C η b , η b = y b ′ θ 0 ( z − 1 ) b ′ θ 1 d z \eta_b=y^{b'\theta_0}(z-1)^{b'\theta_1}\mathrm{d}z η b = y b ′ θ 0 ( z − 1 ) b ′ θ 1 d z , where a 0 = { b θ 0 } = b θ 0 + i b + 1 a_0=\{b\theta_0\}=b\theta_0+i_b+1 a 0 = { b θ 0 } = b θ 0 + i b + 1 and a 1 = { b θ 1 } = b θ 1 + j b + 1 a_1=\{b\theta_1\}=b\theta_1+j_b+1 a 1 = { b θ 1 } = b θ 1 + j b + 1 lie in ( 0 , 1 ) (0,1) ( 0 , 1 ) and a 0 + a 1 < 1 a_0+a_1<1 a 0 + a 1 < 1 . The form ϖ = ( η b / ω ) ω ‾ \varpi=(\eta_b/\omega)\overline{\omega} ϖ = ( η b / ω ) ω satisfies T ∗ ϖ = ζ − b ′ ϖ = ζ − b ϖ T^*\varpi=\zeta^{-b'}\varpi=\zeta^{-b}\varpi T ∗ ϖ = ζ − b ′ ϖ = ζ − b ϖ . In its decomposition ϖ = ∂ G ‾ + H \varpi=\partial\overline{G}+H ϖ = ∂ G + H the antiholomorphic form H H H is unique, so T ∗ H = ζ − b ′ H T^*H=\zeta^{-b'}H T ∗ H = ζ − b ′ H and H ‾ ∈ H b ′ 1 , 0 \overline{H}\in H^{1,0}_{b'} H ∈ H b ′ 1 , 0 , which is 0 0 0 by Lemma 3.1(c), as b ′ x 0 ≡ 0 b'x_0\equiv0 b ′ x 0 ≡ 0 and b ′ ≢ 0 ( m o d n ) b'\not\equiv0\pmod n b ′ ≡ 0 ( mod n ) . So ϖ = ∂ ‾ G \varpi=\overline{\partial}G ϖ = ∂ G , and G ∘ T − ζ − b ′ G G\circ T-\zeta^{-b'}G G ∘ T − ζ − b ′ G is a constant; adding a constant to G G G (possible as ζ − b ′ ≠ 1 \zeta^{-b'}\ne1 ζ − b ′ = 1 ) we get G ∘ T = ζ − b ′ G G\circ T=\zeta^{-b'}G G ∘ T = ζ − b ′ G , so G ( Q j ) = ζ − b ′ j G ( Q 0 ) G(Q_j)=\zeta^{-b'j}G(Q_0) G ( Q j ) = ζ − b ′ j G ( Q 0 ) .
What the lattice property gives. The divisor D = Q 1 − Q 0 D=Q_1-Q_0 D = Q 1 − Q 0 has degree 0 0 0 and is supported on zeros of ω \omega ω ; as ω \omega ω has zeros, X X X has genus at least two, and Corollary 3.7 gives 0 = G ( D ) = ( ζ − b ′ − 1 ) G ( Q 0 ) 0=G(D)=(\zeta^{-b'}-1)G(Q_0) 0 = G ( D ) = ( ζ − b ′ − 1 ) G ( Q 0 ) , that is, G ( Q 0 ) = 0 G(Q_0)=0 G ( Q 0 ) = 0 .
The value of $G(Q_0). Let b 1 ′ = { b ′ θ 1 } ∈ ( 0 , 1 ) b'_1=\{b'\theta_1\}\in(0,1) b 1 ′ = { b ′ θ 1 } ∈ ( 0 , 1 ) and
ξ = y b ′ z − 1 − b ′ θ 0 ( z − 1 ) b 1 ′ − 1 − b ′ θ 1 d z . \xi=y^{b'}z^{-1-b'\theta_0}(z-1)^{b'_1-1-b'\theta_1}\mathrm{d}z. ξ = y b ′ z − 1 − b ′ θ 0 ( z − 1 ) b 1 ′ − 1 − b ′ θ 1 d z .
a meromorphic form of character b ′ b' b ′ (the exponents of z z z and z − 1 z-1 z − 1 are integers). By the order formulas of Lemma 3.1(a), with e 0 = − 1 e_0=-1 e 0 = − 1 , e 1 = b 1 ′ − 1 e_1=b'_1-1 e 1 = b 1 ′ − 1 and h = 1 h=1 h = 1 , it has order ( n / g 1 ) b 1 ′ − 1 ≥ 0 (n/g_1)b'_1-1\ge0 ( n / g 1 ) b 1 ′ − 1 ≥ 0 above 1 1 1 and ( n / g ∞ ) ( 1 − b 1 ′ ) − 1 ≥ 0 (n/g_\infty)(1-b'_1)-1\ge0 ( n / g ∞ ) ( 1 − b 1 ′ ) − 1 ≥ 0 above ∞ \infty ∞ (both ( n / g 1 ) b 1 ′ (n/g_1)b'_1 ( n / g 1 ) b 1 ′ and ( n / g ∞ ) ( 1 − b 1 ′ ) (n/g_\infty)(1-b'_1) ( n / g ∞ ) ( 1 − b 1 ′ ) are positive integers, as 1 − b 1 ′ ≡ b ′ θ ∞ ( m o d 1 ) 1-b'_1\equiv b'\theta_\infty\pmod1 1 − b 1 ′ ≡ b ′ θ ∞ ( mod 1 ) ), and a simple pole at each Q j Q_j Q j , where ξ = U d z / z \xi=U\,\mathrm{d}z/z ξ = U d z / z with U = y b ′ θ 0 ( z − 1 ) b 1 ′ − 1 − b ′ θ 1 U=y^{b'\theta_0}(z-1)^{b'_1-1-b'\theta_1} U = y b ′ θ 0 ( z − 1 ) b 1 ′ − 1 − b ′ θ 1 a function of order 0 0 0 ; so Res Q j ξ = ( n / g 0 ) U j \operatorname{Res}_{Q_j}\xi=(n/g_0)U_j Res Q j ξ = ( n / g 0 ) U j with U j = U ( Q j ) ≠ 0 U_j=U(Q_j)\ne0 U j = U ( Q j ) = 0 . Since T ∗ ξ = ζ − b ′ ξ T^*\xi=\zeta^{-b'}\xi T ∗ ξ = ζ − b ′ ξ and G ∘ T = ζ − b ′ G G\circ T=\zeta^{-b'}G G ∘ T = ζ − b ′ G , the product G ( Q j ) Res Q j ξ G(Q_j)\operatorname{Res}_{Q_j}\xi G ( Q j ) Res Q j ξ does not depend on j j j , and Lemma 3.8 gives
∫ X ϖ ∧ ξ = − 2 π i g 0 G ( Q 0 ) n g 0 U 0 = − 2 π i n U 0 G ( Q 0 ) . \int_X\varpi\wedge\xi=-2\pi i g_0G(Q_0)\frac{n}{g_0}U_0=-2\pi i nU_0G(Q_0). ∫ X ϖ ∧ ξ = − 2 π i g 0 G ( Q 0 ) g 0 n U 0 = − 2 π in U 0 G ( Q 0 ) .
On the other hand ϖ ∧ ξ \varpi\wedge\xi ϖ ∧ ξ is T T T -invariant, so ∫ X = n ∫ C \int_X=n\int_C ∫ X = n ∫ C . As b + b ′ = 2 b+b'=2 b + b ′ = 2 and y y ‾ = ∣ z ∣ 2 θ 0 ∣ z − 1 ∣ 2 θ 1 y\overline{y}=|z|^{2\theta_0}|z-1|^{2\theta_1} y y = ∣ z ∣ 2 θ 0 ∣ z − 1 ∣ 2 θ 1 ,
ϖ ∧ ξ = y y ‾ z k 0 ( z − 1 ) k 1 z ‾ ( z ‾ − 1 ) d z ‾ ∧ d z , \varpi\wedge\xi=y\overline{y}\frac{z^{k_0}(z-1)^{k_1}}{\overline{z}(\overline{z}-1)}\mathrm{d}\overline{z}\wedge\mathrm{d}z, ϖ ∧ ξ = y y z ( z − 1 ) z k 0 ( z − 1 ) k 1 d z ∧ d z ,
k 0 = i b − b ′ θ 0 = a 0 − 1 − 2 θ 0 , k 1 = j b + b 1 ′ − b ′ θ 1 = a 1 + b 1 ′ − 1 − 2 θ 1 , k_0=i_b-b'\theta_0=a_0-1-2\theta_0,\qquad k_1=j_b+b'_1-b'\theta_1=a_1+b'_1-1-2\theta_1, k 0 = i b − b ′ θ 0 = a 0 − 1 − 2 θ 0 , k 1 = j b + b 1 ′ − b ′ θ 1 = a 1 + b 1 ′ − 1 − 2 θ 1 ,
with k 0 , k 1 k_0,k_1 k 0 , k 1 integers; in the notation of (3) ,
ϖ ∧ ξ = ( − 1 ) k 1 + 1 z a 0 − θ 0 − 1 ∣ θ 0 − 1 ( 1 − z ) a 1 + b 1 ′ − θ 1 − 1 ∣ θ 1 − 1 ⋅ 2 i d 2 z . \varpi\wedge\xi=(-1)^{k_1+1}z^{a_0-\theta_0-1|\theta_0-1}(1-z)^{a_1+b'_1-\theta_1-1|\theta_1-1}\cdot2i\,\mathrm{d}^2z. ϖ ∧ ξ = ( − 1 ) k 1 + 1 z a 0 − θ 0 − 1∣ θ 0 − 1 ( 1 − z ) a 1 + b 1 ′ − θ 1 − 1∣ θ 1 − 1 ⋅ 2 i d 2 z .
Here α + α ′ = a 0 > 0 \alpha+\alpha'=a_0>0 α + α ′ = a 0 > 0 , β + β ′ = a 1 + b 1 ′ > 0 \beta+\beta'=a_1+b'_1>0 β + β ′ = a 1 + b 1 ′ > 0 and α + α ′ + β + β ′ = a 0 + a 1 + b 1 ′ < 2 \alpha+\alpha'+\beta+\beta'=a_0+a_1+b'_1<2 α + α ′ + β + β ′ = a 0 + a 1 + b 1 ′ < 2 , so the integral converges absolutely and (3) gives
∫ X ϖ ∧ ξ = 2 π i n ( − 1 ) k 1 + 1 κ , κ = Γ ( a 0 − θ 0 ) Γ ( a 1 + b 1 ′ − θ 1 ) Γ ( θ ∞ ) Γ ( a 0 + a 1 + b 1 ′ − θ 0 − θ 1 ) Γ ( 1 − θ 0 ) Γ ( 1 − θ 1 ) . \int_X\varpi\wedge\xi=2\pi i n(-1)^{k_1+1}\kappa,\qquad
\kappa=\frac{\Gamma(a_0-\theta_0)\Gamma(a_1+b'_1-\theta_1)\Gamma(\theta_\infty)}{\Gamma(a_0+a_1+b'_1-\theta_0-\theta_1)\Gamma(1-\theta_0)\Gamma(1-\theta_1)}. ∫ X ϖ ∧ ξ = 2 π in ( − 1 ) k 1 + 1 κ , κ = Γ ( a 0 + a 1 + b 1 ′ − θ 0 − θ 1 ) Γ ( 1 − θ 0 ) Γ ( 1 − θ 1 ) Γ ( a 0 − θ 0 ) Γ ( a 1 + b 1 ′ − θ 1 ) Γ ( θ ∞ ) .
Hence G ( Q 0 ) = ( − 1 ) k 1 κ / U 0 G(Q_0)=(-1)^{k_1}\kappa/U_0 G ( Q 0 ) = ( − 1 ) k 1 κ / U 0 .
It is not zero. Modulo 1 1 1 , a 0 − θ 0 ≡ ( b − 1 ) θ 0 ≡ θ 0 a_0-\theta_0\equiv(b-1)\theta_0\equiv\theta_0 a 0 − θ 0 ≡ ( b − 1 ) θ 0 ≡ θ 0 , because ( b − 2 ) θ 0 = − b ′ θ 0 ∈ Z (b-2)\theta_0=-b'\theta_0\in\mathbb{Z} ( b − 2 ) θ 0 = − b ′ θ 0 ∈ Z ; a 1 + b 1 ′ − θ 1 ≡ ( b + b ′ − 1 ) θ 1 = θ 1 a_1+b'_1-\theta_1\equiv(b+b'-1)\theta_1=\theta_1 a 1 + b 1 ′ − θ 1 ≡ ( b + b ′ − 1 ) θ 1 = θ 1 ; and a 0 + a 1 + b 1 ′ − θ 0 − θ 1 ≡ θ 0 + θ 1 = 1 − θ ∞ a_0+a_1+b'_1-\theta_0-\theta_1\equiv\theta_0+\theta_1=1-\theta_\infty a 0 + a 1 + b 1 ′ − θ 0 − θ 1 ≡ θ 0 + θ 1 = 1 − θ ∞ . None of θ 0 , θ 1 , θ ∞ \theta_0,\theta_1,\theta_\infty θ 0 , θ 1 , θ ∞ is an integer, so every argument of Γ \Gamma Γ in κ \kappa κ is a real number that is not an integer, and κ \kappa κ is a non-zero real number. This contradicts G ( Q 0 ) = 0 G(Q_0)=0 G ( Q 0 ) = 0 . □
So under the hypotheses of part (b) the derivative along the Teichmüller disc of the relative period of η b \eta_b η b over a rational class bounding Q 1 − Q 0 Q_1-Q_0 Q 1 − Q 0 is ( ζ − b ′ − 1 ) ( − 1 ) k 1 κ / U 0 (\zeta^{-b'}-1)(-1)^{k_1}\kappa/U_0 ( ζ − b ′ − 1 ) ( − 1 ) k 1 κ / U 0 , a non-zero number given in closed form. As b b b is a unit and b ≡ 2 ( m o d n / g ) b\equiv2\pmod{n/g} b ≡ 2 ( mod n / g ) , the number n / g n/g n / g must be odd: part (b) never speaks at a vertex above which the points are fixed by T n / 2 T^{n/2} T n /2 .
Corollary 3.7, applied to all units b b b , gives nothing beyond Theorem 1.2, on any rational triangle. For a unit b ≢ 2 ( m o d n ) b\not\equiv2\pmod n b ≡ 2 ( mod n ) with H b 1 , 0 ≠ 0 H_b^{1,0}\ne0 H b 1 , 0 = 0 and G b ∘ T = ζ b ′ − 2 G b G_b\circ T=\zeta^{b'-2}G_b G b ∘ T = ζ b ′ − 2 G b , the points above a vertex x x x are fixed by T gcd ( x , n ) T^{\gcd(x,n)} T g c d ( x , n ) , so G b G_b G b vanishes there unless ( 2 − b ) x ≡ 0 ( m o d n ) (2-b)x \equiv0 \pmod{n} ( 2 − b ) x ≡ 0 ( mod n ) ; hence G b = 0 G_b=0 G b = 0 on Σ \Sigma Σ unless b b b satisfies the hypotheses of part (b) at some vertex, and then G b G_b G b is not constant on Σ \Sigma Σ , by the proof above. For b ≡ 2 b \equiv2 b ≡ 2 (n n n odd, Δ \Delta Δ obtuse) G 2 G_2 G 2 is constant above each vertex, so it is constant on Σ \Sigma Σ exactly when p ∣ n p \mid n p ∣ n and q ∣ n q \mid n q ∣ n , the case part (a) leaves out; for even n n n , where 2 2 2 is not a unit, only part (b) remains.
Example 3.11. For ( 1 , 3 , 6 ) / 10 (1,3,6)/10 ( 1 , 3 , 6 ) /10 , the vertex 6 6 6 has g = 2 g=2 g = 2 , 6 ∤ 10 6\nmid10 6 ∤ 10 and n / g = 5 n/g=5 n / g = 5 , and b = 7 b=7 b = 7 is a unit with 7 ≡ 2 ( m o d 5 ) 7\equiv2\pmod{5} 7 ≡ 2 ( mod 5 ) , 7 ≢ 2 ( m o d 10 ) 7\not\equiv2\pmod{10} 7 ≡ 2 ( mod 10 ) and [ 7 ] 10 + [ 21 ] 10 + [ 42 ] 10 = 7 + 1 + 2 = 10 [7]_{10}+[21]_{10}+[42]_{10}=7+1+2=10 [ 7 ] 10 + [ 21 ] 10 + [ 42 ] 10 = 7 + 1 + 2 = 10 . With 6 6 6 at 0 0 0 and 1 1 1 at 1 1 1 : θ 0 = 3 / 5 \theta_0=3/5 θ 0 = 3/5 , θ 1 = 1 / 10 \theta_1=1/10 θ 1 = 1/10 , θ ∞ = 3 / 10 \theta_\infty=3/10 θ ∞ = 3/10 , a 0 = 1 / 5 a_0=1/5 a 0 = 1/5 , a 1 = 7 / 10 a_1=7/10 a 1 = 7/10 , b ′ = − 5 b'=-5 b ′ = − 5 , b 1 ′ = 1 / 2 b'_1=1/2 b 1 ′ = 1/2 , k 0 = − 2 k_0=-2 k 0 = − 2 , k 1 = 0 k_1=0 k 1 = 0 , and
ϰ = Γ ( − 2 / 5 ) Γ ( 11 / 10 ) Γ ( 3 / 10 ) Γ ( 7 / 10 ) Γ ( 2 / 5 ) Γ ( 9 / 10 ) . \varkappa=\frac{\Gamma(-2/5)\Gamma(11/10)\Gamma(3/10)}{\Gamma(7/10)\Gamma(2/5)\Gamma(9/10)}. ϰ = Γ ( 7/10 ) Γ ( 2/5 ) Γ ( 9/10 ) Γ ( − 2/5 ) Γ ( 11/10 ) Γ ( 3/10 ) .
For ( 1 , 3 , 8 ) / 12 (1,3,8)/12 ( 1 , 3 , 8 ) /12 , the vertex 8 8 8 has g = 4 g=4 g = 4 and n / g = 3 n/g=3 n / g = 3 , b = 5 b=5 b = 5 , [ 5 ] 12 + [ 15 ] 12 + [ 40 ] 12 = 12 [5]_{12}+[15]_{12}+[40]_{12}=12 [ 5 ] 12 + [ 15 ] 12 + [ 40 ] 12 = 12 , and with 8 8 8 at 0 0 0 and 1 1 1 at 1 1 1 : a 0 = 1 / 3 a_0=1/3 a 0 = 1/3 , a 1 = 5 / 12 a_1=5/12 a 1 = 5/12 , b ′ = − 3 b'=-3 b ′ = − 3 , b 1 ′ = 3 / 4 b'_1=3/4 b 1 ′ = 3/4 , k 0 = − 2 k_0=-2 k 0 = − 2 , k 1 = 0 k_1=0 k 1 = 0 and ϰ = Γ ( − 1 / 3 ) Γ ( 13 / 12 ) Γ ( 1 / 4 ) / ( Γ ( 3 / 4 ) Γ ( 1 / 3 ) Γ ( 11 / 12 ) ) \varkappa=\Gamma(-1/3)\Gamma(13/12)\Gamma(1/4)/(\Gamma(3/4)\Gamma(1/3)\Gamma(11/12)) ϰ = Γ ( − 1/3 ) Γ ( 13/12 ) Γ ( 1/4 ) / ( Γ ( 3/4 ) Γ ( 1/3 ) Γ ( 11/12 )) . So neither triangle is a lattice triangle.
Where the criterion is blind
Theorem 4.1 below says exactly where the criterion is blind when the obtuse angle is at most 2 π / 3 2\pi/3 2 π /3 , for every denominator. Only its part (a) for even n n n is used, in the proof of Theorem 1.1.
The sets S x S_x S x , usable units and witnesses are defined in § 2.3 . We call the range n / 2 < r ≤ 2 n / 3 n/2<r\le2n/3 n /2 < r ≤ 2 n /3 , where the obtuse angle lies in ( π / 2 , 2 π / 3 ] (\pi/2,2\pi/3] ( π /2 , 2 π /3 ] , the window ; there 1 ≤ s ≤ n / 3 1\le s\le n/3 1 ≤ s ≤ n /3 , with s = 2 r − n s=2r-n s = 2 r − n as in § 2.3 . When s s s divides n n n we use the unit condition
(U) every usable unit a a a modulo n n n with [ a p ] n < 2 p [ap]_n<2p [ a p ] n < 2 p satisfies a ≡ 2 ( m o d n / s ) a\equiv2\pmod{n/s} a ≡ 2 ( mod n / s ) .
Three families of window triangles occur:
A \mathcal{A} A : p = 1 p=1 p = 1 , s s s divides n n n and n / s n/s n / s is odd;
B \mathcal{B} B : s s s divides n n n , d = n / s d=n/s d = n / s is odd, 2 ≤ p ≤ s / 2 2\le p\le s/2 2 ≤ p ≤ s /2 , and (U) holds;
C \mathcal{C} C : s = 2 s=2 s = 2 and 4 4 4 divides n n n , and either p ∈ { 1 , 2 } p\in\{1,2\} p ∈ { 1 , 2 } , or p = 4 p=4 p = 4 and n ≡ 4 ( m o d 8 ) n\equiv4\pmod{8} n ≡ 4 ( mod 8 ) .
A member of A \mathcal{A} A is determined by n n n and s s s , namely ( 1 , ( n − s ) / 2 − 1 , ( n + s ) / 2 ) / n (1,(n-s)/2-1,(n+s)/2)/n ( 1 , ( n − s ) /2 − 1 , ( n + s ) /2 ) / n , and a member of C \mathcal{C} C by n n n and p p p , namely ( p , n / 2 − p − 1 , n / 2 + 1 ) / n (p,n/2-p-1,n/2+1)/n ( p , n /2 − p − 1 , n /2 + 1 ) / n . For instance A \mathcal{A} A with s = 1 s=1 s = 1 is ( 1 , ( n − 3 ) / 2 , ( n + 1 ) / 2 ) / n (1,(n-3)/2,(n+1)/2)/n ( 1 , ( n − 3 ) /2 , ( n + 1 ) /2 ) / n , n n n odd. Let E \mathcal{E} E be the list of the thirty window triangles of Table 1, all with n ≤ 48 n\le48 n ≤ 48 .
Table 1 (PDF p. 17). The thirty triangles of E \mathcal{E} E , each a window triangle with n ≤ 48 n\le48 n ≤ 48 . Twenty-five of them have even denominator; the five odd ones are the first, second, third, ninth and tenth.
Theorem 4.1. Let Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n be a window triangle and s = 2 r − n s=2r-n s = 2 r − n .
(a) Δ \Delta Δ has no witness if and only if it lies in A A A , B B B or C C C or is a member of E \mathcal{E} E (Table 1).
(b) Suppose that s s s divides n n n , that n / s n/s n / s is odd and that 2 ≤ p ≤ s / 2 2 \le p \le s/2 2 ≤ p ≤ s /2 . Then Δ \Delta Δ has no witness if and only if it satisfies the unit condition (U), that is, if and only if it lies in B B B . If gcd ( p , n ) = 1 \gcd(p,n)=1 g cd( p , n ) = 1 , this holds exactly when p p p is a prime, n / s n/s n / s divides 2 p − 1 2p-1 2 p − 1 , and every prime ℓ < 2 p \ell< 2p ℓ < 2 p with ℓ ≠ p \ell\ne p ℓ = p and ℓ ≢ 1 ( m o d n / s ) \ell\not\equiv1 \pmod{n/s} ℓ ≡ 1 ( mod n / s ) divides n n n ; if gcd ( p , n ) > 1 \gcd(p,n)>1 g cd( p , n ) > 1 , exactly when p = 3 p=3 p = 3 and n = 3 s n=3s n = 3 s .
In particular, for a prime n ≥ 5 n \ge5 n ≥ 5 the window triangles without a witness are ( 1 , ( n − 3 ) / 2 , ( n + 1 ) / 2 ) / n (1,(n-3)/2,(n+1)/2)/n ( 1 , ( n − 3 ) /2 , ( n + 1 ) /2 ) / n and, for n = 11 n=11 n = 11 , ( 2 , 3 , 6 ) / 11 (2,3,6)/11 ( 2 , 3 , 6 ) /11 ; for n = 2 ℓ n=2\ell n = 2 ℓ with ℓ \ell ℓ an odd prime they are ( 1 , ℓ − 2 , ℓ + 1 ) / 2 ℓ (1,\ell-2,\ell+1)/2\ell ( 1 , ℓ − 2 , ℓ + 1 ) /2 ℓ and, for n = 22 , 26 , 34 n=22,26,34 n = 22 , 26 , 34 , also ( 3 , 7 , 12 ) / 22 (3,7,12)/22 ( 3 , 7 , 12 ) /22 ; ( 1 , 9 , 16 ) / 26 (1,9,16)/26 ( 1 , 9 , 16 ) /26 , ( 5 , 7 , 14 ) / 26 (5,7,14)/26 ( 5 , 7 , 14 ) /26 ; ( 3 , 13 , 18 ) / 34 (3,13,18)/34 ( 3 , 13 , 18 ) /34 . The window triangles without a witness and with p = 1 p=1 p = 1 are the members of A A A , the members of C C C with p = 1 p=1 p = 1 , and ( 1 , 6 , 11 ) / 18 (1,6,11)/18 ( 1 , 6 , 11 ) /18 , ( 1 , 8 , 15 ) / 24 (1,8,15)/24 ( 1 , 8 , 15 ) /24 , ( 1 , 9 , 16 ) / 26 (1,9,16)/26 ( 1 , 9 , 16 ) /26 , ( 1 , 10 , 17 ) / 28 (1,10,17)/28 ( 1 , 10 , 17 ) /28 , ( 1 , 12 , 23 ) / 36 (1,12,23)/36 ( 1 , 12 , 23 ) /36 , ( 1 , 16 , 23 ) / 40 (1,16,23)/40 ( 1 , 16 , 23 ) /40 .
The unit condition (U) is a condition on the units in S p S_p S p alone. The member ( 3 , 5 , 12 ) / 20 (3,5,12)/20 ( 3 , 5 , 12 ) /20 of E \mathcal{E} E satisfies every clause of the explicit condition in part (b) with gcd ( p , n ) = 1 \gcd(p,n)=1 g cd( p , n ) = 1 except 2 ≤ p ≤ s / 2 2 \le p \le s/2 2 ≤ p ≤ s /2 (p = 3 p=3 p = 3 , s = 4 s=4 s = 4 , n / s = 5 n/s=5 n / s = 5 ). The family A A A contains the two lattice triangles ( 1 , 1 , 3 ) / 5 (1,1,3)/5 ( 1 , 1 , 3 ) /5 and ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 .
Part (a) is Theorem 4.15 for even n n n and Theorem C.3 for odd n n n . For all but finitely many denominators these rest on Theorem 4.14 , proved in Appendix § 9 , and on Theorem C.2 , proved in Appendix C; both use the lemmas of Appendix § 9 .§ 1 , which hold for every denominator.
Theorem 4.2. Let n n n be odd and let Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n be an obtuse triangle such that p p p and q q q divide n n n . Then p q pq pq divides n n n ; put I = n / ( p q ) I=n/(pq) I = n / ( pq ) and s = 2 r − n s=2r-n s = 2 r − n . The triangle Δ \Delta Δ has a witness if and only if s > p I s>pI s > p I , and the triangles with s ≤ p I s\le pI s ≤ p I are exactly ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n with n ≥ 5 n\ge5 n ≥ 5 odd, ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 , ( 1 , 5 , 9 ) / 15 (1,5,9)/15 ( 1 , 5 , 9 ) /15 and ( 3 , 7 , 11 ) / 21 (3,7,11)/21 ( 3 , 7 , 11 ) /21 .
The rest of this section proves Theorems 4.1 and 4.2 and, in Section 4.4, applies Theorem 1.2 to A A A and B B B . Throughout, Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n is a window triangle unless said otherwise; then q ≥ ( p + q ) / 2 = ( n − s ) / 4 q\ge(p+q)/2=(n-s)/4 q ≥ ( p + q ) /2 = ( n − s ) /4 .
The region where s s s divides n n n
In this subsection n n n is arbitrary, Δ \Delta Δ is a window triangle, s s s divides n n n and d = n / s d=n/s d = n / s is odd. Since s ≤ n / 3 s\le n/3 s ≤ n /3 , d ≥ 3 d\ge3 d ≥ 3 . As d d d is odd,
r = n + s 2 = s d + 1 2 , p + q = n − r = s d − 1 2 . r=\frac{n+s}{2}=s\frac{d+1}{2}, \qquad p+q=n-r=s\frac{d-1}{2}. r = 2 n + s = s 2 d + 1 , p + q = n − r = s 2 d − 1 .
Proposition 4.3. Let Δ \Delta Δ be a window triangle with s ∣ n s\mid n s ∣ n and d = n / s d=n/s d = n / s odd.
(a) S r S_r S r contains no unit modulo n n n .
(b) If 2 p ≤ s 2p\le s 2 p ≤ s , a unit a a a with [ a p ] n < 2 p [ap]_n<2p [ a p ] n < 2 p lies in S q S_q S q if and only if a ≢ 2 ( m o d d ) a\not\equiv2\pmod d a ≡ 2 ( mod d ) .
(c) If 2 ≤ p ≤ s / 2 2\le p\le s/2 2 ≤ p ≤ s /2 , then Δ \Delta Δ has no witness if and only if it satisfies the unit condition (U), that is, if and only if it lies in B B B .
(d) If p = 1 p=1 p = 1 , that is if Δ ∈ A \Delta\in A Δ ∈ A , then Δ \Delta Δ has no witness.
Proof. (a) For every a a a , a r = s ⋅ a ( d + 1 ) / 2 ≡ s c 1 ( m o d d s ) ar=s\cdot a(d+1)/2\equiv sc_1\pmod{ds} a r = s ⋅ a ( d + 1 ) /2 ≡ s c 1 ( mod d s ) with c 1 = [ a ( d + 1 ) / 2 ] d c_1=[a(d+1)/2]_d c 1 = [ a ( d + 1 ) /2 ] d , and 0 ≤ s c 1 ≤ s ( d − 1 ) < n 0\le sc_1\le s(d-1)<n 0 ≤ s c 1 ≤ s ( d − 1 ) < n , so [ a r ] n = s c 1 [ar]_n=sc_1 [ a r ] n = s c 1 . Thus [ a r ] n < s [ar]_n<s [ a r ] n < s if and only if c 1 = 0 c_1=0 c 1 = 0 , that is d ∣ a ( d + 1 ) / 2 d\mid a(d+1)/2 d ∣ a ( d + 1 ) /2 , that is d ∣ a d\mid a d ∣ a , because 2 ⋅ ( d + 1 ) / 2 ≡ 1 ( m o d d ) 2\cdot(d+1)/2\equiv1\pmod d 2 ⋅ ( d + 1 ) /2 ≡ 1 ( mod d ) makes ( d + 1 ) / 2 (d+1)/2 ( d + 1 ) /2 invertible modulo d d d . A unit modulo n n n is prime to d d d and d > 1 d>1 d > 1 .
(b) Let a a a be a unit and x = [ a p ] n < 2 p ≤ s x=[ap]_n<2p\le s x = [ a p ] n < 2 p ≤ s . Then x ≥ 1 x\ge1 x ≥ 1 , since n ∤ a p n\nmid ap n ∤ a p . Next a ( p + q ) = s ⋅ a ( d − 1 ) / 2 ≡ s c ( m o d n ) a(p+q)=s\cdot a(d-1)/2\equiv sc\pmod n a ( p + q ) = s ⋅ a ( d − 1 ) /2 ≡ sc ( mod n ) with c = [ a ( d − 1 ) / 2 ] d c=[a(d-1)/2]_d c = [ a ( d − 1 ) /2 ] d ; as ( d − 1 ) / 2 (d-1)/2 ( d − 1 ) /2 is invertible modulo d d d (2 ⋅ ( d − 1 ) / 2 ≡ − 1 2\cdot(d-1)/2\equiv-1 2 ⋅ ( d − 1 ) /2 ≡ − 1 ) and d ∤ a d\nmid a d ∤ a , 1 ≤ c ≤ d − 1 1\le c\le d-1 1 ≤ c ≤ d − 1 . Hence a q ≡ s c − x ( m o d n ) aq\equiv sc-x\pmod n a q ≡ sc − x ( mod n ) , and since $1 \le x < s \le sc < n$, [ a q ] n = s c − x [aq]_n = sc - x [ a q ] n = sc − x . So
a ∈ S q ⟺ s c − x < 2 q = s ( d − 1 ) − 2 p ⟺ x > 2 p − s ( d − 1 − c ) a \in S_q \Longleftrightarrow sc - x < 2q = s(d - 1) - 2p \Longleftrightarrow x > 2p - s(d - 1 - c) a ∈ S q ⟺ sc − x < 2 q = s ( d − 1 ) − 2 p ⟺ x > 2 p − s ( d − 1 − c ) .
If c = d − 1 c = d - 1 c = d − 1 this reads x > 2 p x > 2p x > 2 p , which is false. If c ≤ d − 2 c \le d - 2 c ≤ d − 2 , then 2 p − s ( d − 1 − c ) ≤ 2 p − s ≤ 0 < x 2p - s(d - 1 - c) \le2p - s \le0 < x 2 p − s ( d − 1 − c ) ≤ 2 p − s ≤ 0 < x , and it holds. Finally c = d − 1 c = d - 1 c = d − 1 if and only if a ( d − 1 ) / 2 ≡ − 1 ( m o d d ) a(d - 1)/2 \equiv-1 \pmod{d} a ( d − 1 ) /2 ≡ − 1 ( mod d ) , that is, multiplying by − 2 -2 − 2 , a − a d ≡ 2 a - ad \equiv2 a − a d ≡ 2 , that is a ≡ 2 ( m o d d ) a \equiv2 \pmod{d} a ≡ 2 ( mod d ) .
(c) A witness lies in two of the three sets, by (a) not in S r S_r S r , so it is a usable unit of S p ∩ S q S_p \cap S_q S p ∩ S q . By (b) a usable unit of S p S_p S p lies in S q S_q S q exactly when it is ≢ 2 ( m o d d ) \not\equiv2 \pmod{d} ≡ 2 ( mod d ) . So Δ \Delta Δ has a witness if and only if some usable unit a a a with [ a p ] n < 2 p [ap]_n < 2p [ a p ] n < 2 p is ≢ 2 ( m o d d ) \not\equiv2 \pmod{d} ≡ 2 ( mod d ) , which is the negation of (U).
(d) S 1 = { a : [ a ] n < 2 } = { 0 , 1 } S_1 = \{a : [a]_n < 2\} = \{0, 1\} S 1 = { a : [ a ] n < 2 } = { 0 , 1 } ; 0 0 0 is not a unit and 1 1 1 is not usable. A witness would lie in S 1 S_1 S 1 or in S r S_r S r , which is impossible by (a). □
Parts (a) and (d) are remarks of Larsen, Norton and Zykoski [17 Remark 3.6 and p. 7124] , the first with their m = ( d + 1 ) / 2 m = (d + 1)/2 m = ( d + 1 ) /2 . Part (c) is Theorem 4.1 (b). The unit condition (U) can be made explicit. Say that Δ \Delta Δ satisfies condition (B1) if p p p is a prime not dividing n n n , d d d divides 2 p − 1 2p - 1 2 p − 1 , and every prime ℓ < 2 p \ell< 2p ℓ < 2 p with ℓ ≠ p \ell\ne p ℓ = p and ℓ ≢ 1 ( m o d d ) \ell\not\equiv1 \pmod{d} ℓ ≡ 1 ( mod d ) divides n n n .
Proposition 4.4. Let Δ \Delta Δ be a window triangle with s ∣ n s \mid n s ∣ n , d = n / s d = n/s d = n / s odd and 2 ≤ p ≤ s / 2 2 \le p \le s/2 2 ≤ p ≤ s /2 .
(a) If gcd ( p , n ) = 1 \gcd(p,n) = 1 g cd( p , n ) = 1 , Δ \Delta Δ lies in B \mathcal{B} B if and only if it satisfies (B1).
(b) If gcd ( p , n ) > 1 \gcd(p,n) > 1 g cd( p , n ) > 1 , Δ \Delta Δ lies in B \mathcal{B} B if and only if p = 3 p = 3 p = 3 and d = 3 d = 3 d = 3 ; these are the triangles ( 3 , s − 3 , 2 s ) / 3 s (3, s - 3, 2s)/3s ( 3 , s − 3 , 2 s ) /3 s with s ≥ 7 s \ge7 s ≥ 7 and 3 ∤ s 3 \nmid s 3 ∤ s , and for them S 3 S_3 S 3 contains the unit ( 2 , 1 ) (2, 1) ( 2 , 1 ) of Z / n = Z / 3 × Z / s \mathbb{Z}/n = \mathbb{Z}/3 \times\mathbb{Z}/s Z / n = Z /3 × Z / s .
So on the whole region, Δ \Delta Δ lies in B \mathcal{B} B if and only if it satisfies (B1) or ( p , d ) = ( 3 , 3 ) (p,d) = (3,3) ( p , d ) = ( 3 , 3 ) .
The proof of part (b) uses that gcd ( p , q , r ) = 1 \gcd(p,q,r) = 1 g cd( p , q , r ) = 1 ; for triples with a common factor (U) can hold with ( p , d ) ≠ ( 3 , 3 ) (p,d) \ne(3,3) ( p , d ) = ( 3 , 3 ) , for instance for ( p , q , r ) = ( 2 , 2 , 8 ) (p,q,r) = (2,2,8) ( p , q , r ) = ( 2 , 2 , 8 ) , n = 12 n = 12 n = 12 : (U) holds, vacuously, since S 2 S_2 S 2 contains no usable unit, while ( p , d ) = ( 2 , 3 ) (p,d) = (2,3) ( p , d ) = ( 2 , 3 ) . The triangles that Larsen, Norton and Zykoski construct in the proof of [17 Proposition 3.7] , to show that the criterion is blind for arbitrarily large p p p , satisfy (B1) with d = 2 p − 1 d = 2p - 1 d = 2 p − 1 and with every number below 2 p 2p 2 p other than p p p dividing n n n . Condition (B1) is the exact form of the pattern they observed experimentally for p ≠ 1 , 2 , 4 p \ne1, 2, 4 p = 1 , 2 , 4 [17 p. 7124] .
Proof. (a) Let p − 1 p^{-1} p − 1 be the inverse of p p p modulo n n n . As a ↦ a p a \mapsto ap a ↦ a p permutes Z / n \mathbb{Z}/n Z / n , S p = { k p − 1 : 0 ≤ k ≤ 2 p − 1 } S_p = \{kp^{-1} : 0 \le k \le2p - 1\} S p = { k p − 1 : 0 ≤ k ≤ 2 p − 1 } , and k p − 1 kp^{-1} k p − 1 is a unit if and only if gcd ( k , n ) = 1 \gcd(k,n) = 1 g cd( k , n ) = 1 . Next 2 k p − 1 ≡ 2 ( m o d n ) 2kp^{-1} \equiv2 \pmod{n} 2 k p − 1 ≡ 2 ( mod n ) means 2 k ≡ 2 p ( m o d n ) 2k \equiv2p \pmod{n} 2 k ≡ 2 p ( mod n ) , that is k ≡ p ( m o d n / gcd ( 2 , n ) ) k \equiv p \pmod{n/\gcd(2,n)} k ≡ p ( mod n / g cd( 2 , n )) ; for 1 ≤ k ≤ 2 p − 1 1 \le k \le2p - 1 1 ≤ k ≤ 2 p − 1 , ∣ k − p ∣ < p < n / 2 |k - p| < p < n/2 ∣ k − p ∣ < p < n /2 , so this happens only for k = p k = p k = p . Hence the usable units of S p S_p S p are the k p − 1 kp^{-1} k p − 1 with k k k in
K = { k : 1 ≤ k ≤ 2 p − 1 , k ≠ p , gcd ( k , n ) = 1 } . K = \{k : 1 \le k \le2p - 1,\ k \ne p,\ \gcd(k,n) = 1\}. K = { k : 1 ≤ k ≤ 2 p − 1 , k = p , g cd( k , n ) = 1 } .
As d ∣ n d \mid n d ∣ n , p − 1 p^{-1} p − 1 reduces to the inverse of p p p modulo d d d , and k p − 1 ≡ 2 ( m o d d ) kp^{-1} \equiv2 \pmod{d} k p − 1 ≡ 2 ( mod d ) if and only if k ≡ 2 p ( m o d d ) k \equiv2p \pmod{d} k ≡ 2 p ( mod d ) . So (U) says:
( ⋆ ) every k ∈ K satisfies k ≡ 2 p ( m o d d ) . (\star)\qquad\text{every } k \in K \text{ satisfies } k \equiv2p \pmod{d}. ( ⋆ ) every k ∈ K satisfies k ≡ 2 p ( mod d ) .
Suppose ( ⋆ ) (\star) ( ⋆ ) . As p ≥ 2 p \ge2 p ≥ 2 , 1 ∈ K 1 \in K 1 ∈ K , so 2 p ≡ 1 ( m o d d ) 2p \equiv1 \pmod{d} 2 p ≡ 1 ( mod d ) , and ( ⋆ ) (\star) ( ⋆ ) says that every k ∈ K k \in K k ∈ K is ≡ 1 ( m o d d ) \equiv1 \pmod{d} ≡ 1 ( mod d ) . If p p p were composite, each prime factor ℓ \ell ℓ of p p p would satisfy ℓ < p \ell< p ℓ < p and ℓ ∣ n \ell\mid n ℓ ∣ n , so ℓ ∈ K \ell\in K ℓ ∈ K and ℓ ≡ 1 ( m o d d ) \ell\equiv1 \pmod{d} ℓ ≡ 1 ( mod d ) ; multiplying, p ≡ 1 ( m o d d ) p \equiv1 \pmod{d} p ≡ 1 ( mod d ) , and with 2 p ≡ 1 2p \equiv1 2 p ≡ 1 this gives d ∣ p d \mid p d ∣ p , against gcd ( p , n ) = 1 \gcd(p,n) = 1 g cd( p , n ) = 1 . So p p p is a prime not dividing n n n . A prime ℓ < 2 p \ell< 2p ℓ < 2 p with ℓ ≠ p \ell\ne p ℓ = p and ℓ ∤ n \ell\nmid n ℓ ∤ n lies in K K K , so ℓ ≡ 1 ( m o d d ) \ell\equiv1 \pmod{d} ℓ ≡ 1 ( mod d ) : this is the last clause of (B1). Conversely, suppose (B1) and let k ∈ K k \in K k ∈ K . Each prime factor ℓ \ell ℓ of k k k has ℓ < 2 p \ell< 2p ℓ < 2 p , ℓ ∤ n \ell\nmid n ℓ ∤ n and ℓ ≠ p \ell\ne p ℓ = p (as p ∣ k p \mid k p ∣ k and k < 2 p k < 2p k < 2 p would force k = p k = p k = p ), so ℓ ≡ 1 ( m o d d ) \ell\equiv1 \pmod{d} ℓ ≡ 1 ( mod d ) . Hence k ≡ 1 ≡ 2 p ( m o d d ) k \equiv1 \equiv2p \pmod{d} k ≡ 1 ≡ 2 p ( mod d ) , and ( ⋆ ) (\star) ( ⋆ ) holds.
(b) Let h = gcd ( p , n ) > 1 h=\gcd(p,n)>1 h = g cd( p , n ) > 1 . A common divisor of p p p and s s s divides q = ( p + q ) − p q=(p+q)-p q = ( p + q ) − p and r r r , so gcd ( p , s ) = 1 \gcd(p,s)=1 g cd( p , s ) = 1 . Hence h = gcd ( p , s d ) h=\gcd(p,sd) h = g cd( p , s d ) is prime to s s s and divides d d d ; in particular h h h is odd and h ≥ 3 h\ge3 h ≥ 3 . Write p = h p ′ p=hp' p = h p ′ and n = h n ′ n=hn' n = h n ′ ; then gcd ( p ′ , n ′ ) = 1 \gcd(p',n')=1 g cd( p ′ , n ′ ) = 1 and n ′ = s ⋅ d / h n'=s\cdot d/h n ′ = s ⋅ d / h .
Step 1. Every a ≡ 1 ( m o d n ′ ) a\equiv1\pmod{n'} a ≡ 1 ( mod n ′ ) lies in S p S_p S p : [ a p ] n = h [ a p ′ ] n ′ = h p ′ = p < 2 p [ap]_n=h[ap']_{n'}=hp'=p<2p [ a p ] n = h [ a p ′ ] n ′ = h p ′ = p < 2 p , since p ′ < n ′ p'<n' p ′ < n ′ , as p < n p<n p < n .
Step 2. There is a usable unit a ≠ 1 a\ne1 a = 1 with a ≡ 1 ( m o d n ′ ) a\equiv1\pmod{n'} a ≡ 1 ( mod n ′ ) . The units ≡ 1 ( m o d n ′ ) \equiv1\pmod{n'} ≡ 1 ( mod n ′ ) form the kernel of the surjection ( Z / n ) × → ( Z / n ′ ) × (\mathbb{Z}/n)^\times\to(\mathbb{Z}/n')^\times ( Z / n ) × → ( Z / n ′ ) × , of order φ ( n ) / φ ( n ′ ) = ∏ ℓ ∣ h c ℓ \varphi(n)/\varphi(n')=\prod_{\ell\mid h}c_\ell φ ( n ) / φ ( n ′ ) = ∏ ℓ ∣ h c ℓ with c ℓ = ℓ e c_\ell=\ell^e c ℓ = ℓ e if ℓ ∣ n ′ \ell\mid n' ℓ ∣ n ′ and c ℓ = ℓ e − 1 ( ℓ − 1 ) c_\ell=\ell^{e-1}(\ell-1) c ℓ = ℓ e − 1 ( ℓ − 1 ) otherwise; each factor is at least 2 2 2 , as ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 . Take a ≠ 1 a\ne1 a = 1 in it. If n n n is odd, a a a is usable. If n n n is even, the only a ≠ 1 a\ne1 a = 1 with 2 a ≡ 2 2a\equiv2 2 a ≡ 2 is n / 2 + 1 n/2+1 n /2 + 1 , and n / 2 + 1 ≡ 1 ( m o d n ′ ) n/2+1\equiv1\pmod{n'} n /2 + 1 ≡ 1 ( mod n ′ ) would need n ′ ∣ h n ′ / 2 n'\mid hn'/2 n ′ ∣ h n ′ /2 , that is h h h even.
Step 3. If Δ \Delta Δ satisfies (U), then h = d h=d h = d : the a a a of Step 2 lies in S p S_p S p and satisfies a ≡ 1 ( m o d d / h ) a\equiv1\pmod{d/h} a ≡ 1 ( mod d / h ) , as d / h d/h d / h divides n ′ n' n ′ , while the condition requires a ≡ 2 ( m o d d ) a\equiv2\pmod d a ≡ 2 ( mod d ) ; so d / h d/h d / h divides 1 1 1 .
Step 4. If moreover h = d h=d h = d , then d = 3 d=3 d = 3 . Now n ′ = s n'=s n ′ = s and d ∣ p d\mid p d ∣ p , so gcd ( d , s ) = 1 \gcd(d,s)=1 g cd( d , s ) = 1 (as gcd ( p , s ) = 1 \gcd(p,s)=1 g cd( p , s ) = 1 ) and Z / n = Z / d × Z / s \mathbb{Z}/n=\mathbb{Z}/d\times\mathbb{Z}/s Z / n = Z / d × Z / s . Every ( x , 1 ) (x,1) ( x , 1 ) with x x x a unit modulo d d d is a unit modulo n n n and lies in S p S_p S p by Step 1, and for x ≠ 1 x\ne1 x = 1 it is usable (n / 2 + 1 ≡ 1 ( m o d s ) n/2+1\equiv1\pmod s n /2 + 1 ≡ 1 ( mod s ) would need d d d even). The condition forces x ≡ 2 ( m o d d ) x\equiv2\pmod d x ≡ 2 ( mod d ) for each unit x ≠ 1 x\ne1 x = 1 modulo d d d , so φ ( d ) ≤ 2 \varphi(d)\le2 φ ( d ) ≤ 2 , and d d d odd and ≥ 3 \ge3 ≥ 3 gives d = 3 d=3 d = 3 .
Step 5. If moreover h = d = 3 h=d=3 h = d = 3 , then p = 3 p=3 p = 3 . Here 3 ∤ s 3\nmid s 3 ∤ s , p = 3 p ′ p=3p' p = 3 p ′ with gcd ( p ′ , s ) = 1 \gcd(p',s)=1 g cd( p ′ , s ) = 1 , and 6 p ′ ≤ s 6p'\le s 6 p ′ ≤ s . If p ′ ≥ 2 p'\ge2 p ′ ≥ 2 , let a = ( 1 , p ′ − 1 ) ∈ Z / 3 × Z / s a=(1,p'^{-1})\in\mathbb{Z}/3\times\mathbb{Z}/s a = ( 1 , p ′ − 1 ) ∈ Z /3 × Z / s , with p ′ − 1 p'^{-1} p ′ − 1 the inverse of p ′ p' p ′ modulo s s s (p ′ − 1 ≠ 1 p'^{-1}\ne1 p ′ − 1 = 1 , as 2 ≤ p ′ < s 2\le p'<s 2 ≤ p ′ < s ). It is a unit, it lies in S p S_p S p as [ a p ] n = 3 [ a p ′ ] s = 3 < 2 p [ap]_n=3[ap']_s=3<2p [ a p ] n = 3 [ a p ′ ] s = 3 < 2 p , it is ≡ 1 ( m o d 3 ) \equiv1\pmod3 ≡ 1 ( mod 3 ) , and a ≠ 1 a\ne1 a = 1 . It is usable: n / 2 + 1 n/2+1 n /2 + 1 is a unit only if 4 ∣ s 4\mid s 4 ∣ s , and then n / 2 + 1 = ( 1 , s / 2 + 1 ) n/2+1=(1,s/2+1) n /2 + 1 = ( 1 , s /2 + 1 ) with ( s / 2 + 1 ) 2 ≡ 1 ( m o d s ) (s/2+1)^2\equiv1\pmod s ( s /2 + 1 ) 2 ≡ 1 ( mod s ) , so a = n / 2 + 1 a=n/2+1 a = n /2 + 1 would give p ′ ≡ s / 2 + 1 ( m o d s ) p'\equiv s/2+1\pmod s p ′ ≡ s /2 + 1 ( mod s ) , impossible as p ′ ≤ s / 6 p'\le s/6 p ′ ≤ s /6 . So the condition fails, and p ′ = 1 p'=1 p ′ = 1 .
Step 6. Conversely let ( p , d ) = ( 3 , 3 ) (p,d)=(3,3) ( p , d ) = ( 3 , 3 ) . Then n = 3 s n=3s n = 3 s , r = 2 s r=2s r = 2 s , q = s − 3 q=s-3 q = s − 3 , s ≥ 6 s\ge6 s ≥ 6 as 2 p ≤ s 2p\le s 2 p ≤ s , and 3 ∤ s 3\nmid s 3 ∤ s as gcd ( 3 , s − 3 , 2 s ) = 1 \gcd(3,s-3,2s)=1 g cd( 3 , s − 3 , 2 s ) = 1 ; so s ≥ 7 s\ge7 s ≥ 7 and h = 3 h=3 h = 3 . As 3 ∣ n 3\mid n 3 ∣ n , [ 3 a ] n [3a]_n [ 3 a ] n is a multiple of 3 3 3 , so S 3 = { a : 3 a ≡ 0 or 3 ( m o d n ) } = { a : a ≡ 0 or 1 ( m o d s ) } S_3=\{a:3a\equiv0\text{ or }3\pmod n\}=\{a:a\equiv0\text{ or }1\pmod s\} S 3 = { a : 3 a ≡ 0 or 3 ( mod n )} = { a : a ≡ 0 or 1 ( mod s )} . The a ≡ 0 ( m o d s ) a\equiv0\pmod s a ≡ 0 ( mod s ) are not units, and the units ≡ 1 ( m o d s ) \equiv1\pmod s ≡ 1 ( mod s ) are ( 1 , 1 ) = 1 (1,1)=1 ( 1 , 1 ) = 1 , which is not usable, and ( 2 , 1 ) ≡ 2 ( m o d 3 ) (2,1)\equiv2\pmod3 ( 2 , 1 ) ≡ 2 ( mod 3 ) . So (U) holds, and the unit ( 2 , 1 ) (2,1) ( 2 , 1 ) lies in S 3 S_3 S 3 .
For the last sentence: if gcd ( p , n ) > 1 \gcd(p,n)>1 g cd( p , n ) > 1 , (B1) fails, since it requires p p p to be a prime not dividing n n n , and a prime with gcd ( p , n ) > 1 \gcd(p,n)>1 g cd( p , n ) > 1 divides n n n ; so part (b) is the statement. If gcd ( p , n ) = 1 \gcd(p,n)=1 g cd( p , n ) = 1 , ( p , d ) = ( 3 , 3 ) (p,d)=(3,3) ( p , d ) = ( 3 , 3 ) is impossible, since it gives 3 ∣ gcd ( p , n ) 3\mid\gcd(p,n) 3 ∣ g cd( p , n ) ; so part (a) is the statement. □
Odd denominators with p p p and q q q dividing n n n
In this subsection n n n is odd and Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n is any obtuse triangle with p ∣ n p\mid n p ∣ n and q ∣ n q\mid n q ∣ n ; it need not lie in the window. For odd n n n a unit a a a is usable exactly when a ≠ 1 a\ne1 a = 1 , and [ 2 r ] n = s [2r]_n=s [ 2 r ] n = s as before.
We prove Theorem 4.2 at the end of this subsection, with the explicit witness of Proposition 4.5 , which is Proposition 4.9 (i) together with Proposition 4.11 .
Proposition 4.5. In Theorem 4.2 , if s > p I s>pI s > p I , a witness is the residue a a a modulo n n n with a ≡ 1 ( m o d p I ) a\equiv1\pmod{pI} a ≡ 1 ( mod p I ) and a ≡ 1 + I ( q + t ) / 2 ( m o d q I ) a\equiv1+I(q+t)/2\pmod{qI} a ≡ 1 + I ( q + t ) /2 ( mod q I ) , where t t t is any odd integer with ∣ t ∣ p I < s \lvert t\rvert pI<s ∣ t ∣ p I < s and gcd ( 2 + I t , q ) = 1 \gcd(2+It,q)=1 g cd( 2 + I t , q ) = 1 ; for [ a q ] n = q [aq]_n=q [ a q ] n = q and [ a r ] n = ( s − p I t ) / 2 [ar]_n=(s-pIt)/2 [ a r ] n = ( s − p I t ) /2 . Such an a a a exists (Proposition 4.11 ).
Lemma 4.6. gcd ( p , q ) = 1 \gcd(p,q)=1 g cd( p , q ) = 1 ; n = p q I n=pqI n = pq I with I I I a positive odd integer; p p p and q q q are odd; and 2 ( p + q ) < p q I 2(p+q)<pqI 2 ( p + q ) < pq I .
Proof. A common divisor of p p p and q q q divides n n n , hence r = n − p − q r=n-p-q r = n − p − q , so it divides gcd ( p , q , r ) = 1 \gcd(p,q,r)=1 g cd( p , q , r ) = 1 . Then p q ∣ n pq\mid n pq ∣ n , and all divisors of the odd number n n n are odd. Finally 2 r > n 2r>n 2 r > n is 2 ( p + q ) < n 2(p+q)<n 2 ( p + q ) < n . □】【。
Lemma 4.7. For a unit a a a modulo n n n : a ∈ S q a \in S_q a ∈ S q if and only if a ≡ 1 ( m o d p I ) a \equiv1 \pmod{pI} a ≡ 1 ( mod p I ) , if and only if a q ≡ q ( m o d n ) aq \equiv q \pmod n a q ≡ q ( mod n ) ; and a ∈ S p a \in S_p a ∈ S p if and only if a ≡ 1 ( m o d q I ) a \equiv1 \pmod{qI} a ≡ 1 ( mod q I ) , if and only if a p ≡ p ( m o d n ) ap \equiv p \pmod n a p ≡ p ( mod n ) . The only unit in S p ∩ S q S_p \cap S_q S p ∩ S q is 1 1 1 ; so no witness lies in S p ∩ S q S_p \cap S_q S p ∩ S q .
Proof. q q q divides n n n and a q aq a q , hence [ a q ] n [aq]_n [ a q ] n . If [ a q ] n < 2 q [aq]_n < 2q [ a q ] n < 2 q , then [ a q ] n [aq]_n [ a q ] n is 0 0 0 or q q q ; it is not 0 0 0 , since n ∣ a q n \mid aq n ∣ a q with a a a a unit would give n ∣ q n \mid q n ∣ q , while q < n q<n q < n . So a q ≡ q ( m o d n ) aq \equiv q \pmod n a q ≡ q ( mod n ) , that is n = p q I n=pqI n = pq I divides ( a − 1 ) q (a-1)q ( a − 1 ) q , that is p I ∣ a − 1 pI \mid a-1 p I ∣ a − 1 . Conversely a ≡ 1 ( m o d p I ) a \equiv1 \pmod{pI} a ≡ 1 ( mod p I ) gives a q ≡ q aq \equiv q a q ≡ q and [ a q ] n = q < 2 q [aq]_n=q<2q [ a q ] n = q < 2 q . The same holds for p p p . If both hold, a − 1 a-1 a − 1 is divisible by lcm ( p I , q I ) = p q I = n \operatorname{lcm}(pI,qI)=pqI=n lcm ( p I , q I ) = pq I = n .
Lemma 4.8. Let a ≡ 1 ( m o d p I ) a \equiv1 \pmod{pI} a ≡ 1 ( mod p I ) . The odd integers t t t with
a ≡ 1 + I ( q + t ) / 2 ( m o d q I ) a \equiv1 + I(q+t)/2 \pmod{qI} a ≡ 1 + I ( q + t ) /2 ( mod q I ) form exactly one residue class modulo 2 q 2q 2 q , and for each of them a q ≡ q aq \equiv q a q ≡ q and a r ≡ ( s − p I t ) / 2 ( m o d n ) ar \equiv(s-pIt)/2 \pmod n a r ≡ ( s − p I t ) /2 ( mod n ) .
Proof. a ≡ 1 ( m o d I ) a \equiv1 \pmod I a ≡ 1 ( mod I ) ; write a = 1 + I m a=1+Im a = 1 + I m . The congruence says m ≡ ( q + t ) / 2 ( m o d q ) m \equiv(q+t)/2 \pmod q m ≡ ( q + t ) /2 ( mod q ) , that is t ≡ 2 m − q ( m o d 2 q ) t \equiv2m-q \pmod{2q} t ≡ 2 m − q ( mod 2 q ) , and such t t t are odd as q q q is odd. a q ≡ q aq \equiv q a q ≡ q , since n = p I ⋅ q n=pI\cdot q n = p I ⋅ q divides ( a − 1 ) q (a-1)q ( a − 1 ) q . Multiplying the congruence modulo q I qI q I by p p p gives a p ≡ p + p I ( q + t ) / 2 ( m o d n ) ap \equiv p+pI(q+t)/2 \pmod n a p ≡ p + p I ( q + t ) /2 ( mod n ) . Then, modulo n n n ,
a r = a n − a p − a q ≡ − p − q − p I ( q + t ) 2 = − ( p + q ) − n + p I t 2 ≡ n − ( p + q ) − n + p I t 2 = s − p I t 2 , ar=an-ap-aq \equiv-p-q-\frac{pI(q+t)}{2}=-(p+q)-\frac{n+pIt}{2}\equiv n-(p+q)-\frac{n+pIt}{2}=\frac{s-pIt}{2}, a r = an − a p − a q ≡ − p − q − 2 p I ( q + t ) = − ( p + q ) − 2 n + p I t ≡ n − ( p + q ) − 2 n + p I t = 2 s − p I t ,
because s = n − 2 p − 2 q s=n-2p-2q s = n − 2 p − 2 q . Here s s s and p I t pIt p I t are odd, so ( s − p I t ) / 2 (s-pIt)/2 ( s − p I t ) /2 is an integer, and replacing t t t by t + 2 q t+2q t + 2 q changes it by n n n .
Proposition 4.9. (i) The witnesses in S q ∩ S r S_q \cap S_r S q ∩ S r are in bijection with the odd integers t t t such that ∣ t ∣ p I < s |t|pI<s ∣ t ∣ p I < s and gcd ( 2 + I t , q ) = 1 \gcd(2+It,q)=1 g cd( 2 + I t , q ) = 1 , the witness of t t t being the residue a t a_t a t modulo n n n with a t ≡ 1 ( m o d p I ) a_t \equiv1 \pmod{pI} a t ≡ 1 ( mod p I ) and a t ≡ 1 + I ( q + t ) / 2 ( m o d q I ) a_t \equiv1+I(q+t)/2 \pmod{qI} a t ≡ 1 + I ( q + t ) /2 ( mod q I ) ; and [ a t q ] n = q [a_tq]_n=q [ a t q ] n = q , [ a t r ] n = ( s − p I t ) / 2 [a_tr]_n=(s-pIt)/2 [ a t r ] n = ( s − p I t ) /2 .
(ii) The same holds with p p p and q q q exchanged: the witnesses in S p ∩ S r S_p \cap S_r S p ∩ S r are in bijection with the odd t t t such that ∣ t ∣ q I < s |t|qI<s ∣ t ∣ q I < s and gcd ( 2 + I t , p ) = 1 \gcd(2+It,p)=1 g cd( 2 + I t , p ) = 1 .
(iii) Every witness is of exactly one of these two kinds.
Proof. (i) From t t t to a witness. The moduli p I pI p I and q I qI q I have greatest common divisor I I I and least common multiple n n n , and the two prescribed residues agree modulo I I I (both are ≡ 1 \equiv1 ≡ 1 ), so a t a_t a t exists and is unique modulo n n n . It is a unit: let ℓ \ell ℓ be a prime dividing n n n . If ℓ ∣ p I \ell\mid pI ℓ ∣ p I , then a t ≡ 1 ( m o d ℓ ) a_t \equiv1 \pmod\ell a t ≡ 1 ( mod ℓ ) . Otherwise ℓ ∣ q \ell\mid q ℓ ∣ q , and 2 a t ≡ 2 + I q + I t ≡ 2 + I t ( m o d ℓ ) 2a_t \equiv2+Iq+It \equiv2+It \pmod\ell 2 a t ≡ 2 + I q + I t ≡ 2 + I t ( mod ℓ ) , which is not 0 0 0 because gcd ( 2 + I t , q ) = 1 \gcd(2+It,q)=1 g cd( 2 + I t , q ) = 1 ; as ℓ \ell ℓ is odd, ℓ ∤ a t \ell\nmid a_t ℓ ∤ a t . By Lemma 4.8 , a t q ≡ q a_tq \equiv q a t q ≡ q , so [ a t q ] n = q < 2 q [a_tq]_n=q<2q [ a t q ] n = q < 2 q , and a t r ≡ ( s − p I t ) / 2 a_tr \equiv(s-pIt)/2 a t r ≡ ( s − p I t ) /2 ; from − s < p I t < s -s<pIt<s − s < p I t < s we get 0 < ( s − p I t ) / 2 < s < n 0<(s-pIt)/2<s<n 0 < ( s − p I t ) /2 < s < n , so [ a t r ] n = ( s − p I t ) / 2 < s [a_tr]_n=(s-pIt)/2<s [ a t r ] n = ( s − p I t ) /2 < s . Finally a t ≠ 1 a_t\ne1 a t = 1 , because [ r ] n = r ≥ s [r]_n=r\ge s [ r ] n = r ≥ s , so 1 ∉ S r 1\notin S_r 1 ∈ / S r . Hence a t a_t a t is a witness in S q ∩ S r S_q\cap S_r S q ∩ S r .
From a witness to t t t . Let a a a be a unit in S q ∩ S r S_q\cap S_r S q ∩ S r . By Lemma 4.7 , a ≡ 1 ( m o d p I ) a\equiv1\pmod{pI} a ≡ 1 ( mod p I ) . By Lemma 4.8 , as t t t runs over the residue class modulo 2 q 2q 2 q attached to a a a , the numbers ( s − p I t ) / 2 (s-pIt)/2 ( s − p I t ) /2 run over a full residue class modulo n n n , the class of a r ar a r ; choose the t t t with ( s − p I t ) / 2 = [ a r ] n (s-pIt)/2=[ar]_n ( s − p I t ) /2 = [ a r ] n . Then 0 ≤ ( s − p I t ) / 2 < s 0\le(s-pIt)/2<s 0 ≤ ( s − p I t ) /2 < s , and it is not 0 0 0 , as a a a is a unit and 0 < r < n 0<r<n 0 < r < n ; so − s < p I t < s -s<pIt<s − s < p I t < s . And 2 a ≡ 2 + I t ( m o d q ) 2a\equiv2+It\pmod q 2 a ≡ 2 + I t ( mod q ) with a a a a unit gives gcd ( 2 + I t , q ) = 1 \gcd(2+It,q)=1 g cd( 2 + I t , q ) = 1 . By construction a = a t a=a_t a = a t .
Bijection. If a t = a t ′ a_t=a_{t'} a t = a t ′ , then t ≡ t ′ ( m o d 2 q ) t\equiv t'\pmod{2q} t ≡ t ′ ( mod 2 q ) by Lemma 4.8 , and ∣ t ∣ , ∣ t ′ ∣ < s / ( p I ) = q − 2 / I − 2 q / ( p I ) < q |t|,|t'|<s/(pI)=q-2/I-2q/(pI)<q ∣ t ∣ , ∣ t ′ ∣ < s / ( p I ) = q − 2/ I − 2 q / ( p I ) < q , so t = t ′ t=t' t = t ′ .
(ii) Lemmas 4.7 and 4.8 and the argument of (i) never used p ≤ q p\le q p ≤ q . (iii) A witness lies in two of S p S_p S p , S q S_q S q , S r S_r S r , and not in S p ∩ S q S_p\cap S_q S p ∩ S q by Lemma 4.7 ; if it lay in all three it would lie in S p ∩ S q S_p\cap S_q S p ∩ S q .
Lemma 4.10. Let D D D be a product of k ≥ 1 k \ge1 k ≥ 1 different primes, and for each prime ℓ ∣ D \ell\mid D ℓ ∣ D let a residue class β ℓ \beta_\ell β ℓ modulo ℓ \ell ℓ be given. Among any N N N consecutive integers, more than N φ ( D ) / D − 2 k + 1 N\varphi(D)/D - 2^k + 1 N φ ( D ) / D − 2 k + 1 avoid all the classes β ℓ \beta_\ell β ℓ .
Proof. By inclusion and exclusion the number is ∑ e ∣ D ( − 1 ) k e A e \sum_{e\mid D}(-1)^{k_e}A_e ∑ e ∣ D ( − 1 ) k e A e , k e k_e k e the number of prime factors of e e e , where A e A_e A e is the number of the N N N integers lying in β ℓ \beta_\ell β ℓ modulo ℓ \ell ℓ for every ℓ ∣ e \ell\mid e ℓ ∣ e . By the Chinese remainder theorem that is one class modulo e e e , so A e = N / e + ϵ e A_e=N/e+\epsilon_e A e = N / e + ϵ e with ∣ ϵ e ∣ < 1 |\epsilon_e|<1 ∣ ϵ e ∣ < 1 and ϵ 1 = 0 \epsilon_1=0 ϵ 1 = 0 . The main terms add up to N ∏ ℓ ∣ D ( 1 − 1 / ℓ ) = N φ ( D ) / D N\prod_{\ell\mid D}(1-1/\ell)=N\varphi(D)/D N ∏ ℓ ∣ D ( 1 − 1/ ℓ ) = N φ ( D ) / D , and the 2 k − 1 2^k-1 2 k − 1 other error terms are each less than 1 in size. □
Proposition 4.11. If s > p I s>pI s > p I , there is an odd t t t with ∣ t ∣ p I < s |t|pI<s ∣ t ∣ p I < s and gcd ( 2 + I t , q ) = 1 \gcd(2+It,q)=1 g cd( 2 + I t , q ) = 1 . If neither t = 1 t=1 t = 1 nor t = − 1 t=-1 t = − 1 is one, then I ≥ 5 I\ge5 I ≥ 5 and q q q is divisible by two different primes that do not divide I I I .
Proof. t = ± 1 t=\pm1 t = ± 1 satisfy ∣ t ∣ p I < s |t|pI<s ∣ t ∣ p I < s . If gcd ( I + 2 , q ) = 1 \gcd(I+2,q)=1 g cd( I + 2 , q ) = 1 take t = 1 t=1 t = 1 ; if gcd ( I − 2 , q ) = 1 \gcd(I-2,q)=1 g cd( I − 2 , q ) = 1 take t = − 1 t=-1 t = − 1 . Otherwise there are primes ℓ 1 ∣ gcd ( q , I + 2 ) \ell_1\mid\gcd(q,I+2) ℓ 1 ∣ g cd( q , I + 2 ) and ℓ 2 ∣ gcd ( q , I − 2 ) \ell_2\mid\gcd(q,I-2) ℓ 2 ∣ g cd( q , I − 2 ) . They are odd; they are different, since a common prime would divide ( I + 2 ) − ( I − 2 ) = 4 (I+2)-(I-2)=4 ( I + 2 ) − ( I − 2 ) = 4 ; and neither divides I I I , since it would then divide 2. As I − 2 I-2 I − 2 has a prime factor and I I I is odd and positive, I ≥ 5 I\ge5 I ≥ 5 (I − 2 = − 1 , 1 I-2=-1,1 I − 2 = − 1 , 1 for I = 1 , 3 I=1,3 I = 1 , 3 ). So q ≥ 15 q\ge15 q ≥ 15 , and
M = s p I = q − 2 I − 2 q p I ≥ q − 2 5 − 2 q 5 = 3 q − 2 5 . M=\frac{s}{pI}=q-\frac{2}{I}-\frac{2q}{pI}\ge q-\frac{2}{5}-\frac{2q}{5}=\frac{3q-2}{5}. M = p I s = q − I 2 − p I 2 q ≥ q − 5 2 − 5 2 q = 5 3 q − 2 .
Let D D D be the product of the primes dividing q q q but not I I I , and k ≥ 2 k\ge2 k ≥ 2 their number. Write the odd t t t as 2 j + 1 2j+1 2 j + 1 . The condition ∣ t ∣ < M |t|<M ∣ t ∣ < M is − ( M + 1 ) / 2 < j < ( M − 1 ) / 2 -(M+1)/2<j<(M-1)/2 − ( M + 1 ) /2 < j < ( M − 1 ) /2 , an open interval of length M M M , which contains N ≥ M − 1 N\ge M-1 N ≥ M − 1 consecutive integers. For a prime ℓ ∣ q \ell\mid q ℓ ∣ q : if ℓ ∣ I \ell\mid I ℓ ∣ I , then 2 + I t ≡ 2 ≢ 0 ( m o d ℓ ) 2+It\equiv2\not\equiv0\pmod{\ell} 2 + I t ≡ 2 ≡ 0 ( mod ℓ ) ; if ℓ ∣ D \ell\mid D ℓ ∣ D , then ℓ ∣ 2 + I ( 2 j + 1 ) \ell\mid2+I(2j+1) ℓ ∣ 2 + I ( 2 j + 1 ) exactly when j j j lies in one class β ℓ \beta_\ell β ℓ modulo ℓ \ell ℓ , namely − ( I + 2 ) / ( 2 I ) -(I+2)/(2I) − ( I + 2 ) / ( 2 I ) . So gcd ( 2 + I t , q ) = 1 \gcd(2+It,q)=1 g cd( 2 + I t , q ) = 1 exactly when j j j avoids β ℓ \beta_\ell β ℓ for every ℓ ∣ D \ell\mid D ℓ ∣ D , and by Lemma 4.10 the number of such j j j is more than N φ ( D ) / D − 2 k + 1 N\varphi(D)/D-2^k+1 N φ ( D ) / D − 2 k + 1 . Now N ≥ M − 1 ≥ ( 3 q − 7 ) / 5 ≥ ( 3 D − 7 ) / 5 > 0 N\ge M-1\ge(3q-7)/5\ge(3D-7)/5>0 N ≥ M − 1 ≥ ( 3 q − 7 ) /5 ≥ ( 3 D − 7 ) /5 > 0 , so
N φ ( D ) / D ≥ ( 3 D − 7 ) φ ( D ) / ( 5 D ) > ( 3 φ ( D ) − 7 ) / 5 , N\varphi(D)/D\ge(3D-7)\varphi(D)/(5D)>(3\varphi(D)-7)/5, N φ ( D ) / D ≥ ( 3 D − 7 ) φ ( D ) / ( 5 D ) > ( 3 φ ( D ) − 7 ) /5 ,
and φ ( D ) ≥ 2 ⋅ 4 k − 1 ≥ 2 k + 1 \varphi(D)\ge2\cdot4^{k-1}\ge2^{k+1} φ ( D ) ≥ 2 ⋅ 4 k − 1 ≥ 2 k + 1 , as the least prime of D D D is at least 3 and the others at least 5. Hence N φ ( D ) / D > ( 6 ⋅ 2 k − 7 ) / 5 = 2 k − 1 + ( 2 k − 2 ) / 5 ≥ 2 k − 1 N\varphi(D)/D>(6\cdot2^k-7)/5=2^k-1+(2^k-2)/5\ge2^k-1 N φ ( D ) / D > ( 6 ⋅ 2 k − 7 ) /5 = 2 k − 1 + ( 2 k − 2 ) /5 ≥ 2 k − 1 , and the number of good j j j is positive. □
The smallest triangles where t = ± 1 t=\pm1 t = ± 1 both fail are ( 1 , 15 , 89 ) / 105 (1,15,89)/105 ( 1 , 15 , 89 ) /105 , with I = 7 I=7 I = 7 , where t = 3 t=3 t = 3 gives a = 64 a=64 a = 64 , and ( 1 , 21 , 83 ) / 105 (1,21,83)/105 ( 1 , 21 , 83 ) /105 , with I = 5 I=5 I = 5 , where t = 3 t=3 t = 3 gives a = 61 a=61 a = 61 .
Proposition 4.12. s ≤ p I s\le pI s ≤ p I holds exactly for ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n with n ≥ 5 n\ge5 n ≥ 5 odd, ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 , ( 1 , 5 , 9 ) / 15 (1,5,9)/15 ( 1 , 5 , 9 ) /15 and ( 3 , 7 , 11 ) / 21 (3,7,11)/21 ( 3 , 7 , 11 ) /21 .
Proof. s ≤ p I s\le pI s ≤ p I reads p I ( q − 1 ) ≤ 2 ( p + q ) pI(q-1)\le2(p+q) p I ( q − 1 ) ≤ 2 ( p + q ) , and the triangle is obtuse: $2(p+q)<pqI$. If q = 1 q=1 q = 1 , then p = 1 p=1 p = 1 ; the first inequality is 0 ≤ 4 0\le4 0 ≤ 4 , the second is I > 4 I>4 I > 4 , and the triangle is ( 1 , 1 , I − 2 ) / I (1,1,I-2)/I ( 1 , 1 , I − 2 ) / I with I = n ≥ 5 I=n\ge5 I = n ≥ 5 . If q ≥ 3 q\ge3 q ≥ 3 and p = 1 p=1 p = 1 : I ( q − 1 ) ≤ 2 q + 2 < I q I(q-1)\le2q+2<Iq I ( q − 1 ) ≤ 2 q + 2 < I q . The right half gives I > 2 I>2 I > 2 , so I ≥ 3 I\ge3 I ≥ 3 ; the left half gives I ≤ 2 + 4 / ( q − 1 ) I\le2+4/(q-1) I ≤ 2 + 4/ ( q − 1 ) , so q ≤ 5 q\le5 q ≤ 5 . For q = 3 q=3 q = 3 , I ≤ 4 I\le4 I ≤ 4 , so I = 3 I=3 I = 3 and the triangle is ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 (s = 1 ≤ 3 s=1\le3 s = 1 ≤ 3 ). For q = 5 q=5 q = 5 , I ≤ 3 I\le3 I ≤ 3 , so I = 3 I=3 I = 3 and the triangle is ( 1 , 5 , 9 ) / 15 (1,5,9)/15 ( 1 , 5 , 9 ) /15 (s = 3 ≤ 3 s=3\le3 s = 3 ≤ 3 ). If p ≥ 3 p\ge3 p ≥ 3 : p p p and q q q are coprime and odd with p ≤ q p\le q p ≤ q , so q ≥ p + 2 q\ge p+2 q ≥ p + 2 and 2 ( p + q ) ≤ 4 ( q − 1 ) 2(p+q)\le4(q-1) 2 ( p + q ) ≤ 4 ( q − 1 ) . Then p I ( q − 1 ) ≤ 4 ( q − 1 ) pI(q-1)\le4(q-1) p I ( q − 1 ) ≤ 4 ( q − 1 ) , so p = 3 p=3 p = 3 and I = 1 I=1 I = 1 , and the two inequalities read 3 ( q − 1 ) ≤ 6 + 2 q < 3 q 3(q-1)\le6+2q<3q 3 ( q − 1 ) ≤ 6 + 2 q < 3 q : 6 < q ≤ 9 6<q\le9 6 < q ≤ 9 with q q q odd and prime to 3, so q = 7 q=7 q = 7 and the triangle is ( 3 , 7 , 11 ) / 21 (3,7,11)/21 ( 3 , 7 , 11 ) /21 (s = 1 ≤ 3 s=1\le3 s = 1 ≤ 3 ). □$
Proof of Theorem 4.2 and Proposition 4.5. That p q pq pq divides n n n is Lemma 4.6 . If s > p I s > pI s > p I , Propositions 4.11 and 4.9 give a witness. If s ≤ p I s \le pI s ≤ p I , there is no odd t t t with ∣ t ∣ p I < s |t|pI < s ∣ t ∣ p I < s , and none with ∣ t ∣ q I < s |t|qI < s ∣ t ∣ q I < s , as q I ≥ p I qI \ge pI q I ≥ p I ; by Proposition 4.9 there is no witness. The triangles with s ≤ p I s \le pI s ≤ p I are those of Proposition 4.12 , and Proposition 4.5 is Proposition 4.9 (i) with Proposition 4.11 . □
Even denominators
Let n = 2 m n = 2m n = 2 m be even. Then s s s is even; set ς = s / 2 = r − m \varsigma= s/2 = r-m ς = s /2 = r − m , so that
p + q + ς = m , 1 ≤ ς ≤ m / 3 , L = n − 2 q = 2 p + 2 ς p+q+\varsigma=m,\qquad1\le\varsigma\le m/3,\qquad L=n-2q=2p+2\varsigma p + q + ς = m , 1 ≤ ς ≤ m /3 , L = n − 2 q = 2 p + 2 ς
and q ≥ ( m − ς ) / 2 ≥ m / 3 q \ge(m-\varsigma)/2 \ge m/3 q ≥ ( m − ς ) /2 ≥ m /3 , p < m / 2 p < m/2 p < m /2 . There is no window triangle for n = 2 , 4 n=2,4 n = 2 , 4 , so m ≥ 3 m \ge3 m ≥ 3 . Let J ( N ) J(N) J ( N ) be Jacobsthal’s function: the least J ≥ 1 J \ge1 J ≥ 1 such that any J J J consecutive integers contain one prime to N N N ; it depends only on the primes dividing N N N . If N ′ ∣ N N' \mid N N ′ ∣ N , then J ( N ′ ) ≤ J ( N ) J(N') \le J(N) J ( N ′ ) ≤ J ( N ) , since an integer prime to N N N is prime to N ′ N' N ′ ; and J ( N ) ≤ N J(N) \le N J ( N ) ≤ N , as N N N consecutive integers contain one ≡ 1 ( m o d N ) \equiv1 \pmod{N} ≡ 1 ( mod N ) . Let m ∘ m^\circ m ∘ be the product of the odd primes dividing m m m (m ∘ = 1 m^\circ=1 m ∘ = 1 if m m m is a power of 2 2 2 ), and write J = J ( m ∘ ) J=J(m^\circ) J = J ( m ∘ ) . Every other argument of J J J in this section and in Appendix § 9 is odd and squarefree. Let P ( m ) < P 2 ( m ) P(m)<P_2(m) P ( m ) < P 2 ( m ) be the two least odd primes not dividing m m m .
Lemma 4.13. Let N N N be odd and squarefree with k ≥ 1 k \ge1 k ≥ 1 prime factors. Then J ( N ) ≤ 3 J(N) \le3 J ( N ) ≤ 3 if k = 2 k=2 k = 2 , J ( N ) ≤ 3 k J(N) \le3^k J ( N ) ≤ 3 k in general, and J ( N ) ≤ ⌊ ( 2 k − 1 ) N / φ ( N ) ⌋ + 1 J(N) \le\lfloor(2^k-1)N/\varphi(N)\rfloor+1 J ( N ) ≤ ⌊( 2 k − 1 ) N / φ ( N )⌋ + 1 .
Proof. For k = 2 k=2 k = 2 : among three consecutive integers each prime ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 divides at most one. In general, by Lemma 4.10 with D = N D=N D = N and every class β ℓ = 0 \beta_\ell=0 β ℓ = 0 , among M M M consecutive integers more than M φ ( N ) / N − 2 k + 1 M\varphi(N)/N-2^k+1 M φ ( N ) / N − 2 k + 1 are prime to N N N ; this is ≥ 0 \ge0 ≥ 0 , and the number is therefore at least 1 1 1 , once M ≥ ( 2 k − 1 ) N / φ ( N ) M \ge(2^k-1)N/\varphi(N) M ≥ ( 2 k − 1 ) N / φ ( N ) , so J ( N ) ≤ ⌈ ( 2 k − 1 ) N / φ ( N ) ⌉ ≤ ⌊ ( 2 k − 1 ) N / φ ( N ) ⌋ + 1 J(N) \le\lceil(2^k-1)N/\varphi(N)\rceil\le\lfloor(2^k-1)N/\varphi(N)\rfloor+1 J ( N ) ≤ ⌈( 2 k − 1 ) N / φ ( N )⌉ ≤ ⌊( 2 k − 1 ) N / φ ( N )⌋ + 1 . Finally N / φ ( N ) ≤ ( 3 / 2 ) k N/\varphi(N) \le(3/2)^k N / φ ( N ) ≤ ( 3/2 ) k and ( 2 k − 1 ) ( 3 / 2 ) k + 1 ≤ 3 k (2^k-1)(3/2)^k+1 \le3^k ( 2 k − 1 ) ( 3/2 ) k + 1 ≤ 3 k . □
Theorem 4.14. Let n = 2 m n=2m n = 2 m , J = J ( m ∘ ) J=J(m^\circ) J = J ( m ∘ ) , P 2 = P 2 ( m ) P_2=P_2(m) P 2 = P 2 ( m ) , and suppose m > N 0 ( m ) = P 2 ( 31 J 2 − 1 ) m>N_0(m)=P_2(31J^2-1) m > N 0 ( m ) = P 2 ( 31 J 2 − 1 ) . If a window triangle ( p , q , r ) / n (p,q,r)/n ( p , q , r ) / n has no witness, then it lies in A A A , in B B B or in C C C .
The idea of the proof. The proof is in Appendix § 9 ; the table at the end of this subsection lists its cases and the results that settle them. Suppose that Δ \Delta Δ has no witness. Minkowski’s theorem and Jacobsthal’s function give p ≤ 7 J 2 − 1 p \le7J^2-1 p ≤ 7 J 2 − 1 , and so 7 L ≤ 3 n 7L \le3n 7 L ≤ 3 n (Step 1). Step 2 shows that S r S_r S r contains no usable unit. The odd elements of S r S_r S r fall into rows of g ς = gcd ( ς , m ) g_\varsigma=\gcd(\varsigma,m) g ς = g cd( ς , m ) residues spaced n / g ς n/g_\varsigma n / g ς apart (Proposition B.13); for g ς ≥ 2 g_\varsigma\ge2 g ς ≥ 2 , outside a few cases treated apart, the units of a row are spread along it and a usable one lies in S q S_q S q , a witness, and for g ς = 1 g_\varsigma=1 g ς = 1 and r r r even a witness comes from doubling. For g ς = 1 g_\varsigma=1 g ς = 1 and r r r odd, the products of r − 1 r-1 r − 1 with the powers of an odd prime u ∤ m u\nmid m u ∤ m with u L < n uL<n uL < n , a multiplier chain (Lemma B.5), give P 2 L ≥ n P_2L\ge n P 2 L ≥ n , and a rational approximation from Dirichlet’s theorem gives ς ≤ 24 J 2 \varsigma\le24J^2 ς ≤ 24 J 2 ; with p ≤ 7 J 2 − 1 p\le7J^2-1 p ≤ 7 J 2 − 1 this contradicts m > N 0 ( m ) m>N_0(m) m > N 0 ( m ) , and this is where the constant 31 = 7 + 24 31=7+24 31 = 7 + 24 comes from. Once S r S_r S r has no usable unit, either ς ∣ m \varsigma\mid m ς ∣ m with m / ς m/\varsigma m / ς odd, or ς = 1 \varsigma=1 ς = 1 with m m m even (Proposition B.13(b)), and Step 3 finds a witness in S p ∩ S q S_p\cap S_q S p ∩ S q , from the multiplier chain and the rows of S p S_p S p , for every such triangle outside A A A , B B B and C C C . The corresponding threshold of Larsen, Norton and Zykoski above 2 π / 3 2\pi/3 2 π /3 is the hypothesis 24 j ( n ) 2 < n 24j(n)^2<\sqrt n 24 j ( n ) 2 < n of [17 Lemma 7.9] ; here the multiplier chain replaces n \sqrt n n by n / P 2 ( m ) n/P_2(m) n / P 2 ( m ) .
Theorem 4.15. Let n n n be even. A window triangle of denominator n n n has no witness if and only if it lies in A \mathcal{A} A , B \mathcal{B} B or C \mathcal{C} C , or is a member of E \mathcal{E} E with even denominator (Table 1).
Proof. There is no window triangle for n = 2 , 4 n=2,4 n = 2 , 4 . Only if. By Proposition 4.3(c), a triangle without witness with s ∣ n s\mid n s ∣ n , $n/so d d a n d $ 2 ≤ p ≤ s / 2 odd and \$2\le p\le s/2 o dd an d $2 ≤ p ≤ s /2 lies in B \mathcal{B} B ; so it is enough to show that a window triangle of even denominator without witness lies in A \mathcal{A} A or C \mathcal{C} C , or has s ∣ n s\mid n s ∣ n , n / s n/s n / s odd and 2 ≤ p ≤ s / 2 2\le p\le s/2 2 ≤ p ≤ s /2 , or is a member of E \mathcal{E} E with even denominator. If m > N 0 ( m ) m>N_0(m) m > N 0 ( m ) , this is Theorem 4.14. The m ≥ 3 m\ge3 m ≥ 3 with m ≤ N 0 ( m ) m\le N_0(m) m ≤ N 0 ( m ) are 2,567 values, the largest 69,615 (Appendix A.§ 2 , where the list is proved complete). By Lemma B.4, which has no hypothesis on m m m , a window triangle of denominator 2 m 2m 2 m without witness has p q ≤ 4 m J 2 pq\le4mJ^2 pq ≤ 4 m J 2 . For the 2,567 values of m m m an enumeration of the window triangles of denominator 2 m 2m 2 m with p q ≤ 4 m J 2 pq\le4mJ^2 pq ≤ 4 m J 2 finds that each of them without witness lies in A \mathcal{A} A or C \mathcal{C} C , or has s ∣ n s\mid n s ∣ n , n / s n/s n / s odd and 2 ≤ p ≤ s / 2 2\le p\le s/2 2 ≤ p ≤ s /2 , or is a member of E \mathcal{E} E (Appendix A.§ 3 ). If. For A \mathcal{A} A and B \mathcal{B} B this is Proposition 4.3, and for $\mathcal{C} {citet}Proposition 4.16 below. The members of $\mathcal{E}$ have no witness by a direct check of their units ({xref}Appendix A.3`). □
Proposition 4.16. No member of C \mathcal{C} C has a witness.
Proof. Here 4 ∣ n 4\mid n 4 ∣ n , s = 2 s=2 s = 2 , r = n / 2 + 1 r=n/2+1 r = n /2 + 1 , q = n / 2 − 1 − p q=n/2-1-p q = n /2 − 1 − p , and p ≤ q p\le q p ≤ q forces n ≥ 4 p + 2 n\ge4p+2 n ≥ 4 p + 2 . S r S_r S r . r r r is odd and gcd ( r , n ) \gcd(r,n) g cd( r , n ) divides 2 r − n = 2 2r-n=2 2 r − n = 2 , so r r r is a unit and S r = { a : a r ≡ 0 or 1 ( m o d n ) } = { 0 , r − 1 } S_r=\{a: ar\equiv0\text{ or }1\pmod n\}=\{0,r-1\} S r = { a : a r ≡ 0 or 1 ( mod n )} = { 0 , r − 1 } . As r 2 = n 2 / 4 + n + 1 ≡ 1 ( m o d n ) r^2=n^2/4+n+1\equiv1\pmod n r 2 = n 2 /4 + n + 1 ≡ 1 ( mod n ) , r − 1 = r = n / 2 + 1 r^{-1}=r=n/2+1 r − 1 = r = n /2 + 1 , and 2 r ≡ 2 2r\equiv2 2 r ≡ 2 : S r S_r S r has no usable unit. S p S_p S p . For p = 1 p=1 p = 1 , S 1 = { 0 , 1 } S_1=\{0,1\} S 1 = { 0 , 1 } . For p = 2 p=2 p = 2 : [ 2 a ] n [2a]_n [ 2 a ] n is even, so [ 2 a ] n < 4 [2a]_n<4 [ 2 a ] n < 4 means 2 a ≡ 0 2a\equiv0 2 a ≡ 0 or 2 2 2 , that is a ∈ { 0 , n / 2 , 1 , n / 2 + 1 } a\in\{0,n/2,1,n/2+1\} a ∈ { 0 , n /2 , 1 , n /2 + 1 } ; 0 0 0 and n / 2 n/2 n /2 are not units, and 1 1 1 and n / 2 + 1 n/2+1 n /2 + 1 are not usable. For p = 4 p=4 p = 4 and n ≡ 4 ( m o d 8 ) n\equiv4\pmod8 n ≡ 4 ( mod 8 ) : [ 4 a ] n [4a]_n [ 4 a ] n is a multiple of 4 = gcd ( 4 , n ) 4=\gcd(4,n) 4 = g cd( 4 , n ) , so [ 4 a ] n < 8 [4a]_n<8 [ 4 a ] n < 8 means 4 a ≡ 0 4a\equiv0 4 a ≡ 0 or 4 4 4 , that is a ≡ 0 a\equiv0 a ≡ 0 or 1 ( m o d n / 4 ) 1\pmod{n/4} 1 ( mod n /4 ) ; the eight candidates are j n / 4 jn/4 j n /4 and j n / 4 + 1 jn/4+1 j n /4 + 1 , j = 0 , … , 3 j=0,\ldots,3 j = 0 , … , 3 . The j n / 4 jn/4 j n /4 are not units, as n / 4 > 1 n/4>1 n /4 > 1 divides them and n n n ; n / 4 + 1 n/4+1 n /4 + 1 and 3 n / 4 + 1 3n/4+1 3 n /4 + 1 are even, as n / 4 n/4 n /4 is odd; and 1 1 1 , n / 2 + 1 n/2+1 n /2 + 1 are not usable. So S p S_p S p has no usable unit. A witness lies in two of the three sets, hence in S p S_p S p or S r S_r S r : impossible. □
The statement on S p S_p S p is [17 Proposition 3.5] for x = 1 , 2 , 4 x=1,2,4 x = 1 , 2 , 4 .
Remark 4.17. Let 4 ∣ n 4\mid n 4 ∣ n and u = 1 + n / 2 u=1+n/2 u = 1 + n /2 . Then u u u is a unit, 2 u ≡ 2 2u\equiv2 2 u ≡ 2 and u 2 ≡ 1 ( m o d n ) u^2\equiv1\pmod n u 2 ≡ 1 ( mod n ) , so S u x = u S x S_{ux}=uS_x S ux = u S x and multiplication by u u u preserves usable units. If ς \varsigma ς is odd, exactly one of p p p , q q q is even, say e e e , the other o o o is odd, and [ u p ] n [up]_n [ u p ] n , [ u q ] n [uq]_n [ u q ] n , [ u r ] n [ur]_n [ u r ] n are e e e , o + n / 2 o+n/2 o + n /2 , ς \varsigma ς in some order, with sum n n n : a triangle u Δ u\Delta u Δ which has a witness if and only if Δ \Delta Δ has one, and whose unfolding is the same minimal surface as that of Δ \Delta Δ with the eigenform of character u u u . The members ( 2 , m − 3 , m + 1 ) / 2 m (2,m-3,m+1)/2m ( 2 , m − 3 , m + 1 ) /2 m (m ≥ 6 m\ge6 m ≥ 6 ) and ( 4 , m − 5 , m + 1 ) / 2 m (4,m-5,m+1)/2m ( 4 , m − 5 , m + 1 ) /2 m (m ≡ 2 ( m o d 4 ) m\equiv2\pmod4 m ≡ 2 ( mod 4 ) , m ≥ 10 m\ge10 m ≥ 10 ) of C \mathcal{C} C are sent to the triangles ( 1 , 2 , n − 3 ) / n (1,2,n-3)/n ( 1 , 2 , n − 3 ) / n and ( 1 , 4 , n − 5 ) / n (1,4,n-5)/n ( 1 , 4 , n − 5 ) / n of families (2) and (3) of Theorem 2.5; ( 3 , 4 , 9 ) / 16 (3,4,9)/16 ( 3 , 4 , 9 ) /16 , ( 5 , 8 , 15 ) / 28 (5,8,15)/28 ( 5 , 8 , 15 ) /28 , ( 8 , 9 , 23 ) / 40 (8,9,23)/40 ( 8 , 9 , 23 ) /40 to ( 1 , 4 , 11 ) / 16 (1,4,11)/16 ( 1 , 4 , 11 ) /16 , ( 1 , 8 , 19 ) / 28 (1,8,19)/28 ( 1 , 8 , 19 ) /28 , ( 3 , 8 , 29 ) / 40 (3,8,29)/40 ( 3 , 8 , 29 ) /40 of item (7).
Proof of Theorem 4.1. Part (a) is Theorem C.3 for odd n n n (C \mathcal{C} C is empty for odd n n n , and the members of E \mathcal{E} E with odd denominator are the five of Theorem C.3) and Theorem 4.15 for even n n n . Part (b) is Proposition 4.3(c), and its explicit form is Proposition 4.4. In the sentence after Theorem 4.1, the prime case is proved at the end of Appendix C, and the cases n = 2 ℓ n=2\ell n = 2 ℓ and p = 1 p=1 p = 1 follow from part (a). For n = 2 ℓ n=2\ell n = 2 ℓ , ℓ \ell ℓ an odd prime: C \mathcal{C} C needs 4 ∣ n 4\mid n 4 ∣ n ; a member of B \mathcal{B} B has s ∣ 2 ℓ s\mid2\ell s ∣ 2 ℓ with n / s n/s n / s odd and s ≥ 4 s\ge4 s ≥ 4 , so s = 2 ℓ ≥ n / 3 s=2\ell\ge n/3 s = 2 ℓ ≥ n /3 , which is impossible; A \mathcal{A} A needs s = 2 s=2 s = 2 , giving ( 1 , ℓ − 2 , ℓ + 1 ) / 2 ℓ (1,\ell-2,\ell+1)/2\ell ( 1 , ℓ − 2 , ℓ + 1 ) /2 ℓ ; and the members of E \mathcal{E} E with denominator twice an odd prime are ( 3 , 7 , 12 ) / 22 (3,7,12)/22 ( 3 , 7 , 12 ) /22 , ( 1 , 9 , 16 ) / 26 (1,9,16)/26 ( 1 , 9 , 16 ) /26 , ( 5 , 7 , 14 ) / 26 (5,7,14)/26 ( 5 , 7 , 14 ) /26 and ( 3 , 13 , 18 ) / 34 (3,13,18)/34 ( 3 , 13 , 18 ) /34 . For p = 1 p=1 p = 1 : a member of B \mathcal{B} B has p ≥ 2 p\ge2 p ≥ 2 , and the members of E \mathcal{E} E with p = 1 p=1 p = 1 are the six listed there. □
Table 4.
The families A \mathcal{A} A and B \mathcal{B} B
For a member of A \mathcal{A} A or B \mathcal{B} B write n = s d n=sd n = s d with d d d odd, d ≥ 3 d\ge3 d ≥ 3 ; then r = s ( d + 1 ) / 2 r=s(d+1)/2 r = s ( d + 1 ) /2 and p + q = s ( d − 1 ) / 2 p+q=s(d-1)/2 p + q = s ( d − 1 ) /2 . Since gcd ( ( d + 1 ) / 2 , d ) = 1 \gcd((d+1)/2,d)=1 g cd(( d + 1 ) /2 , d ) = 1 , gcd ( r , n ) = s \gcd(r,n)=s g cd( r , n ) = s : the s s s points above the obtuse vertex are zeros of ω \omega ω (r ∤ n r\nmid n r ∤ n ), and n / gcd ( r , n ) = d n/\gcd(r,n)=d n / g cd( r , n ) = d . Also gcd ( p , q , r ) = gcd ( p , s ) \gcd(p,q,r)=\gcd(p,s) g cd( p , q , r ) = g cd( p , s ) , since ( d − 1 ) / 2 (d-1)/2 ( d − 1 ) /2 and ( d + 1 ) / 2 (d+1)/2 ( d + 1 ) /2 are coprime.
Theorem 4.18. (a) Part (a) or part (b) of Theorem 1.2 applies to every member of A \mathcal{A} A other than ( 1 , 1 , 3 ) / 5 (1,1,3)/5 ( 1 , 1 , 3 ) /5 , ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 and ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 .
(b) Part (b) of Theorem 1.2 applies to every member of B \mathcal{B} B , at the obtuse vertex.
(c) Neither part applies to ( 1 , 1 , 3 ) / 5 (1,1,3)/5 ( 1 , 1 , 3 ) /5 , ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 or ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 .
Consequently no member of A ∪ B \mathcal{A}\cup\mathcal{B} A ∪ B other than ( 1 , 1 , 3 ) / 5 (1,1,3)/5 ( 1 , 1 , 3 ) /5 , ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 and ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 is a lattice triangle; the first two are, by Veech’s theorem; the third is not, by Theorem 6.1 .
Proof. A unit a ≡ 2 ( m o d d ) a\equiv2\pmod d a ≡ 2 ( mod d ) that is small on p p p gives a holomorphic form. Let a a a be a unit with a ≡ 2 ( m o d d ) a\equiv2\pmod d a ≡ 2 ( mod d ) and x = [ a p ] n < n − s x=[ap]_n<n-s x = [ a p ] n < n − s . Then a r = s ⋅ a ( d + 1 ) / 2 ar=s\cdot a(d+1)/2 a r = s ⋅ a ( d + 1 ) /2 and a ( d + 1 ) / 2 ≡ d + 1 ≡ 1 ( m o d d ) a(d+1)/2\equiv d+1\equiv1\pmod d a ( d + 1 ) /2 ≡ d + 1 ≡ 1 ( mod d ) , so [ a r ] n = s [ar]_n=s [ a r ] n = s ; and a q ≡ − a p − a r ≡ − ( x + s ) aq\equiv-ap-ar\equiv-(x+s) a q ≡ − a p − a r ≡ − ( x + s ) with 0 < x + s < n 0<x+s<n 0 < x + s < n , so [ a q ] n = n − x − s [aq]_n=n-x-s [ a q ] n = n − x − s . The three residues add up to n n n , so H a 1 , 0 ≠ 0 H^{1,0}_a\ne0 H a 1 , 0 = 0 by Lemma 3.1(b). We refer to this as ( ∗ ) (*) ( ∗ ) .
(b) Let Δ ∈ B \Delta\in\mathcal{B} Δ ∈ B . By Proposition 4.4, either gcd ( p , n ) = 1 \gcd(p,n)=1 g cd( p , n ) = 1 , and then the inverse of p p p modulo n n n is a usable unit of S p S_p S p (it is the element k p − 1 kp^{-1} k p − 1 with k = 1 ≠ p k=1\ne p k = 1 = p in the proof of Proposition 4.4(a)); or ( p , d ) = ( 3 , 3 ) (p,d)=(3,3) ( p , d ) = ( 3 , 3 ) , and then the unit ( 2 , 1 ) (2,1) ( 2 , 1 ) of Z / n = Z / 3 × Z / s \mathbb{Z}/n=\mathbb{Z}/3\times\mathbb{Z}/s Z / n = Z /3 × Z / s lies in S 3 S_3 S 3 (Proposition 4.4(b)) and is usable, since $2a\equiv 2\pmod n$ would give a ≡ 1 ( m o d 3 ) a\equiv1\pmod3 a ≡ 1 ( mod 3 ) , as 3 3 3 divides n / gcd ( 2 , n ) n/\gcd(2,n) n / g cd( 2 , n ) . So S p S_p S p contains a usable unit a a a , and by (U), a ≡ 2 ( m o d d ) a\equiv2\pmod d a ≡ 2 ( mod d ) . Also a ≠ 2 ( m o d n ) a\ne2\pmod n a = 2 ( mod n ) , because [ 2 p ] n = 2 p [2p]_n=2p [ 2 p ] n = 2 p is not less than 2 p 2p 2 p , and [ a p ] n < 2 p ≤ s ≤ n − s [ap]_n<2p\le s\le n-s [ a p ] n < 2 p ≤ s ≤ n − s . By ( ∗ ) (*) ( ∗ ) , H a 1 , 0 ≠ 0 H^{1,0}_a\ne0 H a 1 , 0 = 0 . As gcd ( r , n ) = s ≥ 2 p > 1 \gcd(r,n)=s\ge2p>1 g cd( r , n ) = s ≥ 2 p > 1 , r ∤ n r\nmid n r ∤ n and n / s = d n/s=d n / s = d , part (b) applies at the vertex r r r with $b=a.
(a), n n n odd. Here s s s is odd. Part (a) applies unless q ∣ n q\mid n q ∣ n , as p = 1 p=1 p = 1 . Since q = s ( d − 1 ) / 2 − 1 q=s(d-1)/2-1 q = s ( d − 1 ) /2 − 1 is prime to s s s , q ∣ n q\mid n q ∣ n means q ∣ d q\mid d q ∣ d ; as d d d is odd, either q = d q=d q = d or q ≤ d / 3 q\le d/3 q ≤ d /3 . If q = d q=d q = d , then s ( d − 1 ) = 2 ( d + 1 ) s(d-1)=2(d+1) s ( d − 1 ) = 2 ( d + 1 ) , s = 2 + 4 / ( d − 1 ) s=2+4/(d-1) s = 2 + 4/ ( d − 1 ) , and s s s odd forces d = 5 d=5 d = 5 , s = 3 s=3 s = 3 : the triangle ( 1 , 5 , 9 ) / 15 (1,5,9)/15 ( 1 , 5 , 9 ) /15 .
If q ≤ d / 3 q\le d/3 q ≤ d /3 , then s ( d − 1 ) ≤ 2 d / 3 + 2 s(d-1)\le2d/3+2 s ( d − 1 ) ≤ 2 d /3 + 2 , which forces s = 1 s=1 s = 1 and d ≤ 9 d\le9 d ≤ 9 , with q = ( d − 3 ) / 2 q=(d-3)/2 q = ( d − 3 ) /2 : d = 5 d=5 d = 5 gives ( 1 , 1 , 3 ) / 5 (1,1,3)/5 ( 1 , 1 , 3 ) /5 , d = 9 d=9 d = 9 gives ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 , d = 7 d=7 d = 7 gives q = 2 ∤ 7 q=2\nmid7 q = 2 ∤ 7 , and d = 3 d=3 d = 3 gives no triangle. On ( 1 , 5 , 9 ) / 15 (1,5,9)/15 ( 1 , 5 , 9 ) /15 part (b) applies at the vertex 9 9 9 (gcd ( 9 , 15 ) = 3 \gcd(9,15)=3 g cd( 9 , 15 ) = 3 ) with b = 7 b=7 b = 7 : [ 7 ] 15 + [ 35 ] 15 + [ 63 ] 15 = 7 + 5 + 3 = 15 [7]_{15}+[35]_{15}+[63]_{15}=7+5+3=15 [ 7 ] 15 + [ 35 ] 15 + [ 63 ] 15 = 7 + 5 + 3 = 15 .
(a), n n n even. Here s s s is even. Take b = 2 − k d b=2-kd b = 2 − k d with 1 ≤ k ≤ s − 1 1\le k\le s-1 1 ≤ k ≤ s − 1 ; then b ≡ 2 ( m o d d ) b\equiv2\pmod d b ≡ 2 ( mod d ) , b ≢ 2 ( m o d n ) b\not\equiv2\pmod n b ≡ 2 ( mod n ) , and [ b ] n = n + 2 − k d [b]_n=n+2-kd [ b ] n = n + 2 − k d , which is less than n − s n-s n − s exactly when k d > s + 2 kd>s+2 k d > s + 2 ; then ( ∗ ) (*) ( ∗ ) gives H b 1 , 0 ≠ 0 H_b^{1,0}\ne0 H b 1 , 0 = 0 , once b b b is a unit. As gcd ( k d − 2 , d ) = gcd ( 2 , d ) = 1 \gcd(kd-2,d)=\gcd(2,d)=1 g cd( k d − 2 , d ) = g cd( 2 , d ) = 1 , b b b is a unit exactly when gcd ( k d − 2 , s ) = 1 \gcd(kd-2,s)=1 g cd( k d − 2 , s ) = 1 , that is, when k k k is odd and k ≢ 2 d − 1 ( m o d ℓ ) k\not\equiv2d^{-1}\pmod\ell k ≡ 2 d − 1 ( mod ℓ ) for every odd prime ℓ ∣ s \ell\mid s ℓ ∣ s with ℓ ∤ d \ell\nmid d ℓ ∤ d (for ℓ ∣ d \ell\mid d ℓ ∣ d , k d − 2 ≡ − 2 kd-2\equiv-2 k d − 2 ≡ − 2 ). Call such k k k valid. Validity depends only on k k k modulo s ∗ s^* s ∗ , the product of the primes dividing s s s and not d d d (so 2 ∣ s ∗ 2\mid s^* 2 ∣ s ∗ ); by the Chinese remainder theorem valid classes exist modulo every divisor of s ∗ s^* s ∗ , and k = s k=s k = s is not valid, as s s s is even. We need a valid k k k with ( s + 2 ) / d < k < s (s+2)/d<k<s ( s + 2 ) / d < k < s .
s = 2 s=2 s = 2 : k = 1 k=1 k = 1 is valid, and d > 4 d>4 d > 4 is needed; every d ≥ 5 d\ge5 d ≥ 5 works, and ( s , d ) = ( 2 , 3 ) (s,d)=(2,3) ( s , d ) = ( 2 , 3 ) is ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 .
s ≥ 4 s\ge4 s ≥ 4 and s ∗ < s s^*<s s ∗ < s : then s ≥ 2 s ∗ s\ge2s^* s ≥ 2 s ∗ , and the s ∗ s^* s ∗ consecutive integers in ( s − s ∗ , s ] (s-s^*,s] ( s − s ∗ , s ] contain a valid k k k , which is less than s s s ; it satisfies k > s / 2 ≥ ( s + 2 ) / 3 ≥ ( s + 2 ) / d k>s/2\ge(s+2)/3\ge(s+2)/d k > s /2 ≥ ( s + 2 ) /3 ≥ ( s + 2 ) / d .
s = s ∗ > 2 s=s^*>2 s = s ∗ > 2 , so s s s is squarefree and prime to d d d : let ℓ \ell ℓ be the least odd prime dividing s s s , and k 1 ∈ [ 1 , s / ℓ ] k_1\in[1,s/\ell] k 1 ∈ [ 1 , s / ℓ ] a class valid for the primes of s / ℓ s/\ell s / ℓ . The ℓ \ell ℓ numbers k 1 + j s / ℓ k_1+js/\ell k 1 + j s / ℓ , 0 ≤ j < ℓ 0\le j<\ell 0 ≤ j < ℓ , are valid for the primes of s / ℓ s/\ell s / ℓ and run over all classes modulo ℓ \ell ℓ , of which exactly one is not valid; so each class modulo s / ℓ s/\ell s / ℓ valid for its primes contains ℓ − 1 \ell-1 ℓ − 1 valid classes modulo s s s , and j = ℓ − 1 j=\ell-1 j = ℓ − 1 or j = ℓ − 2 j=\ell-2 j = ℓ − 2 gives a valid k > ( ℓ − 2 ) s / ℓ k>(\ell-2)s/\ell k > ( ℓ − 2 ) s / ℓ , with k < s k<s k < s . If ℓ = 3 \ell=3 ℓ = 3 , then 3 ∤ d 3\nmid d 3 ∤ d , so d ≥ 5 d\ge5 d ≥ 5 and k > s / 3 ≥ ( s + 2 ) / 5 k>s/3\ge(s+2)/5 k > s /3 ≥ ( s + 2 ) /5 ; if ℓ ≥ 5 \ell\ge5 ℓ ≥ 5 , then k > 3 s / 5 ≥ ( s + 2 ) / 3 k>3s/5\ge(s+2)/3 k > 3 s /5 ≥ ( s + 2 ) /3 .
So part (b) applies at the vertex r r r (gcd ( r , n ) = s > 1 \gcd(r,n)=s>1 g cd( r , n ) = s > 1 , n / s = d n/s=d n / s = d ) to every member of A \mathcal{A} A with n n n even except ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 .
(c) On ( 1 , 1 , 3 ) / 5 (1,1,3)/5 ( 1 , 1 , 3 ) /5 both p p p and q q q divide n n n and every gcd ( x , n ) \gcd(x,n) g cd( x , n ) is 1 1 1 . On ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 , n n n is even, and at the vertex 4 4 4 (gcd ( 4 , 6 ) = 2 \gcd(4,6)=2 g cd( 4 , 6 ) = 2 ) the only unit b ≡ 2 ( m o d 3 ) b\equiv2\pmod3 b ≡ 2 ( mod 3 ) with b ≢ 2 ( m o d 6 ) b\not\equiv2\pmod6 b ≡ 2 ( mod 6 ) is b = 5 b=5 b = 5 , for which [ 5 ] 6 + [ 5 ] 6 + [ 20 ] 6 = 12 ≠ 6 [5]_6+[5]_6+[20]_6=12\ne6 [ 5 ] 6 + [ 5 ] 6 + [ 20 ] 6 = 12 = 6 . On ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 the numerators 1 1 1 and 3 3 3 divide 9 9 9 , and gcd ( 5 , 9 ) = 1 \gcd(5,9)=1 g cd( 5 , 9 ) = 1 . □
Cylinder families
In this section we show by Lemma 2.3 that several one-parameter families of obtuse triangles are not lattice triangles, exhibiting two parallel maximal cylinders whose moduli have an irrational ratio, or one maximal cylinder whose share of the area has a Galois conjugate outside ( 0 , 1 ] (0,1] ( 0 , 1 ] . The model of Lemma 5.1 with ( p , q ) = ( 4 , 1 ) (p,q)=(4,1) ( p , q ) = ( 4 , 1 ) is the unfolding drawn as a regular n n n -gon with four regular ( n / 4 ) (n/4) ( n /4 ) -gons attached. With Larsén, Norton and Zykoski find the two cylinders of family (3) [17 §10] , after [11 Figure 1] ; the star of Section 5.3 is the model in which Ward finds the two cylinders of [31 Theorem B] . Theorems 5.8 and 5.9 use an irrational ratio of moduli in the first model for p = 2 p=2 p = 2 , Theorems 5.15 and 5.16 one in the star, and Theorems 5.10 and 5.11 a share of the area in the first model for p = 4 p=4 p = 4 .
Two families with an angle 2 π / n 2\pi/n 2 π / n
In this subsection Δ = P Q R \Delta=PQR Δ = P QR has angles p ψ p\psi p ψ , q ψ q\psi q ψ , r ψ r\psi r ψ at P P P , Q Q Q , R R R , where ψ = π / n \psi=\pi/n ψ = π / n (the order of p , q , r p,q,r p , q , r is not assumed). We use complex coordinates P = 0 P=0 P = 0 , Q = 1 Q=1 Q = 1 , R = ∣ P R ∣ e i φ R=|PR|e^{i\varphi} R = ∣ P R ∣ e i φ , where by the law of sines ∣ P R ∣ = sin q ψ / sin r ψ |PR|=\sin q\psi/\sin r\psi ∣ P R ∣ = sin q ψ / sin r ψ and ∣ Q R ∣ = sin p ψ / sin r ψ |QR|=\sin p\psi/\sin r\psi ∣ QR ∣ = sin p ψ / sin r ψ . Let Θ k \Theta_k Θ k be the rotation by 2 k ψ 2k\psi 2 k ψ (k ∈ Z / n k\in\mathbb{Z}/n k ∈ Z / n ) and κ \kappa κ complex conjugation. The dihedral group D = { Θ k , Θ k κ } D=\{\Theta_k,\Theta_k\kappa\} D = { Θ k , Θ k κ } is generated by the linear parts of the reflections in the sides, κ P Q = κ \kappa_{PQ}=\kappa κ P Q = κ , κ P R = Θ p κ \kappa_{PR}=\Theta_p\kappa κ P R = Θ p κ , κ Q R = Θ − q κ \kappa_{QR}=\Theta_{-q}\kappa κ QR = Θ − q κ (the reflection in a line of direction ϕ \phi ϕ is z ↦ e 2 i ϕ z ˉ z\mapsto e^{2i\phi}\bar{z} z ↦ e 2 i ϕ z ˉ ).
Lemma 5.1. Assume p ∣ n p\mid n p ∣ n , n / p ≥ 3 n/p\ge3 n / p ≥ 3 , p < n / 2 p<n/2 p < n /2 and q < n / 2 q<n/2 q < n /2 . Then X X X is the union of the following pieces, glued by translations as stated.
For i ∈ Z / p i\in\mathbb{Z}/p i ∈ Z / p , the regular ( n / p ) (n/p) ( n / p ) -gon G i G_i G i with vertices Θ k ( R ) \Theta_k(R) Θ k ( R ) , k ≡ i ( m o d p ) k\equiv i\pmod p k ≡ i ( mod p ) , centre 0 0 0 and circumradius ∣ P R ∣ |PR| ∣ P R ∣ ; its side from Θ k ( R ˉ ) = Θ k − p ( R ) \Theta_k(\bar{R})=\Theta_{k-p}(R) Θ k ( R ˉ ) = Θ k − p ( R ) to Θ k ( R ) \Theta_k(R) Θ k ( R ) is called side k k k , and its outward normal has direction 2 k ψ 2k\psi 2 k ψ .
For k ∈ Z / n k\in\mathbb{Z}/n k ∈ Z / n , the isosceles triangle (ear) E k E_k E k with vertices Θ k ( R ˉ ) \Theta_k(\bar{R}) Θ k ( R ˉ ) , Θ k ( Q ) \Theta_k(Q) Θ k ( Q ) , Θ k ( R ) \Theta_k(R) Θ k ( R ) ; its apex is O k = Θ k ( Q ) = e 2 i k ψ O_k=\Theta_k(Q)=e^{2ik\psi} O k = Θ k ( Q ) = e 2 ik ψ , its apex angle is 2 q ψ 2q\psi 2 q ψ and its legs have length ∣ Q R ∣ |QR| ∣ QR ∣ .
The base Θ k ( R R ˉ ) \Theta_k(R\bar{R}) Θ k ( R R ˉ ) of E k E_k E k is glued to side k k k of G k mod p G_{k\mathbin{\operatorname{mod}}p} G k mod p , and the leg Θ k ( Q R ) \Theta_k(QR) Θ k ( QR ) of E k E_k E k to the leg Θ k − q ( Q R ˉ ) \Theta_{k-q}(Q\bar{R}) Θ k − q ( Q R ˉ ) of E k − q E_{k-q} E k − q .
The centres of the G i G_i G i are the points over P P P , the apexes are the points over Q Q Q , and the vertices of the G i G_i G i are the points over R R R . At its apex, E k E_k E k occupies the directions from ( 2 k + n − q ) ψ (2k+n-q)\psi ( 2 k + n − q ) ψ to ( 2 k + n + q ) ψ (2k+n+q)\psi ( 2 k + n + q ) ψ ; across its leg of direction ( 2 k + n − q ) ψ (2k+n-q)\psi ( 2 k + n − q ) ψ lies E k − q E_{k-q} E k − q , across the other E k + q E_{k+q} E k + q .
Proof. For k ∈ Z / n k\in\mathbb{Z}/n k ∈ Z / n let F k = Θ k Δ ∪ Θ k κ Δ F_k=\Theta_k\Delta\cup\Theta_k\kappa\Delta F k = Θ k Δ ∪ Θ k κ Δ . Since κ P Q = κ \kappa_{PQ}=\kappa κ P Q = κ , these two copies are glued along Θ k ( P Q ) \Theta_k(PQ) Θ k ( P Q ) , and F k F_k F k is the quadrilateral Θ k ( P R Q R ˉ ) \Theta_k(PRQ\bar{R}) Θ k ( P R Q R ˉ ) . Its angles are 2 p ψ 2p\psi 2 p ψ at P P P , 2 q ψ 2q\psi 2 q ψ at Q Q Q and r ψ r\psi r ψ at R R R and R ˉ \bar{R} R ˉ , all less than π \pi π , so F k F_k F k is convex and its diagonal Θ k ( R R ˉ ) \Theta_k(R\bar{R}) Θ k ( R R ˉ ) cuts it into the triangle Θ k ( P R R ˉ ) \Theta_k(PR\bar{R}) Θ k ( P R R ˉ ) and the ear E k E_k E k . Each element of D D D is Θ k \Theta_k Θ k or Θ k κ \Theta_k\kappa Θ k κ for exactly one k k k , so X X X is the union of the n n n quadrilaterals F k F_k F k , and the gluings along the copies of P Q PQ P Q are used inside them. The side Θ k ( P R ) \Theta_k(PR) Θ k ( P R ) of Θ k Δ \Theta_k\Delta Θ k Δ is glued to the copy Θ k κ P R Δ = Θ k + p κ Δ ⊂ F k + p \Theta_k\kappa_{PR}\Delta=\Theta_{k+p}\kappa\Delta\subset F_{k+p} Θ k κ P R Δ = Θ k + p κ Δ ⊂ F k + p along Θ k + p κ ( P R ) = Θ k + p ( P R ˉ ) \Theta_{k+p}\kappa(PR)=\Theta_{k+p}(P\bar{R}) Θ k + p κ ( P R ) = Θ k + p ( P R ˉ ) , which is the same segment from 0 0 0 to Θ k ( R ) = Θ k + p ( R ˉ ) \Theta_k(R)=\Theta_{k+p}(\bar{R}) Θ k ( R ) = Θ k + p ( R ˉ ) . Hence F i , F i + p , … , F i + ( n / p − 1 ) p F_i,F_{i+p},\ldots,F_{i+(n/p-1)p} F i , F i + p , … , F i + ( n / p − 1 ) p are glued cyclically around one point over P P P , each with angle 2 p ψ 2p\psi 2 p ψ , in total 2 π 2\pi 2 π ; they form a star centred at 0 0 0 , and the triangles Θ k ( P R R ˉ ) \Theta_k(PR\bar{R}) Θ k ( P R R ˉ ) , k ≡ i ( m o d p ) k\equiv i\pmod p k ≡ i ( mod p ) , tile G i G_i G i , whose sides are the bases of the ears. The side Θ k ( Q R ) \Theta_k(QR) Θ k ( QR ) of Θ k Δ \Theta_k\Delta Θ k Δ is glued to Θ k κ Q R Δ = Θ k − q κ Δ ⊂ F k − q \Theta_k\kappa_{QR}\Delta=\Theta_{k-q}\kappa\Delta\subset F_{k-q} Θ k κ QR Δ = Θ k − q κ Δ ⊂ F k − q along Θ k − q κ ( Q R ) = Θ k − q ( Q R ˉ ) \Theta_{k-q}\kappa(QR)=\Theta_{k-q}(Q\bar{R}) Θ k − q κ ( QR ) = Θ k − q ( Q R ˉ ) , a leg of E k − q E_{k-q} E k − q ; the side Θ k κ ( Q R ) \Theta_k\kappa(QR) Θ k κ ( QR ) of Θ k κ Δ \Theta_k\kappa\Delta Θ k κ Δ is glued to Θ k + q Δ \Theta_{k+q}\Delta Θ k + q Δ along Θ k + q ( Q R ) \Theta_{k+q}(QR) Θ k + q ( QR ) , which is the same rule for k + q k+q k + q . Every copy has three sides, so these are all the gluings. Finally Θ k ( R − Q ) \Theta_k(R-Q) Θ k ( R − Q ) has direction 2 k ψ + π − q ψ 2k\psi+\pi-q\psi 2 k ψ + π − q ψ and Θ k ( R ˉ − Q ) \Theta_k(\bar{R}-Q) Θ k ( R ˉ − Q ) has direction 2 k ψ + π + q ψ 2k\psi+\pi+q\psi 2 k ψ + π + q ψ .
From now on in this subsection assume, besides the hypotheses of Lemma 5.1, that n n n is even and q q q is odd. Horizontal means the direction of P Q PQ P Q , and leaves are oriented by increasing x x x . For an apex O O O and a leaf not through O O O , the height above O O O of the leaf is its y y y -coordinate in the chart of an ear at O O O with O O O at the origin; let χ = ( height ) / ∣ Q R ∣ \chi=(\text{height})/|QR| χ = ( height ) /∣ QR ∣ . The legs at an apex have directions ( 2 k + n ± q ) ψ (2k+n\pm q)\psi ( 2 k + n ± q ) ψ , odd multiples of ψ \psi ψ since n + q n+q n + q is odd. For an odd integer i i i write v ( i ) v(i) v ( i ) for the endpoint (a point over R R R ) of the leg of direction i ψ i\psi i ψ at the apex under consideration, and let
I χ = { i o d d : 0 < i < n , sin ( i ψ ) > χ } , ι ( i ) = n − i . I_\chi=\{i\ \mathrm{odd}:0<i<n,\ \sin(i\psi)>\chi\},\qquad\iota(i)=n-i. I χ = { i odd : 0 < i < n , sin ( i ψ ) > χ } , ι ( i ) = n − i .
The map ι \iota ι preserves I χ I_\chi I χ . For an odd i i i with − n < i < 2 n -n<i<2n − n < i < 2 n , sin ( i ψ ) > χ ≥ 0 \sin(i\psi)>\chi\ge0 sin ( i ψ ) > χ ≥ 0 already forces 0 < i < n 0<i<n 0 < i < n . A chain of I χ I_\chi I χ is a set W = { i b o t , i b o t + 2 q , … , i t o p } ⊂ I χ W=\{i_{\mathrm{bot}},i_{\mathrm{bot}}+2q,\ldots,i_{\mathrm{top}}\}\subset I_\chi W = { i bot , i bot + 2 q , … , i top } ⊂ I χ with i t o p + 2 q ∉ I χ i_{\mathrm{top}}+2q\notin I_\chi i top + 2 q ∈ / I χ and i b o t − 2 q ∉ I χ i_{\mathrm{bot}}-2q\notin I_\chi i bot − 2 q ∈ / I χ (Figure 1 ); then ι ( W ) \iota(W) ι ( W ) is a chain, with top n − i b o t n-i_{\mathrm{bot}} n − i bot and bottom n − i t o p n-i_{\mathrm{top}} n − i top . Since 2 q < n 2q<n 2 q < n , we have i t o p + 2 q < 2 n i_{\mathrm{top}}+2q<2n i top + 2 q < 2 n and i b o t − 2 q > − n i_{\mathrm{bot}}-2q>-n i bot − 2 q > − n . A value of χ \chi χ is critical if it equals sin ( i ψ ) \sin(i\psi) sin ( i ψ ) for some integer i i i .
Figure 1. The model of Lemma 5.1 for (2, 3, 11)/16: the octagons G 0 G_0 G 0 , G 1 G_1 G 1 with their ears E k E_k E k . A leg is labelled i i i when its direction from the apex is i ψ i\psi i ψ ; legs with the same label are glued. Shaded: the cylinder Z ( W ) Z(W) Z ( W ) of the chain W = { 1 , 7 , 13 } W = \{1, 7, 13\} W = { 1 , 7 , 13 } (Theorem 5.9 , q = 3 q = 3 q = 3 ); its lower boundary passes through the apex O O O .
A horizontal leaf that meets no vertex and enters the ears around an apex O O O through a base, at height ∣ Q R ∣ χ |QR|\chi ∣ QR ∣ χ above O O O , turns clockwise about O O O , crossing one leg after another while the far end v ( j ) v(j) v ( j ) of the next leg lies above it, that is, while sin ( j ψ ) > χ \sin(j\psi)>\chi sin ( j ψ ) > χ , and then leaves through a base. A chain of I χ \mathcal{I}_{\chi} I χ is the set of indices of the legs crossed in one such passage (Lemma 5.2 , Figure 1 ). If the polygon the leaf crosses next is symmetric, it re-enters the ears at the next apex at the same height and crosses the legs of ι ( W ) \iota(W) ι ( W ) , and if the polygon after that is symmetric too, it is back at its starting point (Lemmas 5.3 and 5.4 ; after one passage if ι ( W ) = W \iota(W) = W ι ( W ) = W ). Since the chains stay the same while χ \chi χ moves in an open interval, these closed leaves, all of one length, fill an annulus whose upper boundary leaf runs into a point over R R R at which the chain changes, and whose lower boundary leaf runs into such a point or, when a ( W ) = 0 a(W) = 0 a ( W ) = 0 , into the apex O O O . It is a maximal cylinder when those points are singular (Proposition 5.5 ).
Lemma 5.2. Let y > 0 y > 0 y > 0 be a height with χ = y / ∣ Q R ∣ \chi= y/|QR| χ = y /∣ QR ∣ not critical, and suppose a horizontal leaf crosses the base of E k E_k E k from G k m o d p G_{k \bmod p} G k mod p into E k E_k E k at height y y y above O = O k O = O_k O = O k . Let i 1 i_1 i 1 be the representative of 2 k + n − q 2k + n - q 2 k + n − q in ( 0 , 2 n ) (0, 2n) ( 0 , 2 n ) , so that the clockwise leg of E k E_k E k has direction i 1 ψ i_1\psi i 1 ψ . Then 0 < i 1 < n 0 < i_1 < n 0 < i 1 < n , i 1 ∈ I χ i_1 \in\mathcal{I}_{\chi} i 1 ∈ I χ and i 1 + 2 q ∉ I χ i_1 + 2q \notin\mathcal{I}_{\chi} i 1 + 2 q ∈ / I χ ; the leaf crosses, in this order, the legs of directions i 1 ψ , ( i 1 − 2 q ) ψ , … , i N ψ i_1\psi, (i_1 - 2q)\psi, \ldots, i_N\psi i 1 ψ , ( i 1 − 2 q ) ψ , … , i N ψ , where { i N , … , i 1 } \{i_N, \ldots, i_1\} { i N , … , i 1 } is the chain of I χ \mathcal{I}_{\chi} I χ with top i 1 i_1 i 1 ; and it then leaves the ears around O O O through the base of the ear E k ′ E_{k'} E k ′ with 2 k ′ ≡ i N − n − q ( m o d 2 n ) 2k' \equiv i_N - n - q \pmod{2n} 2 k ′ ≡ i N − n − q ( mod 2 n ) , at height y y y above O O O , into G k ′ m o d p G_{k' \bmod p} G k ′ mod p .
Proof. Develop the ears around O O O into the plane, O O O at the origin, one ear after the other across their common legs. The leaf develops into a segment of the line { y = const > 0 } \{y = \operatorname{const} > 0\} { y = const > 0 } , along which the polar angle decreases strictly as x x x increases and stays in ( 0 , π ) (0, \pi) ( 0 , π ) ; since this range of angles is shorter than π \pi π , the development meets each ear at most once and is injective. Inside an ear (the triangle O , v ( j ) , v ( j + 2 q ) O, v(j), v(j + 2q) O , v ( j ) , v ( j + 2 q ) , with the sector [ j ψ , ( j + 2 q ) ψ ] [j\psi, (j + 2q)\psi] [ j ψ , ( j + 2 q ) ψ ] ) the line enters through one side and leaves through another. In E k E_k E k the leaf enters through the base; the other side it crosses has smaller polar angle, so it is the clockwise leg, of direction i 1 ψ i_1\psi i 1 ψ . The line separates v ( i 1 ) v(i_1) v ( i 1 ) (above it) from O O O (below), so ∣ Q R ∣ sin ( i 1 ψ ) > y |QR|\sin(i_1\psi) > y ∣ QR ∣ sin ( i 1 ψ ) > y and i 1 ψ ∈ ( 0 , π ) i_1\psi\in(0, \pi) i 1 ψ ∈ ( 0 , π ) ; it also separates v ( i 1 + 2 q ) v(i_1 + 2q) v ( i 1 + 2 q ) from v ( i 1 ) v(i_1) v ( i 1 ) , so sin ( ( i 1 + 2 q ) ψ ) < χ \sin((i_1 + 2q)\psi) < \chi sin (( i 1 + 2 q ) ψ ) < χ . After crossing a leg of direction j ψ j\psi j ψ the leaf is in the next ear clockwise, with sector [ ( j − 2 q ) ψ , j ψ ] [(j - 2q)\psi, j\psi] [( j − 2 q ) ψ , j ψ ] , entered through its counterclockwise leg; it leaves through the clockwise leg, of direction ( j − 2 q ) ψ (j - 2q)\psi ( j − 2 q ) ψ , if v ( j − 2 q ) v(j-2q) v ( j − 2 q ) lies strictly above the line, that is, if j − 2 q ∈ I χ j-2q \in I_{\chi} j − 2 q ∈ I χ , and through the base otherwise (χ \chi χ is not critical, so the leaf meets no vertex). The ear left through its base has sector [ ( i N − 2 q ) ψ , i N ψ ] [(i_N-2q)\psi,i_N\psi] [( i N − 2 q ) ψ , i N ψ ] , so 2 k ′ + n + q ≡ i N ( m o d 2 n ) 2k' + n + q \equiv i_N \pmod{2n} 2 k ′ + n + q ≡ i N ( mod 2 n ) .
Lemma 5.3. The polygon G j G_j G j is symmetric under ( x , y ) ↦ ( − x , y ) (x,y) \mapsto(-x,y) ( x , y ) ↦ ( − x , y ) about its centre if and only if 2 j ≡ n / 2 ( m o d p ) 2j \equiv n/2 \pmod p 2 j ≡ n /2 ( mod p ) ; for p = 2 p=2 p = 2 this holds for both polygons if and only if 4 ∣ n 4 \mid n 4 ∣ n . If G j G_j G j is symmetric, a horizontal leaf entering G j G_j G j through side k ′ k' k ′ leaves it through side k ′ ′ = n / 2 − k ′ k''=n/2-k' k ′′ = n /2 − k ′ , its heights above O k ′ O_{k'} O k ′ and above O k ′ ′ O_{k''} O k ′′ are equal, and in the chart of G j G_j G j we have O k ′ ′ − O k ′ = ( − 2 cos 2 k ′ ψ , 0 ) O_{k''}-O_{k'}=(-2\cos2k'\psi,0) O k ′′ − O k ′ = ( − 2 cos 2 k ′ ψ , 0 ) .
Proof. The reflection maps the direction ϕ \phi ϕ to π − ϕ \pi-\phi π − ϕ , so the outward normal 2 k ψ 2k\psi 2 k ψ to 2 ( n / 2 − k ) ψ 2(n/2-k)\psi 2 ( n /2 − k ) ψ , and the set { k ≡ j ( m o d p ) } \{k\equiv j\pmod p\} { k ≡ j ( mod p )} is preserved if and only if n / 2 − j ≡ j ( m o d p ) n/2-j\equiv j\pmod p n /2 − j ≡ j ( mod p ) . For p = 2 p=2 p = 2 and j = 0 , 1 j=0,1 j = 0 , 1 this says that n / 2 n/2 n /2 is even. A horizontal line meets a symmetric convex polygon in a segment symmetric about the vertical axis, whose ends lie on mirror sides k ′ k' k ′ and n / 2 − k ′ n/2-k' n /2 − k ′ . Finally e 2 i k ′ ′ ψ = e i ( π − 2 k ′ ψ ) = − cos 2 k ′ ψ + i sin 2 k ′ ψ e^{2ik''\psi}=e^{i(\pi-2k'\psi)}=-\cos2k'\psi+i\sin2k'\psi e 2 i k ′′ ψ = e i ( π − 2 k ′ ψ ) = − cos 2 k ′ ψ + i sin 2 k ′ ψ .
Lemma 5.4. In Lemma 5.2, if G k ′ m o d p G_{k'\bmod p} G k ′ mod p is symmetric, the leaf next enters E k ′ ′ E_{k''} E k ′′ , k ′ ′ = n / 2 − k ′ k''=n/2-k' k ′′ = n /2 − k ′ , through its base at the same height y y y above O k ′ ′ O_{k''} O k ′′ , and the first leg it crosses there has direction ( n − i N ) ψ (n-i_N)\psi ( n − i N ) ψ ; so its next passage traverses ι ( W ) \iota(W) ι ( W ) , where W W W is the chain of the first passage.
Proof. We have 2 k ′ ′ + n − q = 2 n − 2 k ′ − q ≡ 2 n − ( i N − n − q ) − q = 3 n − i N ≡ n − i N ( m o d 2 n ) 2k''+n-q=2n-2k'-q\equiv2n-(i_N-n-q)-q=3n-i_N\equiv n-i_N\pmod{2n} 2 k ′′ + n − q = 2 n − 2 k ′ − q ≡ 2 n − ( i N − n − q ) − q = 3 n − i N ≡ n − i N ( mod 2 n ) , and 0 < n − i N < n 0<n-i_N<n 0 < n − i N < n . Apply Lemma 5.2 to the new passage: it traverses the chain of I χ I_{\chi} I χ with top n − i N n-i_N n − i N , which is ι ( W ) \iota(W) ι ( W ) .
For a set W = { i b o t , … , i t o p } W=\{i_{\mathrm{bot}},\ldots,i_{\mathrm{top}}\} W = { i bot , … , i top } of odd integers in ( 0 , n ) (0,n) ( 0 , n ) in arithmetic progression with difference 2 q 2q 2 q let
b ( W ) = min ( sin i b o t ψ , sin i t o p ψ ) , a ( W ) = max ( 0 , sin ( i t o p + 2 q ) ψ , sin ( i b o t − 2 q ) ψ ) . b(W)=\min(\sin i_{\mathrm{bot}}\psi,\sin i_{\mathrm{top}}\psi),\qquad a(W)=\max(0,\sin(i_{\mathrm{top}}+2q)\psi,\sin(i_{\mathrm{bot}}-2q)\psi). b ( W ) = min ( sin i bot ψ , sin i top ψ ) , a ( W ) = max ( 0 , sin ( i top + 2 q ) ψ , sin ( i bot − 2 q ) ψ ) .
Since sin \sin sin is concave on [ 0 , π ] [0,\pi] [ 0 , π ] , b ( W ) b(W) b ( W ) is the minimum of sin ( i ψ ) \sin(i\psi) sin ( i ψ ) over W W W , and for non-critical χ \chi χ the set W W W is a chain of I χ I_{\chi} I χ if and only if a ( W ) < χ < b ( W ) a(W)<\chi<b(W) a ( W ) < χ < b ( W ) .
Proposition 5.5. Let W = { i b o t , … , i t o p } W=\{i_{\mathrm{bot}},\ldots,i_{\mathrm{top}}\} W = { i bot , … , i top } be as above, with a ( W ) < b ( W ) a(W)<b(W) a ( W ) < b ( W ) ; let E k E_k E k be the ear with 2 k + n − q ≡ i t o p ( m o d 2 n ) 2k+n-q\equiv i_{\mathrm{top}}\pmod{2n} 2 k + n − q ≡ i top ( mod 2 n ) , and O O O its apex. Define k 1 , k 2 ∈ Z / n k_1,k_2\in\mathbb{Z}/n k 1 , k 2 ∈ Z / n by 2 k 1 ≡ i b o t − n − q 2k_1\equiv i_{\mathrm{bot}}-n-q 2 k 1 ≡ i bot − n − q and 2 k 2 ≡ − i t o p − q ( m o d 2 n ) 2k_2\equiv-i_{\mathrm{top}}-q\pmod{2n} 2 k 2 ≡ − i top − q ( mod 2 n ) , and suppose that
2 k 1 ≡ 2 k 2 ≡ n / 2 ( m o d p ) (4) 2k_1\equiv2k_2\equiv n/2\pmod p \tag*{(4)} 2 k 1 ≡ 2 k 2 ≡ n /2 ( mod p ) (4) that is, i b o t ≡ q + n / 2 i_{\mathrm{bot}}\equiv q+n/2 i bot ≡ q + n /2 and i t o p ≡ − q − n / 2 ( m o d p ) i_{\mathrm{top}}\equiv-q-n/2\pmod p i top ≡ − q − n /2 ( mod p ) . Then the horizontal leaves through the base of E k E_k E k at heights y ∈ ( ∣ Q R ∣ a ( W ) , ∣ Q R ∣ b ( W ) ) y\in(|QR|a(W),|QR|b(W)) y ∈ ( ∣ QR ∣ a ( W ) , ∣ QR ∣ b ( W )) above O O O are closed and fill an open annulus of closed leaves Z ( W ) = Z ( ι ( W ) ) Z(W)=Z(\iota(W)) Z ( W ) = Z ( ι ( W )) , of height ∣ Q R ∣ ( b ( W ) − a ( W ) ) |QR|(b(W)-a(W)) ∣ QR ∣ ( b ( W ) − a ( W )) and circumference
L ( W ) = { 2 [ cos ( ( i b o t − q ) ψ ) − cos ( ( i t o p + q ) ψ ) ] , ι ( W ) ≠ W , 2 cos ( ( i b o t − q ) ψ ) , ι ( W ) = W . L(W)=
\begin{cases}
2[\cos((i_{\mathrm{bot}}-q)\psi)-\cos((i_{\mathrm{top}}+q)\psi)], & \iota(W)\ne W,\\
2\cos((i_{\mathrm{bot}}-q)\psi), & \iota(W)=W.
\end{cases} L ( W ) = { 2 [ cos (( i bot − q ) ψ ) − cos (( i top + q ) ψ )] , 2 cos (( i bot − q ) ψ ) , ι ( W ) = W , ι ( W ) = W . Its upper boundary contains a point over R R R ; its lower boundary contains a point over R R R if a ( W ) > 0 a(W)>0 a ( W ) > 0 , and the apex O O O if a ( W ) = 0 a(W)=0 a ( W ) = 0 . If moreover r ∤ n r\nmid n r ∤ n , then Z ( W ) Z(W) Z ( W ) is a maximal cylinder if a ( W ) > 0 a(W)>0 a ( W ) > 0 , and also if a ( W ) = 0 a(W)=0 a ( W ) = 0 and q ∤ n q\nmid n q ∤ n .
Proof. For a ( W ) < χ < b ( W ) a(W)<\chi<b(W) a ( W ) < χ < b ( W ) the line at height y = ∣ Q R ∣ χ y=|QR|\chi y = ∣ QR ∣ χ meets the base of E k E_k E k , which runs from v ( i t o p + 2 q ) v(i_{\mathrm{top}}+2q) v ( i top + 2 q ) (below the line) to v ( i t o p ) v(i_{\mathrm{top}}) v ( i top ) (above it), in one interior point, and the leaf crosses there into E k E_k E k . Let T \mathcal{T} T be the open subsegment of the base of these points. For non-critical χ \chi χ , by Lemma 5.2 the leaf traverses W W W and leaves the ears around O O O through the base of E k 1 E_{k_1} E k 1 into G k 1 m o d p G_{k_1\bmod p} G k 1 mod p , which is symmetric by (5.1) and Lemma 5.3; by Lemma 5.4 it next traverses ι ( W ) \iota(W) ι ( W ) , whose bottom is n − i t o p n-i_{\mathrm{top}} n − i top , and leaves through the base of E k 2 E_{k_2} E k 2 (as
2 k 2 ′ ≡ ( n − i t o p ) − n − q ( m o d 2 n ) 2k_2' \equiv(n-i_{\mathrm{top}})-n-q \pmod{2n} 2 k 2 ′ ≡ ( n − i top ) − n − q ( mod 2 n ) into G k 2 G_{k_2} G k 2 mod p p p , also symmetric; and by Lemma 5.4 again it re-enters under n − ( n − i t o p ) = i t o p n-(n-i_{\mathrm{top}})=i_{\mathrm{top}} n − ( n − i top ) = i top , that is, through the base of E k E_k E k at height y y y above its apex: the point it started from. If ι ( W ) = W \iota(W)=W ι ( W ) = W it is back after one passage.
Length. In a passage the leaf runs, in the developed coordinates of its apex, from ( x i n , y ) (x_{\mathrm{in}},y) ( x in , y ) to ( x o u t , y ) (x_{\mathrm{out}},y) ( x out , y ) ; then, by Lemma 5.3 , across the polygon a length − 2 cos 2 k ′ ψ + x i n ′ − x o u t -2\cos2k'\psi+x'_{\mathrm{in}}-x_{\mathrm{out}} − 2 cos 2 k ′ ψ + x in ′ − x out . Over one period the x x x ’s telescope, and the length is the sum of − 2 cos 2 k ′ ψ -2\cos2k'\psi − 2 cos 2 k ′ ψ over the exit ears k ′ k' k ′ of the period. By Lemma 5.2 , 2 k ′ ψ ≡ ( i b o t − q ) ψ − π 2k'\psi\equiv(i_{\mathrm{bot}}-q)\psi-\pi 2 k ′ ψ ≡ ( i bot − q ) ψ − π for the exit after W W W , and 2 k ′ ψ ≡ ( n − i t o p − q ) ψ − π 2k'\psi\equiv(n-i_{\mathrm{top}}-q)\psi-\pi 2 k ′ ψ ≡ ( n − i top − q ) ψ − π after ι ( W ) \iota(W) ι ( W ) ; this gives L ( W ) \mathcal{L}(W) L ( W ) .
Annulus. Along these leaves the only points over Q Q Q or R R R that can be met are the points v ( j ) v(j) v ( j ) , at heights ∣ Q R ∣ sin ( j ψ ) |QR|\sin(j\psi) ∣ QR ∣ sin ( j ψ ) above the apex, for j j j in W W W , in ι ( W ) \iota(W) ι ( W ) and their neighbours in the progressions; the apexes, at height 0 0 0 ; and in the polygons the ends of the sides crossed, which are ends of bases of the ears involved and so again points v ( j ) v(j) v ( j ) of this list. For j j j in W W W or in ι ( W ) \iota(W) ι ( W ) the height is at least ∣ Q R ∣ b ( W ) |QR|b(W) ∣ QR ∣ b ( W ) , for a neighbour it is at most ∣ Q R ∣ a ( W ) |QR|a(W) ∣ QR ∣ a ( W ) (since sin ( n − i ) ψ = sin i ψ \sin(n-i)\psi=\sin i\psi sin ( n − i ) ψ = sin i ψ , ι \iota ι preserves the sines, and a ( ι ( W ) ) = a ( W ) a(\iota(W))=a(W) a ( ι ( W )) = a ( W ) , b ( ι ( W ) ) = b ( W ) b(\iota(W))=b(W) b ( ι ( W )) = b ( W ) ), and 0 ≤ ∣ Q R ∣ a ( W ) 0\leq|QR|a(W) 0 ≤ ∣ QR ∣ a ( W ) . So for every χ ∈ ( a ( W ) , b ( W ) ) \chi\in(a(W),b(W)) χ ∈ ( a ( W ) , b ( W )) , critical or not, W W W and ι ( W ) \iota(W) ι ( W ) are chains of I χ \mathcal{I}_{\chi} I χ , the leaf meets no point over Q Q Q or R R R , and the proofs of Lemmas 5.2 and 5.4 , which use that χ \chi χ is not critical only to know that the leaf meets no such point, apply unchanged. A leaf may pass through the centre of a polygon, a point over P P P ; this point is regular, since p ∣ n p\mid n p ∣ n , so the leaf is still a closed regular leaf. So the map T × R / L ( W ) Z → X \mathcal{T}\times\mathbb{R}/\mathcal{L}(W)\mathbb{Z}\to X T × R / L ( W ) Z → X that sends a point of T \mathcal{T} T and a time to the point reached at that time on the leaf from that point is a local homeomorphism, affine in the charts. It is injective: a rightward leaf crosses the base of E k E_k E k only into E k E_k E k , which happens once per period, at the same height above O O O each time (Lemma 5.3 ). Its image is an open annulus of closed leaves, of circumference L ( W ) \mathcal{L}(W) L ( W ) and height ∣ Q R ∣ ( b ( W ) − a ( W ) ) |QR|(b(W)-a(W)) ∣ QR ∣ ( b ( W ) − a ( W )) , and the leaves of ι ( W ) \iota(W) ι ( W ) are the same leaves.
Boundary. As y ↑ ∣ Q R ∣ b ( W ) y\uparrow|QR|b(W) y ↑ ∣ QR ∣ b ( W ) the leaves converge to the leaf from the upper end of T \mathcal{T} T , which runs into v ( i ) v(i) v ( i ) with sin ( i ψ ) = b ( W ) \sin(i\psi)=b(W) sin ( i ψ ) = b ( W ) ; as y ↓ ∣ Q R ∣ a ( W ) y\downarrow|QR|a(W) y ↓ ∣ QR ∣ a ( W ) , to the leaf through O O O (if a = 0 a=0 a = 0 ) or through the neighbouring vertex (if a > 0 a>0 a > 0 ). An open annulus of closed leaves is a maximal cylinder if each of its boundary components contains a singular point. If r ∤ n r\nmid n r ∤ n , the gcd ( r , n ) \gcd(r,n) g cd( r , n ) points over R R R have cone angle 2 π r / gcd ( r , n ) > 2 π 2\pi r/\gcd(r,n)>2\pi 2 π r / g cd( r , n ) > 2 π , and if q ∤ n q\nmid n q ∤ n so have the points over Q Q Q ; this gives the last sentence of the statement.
Lemma 5.6. If p = 2 p=2 p = 2 and 4 ∣ n 4\mid n 4 ∣ n , then (4) holds for every W W W , and r ∤ n r\nmid n r ∤ n .
Proof. Both sides of (4) are even, as 4 ∣ n 4\mid n 4 ∣ n . Next, r = n − 2 − q r=n-2-q r = n − 2 − q is odd, as n n n is even and q q q is odd; r > n / 2 − 2 r>n/2-2 r > n /2 − 2 , as q < n / 2 q<n/2 q < n /2 ; and n ≥ 8 n\geq8 n ≥ 8 , as 4 ∣ n 4\mid n 4 ∣ n and n / p ≥ 3 n/p\geq3 n / p ≥ 3 . An odd divisor of n n n divides n / 4 n/4 n /4 , so it is at most n / 4 < n / 2 − 1 ≤ r n/4<n/2-1\leq r n /4 < n /2 − 1 ≤ r . Hence r ∤ n r\nmid n r ∤ n .
The map Δ Δ → κ δ Δ \Delta\Delta\to\kappa\delta\Delta ΔΔ → κ δ Δ , z ↦ z ˉ z\mapsto\bar z z ↦ z ˉ in each copy, respects the gluings, so it is an isometry κ ^ \hat\kappa κ ^ of X X X with derivative diag ( 1 , − 1 ) \operatorname{diag}(1,-1) diag ( 1 , − 1 ) . It sends E k E_k E k to E − k E_{-k} E − k , apexes to apexes, polygons to polygons, rightward horizontal leaves to rightward horizontal leaves, and height y y y above O k O_k O k to height − y -y − y above O − k O_{-k} O − k ; if the clockwise leg of E k E_k E k has index i 1 i_1 i 1 , that of E − k E_{-k} E − k has index 2 n − i 1 − 2 q 2n-i_1-2q 2 n − i 1 − 2 q .
Proposition 5.7. In Proposition 5.5 , suppose that r ∤ n r\nmid n r ∤ n , that q ∣ n q\mid n q ∣ n , that a ( W ) = 0 a(W)=0 a ( W ) = 0 with sin ( i t o p + 2 q ) ψ < 0 \sin(i_{\mathrm{top}}+2q)\psi<0 sin ( i top + 2 q ) ψ < 0 , and that for small χ > 0 \chi>0 χ > 0 the chain of I χ \mathcal{I}_{\chi} I χ with top 2 n − i t o p − 2 q 2n-i_{\mathrm{top}}-2q 2 n − i top − 2 q is W W W or ι ( W ) \iota(W) ι ( W ) . Then Z ( W ) ∪ γ 0 κ ^ ( Z ( W ) ) Z(W)\cup\gamma_0\hat\kappa(Z(W)) Z ( W ) ∪ γ 0 κ ^ ( Z ( W )) , where γ 0 \gamma_0 γ 0 is the leaf at height 0 0 0 through O O O , is a maximal cylinder of circumference L ( W ) \mathcal{L}(W) L ( W ) and height 2 ∣ Q R ∣ b ( W ) 2|QR|b(W) 2∣ QR ∣ b ( W ) .
Proof. The base of E k E_k E k has its ends at heights ∣ Q R ∣ sin ( i t o p + 2 q ) ψ < 0 |QR|\sin(i_{\mathrm{top}}+2q)\psi<0 ∣ QR ∣ sin ( i top + 2 q ) ψ < 0 and ∣ Q R ∣ sin ( i t o p ψ ) > 0 |QR|\sin(i_{\mathrm{top}}\psi)>0 ∣ QR ∣ sin ( i top ψ ) > 0 , so it contains a point at height $0$ in its interior. The leaf γ 0 \gamma_0 γ 0 from that point is the limit of the leaves of Z ( W ) Z(W) Z ( W ) ; it meets no point over R R R (their heights above an apex are $|QR|\sin(j\psi)\ne 0 with j j j odd, and polygon chords meet vertices only at ends of bases of ears), only regular points over Q Q Q , and possibly centres of polygons, points over P P P , which are regular since p ∣ n p \mid n p ∣ n ; so it is a regular closed leaf, of length L ( W ) \mathcal{L}(W) L ( W ) . The leaves just below γ 0 \gamma_0 γ 0 cross the base of E k E_k E k at small negative heights; they are the κ ^ \hat{\kappa} κ ^ -images of leaves entering E − k E_{-k} E − k at small positive heights, whose first leg has index 2 n − i t o p − 2 q 2n-i_{\mathrm{top}}-2q 2 n − i top − 2 q , so by hypothesis they are leaves of Z ( W ) Z(W) Z ( W ) , and the leaves just below γ 0 \gamma_0 γ 0 lie in κ ^ ( Z ( W ) ) \hat{\kappa}(Z(W)) κ ^ ( Z ( W )) . The annuli Z ( W ) Z(W) Z ( W ) and κ ^ ( Z ( W ) ) \hat{\kappa}(Z(W)) κ ^ ( Z ( W )) are disjoint: a leaf of Z ( W ) Z(W) Z ( W ) enters each ear of its period at the same positive height above the apex, while its image enters the image ears at the negative height. The union is an annulus foliated by closed leaves of length L ( W ) \mathcal{L}(W) L ( W ) , of height 2 ∣ Q R ∣ b ( W ) 2|QR|b(W) 2∣ QR ∣ b ( W ) , whose two boundary components contain points over R R R , which are singular since r ∤ n r\nmid n r ∤ n .
In the irrationality steps below we use the standard fact that, for N ≥ 3 N \ge3 N ≥ 3 and gcd ( j , N ) = 1 \gcd(j,N)=1 g cd( j , N ) = 1 , cos ( 2 π j / N ) \cos(2\pi j/N) cos ( 2 π j / N ) has degree φ ( N ) / 2 \varphi(N)/2 φ ( N ) /2 over Q \mathbb{Q} Q .
Theorem 5.8. For every even m ≥ 6 m \ge6 m ≥ 6 the triangle ( 2 , m − 3 , m + 1 ) / 2 m (2,m-3,m+1)/2m ( 2 , m − 3 , m + 1 ) /2 m is not a lattice triangle. Let ψ = π / 2 m \psi=\pi/2m ψ = π /2 m , so that ∣ Q R ∣ = sin 2 ψ / sin ( m + 1 ) ψ |QR|=\sin2\psi/\sin(m+1)\psi ∣ QR ∣ = sin 2 ψ / sin ( m + 1 ) ψ .
(a) If m ≥ 8 m \ge8 m ≥ 8 , in the direction of the side joining the vertices of angles 2 ψ 2\psi 2 ψ and ( m − 3 ) ψ (m-3)\psi ( m − 3 ) ψ , with that side of length 1, the unfolding has two maximal cylinders, of circumference 4 sin 5 ψ cos 3 ψ 4\sin5\psi\cos3\psi 4 sin 5 ψ cos 3 ψ and height ∣ Q R ∣ ( sin 5 ψ − sin ψ ) |QR|(\sin5\psi-\sin\psi) ∣ QR ∣ ( sin 5 ψ − sin ψ ) , and of circumference 4 cos 3 ψ cos ψ 4\cos3\psi\cos\psi 4 cos 3 ψ cos ψ and height ∣ Q R ∣ cos ψ |QR|\cos\psi ∣ QR ∣ cos ψ ; the ratio of their moduli is 1 − sin ψ / sin 5 ψ 1-\sin\psi/\sin5\psi 1 − sin ψ / sin 5 ψ , which is irrational.
(b) If m = 6 m=6 m = 6 , the triangle is ( 2 , 3 , 7 ) / 12 (2,3,7)/12 ( 2 , 3 , 7 ) /12 ; in the same direction it has a maximal cylinder of circumference 2 and height 2 ∣ Q R ∣ sin 3 ψ 2|QR|\sin3\psi 2∣ QR ∣ sin 3 ψ , and the first cylinder of (a); the ratio of their moduli is ( 3 − 1 ) / 2 (\sqrt{3}-1)/2 ( 3 − 1 ) /2 .
These triangles form the part of C \mathcal{C} C with p = 2 p=2 p = 2 . For m = 4 m=4 m = 4 the triangle is ( 1 , 2 , 5 ) / 8 (1,2,5)/8 ( 1 , 2 , 5 ) /8 , a lattice triangle. For m = 6 m=6 m = 6 the statement is Vorobets’s [30 Theorem 4.5] , proved there with two horizontal cylinders. For odd m m m the integers 2 2 2 , m − 3 m-3 m − 3 , m + 1 m+1 m + 1 have greatest common divisor 2, and the triangle is ( 1 , ( m − 3 ) / 2 , ( m + 1 ) / 2 ) / m (1,(m-3)/2,(m+1)/2)/m ( 1 , ( m − 3 ) /2 , ( m + 1 ) /2 ) / m , which is treated in § 4 and § 6 .
Proof. Let m ≥ 6 m \ge6 m ≥ 6 be even, n = 2 m n=2m n = 2 m , p = 2 p=2 p = 2 , q = m − 3 q=m-3 q = m − 3 , r = m + 1 r=m+1 r = m + 1 ; then 4 ∣ n 4\mid n 4 ∣ n , the hypotheses of Lemma 5.1 hold, and by Lemma 5.6 so does (4) , for every W W W , and r ∤ n r\nmid n r ∤ n . The vertex P P P gives two regular points, R R R one point of cone angle 2 π ( m + 1 ) 2\pi(m+1) 2 π ( m + 1 ) (as gcd ( m + 1 , 2 m ) = 1 \gcd(m+1,2m)=1 g cd( m + 1 , 2 m ) = 1 ), and Q Q Q gcd ( m − 3 , 6 ) ∈ { 1 , 3 } \gcd(m-3,6)\in\{1,3\} g cd( m − 3 , 6 ) ∈ { 1 , 3 } points of cone angle 2 π ( m − 3 ) / gcd ( m − 3 , 6 ) 2\pi(m-3)/\gcd(m-3,6) 2 π ( m − 3 ) / g cd( m − 3 , 6 ) , which are regular exactly for m ∈ { 4 , 6 } m\in\{4,6\} m ∈ { 4 , 6 } . By (1) , 2 genus ( X ) − 2 = 2 m − 3 − gcd ( m − 3 , 6 ) ≥ 6 2\operatorname{genus}(X)-2=2m-3-\gcd(m-3,6)\ge6 2 genus ( X ) − 2 = 2 m − 3 − g cd( m − 3 , 6 ) ≥ 6 . Since m ψ = π / 2 m\psi=\pi/2 m ψ = π /2 , sin ( m + j ) ψ = cos j ψ \sin(m+j)\psi=\cos j\psi sin ( m + j ) ψ = cos j ψ .
The chain W c = { 2 m − 5 } \mathcal{W}_c=\{2m-5\} W c = { 2 m − 5 } : its neighbours in the progression are 2 m − 5 + 2 q = 2 n − 11 2m-5+2q=2n-11 2 m − 5 + 2 q = 2 n − 11 , with sin ( 2 n − 11 ) ψ = − sin 11 ψ < 0 \sin(2n-11)\psi=-\sin11\psi<0 sin ( 2 n − 11 ) ψ = − sin 11 ψ < 0 as n ≥ 12 n\ge12 n ≥ 12 , and 2 m − 5 − 2 q = 1 2m-5-2q=1 2 m − 5 − 2 q = 1 . So a ( W c ) = sin ψ a(\mathcal{W}_c)=\sin\psi a ( W c ) = sin ψ and b ( W c ) = sin ( 2 m − 5 ) ψ = sin 5 ψ > sin ψ b(\mathcal{W}_c)=\sin(2m-5)\psi=\sin5\psi>\sin\psi b ( W c ) = sin ( 2 m − 5 ) ψ = sin 5 ψ > sin ψ ; ℓ ( W c ) = { 5 } ≠ W c \ell(\mathcal{W}_c)=\{5\}\ne\mathcal{W}_c ℓ ( W c ) = { 5 } = W c , and L ( W c ) = 2 [ cos ( m − 2 ) ψ − cos ( 3 m − 8 ) ψ ] = 2 [ sin 2 ψ + sin 8 ψ ] = 4 sin 5 ψ cos 3 ψ \mathcal{L}(\mathcal{W}_c)=2[\cos(m-2)\psi-\cos(3m-8)\psi]=2[\sin2\psi+\sin8\psi]=4\sin5\psi\cos3\psi L ( W c ) = 2 [ cos ( m − 2 ) ψ − cos ( 3 m − 8 ) ψ ] = 2 [ sin 2 ψ + sin 8 ψ ] = 4 sin 5 ψ cos 3 ψ . By Proposition 5.5 , Z c = Z ( W c ) Z_c=Z(\mathcal{W}_c) Z c = Z ( W c ) is a maximal annulus of height ∣ Q R ∣ ( sin 5 ψ − sin ψ ) |QR|(\sin5\psi-\sin\psi) ∣ QR ∣ ( sin 5 ψ − sin ψ ) and modulus μ c = ∣ Q R ∣ ( sin 5 ψ − sin ψ ) / ( 4 sin 5 ψ cos 3 ψ ) \mu_c=|QR|(\sin5\psi-\sin\psi)/(4\sin5\psi\cos3\psi) μ c = ∣ QR ∣ ( sin 5 ψ − sin ψ ) / ( 4 sin 5 ψ cos 3 ψ ) .
Part (a): let m ≥ 8 m\ge8 m ≥ 8 and W d = { m + 1 } \mathcal{W}_d=\{m+1\} W d = { m + 1 } . Its neighbours are m + 1 + 2 q = 3 m − 5 m+1+2q=3m-5 m + 1 + 2 q = 3 m − 5 , with sin ( 3 m − 5 ) ψ = − cos 5 ψ < 0 \sin(3m-5)\psi=-\cos5\psi<0 sin ( 3 m − 5 ) ψ = − cos 5 ψ < 0 , and m + 1 − 2 q = 7 − m ∈ ( − n , 0 ) m+1-2q=7-m\in(-n,0) m + 1 − 2 q = 7 − m ∈ ( − n , 0 ) . So a ( W d ) = 0 a(\mathcal{W}_d)=0 a ( W d ) = 0 , b ( W d ) = cos ψ b(\mathcal{W}_d)=\cos\psi b ( W d ) = cos ψ , ℓ ( W d ) = { m − 1 } ≠ W d \ell(\mathcal{W}_d)=\{m-1\}\ne\mathcal{W}_d ℓ ( W d ) = { m − 1 } = W d , and L ( W d ) = 2 [ cos 4 ψ − cos ( 2 m − 2 ) ψ ] = 4 cos 3 ψ cos ψ \mathcal{L}(\mathcal{W}_d)=2[\cos4\psi-\cos(2m-2)\psi]=4\cos3\psi\cos\psi L ( W d ) = 2 [ cos 4 ψ − cos ( 2 m − 2 ) ψ ] = 4 cos 3 ψ cos ψ . The points over Q Q Q are singular, so Z d = Z ( W d ) Z_d=Z(\mathcal{W}_d) Z d = Z ( W d ) is maximal, of modulus μ d = ∣ Q R ∣ / ( 4 cos 3 ψ ) \mu_d=|QR|/(4\cos3\psi) μ d = ∣ QR ∣/ ( 4 cos 3 ψ ) . Then μ c / μ d = 1 − sin ψ / sin 5 ψ \mu_c/\mu_d=1-\sin\psi/\sin5\psi μ c / μ d = 1 − sin ψ / sin 5 ψ . With x = 2 cos 2 ψ = 2 cos ( π / m ) x=2\cos2\psi=2\cos(\pi/m) x = 2 cos 2 ψ = 2 cos ( π / m ) we have $4\cos^2\psi=2+x$ and sin 5 ψ / sin ψ = 16 cos 4 ψ − 12 cos 2 ψ + 1 = x 2 + x − 1 \sin5\psi/\sin\psi=16\cos^4\psi-12\cos^2\psi+1=x^2+x-1 sin 5 ψ / sin ψ = 16 cos 4 ψ − 12 cos 2 ψ + 1 = x 2 + x − 1 . If the ratio were rational, x x x would have degree at most 2 over Q \mathbb{Q} Q ; but [ Q ( 2 cos ( 2 π / 2 m ) ) : Q ] = φ ( 2 m ) / 2 [\mathbb{Q}(2\cos(2\pi/2m)):\mathbb{Q}]=\varphi(2m)/2 [ Q ( 2 cos ( 2 π /2 m )) : Q ] = φ ( 2 m ) /2 , and φ ( 2 m ) ≤ 4 \varphi(2m)\le4 φ ( 2 m ) ≤ 4 with $m\ge 3】【。 only for 2 m ∈ { 6 , 8 , 10 , 12 } 2m \in\{6,8,10,12\} 2 m ∈ { 6 , 8 , 10 , 12 } . So for even m ≥ 8 m \ge8 m ≥ 8 the ratio is irrational, and Lemma 2.3 (i) applies.
Part (b): m = 6 m=6 m = 6 , n = 12 n=12 n = 12 , q = 3 q=3 q = 3 , and the points over Q \mathbb{Q} Q are regular. Take W a = { 3 , 9 } W_a=\{3,9\} W a = { 3 , 9 } : its neighbours 15 15 15 and − 3 -3 − 3 have negative sines, so a ( W a ) = 0 a(W_a)=0 a ( W a ) = 0 , b ( W a ) = sin 3 ψ b(W_a)=\sin3\psi b ( W a ) = sin 3 ψ , ι ( W a ) = W a \iota(W_a)=W_a ι ( W a ) = W a and L ( W a ) = 2 \mathcal{L}(W_a)=2 L ( W a ) = 2 . The index 2 n − i t o p − 2 q = 24 − 9 − 6 2n-i_{\mathrm{top}}-2q=24-9-6 2 n − i top − 2 q = 24 − 9 − 6 is 9 = i t o p 9=i_{\mathrm{top}} 9 = i top , so Proposition 5.7 gives a maximal cylinder Z ^ a \widehat{Z}_a Z a of circumference 2 2 2 , height 2 ∣ Q R ∣ sin 3 ψ 2|QR|\sin3\psi 2∣ QR ∣ sin 3 ψ and modulus ∣ Q R ∣ sin 3 ψ |QR|\sin3\psi ∣ QR ∣ sin 3 ψ . Since sin 6 ψ = 1 \sin6\psi=1 sin 6 ψ = 1 and sin 5 ψ = cos ψ \sin5\psi=\cos\psi sin 5 ψ = cos ψ ,
μ c μ ( Z ^ a ) = 1 − sin ψ / sin 5 ψ 4 sin 3 ψ cos 3 ψ = 1 − tan ( π / 12 ) 2 = 3 − 1 2 , \frac{\mu_c}{\mu(\widehat{Z}_a)}
=\frac{1-\sin\psi/\sin5\psi}{4\sin3\psi\cos3\psi}
=\frac{1-\tan(\pi/12)}{2}
=\frac{\sqrt{3}-1}{2}, μ ( Z a ) μ c = 4 sin 3 ψ cos 3 ψ 1 − sin ψ / sin 5 ψ = 2 1 − tan ( π /12 ) = 2 3 − 1 ,
which is irrational, and Lemma 2.3 (i) applies. □
At ( 1 , 2 , 5 ) / 8 (1,2,5)/8 ( 1 , 2 , 5 ) /8 (m = 4 m=4 m = 4 ) the chains { 3 } \{3\} { 3 } and { 5 } \{5\} { 5 } satisfy a = b a=b a = b and give no cylinder.
Theorem 5.9. For every odd q ≥ 3 q \ge3 q ≥ 3 the triangle ( 2 , q , 3 q + 2 ) / ( 4 q + 4 ) (2,q,3q+2)/(4q+4) ( 2 , q , 3 q + 2 ) / ( 4 q + 4 ) is not a lattice triangle. With ψ = π / ( 4 q + 4 ) \psi=\pi/(4q+4) ψ = π / ( 4 q + 4 ) and ∣ Q R ∣ = sin 2 ψ / sin ( 3 q + 2 ) ψ |QR|=\sin2\psi/\sin(3q+2)\psi ∣ QR ∣ = sin 2 ψ / sin ( 3 q + 2 ) ψ , in the direction of the side joining the vertices of angles 2 ψ 2\psi 2 ψ and q ψ q\psi q ψ , with that side of length 1 1 1 , the unfolding has two maximal cylinders, of circumference 4 cos ψ sin ( π / 4 + 3 ψ ) 4\cos\psi\sin(\pi/4+3\psi) 4 cos ψ sin ( π /4 + 3 ψ ) and height ∣ Q R ∣ sin ψ |QR|\sin\psi ∣ QR ∣ sin ψ , and of circumference 2 cos 2 ψ 2\cos2\psi 2 cos 2 ψ and height ∣ Q R ∣ sin ( π / 4 + ψ ) |QR|\sin(\pi/4+\psi) ∣ QR ∣ sin ( π /4 + ψ ) ; the ratio of their moduli is tan ψ / ( 1 + 2 sin 2 ψ ) \tan\psi/(1+2\sin2\psi) tan ψ / ( 1 + 2 sin 2 ψ ) , which is irrational.
Proof. Here n = 4 q + 4 n=4q+4 n = 4 q + 4 , m = n / 2 = 2 q + 2 m=n/2=2q+2 m = n /2 = 2 q + 2 , so 4 ∣ n 4\mid n 4 ∣ n and 2 q = m − 2 2q=m-2 2 q = m − 2 ; q ψ = π / 4 − ψ q\psi=\pi/4-\psi q ψ = π /4 − ψ and ( q + 2 ) ψ = π / 4 + ψ (q+2)\psi=\pi/4+\psi ( q + 2 ) ψ = π /4 + ψ . The hypotheses of Lemma 5.1 hold, and by Lemma 5.6 so do (4) and r ∤ n r\nmid n r ∤ n . The points over Q \mathbb{Q} Q are singular, since gcd ( q , 4 q + 4 ) = 1 \gcd(q,4q+4)=1 g cd( q , 4 q + 4 ) = 1 and q ≥ 3 q\ge3 q ≥ 3 , and the point over R R R is singular, since gcd ( 3 q + 2 , 4 q + 4 ) = gcd ( q + 2 , 4 ) = 1 \gcd(3q+2,4q+4)=\gcd(q+2,4)=1 g cd( 3 q + 2 , 4 q + 4 ) = g cd( q + 2 , 4 ) = 1 . By (1) , 2 genus ( X ) − 2 = 4 q 2\operatorname{genus}(X)-2=4q 2 genus ( X ) − 2 = 4 q .
The chain W T = { 1 , 2 q + 1 , 4 q + 1 } W_T=\{1,2q+1,4q+1\} W T = { 1 , 2 q + 1 , 4 q + 1 } : the sines are sin ψ \sin\psi sin ψ , cos ψ \cos\psi cos ψ and sin 3 ψ \sin3\psi sin 3 ψ , so b ( W T ) = sin ψ b(W_T)=\sin\psi b ( W T ) = sin ψ ; the neighbours are 6 q + 1 = 3 m − 5 6q+1=3m-5 6 q + 1 = 3 m − 5 , with sine − cos 5 ψ < 0 -\cos5\psi<0 − cos 5 ψ < 0 as n > 10 n>10 n > 10 , and 1 − 2 q ∈ ( − n , 0 ) 1-2q\in(-n,0) 1 − 2 q ∈ ( − n , 0 ) ; so a ( W T ) = 0 a(W_T)=0 a ( W T ) = 0 . Also ι ( W T ) = { 3 , 2 q + 3 , 4 q + 3 } ≠ W T \iota(W_T)=\{3,2q+3,4q+3\}\ne W_T ι ( W T ) = { 3 , 2 q + 3 , 4 q + 3 } = W T , and
L ( W T ) = 2 [ cos ( q − 1 ) ψ + cos ( q − 3 ) ψ ] = 4 cos ψ cos ( q − 2 ) ψ = 4 cos ψ sin ( π / 4 + 3 ψ ) . \mathcal{L}(W_T)=2[\cos(q-1)\psi+\cos(q-3)\psi]=4\cos\psi\cos(q-2)\psi=4\cos\psi\sin(\pi/4+3\psi). L ( W T ) = 2 [ cos ( q − 1 ) ψ + cos ( q − 3 ) ψ ] = 4 cos ψ cos ( q − 2 ) ψ = 4 cos ψ sin ( π /4 + 3 ψ ) .
So Z T = Z ( W T ) Z_T=Z(W_T) Z T = Z ( W T ) is maximal, with height ∣ Q R ∣ sin ψ |QR|\sin\psi ∣ QR ∣ sin ψ .
The chain W S = { q + 2 , 3 q + 2 } W_S=\{q+2,3q+2\} W S = { q + 2 , 3 q + 2 } satisfies ι ( W S ) = W S \iota(W_S)=W_S ι ( W S ) = W S ; its neighbours 5 q + 2 ∈ ( n , 2 n ) 5q+2\in(n,2n) 5 q + 2 ∈ ( n , 2 n ) and 2 − q < 0 2-q<0 2 − q < 0 have negative sines, so a ( W S ) = 0 a(W_S)=0 a ( W S ) = 0 , b ( W S ) = sin ( q + 2 ) ψ = sin ( π / 4 + ψ ) b(W_S)=\sin(q+2)\psi=\sin(\pi/4+\psi) b ( W S ) = sin ( q + 2 ) ψ = sin ( π /4 + ψ ) and L ( W S ) = 2 cos 2 ψ \mathcal{L}(W_S)=2\cos2\psi L ( W S ) = 2 cos 2 ψ . So Z S = Z ( W S ) Z_S=Z(W_S) Z S = Z ( W S ) is maximal, with height ∣ Q R ∣ sin ( π / 4 + ψ ) |QR|\sin(\pi/4+\psi) ∣ QR ∣ sin ( π /4 + ψ ) .
Since 2 sin ( π / 4 + 3 ψ ) sin ( π / 4 + ψ ) = cos 2 ψ − cos ( π / 2 + 4 ψ ) = cos 2 ψ ( 1 + 2 sin 2 ψ ) 2\sin(\pi/4+3\psi)\sin(\pi/4+\psi)=\cos2\psi-\cos(\pi/2+4\psi)=\cos2\psi(1+2\sin2\psi) 2 sin ( π /4 + 3 ψ ) sin ( π /4 + ψ ) = cos 2 ψ − cos ( π /2 + 4 ψ ) = cos 2 ψ ( 1 + 2 sin 2 ψ ) , the ratio of the moduli is
ρ = tan ψ 1 + 2 sin 2 ψ = sin 2 ψ ( 1 + cos 2 ψ ) ( 1 + 2 sin 2 ψ ) . \rho=\frac{\tan\psi}{1+2\sin2\psi}
=\frac{\sin2\psi}{(1+\cos2\psi)(1+2\sin2\psi)}. ρ = 1 + 2 sin 2 ψ tan ψ = ( 1 + cos 2 ψ ) ( 1 + 2 sin 2 ψ ) sin 2 ψ .
Let ζ = e 2 π i / n \zeta=e^{2\pi i/n} ζ = e 2 π i / n , so that cos 2 ψ , sin 2 ψ ∈ Q ( ζ ) \cos2\psi,\sin2\psi\in\mathbb{Q}(\zeta) cos 2 ψ , sin 2 ψ ∈ Q ( ζ ) (as i = ζ n / 4 i=\zeta^{n/4} i = ζ n /4 ). Since q q q is odd, 8 ∣ n 8\mid n 8 ∣ n , and n / 2 + 1 n/2+1 n /2 + 1 is odd and prime to n n n ; so σ n / 2 + 1 : ζ ↦ ζ n / 2 + 1 = − ζ \sigma_{n/2+1}:\zeta\mapsto\zeta^{n/2+1}=-\zeta σ n /2 + 1 : ζ ↦ ζ n /2 + 1 = − ζ is an automorphism of Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) . It fixes i i i (as n / 4 n/4 n /4 is even) and changes the signs of cos 2 ψ \cos2\psi cos 2 ψ and sin 2 ψ \sin2\psi sin 2 ψ , so
σ n / 2 + 1 ( ρ ) = − sin 2 ψ ( 1 − cos 2 ψ ) ( 1 − 2 sin 2 ψ ) < 0 , \sigma_{n/2+1}(\rho)
=-\frac{\sin2\psi}{(1-\cos2\psi)(1-2\sin2\psi)}<0, σ n /2 + 1 ( ρ ) = − ( 1 − cos 2 ψ ) ( 1 − 2 sin 2 ψ ) sin 2 ψ < 0 ,
because sin 2 ψ < 1 / 2 \sin2\psi<1/2 sin 2 ψ < 1/2 for n > 12 n>12 n > 12 , while ρ > 0 \rho>0 ρ > 0 . So ρ \rho ρ is irrational, and Lemma 2.3 (i) applies. □
For q = 1 q=1 q = 1 the triangle is ( 1 , 2 , 5 ) / 8 (1,2,5)/8 ( 1 , 2 , 5 ) /8 again. For odd q ≥ 3 q\ge3 q ≥ 3 , ( 2 , q , 3 q + 2 ) / ( 4 q + 4 ) (2,q,3q+2)/(4q+4) ( 2 , q , 3 q + 2 ) / ( 4 q + 4 ) is family (5) of Theorem 2.5 , and its obtuse angle exceeds 2 π / 3 2\pi/3 2 π /3 .
Two families with an angle 4 π / n 4\pi/n 4 π / n : one cylinder and its share
We keep the notation of Section 5.1, now with p = 4 p=4 p = 4 ; Propositions 5.5 and 5.7 are stated for every p p p .
Theorem 5.10. Let m ≡ 2 ( m o d 4 ) m \equiv2 \pmod{4} m ≡ 2 ( mod 4 ) with m ≥ 10 m \ge10 m ≥ 10 , n = 2 m n=2m n = 2 m and ϑ = 2 π / n = π / m \vartheta=2\pi/n=\pi/m ϑ = 2 π / n = π / m . The triangle ( 4 , m − 5 , m + 1 ) / 2 m (4,m-5,m+1)/2m ( 4 , m − 5 , m + 1 ) /2 m is not a lattice triangle. In the direction of the side joining its vertices of angles 4 π / n 4\pi/n 4 π / n and ( m − 5 ) π / n (m-5)\pi/n ( m − 5 ) π / n , its unfolding has a maximal cylinder Z Z Z with
n ρ ( Z ) = 4 cos ϑ ( cos ϑ − cos 4 ϑ ) ( m ≥ 14 ) , n\rho(Z)=4\cos\vartheta(\cos\vartheta-\cos4\vartheta) \qquad(m\ge14), n ρ ( Z ) = 4 cos ϑ ( cos ϑ − cos 4 ϑ ) ( m ≥ 14 ) , n ρ ( Z ) = 8 cos ϑ ( cos ϑ − cos 4 ϑ ) ( m = 10 ) , n\rho(Z)=8\cos\vartheta(\cos\vartheta-\cos4\vartheta) \qquad(m=10), n ρ ( Z ) = 8 cos ϑ ( cos ϑ − cos 4 ϑ ) ( m = 10 ) , and σ ℓ ( ρ ( Z ) ) < 0 \sigma_{\ell}(\rho(Z))<0 σ ℓ ( ρ ( Z )) < 0 for ℓ = m / 2 − 2 \ell=m/2-2 ℓ = m /2 − 2 if m ≥ 22 m\ge22 m ≥ 22 , and for ℓ = 7 , 11 , 13 \ell=7,11,13 ℓ = 7 , 11 , 13 if m = 10 , 14 , 18 m=10,14,18 m = 10 , 14 , 18 .
These triangles form the part of C C C with p = 4 p=4 p = 4 . For m = 6 m=6 m = 6 the triangle is ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 , Hooper’s lattice triangle; the proof below needs n ≥ 20 n\ge20 n ≥ 20 .
Proof. Here p = 4 p=4 p = 4 , q = m − 5 q=m-5 q = m − 5 , r = m + 1 r=m+1 r = m + 1 , ψ = π / n \psi=\pi/n ψ = π / n , so that m ψ = π / 2 m\psi=\pi/2 m ψ = π /2 and ϑ = 2 ψ \vartheta=2\psi ϑ = 2 ψ . The hypotheses of Lemma 5.1 hold, n n n is even and q q q is odd; since n / 2 = m ≡ 2 ( m o d 4 ) n/2=m\equiv2\pmod{4} n /2 = m ≡ 2 ( mod 4 ) , the polygon G j G_j G j is symmetric if and only if j j j is odd (Lemma 5.3 ). The vertex R R R gives one point of cone angle 2 π ( m + 1 ) 2\pi(m+1) 2 π ( m + 1 ) , since gcd ( m + 1 , 2 m ) = 1 \gcd(m+1,2m)=1 g cd( m + 1 , 2 m ) = 1 ; it is singular, and r ∤ n r\nmid n r ∤ n , as Propositions 5.5 and 5.7 require. The points over Q Q Q are singular if and only if q ∤ n q\nmid n q ∤ n , that is, ( m − 5 ) ∣ 10 (m-5)\mid10 ( m − 5 ) ∣ 10 , that is, m ≠ 6 , 10 m\ne6,10 m = 6 , 10 .
Let W = { 3 , n − 7 } \mathcal{W}=\{3,n-7\} W = { 3 , n − 7 } (Figure 2 , left, for m = 14 m=14 m = 14 ), a progression with difference n − 10 = 2 q n-10=2q n − 10 = 2 q . Since 10 ψ < π 10\psi<\pi 10 ψ < π we have sin ( n − 7 ) ψ = sin 7 ψ > sin 3 ψ > 0 \sin(n-7)\psi=\sin7\psi>\sin3\psi>0 sin ( n − 7 ) ψ = sin 7 ψ > sin 3 ψ > 0 . The neighbours are ( n − 7 ) + 2 q = 2 n − 17 (n-7)+2q=2n-17 ( n − 7 ) + 2 q = 2 n − 17 , with sin ( 2 n − 17 ) ψ = − sin 17 ψ < 0 \sin(2n-17)\psi=-\sin17\psi<0 sin ( 2 n − 17 ) ψ = − sin 17 ψ < 0 because 17 ψ < π 17\psi<\pi 17 ψ < π , and 3 − 2 q = 13 − n < 0 3-2q=13-n<0 3 − 2 q = 13 − n < 0 . So W \mathcal{W} W is a chain of I χ I_{\chi} I χ for 0 < χ < sin 3 ψ 0<\chi<\sin3\psi 0 < χ < sin 3 ψ , with a ( W ) = 0 a(\mathcal{W})=0 a ( W ) = 0 and b ( W ) = sin 3 ψ b(\mathcal{W})=\sin3\psi b ( W ) = sin 3 ψ , and ι ( W ) = { 7 , n − 3 } ≠ W \iota(\mathcal{W})=\{7,n-3\}\ne\mathcal{W} ι ( W ) = { 7 , n − 3 } = W . The exit ears are given by 2 k 1 ≡ 3 − n − q = 8 − 3 m 2k_1\equiv3-n-q=8-3m 2 k 1 ≡ 3 − n − q = 8 − 3 m and 2 k 2 ≡ 7 − n − q = 12 − 3 m ( m o d 4 m ) 2k_2\equiv7-n-q=12-3m\pmod{4m} 2 k 2 ≡ 7 − n − q = 12 − 3 m ( mod 4 m ) , so k 1 ≡ m / 2 + 4 k_1\equiv m/2+4 k 1 ≡ m /2 + 4 and k 2 ≡ m / 2 + 6 ( m o d n ) k_2\equiv m/2+6\pmod n k 2 ≡ m /2 + 6 ( mod n ) ; both are odd, as m / 2 m/2 m /2 is odd, so 2 k 1 ≡ 2 k 2 ≡ 2 ≡ n / 2 ( m o d 4 ) 2k_1\equiv2k_2\equiv2\equiv n/2\pmod4 2 k 1 ≡ 2 k 2 ≡ 2 ≡ n /2 ( mod 4 ) : this is (4) , and Proposition 5.5 applies, with
Figure 2. The model of Lemma 5.1 with p = 4 p = 4 p = 4 : the regular ( n / 4 ) (n/4) ( n /4 ) -gons G 0 , … , G 3 G_0,\ldots,G_3 G 0 , … , G 3 with their ears; a leg is labelled i i i when its direction from its apex is i ψ i\psi i ψ , and legs with the same label are glued. Left, ( 4 , 9 , 15 ) / 28 (4,9,15)/28 ( 4 , 9 , 15 ) /28 (Theorem 5.10 , m = 14 m = 14 m = 14 ); right, ( 4 , 7 , 25 ) / 36 (4,7,25)/36 ( 4 , 7 , 25 ) /36 (Theorem 5.11 , q = 7 q = 7 q = 7 ): shaded, the maximal cylinder Z ( W ) Z(W) Z ( W ) of the chain W = { 3 , 21 } W = \{3,21\} W = { 3 , 21 } , resp. { 9 , 23 } \{9,23\} { 9 , 23 } , whose share of the area has a negative conjugate; its lower boundary passes through the apex O O O , and on the right one of its leaves crosses the centre of G 3 G_3 G 3 . For ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 (n = 12 n = 12 n = 12 , q = 1 q = 1 q = 1 ) the ears close up into a regular 12-gon and the model is Hooper’s Figure 1 [11 ] , turned through 13 π / 12 13\pi/12 13 π /12 .
L ( W ) = 2 [ cos ( 3 − q ) ψ − cos ( n − 7 + q ) ψ ] = 2 [ sin 8 ψ + sin 12 ψ ] = 4 sin 10 ψ cos 2 ψ , \mathcal{L}(\mathcal{W})=2[\cos(3-q)\psi-\cos(n-7+q)\psi]=2[\sin8\psi+\sin12\psi]=4\sin10\psi\cos2\psi, L ( W ) = 2 [ cos ( 3 − q ) ψ − cos ( n − 7 + q ) ψ ] = 2 [ sin 8 ψ + sin 12 ψ ] = 4 sin 10 ψ cos 2 ψ ,
since ( 3 − q ) ψ = 8 ψ − π / 2 (3-q)\psi=8\psi-\pi/2 ( 3 − q ) ψ = 8 ψ − π /2 and ( n − 7 + q ) ψ = 3 π / 2 − 12 ψ (n-7+q)\psi=3\pi/2-12\psi ( n − 7 + q ) ψ = 3 π /2 − 12 ψ .
The share. The area of X X X is 2 n 2n 2 n times the area 1 2 ∣ P Q ∣ ∣ Q R ∣ sin q ψ \frac{1}{2}|PQ||QR|\sin q\psi 2 1 ∣ P Q ∣∣ QR ∣ sin q ψ of Δ \Delta Δ , that is, n ∣ Q R ∣ sin q ψ = n ∣ Q R ∣ cos 5 ψ n|QR|\sin q\psi=n|QR|\cos5\psi n ∣ QR ∣ sin q ψ = n ∣ QR ∣ cos 5 ψ . If m ≥ 14 m\ge14 m ≥ 14 , the points over Q Q Q are singular and Z ( W ) Z(\mathcal{W}) Z ( W ) is a maximal cylinder of height ∣ Q R ∣ sin 3 ψ |QR|\sin3\psi ∣ QR ∣ sin 3 ψ , so
n ρ = 4 sin 3 ψ sin 10 ψ cos 2 ψ cos 5 ψ = 8 sin 3 ψ sin 5 ψ cos 2 ψ = 4 cos 2 ψ ( cos 2 ψ − cos 8 ψ ) = 4 cos ϑ ( cos ϑ − cos 4 ϑ ) . \begin{aligned}
n\rho=\frac{4\sin3\psi\sin10\psi\cos2\psi}{\cos5\psi}=8\sin3\psi\sin5\psi\cos2\psi\\
&=4\cos2\psi(\cos2\psi-\cos8\psi)=4\cos\vartheta(\cos\vartheta-\cos4\vartheta).
\end{aligned} n ρ = cos 5 ψ 4 sin 3 ψ sin 10 ψ cos 2 ψ = 8 sin 3 ψ sin 5 ψ cos 2 ψ = 4 cos 2 ψ ( cos 2 ψ − cos 8 ψ ) = 4 cos ϑ ( cos ϑ − cos 4 ϑ ) .
If m = 10 m=10 m = 10 , the points over Q Q Q are regular, and we apply Proposition 5.7 (r ∤ n r\nmid n r ∤ n and q ∣ n q\mid n q ∣ n ): sin ( i t o p + 2 q ) ψ = − sin 17 ψ < 0 \sin(i_{\mathrm{top}}+2q)\psi=-\sin17\psi<0 sin ( i top + 2 q ) ψ = − sin 17 ψ < 0 , and 2 n − i t o p − 2 q = 17 = n − 3 2n-i_{\mathrm{top}}-2q=17=n-3 2 n − i top − 2 q = 17 = n − 3 is the top of ι ( W ) \iota(\mathcal{W}) ι ( W ) , which is the chain of I χ I_{\chi} I χ with top 17 17 17 for 0 < χ < sin 3 ψ 0<\chi<\sin3\psi 0 < χ < sin 3 ψ (its neighbours 27 27 27 and − 3 -3 − 3 have negative sines). So the maximal cylinder containing Z ( W ) Z(\mathcal{W}) Z ( W ) has the same circumference and height 2 ∣ Q R ∣ sin 3 ψ 2|QR|\sin3\psi 2∣ QR ∣ sin 3 ψ , and n ρ = 8 cos ϑ ( cos ϑ − cos 4 ϑ ) n\rho=8\cos\vartheta(\cos\vartheta-\cos4\vartheta) n ρ = 8 cos ϑ ( cos ϑ − cos 4 ϑ ) .
The conjugates. n ρ n\rho n ρ is a polynomial in cos ϑ \cos\vartheta cos ϑ with rational coefficients, so ρ ∈ Q ( ζ ) + \rho\in\mathbb{Q}(\zeta)^+ ρ ∈ Q ( ζ ) + , and for a unit ℓ \ell ℓ modulo n n n , σ ℓ ( n ρ ) \sigma_{\ell}(n\rho) σ ℓ ( n ρ ) is the same expression with ϑ \vartheta ϑ replaced by ℓ ϑ \ell\vartheta ℓ ϑ ; it has the sign of σ ℓ ( ρ ) \sigma_{\ell}(\rho) σ ℓ ( ρ ) . (i) Let m ≥ 22 m\ge22 m ≥ 22 and ℓ = m / 2 − 2 \ell=m/2-2 ℓ = m /2 − 2 . It is odd, and gcd ( ℓ , m / 2 ) = gcd ( 2 , m / 2 ) = 1 \gcd(\ell,m/2)=\gcd(2,m/2)=1 g cd( ℓ , m /2 ) = g cd( 2 , m /2 ) = 1 , so ℓ \ell ℓ is a unit modulo n n n . As ℓ ϑ = π / 2 − 2 ϑ \ell\vartheta=\pi/2-2\vartheta ℓ ϑ = π /2 − 2 ϑ , cos ℓ ϑ = sin 2 ϑ \cos\ell\vartheta=\sin2\vartheta cos ℓ ϑ = sin 2 ϑ and cos 4 ℓ ϑ = cos ( 2 π − 8 ϑ ) = cos 8 ϑ \cos4\ell\vartheta=\cos(2\pi-8\vartheta)=\cos8\vartheta cos 4 ℓ ϑ = cos ( 2 π − 8 ϑ ) = cos 8 ϑ , so σ ℓ ( n ρ ) = 4 sin 2 ϑ ( sin 2 ϑ − cos 8 ϑ ) \sigma_{\ell}(n\rho)=4\sin2\vartheta(\sin2\vartheta-\cos8\vartheta) σ ℓ ( n ρ ) = 4 sin 2 ϑ ( sin 2 ϑ − cos 8 ϑ ) . Since 10 ϑ < π / 2 10\vartheta<\pi/2 10 ϑ < π /2 , we have 0 < 2 ϑ < π / 2 − 8 ϑ < π / 2 0<2\vartheta<\pi/2-8\vartheta<\pi/2 0 < 2 ϑ < π /2 − 8 ϑ < π /2 , so sin 2 ϑ < cos 8 ϑ \sin2\vartheta<\cos8\vartheta sin 2 ϑ < cos 8 ϑ and σ ℓ ( n ρ ) < 0 \sigma_{\ell}(n\rho)<0 σ ℓ ( n ρ ) < 0 . (ii) m = 14 m=14 m = 14 , ℓ = 11 \ell=11 ℓ = 11 : cos ( 11 π / 14 ) = − cos ( 3 π / 14 ) \cos(11\pi/14)=-\cos(3\pi/14) cos ( 11 π /14 ) = − cos ( 3 π /14 ) and cos ( 44 π / 14 ) = cos ( 8 π / 7 ) = − cos ( π / 7 ) \cos(44\pi/14)=\cos(8\pi/7)=-\cos(\pi/7) cos ( 44 π /14 ) = cos ( 8 π /7 ) = − cos ( π /7 ) , so σ 11 ( n ρ ) = − 4 cos ( 3 π / 14 ) ( cos ( π / 7 ) − cos ( 3 π / 14 ) ) < 0 \sigma_{11}(n\rho)=-4\cos(3\pi/14)(\cos(\pi/7)-\cos(3\pi/14))<0 σ 11 ( n ρ ) = − 4 cos ( 3 π /14 ) ( cos ( π /7 ) − cos ( 3 π /14 )) < 0 . (iii) m = 18 m=18 m = 18 , ℓ = 13 \ell=13 ℓ = 13 : cos ( 13 π / 18 ) = − cos ( 5 π / 18 ) \cos(13\pi/18)=-\cos(5\pi/18) cos ( 13 π /18 ) = − cos ( 5 π /18 ) and cos ( 52 π / 18 ) = \cos(52\pi/18)= cos ( 52 π /18 ) = cos ( 8 π / 9 ) = − cos ( π / 9 ) \cos(8\pi/9) = -\cos(\pi/9) cos ( 8 π /9 ) = − cos ( π /9 ) , so σ 13 ( n ρ ) = − 4 cos ( 5 π / 18 ) ( cos ( π / 9 ) − cos ( 5 π / 18 ) ) < 0 \sigma_{13}(n\rho) = -4\cos(5\pi/18)(\cos(\pi/9) - \cos(5\pi/18)) < 0 σ 13 ( n ρ ) = − 4 cos ( 5 π /18 ) ( cos ( π /9 ) − cos ( 5 π /18 )) < 0 . (iv) m = 10 m = 10 m = 10 , ℓ = 7 \ell= 7 ℓ = 7 : cos ( 7 π / 10 ) = − cos ( 3 π / 10 ) \cos(7\pi/10) = -\cos(3\pi/10) cos ( 7 π /10 ) = − cos ( 3 π /10 ) and cos ( 28 π / 10 ) = − cos ( π / 5 ) \cos(28\pi/10) = -\cos(\pi/5) cos ( 28 π /10 ) = − cos ( π /5 ) , so σ 7 ( n ρ ) = − 8 cos ( 3 π / 10 ) ( cos ( π / 5 ) − cos ( 3 π / 10 ) ) < 0 \sigma_7(n\rho) = -8\cos(3\pi/10)(\cos(\pi/5) - \cos(3\pi/10)) < 0 σ 7 ( n ρ ) = − 8 cos ( 3 π /10 ) ( cos ( π /5 ) − cos ( 3 π /10 )) < 0 . In each case Lemma 2.3(ii) applies. □
In general 4 cos a ( cos a − cos 4 a ) = 8 cos a sin ( 5 a / 2 ) sin ( 3 a / 2 ) 4\cos a(\cos a - \cos4a) = 8\cos a\sin(5a/2)\sin(3a/2) 4 cos a ( cos a − cos 4 a ) = 8 cos a sin ( 5 a /2 ) sin ( 3 a /2 ) is negative for a ∈ ( 0 , π ) a \in(0,\pi) a ∈ ( 0 , π ) exactly when a ∈ ( 2 π / 5 , π / 2 ) ∪ ( 2 π / 3 , 4 π / 5 ) a \in(2\pi/5,\pi/2)\cup(2\pi/3,4\pi/5) a ∈ ( 2 π /5 , π /2 ) ∪ ( 2 π /3 , 4 π /5 ) ; so σ ℓ ( ρ ( Z ) ) < 0 \sigma_\ell(\rho(Z)) < 0 σ ℓ ( ρ ( Z )) < 0 exactly when ± ℓ / n \pm\ell/n ± ℓ / n is congruent modulo 1 to a number in ( 1 / 5 , 1 / 4 ) ∪ ( 1 / 3 , 2 / 5 ) (1/5,1/4)\cup(1/3,2/5) ( 1/5 , 1/4 ) ∪ ( 1/3 , 2/5 ) .
Theorem 5.11. For every odd q ≥ 3 q \ge3 q ≥ 3 the triangle ( 4 , q , 3 q + 4 ) / ( 4 q + 8 ) (4,q,3q+4)/(4q+8) ( 4 , q , 3 q + 4 ) / ( 4 q + 8 ) is not a lattice triangle. Let n = 4 q + 8 n = 4q+8 n = 4 q + 8 , ψ = π / n \psi= \pi/n ψ = π / n and ϑ = 2 ψ \vartheta= 2\psi ϑ = 2 ψ .
(a) If q ≥ 7 q \ge7 q ≥ 7 , in the direction of the side joining the vertices of angles 4 ψ 4\psi 4 ψ and q ψ q\psi q ψ the unfolding has a maximal cylinder Z Z Z with n ρ ( Z ) = 2 ( 1 + cos 2 ϑ + sin 2 ϑ ) n\rho(Z) = 2(1+\cos2\vartheta+\sin2\vartheta) n ρ ( Z ) = 2 ( 1 + cos 2 ϑ + sin 2 ϑ ) , and σ ℓ ( ρ ( Z ) ) < 0 \sigma_\ell(\rho(Z)) < 0 σ ℓ ( ρ ( Z )) < 0 for ℓ = q \ell= q ℓ = q if q ≡ 3 ( m o d 4 ) q \equiv3 \pmod{4} q ≡ 3 ( mod 4 ) and for ℓ = q − 2 \ell= q-2 ℓ = q − 2 if q ≡ 1 ( m o d 4 ) q \equiv1 \pmod{4} q ≡ 1 ( mod 4 ) .
(b) If q = 5 q = 5 q = 5 , in the same direction the unfolding has a maximal cylinder Z Z Z with n ρ ( Z ) = 8 cos 2 4 ψ cos 2 ψ sin 3 ψ / sin 5 ψ n\rho(Z) = 8\cos^2 4\psi\cos2\psi\sin3\psi/\sin5\psi n ρ ( Z ) = 8 cos 2 4 ψ cos 2 ψ sin 3 ψ / sin 5 ψ , and σ 11 ( ρ ( Z ) ) < 0 \sigma_{11}(\rho(Z)) < 0 σ 11 ( ρ ( Z )) < 0 .
(c) If q = 3 q = 3 q = 3 , the triangle is ( 3 , 4 , 13 ) / 20 (3,4,13)/20 ( 3 , 4 , 13 ) /20 , which is not a lattice triangle by Theorem 6.1 .
For q ≥ 5 q \ge5 q ≥ 5 these triangles form family (6) of Theorem 2.5 : their obtuse angle exceeds 2 π / 3 2\pi/3 2 π /3 exactly when q > 4 q > 4 q > 4 . For q = 3 q = 3 q = 3 the obtuse angle is 13 π / 20 < 2 π / 3 13\pi/20 < 2\pi/3 13 π /20 < 2 π /3 , and ( 3 , 4 , 13 ) / 20 (3,4,13)/20 ( 3 , 4 , 13 ) /20 is a member of E \mathcal{E} E . For q = 1 q = 1 q = 1 the triangle is ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 once more.
Proof. Here p = 4 p = 4 p = 4 , r = 3 q + 4 r = 3q+4 r = 3 q + 4 , ( q + 2 ) ψ = π / 4 (q+2)\psi= \pi/4 ( q + 2 ) ψ = π /4 and q ψ = π / 4 − 2 ψ q\psi= \pi/4 - 2\psi q ψ = π /4 − 2 ψ . The hypotheses of Lemma 5.1 hold, n n n is even and q q q is odd; n / 2 = 2 q + 4 ≡ 2 ( m o d 4 ) n/2 = 2q+4 \equiv2 \pmod{4} n /2 = 2 q + 4 ≡ 2 ( mod 4 ) , so G j G_j G j is symmetric if and only if j j j is odd. Since gcd ( 3 q + 4 , 4 q + 8 ) = gcd ( q + 4 , 8 ) = 1 \gcd(3q + 4, 4q + 8) = \gcd(q + 4, 8) = 1 g cd( 3 q + 4 , 4 q + 8 ) = g cd( q + 4 , 8 ) = 1 , the vertex R R R gives one singular point, and r ∤ n r \nmid n r ∤ n , as Proposition 5.5 requires; since gcd ( q , 4 q + 8 ) = gcd ( q , 8 ) = 1 \gcd(q, 4q + 8) = \gcd(q, 8) = 1 g cd( q , 4 q + 8 ) = g cd( q , 8 ) = 1 , the vertex Q Q Q gives one point of cone angle 2 π q 2\pi q 2 π q , singular for q ≥ 3 q \ge3 q ≥ 3 .
Let q ≥ 5 q \ge5 q ≥ 5 and W = { q + 2 , 3 q + 2 } W = \{q + 2, 3q + 2\} W = { q + 2 , 3 q + 2 } (Figure 2 , right, for q = 7 q = 7 q = 7 ). Its sines are sin ( π / 4 ) \sin(\pi/4) sin ( π /4 ) and sin ( 3 q + 2 ) ψ = sin ( q + 6 ) ψ > sin ( π / 4 ) \sin(3q + 2)\psi= \sin(q + 6)\psi> \sin(\pi/4) sin ( 3 q + 2 ) ψ = sin ( q + 6 ) ψ > sin ( π /4 ) , because π / 4 < ( q + 6 ) ψ < 3 π / 4 \pi/4 < (q + 6)\psi< 3\pi/4 π /4 < ( q + 6 ) ψ < 3 π /4 . The neighbours are ( 3 q + 2 ) + 2 q = n + ( q − 6 ) (3q + 2) + 2q = n + (q - 6) ( 3 q + 2 ) + 2 q = n + ( q − 6 ) , with sine − sin ( q − 6 ) ψ -\sin(q - 6)\psi − sin ( q − 6 ) ψ , and 2 − q < 0 2 - q < 0 2 − q < 0 . So for q ≥ 7 q \ge7 q ≥ 7 , a ( W ) = 0 a(W) = 0 a ( W ) = 0 ; for q = 5 q = 5 q = 5 , a ( W ) = sin ψ a(W) = \sin\psi a ( W ) = sin ψ ; in both cases b ( W ) = sin ( π / 4 ) > a ( W ) b(W) = \sin(\pi/4) > a(W) b ( W ) = sin ( π /4 ) > a ( W ) , and ι ( W ) = { q + 6 , 3 q + 6 } ≠ W \iota(W) = \{q + 6, 3q + 6\} \ne W ι ( W ) = { q + 6 , 3 q + 6 } = W .
The exit ears are given by $2k_1 \equiv (q + 2) - n - q = 2 - na n d $ 2 k 2 ≡ − ( 3 q + 2 ) − q = 6 − n and \$2k_2 \equiv-(3q + 2) - q = 6 - n an d $2 k 2 ≡ − ( 3 q + 2 ) − q = 6 − n (mod 2 n 2n 2 n ), so k 1 ≡ − 2 q − 3 k_1 \equiv-2q - 3 k 1 ≡ − 2 q − 3 and k 2 ≡ − 2 q − 1 k_2 \equiv-2q - 1 k 2 ≡ − 2 q − 1 (mod n n n ), both odd, so 2 k 1 ≡ 2 k 2 ≡ 2 ≡ n / 2 2k_1 \equiv2k_2 \equiv2 \equiv n/2 2 k 1 ≡ 2 k 2 ≡ 2 ≡ n /2 (mod 4 4 4 ), which is (5.1). So Proposition 5.5 applies:
L ( W ) = 2 [ cos 2 ψ − cos ( 4 q + 2 ) ψ ] = 2 [ cos 2 ψ + cos 6 ψ ] = 4 cos 2 ψ cos 4 ψ , \mathcal{L}(W) = 2[\cos2\psi- \cos(4q + 2)\psi] = 2[\cos2\psi+ \cos6\psi] = 4\cos2\psi\cos4\psi, L ( W ) = 2 [ cos 2 ψ − cos ( 4 q + 2 ) ψ ] = 2 [ cos 2 ψ + cos 6 ψ ] = 4 cos 2 ψ cos 4 ψ ,
as ( 4 q + 2 ) ψ = π − 6 ψ (4q + 2)\psi= \pi- 6\psi ( 4 q + 2 ) ψ = π − 6 ψ , and Z ( W ) Z(W) Z ( W ) is a maximal cylinder: for q ≥ 7 q \ge7 q ≥ 7 its lower boundary contains the apex O O O , a point over Q Q Q , and for q = 5 q = 5 q = 5 a point over R R R .
Part (a). The height is ∣ Q R ∣ sin ( π / 4 ) |QR|\sin(\pi/4) ∣ QR ∣ sin ( π /4 ) and the area of X X X is n ∣ Q R ∣ sin q ψ n|QR|\sin q\psi n ∣ QR ∣ sin q ψ , so
n ρ = 4 sin ( π / 4 ) cos 2 ψ cos 4 ψ sin ( π / 4 − 2 ψ ) = 4 cos 2 ψ cos 4 ψ cos 2 ψ − sin 2 ψ = 4 cos 2 ψ ( cos 2 ψ + sin 2 ψ ) = 2 ( 1 + cos 2 ϑ + sin 2 ϑ ) . \begin{aligned}
n\rho= \frac{4\sin(\pi/4)\cos2\psi\cos4\psi}{\sin(\pi/4 - 2\psi)} = \frac{4\cos2\psi\cos4\psi}{\cos2\psi- \sin2\psi} \\
&= 4\cos2\psi(\cos2\psi+ \sin2\psi) = 2(1 + \cos2\vartheta+ \sin2\vartheta).
\end{aligned} n ρ = sin ( π /4 − 2 ψ ) 4 sin ( π /4 ) cos 2 ψ cos 4 ψ = cos 2 ψ − sin 2 ψ 4 cos 2 ψ cos 4 ψ = 4 cos 2 ψ ( cos 2 ψ + sin 2 ψ ) = 2 ( 1 + cos 2 ϑ + sin 2 ϑ ) .
Since ( q + 2 ) ϑ = π / 2 (q + 2)\vartheta= \pi/2 ( q + 2 ) ϑ = π /2 , sin 2 ϑ = cos q ϑ \sin2\vartheta= \cos q\vartheta sin 2 ϑ = cos q ϑ , so n ρ = 2 + 2 cos 2 ϑ + 2 cos q ϑ ∈ Q ( ζ ) + n\rho= 2 + 2\cos2\vartheta+ 2\cos q\vartheta\in\mathbb{Q}(\zeta)^+ n ρ = 2 + 2 cos 2 ϑ + 2 cos q ϑ ∈ Q ( ζ ) + and σ ℓ ( n ρ ) = 2 + 2 cos 2 ℓ ϑ + 2 cos q ℓ ϑ \sigma_\ell(n\rho) = 2 + 2\cos2\ell\vartheta+ 2\cos q\ell\vartheta σ ℓ ( n ρ ) = 2 + 2 cos 2 ℓ ϑ + 2 cos q ℓ ϑ . For odd ℓ \ell ℓ , cos q ℓ ϑ = cos ( ℓ π / 2 − 2 ℓ ϑ ) = ε ℓ sin 2 ℓ ϑ \cos q\ell\vartheta= \cos(\ell\pi/2 - 2\ell\vartheta) = \varepsilon_\ell\sin2\ell\vartheta cos q ℓ ϑ = cos ( ℓ π /2 − 2 ℓ ϑ ) = ε ℓ sin 2 ℓ ϑ with ε ℓ = 1 \varepsilon_\ell= 1 ε ℓ = 1 if ℓ ≡ 1 \ell\equiv1 ℓ ≡ 1 and ε ℓ = − 1 \varepsilon_\ell= -1 ε ℓ = − 1 if ℓ ≡ 3 \ell\equiv3 ℓ ≡ 3 (mod 4 4 4 ). If q ≡ 3 q \equiv3 q ≡ 3 (mod 4 4 4 ), ℓ = q \ell= q ℓ = q is a unit (gcd ( q , 8 ) = 1 \gcd(q, 8) = 1 g cd( q , 8 ) = 1 ), 2 q ϑ = π − 4 ϑ 2q\vartheta= \pi- 4\vartheta 2 q ϑ = π − 4 ϑ , and σ q ( n ρ ) = 2 ( 1 − cos 4 ϑ − sin 4 ϑ ) = 2 − 2 2 sin ( 4 ϑ + π / 4 ) < 0 \sigma_q(n\rho) = 2(1 - \cos4\vartheta- \sin4\vartheta) = 2 - 2\sqrt{2}\sin(4\vartheta+ \pi/4) < 0 σ q ( n ρ ) = 2 ( 1 − cos 4 ϑ − sin 4 ϑ ) = 2 − 2 2 sin ( 4 ϑ + π /4 ) < 0 , because 0 < 4 ϑ < π / 2 0 < 4\vartheta< \pi/2 0 < 4 ϑ < π /2 . If q ≡ 1 q \equiv1 q ≡ 1 (mod 4 4 4 ), then q ≥ 9 q \ge9 q ≥ 9 and ℓ = q − 2 ≡ 3 \ell= q - 2 \equiv3 ℓ = q − 2 ≡ 3 (mod 4 4 4 ) is a unit (it is odd and gcd ( q − 2 , q + 2 ) \gcd(q - 2, q + 2) g cd( q − 2 , q + 2 ) divides 4 4 4 ); 2 ( q − 2 ) ϑ = π − 8 ϑ 2(q - 2)\vartheta= \pi- 8\vartheta 2 ( q − 2 ) ϑ = π − 8 ϑ , and σ q − 2 ( n ρ ) = 2 ( 1 − cos 8 ϑ − sin 8 ϑ ) < 0 \sigma_{q-2}(n\rho) = 2(1 - \cos8\vartheta- \sin8\vartheta) < 0 σ q − 2 ( n ρ ) = 2 ( 1 − cos 8 ϑ − sin 8 ϑ ) < 0 , because 0 < 8 ϑ < π / 2 0 < 8\vartheta< \pi/2 0 < 8 ϑ < π /2 for n ≥ 44 n \ge44 n ≥ 44 .
Part (b). Now ψ = π / 28 \psi= \pi/28 ψ = π /28 and π / 4 = 7 ψ \pi/4 = 7\psi π /4 = 7 ψ ; the height is ∣ Q R ∣ ( sin 7 ψ − sin ψ ) = 2 ∣ Q R ∣ cos 4 ψ sin 3 ψ |QR|(\sin7\psi- \sin\psi) = 2|QR|\cos4\psi\sin3\psi ∣ QR ∣ ( sin 7 ψ − sin ψ ) = 2∣ QR ∣ cos 4 ψ sin 3 ψ and sin q ψ = sin 5 ψ \sin q\psi= \sin5\psi sin q ψ = sin 5 ψ , so n ρ = 8 cos 2 4 ψ cos 2 ψ sin 3 ψ / sin 5 ψ n\rho= 8\cos^2 4\psi\cos2\psi\sin3\psi/\sin5\psi n ρ = 8 cos 2 4 ψ cos 2 ψ sin 3 ψ / sin 5 ψ . With ζ = e i ϑ \zeta= e^{i\vartheta} ζ = e i ϑ we have sin 3 ψ / sin 5 ψ = ζ ( ζ 3 − 1 ) / ( ζ 5 − 1 ) ∈ Q ( ζ ) \sin3\psi/\sin5\psi= \zeta(\zeta^3 - 1)/(\zeta^5 - 1) \in\mathbb{Q}(\zeta) sin 3 ψ / sin 5 ψ = ζ ( ζ 3 − 1 ) / ( ζ 5 − 1 ) ∈ Q ( ζ ) , so ρ ∈ Q ( ζ ) + \rho\in\mathbb{Q}(\zeta)^+ ρ ∈ Q ( ζ ) + , and σ ℓ \sigma_\ell σ ℓ maps sin 3 ψ / sin 5 ψ \sin3\psi/\sin5\psi sin 3 ψ / sin 5 ψ to sin 3 ℓ ψ / sin 5 ℓ ψ \sin3\ell\psi/\sin5\ell\psi sin 3 ℓ ψ / sin 5 ℓ ψ . So σ 11 ( n ρ ) = 8 cos 2 ( 44 ψ ) cos ( 22 ψ ) sin ( 33 ψ ) / sin ( 55 ψ ) \sigma_{11}(n\rho) = 8\cos^2(44\psi)\cos(22\psi)\sin(33\psi)/\sin(55\psi) σ 11 ( n ρ ) = 8 cos 2 ( 44 ψ ) cos ( 22 ψ ) sin ( 33 ψ ) / sin ( 55 ψ ) , where cos 22 ψ = cos ( 11 π / 14 ) < 0 \cos22\psi= \cos(11\pi/14) < 0 cos 22 ψ = cos ( 11 π /14 ) < 0 , sin 33 ψ = sin ( 33 π / 28 ) < 0 \sin33\psi= \sin(33\pi/28) < 0 sin 33 ψ = sin ( 33 π /28 ) < 0 and sin 55 ψ = − sin ( π / 28 ) < 0 \sin55\psi= -\sin(\pi/28) < 0 sin 55 ψ = − sin ( π /28 ) < 0 ; so σ 11 ( ρ ) < 0 \sigma_{11}(\rho) < 0 σ 11 ( ρ ) < 0 .
Part (c) is Theorem 6.1 .
In parts (a) and (b), Lemma 2.3 (ii) applies. □
At q = 3 q = 3 q = 3 the cylinder Z ( W ) Z(W) Z ( W ) of the proof still exists, with a ( W ) = sin 3 ψ a(W) = \sin3\psi a ( W ) = sin 3 ψ , but its share n ρ = 4 cos 2 ( π / 10 ) − 2 cos ( π / 10 ) n\rho= 4\cos^2(\pi/10) - 2\cos(\pi/10) n ρ = 4 cos 2 ( π /10 ) − 2 cos ( π /10 ) is totally positive, and we find that all eight maximal cylinders in the direction of P Q PQ P Q have the same modulus, so that direction is consistent with Lemma 2.3 and Theorem 6.1 uses another one.
Two families with an angle π / n \pi/n π / n : the star
In this subsection Δ = ( 1 , q , r ) / n \Delta= (1,q,r)/n Δ = ( 1 , q , r ) / n has vertices A A A , B B B , C C C with angles ψ \psi ψ , q ψ q\psi q ψ , r ψ r\psi r ψ , where ψ = π / n \psi= \pi/n ψ = π / n as before, is obtuse at C C C , that is, r > n / 2 r > n/2 r > n /2 , and has ∣ B C ∣ = 1 |BC| = 1 ∣ B C ∣ = 1 . Place A A A at the origin O O O and, for a real number ϕ \phi ϕ ,
B k = ∣ A B ∣ e i ( 2 k ψ + ϕ ) , C k = ∣ A C ∣ e i ( ( 2 k + 1 ) ψ + ϕ ) ( k ∈ Z / n ) , B_k = |AB|e^{i(2k\psi+\phi)}, \qquad C_k = |AC|e^{i((2k+1)\psi+\phi)} \qquad(k \in\mathbb{Z}/n), B k = ∣ A B ∣ e i ( 2 k ψ + ϕ ) , C k = ∣ A C ∣ e i (( 2 k + 1 ) ψ + ϕ ) ( k ∈ Z / n ) ,
∣ A B ∣ = sin r ψ sin ψ , ∣ A C ∣ = sin q ψ sin ψ . |AB| = \frac{\sin r\psi}{\sin\psi}, \qquad|AC| = \frac{\sin q\psi}{\sin\psi}. ∣ A B ∣ = sin ψ sin r ψ , ∣ A C ∣ = sin ψ sin q ψ .
The star P P P is the union of the 2 n 2n 2 n triangles Δ 2 k = O B k C k \Delta_{2k}=OB_kC_k Δ 2 k = O B k C k and Δ 2 k + 1 = O C k B k + 1 \Delta_{2k+1}=OC_kB_{k+1} Δ 2 k + 1 = O C k B k + 1 , each a copy of Δ \Delta Δ with the angle ψ \psi ψ at O O O . Its sides are b k = B k C k b_k=B_kC_k b k = B k C k and c k = C k B k + 1 c_k=C_kB_{k+1} c k = C k B k + 1 , all of length 1 1 1 ; its interior angles are 2 q ψ 2q\psi 2 q ψ at the B k B_k B k and 2 r ψ > π 2r\psi>\pi 2 r ψ > π at the C k C_k C k , which we call the inner vertices. For ϕ = 0 \phi=0 ϕ = 0 the star is ∣ A B ∣ |AB| ∣ A B ∣ times the model of Lemma 5.1 for p = 1 p=1 p = 1 and ( P , Q , R ) = ( A , B , C ) (P,Q,R)=(A,B,C) ( P , Q , R ) = ( A , B , C ) (its hypotheses hold, as q < n / 2 q<n/2 q < n /2 ): B k = ∣ A B ∣ Θ k ( Q ) B_k=|AB|\Theta_k(Q) B k = ∣ A B ∣ Θ k ( Q ) and C k = ∣ A B ∣ Θ k ( R ) C_k=|AB|\Theta_k(R) C k = ∣ A B ∣ Θ k ( R ) , the regular n n n -gon G 0 G_0 G 0 is the union of the triangles O C k − 1 C k OC_{k-1}C_k O C k − 1 C k , the ear E k E_k E k is C k − 1 B k C k C_{k-1}B_kC_k C k − 1 B k C k , and the gluing of the leg Θ k ( Q R ) \Theta_k(QR) Θ k ( QR ) to Θ k − q ( Q R ‾ ) \Theta_{k-q}(\overline{QR}) Θ k − q ( QR ) is that of b k b_k b k to c k − q − 1 c_{k-q-1} c k − q − 1 in Lemma 5.12 .
Lemma 5.12. The unfolding X X X is P P P with b k b_k b k glued by translation to c k − q − 1 c_{k-q-1} c k − q − 1 , the gluing matching B k B_k B k with B k − q B_{k-q} B k − q and C k C_k C k with C k − q − 1 C_{k-q-1} C k − q − 1 ; the point O O O is a regular point. The B k B_k B k form gcd ( q , n ) \gcd(q,n) g cd( q , n ) points of cone angle 2 π q / gcd ( q , n ) 2\pi q/\gcd(q,n) 2 π q / g cd( q , n ) and the C k C_k C k form gcd ( r , n ) \gcd(r,n) g cd( r , n ) points of cone angle 2 π r / gcd ( r , n ) 2\pi r/\gcd(r,n) 2 π r / g cd( r , n ) .
Proof. It suffices to take ϕ = 0 \phi=0 ϕ = 0 . The linear part of Δ 2 k \Delta_{2k} Δ 2 k is the rotation Θ k \Theta_k Θ k and that of Δ 2 k + 1 \Delta_{2k+1} Δ 2 k + 1 is Θ k + 1 κ \Theta_{k+1}\kappa Θ k + 1 κ , κ \kappa κ complex conjugation; these 2 n 2n 2 n elements are distinct, so they are all of the dihedral group, and every copy of Δ \Delta Δ in X X X is one Δ j \Delta_j Δ j . The gluings across the sides through A A A put Δ j \Delta_j Δ j next to Δ j ± 1 \Delta_{j\pm1} Δ j ± 1 around A A A , which gives P P P , and A A A is a regular point. Crossing the side B C BC B C of the copy with linear part δ \delta δ leads to the copy with linear part δ κ B C \delta\kappa_{BC} δ κ B C , where κ B C = Θ − q κ \kappa_{BC}=\Theta_{-q}\kappa κ B C = Θ − q κ is the linear part of the reflection in B C BC B C (which has direction π − q ψ \pi-q\psi π − q ψ ). Since Θ k κ B C = Θ ( k − q − 1 ) + 1 κ \Theta_k\kappa_{BC}=\Theta_{(k-q-1)+1}\kappa Θ k κ B C = Θ ( k − q − 1 ) + 1 κ , the side b k b_k b k of Δ 2 k \Delta_{2k} Δ 2 k is glued to the side c k − q − 1 = C k − q − 1 B k − q c_{k-q-1}=C_{k-q-1}B_{k-q} c k − q − 1 = C k − q − 1 B k − q of Δ 2 ( k − q − 1 ) + 1 \Delta_{2(k-q-1)+1} Δ 2 ( k − q − 1 ) + 1 , matching the copies of B B B and the copies of C C C . The cone angles are as in § 2.1 . □
Two properties of P P P are used. (a) P P P is star-shaped about O O O : Δ j \Delta_j Δ j fills the closed sector between the arguments of its two outer vertices, an angle ψ < π \psi<\pi ψ < π ; these sectors tile the full turn, and inside its sector a ray from O O O meets the outer side of Δ j \Delta_j Δ j once. So each ray from O O O meets ∂ P \partial P ∂ P in exactly one point β ( α ) \beta(\alpha) β ( α ) , α \alpha α the argument of the ray, and the open segment from O O O to β ( α ) \beta(\alpha) β ( α ) lies in the interior of P P P . (b) Mirrors: the reflection in the line through O O O and any vertex maps P P P to itself (it sends the vertex at argument β + j ψ \beta+j\psi β + j ψ to the one at β − j ψ \beta-j\psi β − j ψ , which is of the same type, B B B or C C C ).
Lemma 5.13 (chords). Let h > 0 h>0 h > 0 and let p + , p − ∈ ∂ P p_+,p_-\in\partial P p + , p − ∈ ∂ P be points of height h h h , neither a vertex, lying on two different sides of P P P . Suppose that every vertex of P P P met when going counterclockwise along ∂ P \partial P ∂ P from p + p_+ p + to p − p_- p − has height greater than h h h . Then the open horizontal segment from p − p_- p − to p + p_+ p + lies in the interior of P P P , and p − p_- p − is to the left of p + p_+ p + .
Proof. Height is affine along each side, equal to h h h at p + p_+ p + and p − p_- p − and greater than h h h at the vertices between, so the open boundary arc from p + p_+ p + to p − p_- p − lies in { y > h } \{y>h\} { y > h } ; by (a) it is the set of β ( α ) \beta(\alpha) β ( α ) with α \alpha α in an interval ( α + , α − ) ⊂ ( 0 , π ) (\alpha_+,\alpha_-)\subset(0,\pi) ( α + , α − ) ⊂ ( 0 , π ) , where α + , α − \alpha_+,\alpha_- α + , α − are the arguments of p + , p − p_+,p_- p + , p − . On the line y = h y=h y = h the argument decreases strictly from left to right, so the open segment between p − p_- p − and p + p_+ p + is { w : Im w = h , α + < arg w < α − } \{w:\operatorname{Im}w=h,\ \alpha_+<\arg w<\alpha_-\} { w : Im w = h , α + < arg w < α − } , and for such w w w , ∣ w ∣ = h / sin ( arg w ) < ∣ β ( arg w ) ∣ |w|=h/\sin(\arg w)<|\beta(\arg w)| ∣ w ∣ = h / sin ( arg w ) < ∣ β ( arg w ) ∣ , because β ( arg w ) \beta(\arg w) β ( arg w ) has height greater than h h h . So w w w lies on the open segment from O O O to β ( arg w ) \beta(\arg w) β ( arg w ) , which is in the interior of P P P by (a). Finally α − → α + \alpha_-\to\alpha_+ α − → α + gives x ( p − ) < x ( p + ) x(p_-)<x(p_+) x ( p − ) < x ( p + ) . □
Strips. Let e + e_+ e + be a side of P P P that rises when ∂ P \partial P ∂ P is traversed counterclockwise, e − e_- e − a side that falls, and 0 ≤ a < b 0\leq a<b 0 ≤ a < b such that both sides cross every height in [ a , b ] [a,b] [ a , b ] . If every vertex on the counterclockwise arc from the upper end of e + e_+ e + to the upper end of e − e_- e − (both ends included) has height at least b b b , then by Lemma 5.13 (the two sides are different, one rising and one falling), for every h ∈ ( a , b ) h\in(a,b) h ∈ ( a , b ) , the open horizontal segment at height h h h from e − e_- e − to e + e_+ e + lies in the interior of P \mathcal{P} P . The union of these segments is the strip of heights ( a , b ) (a,b) ( a , b ) with left side e − e_- e − and right side e + e_+ e + . By (b), the same holds after reflection in a horizontal mirror of P \mathcal{P} P .
Lemma 5.14 (annuli). Let S 1 , … , S k S_1,\ldots,S_k S 1 , … , S k be disjoint strips , S i S_i S i of heights ( a i , a i + h ) (a_i,a_i+h) ( a i , a i + h ) with left side e ˉ i − \bar{e}_i^- e ˉ i − and right side e ˉ i + \bar{e}_i^+ e ˉ i + , and suppose that for each i i i the gluing of Lemma 5.12 maps e ˉ i + \bar{e}_i^+ e ˉ i + to e ˉ i + 1 − \bar{e}_{i+1}^- e ˉ i + 1 − (indices modulo k k k ), sending the point of e ˉ i + \bar{e}_i^+ e ˉ i + at height a i + t a_i+t a i + t to the point of e ˉ i + 1 − \bar{e}_{i+1}^- e ˉ i + 1 − at height a i + 1 + t a_{i+1}+t a i + 1 + t for 0 < t < h 0<t<h 0 < t < h . Then the S i S_i S i , with the glued open pieces of the e ˉ i + \bar{e}_i^+ e ˉ i + , form an open annulus Y ⊂ X Y\subset X Y ⊂ X foliated by closed horizontal leaves, of height h h h and circumference ∑ i ℓ i ( a i + t ) \sum_i \ell_i(a_i+t) ∑ i ℓ i ( a i + t ) , where ℓ i ( y ) \ell_i(y) ℓ i ( y ) is the length of the chord of S i S_i S i at height y y y ; this sum does not depend on t t t . If each boundary component of Y Y Y contains a point of Σ \Sigma Σ , then Y Y Y is a maximal cylinder.
Proof. The horizontal line through S 1 S_1 S 1 at height a 1 + t a_1+t a 1 + t crosses S 1 , … , S k S_1,\ldots,S_k S 1 , … , S k in turn and re-enters S 1 S_1 S 1 at height a 1 + t a_1+t a 1 + t : it is closed. Distinct values of t t t give disjoint leaves, and each strip is crossed once per turn, so Y Y Y is an embedded annulus. The leaves are homotopic in Y Y Y and the holonomy of a closed curve is a homotopy invariant, so the circumference does not depend on t t t . □ \square □
Theorem 5.15. For every even m ≥ 8 m\ge8 m ≥ 8 the triangle ( 1 , m − 2 , m + 1 ) / 2 m (1,m-2,m+1)/2m ( 1 , m − 2 , m + 1 ) /2 m is not a lattice triangle. With ψ = π / 2 m \psi=\pi/2m ψ = π /2 m , in the direction of the side joining the vertices of angles ψ \psi ψ and ( m + 1 ) ψ (m+1)\psi ( m + 1 ) ψ the unfolding has two maximal cylinders Z Z Z and Z ′ Z' Z ′ of moduli
μ ( Z ) = sin ψ 4 cos ψ cos 2 ψ , μ ( Z ′ ) = sin ψ 16 cos 3 ψ cos 2 ψ , \mu(Z)=\frac{\sin\psi}{4\cos\psi\cos2\psi},\qquad\mu(Z')=\frac{\sin\psi}{16\cos^3\psi\cos2\psi}, μ ( Z ) = 4 cos ψ cos 2 ψ sin ψ , μ ( Z ′ ) = 16 cos 3 ψ cos 2 ψ sin ψ , whose ratio 4 cos 2 ψ = 2 + 2 cos ( π / m ) 4\cos^2\psi=2+2\cos(\pi/m) 4 cos 2 ψ = 2 + 2 cos ( π / m ) is irrational.
These triangles form the part of C \mathcal{C} C with p = 1 p=1 p = 1 ; for m = 4 m=4 m = 4 and m = 6 m=6 m = 6 the triangle is the lattice triangle ( 1 , 2 , 5 ) / 8 (1,2,5)/8 ( 1 , 2 , 5 ) /8 , respectively ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 . For odd m m m the triangle lies in A \mathcal{A} A (with s = 2 s=2 s = 2 ): for m = 3 m=3 m = 3 it is the lattice triangle ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 , and for odd m ≥ 5 m\ge5 m ≥ 5 it is treated in § 4 (Theorem 4.18 ).
Proof. Here n = 2 m n=2m n = 2 m , q = m − 2 q=m-2 q = m − 2 , r = m + 1 r=m+1 r = m + 1 , ∣ A B ∣ = cot ψ |AB|=\cot\psi ∣ A B ∣ = cot ψ and ∣ A C ∣ = cos 2 ψ / sin ψ |AC|=\cos2\psi/\sin\psi ∣ A C ∣ = cos 2 ψ / sin ψ . Position the star so that an inner vertex, the notch , lies on the positive y y y -axis (Figure 3 ), and let V j V_j V j (j ∈ Z / 4 m j\in\mathbb{Z}/4m j ∈ Z /4 m ) be the vertex at argument π / 2 − j ψ \pi/2-j\psi π /2 − j ψ : V j = ∣ V j ∣ ( sin j ψ , cos j ψ ) V_j=|V_j|(\sin j\psi,\cos j\psi) V j = ∣ V j ∣ ( sin j ψ , cos j ψ ) , with ∣ V j ∣ = ∣ A C ∣ |V_j|=|AC| ∣ V j ∣ = ∣ A C ∣ for j j j even (a copy of C C C ) and ∣ V j ∣ = ∣ A B ∣ |V_j|=|AB| ∣ V j ∣ = ∣ A B ∣ for j j j odd (a copy of B B B ). Then V 0 V_0 V 0 is the notch, and V m V_m V m lies on the positive x x x -axis. Write y j y_j y j for the height of V j V_j V j and Ξ j \Xi_j Ξ j for the side with ends V j , V j − 1 V_j,V_{j-1} V j , V j − 1 .
Figure 3. Ward’s star for m = 8 m = 8 m = 8 , that is for ( 1 , 6 , 9 ) / 16 (1, 6, 9)/16 ( 1 , 6 , 9 ) /16 , placed as in the proof of Theorem 5.15 , the notch V 0 V_0 V 0 on the positive y y y -axis; for odd j j j the side Ξ j \Xi_j Ξ j is glued to Ξ j + 2 m − 3 \Xi_{j+2m-3} Ξ j + 2 m − 3 . Shaded: the cylinders Z ′ Z' Z ′ (dark), of height sin ψ \sin\psi sin ψ , and Z Z Z (light), of height cos ψ \cos\psi cos ψ ; dashed: their core curves.
Mirrors. The reflection x ↦ − x x\mapsto-x x ↦ − x sends V j V_j V j to V − j V_{-j} V − j , and y ↦ − y y\mapsto-y y ↦ − y sends V j V_j V j to V 2 m − j V_{2m-j} V 2 m − j .
Gluing. For j j j odd, Ξ j \Xi_j Ξ j is glued to Ξ j + 2 m − 3 \Xi_{j+2m-3} Ξ j + 2 m − 3 , with V j ↔ V j + 2 m − 4 V_j\leftrightarrow V_{j+2m-4} V j ↔ V j + 2 m − 4 and V j − 1 ↔ V j + 2 m − 3 V_{j-1}\leftrightarrow V_{j+2m-3} V j − 1 ↔ V j + 2 m − 3 ; equivalently, a side Ξ i \Xi_i Ξ i with i i i even is glued to Ξ i + 2 m + 3 \Xi_{i+2m+3} Ξ i + 2 m + 3 . Indeed, in the labels of Lemma 5.12 , B k = V j 0 − 2 k B_k=V_{j_0-2k} B k = V j 0 − 2 k and C k = V j 0 − 2 k − 1 C_k=V_{j_0-2k-1} C k = V j 0 − 2 k − 1 for a fixed odd j 0 j_0 j 0 , so b k = Ξ j 0 − 2 k b_k=\Xi_{j_0-2k} b k = Ξ j 0 − 2 k and c k − q − 1 = Ξ j 0 − 2 k + 2 q + 1 c_{k-q-1}=\Xi_{j_0-2k+2q+1} c k − q − 1 = Ξ j 0 − 2 k + 2 q + 1 , with B k ↔ B k − q − 1 B_k\leftrightarrow B_{k-q-1} B k ↔ B k − q − 1 and C k ↔ C k − q − 1 C_k\leftrightarrow C_{k-q-1} C k ↔ C k − q − 1 , where k − q + 1 k-q+1 k − q + 1 and q + 1 q+1 q + 1 are 2 m − 2 2m-2 2 m − 2 and 2 m 2m 2 m .
Heights. For 1 ≤ j ≤ m 1\le j\le m 1 ≤ j ≤ m , w j : = y j − 1 − y j w_j:=y_{j-1}-y_j w j := y j − 1 − y j equals sin ( j + 1 ) ψ \sin(j+1)\psi sin ( j + 1 ) ψ if j j j is even and sin ( j − 2 ) ψ \sin(j-2)\psi sin ( j − 2 ) ψ if j j j is odd: for j j j even, y j − 1 − y j = [ cos ψ cos ( j − 1 ) ψ − cos 2 ψ cos j ψ ] / sin ψ = [ cos j ψ − cos ( j + 2 ) ψ ] / ( 2 sin ψ ) y_{j-1}-y_j=[\cos\psi\cos(j-1)\psi-\cos2\psi\cos j\psi]/\sin\psi=[\cos j\psi-\cos(j+2)\psi]/(2\sin\psi) y j − 1 − y j = [ cos ψ cos ( j − 1 ) ψ − cos 2 ψ cos j ψ ] / sin ψ = [ cos j ψ − cos ( j + 2 ) ψ ] / ( 2 sin ψ ) , and for j j j odd, [ cos 2 ψ cos ( j − 1 ) ψ − cos ψ cos j ψ ] / sin ψ = [ cos ( j − 3 ) ψ − cos ( j − 1 ) ψ ] / ( 2 sin ψ ) [\cos2\psi\cos(j-1)\psi-\cos\psi\cos j\psi]/\sin\psi=[\cos(j-3)\psi-\cos(j-1)\psi]/(2\sin\psi) [ cos 2 ψ cos ( j − 1 ) ψ − cos ψ cos j ψ ] / sin ψ = [ cos ( j − 3 ) ψ − cos ( j − 1 ) ψ ] / ( 2 sin ψ ) . Hence
0 = y m < y m − 1 < ⋯ < y 3 < y 2 < y 0 < y 1 , 0=y_m<y_{m-1}<\cdots<y_3<y_2<y_0<y_1, 0 = y m < y m − 1 < ⋯ < y 3 < y 2 < y 0 < y 1 ,
with y 1 − y 0 = sin ψ y_1-y_0=\sin\psi y 1 − y 0 = sin ψ and y 0 − y 2 = 2 sin ψ cos 2 ψ y_0-y_2=2\sin\psi\cos2\psi y 0 − y 2 = 2 sin ψ cos 2 ψ .
Strips. For 3 ≤ j ≤ m 3\le j\le m 3 ≤ j ≤ m let S j S_j S j be the strip of heights ( y j , y j − 1 ) (y_j,y_{j-1}) ( y j , y j − 1 ) with right side Ξ j \Xi_j Ξ j (rising) and left side its mirror image Ξ 1 − j \Xi_{1-j} Ξ 1 − j ; the vertices on the arc between their upper ends are the V i V_i V i with ∣ i ∣ ≤ j − 1 |i| \le j - 1 ∣ i ∣ ≤ j − 1 , of height at least y j − 1 y_{j-1} y j − 1 by the order above. The right tip is the strip of heights ( y 0 , y 1 ) (y_0,y_1) ( y 0 , y 1 ) with right side Ξ 2 \Xi_2 Ξ 2 and left side Ξ 1 \Xi_1 Ξ 1 , and the left tip its mirror image, with right side Ξ 0 \Xi_0 Ξ 0 and left side Ξ − 1 \Xi_{-1} Ξ − 1 . Reflecting in the x x x -axis, S j ′ S'_j S j ′ is the strip of heights ( − y j − 1 , − y j ) (-y_{j-1},-y_j) ( − y j − 1 , − y j ) with right side Ξ 2 m + 1 − j \Xi_{2m+1-j} Ξ 2 m + 1 − j and left side Ξ 2 m + j \Xi_{2m+j} Ξ 2 m + j . S j S_j S j and S j ′ S'_j S j ′ have height w j w_j w j . Strips of disjoint height ranges, or on the two sides of the y y y -axis, are disjoint.
The cylinder Z ′ Z' Z ′ (for m ≥ 5 m \ge5 m ≥ 5 ) is made of the right tip, the top part of S 5 ′ S'_5 S 5 ′ of height sin ψ \sin\psi sin ψ , the left tip, and S 3 ′ S'_3 S 3 ′ . The side Ξ 2 \Xi_2 Ξ 2 of the right tip is glued to Ξ 2 m + 5 \Xi_{2m+5} Ξ 2 m + 5 , the left side of S 5 ′ S'_5 S 5 ′ , with V 1 ↔ V 2 m + 5 V_1 \leftrightarrow V_{2m+5} V 1 ↔ V 2 m + 5 , so the apex V 1 V_1 V 1 goes to the upper end of Ξ 2 m + 5 \Xi_{2m+5} Ξ 2 m + 5 , at the top height − y 5 -y_5 − y 5 of S 5 ′ S'_5 S 5 ′ ; the tip has height sin ψ < w 5 = sin 3 ψ \sin\psi< w_5 = \sin3\psi sin ψ < w 5 = sin 3 ψ , so it is glued to the top part of S 5 ′ S'_5 S 5 ′ . The right side Ξ 2 m − 4 \Xi_{2m-4} Ξ 2 m − 4 of S 5 ′ S'_5 S 5 ′ is glued to Ξ − 1 \Xi_{-1} Ξ − 1 , the left side of the left tip, with V 2 m − 5 ↔ V − 1 V_{2m-5} \leftrightarrow V_{-1} V 2 m − 5 ↔ V − 1 . The side Ξ 0 \Xi_0 Ξ 0 of the left tip is glued to Ξ 2 m + 3 \Xi_{2m+3} Ξ 2 m + 3 , the left side of S 3 ′ S'_3 S 3 ′ , apex to top, and w 3 = sin ψ w_3 = \sin\psi w 3 = sin ψ . Finally Ξ 2 m − 2 \Xi_{2m-2} Ξ 2 m − 2 , the right side of S 3 ′ S'_3 S 3 ′ , is glued to Ξ 1 \Xi_1 Ξ 1 , the left side of the right tip; both have height sin ψ \sin\psi sin ψ . By Lemma 5.14, Z ′ Z' Z ′ is an annulus of height sin ψ \sin\psi sin ψ . On its upper boundary the tips have width zero and the chords of S 5 ′ S'_5 S 5 ′ and S 3 ′ S'_3 S 3 ′ join the mirror images of V ± 5 V_{\pm5} V ± 5 and V ± 3 V_{\pm3} V ± 3 , so its circumference is 2 x ( V 5 ) + 2 x ( V 3 ) = 2 cot ψ ( sin 5 ψ + sin 3 ψ ) = 16 cos 3 ψ cos 2 ψ 2x(V_5) + 2x(V_3) = 2\cot\psi(\sin5\psi+ \sin3\psi) = 16\cos^3\psi\cos2\psi 2 x ( V 5 ) + 2 x ( V 3 ) = 2 cot ψ ( sin 5 ψ + sin 3 ψ ) = 16 cos 3 ψ cos 2 ψ . Its upper boundary contains V 1 V_1 V 1 and its lower boundary the notch V 0 V_0 V 0 .
The cylinder Z Z Z . Z = S m ∪ S m − 2 Z = S_m \cup S_{m-2} Z = S m ∪ S m − 2 : the side Ξ m \Xi_m Ξ m is glued to Ξ 3 m + 3 = Ξ 1 − ( m − 2 ) \Xi_{3m+3} = \Xi_{1-(m-2)} Ξ 3 m + 3 = Ξ 1 − ( m − 2 ) , the left side of S m − 2 S_{m-2} S m − 2 , with V m ↔ V 2 − m V_m \leftrightarrow V_{2-m} V m ↔ V 2 − m , bottom to bottom; and Ξ m − 2 \Xi_{m-2} Ξ m − 2 is glued to Ξ 3 m + 1 = Ξ 1 − m \Xi_{3m+1} = \Xi_{1-m} Ξ 3 m + 1 = Ξ 1 − m , the left side of S m S_m S m . Both strips have height cos ψ \cos\psi cos ψ (w m = sin ( m + 1 ) ψ w_m = \sin(m+1)\psi w m = sin ( m + 1 ) ψ , w m − 2 = sin ( m − 1 ) ψ w_{m-2} = \sin(m-1)\psi w m − 2 = sin ( m − 1 ) ψ ), and on the lower boundary the circumference is 2 x ( V m ) + 2 x ( V m − 2 ) = 2 ∣ A C ∣ ( 1 + cos 2 ψ ) = 4 cos 2 ψ cos 2 ψ / sin ψ 2x(V_m) + 2x(V_{m-2}) = 2|AC|(1+\cos2\psi) = 4\cos^2\psi\cos2\psi/\sin\psi 2 x ( V m ) + 2 x ( V m − 2 ) = 2∣ A C ∣ ( 1 + cos 2 ψ ) = 4 cos 2 ψ cos 2 ψ / sin ψ . So μ ( Z ) = sin ψ / ( 4 cos ψ cos 2 ψ ) \mu(Z) = \sin\psi/(4\cos\psi\cos2\psi) μ ( Z ) = sin ψ / ( 4 cos ψ cos 2 ψ ) ; this needs m − 2 ≥ 3 m-2 \ge3 m − 2 ≥ 3 . The two boundary components of Z Z Z contain V m V_m V m and V m − 1 V_{m-1} V m − 1 .
Maximality. The C k C_k C k form one point of cone angle 2 π ( m + 1 ) 2\pi(m+1) 2 π ( m + 1 ) , since gcd ( m + 1 , 2 m ) = 1 \gcd(m+1,2m)=1 g cd( m + 1 , 2 m ) = 1 , and the B k B_k B k form gcd ( m − 2 , 4 ) \gcd(m-2,4) g cd( m − 2 , 4 ) points of cone angle 2 π ( m − 2 ) / gcd ( m − 2 , 4 ) ≥ 4 π 2\pi(m-2)/\gcd(m-2,4) \ge4\pi 2 π ( m − 2 ) / g cd( m − 2 , 4 ) ≥ 4 π , since m ≥ 8 m \ge8 m ≥ 8 . So every vertex of the star is in Σ \Sigma Σ , and by Lemma 5.14 Z Z Z and Z ′ Z' Z ′ are maximal cylinders, both horizontal.
Irrationality. μ ( Z ) / μ ( Z ′ ) = 4 cos 2 ψ = 2 + 2 cos ( 2 π / 2 m ) \mu(Z)/\mu(Z') = 4\cos^2\psi= 2 + 2\cos(2\pi/2m) μ ( Z ) / μ ( Z ′ ) = 4 cos 2 ψ = 2 + 2 cos ( 2 π /2 m ) , and cos ( 2 π / 2 m ) \cos(2\pi/2m) cos ( 2 π /2 m ) has degree φ ( 2 m ) / 2 ≥ 2 \varphi(2m)/2 \ge2 φ ( 2 m ) /2 ≥ 2 for m ≥ 4 m \ge4 m ≥ 4 . By Lemma 2.3(i), Δ \Delta Δ is not a lattice triangle.
The direction. V m V_m V m is a copy of C C C , and the horizontal direction is that of the side A C AC A C . □
At m = 6 m=6 m = 6 , that is at ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 , the annuli Z Z Z and Z ′ Z' Z ′ exist with the moduli of the theorem, but the copies of B B B are regular points and Z Z Z and Z ′ Z' Z ′ are not maximal; the maximal cylinders in that direction are four, with moduli 1 / ( 5 + 3 3 ) 1/(5+3\sqrt{3}) 1/ ( 5 + 3 3 ) and 1 / ( 10 + 6 3 ) 1/(10+6\sqrt{3}) 1/ ( 10 + 6 3 ) in pairs [11 (5)] .
Theorem 5.16. For every q ≥ 3 q \ge3 q ≥ 3 the triangle ( 1 , q , 3 q + 1 ) / ( 4 q + 2 ) (1,q,3q+1)/(4q+2) ( 1 , q , 3 q + 1 ) / ( 4 q + 2 ) is not a lattice triangle. With ε = sin ( π / ( 4 q + 2 ) ) \varepsilon= \sin(\pi/(4q+2)) ε = sin ( π / ( 4 q + 2 )) , in the direction of the side joining the vertices of angles π / ( 4 q + 2 ) \pi/(4q+2) π / ( 4 q + 2 ) and q π / ( 4 q + 2 ) q\pi/(4q+2) q π / ( 4 q + 2 ) the unfolding has two maximal cylinders U U U and V V V of moduli
μ ( U ) = ε 2 cos ψ ( 1 + 2 ε ) , μ ( V ) = ε 4 cos ψ ( 1 + ε ) , ψ = π 4 q + 2 , \mu(U) = \frac{\varepsilon^2}{\cos\psi(1+2\varepsilon)}, \qquad\mu(V) = \frac{\varepsilon}{4\cos\psi(1+\varepsilon)}, \qquad\psi= \frac{\pi}{4q+2}, μ ( U ) = cos ψ ( 1 + 2 ε ) ε 2 , μ ( V ) = 4 cos ψ ( 1 + ε ) ε , ψ = 4 q + 2 π ,
whose ratio μ ( V ) / μ ( U ) = ( 1 + 2 ε ) / ( 4 ε ( 1 + ε ) ) \mu(V)/\mu(U) = (1+2\varepsilon)/(4\varepsilon(1+\varepsilon)) μ ( V ) / μ ( U ) = ( 1 + 2 ε ) / ( 4 ε ( 1 + ε )) is irrational.
This is family (4) of Theorem 2.5 ; its obtuse angle exceeds 2 π / 3 2\pi/3 2 π /3 for q ≥ 2 q \ge2 q ≥ 2 . For q q q odd it is also a theorem of Ward [31 Theorem B] . For q = 1 q=1 q = 1 and q = 2 q=2 q = 2 the triangle is ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 , respectively ( 1 , 2 , 7 ) / 10 (1,2,7)/10 ( 1 , 2 , 7 ) /10 , both lattice triangles.
Proof. Here n = 4 q + 2 n = 4q+2 n = 4 q + 2 , ε = sin ψ \varepsilon= \sin\psi ε = sin ψ , Ψ : = q ψ = π / 4 − ψ / 2 \Psi:= q\psi= \pi/4 - \psi/2 Ψ := q ψ = π /4 − ψ /2 and r ψ = 3 π / 4 − ψ / 2 r\psi= 3\pi/4 - \psi/2 r ψ = 3 π /4 − ψ /2 ; so ∣ A B ∣ = cos Ψ / ε |AB| = \cos\Psi/\varepsilon ∣ A B ∣ = cos Ψ/ ε , ∣ A C ∣ = sin Ψ / ε |AC| = \sin\Psi/\varepsilon ∣ A C ∣ = sin Ψ/ ε and tan Ψ = cos ψ / ( 1 + ε ) \tan\Psi= \cos\psi/(1+\varepsilon) tan Ψ = cos ψ / ( 1 + ε ) . Take ϕ = 0 \phi= 0 ϕ = 0 : B k = ∣ A B ∣ e 2 i k ψ B_k = |AB|e^{2ik\psi} B k = ∣ A B ∣ e 2 ik ψ and C k = ∣ A C ∣ e i ( 2 k + 1 ) ψ C_k = |AC|e^{i(2k+1)\psi} C k = ∣ A C ∣ e i ( 2 k + 1 ) ψ . Since ( 2 q + 1 ) ψ = π / 2 (2q+1)\psi= \pi/2 ( 2 q + 1 ) ψ = π /2 , the inner vertex C q C_q C q lies on the positive y y y -axis, B 0 B_0 B 0 on the positive x x x -axis and B 2 q + 1 B_{2q+1} B 2 q + 1 on the negative x x x -axis. The side b k b_k b k rises by sin ( q − 2 k ) ψ \sin(q-2k)\psi sin ( q − 2 k ) ψ and c k c_k c k by sin ( 2 k + q + 2 ) ψ \sin(2k+q+2)\psi sin ( 2 k + q + 2 ) ψ when traversed counterclockwise. By Lemma 5.12 , c k c_k c k is glued to b k + q + 1 b_{k+q+1} b k + q + 1 , with C k ↔ C k + q + 1 C_k \leftrightarrow C_{k+q+1} C k ↔ C k + q + 1 and B k + 1 ↔ B k + q + 1 B_{k+1} \leftrightarrow B_{k+q+1} B k + 1 ↔ B k + q + 1 . The mirror x ↦ − x x \mapsto-x x ↦ − x maps B k ↦ B 2 q + 1 − k B_k \mapsto B_{2q+1-k} B k ↦ B 2 q + 1 − k , C k ↦ C 2 q − k C_k \mapsto C_{2q-k} C k ↦ C 2 q − k , and y ↦ − y y \mapsto-y y ↦ − y maps B k ↦ B − k B_k \mapsto B_{-k} B k ↦ B − k , C k ↦ C − k − 1 C_k \mapsto C_{-k-1} C k ↦ C − k − 1 . The heights are y ( B k ) = ∣ A B ∣ sin 2 k ψ y(B_k)=|AB|\sin2k\psi y ( B k ) = ∣ A B ∣ sin 2 k ψ and y ( C k ) = ∣ A C ∣ sin ( 2 k + 1 ) ψ y(C_k)=|AC|\sin(2k+1)\psi y ( C k ) = ∣ A C ∣ sin ( 2 k + 1 ) ψ .
Heights used. (i) y ( C 0 ) = sin Ψ y(C_0)=\sin\Psi y ( C 0 ) = sin Ψ and y ( B 1 ) − y ( C 0 ) = sin ( q + 2 ) ψ = sin Ψ + 2 ε sin Ψ y(B_1)-y(C_0)=\sin(q+2)\psi=\sin\Psi+2\varepsilon\sin\Psi y ( B 1 ) − y ( C 0 ) = sin ( q + 2 ) ψ = sin Ψ + 2 ε sin Ψ , because sin ( q + 2 ) ψ − sin q ψ = 2 sin ψ cos ( q + 1 ) ψ \sin(q+2)\psi-\sin q\psi=2\sin\psi\cos(q+1)\psi sin ( q + 2 ) ψ − sin q ψ = 2 sin ψ cos ( q + 1 ) ψ and ( q + 1 ) ψ = π / 2 − Ψ (q+1)\psi=\pi/2-\Psi ( q + 1 ) ψ = π /2 − Ψ . (ii) y ( C 1 ) − y ( B 1 ) = [ sin Ψ sin 3 ψ − cos Ψ sin 2 ψ ] / ε = cos Ψ cos ψ ( 1 − 2 ε − 4 ε 2 ) / ( 1 + ε ) y(C_1)-y(B_1)=[\sin\Psi\sin3\psi-\cos\Psi\sin2\psi]/\varepsilon=\cos\Psi\cos\psi(1-2\varepsilon-4\varepsilon^2)/(1+\varepsilon) y ( C 1 ) − y ( B 1 ) = [ sin Ψ sin 3 ψ − cos Ψ sin 2 ψ ] / ε = cos Ψ cos ψ ( 1 − 2 ε − 4 ε 2 ) / ( 1 + ε ) , using sin Ψ ( 1 + ε ) = cos Ψ cos ψ \sin\Psi(1+\varepsilon)=\cos\Psi\cos\psi sin Ψ ( 1 + ε ) = cos Ψ cos ψ . The root of 1 − 2 ε − 4 ε 2 1-2\varepsilon-4\varepsilon^2 1 − 2 ε − 4 ε 2 in ( 0 , 1 ) (0,1) ( 0 , 1 ) is sin ( π / 10 ) \sin(\pi/10) sin ( π /10 ) , so y ( C 1 ) > y ( B 1 ) y(C_1)>y(B_1) y ( C 1 ) > y ( B 1 ) exactly when ψ < π / 10 \psi<\pi/10 ψ < π /10 , that is, q ≥ 3 q\ge3 q ≥ 3 . (iii) For 1 ≤ k ≤ 2 q 1\le k\le2q 1 ≤ k ≤ 2 q , y ( B k ) ≥ y ( B 1 ) y(B_k)\ge y(B_1) y ( B k ) ≥ y ( B 1 ) ; for 1 ≤ k ≤ 2 q − 1 1\le k\le2q-1 1 ≤ k ≤ 2 q − 1 , y ( C k ) ≥ y ( C 1 ) > y ( B 1 ) y(C_k)\ge y(C_1)>y(B_1) y ( C k ) ≥ y ( C 1 ) > y ( B 1 ) by (ii); for 0 ≤ k ≤ 2 q 0\le k\le2q 0 ≤ k ≤ 2 q , y ( C k ) ≥ y ( C 0 ) = sin Ψ y(C_k)\ge y(C_0)=\sin\Psi y ( C k ) ≥ y ( C 0 ) = sin Ψ . (iv) y ( B q ) − y ( C q ) = sin Ψ y(B_q)-y(C_q)=\sin\Psi y ( B q ) − y ( C q ) = sin Ψ and y ( C q ) − y ( C q − 1 ) = ∣ A C ∣ ( 1 − cos 2 ψ ) = 2 ε sin Ψ > 0 y(C_q)-y(C_{q-1})=|AC|(1-\cos2\psi)=2\varepsilon\sin\Psi>0 y ( C q ) − y ( C q − 1 ) = ∣ A C ∣ ( 1 − cos 2 ψ ) = 2 ε sin Ψ > 0 .
Strips (Figure 4 ). M M M : heights ( sin Ψ , y ( B 1 ) ) (\sin\Psi,y(B_1)) ( sin Ψ , y ( B 1 )) , right side c 0 c_0 c 0 , left side b 2 q b_{2q} b 2 q , height sin Ψ + 2 ε sin Ψ \sin\Psi+2\varepsilon\sin\Psi sin Ψ + 2 ε sin Ψ ; the arc from B 1 B_1 B 1 to B 2 q B_{2q} B 2 q passes through B 1 , C 1 , … , C 2 q − 1 , B 2 q B_1,C_1,\ldots,C_{2q-1},B_{2q} B 1 , C 1 , … , C 2 q − 1 , B 2 q , all of height at least y ( B 1 ) y(B_1) y ( B 1 ) by (iii). M 0 M_0 M 0 : heights ( 0 , sin Ψ ) (0,\sin\Psi) ( 0 , sin Ψ ) , right side b 0 b_0 b 0 , left side c 2 q c_{2q} c 2 q ; the vertices on the arc from C 0 C_0 C 0 to C 2 q C_{2q} C 2 q have height at least sin Ψ \sin\Psi sin Ψ by (iii). Its mirror image M ′ M' M ′ in the x x x -axis has heights ( − sin Ψ , 0 ) (-\sin\Psi,0) ( − sin Ψ , 0 ) , left side b 2 q + 1 b_{2q+1} b 2 q + 1 and right side c − 1 c_{-1} c − 1 . The slab: heights ( y ( C q − 1 ) , y ( C q ) ) (y(C_{q-1}),y(C_q)) ( y ( C q − 1 ) , y ( C q )) , right side c q − 1 c_{q-1} c q − 1 , left side b q + 1 b_{q+1} b q + 1 ; the arc from B q B_q B q to B q + 1 B_{q+1} B q + 1 is B q , C q , B q + 1 B_q,C_q,B_{q+1} B q , C q , B q + 1 , of heights at least y ( C q ) y(C_q) y ( C q ) by (iv). The tips : heights ( y ( C q ) , y ( B q ) ) (y(C_q),y(B_q)) ( y ( C q ) , y ( B q )) ; the right tip has right side c q − 1 c_{q-1} c q − 1 and left side b q b_q b q , the left tip right side c q c_q c q and left side b q + 1 b_{q+1} b q + 1 . These strips are pairwise disjoint: the slab and the two tips lie at heights above y ( C q − 1 ) ≥ y ( C 1 ) > y ( B 1 ) y(C_{q-1}) \ge y(C_1) > y(B_1) y ( C q − 1 ) ≥ y ( C 1 ) > y ( B 1 ) , the top of M M M ; the tips lie in x > 0 x>0 x > 0 and x < 0 x<0 x < 0 ; M ′ M' M ′ lies below the x x x -axis; and U U U and V V V below use complementary parts of M M M .
Figure 4. The star of Lemma 5.12 for ( 1 , 3 , 10 ) / 14 (1,3,10)/14 ( 1 , 3 , 10 ) /14 (q = 3 q=3 q = 3 in Theorem 5.16), B 0 B_0 B 0 on the positive x x x -axis; b k b_k b k is glued to c k − 4 c_{k-4} c k − 4 . Shaded: the cylinders U U U (dark), of height 2 ε sin Ψ 2\varepsilon\sin\Psi 2 ε sin Ψ , and V V V (light), of height sin Ψ \sin\Psi sin Ψ ; dashed: their core curves.
The cylinder U U U is the lower part of M M M , of heights ( sin Ψ , sin Ψ + 2 ε sin Ψ ) (\sin\Psi,\sin\Psi+2\varepsilon\sin\Psi) ( sin Ψ , sin Ψ + 2 ε sin Ψ ) , together with the slab. The gluing c 0 ↔ b q + 1 c_0\leftrightarrow b_{q+1} c 0 ↔ b q + 1 takes C 0 C_0 C 0 to C q + 1 C_{q+1} C q + 1 , and y ( C q + 1 ) = y ( C q − 1 ) y(C_{q+1})=y(C_{q-1}) y ( C q + 1 ) = y ( C q − 1 ) by the mirror, so the bottom of M M M goes to the bottom of the slab and the lower part of M M M onto the slab (iv). The gluing c q − 1 ↔ b 2 q c_{q-1}\leftrightarrow b_{2q} c q − 1 ↔ b 2 q takes C q − 1 C_{q-1} C q − 1 to C 2 q C_{2q} C 2 q , of height y ( C 0 ) y(C_0) y ( C 0 ) : bottom to bottom, back into M M M through its left side. By Lemma 5.14 , U U U is an annulus of height 2 ε sin Ψ 2\varepsilon\sin\Psi 2 ε sin Ψ ; on its lower boundary its circumference is 2 x ( C 0 ) + 2 x ( C q − 1 ) = 2 ∣ A C ∣ ( cos ψ + sin 2 ψ ) = 2 sin Ψ cos ψ ( 1 + 2 ε ) / ε 2x(C_0)+2x(C_{q-1})=2|AC|(\cos\psi+\sin2\psi)=2\sin\Psi\cos\psi(1+2\varepsilon)/\varepsilon 2 x ( C 0 ) + 2 x ( C q − 1 ) = 2∣ A C ∣ ( cos ψ + sin 2 ψ ) = 2 sin Ψ cos ψ ( 1 + 2 ε ) / ε , as ( 2 q − 1 ) ψ = π / 2 − 2 ψ (2q-1)\psi=\pi/2-2\psi ( 2 q − 1 ) ψ = π /2 − 2 ψ . So μ ( U ) = ε 2 / ( cos ψ ( 1 + 2 ε ) ) \mu(U)=\varepsilon^2/(\cos\psi(1+2\varepsilon)) μ ( U ) = ε 2 / ( cos ψ ( 1 + 2 ε )) . The lower boundary passes through C 0 C_0 C 0 , the upper boundary through C q C_q C q .
The cylinder V V V is the right tip, the upper part of M M M (heights ( y ( B 1 ) − sin Ψ , y ( B 1 ) ) (y(B_1)-\sin\Psi,y(B_1)) ( y ( B 1 ) − sin Ψ , y ( B 1 )) ), the left tip and M ′ M' M ′ . The gluing c q − 1 ↔ b 2 q c_{q-1}\leftrightarrow b_{2q} c q − 1 ↔ b 2 q takes B q B_q B q to B 2 q B_{2q} B 2 q , of height y ( B 1 ) y(B_1) y ( B 1 ) : the apex of the right tip goes to the top of M M M , and the tip (of height sin Ψ \sin\Psi sin Ψ ) onto the upper part of M M M . The gluing c 0 ↔ b q + 1 c_0\leftrightarrow b_{q+1} c 0 ↔ b q + 1 takes B 1 B_1 B 1 to B q + 1 B_{q+1} B q + 1 , of height y ( B q ) y(B_q) y ( B q ) : the upper part of M M M goes onto the left tip. The gluing c q ↔ b 2 q + 1 c_q\leftrightarrow b_{2q+1} c q ↔ b 2 q + 1 takes B q + 1 B_{q+1} B q + 1 to B 2 q + 1 B_{2q+1} B 2 q + 1 (height 0 0 0 , the top of M ′ M' M ′ ) and C q C_q C q to C 2 q + 1 C_{2q+1} C 2 q + 1 (height − sin Ψ -\sin\Psi − sin Ψ ): the left tip goes onto M ′ M' M ′ . The gluing c − 1 ↔ b q c_{-1}\leftrightarrow b_q c − 1 ↔ b q takes B 0 B_0 B 0 to B q B_q B q : M ′ M' M ′ goes onto the right tip. By Lemma 5.14 , V V V has height sin Ψ \sin\Psi sin Ψ ; on its upper boundary the tips have width zero and the chords are B 2 q B 1 B_{2q}B_1 B 2 q B 1 and B 2 q + 1 B 0 B_{2q+1}B_0 B 2 q + 1 B 0 , so its circumference is 2 x ( B 1 ) + 2 x ( B 0 ) = 2 ∣ A B ∣ ( 1 + cos 2 ψ ) = 4 cos Ψ cos 2 ψ / ε 2x(B_1) + 2x(B_0) = 2|AB|(1 + \cos2\psi) = 4\cos\Psi\cos^2\psi/\varepsilon 2 x ( B 1 ) + 2 x ( B 0 ) = 2∣ A B ∣ ( 1 + cos 2 ψ ) = 4 cos Ψ cos 2 ψ / ε , and μ ( V ) = ε / ( 4 cos ψ ( 1 + ε ) ) \mu(V) = \varepsilon/(4\cos\psi(1+\varepsilon)) μ ( V ) = ε / ( 4 cos ψ ( 1 + ε )) . The upper boundary passes through B q B_q B q , the lower boundary through C q C_q C q .
Maximality. For q q q even the C k C_k C k form one point of cone angle 2 π ( 3 q + 1 ) 2\pi(3q+1) 2 π ( 3 q + 1 ) and the B k B_k B k two points of cone angle π q \pi q π q ; for q q q odd the B k B_k B k form one point of cone angle 2 π q 2\pi q 2 π q and the C k C_k C k two points of cone angle π ( 3 q + 1 ) \pi(3q+1) π ( 3 q + 1 ) . For q ≥ 3 q \ge3 q ≥ 3 all exceed 2 π 2\pi 2 π , so U U U and V V V are maximal cylinders, both horizontal; the horizontal direction is that of the side A B AB A B .
Irrationality. Let ϱ = μ ( V ) / μ ( U ) = ( 1 + 2 ε ) / ( 4 ε ( 1 + ε ) ) \varrho= \mu(V)/\mu(U) = (1+2\varepsilon)/(4\varepsilon(1+\varepsilon)) ϱ = μ ( V ) / μ ( U ) = ( 1 + 2 ε ) / ( 4 ε ( 1 + ε )) . If ϱ ∈ Q \varrho\in\mathbb{Q} ϱ ∈ Q , then ε \varepsilon ε is a root of 4 ϱ x 2 + ( 4 ϱ − 2 ) x − 1 4\varrho x^2 + (4\varrho- 2)x - 1 4 ϱ x 2 + ( 4 ϱ − 2 ) x − 1 , a non-zero rational polynomial of degree 2. But ε = cos ( q π / ( 2 q + 1 ) ) \varepsilon= \cos(q\pi/(2q+1)) ε = cos ( q π / ( 2 q + 1 )) is, up to sign, cos ( 2 π j / ( 2 q + 1 ) ) \cos(2\pi j/(2q+1)) cos ( 2 π j / ( 2 q + 1 )) with gcd ( j , 2 q + 1 ) = 1 \gcd(j,2q+1)=1 g cd( j , 2 q + 1 ) = 1 (j = q / 2 j=q/2 j = q /2 for q q q even, and ε = − cos ( 2 π ( q + 1 ) / 2 / ( 2 q + 1 ) ) \varepsilon=-\cos(2\pi(q+1)/2/(2q+1)) ε = − cos ( 2 π ( q + 1 ) /2/ ( 2 q + 1 )) for q q q odd), so its degree is φ ( 2 q + 1 ) / 2 ≥ 3 \varphi(2q+1)/2 \ge3 φ ( 2 q + 1 ) /2 ≥ 3 for q ≥ 3 q \ge3 q ≥ 3 . So ϱ ∉ Q \varrho\notin\mathbb{Q} ϱ ∈ / Q , and Lemma 2.3(i) applies. □
Single triangles
The list F \mathcal{F} F of Table 2 consists of the triangles that no general theorem of this paper reaches: ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 of A \mathcal{A} A , to which Theorem 1.2 does not apply (Theorem 4.18 (c)); the members of E \mathcal{E} E to which Theorem 1.2 does not apply, among them ( 3 , 4 , 13 ) / 20 (3,4,13)/20 ( 3 , 4 , 13 ) /20 , the case q = 3 q=3 q = 3 of Theorem 5.11 ; and the triangles ( 1 , 4 , 11 ) / 16 (1,4,11)/16 ( 1 , 4 , 11 ) /16 , ( 2 , 3 , 17 ) / 22 (2,3,17)/22 ( 2 , 3 , 17 ) /22 , ( 3 , 8 , 29 ) / 40 (3,8,29)/40 ( 3 , 8 , 29 ) /40 and ( 2 , 11 , 29 ) / 42 (2,11,29)/42 ( 2 , 11 , 29 ) /42 of item (7) of Theorem 2.5 , to which it does not apply either. All of them have denominator at most 48.
Table 2. The triangles of Theorem 6.1, those that no general theorem of the paper reaches. Column 3: the list or family of the triangle (7): item (7) of Theorem 2.5. Column 4: the direction d d d , as the angle θ ∈ [ 0 , π ) \theta\in[0,\pi) θ ∈ [ 0 , π ) from A B AB A B to the line R d \mathbb{R}d R d , counterclockwise, in the coordinates of Appendix A.§ 1 , with A = 0 A=0 A = 0 and B B B on the positive real axis; in row 5, where θ / π \theta/\pi θ / π is irrational, d d d itself, as a polynomial in z = e i π / n z=e^{i\pi/n} z = e iπ / n . Column 5: the minimal polynomial of μ ( Z 1 ) / μ ( Z 2 ) \mu(Z_1)/\mu(Z_2) μ ( Z 1 ) / μ ( Z 2 ) over Q \mathbb{Q} Q , Z 1 Z_1 Z 1 and Z 2 Z_2 Z 2 being the first two cylinders of the certificate.
Table 2.
Theorem 6.1. No triangle of Table 2 is a lattice triangle. For each of them, in the direction d d d given in the table, the unfolding has two maximal cylinders Z 1 Z_1 Z 1 , Z 2 Z_2 Z 2 such that μ ( Z 1 ) / μ ( Z 2 ) \mu(Z_1)/\mu(Z_2) μ ( Z 1 ) / μ ( Z 2 ) has the minimal polynomial given in the table, of degree between 2 and 8. The direction d d d , as an element of Z [ z ] \mathbb{Z}[z] Z [ z ] with z = e i π / n z=e^{i\pi/n} z = e iπ / n , is in the deposit [13 ] .
Proof. The cylinders Z 1 Z_1 Z 1 and Z 2 Z_2 Z 2 , their maximality, and the minimal polynomial of the ratio of their moduli are verified for each triangle by an exact computation, described in Appendix A.§ 1 . A polynomial of degree at least 2 that is irreducible over Q \mathbb{Q} Q has no rational root, so the ratio is irrational, and Lemma 2.3(i) applies. □
The certificates for the members of E \mathcal{E} E in Table 2 with even denominator are used in the window case of the proof of Theorem 1.1 , those for the triangles of item (7) in Theorem 1.3 , and those for ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 and ( 3 , 7 , 11 ) / 21 (3,7,11)/21 ( 3 , 7 , 11 ) /21 in Theorem 1.4 . Figure 5 shows the certificate of ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 .
Figure 5. The unfolding of ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 , the star of Lemma 5.12 turned so that the direction d d d of Table 2 , row 1, is horizontal. Sides with the same numeral are glued. Filled dots: the copies of the vertex C C C , which form the one point of Σ \Sigma Σ , of cone angle 10 π 10\pi 10 π . Hollow dots: the copies of B B B and the point O O O over A A A , which form four points of cone angle 2 π 2\pi 2 π . Shaded: the cylinders Z 1 Z_1 Z 1 (dark) and Z 2 Z_2 Z 2 (light) of the certificate.
The unfolding of ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 has genus 3, and Σ \Sigma Σ is the one point over C C C . In its certificate, with z = e i π / 9 z=e^{i\pi/9} z = e iπ /9 , d = − 1 + z + 2 z 3 − 2 z 4 d=-1+z+2z^3-2z^4 d = − 1 + z + 2 z 3 − 2 z 4 is a positive multiple of 1 + z 1+z 1 + z , at the angle π / 18 \pi/18 π /18 to A B AB A B . The cylinder Z 1 Z_1 Z 1 consists of three strips, each with an itinerary of sixteen steps, which meet along a leaf through a point over B B B and along a leaf through the point over A A A . The cylinder Z 2 Z_2 Z 2 consists of two strips, each with an itinerary of eight steps, which meet along a leaf through a point over B B B . The outer boundary leaves of both pass through the point over C C C . By Lemma A.1,
μ ( Z 1 ) = sin π 9 + sin 2 π 9 − sin π 3 , μ ( Z 2 ) = sin 2 π 9 − sin π 9 , \mu(Z_1)=\sin\frac{\pi}{9}+\sin\frac{2\pi}{9}-\sin\frac{\pi}{3},\qquad
\mu(Z_2)=\sin\frac{2\pi}{9}-\sin\frac{\pi}{9}, μ ( Z 1 ) = sin 9 π + sin 9 2 π − sin 3 π , μ ( Z 2 ) = sin 9 2 π − sin 9 π ,
and μ ( Z 1 ) / μ ( Z 2 ) ≈ 0.3949 \mu(Z_1)/\mu(Z_2)\approx0.3949 μ ( Z 1 ) / μ ( Z 2 ) ≈ 0.3949 has the minimal polynomial x 3 − x + 1 3 x^3-x+\frac{1}{3} x 3 − x + 3 1 , of degree 3, as in row 1 of Table 2 .
That the triangles of Table 2 are not lattice triangles also follows from [6 Theorem 4.1] for the nineteen of them other than ( 1 , 4 , 11 ) / 16 (1,4,11)/16 ( 1 , 4 , 11 ) /16 and ( 3 , 4 , 13 ) / 20 (3,4,13)/20 ( 3 , 4 , 13 ) /20 . These two Delcroix and Rüth list as orbit closures of rank two, and for ( 3 , 4 , 13 ) / 20 (3,4,13)/20 ( 3 , 4 , 13 ) /20 they prove that the orbit closure has complex dimension at least 4 [6 Theorem 1.2] , so that this triangle is not a lattice triangle. That the triangles of item (7) of Theorem 2.5 are not lattice triangles was shown with the program of Rüth, Delecroix and Eskin [17 pp. 7120 and 7125] .
The classification
In this section we prove Theorem 1.4 , Corollary 1.5, Theorem 1.3 and Theorem 1.1 , in this order.
The members of E \mathcal{E} E and the triples of item (7) of Theorem 2.5 that are not in Table 2 are reached by Theorem 1.2 : ( 2 , 3 , 6 ) / 11 (2,3,6)/11 ( 2 , 3 , 6 ) /11 , ( 2 , 5 , 8 ) / 15 (2,5,8)/15 ( 2 , 5 , 8 ) /15 , ( 3 , 4 , 8 ) / 15 (3,4,8)/15 ( 3 , 4 , 8 ) /15 and ( 2 , 7 , 12 ) / 21 (2,7,12)/21 ( 2 , 7 , 12 ) /21 by part (a), and the triangles of the following lemma by part (b).
Lemma 7.1. Theorem 1.2(b) applies to each of the thirteen triangles below at the vertex x x x with the unit b b b , so none of them is a lattice triangle. Three of them, ( 1 , 3 , 16 ) / 20 (1,3,16)/20 ( 1 , 3 , 16 ) /20 , ( 1 , 4 , 21 ) / 26 (1,4,21)/26 ( 1 , 4 , 21 ) /26 and ( 1 , 8 , 19 ) / 28 (1,8,19)/28 ( 1 , 8 , 19 ) /28 , are triangles of item (7) of Theorem 2.5, and the other ten are members of E \mathcal{E} E with even denominator. Here g = gcd ( x , n ) g=\gcd(x,n) g = g cd( x , n ) , and the last column gives [ b p ] n [bp]_n [ b p ] n , [ b q ] n [bq]_n [ b q ] n , [ b r ] n [br]_n [ b r ] n .
Table 7.1.
Proof. In each row 1 < g < x 1<g<x 1 < g < x , so g ∤ x g\nmid x g ∤ x . Also gcd ( b , n ) = 1 \gcd(b,n)=1 g cd( b , n ) = 1 , and b − 2 b-2 b − 2 is n / g n/g n / g , except for ( 5 , 8 , 15 ) / 28 (5,8,15)/28 ( 5 , 8 , 15 ) /28 and ( 3 , 16 , 21 ) / 40 (3,16,21)/40 ( 3 , 16 , 21 ) /40 , where it is 3 n / g 3n/g 3 n / g and 5 n / g 5n/g 5 n / g ; so b ≡ 2 ( m o d n / g ) b\equiv2\pmod{n/g} b ≡ 2 ( mod n / g ) , and 0 < b − 2 < n 0 < b - 2 < n 0 < b − 2 < n gives b ≠ 2 ( m o d n ) b \ne2 \pmod{n} b = 2 ( mod n ) . The three residues add up to n n n . These are the hypotheses of Theorem 1.2 (b). □
Proof of Theorem 1.4. Let n n n be odd and Δ = ( p , q , r ) / n \Delta= (p,q,r)/n Δ = ( p , q , r ) / n obtuse. If p ∤ n p \nmid n p ∤ n or q ∤ n q \nmid n q ∤ n , then Δ \Delta Δ is not a lattice triangle by Theorem 1.2 (a). Otherwise p p p and q q q divide n n n . By Theorem 4.2 , Δ \Delta Δ has a witness, and so is not a lattice triangle (Proposition 2.4 ), unless it is ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n , ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 , ( 1 , 5 , 9 ) / 15 (1,5,9)/15 ( 1 , 5 , 9 ) /15 or ( 3 , 7 , 11 ) / 21 (3,7,11)/21 ( 3 , 7 , 11 ) /21 . The first is a lattice triangle for every n ≥ 5 n \ge5 n ≥ 5 by Veech’s theorem. ( 1 , 5 , 9 ) / 15 (1,5,9)/15 ( 1 , 5 , 9 ) /15 is the member ( s , d ) = ( 3 , 5 ) (s,d) = (3,5) ( s , d ) = ( 3 , 5 ) of A \mathcal{A} A , and Theorem 1.2 (b) applies to it at the vertex 9 9 9 with b = 7 b = 7 b = 7 (proof of Theorem 4.18 ); ( 1 , 3 , 5 ) / 9 (1,3,5)/9 ( 1 , 3 , 5 ) /9 and ( 3 , 7 , 11 ) / 21 (3,7,11)/21 ( 3 , 7 , 11 ) /21 are not lattice triangles by Theorem 6.1 . So the obtuse lattice triangles of odd denominator n n n are exactly the triangles ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n with n ≥ 5 n \ge5 n ≥ 5 . □
Proof of Corollary 1.5. There is no obtuse triangle of denominator 2 2 2 or 3 3 3 . Let n ≥ 5 n \ge5 n ≥ 5 be prime and Δ = ( p , q , r ) / n \Delta= (p,q,r)/n Δ = ( p , q , r ) / n obtuse. Then p , q < n p,q < n p , q < n , so p ∣ n p \mid n p ∣ n and q ∣ n q \mid n q ∣ n only if p = q = 1 p = q = 1 p = q = 1 . Otherwise Theorem 1.2 (a) applies and Δ \Delta Δ is not a lattice triangle; and ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n is a lattice triangle by Veech’s theorem. □
Proof of Theorem 1.3. Let Δ = ( p , q , r ) / n \Delta= (p,q,r)/n Δ = ( p , q , r ) / n with r > 2 n / 3 r > 2n/3 r > 2 n /3 . For n n n odd this is Theorem 1.4 (n − 2 > 2 n / 3 n - 2 > 2n/3 n − 2 > 2 n /3 exactly when n ≥ 7 n \ge7 n ≥ 7 ); let n n n be even. If Δ \Delta Δ has a witness, it is not a lattice triangle (Proposition 2.4 ). Otherwise, by Theorem 2.5 in the direction “no witness implies listed”, Δ \Delta Δ lies in one of the families (1) to (6) or is one of the seven triangles of (7).
Family (1): ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n , with n − 2 > 2 n / 3 n - 2 > 2n/3 n − 2 > 2 n /3 , that is n ≥ 7 n \ge7 n ≥ 7 . It is a lattice triangle by Veech’s theorem.
Family (2): ( 1 , 2 , 2 k − 3 ) / 2 k (1,2,2k-3)/2k ( 1 , 2 , 2 k − 3 ) /2 k , with 2 k − 3 > 4 k / 3 2k - 3 > 4k/3 2 k − 3 > 4 k /3 , that is k ≥ 5 k \ge5 k ≥ 5 . It is a lattice triangle by the theorem of Vorobets and Ward.
Family (3): ( 1 , 4 , r ) / ( r + 5 ) (1,4,r)/(r+5) ( 1 , 4 , r ) / ( r + 5 ) with r ≡ 7 ( m o d 8 ) r \equiv7 \pmod{8} r ≡ 7 ( mod 8 ) . For n = 12 n = 12 n = 12 , r = 7 < 2 n / 3 r = 7 < 2n/3 r = 7 < 2 n /3 ; so n ≥ 20 n \ge20 n ≥ 20 , and the triangle is not a lattice triangle by [17 §10] .
Family (4): ( 1 , q , 3 q + 1 ) / ( 4 q + 2 ) (1,q,3q+1)/(4q+2) ( 1 , q , 3 q + 1 ) / ( 4 q + 2 ) , with 3 q + 1 > 2 ( 4 q + 2 ) / 3 3q + 1 > 2(4q + 2)/3 3 q + 1 > 2 ( 4 q + 2 ) /3 , that is q ≥ 2 q \ge2 q ≥ 2 . For q = 2 q = 2 q = 2 it is ( 1 , 2 , 7 ) / 10 (1,2,7)/10 ( 1 , 2 , 7 ) /10 , of family (2). For q ≥ 3 q \ge3 q ≥ 3 it is not a lattice triangle by Theorem 5.16.
Family (5): ( 2 , q , 3 q + 2 ) / ( 4 q + 4 ) (2,q,3q+2)/(4q+4) ( 2 , q , 3 q + 2 ) / ( 4 q + 4 ) . In lowest terms q q q is odd (for even q q q the triple has the common factor 2 2 2 and the triangle lies in family (4)), and p ≤ q p \le q p ≤ q gives q ≥ 3 q \ge3 q ≥ 3 . It is not a lattice triangle by Theorem 5.9 .
Family (6): ( 4 , q , 3 q + 4 ) / ( 4 q + 8 ) (4,q,3q+4)/(4q+8) ( 4 , q , 3 q + 4 ) / ( 4 q + 8 ) , with q q q odd in lowest terms (for q ≡ 2 ( m o d 4 ) q \equiv2 \pmod{4} q ≡ 2 ( mod 4 ) the triangle lies in family (5), for 4 ∣ q 4 \mid q 4 ∣ q in family (4)), and 3 q + 4 > 2 ( 4 q + 8 ) / 3 3q + 4 > 2(4q + 8)/3 3 q + 4 > 2 ( 4 q + 8 ) /3 , that is q ≥ 5 q \ge5 q ≥ 5 . It is not a lattice triangle by Theorem 5.11 .
The seven triangles of (7) are not lattice triangles: ( 1 , 3 , 16 ) / 20 (1,3,16)/20 ( 1 , 3 , 16 ) /20 , ( 1 , 4 , 21 ) / 26 (1,4,21)/26 ( 1 , 4 , 21 ) /26 and ( 1 , 8 , 19 ) / 28 (1,8,19)/28 ( 1 , 8 , 19 ) /28 by Lemma 7.1, and ( 1 , 4 , 11 ) / 16 (1,4,11)/16 ( 1 , 4 , 11 ) /16 , ( 2 , 3 , 17 ) / 22 (2,3,17)/22 ( 2 , 3 , 17 ) /22 , ( 3 , 8 , 29 ) / 40 (3,8,29)/40 ( 3 , 8 , 29 ) /40 and ( 2 , 11 , 29 ) / 42 (2,11,29)/42 ( 2 , 11 , 29 ) /42 by Theorem 6.1 .
So Δ \Delta Δ is a lattice triangle exactly when it is ( 1 , 1 , n − 2 ) / n (1,1,n-2)/n ( 1 , 1 , n − 2 ) / n with n ≥ 7 n \ge7 n ≥ 7 or ( 1 , 2 , 2 k − 3 ) / 2 k (1,2,2k-3)/2k ( 1 , 2 , 2 k − 3 ) /2 k with k ≥ 5 k \ge5 k ≥ 5 . □
Proof of Theorem 1.1. Let Δ = ( p , q , r ) / n \Delta= (p,q,r)/n Δ = ( p , q , r ) / n be obtuse. If n n n is odd, the statement is Theorem 1.4 , since ( 1 , 2 , 2 k − 3 ) / 2 k (1,2,2k-3)/2k ( 1 , 2 , 2 k − 3 ) /2 k and ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 have even denominator. If n n n is even and r > 2 n / 3 r > 2n/3 r > 2 n /3 , it is Theorem 1.3 , since the triangles of the statement with n n n even and r ≤ 2 n / 3 r \le2n/3 r ≤ 2 n /3 are ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 , ( 1 , 2 , 5 ) / 8 (1,2,5)/8 ( 1 , 2 , 5 ) /8 and ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 . Let n n n be even and n / 2 < r ≤ 2 n / 3 n/2 < r \le2n/3 n /2 < r ≤ 2 n /3 . These three triangles are lattice triangles, by the theorems of Veech, of Vorobets and Ward, and of Hooper. Let Δ \Delta Δ be any other triangle. If Δ \Delta Δ has a witness, it is not a lattice triangle (Proposition 2.4 ). If it has none, then by Theorem 4.1 (a) it lies in A \mathcal{A} A , B \mathcal{B} B or C \mathcal{C} C or is a member of E \mathcal{E} E with even denominator. The members of A \mathcal{A} A with n n n even other than ( 1 , 1 , 4 ) / 6 (1,1,4)/6 ( 1 , 1 , 4 ) /6 , and the members of B \mathcal{B} B , are not lattice triangles by Theorem 4.18. The members of C \mathcal{C} C are ( 1 , m − 2 , m + 1 ) / 2 m (1,m-2,m+1)/2m ( 1 , m − 2 , m + 1 ) /2 m with m ≥ 4 m \ge4 m ≥ 4 even, ( 2 , m − 3 , m + 1 ) / 2 m (2,m-3,m+1)/2m ( 2 , m − 3 , m + 1 ) /2 m with m ≥ 6 m \ge6 m ≥ 6 even, and ( 4 , m − 5 , m + 1 ) / 2 m (4,m-5,m+1)/2m ( 4 , m − 5 , m + 1 ) /2 m with m ≡ 2 ( m o d 4 ) m \equiv2 \pmod{4} m ≡ 2 ( mod 4 ) , m ≥ 10 m \ge10 m ≥ 10 ; the first with m = 4 m=4 m = 4 and m = 6 m=6 m = 6 are ( 1 , 2 , 5 ) / 8 (1,2,5)/8 ( 1 , 2 , 5 ) /8 and ( 1 , 4 , 7 ) / 12 (1,4,7)/12 ( 1 , 4 , 7 ) /12 , and the others are not lattice triangles by Theorems 5.15, 5.8 and 5.10. Of the members of E \mathcal{E} E with even denominator, those of Lemma 7.1 are not lattice triangles by that lemma, and the others, which are in Table 2 , by Theorem 6.1. □
Appendix A. The computations
This appendix describes the computations the proofs use; the programs and their outputs are in the deposit [13 ] . The arithmetic on which a proof rests is exact, in integers, rationals or elements of Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) for a root of unity ζ \zeta ζ , and signs of real elements of Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) are decided by interval arithmetic with an exact zero test.
Cylinder certificates
This subsection proves that the cylinders of Theorem 6.1 exist and are maximal, and that the ratio of their moduli has the minimal polynomial given in Table 2 . Let Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n and let X X X be its unfolding, with singular set Σ \Sigma Σ as in § 2.1 ; every surface of Table 2 has genus between 3 and 20.
Coordinates. Let z = e i π / n z=e^{i\pi/n} z = e iπ / n . The copy of Δ \Delta Δ with the identity as linear part has vertices A = 0 A=0 A = 0 , B = − ( z r − z − r ) 2 = 4 sin 2 ( r π / n ) B=-(z^r-z^{-r})^2=4\sin^2(r\pi/n) B = − ( z r − z − r ) 2 = 4 sin 2 ( r π / n ) and C = − ( z r − z − r ) ( z q − z − q ) z p C=-(z^r-z^{-r})(z^q-z^{-q})z^p C = − ( z r − z − r ) ( z q − z − q ) z p , so that C / B = ( sin ( q π / n ) / sin ( r π / n ) ) e i p π / n C/B=(\sin(q\pi/n)/\sin(r\pi/n))e^{ip\pi/n} C / B = ( sin ( q π / n ) / sin ( r π / n )) e i p π / n and the angles at A A A , B B B , C C C are p π / n p\pi/n p π / n , q π / n q\pi/n q π / n , r π / n r\pi/n r π / n . The other copies are its images under the maps w ↦ z 2 k w w\mapsto z^{2k}w w ↦ z 2 k w and w ↦ z 2 k w ‾ w\mapsto z^{2k}\overline{w} w ↦ z 2 k w , and all their vertices lie in Z [ z ] \mathbb{Z}[z] Z [ z ] . For a direction d ∈ Q ( z ) ∖ { 0 } d\in\mathbb{Q}(z)\setminus\{0\} d ∈ Q ( z ) ∖ { 0 } the height of a point w w w of a copy is h ( w ) = w d ‾ − w ‾ d h(w)=w\overline{d}-\overline{w}d h ( w ) = w d − w d ; it is purely imaginary, and Im h ( w ) / ( 2 ∣ d ∣ ) \operatorname{Im}h(w)/(2\lvert d\rvert) Im h ( w ) / ( 2 ∣ d ∣) is the signed distance from w w w to the line R d \mathbb{R}d R d , positive on the left. Crossing a side out of a copy changes heights by a fixed element of Q ( z ) \mathbb{Q}(z) Q ( z ) (the height of the translation part of the gluing, with a minus sign) and adds that translation part to the holonomy.
A certificate for a cylinder in direction d d d consists of a holonomy c c c , a number V \mathcal{V} V , and a list of strips. A strip is given by a start (a copy, a side of it through which the leaf enters, and a height a 0 + b ϵ a_0+b\epsilon a 0 + b ϵ with a 0 ∈ Q ( z ) a_0\in\mathbb{Q}(z) a 0 ∈ Q ( z ) , b ∈ { − 1 , 0 , 1 } b\in\{-1,0,1\} b ∈ { − 1 , 0 , 1 } and ϵ \epsilon ϵ a positive infinitesimal; heights of this form are compared lexicographically), an itinerary (the list of triples: copy, entry side, exit side), and two heights lo < hi \operatorname{lo}<\operatorname{hi} lo < hi . The conditions on a certificate are:
(v1) the leaf entering the start copy through the start side at height a 0 + b ϵ a_0+b\epsilon a 0 + b ϵ , followed copy by copy (it leaves each copy through the other side whose two end heights lie on opposite sides of its own height), returns for the first time to the start side of the start copy at the same height after exactly the stored itinerary;
(v2) lo \operatorname{lo} lo (respectively hi \operatorname{hi} hi ) is the largest (respectively smallest) height, carried back to the start by the accumulated height changes, of a corner of a copy of the itinerary lying below (respectively above) a 0 + b ϵ a_0+b\epsilon a 0 + b ϵ ;
(v3) the accumulated holonomy is c c c , h ( c ) = 0 h(c)=0 h ( c ) = 0 and Re ( c d ‾ ) > 0 \operatorname{Re}(c\overline{d})>0 Re ( c d ) > 0 ;
(v4) exactly one strip has a corner in Σ \Sigma Σ among the corners attaining its lo \operatorname{lo} lo , and exactly one has a corner in Σ \Sigma Σ among those attaining its hi \operatorname{hi} hi ; starting from the first, each next strip has no corner in Σ \Sigma Σ attaining its lo \operatorname{lo} lo and one of the corners attaining its lo \operatorname{lo} lo is the same point of X X X as one of the corners attaining the current hi \operatorname{hi} hi ; the chain uses every strip once and ends at the second;
(v5) V \mathcal{V} V is the sum of hi − lo \operatorname{hi}-\operatorname{lo} hi − lo over the strips.
Lemma A.1. Let X X X have genus at least two. If a certificate in direction d d d passes (v1)–(v5), then the union of the leaves of its strips and of the leaves along which consecutive strips meet is a cylinder Z Z Z of X X X in direction d d d in the sense of § 2.2 , with core holonomy c c c , height H = V / ( 2 i ∣ d ∣ ) H = \mathcal{V}/(2i|d|) H = V / ( 2 i ∣ d ∣ ) , modulus μ = V / ( 2 i c d ˉ ) \mu= \mathcal{V}/(2i c\bar{d}) μ = V / ( 2 i c d ˉ ) , and area ( Z ) / area ( X ) = V c d ˉ / ( n d d ˉ ( B C ‾ − B ′ C ′ ‾ ) ) \operatorname{area}(Z)/\operatorname{area}(X) = \mathcal{V}c\bar{d}/(n d\bar{d}(\overline{BC}-\overline{B'C'}) ) area ( Z ) / area ( X ) = V c d ˉ / ( n d d ˉ ( B C − B ′ C ′ )) .
Proof. (i) Each strip is an annulus of closed leaves. Let η \eta η lie strictly between lo and hi. By (v2), at every step of the itinerary each corner of the copy lies strictly on the same side of η \eta η as of a 0 + b ϵ a_0+b\epsilon a 0 + b ϵ , so the leaf at height η \eta η leaves each copy through the same side as the leaf at a 0 + b ϵ a_0+b\epsilon a 0 + b ϵ and follows the same itinerary. The height changes depend only on the steps, so after the whole itinerary the leaf at η \eta η is back on the start side at height η \eta η , the total change being 0 by (v1); its holonomy is c c c by (v3), and it meets no corner, so it misses Σ \Sigma Σ .
(ii) Embedded. Let ℓ \ell ℓ be the length of the itinerary, and suppose the leaf at η \eta η enters the start copy through the start side at a step k k k with 0 < k < ℓ 0<k<\ell 0 < k < ℓ , at a height η \eta η strictly between lo and hi. By (i) it follows the itinerary from there, so the sequence of steps, which has period ℓ \ell ℓ , is invariant under the shift by k k k and so has period e = gcd ( k , ℓ ) < ℓ e=\gcd(k,\ell)<\ell e = g cd( k , ℓ ) < ℓ . The height changes are functions of the steps, so their sum over e e e steps is e / ℓ e/\ell e / ℓ times their sum over ℓ \ell ℓ steps, which is 0. Then the leaf at a 0 + b ϵ a_0+b\epsilon a 0 + b ϵ returns to the start side at the same height after e < ℓ e<\ell e < ℓ steps, against the first return in (v1). So distinct heights give distinct leaves (two leaves in direction d d d through one regular point coincide), and η ↦ \eta\mapsto η ↦ leaf is an embedding of an annulus of height ( h i − l o ) / ( 2 i ∣ d ∣ ) (\mathrm{hi}-\mathrm{lo})/(2i|d|) ( hi − lo ) / ( 2 i ∣ d ∣ ) foliated by closed leaves of holonomy c c c .
(iii) Joints. Let ∂ + S \partial^+S ∂ + S be the limit leaf at the upper end hi of a strip S S S whose corners attaining hi are all regular. The vertices on ∂ + S \partial^+S ∂ + S are the corners attaining hi; through these regular points ∂ + S \partial^+S ∂ + S is a straight line, so ∂ + S \partial^+S ∂ + S is a closed geodesic in direction d d d missing Σ \Sigma Σ , of holonomy c c c (a closed regular leaf has a neighbourhood of parallel closed leaves with its own holonomy, and the leaves of S S S near ∂ + S \partial^+S ∂ + S have holonomy c c c ). The next strip S ′ S' S ′ of the chain has a corner attaining its lo at the same point of X X X as a corner attaining hi (v4); that corner is a regular point and lies on exactly one leaf in direction d d d , so the lower limit leaf of S ′ S' S ′ is ∂ + S \partial^+S ∂ + S . Heights increase to the left of d d d , so S S S lies to the right of ∂ + S \partial^+S ∂ + S and S ′ S' S ′ to its left. By induction the union Z 0 Z_0 Z 0 of the strips and the joint leaves is a connected open set of closed regular leaves of holonomy c c c , so it lies in one cylinder Z Z Z .
(iv) Maximal. The frontier of Z 0 Z_0 Z 0 lies in the two outer limit leaves, the lower one of the first strip and the upper one of the last. By (v4) each of them passes through a point of Σ \Sigma Σ , and from any of its points the leaf in direction d d d follows it (it is straight through regular points) until it meets Σ \Sigma Σ ; so it is not a closed regular leaf. Hence Z 0 Z_0 Z 0 is closed in the union of the closed regular leaves, and being open and connected, Z 0 = Z Z_0=Z Z 0 = Z .
(v) Height and the formulas. The leaves of Z Z Z are parametrized, strip after strip, by heights; inside a strip the parametrization is η ↦ η \eta\mapsto\eta η ↦ η , and at each joint the next strip starts at the height where the previous one ended, so in the transverse distance the parametrization is a translation t ↦ t + c o n s t t\mapsto t+\mathrm{const} t ↦ t + const on each piece, with the pieces abutting. It is therefore an isometry from an interval of length ∑ ( h i − l o ) / ( 2 i ∣ d ∣ ) \sum(\mathrm{hi}-\mathrm{lo})/(2i|d|) ∑ ( hi − lo ) / ( 2 i ∣ d ∣ ) onto the space of leaves of Z Z Z , and H = V / ( 2 i ∣ d ∣ ) H=\mathcal{V}/(2i|d|) H = V / ( 2 i ∣ d ∣ ) . As c = λ d c=\lambda d c = λ d with λ > 0 \lambda>0 λ > 0 by (v3), ∣ c ∣ = c ˉ d / ∣ d ∣ |c|=\bar{c}d/|d| ∣ c ∣ = c ˉ d /∣ d ∣ , so μ = H / ∣ c ∣ = V / ( 2 i c d ˉ ) \mu=H/|c|=\mathcal{V}/(2i c\bar{d}) μ = H /∣ c ∣ = V / ( 2 i c d ˉ ) and area ( Z ) = H ∣ c ∣ = V c d ˉ / ( 2 i d d ˉ ) \operatorname{area}(Z)=H|c|=\mathcal{V}c\bar{d}/(2i d\bar{d}) area ( Z ) = H ∣ c ∣ = V c d ˉ / ( 2 i d d ˉ ) . The copy of Δ \Delta Δ has area Im ( B C ‾ ) / 2 = ( B C ‾ − B ′ C ′ ‾ ) / ( 4 i ) \operatorname{Im}(\overline{BC})/2=(\overline{BC}-\overline{B'C'})/(4i) Im ( B C ) /2 = ( B C − B ′ C ′ ) / ( 4 i ) , and X X X is made of 2 n 2n 2 n copies. □
A vertex of cone angle 2 π 2\pi 2 π on a boundary leaf does not stop the cylinder (§ 2.2 ): the chain of (v4) continues through it.
By Lemma A.1, if two cylinders in direction d d d have certificates with holonomies c k , c l c_k,c_l c k , c l and numbers V k , V l \mathcal{V}_k,\mathcal{V}_l V k , V l , the ratio of their moduli is μ k / μ l = V k c l d ˉ / ( V l c k d ˉ ) ∈ Q ( z ) \mu_k/\mu_l=\mathcal{V}_k c_l\bar{d}/(\mathcal{V}_l c_k\bar{d})\in\mathbb{Q}(z) μ k / μ l = V k c l d ˉ / ( V l c k d ˉ ) ∈ Q ( z ) , and its minimal polynomial over Q \mathbb{Q} Q is computed exactly. The certificates of the triangles of Table 2 , each with its direction and two or three cylinders, and the programs that verify (v1)–(v5) and compute the minimal polynomials are in the deposit [13 ] . The directions were found with the flatsurf software [5 ] .
The denominators not covered by the general theorems
Theorems 4.14 and C.2 leave finitely many denominators, which are determined here, with a proof that the lists are complete; the lists are in the deposit [13 ] . Let π i \pi_i π i be the i i i -th odd prime, Ω k = π 1 ⋯ π k \Omega_k=\pi_1\cdots\pi_k Ω k = π 1 ⋯ π k and β k = ∏ i ≤ k π i / ( π i − 1 ) \beta_k=\prod_{i\leq k}\pi_i/(\pi_i-1) β k = ∏ i ≤ k π i / ( π i − 1 ) . If N N N is odd and squarefree with k k k prime factors, then N ≥ Ω k N\geq\Omega_k N ≥ Ω k and N / φ ( N ) ≤ β k N/\varphi(N)\leq\beta_k N / φ ( N ) ≤ β k , so by Lemma 4.13
J ( N ) ≤ J k , J 1 = 2 , J 2 = 3 , J k = ⌊ ( 2 k − 1 ) β k ⌋ + 1 ( k ≥ 3 ) . J(N)\leq J_k,\qquad J_1=2,\qquad J_2=3,\qquad J_k=\left\lfloor(2^k-1)\beta_k\right\rfloor+1\ (k\geq3). J ( N ) ≤ J k , J 1 = 2 , J 2 = 3 , J k = ⌊ ( 2 k − 1 ) β k ⌋ + 1 ( k ≥ 3 ) .
For k = 0 k=0 k = 0 we have N = 1 N=1 N = 1 and J ( 1 ) = 1 J(1)=1 J ( 1 ) = 1 , and we put Ω 0 = 1 \Omega_0=1 Ω 0 = 1 , J 0 = 1 J_0=1 J 0 = 1 .
Odd n n n . Theorem C.2 applies unless 5 n ≤ 324 J ( n ) 2 5n\leq324J(n)^2 5 n ≤ 324 J ( n ) 2 . Let n ≥ 5 n\geq5 n ≥ 5 satisfy this inequality, with radical N N N having k ≥ 1 k\geq1 k ≥ 1 primes. Then 5 Ω k ≤ 5 N ≤ 5 n ≤ 324 J k 2 5\Omega_k\leq5N\leq5n\leq324J_k^2 5 Ω k ≤ 5 N ≤ 5 n ≤ 324 J k 2 . For k = 8 , 9 , 10 k=8,9,10 k = 8 , 9 , 10 one has J k = 780 , 1 , 618 , 3 , 347 J_k=780,1,\!618,3,\!347 J k = 780 , 1 , 618 , 3 , 347 and 5 Ω k > 324 J k 2 5\Omega_k>324J_k^2 5 Ω k > 324 J k 2 ; for k = 11 k=11 k = 11 , 5 Ω 11 > 324 ⋅ 9 11 5\Omega_{11}>324\cdot9^{11} 5 Ω 11 > 324 ⋅ 9 11 , and since π k + 1 > 9 \pi_{k+1}>9 π k + 1 > 9 for k ≥ 11 k\geq11 k ≥ 11 , 5 Ω k > 324 ⋅ 9 k ≥ 324 J ( N ) 2 5\Omega_k>324\cdot9^k\geq324J(N)^2 5 Ω k > 324 ⋅ 9 k ≥ 324 J ( N ) 2 for every k ≥ 11 k\geq11 k ≥ 11 (Lemma 4.13 , J ( N ) ≤ 3 k J(N)\leq3^k J ( N ) ≤ 3 k ). So k ≤ 7 k\leq7 k ≤ 7 and 5 N ≤ 324 J k 2 5N\leq324J_k^2 5 N ≤ 324 J k 2 . Each odd squarefree N N N with k ≤ 7 k\leq7 k ≤ 7 primes and 5 N ≤ 324 J k 2 5N\leq324J_k^2 5 N ≤ 324 J k 2 is discarded if 5 N > 324 E ( N ) 2 5N>324E(N)^2 5 N > 324 E ( N ) 2 , where E ( N ) E(N) E ( N ) is the least of the bounds of Lemma 4.13 at N N N itself; otherwise J ( N ) J(N) J ( N ) is computed exactly, as the largest gap between consecutive integers prime to N N N , and every odd n n n with radical N N N and 5 n ≤ 324 J ( N ) 2 5n\leq324J(N)^2 5 n ≤ 324 J ( N ) 2 is listed. The result is the list of 320 values n ≥ 5 n\geq5 n ≥ 5 of Theorem C.3 , the largest 3,135, with J = 7 J=7 J = 7 .
Even n = 2 m n=2m n = 2 m . Theorem 4.14 applies unless m ≤ N 0 ( m ) = P 2 ( m ) ( 31 J ( m ∘ ) 2 − 1 ) m\leq N_0(m)=P_2(m)(31J(m^\circ)^2-1) m ≤ N 0 ( m ) = P 2 ( m ) ( 31 J ( m ∘ ) 2 − 1 ) . Let m m m satisfy this inequality, with k ≥ 0 k\geq0 k ≥ 0 odd primes and odd radical N = m ∘ N=m^\circ N = m ∘ . Then P 2 ( m ) ≤ π k + 2 P_2(m)\leq\pi_{k+2} P 2 ( m ) ≤ π k + 2 and m ≥ N ≥ Ω k m\geq N\geq\Omega_k m ≥ N ≥ Ω k , so Ω k ≤ π k + 2 ( 31 J k 2 − 1 ) \Omega_k\leq\pi_{k+2}(31J_k^2-1) Ω k ≤ π k + 2 ( 31 J k 2 − 1 ) . This fails for 9 ≤ k ≤ 12 9\leq k\leq12 9 ≤ k ≤ 12 , by exact computation; and for k ≥ 13 k\geq13 k ≥ 13 , J ( N ) ≤ 3 k J(N)\leq3^k J ( N ) ≤ 3 k gives m < 31 ⋅ 9 k π k + 2 m<31\cdot9^k\pi_{k+2} m < 31 ⋅ 9 k π k + 2 , which fails at k = 13 k=13 k = 13 , while from k k k to k + 1 k+1 k + 1 the left side Ω k \Omega_k Ω k is multiplied by π k + 1 ≥ 47 \pi_{k+1}\geq47 π k + 1 ≥ 47 and the right side by 9 π k + 3 / π k + 2 < 18 9\pi_{k+3}/\pi_{k+2}<18 9 π k + 3 / π k + 2 < 18 (Bertrand’s postulate). So k ≤ 8 k\leq8 k ≤ 8 and N ≤ π k + 2 ( 31 J k 2 − 1 ) N\leq\pi_{k+2}(31J_k^2-1) N ≤ π k + 2 ( 31 J k 2 − 1 ) . One enumerates every set of k ≤ 8 k\leq8 k ≤ 8 odd primes whose product satisfies this bound (for k = 2 k=2 k = 2 , the larger range given by the third bound of Lemma 4.13 ), discards a set when its product exceeds P 2 ( 31 E 2 − 1 ) P_2(31E^2-1) P 2 ( 31 E 2 − 1 ) , with P 2 P_2 P 2 the second least odd prime outside the set and E E E the bound of Lemma 4.13 at the set itself (for the empty set, E = J ( 1 ) = 1 E=J(1)=1 E = J ( 1 ) = 1 ), computes J J J exactly for the rest by an exhaustive search over the ways of covering an initial run of integers by one residue class per prime, and lists every m ≥ 3 m\geq3 m ≥ 3 with exactly those odd primes and m ≤ N 0 ( m ) m\leq N_0(m) m ≤ N 0 ( m ) . The result is the list of 2,567 values of Theorem 4.15 , the largest 69,615 = 32 ⋅ 5 ⋅ 7 ⋅ 13 ⋅ 17 32\cdot5\cdot7\cdot13\cdot17 32 ⋅ 5 ⋅ 7 ⋅ 13 ⋅ 17 .
Enumerations of the criterion
Odd n n n . For the 320 values of Appendix A.§ 2 an enumeration of the window triangles finds that those without witness lie in A \mathcal{A} A or B \mathcal{B} B or are the five odd members of E \mathcal{E} E .
Even n n n . For the 2,567 values of m m m of Appendix A.§ 2 an enumeration of the window triangles of denominator 2 m 2m 2 m with p q ≤ 4 m J 2 pq\leq4mJ^2 pq ≤ 4 m J 2 finds that those without witness lie in A \mathcal{A} A or C \mathcal{C} C , have s ∣ n s\mid n s ∣ n , n / s n/s n / s odd and 2 ≤ p ≤ s / 2 2\leq p\leq s/2 2 ≤ p ≤ s /2 (and so lie in B \mathcal{B} B , by Proposition 4.3(c)), or are the twenty-five even members of E \mathcal{E} E .
The thirty members of E \mathcal{E} E . Each has no witness, by a direct test of all usable units, and none lies in A \mathcal{A} A , B \mathcal{B} B or C \mathcal{C} C .
Residues for every denominator, and the proof of the theorem for large even denominators
This appendix proves Theorem Theorem 4.14 : for m > N 0 ( m ) m > N_0(m) m > N 0 ( m ) , a window triangle of denominator n = 2 m n = 2m n = 2 m without witness lies in A \mathcal{A} A , B \mathcal{B} B or C \mathcal{C} C . Appendix B.1 proves the lemmas on residues that hold for every denominator; Appendix C uses them as well. Appendix B.2 proves the lemmas special to even n n n and the theorem. The plan of the proof follows the statement in Section 4.3, and a table of its cases follows Proposition 4.16.
Lemmas for every denominator
In this subsection n n n is any integer and Δ = ( p , q , r ) / n \Delta= (p,q,r)/n Δ = ( p , q , r ) / n is a window triangle (so n ≥ 5 n \ge5 n ≥ 5 ), s = 2 r − n s = 2r-n s = 2 r − n and L = n − 2 q = 2 p + s L = n-2q = 2p+s L = n − 2 q = 2 p + s ; the thresholds of S p S_p S p , S q S_q S q , S r S_r S r are 2 p 2p 2 p , 2 q 2q 2 q , s s s , and 2 p + 2 q + s = n 2p+2q+s=n 2 p + 2 q + s = n . Let n ∘ n^\circ n ∘ be the product of the odd primes dividing n n n (n ∘ = 1 n^\circ= 1 n ∘ = 1 if n n n is a power of 2), and J = J ( n ∘ ) J = J(n^\circ) J = J ( n ∘ ) . For n = 2 m n = 2m n = 2 m this is the J = J ( m ∘ ) J = J(m^\circ) J = J ( m ∘ ) of Section 4.3, and for odd n n n it is J ( n ) J(n) J ( n ) , as J J J depends only on the primes dividing its argument.
Lemma B.1. The units b b b modulo n n n with 2 b ≡ 2 ( m o d n ) 2b \equiv2 \pmod n 2 b ≡ 2 ( mod n ) are b = 1 b=1 b = 1 and, if 4 ∣ n 4 \mid n 4 ∣ n , b = 1 + n / 2 b=1+n/2 b = 1 + n /2 . Such a unit lying in S p S_p S p has [ b p ] n = p [bp]_n=p [ b p ] n = p . If one lies in S r S_r S r , then 4 ∣ n 4 \mid n 4 ∣ n , r r r is odd, b = 1 + n / 2 b=1+n/2 b = 1 + n /2 and [ b r ] n = s / 2 [br]_n=s/2 [ b r ] n = s /2 .
Proof. 2 b ≡ 2 ( m o d n ) 2b \equiv2 \pmod n 2 b ≡ 2 ( mod n ) means b ≡ 1 b \equiv1 b ≡ 1 modulo n / gcd ( 2 , n ) n/\gcd(2,n) n / g cd( 2 , n ) , so b = 1 b=1 b = 1 , or n n n is even and b = 1 + n / 2 b=1+n/2 b = 1 + n /2 , which is prime to n / 2 n/2 n /2 , and so a unit exactly when it is odd, that is when n / 2 n/2 n /2 is even. [ 1 ⋅ x ] n = x [1\cdot x]_n=x [ 1 ⋅ x ] n = x for x ∈ { p , r } x \in\{p,r\} x ∈ { p , r } , and r ≥ s r \ge s r ≥ s , so 1 ∉ S r 1 \notin S_r 1 ∈ / S r . For 4 ∣ n 4 \mid n 4 ∣ n , ( 1 + n / 2 ) x ≡ x (1+n/2)x \equiv x ( 1 + n /2 ) x ≡ x if x x x is even and ≡ x + n / 2 \equiv x+n/2 ≡ x + n /2 if x x x is odd. So [ ( 1 + n / 2 ) p ] n [(1+n/2)p]_n [( 1 + n /2 ) p ] n is p p p or p + n / 2 ≥ 2 p p+n/2 \ge2p p + n /2 ≥ 2 p (as p < n / 2 p<n/2 p < n /2 ), and [ ( 1 + n / 2 ) r ] n [(1+n/2)r]_n [( 1 + n /2 ) r ] n is r ≥ s r \ge s r ≥ s or, for r r r odd, r + n / 2 − n = s / 2 r+n/2-n=s/2 r + n /2 − n = s /2 . □ \square □
Good progressions. For integers c c c , b b b and an odd squarefree N N N , the progression c + i b c+ib c + ib (i ∈ Z i \in\mathbb{Z} i ∈ Z ) is good modulo N N N if gcd ( c , b , N ) = 1 \gcd(c,b,N)=1 g cd( c , b , N ) = 1 (the term is from [17 Definition 7.4] ). Any J ( N ) J(N) J ( N ) consecutive terms of a good progression contain one prime to N N N . Indeed, for a prime ℓ ∣ N \ell\mid N ℓ ∣ N : if ℓ ∣ b \ell\mid b ℓ ∣ b , then ℓ ∤ c + i b \ell\nmid c+ib ℓ ∤ c + ib for every i i i ; if ℓ ∤ b \ell\nmid b ℓ ∤ b , the i i i with ℓ ∣ c + i b \ell\mid c+ib ℓ ∣ c + ib form one class modulo ℓ \ell ℓ . Choosing i 0 i_0 i 0 in all these classes, c + i b c+ib c + ib is prime to N N N if and only if i − i 0 i-i_0 i − i 0 is prime to the product N ′ N' N ′ of the primes ℓ ∣ N \ell\mid N ℓ ∣ N with ℓ ∤ b \ell\nmid b ℓ ∤ b , and J ( N ′ ) ≤ J ( N ) J(N') \le J(N) J ( N ′ ) ≤ J ( N ) .
Lemma B.2. Let ( u , v , w ) (u,v,w) ( u , v , w ) be ( p , q , r ) (p,q,r) ( p , q , r ) in any order, and u ˉ = [ 2 u ] n \bar{u}=[2u]_n u ˉ = [ 2 u ] n , v ˉ = [ 2 v ] n \bar{v}=[2v]_n v ˉ = [ 2 v ] n , w ˉ = [ 2 w ] n \bar{w}=[2w]_n w ˉ = [ 2 w ] n , so that u ˉ + v ˉ + w ˉ = n \bar{u}+\bar{v}+\bar{w}=n u ˉ + v ˉ + w ˉ = n . Let a ∈ Z / n a \in\mathbb{Z}/n a ∈ Z / n with x = [ a u ] n < u ˉ x=[au]_n<\bar{u} x = [ a u ] n < u ˉ , and y = [ a v ] n y=[av]_n y = [ a v ] n . Then a ∉ S v ∪ S w a \notin S_v \cup S_w a ∈ / S v ∪ S w if and only if v ˉ ≤ y ≤ v ˉ + u ˉ − x \bar{v} \le y \le\bar{v}+\bar{u}-x v ˉ ≤ y ≤ v ˉ + u ˉ − x or n − x < y < n n-x<y<n n − x < y < n .
Proof. Let z = [ a w ] n z=[aw]_n z = [ a w ] n . As u + v + w = n u+v+w=n u + v + w = n , x + y + z ∈ { 0 , n , 2 n } x+y+z \in\{0,n,2n\} x + y + z ∈ { 0 , n , 2 n } . If the sum is 0, then y = 0 < v ˉ y=0<\bar{v} y = 0 < v ˉ and a ∈ S v a \in S_v a ∈ S v . If it is n n n , then z = n − x − y z=n-x-y z = n − x − y , and z ≥ w ˉ z \ge\bar{w} z ≥ w ˉ if and only if y ≤ u ˉ + v ˉ − x y \le\bar{u}+\bar{v}-x y ≤ u ˉ + v ˉ − x ; together with y ≥ v ˉ y \ge\bar{v} y ≥ v ˉ this is the first interval. If it is 2 n 2n 2 n , then z = 2 n − x − y < n z=2n-x-y<n z = 2 n − x − y < n forces y > n − x > n − u ˉ ≥ v ˉ y>n-x>n-\bar{u}\ge\bar{v} y > n − x > n − u ˉ ≥ v ˉ , and z ≥ n − x + 1 > v ˉ + w ˉ ≥ w ˉ z\ge n-x+1>\bar{v}+\bar{w}\ge\bar{w} z ≥ n − x + 1 > v ˉ + w ˉ ≥ w ˉ , so a ∉ S v ∪ S w a \notin S_v \cup S_w a ∈ / S v ∪ S w ; conversely n − x < y < n n-x<y<n n − x < y < n forces x + y > n x+y>n x + y > n , so the sum is 2 n 2n 2 n . Finally y y y in the first interval gives x + y ≤ u ˉ + v ˉ < n x+y\le\bar{u}+\bar{v}<n x + y ≤ u ˉ + v ˉ < n and y ≥ 1 y\ge1 y ≥ 1 , so the sum is n n n . □ \square □
For x ∈ { p , r } x \in\{p,r\} x ∈ { p , r } let τ = [ 2 x ] n \tau=[2x]_n τ = [ 2 x ] n and let τ ′ \tau' τ ′ be the other of 2 p , s 2p,s 2 p , s , so that S x = { a : [ a x ] n < τ } S_x=\{a:[ax]_n<\tau\} S x = { a : [ a x ] n < τ } and τ + τ ′ = L \tau+\tau'=L τ + τ ′ = L . For a residue a a a write X = [ a x ] n \mathcal{X}=[ax]_n X = [ a x ] n and Y = [ − a q ] n \mathcal{Y}=[-aq]_n Y = [ − a q ] n .
Lemma B.3. (a) a ∉ S q a \notin S_q a ∈ / S q if and only if Y ∈ [ 1 , L ] \mathcal{Y} \in[1,L] Y ∈ [ 1 , L ] .
(b) L X ≡ τ Y ( m o d n ) L\mathcal{X} \equiv\tau\mathcal{Y} \pmod n L X ≡ τ Y ( mod n ) .
(c) If X < τ \mathcal{X}<\tau X < τ , then a ∉ S q ∪ S x ′ a \notin S_q \cup S_{x'} a ∈ / S q ∪ S x ′ , x ′ x' x ′ the other of p , r p,r p , r , if and only if 1 ≤ Y ≤ X − 1 1 \le\mathcal{Y} \le\mathcal{X}-1 1 ≤ Y ≤ X − 1 or X + τ ′ ≤ Y ≤ L \mathcal{X}+\tau' \le\mathcal{Y} \le L X + τ ′ ≤ Y ≤ L .
(d) If 1 ≤ X < τ 1 \le\mathcal{X} < \tau 1 ≤ X < τ and X + τ ′ ≤ Y ≤ L \mathcal{X}+\tau' \le\mathcal{Y} \le L X + τ ′ ≤ Y ≤ L (“above”), then τ Y − L X \tau\mathcal{Y}-L\mathcal{X} τ Y − L X is a multiple of n n n with n ≤ τ Y − L X ≤ L ( τ − X ) n \le\tau\mathcal{Y}-L\mathcal{X} \le L(\tau-\mathcal{X}) n ≤ τ Y − L X ≤ L ( τ − X ) . If 1 ≤ Y ≤ X − 1 1 \le\mathcal{Y} \le\mathcal{X}-1 1 ≤ Y ≤ X − 1 (“below”), then L X − τ Y L\mathcal{X}-\tau\mathcal{Y} L X − τ Y is a multiple of n n n with n ≤ L X − τ Y ≤ L X − τ n \le L\mathcal{X}-\tau\mathcal{Y} \le L\mathcal{X}-\tau n ≤ L X − τ Y ≤ L X − τ .
Proof. (a) a ∉ S q a \notin S_q a ∈ / S q means [ a q ] n ≥ 2 q [aq]_n \ge2q [ a q ] n ≥ 2 q , that is a q ≢ 0 aq \not\equiv0 a q ≡ 0 and n − [ a q ] n ≤ n − 2 q = L n-[aq]_n \le n-2q=L n − [ a q ] n ≤ n − 2 q = L . (b) L X − τ Y ≡ a ( L x + τ q ) L\mathcal{X}-\tau\mathcal{Y} \equiv a(Lx+\tau q) L X − τ Y ≡ a ( Lx + τ q ) ; for x = p x=p x = p this is a p ( L + 2 q ) = a p n ap(L+2q)=apn a p ( L + 2 q ) = a p n , and for x = r x=r x = r , with L = n − 2 q L=n-2q L = n − 2 q and τ = 2 r − n \tau=2r-n τ = 2 r − n , it is a ( ( n − 2 q ) r + ( 2 r − n ) q ) = a n ( r − q ) a((n-2q)r+(2r-n)q)=an(r-q) a (( n − 2 q ) r + ( 2 r − n ) q ) = an ( r − q ) . (c) Apply Lemma B.2 with ( u , v , w ) = ( x , x ′ , q ) (u,v,w)=(x,x',q) ( u , v , w ) = ( x , x ′ , q ) , so ( u ‾ , v ‾ ) = ( τ , τ ′ ) (\overline{u},\overline{v})=(\tau,\tau') ( u , v ) = ( τ , τ ′ ) and the x x x of that lemma is X \mathcal{X} X : with y = [ a x ′ ] n y=[ax']_n y = [ a x ′ ] n , a ∉ S x ′ ∪ S q a \notin S_{x'} \cup S_q a ∈ / S x ′ ∪ S q if and only if τ ′ ≤ y ≤ L − X \tau' \le y \le L-\mathcal{X} τ ′ ≤ y ≤ L − X or n − X < y < n n-\mathcal{X}<y<n n − X < y < n . As − a q ≡ a x + a x ′ -aq \equiv ax+ax' − a q ≡ a x + a x ′ , Y ≡ X + y \mathcal{Y}\equiv\mathcal{X}+y Y ≡ X + y . In the first case X + y ∈ [ X + τ ′ , L ] \mathcal{X}+y\in[\mathcal{X}+\tau',L] X + y ∈ [ X + τ ′ , L ] and L < n L<n L < n , so Y = X + y \mathcal{Y}=\mathcal{X}+y Y = X + y ; in the second X + y ∈ ( n , n + X ) \mathcal{X}+y\in(n,n+\mathcal{X}) X + y ∈ ( n , n + X ) , so Y = X + y − n ∈ [ 1 , X − 1 ] \mathcal{Y}=\mathcal{X}+y-n\in[1,\mathcal{X}-1] Y = X + y − n ∈ [ 1 , X − 1 ] . Conversely Y ∈ [ X + τ ′ , L ] \mathcal{Y}\in[\mathcal{X}+\tau',L] Y ∈ [ X + τ ′ , L ] gives y = Y − X ∈ [ τ ′ , L − X ] y=\mathcal{Y}-\mathcal{X}\in[\tau',L-\mathcal{X}] y = Y − X ∈ [ τ ′ , L − X ] , and Y ∈ [ 1 , X − 1 ] \mathcal{Y}\in[1,\mathcal{X}-1] Y ∈ [ 1 , X − 1 ] gives y = n + Y − X ∈ ( n − X , n ) y=n+\mathcal{Y}-\mathcal{X}\in(n-\mathcal{X},n) y = n + Y − X ∈ ( n − X , n ) . (d) Above: τ Y − L X ≥ τ ( X + τ ′ ) − ( τ + τ ′ ) X = τ ′ ( τ − X ) > 0 \tau\mathcal{Y}-L\mathcal{X}\ge\tau(\mathcal{X}+\tau')-(\tau+\tau')\mathcal{X}=\tau'(\tau-\mathcal{X})>0 τ Y − L X ≥ τ ( X + τ ′ ) − ( τ + τ ′ ) X = τ ′ ( τ − X ) > 0 and τ Y − L X ≤ τ L − L X \tau\mathcal{Y}-L\mathcal{X}\le\tau L-L\mathcal{X} τ Y − L X ≤ τ L − L X . Below: L X − τ Y ≥ L X − τ ( X − 1 ) = τ ′ X + τ > 0 L\mathcal{X}-\tau\mathcal{Y}\ge L\mathcal{X}-\tau(\mathcal{X}-1)=\tau'\mathcal{X}+\tau>0 L X − τ Y ≥ L X − τ ( X − 1 ) = τ ′ X + τ > 0 and L X − τ Y ≤ L X − τ L\mathcal{X}-\tau\mathcal{Y}\le L\mathcal{X}-\tau L X − τ Y ≤ L X − τ . Both are multiples of n n n by (b). □
Lemma B.4. Let ϵ = gcd ( 2 , n ) \epsilon=\gcd(2,n) ϵ = g cd( 2 , n ) . If p q > ϵ n J 2 pq>\epsilon nJ^2 pq > ϵ n J 2 , there is a witness in S p ∩ S q S_p\cap S_q S p ∩ S q .
Proof. Let Λ = { ( x , z ) ∈ Z 2 : ( x , z ) ≡ ( b p , b q ) ( m o d n ) \Lambda=\{(x,z)\in\mathbb{Z}^2:(x,z)\equiv(bp,bq)\pmod n Λ = {( x , z ) ∈ Z 2 : ( x , z ) ≡ ( b p , b q ) ( mod n ) for some b ≡ 0 ( m o d ϵ ) } b\equiv0\pmod\epsilon\} b ≡ 0 ( mod ϵ )} . As gcd ( p , q , n ) \gcd(p,q,n) g cd( p , q , n ) divides r = n − p − q r=n-p-q r = n − p − q , it is 1 1 1 , so b ↦ ( b p , b q ) ( m o d n ) b\mapsto(bp,bq)\pmod n b ↦ ( b p , b q ) ( mod n ) is injective on Z / n \mathbb{Z}/n Z / n ; the b ≡ 0 ( m o d ϵ ) b\equiv0\pmod\epsilon b ≡ 0 ( mod ϵ ) form a subgroup of order n / ϵ n/\epsilon n / ϵ , so Λ \Lambda Λ is a lattice of index n 2 / ( n / ϵ ) = ϵ n n^2/(n/\epsilon)=\epsilon n n 2 / ( n / ϵ ) = ϵ n in Z 2 \mathbb{Z}^2 Z 2 . The open box ∣ x ∣ < p / J |x|<p/J ∣ x ∣ < p / J , ∣ z ∣ < q / J |z|<q/J ∣ z ∣ < q / J has area 4 p q / J 2 > 4 ϵ n 4pq/J^2>4\epsilon n 4 pq / J 2 > 4 ϵ n , so by Minkowski’s theorem it contains a point ( x , z ) ≠ 0 (x,z)\ne0 ( x , z ) = 0 of Λ \Lambda Λ ; let b ≡ 0 ( m o d ϵ ) b\equiv0\pmod\epsilon b ≡ 0 ( mod ϵ ) be such that ( x , z ) ≡ ( b p , b q ) (x,z)\equiv(bp,bq) ( x , z ) ≡ ( b p , b q ) . For 1 ≤ i ≤ J 1\le i\le J 1 ≤ i ≤ J let a i = 1 + i b a_i=1+ib a i = 1 + ib . Then a i p ≡ p + i x a_ip\equiv p+ix a i p ≡ p + i x and a i q ≡ q + i z a_iq\equiv q+iz a i q ≡ q + i z , with 0 < p + i x < 2 p 0<p+ix<2p 0 < p + i x < 2 p and 0 < q + i z < 2 q 0<q+iz<2q 0 < q + i z < 2 q as ∣ i x ∣ < p |ix|<p ∣ i x ∣ < p and ∣ i z ∣ < q |iz|<q ∣ i z ∣ < q ; so a i ∈ S p ∩ S q a_i\in S_p\cap S_q a i ∈ S p ∩ S q . If 2 a i ≡ 2 2a_i\equiv2 2 a i ≡ 2 , then 2 i b ≡ 0 2ib\equiv0 2 ib ≡ 0 , so 2 i x ≡ 2 i z ≡ 0 ( m o d n ) 2ix\equiv2iz\equiv0\pmod n 2 i x ≡ 2 i z ≡ 0 ( mod n ) , and ∣ 2 i x ∣ < 2 p < n |2ix|<2p<n ∣2 i x ∣ < 2 p < n , ∣ 2 i z ∣ < 2 q < n |2iz|<2q<n ∣2 i z ∣ < 2 q < n give x = z = 0 x=z=0 x = z = 0 ; so 2 a i ≠ 2 2a_i\ne2 2 a i = 2 . The a i a_i a i are J J J consecutive terms of the progression 1 + i b 1+ib 1 + ib , so one of them is prime to n n n (as 1 + i b 1+ib 1 + ib is a good progression modulo n n n ); it is odd when n n n is even, as b b b is then even, so it is a unit, and a witness. □
Lemma B.4 is a variant of [17 Proposition 7.5] , which finds a usable unit in S p ∩ S q S_p\cap S_q S p ∩ S q when 4 ⌊ p / 2 j ( n ) ⌋ ⌊ q / 2 j ( n ) ⌋ > n 4\lfloor p/2j(n)\rfloor\lfloor q/2j(n)\rfloor>n 4 ⌊ p /2 j ( n )⌋ ⌊ q /2 j ( n )⌋ > n , j j j being Jacobsthal’s function, by the same progression through a = 1 a=1 a = 1 .
Lemma B.5. Let x ∈ { p , r } x\in\{p,r\} x ∈ { p , r } and let a 0 a_0 a 0 be a unit of S x S_x S x with X 0 = [ a 0 x ] n ≥ 1 \mathcal{X}_0=[a_0x]_n\ge1 X 0 = [ a 0 x ] n ≥ 1 and X 0 + τ ′ ≤ Y 0 = [ − a 0 q ] n ≤ L \mathcal{X}_0+\tau'\le\mathcal{Y}_0=[-a_0q]_n\le L X 0 + τ ′ ≤ Y 0 = [ − a 0 q ] n ≤ L . Let u ≥ 2 u\ge2 u ≥ 2 be an integer prime to n n n with u L < n uL<n uL < n , let a j = u j a 0 a_j=u^ja_0 a j = u j a 0 , and let κ ≥ 0 \kappa\ge0 κ ≥ 0 be the largest integer with u κ X 0 < τ u^\kappa\mathcal{X}_0<\tau u κ X 0 < τ . If 0 ≤ j 1 ≤ κ 0\le j_1\le\kappa 0 ≤ j 1 ≤ κ and a 1 , … , a j 1 a_1,\ldots,a_{j_1} a 1 , … , a j 1 are usable and not witnesses, then [ a j x ] n = u j X 0 [a_jx]_n=u^j\mathcal{X}_0 [ a j x ] n = u j X 0 and [ − a j q ] n = u j Y 0 ≤ L [-a_jq]_n=u^j\mathcal{Y}_0\le L [ − a j q ] n = u j Y 0 ≤ L for 0 ≤ j ≤ j 1 0\le j\le j_1 0 ≤ j ≤ j 1 , and
n ≤ ( u κ + 1 − j 1 − 1 ) L X 0 . n\le(u^{\kappa+1-j_1}-1)L\mathcal{X}_0. n ≤ ( u κ + 1 − j 1 − 1 ) L X 0 .
In particular, if ( u − 1 ) L X 0 < n (u-1)L\mathcal{X}_0<n ( u − 1 ) L X 0 < n , then κ ≥ 1 \kappa\ge1 κ ≥ 1 and some a j a_j a j with 1 ≤ j ≤ κ 1\le j\le\kappa 1 ≤ j ≤ κ is a witness or is not usable; for the least such j j j , [ a j x ] n = u j X 0 [a_jx]_n=u^j\mathcal{X}_0 [ a j x ] n = u j X 0 .
Proof. For j = 0 j=0 j = 0 the two equalities are the hypothesis. Suppose they hold for some j < j 1 j<j_1 j < j 1 . Then j + 1 ≤ κ j+1\le\kappa j + 1 ≤ κ , so u j + 1 X 0 < τ u^{j+1}\mathcal{X}_0<\tau u j + 1 X 0 < τ ; hence [ a j + 1 x ] n = u j + 1 X 0 [a_{j+1}x]_n=u^{j+1}\mathcal{X}_0 [ a j + 1 x ] n = u j + 1 X 0 and a j + 1 ∈ S x a_{j+1}\in S_x a j + 1 ∈ S x . Also 1 ≤ u j Y 0 ≤ u L < n 1\le u^j\mathcal{Y}_0\le uL<n 1 ≤ u j Y 0 ≤ uL < n , so [ − a j + 1 q ] n = u j + 1 Y 0 [-a_{j+1}q]_n=u^{j+1}\mathcal{Y}_0 [ − a j + 1 q ] n = u j + 1 Y 0 . As a j + 1 a_{j+1} a j + 1 is a usable unit of S x S_x S x and not a witness, it is not in S q S_q S q , so u j + 1 Y 0 ≤ L u^{j+1}\mathcal{Y}_0\le L u j + 1 Y 0 ≤ L by Lemma B.3(a). This proves the equalities. Next, u κ + 1 X 0 ≥ τ u^{\kappa+1}\mathcal{X}_0\ge\tau u κ + 1 X 0 ≥ τ and u j 1 Y 0 ≤ L u^{j_1}\mathcal{Y}_0\le L u j 1 Y 0 ≤ L give τ Y 0 ≤ u κ + 1 X 0 Y 0 = u κ + 1 − j 1 X 0 ⋅ u j 1 Y 0 ≤ u κ + 1 − j 1 X 0 L \tau\mathcal{Y}_0\le u^{\kappa+1}\mathcal{X}_0\mathcal{Y}_0=u^{\kappa+1-j_1}\mathcal{X}_0\cdot u^{j_1}\mathcal{Y}_0\le u^{\kappa+1-j_1}\mathcal{X}_0L τ Y 0 ≤ u κ + 1 X 0 Y 0 = u κ + 1 − j 1 X 0 ⋅ u j 1 Y 0 ≤ u κ + 1 − j 1 X 0 L , and by Lemma B.3(d), n ≤ τ Y 0 − L X 0 ≤ ( u κ + 1 − j 1 − 1 ) L X 0 n\le\tau\mathcal{Y}_0-L\mathcal{X}_0\le(u^{\kappa+1-j_1}-1)L\mathcal{X}_0 n ≤ τ Y 0 − L X 0 ≤ ( u κ + 1 − j 1 − 1 ) L X 0 . If ( u − 1 ) L X 0 < n (u-1)L\mathcal{X}_0<n ( u − 1 ) L X 0 < n , this rules out j 1 = κ j_1=\kappa j 1 = κ ; as j 1 = 0 j_1=0 j 1 = 0 is always allowed, κ ≥ 1 \kappa\ge1 κ ≥ 1 . Let j 1 j_1 j 1 be the largest integer in [ 0 , κ ] [0,\kappa] [ 0 , κ ] such that a 1 , … , a j 1 a_1,\ldots,a_{j_1} a 1 , … , a j 1 are usable and not witnesses; then j 1 < κ j_1<\kappa j 1 < κ , and j = j 1 + 1 ≤ κ j=j_1+1\le\kappa j = j 1 + 1 ≤ κ is the least index with a j a_j a j a witness or not usable, and a j x = u ⋅ u j 1 X 0 = u j X 0 < τ a_jx=u\cdot u^{j_1}\mathcal{X}_0=u^j\mathcal{X}_0<\tau a j x = u ⋅ u j 1 X 0 = u j X 0 < τ , so [ a j x ] n = u j X 0 [a_jx]_n=u^j\mathcal{X}_0 [ a j x ] n = u j X 0 . □
Corollary B.6. Suppose that Δ \Delta Δ has no witness, and let u u u be a prime not dividing n n n with u L < n uL<n uL < n .
(a) If gcd ( r , n ) = 1 \gcd(r,n)=1 g cd( r , n ) = 1 and s ≥ 3 s\ge3 s ≥ 3 , then 4 ∣ n 4\mid n 4 ∣ n and s / 2 s/2 s /2 is a power of u u u .
(b) If gcd ( p , n ) = 1 \gcd(p,n)=1 g cd( p , n ) = 1 and p ≥ 2 p\ge2 p ≥ 2 , then p p p is a power of u u u , and ( u 2 − 1 ) L ≥ n (u^2-1)L\ge n ( u 2 − 1 ) L ≥ n .
In both the exponent is at least 1. Hence, if (a) or (b) applies, at most one prime u ∤ n u\nmid n u ∤ n has u L < n uL<n uL < n ; for n = 2 m n=2m n = 2 m , where the two least primes not dividing n n n are P ( m ) < P 2 ( m ) P(m)<P_2(m) P ( m ) < P 2 ( m ) , this gives P 2 ( m ) L ≥ n P_2(m)L\ge n P 2 ( m ) L ≥ n .
Proof. Let x = r x=r x = r in (a) and x = p x=p x = p in (b), and a 0 = x − 1 ( m o d n ) a_0=x^{-1}\pmod n a 0 = x − 1 ( mod n ) : a unit of S x S_x S x with X 0 = 1 X_0=1 X 0 = 1 , as τ = s ≥ 3 \tau=s\ge3 τ = s ≥ 3 or τ = 2 p ≥ 4 \tau=2p\ge4 τ = 2 p ≥ 4 . It is usable by Lemma B.1 , since X 0 = 1 X_0=1 X 0 = 1 differs from s / 2 s/2 s /2 in (a) and from p p p in (b). It is not a witness, so it is not in S q ∪ S x ′ S_q\cup S_{x'} S q ∪ S x ′ , and by Lemma B.3 (c), Y 0 = [ − a 0 q ] n ∈ [ 1 + τ , L ] Y_0=[-a_0q]_n\in[1+\tau,L] Y 0 = [ − a 0 q ] n ∈ [ 1 + τ , L ] : above. Lemma B.5 applies with u u u , and ( u − 1 ) L X 0 < u L < n (u-1)LX_0<uL<n ( u − 1 ) L X 0 < uL < n : some a j = u j a 0 a_j=u^ja_0 a j = u j a 0 with 1 ≤ j ≤ κ 1\le j\le\kappa 1 ≤ j ≤ κ is a witness or not usable, and for the least such j j j , [ a j x ] n = u j [a_jx]_n=u^j [ a j x ] n = u j . There is no witness, so a j a_j a j is a unit of S x S_x S x that is not usable, and by Lemma B.1 : in (a), 4 ∣ n 4\mid n 4 ∣ n and u j = s / 2 u^j=s/2 u j = s /2 ; in (b), u j = p u^j=p u j = p . In (b) moreover κ = j \kappa=j κ = j , as u j + 1 = u p ≥ 2 p = τ u^{j+1}=up\ge2p=\tau u j + 1 = u p ≥ 2 p = τ , and a 1 , … , a j − 1 a_1,\ldots,a_{j-1} a 1 , … , a j − 1 are usable and not witnesses, so Lemma B.5 with j 1 = j − 1 = κ − 1 j_1=j-1=\kappa-1 j 1 = j − 1 = κ − 1 gives n ≤ ( u 2 − 1 ) L X 0 = ( u 2 − 1 ) L n\le(u^2-1)LX_0=(u^2-1)L n ≤ ( u 2 − 1 ) L X 0 = ( u 2 − 1 ) L . Finally, a number at least 2 is not a power of two different primes; for n = 2 m n=2m n = 2 m , P 2 ( m ) L < n P_2(m)L<n P 2 ( m ) L < n would give P ( m ) L < n P(m)L<n P ( m ) L < n as well. □
Lemma B.7. Let g g g be a divisor of n n n , with n / g n/g n / g even if n n n is even, such that gcd ( q , g ) = 1 \gcd(q,g)=1 g cd( q , g ) = 1 . Let a 0 a_0 a 0 be a residue modulo n n n with gcd ( a 0 , n / g ) = 1 \gcd(a_0,n/g)=1 g cd( a 0 , n / g ) = 1 , R = { a 0 + ( n / g ) i : i ∈ Z / g } R=\{a_0+(n/g)i:i\in\mathbb{Z}/g\} R = { a 0 + ( n / g ) i : i ∈ Z / g } (the row of $a_0), Y 0 = [ − a 0 q ] n Y_0=[-a_0q]_n Y 0 = [ − a 0 q ] n , and Υ = ⌊ ( L − 1 ) g / n ⌋ + 1 \Upsilon=\lfloor(L-1)g/n\rfloor+1 Υ = ⌊( L − 1 ) g / n ⌋ + 1 . Let g ∗ g^* g ∗ be the product of the primes ℓ ∣ g \ell\mid g ℓ ∣ g with ℓ ∤ n / g \ell\nmid n/g ℓ ∤ n / g ; they are odd.
(i) a ↦ [ − a q ] n a\mapsto[-aq]_n a ↦ [ − a q ] n maps R R R bijectively onto { [ Y 0 + ( n / g ) t ] n : t ∈ Z / g } \{[Y_0+(n/g)t]_n:t\in\mathbb{Z}/g\} {[ Y 0 + ( n / g ) t ] n : t ∈ Z / g } , the element a 0 + ( n / g ) i a_0+(n/g)i a 0 + ( n / g ) i going to t ≡ − q i ( m o d g ) t\equiv-qi\pmod g t ≡ − q i ( mod g ) , which we call its index.
(ii) The elements of R R R outside S q S_q S q have cyclically consecutive values of t t t , at most Υ \Upsilon Υ of them.
(iii) For each prime ℓ ∣ g ∗ \ell\mid g^* ℓ ∣ g ∗ there is a class t ℓ t_\ell t ℓ modulo ℓ \ell ℓ such that the element of index t t t is a unit if and only if t ≢ t ℓ ( m o d ℓ ) t\not\equiv t_\ell\pmod\ell t ≡ t ℓ ( mod ℓ ) for every ℓ ∣ g ∗ \ell\mid g^* ℓ ∣ g ∗ . In particular any J ( g ∗ ) J(g^*) J ( g ∗ ) consecutive values of t t t contain one whose element is a unit.
(iv) Let ν \nu ν be the number of units of R R R that are not usable, and suppose g − Υ ≥ ( 1 + ν ) J ( g ∗ ) g-\Upsilon\ge(1+\nu)J(g^*) g − Υ ≥ ( 1 + ν ) J ( g ∗ ) . Then R R R contains a usable unit lying in S q S_q S q .
If g g g divides n n n and one of p , r p,r p , r , then gcd ( q , g ) = 1 \gcd(q,g)=1 g cd( q , g ) = 1 .
Proof. A prime ℓ ∣ g \ell\mid g ℓ ∣ g with ℓ ∤ n / g \ell\nmid n/g ℓ ∤ n / g is odd: if ℓ = 2 \ell=2 ℓ = 2 , then n n n is even and n / g n/g n / g odd. (i) − ( a 0 + ( n / g ) i ) q ≡ Y 0 + ( n / g ) ( − q i ) ( m o d n ) -(a_0+(n/g)i)q\equiv Y_0+(n/g)(-qi)\pmod n − ( a 0 + ( n / g ) i ) q ≡ Y 0 + ( n / g ) ( − q i ) ( mod n ) , and i ↦ − q i i\mapsto-qi i ↦ − q i is a bijection of Z / g \mathbb{Z}/g Z / g . (ii) The g g g points [ Y 0 + ( n / g ) t ] n [Y_0+(n/g)t]_n [ Y 0 + ( n / g ) t ] n are equally spaced on Z / n \mathbb{Z}/n Z / n with gap n / g n/g n / g ; outside S q S_q S q means in [ 1 , L ] [1,L] [ 1 , L ] (Lemma B.3 (a)), an arc that holds at most ⌊ ( L − 1 ) / ( n / g ) ⌋ + 1 \lfloor(L-1)/(n/g)\rfloor+1 ⌊( L − 1 ) / ( n / g )⌋ + 1 of them, consecutive ones. (iii) For a prime ℓ ∤ n / g \ell\nmid n/g ℓ ∤ n / g , a 0 + ( n / g ) i ≡ a 0 ≢ 0 ( m o d ℓ ) a_0+(n/g)i\equiv a_0\not\equiv0\pmod\ell a 0 + ( n / g ) i ≡ a 0 ≡ 0 ( mod ℓ ) . For a prime ℓ ∣ g ∗ \ell\mid g^* ℓ ∣ g ∗ , n / g n/g n / g is invertible modulo ℓ \ell ℓ , so ℓ ∣ a 0 + ( n / g ) i \ell\mid a_0+(n/g)i ℓ ∣ a 0 + ( n / g ) i for exactly one class of i i i modulo ℓ \ell ℓ , that is, q q q being invertible modulo ℓ \ell ℓ , for one class of t t t . Every prime of n n n divides n / g n/g n / g or g ∗ g^* g ∗ . The last sentence follows by the Chinese remainder theorem: with t 0 ≡ t ℓ ( m o d ℓ ) t_0\equiv t_\ell\pmod\ell t 0 ≡ t ℓ ( mod ℓ ) for every ℓ ∣ g ∗ \ell\mid g^* ℓ ∣ g ∗ , the element of index t t t is a unit if and only if t − t 0 t-t_0 t − t 0 is prime to g ∗ g^* g ∗ , and J ( g ∗ ) ≤ g ∗ ≤ g J(g^*)\le g^*\le g J ( g ∗ ) ≤ g ∗ ≤ g . (iv) By (ii) the t t t whose element lies in S q S_q S q include g − Υ g-\Upsilon g − Υ cyclically consecutive values; cut 1 + ν 1+\nu 1 + ν disjoint blocks of J ( g ∗ ) J(g^*) J ( g ∗ ) consecutive values out of them. By (iii) each block holds a unit; these 1 + ν 1+\nu 1 + ν units are distinct and lie in S q S_q S q , and at most ν \nu ν of them are not usable. For the last sentence: a common prime of q q q and g g g divides n n n , q q q and one of p , r p,r p , r , hence the third numerator, against gcd ( p , q , r ) = 1 \gcd(p,q,r)=1 g cd( p , q , r ) = 1 . □
For g ≥ 4 g \ge4 g ≥ 4 the units can be named, and Jacobsthal’s function is not needed. On Z / g \mathbb{Z}/g Z / g the gap between two cyclically consecutive points of a set is their circular distance.
Lemma B.8. In Lemma B.7 let g ≥ 4 g \ge4 g ≥ 4 , and suppose 7 L ≤ 3 n 7L \le3n 7 L ≤ 3 n and either ν = 0 \nu= 0 ν = 0 , or ν = 1 \nu= 1 ν = 1 and g ≥ 7 g \ge7 g ≥ 7 . Then R R R contains a usable unit lying in S q S_q S q .
Proof. Choose t 0 ∈ Z / g t_0 \in\mathbb{Z}/g t 0 ∈ Z / g with t 0 ≡ t ℓ ( m o d ℓ ) t_0 \equiv t_\ell\pmod{\ell} t 0 ≡ t ℓ ( mod ℓ ) for each ℓ ∣ g ∗ \ell\mid g^* ℓ ∣ g ∗ (Chinese remainder theorem); then the element of index t t t is a unit if and only if gcd ( t − t 0 , g ∗ ) = 1 \gcd(t-t_0,g^*)=1 g cd( t − t 0 , g ∗ ) = 1 . Let g ′ = ⌊ ( g − 1 ) / 2 ⌋ g'=\lfloor(g-1)/2\rfloor g ′ = ⌊( g − 1 ) /2 ⌋ . Each of ± 1 \pm1 ± 1 , ± 2 \pm2 ± 2 , ± g ′ \pm g' ± g ′ is prime to every odd prime dividing g g g , as 2 g ′ 2g' 2 g ′ is g − 1 g-1 g − 1 or g − 2 g-2 g − 2 ; so the elements of indices t 0 ± 1 t_0\pm1 t 0 ± 1 , t 0 ± 2 t_0\pm2 t 0 ± 2 , t 0 ± g ′ t_0\pm g' t 0 ± g ′ are units. Let Π = { t 0 ± 1 , t 0 ± g ′ } \Pi=\{t_0\pm1,t_0\pm g'\} Π = { t 0 ± 1 , t 0 ± g ′ } if ν = 0 \nu=0 ν = 0 , and Π = { t 0 ± 1 , t 0 ± 2 , t 0 ± g ′ } \Pi=\{t_0\pm1,t_0\pm2,t_0\pm g'\} Π = { t 0 ± 1 , t 0 ± 2 , t 0 ± g ′ } without the index of the unit that is not usable if ν = 1 \nu=1 ν = 1 ; every element with index in Π \Pi Π is a usable unit. The gaps of Π \Pi Π are at most max ( 2 , g ′ − 1 ) \max(2,g'-1) max ( 2 , g ′ − 1 ) if ν = 0 \nu=0 ν = 0 (they are g ′ − 1 g'-1 g ′ − 1 , g − 2 g ′ g-2g' g − 2 g ′ , g ′ − 1 g'-1 g ′ − 1 , 2 2 2 , and g − 2 g ′ ≤ 2 g-2g'\le2 g − 2 g ′ ≤ 2 ), and at most max ( 3 , g ′ ) \max(3,g') max ( 3 , g ′ ) if ν = 1 \nu=1 ν = 1 (for g ≥ 7 g\ge7 g ≥ 7 the six points are distinct, with gaps 1 1 1 , g ′ − 2 g'-2 g ′ − 2 , g − 2 g ′ g-2g' g − 2 g ′ , g ′ − 2 g'-2 g ′ − 2 , 1 1 1 , 2 2 2 , and removing one point adds two neighbouring gaps). Both bounds are less than ⌈ 4 g / 7 ⌉ \lceil4g/7\rceil ⌈ 4 g /7 ⌉ , since g ′ < g / 2 < 4 g / 7 g'<g/2<4g/7 g ′ < g /2 < 4 g /7 , ⌈ 4 g / 7 ⌉ ≥ 3 \lceil4g/7\rceil\ge3 ⌈ 4 g /7 ⌉ ≥ 3 for g ≥ 4 g\ge4 g ≥ 4 and ⌈ 4 g / 7 ⌉ ≥ 4 \lceil4g/7\rceil\ge4 ⌈ 4 g /7 ⌉ ≥ 4 for g ≥ 6 g\ge6 g ≥ 6 . As 7 L ≤ 3 n 7L\le3n 7 L ≤ 3 n , Υ ≤ ⌊ 3 g / 7 ⌋ + 1 \Upsilon\le\lfloor3g/7\rfloor+1 Υ ≤ ⌊ 3 g /7 ⌋ + 1 . If no element with index in Π \Pi Π were in S q S_q S q , then by Lemma B.7 (ii) Π \Pi Π would lie in a set of at most Υ \Upsilon Υ cyclically consecutive indices, and the at least g − Υ g-\Upsilon g − Υ indices outside it would lie between two cyclically consecutive points of Π \Pi Π , which would have a gap at least g − Υ + 1 ≥ g − ⌊ 3 g / 7 ⌋ = ⌈ 4 g / 7 ⌉ g-\Upsilon+1\ge g-\lfloor3g/7\rfloor=\lceil4g/7\rceil g − Υ + 1 ≥ g − ⌊ 3 g /7 ⌋ = ⌈ 4 g /7 ⌉ . This is a contradiction. □
Lemma B.9. Suppose that 3 3 3 divides r r r and n n n , that 9 ∤ n 9\nmid n 9 ∤ n , 3 s < n 3s<n 3 s < n and 3 ( L − 1 ) < 2 n 3(L-1)<2n 3 ( L − 1 ) < 2 n . Let a 0 a_0 a 0 be a residue with gcd ( a 0 , n / 3 ) = 1 \gcd(a_0,n/3)=1 g cd( a 0 , n /3 ) = 1 and X = [ a 0 r ] n X=[a_0r]_n X = [ a 0 r ] n , 1 ≤ X < s 1\le X<s 1 ≤ X < s , 2 X ≠ s 2X\ne s 2 X = s , and let R R R be its row for g = 3 g=3 g = 3 . If no element of R R R is a witness, then R R R contains exactly two units, and the smaller value Y Y Y of [ − a q ] n [-a_q]_n [ − a q ] n on them satisfies 1 ≤ Y ≤ X − 1 1\le Y\le X-1 1 ≤ Y ≤ X − 1 and Y ≤ L − n / 3 Y\le L-n/3 Y ≤ L − n /3 , and L X − s Y LX-sY L X − s Y is a multiple of n n n with n ≤ L X − s Y ≤ L X − s n\le LX-sY\le LX-s n ≤ L X − s Y ≤ L X − s .
Proof. As 3 ∣ r 3\mid r 3 ∣ r , every a ∈ R a\in R a ∈ R has a r ≡ a 0 r ar\equiv a_0r a r ≡ a 0 r , so [ a r ] n = X < s [ar]_n=X<s [ a r ] n = X < s and R ⊂ S r R\subset S_r R ⊂ S r . The divisor g = 3 g=3 g = 3 satisfies the hypotheses of Lemma B.7 (n / 3 n/3 n /3 is even if n n n is, and gcd ( q , 3 ) = 1 \gcd(q,3)=1 g cd( q , 3 ) = 1 as 3 3 3 divides r r r and n n n ), and g ∗ = 3 g^*=3 g ∗ = 3 as 9 ∤ n 9\nmid n 9 ∤ n ; by Lemma B.7 (iii) exactly one element of R R R is divisible by 3 3 3 , and the other two are units. They are usable by Lemma B.1 , as X ≠ s / 2 X\ne s/2 X = s /2 . They are not witnesses, so they are not in S q S_q S q , and their values Y ≤ Y ′ Y\le Y' Y ≤ Y ′ lie in [ 1 , L ] [1,L] [ 1 , L ] (Lemma B.3 (a)) and differ by ± n / 3 \pm n/3 ± n /3 modulo n n n (Lemma B.7 (i)). As Y ′ − Y ≤ L − 1 < 2 n / 3 Y'-Y\le L-1<2n/3 Y ′ − Y ≤ L − 1 < 2 n /3 , Y ′ − Y = n / 3 Y'-Y=n/3 Y ′ − Y = n /3 , so Y ≤ L − n / 3 = 2 p + s − n / 3 < 2 p Y\le L-n/3=2p+s-n/3<2p Y ≤ L − n /3 = 2 p + s − n /3 < 2 p . By Lemma B.3 (c) with x = r x=r x = r , τ = s \tau=s τ = s , τ ′ = 2 p \tau'=2p τ ′ = 2 p , Y ∈ [ 1 , X − 1 ] ∪ [ X + 2 p , L ] Y\in[1,X-1]\cup[X+2p,L] Y ∈ [ 1 , X − 1 ] ∪ [ X + 2 p , L ] , so Y ≤ X − 1 Y\le X-1 Y ≤ X − 1 , and Lemma B.3 (d) gives the rest. □
Lemma B.10. Let g ≥ 3 g\ge3 g ≥ 3 divide p p p and n n n , with n / g n/g n / g even if n n n is even; let a i = 1 + ( n / g ) i a_i=1+(n/g)i a i = 1 + ( n / g ) i for i ∈ Z / g i\in\mathbb{Z}/g i ∈ Z / g , t = [ i q ] g t=[iq]_g t = [ i q ] g , and g ∗ g^* g ∗ as in Lemma B.7 . Then [ a i p ] n = p [a_i p]_n=p [ a i p ] n = p , so a i ∈ S p a_i\in S_p a i ∈ S p ; a i ∈ S q a_i\in S_q a i ∈ S q if and only if ( n / g ) t < q (n/g)t<q ( n / g ) t < q or ( n / g ) t ≥ n − q (n/g)t\ge n-q ( n / g ) t ≥ n − q ; and a unit a i a_i a i is usable if and only if g ∤ 2 i g\nmid2i g ∤ 2 i .
(a) If n / g < q n/g<q n / g < q and g ∗ g^* g ∗ is 1 1 1 or a prime, some a i a_i a i is a usable unit in S p ∩ S q S_p\cap S_q S p ∩ S q .
(b) If t max = max { t ≥ 0 : ( n / g ) t < q } ≥ J ( g ∗ ) t_{\max}=\max\{t\ge0:(n/g)t<q\}\ge J(g^*) t m a x = max { t ≥ 0 : ( n / g ) t < q } ≥ J ( g ∗ ) , some a i a_i a i is a usable unit in S p ∩ S q S_p\cap S_q S p ∩ S q .
Proof. a i p ≡ p ( m o d n ) a_i p\equiv p\pmod n a i p ≡ p ( mod n ) as g ∣ p g\mid p g ∣ p . By Lemma B.7 , gcd ( q , g ) = 1 \gcd(q,g)=1 g cd( q , g ) = 1 , so i ↦ t i\mapsto t i ↦ t is a bijection of Z / g \mathbb{Z}/g Z / g , and a i q ≡ q + ( n / g ) t = v ∈ [ q , n + q ) a_iq\equiv q+(n/g)t=v\in[q,n+q) a i q ≡ q + ( n / g ) t = v ∈ [ q , n + q ) . If v < n v<n v < n , [ a i q ] n = v < 2 q [a_iq]_n=v<2q [ a i q ] n = v < 2 q if and only if ( n / g ) t < q (n/g)t<q ( n / g ) t < q ; if v ≥ n v\ge n v ≥ n , [ a i q ] n = v − n < q [a_iq]_n=v-n<q [ a i q ] n = v − n < q . Next, 2 a i ≡ 2 2a_i\equiv2 2 a i ≡ 2 if and only if n ∣ ( 2 n / g ) i n\mid(2n/g)i n ∣ ( 2 n / g ) i , that is g ∣ 2 i g\mid2i g ∣ 2 i . The row of 1 1 1 satisfies the hypotheses of Lemma B.7 , so by its part (iii), a i a_i a i is a unit if and only if i i i , equivalently t t t , avoids one class modulo ℓ \ell ℓ for each prime ℓ ∣ g ∗ \ell\mid g^* ℓ ∣ g ∗ , and that class is not 0 0 0 , as a 0 = 1 a_0=1 a 0 = 1 . (a) t = 1 t=1 t = 1 and t = g − 1 t=g-1 t = g − 1 both give elements of S q S_q S q , as n / g < q n/g<q n / g < q and ( n / g ) ( g − 1 ) = n − n / g > n − q (n/g)(g-1)=n-n/g>n-q ( n / g ) ( g − 1 ) = n − n / g > n − q ; they are i = ± q − 1 ( m o d g ) i=\pm q^{-1}\pmod g i = ± q − 1 ( mod g ) , and g ∤ 2 i g\nmid2i g ∤ 2 i as g ≥ 3 g\ge3 g ≥ 3 and gcd ( i , g ) = 1 \gcd(i,g)=1 g cd( i , g ) = 1 . If g ∗ = 1 g^*=1 g ∗ = 1 both are units; if g ∗ = ℓ g^*=\ell g ∗ = ℓ is a prime, i i i and − i -i − i lie in different classes modulo ℓ \ell ℓ , as ℓ \ell ℓ is odd and ℓ ∤ i \ell\nmid i ℓ ∤ i , so one of them avoids the excluded class. (b) t = 1 , … , t max t=1,\ldots,t_{\max} t = 1 , … , t m a x give elements of S q S_q S q ; they are at least J ( g ∗ ) J(g^*) J ( g ∗ ) consecutive values, so, as in Lemma B.7 (iii), one of them gives a unit; and 0 < 2 t < g 0 < 2t < g 0 < 2 t < g , because ( n / g ) t < q < n / 2 (n/g)t < q < n/2 ( n / g ) t < q < n /2 , so g ∤ 2 t g \nmid2t g ∤ 2 t .
Even denominators
In this subsection n = 2 m n = 2m n = 2 m . We keep the notation and the inequalities stated at the start of § 4.3 : Δ = ( p , q , r ) / n \Delta= (p,q,r)/n Δ = ( p , q , r ) / n is a window triangle unless said otherwise, ς = s / 2 = r − m \varsigma= s/2 = r-m ς = s /2 = r − m , L = n − 2 q = 2 p + 2 ς L = n-2q = 2p+2\varsigma L = n − 2 q = 2 p + 2 ς , J = J ( m ∘ ) J = J(m^\circ) J = J ( m ∘ ) , and P ( m ) < P 2 ( m ) P(m) < P_2(m) P ( m ) < P 2 ( m ) are the two least odd primes not dividing m m m .
A unit modulo n n n is an odd a a a with gcd ( a , m ) = 1 \gcd(a,m)=1 g cd( a , m ) = 1 ; by Lemma B.1 it is usable unless a ≡ 1 ( m o d m ) a \equiv1 \pmod m a ≡ 1 ( mod m ) , that is unless a = 1 a=1 a = 1 or (m m m even and a = 1 + m a=1+m a = 1 + m ). For odd a a a , a m ≡ m ( m o d n ) am \equiv m \pmod n am ≡ m ( mod n ) , so a r ≡ m + a ς ar \equiv m+a\varsigma a r ≡ m + a ς and
a ∈ S r ⟺ [ a ς ] n ∈ [ m , m + 2 ς ) . (5) a \in S_r \Longleftrightarrow[a\varsigma]_n \in[m,m+2\varsigma). \tag*{(5)} a ∈ S r ⟺ [ a ς ] n ∈ [ m , m + 2 ς ) . (5)
Corollary B.11. If m ≥ 74 J 2 m \ge74J^2 m ≥ 74 J 2 and Δ \Delta Δ has no witness, then p ≤ 7 J 2 − 1 p \le7J^2-1 p ≤ 7 J 2 − 1 .
Proof. By Lemma B.4, with ϵ = 2 \epsilon=2 ϵ = 2 , p q ≤ 4 m J 2 pq \le4mJ^2 pq ≤ 4 m J 2 . Now q = m − ς − p ≥ 2 m / 3 − p q=m-\varsigma-p \ge2m/3-p q = m − ς − p ≥ 2 m /3 − p , and p ↦ p ( m − ς − p ) p \mapsto p(m-\varsigma-p) p ↦ p ( m − ς − p ) is increasing for p ≤ ( m − ς ) / 2 p \le(m-\varsigma)/2 p ≤ ( m − ς ) /2 , which holds as p ≤ q p \le q p ≤ q . So for p ≥ 7 J 2 p \ge7J^2 p ≥ 7 J 2 , p q ≥ 7 J 2 ( 2 m / 3 − 7 J 2 ) pq \ge7J^2(2m/3-7J^2) pq ≥ 7 J 2 ( 2 m /3 − 7 J 2 ) , which exceeds 4 m J 2 4mJ^2 4 m J 2 when m > 73.5 J 2 m>73.5J^2 m > 73.5 J 2 .
Lemma B.12. If 2 ς ( p + ς ) < m 2\varsigma(p+\varsigma)<m 2 ς ( p + ς ) < m , every odd a ∈ S r a \in S_r a ∈ S r lies in S p ∪ S q S_p \cup S_q S p ∪ S q . If 2 p ( p + ς ) < m 2p(p+\varsigma)<m 2 p ( p + ς ) < m , every odd a ∈ S p a \in S_p a ∈ S p lies in S q ∪ S r S_q \cup S_r S q ∪ S r .
Proof. Let a a a be odd in S r S_r S r , c = [ a ς ] n − m ∈ [ 0 , 2 ς ) c=[a\varsigma]_n-m \in[0,2\varsigma) c = [ a ς ] n − m ∈ [ 0 , 2 ς ) by (B.1), so that [ a r ] n = c [ar]_n=c [ a r ] n = c , and x = [ a p ] n x=[ap]_n x = [ a p ] n . Suppose a ∉ S p ∪ S q a \notin S_p \cup S_q a ∈ / S p ∪ S q . By Lemma B.2 with ( u , v , w ) = ( r , p , q ) (u,v,w)=(r,p,q) ( u , v , w ) = ( r , p , q ) , either 2 p ≤ x ≤ 2 p + 2 ς − c 2p \le x \le2p+2\varsigma-c 2 p ≤ x ≤ 2 p + 2 ς − c or n − c < x < n n-c<x<n n − c < x < n . Multiplying a ς ≡ m + c a\varsigma\equiv m+c a ς ≡ m + c by p p p and a p ≡ x ap \equiv x a p ≡ x by ς \varsigma ς gives ς x ≡ p m + p c ( m o d 2 m ) \varsigma x \equiv pm+pc \pmod{2m} ς x ≡ p m + p c ( mod 2 m ) , so ς x − p c ≡ 0 \varsigma x-pc \equiv0 ς x − p c ≡ 0 or m ( m o d 2 m ) m \pmod{2m} m ( mod 2 m ) . In the first case 0 < p ( 2 ς − c ) ≤ ς x − p c ≤ ς ( 2 p + 2 ς ) < m 0<p(2\varsigma-c)\le\varsigma x-pc\le\varsigma(2p+2\varsigma)<m 0 < p ( 2 ς − c ) ≤ ς x − p c ≤ ς ( 2 p + 2 ς ) < m , impossible. In the second x = n − j x=n-j x = n − j with 1 ≤ j ≤ c − 1 1\le j\le c-1 1 ≤ j ≤ c − 1 , and ς x − p c ≡ − ( j ς + p c ) \varsigma x-pc\equiv-(j\varsigma+pc) ς x − p c ≡ − ( j ς + p c ) with 0 < j ς + p c < 2 ς ( ς + p ) < m 0<j\varsigma+pc<2\varsigma(\varsigma+p)<m 0 < j ς + p c < 2 ς ( ς + p ) < m , impossible.
For the second statement let a a a be odd, x = [ a p ] n < 2 p x=[ap]_n<2p x = [ a p ] n < 2 p , y = [ a r ] n y=[ar]_n y = [ a r ] n , and suppose a ∉ S q ∪ S r a \notin S_q \cup S_r a ∈ / S q ∪ S r . By Lemma B.2 with ( u , v , w ) = ( p , r , q ) (u,v,w)=(p,r,q) ( u , v , w ) = ( p , r , q ) , 2 ς ≤ y ≤ 2 ς + 2 p − x 2\varsigma\le y\le2\varsigma+2p-x 2 ς ≤ y ≤ 2 ς + 2 p − x or n − x < y < n n-x<y<n n − x < y < n . From y ≡ m + a ς y\equiv m+a\varsigma y ≡ m + a ς , p y − ς x ≡ p m ( m o d 2 m ) py-\varsigma x\equiv pm \pmod{2m} p y − ς x ≡ p m ( mod 2 m ) , which is 0 0 0 or m m m modulo 2 m 2m 2 m . In the first case 0 < ς ( 2 p − x ) ≤ p y − ς x ≤ p ( 2 ς + 2 p ) < m 0<\varsigma(2p-x)\le py-\varsigma x\le p(2\varsigma+2p)<m 0 < ς ( 2 p − x ) ≤ p y − ς x ≤ p ( 2 ς + 2 p ) < m ; in the second y = n − j y=n-j y = n − j with 1 ≤ j < x 1\le j<x 1 ≤ j < x and p y − ς x ≡ − ( p j + ς x ) py-\varsigma x\equiv-(pj+\varsigma x) p y − ς x ≡ − ( p j + ς x ) with 0 < p j + ς x < 2 p ( p + ς ) < m 0<pj+\varsigma x<2p(p+\varsigma)<m 0 < p j + ς x < 2 p ( p + ς ) < m . Both are impossible.
Proposition B.13. Let g ς = gcd ( ς , m ) g_\varsigma=\gcd(\varsigma,m) g ς = g cd( ς , m ) , ς = g ς ς ′ \varsigma=g_\varsigma\varsigma' ς = g ς ς ′ , m = g ς m ′ m=g_\varsigma m' m = g ς m ′ , and C 0 ′ = { c ′ ∈ Z : 0 ≤ c ′ < 2 ς ′ , c ′ ≡ m ′ + ς ′ ( m o d 2 ) } C'_0=\{c'\in\mathbb{Z}:0\le c'<2\varsigma',\ c'\equiv m'+\varsigma'\pmod2\} C 0 ′ = { c ′ ∈ Z : 0 ≤ c ′ < 2 ς ′ , c ′ ≡ m ′ + ς ′ ( mod 2 )} .
(a) For each c ′ ∈ C 0 ′ c'\in C'_0 c ′ ∈ C 0 ′ there is exactly one odd residue β ( c ′ ) \beta(c') β ( c ′ ) modulo 2 m ′ 2m' 2 m ′ with β ( c ′ ) ς ′ ≡ m ′ + c ′ ( m o d 2 m ′ ) \beta(c')\varsigma'\equiv m'+c'\pmod{2m'} β ( c ′ ) ς ′ ≡ m ′ + c ′ ( mod 2 m ′ ) . The odd elements of S r S_r S r are exactly the a a a with a ≡ β ( c ′ ) ( m o d 2 m ′ ) a\equiv\beta(c')\pmod{2m'} a ≡ β ( c ′ ) ( mod 2 m ′ ) for some c ′ ∈ C 0 ′ c'\in C'_0 c ′ ∈ C 0 ′ , and then [ a ς ] n = m + g ς c ′ [a\varsigma]_n=m+g_\varsigma c' [ a ς ] n = m + g ς c ′ . For such an a a a , gcd ( a , m / g ς ) = 1 \gcd(a,m/g_\varsigma)=1 g cd( a , m / g ς ) = 1 if and only if gcd ( c ′ , m ′ ) = 1 \gcd(c',m')=1 g cd( c ′ , m ′ ) = 1 ; so a a a is a unit if and only if gcd ( c ′ , m ′ ) = 1 \gcd(c',m')=1 g cd( c ′ , m ′ ) = 1 and gcd ( a , g ς ) = 1 \gcd(a,g_\varsigma)=1 g cd( a , g ς ) = 1 ; and a ≡ 1 ( m o d m ) a\equiv1\pmod m a ≡ 1 ( mod m ) is possible only for c ′ = ς ′ c'=\varsigma' c ′ = ς ′ .
(b) S r S_r S r contains no usable unit if and only if ς ∣ m \varsigma\mid m ς ∣ m and m / ς m/\varsigma m / ς is odd, or ς = 1 \varsigma=1 ς = 1 and m m m is even.
Proof. (a) By (B.1) an odd a a a is in S r S_r S r if and only if a ς ≡ m + c ( m o d 2 m ) a\varsigma\equiv m+c\pmod{2m} a ς ≡ m + c ( mod 2 m ) with $0\le c<2\varsigma$. Reducing modulo g ς g_\varsigma g ς gives g ς ∣ c g_\varsigma\mid c g ς ∣ c , and with c = g ς c ′ c=g_\varsigma c' c = g ς c ′ the congruence is a ς ′ ≡ m ′ + c ′ ( m o d 2 m ′ ) a\varsigma'\equiv m'+c'\pmod{2m'} a ς ′ ≡ m ′ + c ′ ( mod 2 m ′ ) , 0 ≤ c ′ < 2 ς ′ 0\le c'<2\varsigma' 0 ≤ c ′ < 2 ς ′ , where gcd ( ς ′ , m ′ ) = 1 \gcd(\varsigma',m')=1 g cd( ς ′ , m ′ ) = 1 . If an odd a a a solves it, then c ′ ≡ m ′ + ς ′ ( m o d 2 ) c'\equiv m'+\varsigma'\pmod2 c ′ ≡ m ′ + ς ′ ( mod 2 ) . Conversely let c ′ ≡ m ′ + ς ′ ( m o d 2 ) c'\equiv m'+\varsigma'\pmod2 c ′ ≡ m ′ + ς ′ ( mod 2 ) . If m ′ m' m ′ is odd, a ≡ c ′ ς ′ − 1 ( m o d m ′ ) a\equiv c'\varsigma'^{-1}\pmod{m'} a ≡ c ′ ς ′ − 1 ( mod m ′ ) has exactly one odd lift modulo 2 m ′ 2m' 2 m ′ , and the congruence holds modulo m ′ m' m ′ and modulo 2 2 2 . If m ′ m' m ′ is even, ς ′ \varsigma' ς ′ is odd and a unit modulo 2 m ′ 2m' 2 m ′ , the only solution is ( m ′ + c ′ ) ς ′ − 1 (m'+c')\varsigma'^{-1} ( m ′ + c ′ ) ς ′ − 1 , and it is odd as m ′ + c ′ m'+c' m ′ + c ′ is. For a prime ℓ ∣ m ′ \ell\mid m' ℓ ∣ m ′ , ℓ ∣ a \ell\mid a ℓ ∣ a if and only if ℓ ∣ m ′ + c ′ \ell\mid m'+c' ℓ ∣ m ′ + c ′ , that is ℓ ∣ c ′ \ell\mid c' ℓ ∣ c ′ (for ℓ = 2 \ell=2 ℓ = 2 both are false, c ′ c' c ′ being odd); the primes of g ς g_\varsigma g ς not dividing m ′ m' m ′ are accounted for by $\gcd(a,g_\varsigma)=1. Finally a ≡ 1 ( m o d m ) a \equiv1 \pmod{m} a ≡ 1 ( mod m ) gives g c ′ ≡ a ς ≡ ς = g ς ′ ( m o d g m ′ ) gc' \equiv a\varsigma\equiv\varsigma= g\varsigma' \pmod{gm'} g c ′ ≡ a ς ≡ ς = g ς ′ ( mod g m ′ ) , so c ′ ≡ ς ′ ( m o d m ′ ) c' \equiv\varsigma' \pmod{m'} c ′ ≡ ς ′ ( mod m ′ ) , and ∣ c ′ − ς ′ ∣ ≤ ς ′ < m ′ |c'-\varsigma'| \le\varsigma' < m' ∣ c ′ − ς ′ ∣ ≤ ς ′ < m ′ .
(b) ς ′ = 1 \varsigma'=1 ς ′ = 1 , m ′ m' m ′ odd. C 0 ′ = { 0 } C'_0=\{0\} C 0 ′ = { 0 } and gcd ( 0 , m ′ ) = m ′ ≥ 3 \gcd(0,m')=m'\ge3 g cd( 0 , m ′ ) = m ′ ≥ 3 : no unit at all. ς ′ = 1 \varsigma'=1 ς ′ = 1 , m ′ m' m ′ even. C 0 ′ = { 1 } C'_0=\{1\} C 0 ′ = { 1 } and β ( 1 ) = m ′ + 1 \beta(1)=m'+1 β ( 1 ) = m ′ + 1 , so the odd elements of S r S_r S r are the 1 + m ′ j 1+m'j 1 + m ′ j with j j j odd modulo 2 g s 2g_s 2 g s . If g s = 1 g_s=1 g s = 1 the only one is 1 + m 1+m 1 + m , which is not usable. If g s ≥ 2 g_s\ge2 g s ≥ 2 , choose an odd j j j with m ′ j ≡ 1 ( m o d ℓ ) m'j\equiv1\pmod{\ell} m ′ j ≡ 1 ( mod ℓ ) for every odd prime ℓ ∣ g s \ell\mid g_s ℓ ∣ g s with ℓ ∤ m ′ \ell\nmid m' ℓ ∤ m ′ (j = 1 j=1 j = 1 if there is none). Then gcd ( 1 + m ′ j , g s ) = 1 \gcd(1+m'j,g_s)=1 g cd( 1 + m ′ j , g s ) = 1 (1 + m ′ j 1+m'j 1 + m ′ j is ≡ 1 ( m o d ℓ ) \equiv1\pmod{\ell} ≡ 1 ( mod ℓ ) if ℓ ∣ m ′ \ell\mid m' ℓ ∣ m ′ and ≡ 2 \equiv2 ≡ 2 otherwise, and 2 ∣ g s 2\mid g_s 2 ∣ g s implies 2 ∣ m ′ 2\mid m' 2 ∣ m ′ here), so a = 1 + m ′ j a=1+m'j a = 1 + m ′ j is a unit; and a ≡ 1 ( m o d m ) a\equiv1\pmod m a ≡ 1 ( mod m ) would need g s ∣ j g_s\mid j g s ∣ j , which fails (if some ℓ ∣ g s \ell\mid g_s ℓ ∣ g s has ℓ ∤ m ′ \ell\nmid m' ℓ ∤ m ′ , then ℓ ∤ j \ell\nmid j ℓ ∤ j ; otherwise j = 1 < g s j=1<g_s j = 1 < g s ). So a a a is a usable unit. ς ′ ≥ 2 \varsigma'\ge2 ς ′ ≥ 2 . Set c ′ = 1 c'=1 c ′ = 1 if m ′ + ς ′ m'+\varsigma' m ′ + ς ′ is odd and c ′ = 2 c'=2 c ′ = 2 otherwise (then m ′ m' m ′ and c ′ c' c ′ are both odd, being coprime). In both cases c ′ ∈ C 0 ′ c'\in C'_0 c ′ ∈ C 0 ′ , gcd ( c ′ , m ′ ) = 1 \gcd(c',m')=1 g cd( c ′ , m ′ ) = 1 and c ′ ≠ ς ′ c'\ne\varsigma' c ′ = ς ′ . Among the g s g_s g s lifts a = β ( c ′ ) + 2 m ′ i a=\beta(c')+2m'i a = β ( c ′ ) + 2 m ′ i choose i i i with ℓ ∤ a \ell\nmid a ℓ ∤ a for every prime ℓ ∣ g s \ell\mid g_s ℓ ∣ g s with ℓ ∤ m ′ \ell\nmid m' ℓ ∤ m ′ (one class modulo the odd prime ℓ \ell ℓ to avoid); for ℓ ∣ g s \ell\mid g_s ℓ ∣ g s with ℓ ∣ m ′ \ell\mid m' ℓ ∣ m ′ , ℓ ∤ a \ell\nmid a ℓ ∤ a already. By (a), a a a is a unit and a ≠ 1 ( m o d m ) a\ne1\pmod m a = 1 ( mod m ) . □
The case ς ′ = 1 \varsigma'=1 ς ′ = 1 , m ′ m' m ′ odd of (b) is Proposition 4.3(a). The corresponding statement for S p S_p S p is [17 Proposition 3.5] .
Lemma B.14. Let g = gcd ( p , m ) g=\gcd(p,m) g = g cd( p , m ) , p = g p ′ ′ p=gp'' p = g p ′′ , m = g m ′ ′ m=gm'' m = g m ′′ , and C 0 ′ ′ = { x ∈ Z : 0 ≤ x < 2 p ′ ′ , x ≡ p ′ ′ ( m o d 2 ) } C''_0=\{x\in\mathbb{Z}:0\le x<2p'',\ x\equiv p''\pmod2\} C 0 ′′ = { x ∈ Z : 0 ≤ x < 2 p ′′ , x ≡ p ′′ ( mod 2 )} . For each x ∈ C 0 ′ ′ x\in C''_0 x ∈ C 0 ′′ there is exactly one odd residue γ ( x ) \gamma(x) γ ( x ) modulo 2 m ′ ′ 2m'' 2 m ′′ with γ ( x ) p ′ ′ ≡ x ( m o d 2 m ′ ′ ) \gamma(x)p''\equiv x\pmod{2m''} γ ( x ) p ′′ ≡ x ( mod 2 m ′′ ) . The odd elements of S p S_p S p are exactly the a a a with a ≡ γ ( x ) ( m o d 2 m ′ ′ ) a\equiv\gamma(x)\pmod{2m''} a ≡ γ ( x ) ( mod 2 m ′′ ) for some x ∈ C 0 ′ ′ x\in C''_0 x ∈ C 0 ′′ , and then [ a p ] n = g x [ap]_n=gx [ a p ] n = g x . Such an a a a is a unit if and only if gcd ( x , m ′ ′ ) = 1 \gcd(x,m'')=1 g cd( x , m ′′ ) = 1 and gcd ( a , g ) = 1 \gcd(a,g)=1 g cd( a , g ) = 1 ; and a ≡ 1 ( m o d m ) a\equiv1\pmod m a ≡ 1 ( mod m ) is possible only for x = p ′ ′ x=p'' x = p ′′ .
Proof. An odd a a a is in S p S_p S p if and only if a p ≡ X ( m o d 2 m ) ap\equiv X\pmod{2m} a p ≡ X ( mod 2 m ) with 0 ≤ X < 2 p 0\le X<2p 0 ≤ X < 2 p . Reducing modulo g g g gives g ∣ X g\mid X g ∣ X , and with X = g x X=gx X = g x the congruence is a p ′ ′ ≡ x ( m o d 2 m ′ ′ ) ap''\equiv x\pmod{2m''} a p ′′ ≡ x ( mod 2 m ′′ ) , 0 ≤ x < 2 p ′ ′ 0\le x<2p'' 0 ≤ x < 2 p ′′ , where gcd ( p ′ ′ , m ′ ′ ) = 1 \gcd(p'',m'')=1 g cd( p ′′ , m ′′ ) = 1 . If an odd a a a solves it, then x ≡ p ′ ′ ( m o d 2 ) x\equiv p''\pmod2 x ≡ p ′′ ( mod 2 ) . Conversely let x ≡ p ′ ′ ( m o d 2 ) x\equiv p''\pmod2 x ≡ p ′′ ( mod 2 ) . If m ′ ′ m'' m ′′ is odd, a ≡ x p ′ ′ − 1 ( m o d m ′ ′ ) a\equiv xp''^{-1}\pmod{m''} a ≡ x p ′′ − 1 ( mod m ′′ ) has exactly one odd lift modulo 2 m ′ ′ 2m'' 2 m ′′ , and the congruence holds modulo m ′ ′ m'' m ′′ and modulo 2 2 2 . If m ′ ′ m'' m ′′ is even, p ′ ′ p'' p ′′ is odd and a a a a unit modulo 2 m ′ ′ 2m'' 2 m ′′ , the only solution is x p ′ ′ − 1 xp''^{-1} x p ′′ − 1 , and it is odd as x x x is. For a prime ℓ ∣ m ′ ′ \ell\mid m'' ℓ ∣ m ′′ , ℓ ∣ a \ell\mid a ℓ ∣ a if and only if ℓ ∣ x \ell\mid x ℓ ∣ x (for ℓ = 2 \ell=2 ℓ = 2 both are false); the primes of g g g not dividing m ′ ′ m'' m ′′ are accounted for by gcd ( a , g ) = 1 \gcd(a,g)=1 g cd( a , g ) = 1 . Finally a ≡ 1 ( m o d m ) a\equiv1\pmod m a ≡ 1 ( mod m ) gives g x ≡ a p ≡ p = g p ′ ′ ( m o d g m ′ ′ ) gx\equiv ap\equiv p=gp''\pmod{gm''} g x ≡ a p ≡ p = g p ′′ ( mod g m ′′ ) , so x ≡ p ′ ′ ( m o d m ′ ′ ) x\equiv p''\pmod{m''} x ≡ p ′′ ( mod m ′′ ) , and ∣ x − p ′ ′ ∣ ≤ p ′ ′ < m ′ ′ |x-p''|\le p''<m'' ∣ x − p ′′ ∣ ≤ p ′′ < m ′′ as p < m / 2 p<m/2 p < m /2 . □
A row of S p S_p S p is the set of the g g g residues a ≡ γ ( x ) ( m o d 2 m ′ ′ ) a\equiv\gamma(x)\pmod{2m''} a ≡ γ ( x ) ( mod 2 m ′′ ) for one x ∈ C 0 ′ ′ x\in C''_0 x ∈ C 0 ′′ , and a row of S r S_r S r the set of the g s g_s g s residues a ≡ β ( c ′ ) ( m o d 2 m ′ ) a\equiv\beta(c')\pmod{2m'} a ≡ β ( c ′ ) ( mod 2 m ′ ) for one c ′ ∈ C 0 ′ c'\in C'_0 c ′ ∈ C 0 ′ . In both cases a row is the row { a 0 + ( n / g ) i : i ∈ Z / g } \{a_0+(n/g)i:i\in\mathbb{Z}/g\} { a 0 + ( n / g ) i : i ∈ Z / g } of Lemma B.7 , with g = gcd ( p , m ) g=\gcd(p,m) g = g cd( p , m ) , respectively g = g s g=g_s g = g s , and g g g satisfies its hypotheses: g ∣ m g\mid m g ∣ m , so n / g n/g n / g is even, and g g g divides n n n and one of p , r p,r p , r (r = m + ς r=m+\varsigma r = m + ς ), so gcd ( q , g ) = 1 \gcd(q,g)=1 g cd( q , g ) = 1 . As a 0 a_0 a 0 is odd, gcd ( a 0 , n / g ) = 1 \gcd(a_0,n/g)=1 g cd( a 0 , n / g ) = 1 means gcd ( a 0 , m / g ) = 1 \gcd(a_0,m/g)=1 g cd( a 0 , m / g ) = 1 : for a row of S r S_r S r this holds if and only if gcd ( c ′ , m ′ ) = 1 \gcd(c',m')=1 g cd( c ′ , m ′ ) = 1 , and then ν = 0 \nu=0 ν = 0 unless c ′ = ς ′ c'=\varsigma' c ′ = ς ′ ; for a row of S p S_p S p , if and only if gcd ( x , m ′ ′ ) = 1 \gcd(x,m'')=1 g cd( x , m ′′ ) = 1 , and then ν = 0 \nu=0 ν = 0 unless x = p ′ ′ x=p'' x = p ′′ (Proposition B.13(a), Lemma B.14 ).
Lemma B.15. Let gcd ( ς , m ) = 1 \gcd(\varsigma,m)=1 g cd( ς , m ) = 1 , r r r odd, 7 L ≤ 3 n 7L\le3n 7 L ≤ 3 n and ς ≥ 24 J 2 + 1 \varsigma\ge24J^2+1 ς ≥ 24 J 2 + 1 . Then Δ \Delta Δ has a witness in S r S_r S r .
Proof. The argument is that of [17 Lemma 7.9] , with q q q and r r r exchanged. Let M = 2 J M=2J M = 2 J . Any M M M consecutive terms of a progression c + i k c+ik c + ik with gcd ( c , k , n ) = 1 \gcd(c,k,n)=1 g cd( c , k , n ) = 1 contain a unit of n n n : if k k k is even, all terms are odd and J J J consecutive ones suffice; if k k k is odd, the odd terms among M M M consecutive ones are J J J consecutive terms of a progression with difference 2 k 2k 2 k . As gcd ( r , n ) = 1 \gcd(r,n)=1 g cd( r , n ) = 1 , let w = [ − q r − 1 ] n w=[-qr^{-1}]_n w = [ − q r − 1 ] n and, for an integer X X X , a X = X r − 1 ( m o d n ) aX=Xr^{-1}\pmod n a X = X r − 1 ( mod n ) ; then [ a X r ] n = [ X ] n [aXr]_n=[X]_n [ a X r ] n = [ X ] n and [ − a X q ] n = [ X w ] n [-aXq]_n=[Xw]_n [ − a X q ] n = [ X w ] n . For 1 ≤ X < ς 1\le X<\varsigma 1 ≤ X < ς with gcd ( X , n ) = 1 \gcd(X,n)=1 g cd( X , n ) = 1 , a X aX a X is a unit of S r S_r S r , usable by Lemma B.1 (as X < ς = s / 2 X<\varsigma=s/2 X < ς = s /2 ), and it lies in S q S_q S q if and only if [ X w ] n ∉ [ 1 , L ] [Xw]_n\notin[1,L] [ X w ] n ∈ / [ 1 , L ] . Suppose a 1 a_1 a 1 is not a witness. Then w ∈ [ 1 , L ] w \in[1,L] w ∈ [ 1 , L ] , and by Lemma B.3 (c), w ≥ 1 + 2 p ≥ 3 w \ge1+2p \ge3 w ≥ 1 + 2 p ≥ 3 . Let α = w / n ≤ L / n ≤ 3 / 7 \alpha= w/n \le L/n \le3/7 α = w / n ≤ L / n ≤ 3/7 and T = 6 M T=6M T = 6 M . By Dirichlet’s theorem there are coprime integers h h h , k k k with 1 ≤ k ≤ T 1 \le k \le T 1 ≤ k ≤ T and ∣ k α − h ∣ < 1 / T |k\alpha-h|<1/T ∣ k α − h ∣ < 1/ T .
k = 1 k=1 k = 1 . h = 1 h=1 h = 1 would give α > 1 − 1 / T > 3 / 7 \alpha>1-1/T>3/7 α > 1 − 1/ T > 3/7 ; so h = 0 h=0 h = 0 and w < n / T w<n/T w < n / T . For L / w < X < n / w L/w<X<n/w L / w < X < n / w we have L < X w < n L<Xw<n L < X w < n , so a X ∈ S q a_X\in S_q a X ∈ S q . The interval ( L / w , min ( n / w , ς ) ) (L/w,\min(n/w,\varsigma)) ( L / w , min ( n / w , ς )) has length more than M M M :
n − L w > 4 7 T > M , \frac{n-L}{w}>\frac{4}{7}T>M, w n − L > 7 4 T > M ,
and ς − L / w ≥ ς − ( 2 p + 2 ς ) / ( 2 p + 1 ) ≥ ς − ( 2 ς + 2 ) / 3 = ( ς − 2 ) / 3 > M \varsigma-L/w\ge\varsigma-(2p+2\varsigma)/(2p+1)\ge\varsigma-(2\varsigma+2)/3=(\varsigma-2)/3>M ς − L / w ≥ ς − ( 2 p + 2 ς ) / ( 2 p + 1 ) ≥ ς − ( 2 ς + 2 ) /3 = ( ς − 2 ) /3 > M . So it contains M M M consecutive integers, one of them a unit X X X of n n n with 1 ≤ X < ς 1\le X<\varsigma 1 ≤ X < ς , and a X a_X a X is a witness.
k = 2 k=2 k = 2 . Then h = 1 h=1 h = 1 and α > 1 / 2 − 1 / ( 2 T ) > 3 / 7 \alpha>1/2-1/(2T)>3/7 α > 1/2 − 1/ ( 2 T ) > 3/7 , which is impossible.
k ≥ 3 k\ge3 k ≥ 3 . Choose an integer e e e prime to k k k : if k α < h k\alpha<h k α < h , e = k − 1 e=k-1 e = k − 1 ; if k α ≥ h k\alpha\ge h k α ≥ h , e = 2 , 3 , 3 , 5 e=2,3,3,5 e = 2 , 3 , 3 , 5 for k = 3 , 4 , 5 , 6 k=3,4,5,6 k = 3 , 4 , 5 , 6 , and for k ≥ 7 k\ge7 k ≥ 7 , e = ( k + 1 ) / 2 e=(k+1)/2 e = ( k + 1 ) /2 (k k k odd), k / 2 + 1 k/2+1 k /2 + 1 (4 ∣ k 4\mid k 4 ∣ k ), k / 2 + 2 k/2+2 k /2 + 2 (k ≡ 2 ( m o d 4 ) k\equiv2\pmod4 k ≡ 2 ( mod 4 ) ). The choices of e e e are prime to k k k : gcd ( ( k + 1 ) / 2 , k ) = 1 \gcd((k+1)/2,k)=1 g cd(( k + 1 ) /2 , k ) = 1 ; gcd ( k / 2 + 1 , k ) \gcd(k/2+1,k) g cd( k /2 + 1 , k ) divides gcd ( k / 2 + 1 , 2 ) = 1 \gcd(k/2+1,2)=1 g cd( k /2 + 1 , 2 ) = 1 for 4 ∣ k 4\mid k 4 ∣ k ; gcd ( k / 2 + 2 , k ) \gcd(k/2+2,k) g cd( k /2 + 2 , k ) divides gcd ( k / 2 + 2 , 4 ) = 1 \gcd(k/2+2,4)=1 g cd( k /2 + 2 , 4 ) = 1 for k ≡ 2 ( m o d 4 ) k\equiv2\pmod4 k ≡ 2 ( mod 4 ) ; and e / k ≤ 1 / 2 + 2 / 7 < 5 / 6 e/k\le1/2+2/7<5/6 e / k ≤ 1/2 + 2/7 < 5/6 for k ≥ 7 k\ge7 k ≥ 7 . In the first case e / k ≥ 2 / 3 ≥ 3 / 7 + 1 / 6 e/k\ge2/3\ge3/7+1/6 e / k ≥ 2/3 ≥ 3/7 + 1/6 ; in the second 3 / 7 < 1 / 2 < e / k ≤ 5 / 6 3/7<1/2<e/k\le5/6 3/7 < 1/2 < e / k ≤ 5/6 . Let c ∈ [ 1 , k − 1 ] c\in[1,k-1] c ∈ [ 1 , k − 1 ] with c h ≡ e ( m o d k ) ch\equiv e\pmod k c h ≡ e ( mod k ) , so gcd ( c , k ) = 1 \gcd(c,k)=1 g cd( c , k ) = 1 , and X i = c + i k X_i=c+ik X i = c + ik for 0 ≤ i ≤ M − 1 0\le i\le M-1 0 ≤ i ≤ M − 1 . Then 1 ≤ X i < M k ≤ 6 M 2 ≤ ς − 1 1\le X_i<Mk\le6M^2\le\varsigma-1 1 ≤ X i < M k ≤ 6 M 2 ≤ ς − 1 , and
X i α = X i h k + X i ( k α − h ) k ≡ e k + X i ( k α − h ) k ( m o d 1 ) X_i\alpha=\frac{X_ih}{k}+\frac{X_i(k\alpha-h)}{k}\equiv\frac{e}{k}+\frac{X_i(k\alpha-h)}{k}\pmod1 X i α = k X i h + k X i ( k α − h ) ≡ k e + k X i ( k α − h ) ( mod 1 )
with ∣ X i ( k α − h ) / k ∣ < M / T = 1 / 6 |X_i(k\alpha-h)/k|<M/T=1/6 ∣ X i ( k α − h ) / k ∣ < M / T = 1/6 . So the fractional part of X i α X_i\alpha X i α is e / k + X i ( k α − h ) / k ∈ ( 3 / 7 , 1 ) e/k+X_i(k\alpha-h)/k\in(3/7,1) e / k + X i ( k α − h ) / k ∈ ( 3/7 , 1 ) in both cases, that is L < [ X i w ] n < n L<[X_iw]_n<n L < [ X i w ] n < n , and a X i ∈ S q a_{X_i}\in S_q a X i ∈ S q . One of the X i X_i X i is a unit of n n n , and a X i a_{X_i} a X i is a witness. □
Lemma B.16. Let m m m and ς \varsigma ς be odd, gcd ( ς , m ) = 1 \gcd(\varsigma,m)=1 g cd( ς , m ) = 1 , ς ≥ 9 \varsigma\ge9 ς ≥ 9 and 8 L ≤ 7 m 8L\le7m 8 L ≤ 7 m . Then S q ∩ S r S_q\cap S_r S q ∩ S r contains a usable unit.
Proof. r = m + ς r=m+\varsigma r = m + ς and p + q = m − ς p+q=m-\varsigma p + q = m − ς are even, so p p p and q q q are odd. For an odd residue a a a let a ∗ = 2 a + m ( m o d n ) a^*=2a+m\pmod n a ∗ = 2 a + m ( mod n ) ; it is odd, gcd ( a ∗ , m ) = gcd ( a , m ) \gcd(a^*,m)=\gcd(a,m) g cd( a ∗ , m ) = g cd( a , m ) , [ a ∗ r ] n = [ 2 a r ] n [a^*r]_n=[2ar]_n [ a ∗ r ] n = [ 2 a r ] n as r r r is even, and [ − a ∗ q ] n = [ 2 ( − a q ) + m ] n [-a^*q]_n=[2(-aq)+m]_n [ − a ∗ q ] n = [ 2 ( − a q ) + m ] n as q q q is odd. As m m m is odd, every unit other than 1 is usable, and 1 ∉ S r 1\notin S_r 1 ∈ / S r . Let a 0 a_0 a 0 be the odd residue with a 0 ς ≡ m + 2 ( m o d 2 m ) a_0\varsigma\equiv m+2\pmod{2m} a 0 ς ≡ m + 2 ( mod 2 m ) (Proposition B.13(a) with g ς = 1 g_\varsigma=1 g ς = 1 , c ′ = 2 c'=2 c ′ = 2 ): a unit with [ a 0 r ] n = 2 [a_0r]_n=2 [ a 0 r ] n = 2 . Set a j + 1 = a j ∗ a_{j+1}=a_j^* a j + 1 = a j ∗ ; then [ a j r ] n = 2 j + 1 < s [a_jr]_n=2^{j+1}<s [ a j r ] n = 2 j + 1 < s for j ≤ 3 j\le3 j ≤ 3 , so a 0 , … , a 3 a_0,\ldots,a_3 a 0 , … , a 3 are units of S r S_r S r . Suppose none is in S q S_q S q : Y j = [ − a j q ] n ∈ [ 1 , L ] Y_j=[-a_jq]_n\in[1,L] Y j = [ − a j q ] n ∈ [ 1 , L ] , with L < m L<m L < m . If Y j < m / 2 Y_j<m/2 Y j < m /2 , then Y j + 1 = 2 Y j + m > L Y_{j+1}=2Y_j+m>L Y j + 1 = 2 Y j + m > L ; so Y j > m / 2 Y_j>m/2 Y j > m /2 and Y j + 1 = 2 Y j − m Y_{j+1}=2Y_j-m Y j + 1 = 2 Y j − m for j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 . Then Y 1 > m / 2 Y_1>m/2 Y 1 > m /2 gives Y 0 > 3 m / 4 Y_0>3m/4 Y 0 > 3 m /4 , and Y 2 = 4 Y 0 − 3 m > m / 2 Y_2=4Y_0-3m>m/2 Y 2 = 4 Y 0 − 3 m > m /2 gives Y 0 > 7 m / 8 ≥ L Y_0>7m/8\ge L Y 0 > 7 m /8 ≥ L , a contradiction. □
Lemma B.17. Let gcd ( ς , m ) = 3 \gcd(\varsigma,m)=3 g cd( ς , m ) = 3 , 9 ∤ m 9\nmid m 9 ∤ m , 3 L > n 3L>n 3 L > n , ς ≥ 9 \varsigma\ge9 ς ≥ 9 , ς > 6 p \varsigma>6p ς > 6 p and 18 p < m 18p<m 18 p < m . Then Δ \Delta Δ has a witness.
Proof. Write m = 3 m ′ m=3m' m = 3 m ′ , ς = 3 ς ′ \varsigma=3\varsigma' ς = 3 ς ′ with 3 ∤ m ′ 3\nmid m' 3 ∤ m ′ , gcd ( ς ′ , m ′ ) = 1 \gcd(\varsigma',m')=1 g cd( ς ′ , m ′ ) = 1 , ς ′ ≥ 3 \varsigma'\ge3 ς ′ ≥ 3 , and f = m − 3 ς f=m-3\varsigma f = m − 3 ς . Here 3 ∣ f 3\mid f 3 ∣ f , and f > 0 f>0 f > 0 , since ς = m / 3 \varsigma=m/3 ς = m /3 would give gcd ( ς , m ) = m / 3 = 3 \gcd(\varsigma,m)=m/3=3 g cd( ς , m ) = m /3 = 3 and ς = 3 \varsigma=3 ς = 3 , so 3 ς < n 3\varsigma<n 3 ς < n . Let c ′ = 1 c'=1 c ′ = 1 if m ′ ± ς ′ m'\pm\varsigma' m ′ ± ς ′ is odd and c ′ = 2 c'=2 c ′ = 2 otherwise (then m ′ m' m ′ is odd); note gcd ( c ′ , m ′ ) = 1 \gcd(c',m')=1 g cd( c ′ , m ′ ) = 1 . Its row is the row of an odd a 0 a_0 a 0 with gcd ( a 0 , n / 3 ) = 1 \gcd(a_0,n/3)=1 g cd( a 0 , n /3 ) = 1 and X = [ a 0 r ] n = 3 c ′ ∈ { 3 , 6 } \mathcal{X}=[a_0r]_n=3c'\in\{3,6\} X = [ a 0 r ] n = 3 c ′ ∈ { 3 , 6 } (Proposition B.13(a) and (B.1)); X < s \mathcal{X}<s X < s and 2 X ≠ s 2\mathcal{X}\ne s 2 X = s , as ς ≥ 9 \varsigma\ge9 ς ≥ 9 . Also 3 ∣ r = m + ς 3\mid r=m+\varsigma 3 ∣ r = m + ς , 9 ∤ n 9\nmid n 9 ∤ n , and L < 2 p + 2 m / 3 < 7 m / 9 L<2p+2m/3<7m/9 L < 2 p + 2 m /3 < 7 m /9 , so 3 ( L − 1 ) < 2 n 3(L-1)<2n 3 ( L − 1 ) < 2 n . Suppose no element of the row is a witness. By Lemma B.9 , L X − s Y L\mathcal{X}-s\mathcal{Y} L X − s Y is a positive multiple of n n n , at most L X − s L\mathcal{X}-s L X − s , for an integer Y ≥ 1 \mathcal{Y}\ge1 Y ≥ 1 . For X = 3 \mathcal{X}=3 X = 3 , L X − s = 6 p + 4 ς < 2 m L\mathcal{X}-s=6p+4\varsigma<2m L X − s = 6 p + 4 ς < 2 m , because 2 ς ≤ 2 m / 3 2\varsigma\le2m/3 2 ς ≤ 2 m /3 and 3 p < m / 3 3p<m/3 3 p < m /3 ; a contradiction. For X = 6 \mathcal{X}=6 X = 6 , L X − s = 12 p + 10 ς < 4 m L\mathcal{X}-s=12p+10\varsigma<4m L X − s = 12 p + 10 ς < 4 m , so L X − s Y = n L\mathcal{X}-s\mathcal{Y}=n L X − s Y = n and Y = ( 6 L − n ) / s = ( 6 p + 6 ς − f ) / s = 3 + ( 6 p − f ) / s \mathcal{Y}=(6L-n)/s=(6p+6\varsigma-f)/s=3+(6p-f)/s Y = ( 6 L − n ) / s = ( 6 p + 6 ς − f ) / s = 3 + ( 6 p − f ) / s . But 3 L > n 3L>n 3 L > n means f < 3 p f<3p f < 3 p , so 0 < 6 p − f < 6 p < s 0<6p-f<6p<s 0 < 6 p − f < 6 p < s , and Y \mathcal{Y} Y is not an integer; a contradiction. □
Proof of Theorem 4.14. As P 2 ≥ 5 P_2\ge5 P 2 ≥ 5 , m ≥ 155 J 2 − 5 ≥ 150 m\ge155J^2-5\ge150 m ≥ 155 J 2 − 5 ≥ 150 ; and J = 1 J=1 J = 1 only for m m m a power of 2, when m ≥ 256 m\ge256 m ≥ 256 . Suppose Δ \Delta Δ has no witness. Step 1: p p p is small. m ≥ 74 J 2 m \ge74J^2 m ≥ 74 J 2 , so p ≤ 7 J 2 − 1 p \le7J^2 - 1 p ≤ 7 J 2 − 1 by Corollary B.11. Hence 21 p < m 21p < m 21 p < m , and 7 L = 14 p + 14 ς ≤ 14 p + 14 m / 3 < 6 m 7L = 14p + 14\varsigma\le14p + 14m/3 < 6m 7 L = 14 p + 14 ς ≤ 14 p + 14 m /3 < 6 m : so 7 L ≤ 3 n 7L \le3n 7 L ≤ 3 n .
Step 2: S r S_r S r has no usable unit. Suppose it has one, and let g ς = gcd ( ς , m ) g_{\varsigma} = \gcd(\varsigma,m) g ς = g cd( ς , m ) , ς = g ς ς ′ \varsigma= g_{\varsigma}\varsigma' ς = g ς ς ′ , m = g ς m ′ m = g_{\varsigma}m' m = g ς m ′ . By Proposition B.13(b) we are not in the cases ( ς ′ = 1 , m ′ odd ) (\varsigma' = 1, m' \text{ odd}) ( ς ′ = 1 , m ′ odd ) and ( ς = 1 , m even ) (\varsigma= 1, m \text{ even}) ( ς = 1 , m even ) .
(2a) ς ′ = 1 \varsigma' = 1 ς ′ = 1 . Then m ′ m' m ′ is even, g ς = ς ≥ 2 g_{\varsigma} = \varsigma\ge2 g ς = ς ≥ 2 , and the only row of S r S_r S r is c ′ = 1 c' = 1 c ′ = 1 , with ν ≤ 1 \nu\le1 ν ≤ 1 , as 1 ∉ S r 1 \notin S_r 1 ∈ / S r . If ς ≤ 8 \varsigma\le8 ς ≤ 8 , then 2 ς ( p + ς ) ≤ 16 ( 7 J 2 + 7 ) < m 2\varsigma(p+\varsigma) \le16(7J^2+7) < m 2 ς ( p + ς ) ≤ 16 ( 7 J 2 + 7 ) < m (for J ≥ 2 J \ge2 J ≥ 2 , 112 J 2 + 112 < 155 J 2 − 5 112J^2+112 < 155J^2-5 112 J 2 + 112 < 155 J 2 − 5 ; for J = 1 J=1 J = 1 , p ≤ 6 p \le6 p ≤ 6 and 224 < 256 224 < 256 224 < 256 ), and by Lemma B.12 the usable unit of S r S_r S r is a witness. If ς ≥ 9 \varsigma\ge9 ς ≥ 9 , Lemma B.8 (g = g ς = ς g = g_{\varsigma} = \varsigma g = g ς = ς , the row c ′ = 1 c'=1 c ′ = 1 , ν ≤ 1 \nu\le1 ν ≤ 1 ) gives a witness.
(2b) ς ′ ≥ 2 \varsigma' \ge2 ς ′ ≥ 2 and g ς ≥ 2 g_{\varsigma} \ge2 g ς ≥ 2 . Take the row c ′ = 1 c'=1 c ′ = 1 if m ′ + ς ′ m' + \varsigma' m ′ + ς ′ is odd and c ′ = 2 c'=2 c ′ = 2 otherwise: gcd ( c ′ , m ′ ) = 1 \gcd(c',m')=1 g cd( c ′ , m ′ ) = 1 and c ′ ≠ ς ′ c' \ne\varsigma' c ′ = ς ′ , so ν = 0 \nu=0 ν = 0 . If g ς ≥ 4 g_{\varsigma} \ge4 g ς ≥ 4 , Lemma B.8 gives a witness. If g ς = 2 g_{\varsigma}=2 g ς = 2 : Υ = 1 \Upsilon=1 Υ = 1 as ( L − 1 ) g ς / n < 6 / 7 (L-1)g_{\varsigma}/n < 6/7 ( L − 1 ) g ς / n < 6/7 , and g ς − Υ = 1 = J ( 1 ) g_{\varsigma}-\Upsilon=1=J(1) g ς − Υ = 1 = J ( 1 ) . If g ς = 3 g_{\varsigma}=3 g ς = 3 and 9 ∣ m 9 \mid m 9 ∣ m : g ς ∗ = 1 g_{\varsigma}^{*}=1 g ς ∗ = 1 , Υ ≤ 2 \Upsilon\le2 Υ ≤ 2 and g ς − Υ ≥ 1 g_{\varsigma}-\Upsilon\ge1 g ς − Υ ≥ 1 . In these two cases Lemma B.7 (iv) gives a witness. If g ς = 3 g_{\varsigma}=3 g ς = 3 and 9 ∤ m 9 \nmid m 9 ∤ m : if 3 L ≤ n 3L \le n 3 L ≤ n , then Υ = 1 \Upsilon=1 Υ = 1 and g ς − Υ = 2 = J ( 3 ) g_{\varsigma}-\Upsilon=2=J(3) g ς − Υ = 2 = J ( 3 ) , and Lemma B.7 (iv) gives a witness; if 3 L > n 3L>n 3 L > n , then ς > m / 3 − p > 6 p \varsigma>m/3-p>6p ς > m /3 − p > 6 p (as 21 p < m 21p<m 21 p < m ), so ς ≥ 9 \varsigma\ge9 ς ≥ 9 , and 18 p < m 18p<m 18 p < m , and Lemma B.17 gives a witness.
(2c) g ς = 1 g_{\varsigma}=1 g ς = 1 and m , ς m,\varsigma m , ς both odd; then ς ≥ 3 \varsigma\ge3 ς ≥ 3 . If ς ≥ 9 \varsigma\ge9 ς ≥ 9 , 8 L ≤ 7 m 8L\le7m 8 L ≤ 7 m follows from 7 L ≤ 3 n 7L\le3n 7 L ≤ 3 n , and Lemma B.16 gives a witness. If ς ≤ 7 \varsigma\le7 ς ≤ 7 , then 2 ς ( p + ς ) ≤ 14 ( 7 J 2 + 6 ) < m 2\varsigma(p+\varsigma)\le14(7J^2+6)<m 2 ς ( p + ς ) ≤ 14 ( 7 J 2 + 6 ) < m (J ≥ 2 J\ge2 J ≥ 2 as m m m is odd), and Lemma B.12 gives one.
(2d) g ς = 1 g_{\varsigma}=1 g ς = 1 and r r r odd, the remaining case; here ς ≥ 2 \varsigma\ge2 ς ≥ 2 , and gcd ( r , n ) = 1 \gcd(r,n)=1 g cd( r , n ) = 1 . By the last sentence of Corollary B.6 , which applies by its part (a), P 2 L ≥ n P_2L\ge n P 2 L ≥ n ; and by Lemma B.15 , ς ≤ 24 J 2 \varsigma\le24J^2 ς ≤ 24 J 2 . So n ≤ P 2 ( 2 p + 2 ς ) ≤ P 2 ( 62 J 2 − 2 ) = 2 N 0 ( m ) n\le P_2(2p+2\varsigma)\le P_2(62J^2-2)=2N_0(m) n ≤ P 2 ( 2 p + 2 ς ) ≤ P 2 ( 62 J 2 − 2 ) = 2 N 0 ( m ) , that is m ≤ N 0 ( m ) m\le N_0(m) m ≤ N 0 ( m ) , a contradiction.
Step 3. So S r S_r S r has no usable unit, and by Proposition B.13(b): (i) ς ∣ m \varsigma\mid m ς ∣ m with d = m / ς d=m/\varsigma d = m / ς odd, or (ii) ς = 1 \varsigma=1 ς = 1 and m m m is even. In case (i) with p = 1 p=1 p = 1 , Δ ∈ A \Delta\in A Δ ∈ A ; in case (i) with 2 ≤ p ≤ ς 2\le p\le\varsigma 2 ≤ p ≤ ς , it lies in B B B by Proposition 4.3(c); in case (ii) with p ∈ { 1 , 2 } p\in\{1,2\} p ∈ { 1 , 2 } , or p = 4 p=4 p = 4 and m ≡ 2 ( m o d 4 ) m\equiv2\pmod{4} m ≡ 2 ( mod 4 ) , Δ ∈ C \Delta\in C Δ ∈ C . In all other cases p > ς p>\varsigma p > ς , and we show that there is a witness in S p ∩ S q S_p\cap S_q S p ∩ S q . Then L = 2 p + 2 ς ≤ 4 p − 2 ≤ 28 J 2 − 6 L=2p+2\varsigma\le4p-2\le28J^2-6 L = 2 p + 2 ς ≤ 4 p − 2 ≤ 28 J 2 − 6 , so 10 L < n 10L<n 10 L < n and q = m − p − ς > 2 m / 3 q=m-p-\varsigma>2m/3 q = m − p − ς > 2 m /3 . The case p = 2 p=2 p = 2 does not occur: in (i), ς = 1 < p \varsigma=1<p ς = 1 < p makes m m m odd and p p p , q = m − 3 q=m-3 q = m − 3 , r = m + 1 r=m+1 r = m + 1 all even; in (ii), p = 2 p=2 p = 2 is in C C C . So p ≥ 3 p\ge3 p ≥ 3 . Let g = gcd ( p , m ) g=\gcd(p,m) g = g cd( p , m ) , p = g p ′ ′ p=gp'' p = g p ′′ , m = g m ′ ′ m=gm'' m = g m ′′ .
(3a) g = 1 g=1 g = 1 . Then p p p is odd: if p p p were even, m m m would be odd, which excludes (ii), and in (i) ς \varsigma ς , a divisor of the odd m m m , would be odd, so r = m + ς r=m+\varsigma r = m + ς and q = m − ς − p q=m-\varsigma-p q = m − ς − p would be even together with p p p . So gcd ( p , n ) = 1 \gcd(p,n)=1 g cd( p , n ) = 1 . Now P 2 L ≤ P 2 ( 28 J 2 − 6 ) < P 2 ( 62 J 2 − 2 ) = 2 N 0 ( m ) < n P_2L\le P_2(28J^2-6)<P_2(62J^2-2)=2N_0(m)<n P 2 L ≤ P 2 ( 28 J 2 − 6 ) < P 2 ( 62 J 2 − 2 ) = 2 N 0 ( m ) < n , against the last sentence of Corollary B.6 , which applies by its part (b).
(3b) g ≥ 2 g\ge2 g ≥ 2 and p ′ ′ ≥ 3 p''\ge3 p ′′ ≥ 3 . Take the row x = 1 x=1 x = 1 if p ′ ′ p'' p ′′ is odd and x = 2 x=2 x = 2 if p ′ ′ p'' p ′′ is even (then m ′ ′ m'' m ′′ is odd): gcd ( x , m ′ ′ ) = 1 \gcd(x,m'')=1 g cd( x , m ′′ ) = 1 and x ≠ p ′ ′ x\ne p'' x = p ′′ , so ν = 0 \nu=0 ν = 0 . If g ≥ 4 g\ge4 g ≥ 4 , Lemma B.8 gives a usable unit of S p S_p S p in S q S_q S q . If g ∈ { 2 , 3 } g\in\{2,3\} g ∈ { 2 , 3 } , Υ = 1 \Upsilon=1 Υ = 1 as 10 L < n 10L<n 10 L < n , and g − Υ = g − 1 ≥ J ( g ∗ ) g-\Upsilon=g-1\ge J(g^*) g − Υ = g − 1 ≥ J ( g ∗ ) ; Lemma B.7 (iv) gives one.
(3c) g ≥ 3 g\ge3 g ≥ 3 and p ′ ′ ≤ 2 p''\le2 p ′′ ≤ 2 . Then n / g ≤ 2 m / 3 < q n/g\le2m/3<q n / g ≤ 2 m /3 < q , and g g g divides p p p and n n n with n / g n/g n / g even. If g ∗ g^* g ∗ is 1 1 1 or a prime, Lemma B.10 (c) gives one. If q ∗ q^* q ∗ has k ≥ 2 k\ge2 k ≥ 2 primes, then q ≥ 15 ⋅ 7 k − 2 q\ge15\cdot7^{k-2} q ≥ 15 ⋅ 7 k − 2 and t max ≥ ⌈ g / 3 ⌉ ≥ 5 ⋅ 7 k − 2 ≥ J ( g ∗ ) t_{\max}\ge\lceil g/3\rceil\ge5\cdot7^{k-2}\ge J(g^*) t m a x ≥ ⌈ g /3 ⌉ ≥ 5 ⋅ 7 k − 2 ≥ J ( g ∗ ) (J ( g ∗ ) ≤ 3 J(g^*)\le3 J ( g ∗ ) ≤ 3 for k = 2 k=2 k = 2 and ≤ 3 k \le3^k ≤ 3 k for k ≥ 3 k\ge3 k ≥ 3 , by Lemma 4.13 ), and Lemma B.10 (b) gives one.
(3d) g = 2 g=2 g = 2 and p ′ ′ ≤ 2 p''\le2 p ′′ ≤ 2 . Then p = 4 p=4 p = 4 , and m ′ ′ = m / 2 m''=m/2 m ′′ = m /2 is odd, so m ≡ 2 ( m o d 4 ) m\equiv2\pmod4 m ≡ 2 ( mod 4 ) . In case (ii) this is C C C . In case (i), m m m is even and m / ς m/\varsigma m / ς odd, so ς \varsigma ς is even, and p p p , q = m − p − ς q=m-p-\varsigma q = m − p − ς , r = m + ς r=m+\varsigma r = m + ς are all even: impossible. □
Appendix C. Odd denominators
This appendix proves Theorem 4.1 (a) for odd n n n (Theorem C.3 ), with the lemmas of Appendix § 9 .§ 1 . As in § 4 , Δ = ( p , q , r ) / n \Delta=(p,q,r)/n Δ = ( p , q , r ) / n is a window triangle, n / 2 < r ≤ 2 n / 3 n/2<r\le2n/3 n /2 < r ≤ 2 n /3 , unless said otherwise, and s = 2 r − n s=2r-n s = 2 r − n , so that 1 ≤ s ≤ n / 3 1\le s\le n/3 1 ≤ s ≤ n /3 .
Let n n n be odd, n ≥ 5 n \ge5 n ≥ 5 . Then s s s is odd, L = n − 2 q = 2 p + s L=n-2q=2p+s L = n − 2 q = 2 p + s , gcd ( s , n ) = gcd ( 2 r , n ) = gcd ( r , n ) \gcd(s,n)=\gcd(2r,n)=\gcd(r,n) g cd( s , n ) = g cd( 2 r , n ) = g cd( r , n ) , and the J J J of Appendix § 9 .§ 1 is J = J ( n ) J=J(n) J = J ( n ) , with J J J Jacobsthal’s function as defined in § 4.3 .
A unit is a unit modulo n n n ; by Lemma B.1 every unit other than 1 is usable, and 1 ∉ S r 1 \notin S_r 1 ∈ / S r . For x ∈ { p , r } x \in\{p,r\} x ∈ { p , r } and g = gcd ( x , n ) g=\gcd(x,n) g = g cd( x , n ) , the residues a a a with a x ≡ g ( m o d n ) ax \equiv g \pmod n a x ≡ g ( mod n ) form the row (Lemma B.7 ) of a 0 a_0 a 0 , the inverse of x / g x/g x / g modulo n / g n/g n / g , so that gcd ( a 0 , n / g ) = 1 \gcd(a_0,n/g)=1 g cd( a 0 , n / g ) = 1 ; they all have [ a x ] n = g [ax]_n=g [ a x ] n = g . The divisor g g g satisfies the hypotheses of Lemma B.7 .
Corollary C.1. If Δ \Delta Δ has no witness and 5 n > 324 J 2 5n>324J^2 5 n > 324 J 2 , then p < n / 18 p<n/18 p < n /18 .
Proof. By Lemma B.4, with ϵ = 1 \epsilon=1 ϵ = 1 , p q ≤ n J 2 pq\le nJ^2 pq ≤ n J 2 . Suppose p ≥ n / 18 p\ge n/18 p ≥ n /18 . As p ≤ q p\le q p ≤ q and p + q = ( n − s ) / 2 ≥ n / 3 p+q=(n-s)/2\ge n/3 p + q = ( n − s ) /2 ≥ n /3 (because s ≤ n / 3 s\le n/3 s ≤ n /3 ), q ≥ max ( p , n / 3 − p ) q\ge\max(p,n/3-p) q ≥ max ( p , n /3 − p ) . For n / 18 ≤ p ≤ n / 6 n/18\le p\le n/6 n /18 ≤ p ≤ n /6 the concave function p ( n / 3 − p ) p(n/3-p) p ( n /3 − p ) is at least min ( 5 n 2 / 324 , n 2 / 36 ) = 5 n 2 / 324 \min(5n^2/324,n^2/36)=5n^2/324 min ( 5 n 2 /324 , n 2 /36 ) = 5 n 2 /324 , and for p > n / 6 p>n/6 p > n /6 , p q ≥ p 2 > n 2 / 36 pq\ge p^2>n^2/36 pq ≥ p 2 > n 2 /36 . So n J 2 ≥ 5 n 2 / 324 nJ^2\ge5n^2/324 n J 2 ≥ 5 n 2 /324 , which contradicts 5 n > 324 J 2 5n>324J^2 5 n > 324 J 2 .
Theorem C.2. Let n n n be odd with 5 n > 324 J ( n ) 2 5n>324J(n)^2 5 n > 324 J ( n ) 2 . If a window triangle ( p , q , r ) / n (p,q,r)/n ( p , q , r ) / n has no witness, then s ∣ n s\mid n s ∣ n ; if moreover p ≥ 2 p\ge2 p ≥ 2 , then 2 p ≤ s 2p\le s 2 p ≤ s and Δ \Delta Δ lies in B B B .
Proof. Suppose Δ \Delta Δ has no witness. By Corollary C.1, p < n / 18 p<n/18 p < n /18 , so L = 2 p + s < n / 9 + n / 3 < n / 2 L=2p+s<n/9+n/3<n/2 L = 2 p + s < n /9 + n /3 < n /2 .
s s s divides n n n . Suppose not. Then s ≥ 3 s\ge3 s ≥ 3 (as s s s is odd), s ≠ n / 3 s\ne n/3 s = n /3 , and g = gcd ( s , n ) = gcd ( r , n ) < s g=\gcd(s,n)=\gcd(r,n)<s g = g cd( s , n ) = g cd( r , n ) < s is odd. If g = 1 g=1 g = 1 , Corollary B.6 (a) with u = 2 u=2 u = 2 , a prime not dividing n n n with 2 L < n 2L<n 2 L < n , would give 4 ∣ n 4\mid n 4 ∣ n . If g = 3 g=3 g = 3 and 9 ∤ n 9\nmid n 9 ∤ n , apply Lemma B.9 to the row of the a a a with a r ≡ 3 ar\equiv3 a r ≡ 3 : 3 3 3 divides r r r and n n n , 3 s < n 3s<n 3 s < n , 3 ( L − 1 ) < 2 n 3(L-1)<2n 3 ( L − 1 ) < 2 n , and X = 3 < s X=3<s X = 3 < s with 2 X ≠ s 2X\ne s 2 X = s ; it gives n ≤ 3 L − s = 6 p + 2 s < n / 3 + 2 n / 3 = n n\le3L-s=6p+2s<n/3+2n/3=n n ≤ 3 L − s = 6 p + 2 s < n /3 + 2 n /3 = n , which is false. Otherwise take the row of the a a a with a r ≡ g ar\equiv g a r ≡ g : its elements have [ a r ] n = g < s [ar]_n=g<s [ a r ] n = g < s , so they lie in S r S_r S r , and 1 is not among them, as [ 1 ⋅ r ] n = r ≠ g [1\cdot r]_n=r\ne g [ 1 ⋅ r ] n = r = g ; so ν = 0 \nu=0 ν = 0 . As ( L − 1 ) g / n < g / 2 (L-1)g/n<g/2 ( L − 1 ) g / n < g /2 , Υ ≤ ( g + 1 ) / 2 \Upsilon\le(g+1)/2 Υ ≤ ( g + 1 ) /2 and g − Υ ≥ ( g − 1 ) / 2 g-\Upsilon\ge(g-1)/2 g − Υ ≥ ( g − 1 ) /2 ; and ( g − 1 ) / 2 ≥ J ( g ∗ ) (g-1)/2\ge J(g^*) ( g − 1 ) /2 ≥ J ( g ∗ ) : if g ∗ = 1 g^*=1 g ∗ = 1 , as g ≥ 3 g\ge3 g ≥ 3 ; if g ∗ g^* g ∗ is a prime, then g ≥ 5 g\ge5 g ≥ 5 (since g = 3 g=3 g = 3 with g ∗ = 3 g^*=3 g ∗ = 3 means 9 ∤ n 9\nmid n 9 ∤ n ) and J ( g ∗ ) = 2 J(g^*)=2 J ( g ∗ ) = 2 ; if g ∗ g^* g ∗ has two primes, g ≥ g ∗ ≥ 15 g\ge g^*\ge15 g ≥ g ∗ ≥ 15 , so ( g − 1 ) / 2 ≥ 7 (g-1)/2\ge7 ( g − 1 ) /2 ≥ 7 , and J ( g ∗ ) ≤ 3 J(g^*)\le3 J ( g ∗ ) ≤ 3 ; if g ∗ g^* g ∗ has k ≥ 3 k\ge3 k ≥ 3 primes, g ≥ g ∗ ≥ 3 ⋅ 5 ⋅ 7 ⋅ 11 k − 3 g\ge g^*\ge3\cdot5\cdot7\cdot11^{k-3} g ≥ g ∗ ≥ 3 ⋅ 5 ⋅ 7 ⋅ 1 1 k − 3 and ( g − 1 ) / 2 ≥ 52 ⋅ 11 k − 3 ≥ 3 k ≥ J ( g ∗ ) (g-1)/2\ge52\cdot11^{k-3}\ge3^k\ge J(g^*) ( g − 1 ) /2 ≥ 52 ⋅ 1 1 k − 3 ≥ 3 k ≥ J ( g ∗ ) (Lemma 4.13 ). So Lemma B.7 (iv) gives a usable unit in S r ∩ S q S_r\cap S_q S r ∩ S q , a witness. In each case this is a contradiction.
If p ≥ 2 p\ge2 p ≥ 2 , then 2 p ≤ s 2p\le s 2 p ≤ s . Suppose s ∣ n s\mid n s ∣ n , p ≥ 2 p\ge2 p ≥ 2 and 2 p > s 2p>s 2 p > s . Then L < 4 p < 2 n / 9 L<4p<2n/9 L < 4 p < 2 n /9 , so 3 L < n 3L<n 3 L < n . Let g = gcd ( p , n ) g=\gcd(p,n) g = g cd( p , n ) . If g = 1 g=1 g = 1 , Corollary B.6 (b) with u = 2 u=2 u = 2 would give 3 L ≥ n 3L\ge n 3 L ≥ n . If 3 ≤ g < p 3\le g<p 3 ≤ g < p , take the row of the a a a with a p ≡ g ap\equiv g a p ≡ g : its elements lie in S p S_p S p , as g < 2 p g<2p g < 2 p , and 1 is not among them, as [ 1 ⋅ p ] n = p ≠ g [1\cdot p]_n=p\ne g [ 1 ⋅ p ] n = p = g ; so ν = 0 \nu=0 ν = 0 . If g ≥ 5 g\ge5 g ≥ 5 , Lemma B.8 applies, as 7 L < 3 n 7L<3n 7 L < 3 n ; if g = 3 g=3 g = 3 , then Υ = 1 \Upsilon=1 Υ = 1 as 3 L < n 3L<n 3 L < n , and g − Υ = 2 ≥ J ( g ∗ ) g-\Upsilon=2\ge J(g^*) g − Υ = 2 ≥ J ( g ∗ ) , so Lemma B.7 (iv) applies. Either gives a usable unit in S p ∩ S q S_p\cap S_q S p ∩ S q , a witness. If g = p g=p g = p , so p ∣ n p\mid n p ∣ n and p ≥ 3 p\ge3 p ≥ 3 : q = ( n − s ) / 2 − p > n / 2 − 2 p > 7 n / 18 > n / 3 ≥ n / p q=(n-s)/2-p>n/2-2p>7n/18>n/3\ge n/p q = ( n − s ) /2 − p > n /2 − 2 p > 7 n /18 > n /3 ≥ n / p (as s < 2 p < n / 9 s<2p<n/9 s < 2 p < n /9 ), and t max = ⌈ q / p ⌉ − 1 > 7 p / 18 − 1 t_{\max}=\lceil q/p\rceil-1>7p/18-1 t m a x = ⌈ q / p ⌉ − 1 > 7 p /18 − 1 in Lemma B.10 with g = p g=p g = p . If g ∗ g^* g ∗ has at most one prime, part (a) of that lemma applies; if two, p ≥ 15 p\ge15 p ≥ 15 and t max ≥ 5 ≥ J ( g ∗ ) t_{\max}\ge5\ge J(g^*) t m a x ≥ 5 ≥ J ( g ∗ ) ; if k ≥ 3 k\ge3 k ≥ 3 , p ≥ 105 ⋅ 11 k − 3 p\ge105\cdot11^{k-3} p ≥ 105 ⋅ 1 1 k − 3 and t max > 39 ⋅ 11 k − 3 ≥ 3 k ≥ J ( g ∗ ) t_{\max}>39\cdot11^{k-3}\ge3^k\ge J(g^*) t m a x > 39 ⋅ 1 1 k − 3 ≥ 3 k ≥ J ( g ∗ ) ; in these two cases part (b) applies. It gives a witness. In each case this is a contradiction.
The unit condition. As n n n is odd, n / s n/s n / s is odd; with s ∣ n s\mid n s ∣ n and 2 ≤ p ≤ s / 2 2\le p\le s/2 2 ≤ p ≤ s /2 , Proposition 4.3(c) gives (U), and Δ \Delta Δ lies in B B B .
For p = 1 p=1 p = 1 the theorem says that a window triangle ( 1 , q , r ) / n (1,q,r)/n ( 1 , q , r ) / n without a witness has s ∣ n s\mid n s ∣ n , that is, lies in A A A .
Theorem C.3. Let n n n be odd. A window triangle of denominator n n n has no witness if and only if it lies in A \mathcal{A} A or in B \mathcal{B} B , or is one of ( 2 , 3 , 6 ) / 11 (2,3,6)/11 ( 2 , 3 , 6 ) /11 , ( 2 , 5 , 8 ) / 15 (2,5,8)/15 ( 2 , 5 , 8 ) /15 , ( 3 , 4 , 8 ) / 15 (3,4,8)/15 ( 3 , 4 , 8 ) /15 , ( 2 , 7 , 12 ) / 21 (2,7,12)/21 ( 2 , 7 , 12 ) /21 , ( 3 , 7 , 11 ) / 21 (3,7,11)/21 ( 3 , 7 , 11 ) /21 .
Proof. Only if. If 5 n > 324 J ( n ) 2 5n > 324J(n)^2 5 n > 324 J ( n ) 2 , this is Theorem C.2 . The odd n ≥ 5 n \ge5 n ≥ 5 with 5 n ≤ 324 J ( n ) 2 5n \le324J(n)^2 5 n ≤ 324 J ( n ) 2 are 320 values, the largest 3,135 = 3 ⋅ 5 ⋅ 11 ⋅ 19 3{,}135 = 3 \cdot5 \cdot11 \cdot19 3 , 135 = 3 ⋅ 5 ⋅ 11 ⋅ 19 , with J = 7 J = 7 J = 7 (n = 3 n = 3 n = 3 also satisfies the inequality and has no window triangle); for them an enumeration of all window triangles finds no witness-free triangle other than those stated (Appendix A.§ 3 ; the list of 320 values is proved complete in Appendix A.§ 2 ). If. Members of A \mathcal{A} A and of B \mathcal{B} B have no witness by Proposition 4.3. The five single triangles have no witness by a direct check of the units (Appendix A). □
For prime n n n , Theorem C.3 gives the prime case stated after Theorem 4.1 : s ∣ n s \mid n s ∣ n forces s = 1 s = 1 s = 1 , which excludes B \mathcal{B} B , and the only single triangle of prime denominator is ( 2 , 3 , 6 ) / 11 (2,3,6)/11 ( 2 , 3 , 6 ) /11 .
Declarations
Use of AI. This paper and its results were produced by an AI research system that the author designed, built and directs, running on Anthropic’s Claude Fable 5.2 in Claude Code and OpenAI’s GPT-6 Astra in Codex. The author chooses the problems and prioritizes the work. Separate AI agents checked each result, and a second program confirmed the computations the proofs use, with any exceptions listed in the public deposit. The system also searched the literature and checked the citations against reliable sources. The author read the paper, reran the computations from the deposit and takes full responsibility for all of the paper’s contents.
Data availability. The programs and data used in this paper, with instructions for rerunning them, are deposited at Zenodo [13 ] .
References
Toronto, Canada
Email address: manvirjaswal@icloud.com
URL: https://manvirjaswal.github.io