Preprint
Article

This version is not peer-reviewed.

Black Hole Indices of N = 8 String Theory Are Positive

Submitted:

11 November 2026

Posted:

12 November 2026

Read the latest preprint version here

Abstract
Sen predicted that the protected indices of single-centred black holes in type II string theory on T6 and T5 are positive. We prove this for every charge whose index is known: four-dimensional dyons, five-dimensional rotating black holes, and five-dimensional black holes of fixed angular momentum.
Keywords: 
;  ;  

MSC:  11F50 (primary); 11P82; 81T30; 83E30

1. Introduction

The entropy of a supersymmetric black hole in string theory can be compared with a count of microscopic states. Strominger and Vafa did this for five-dimensional black holes made of D1- and D5-branes with momentum [34], Breckenridge, Myers, Peet and Vafa added angular momentum [7], and Maldacena, Moore and Strominger found the exact count for type II string theory on a torus [21]. In four dimensions the analogous formula was derived by Shih, Strominger and Yin from the relation between four- and five-dimensional black holes [17,33], and Sen extended it to all charges of a certain primitivity type [27,30]. These formulae count the states with a sign: what is computed at weak coupling is a helicity trace index, the quantity that does not change when the coupling is increased until the black hole forms [3,4,20,26].
A black hole is more than the integer that counts it. Sen observed that a single-centred black hole whose only hair outside the horizon are the zero modes of the broken supersymmetries has a horizon of zero angular momentum, so that its microstates all contribute to the index with the same sign. Once the zero modes are removed, the index must therefore be a positive integer [10,14,31]. For the black holes of type II string theory on T 6 the multi-centred configurations that could spoil this do not contribute [27,30], so the prediction applies to the microscopic formula itself. It is a nontrivial statement about Fourier coefficients: the formulae are sums of coefficients of the Jacobi form ϕ − 2 , 1 , whose signs alternate.1 Bringmann and Murthy stated the prediction for the four- and five-dimensional black holes of type II on T 6 and T 5 and proved it when the index is a single Fourier coefficient [8]. Sen introduced the five-dimensional index at fixed angular momentum [32], which the same argument predicts to be positive [8].
This paper proves the prediction in full, for every charge for which the index is known.

The indices

Write q = e 2 π i τ and ζ = e 2 π i z , and define the integers c ^ ( D ) by
− ϑ 1 ( τ , z ) 2 η ( τ ) 6 = ∑ k , l ∈ Z c ^ ( 4 k − l 2 ) q k ζ l ,
so that c ^ ( − 1 ) = 1 , c ^ ( 0 ) = − 2 , c ^ ( 3 ) = 8 , c ^ ( 4 ) = − 12 , c ^ ( 7 ) = 39 , c ^ ( 8 ) = − 56 , and c ^ ( D ) = 0 when D < − 1 or D ≡ 1 , 2 ( mod 4 ) . The left side is the weak Jacobi form ϕ − 2 , 1 of weight − 2 and index 1 [16]; Section 2 explains why a coefficient depends only on 4 k − l 2 .
In type IIB on T 4 × S 1 × S ˜ 1 , consider D1-, D3- and D5-branes wrapped on S ˜ 1 with charge vector Q and on S 1 with charge vector P; both lie in the even lattice of signature ( 4 , 4 ) formed by the even homology of T 4 , and Q 2 , P 2 , Q · P are the intersection products [30]. Put
Δ = Q 2 P 2 − ( Q · P ) 2 , ℓ 1 = gcd i < j { Q i P j − Q j P i } , ℓ 2 = gcd 1 2 Q 2 , 1 2 P 2 , Q · P .
A black hole with these charges exists when Δ > 0 ; its entropy is π & # x 00394 ; [19]. When gcd ( ℓ 1 , ℓ 2 ) = 1 the index of the states with charges ( Q , P ) is [30]2
d ( Q , P ) = ( − 1 ) Q · P + 1 ∑ s ∣ ℓ 1 ℓ 2 s c ^ Δ / s 2 .
In type IIB on T 4 × S 1 , a system of Q 1 D1-branes and Q 5 D5-branes with n units of momentum along S 1 and angular momentum J 3 L = J / 2 forms a rotating five-dimensional black hole when 4 Q 1 Q 5 n − J 2 > 0 ; its entropy is π 4 Q 1 Q 5 n − J 2 [7]. Put M = Q 1 Q 5 n and, for l ∈ Z ,
F ( l ) = ∑ s ∣ gcd ( n , Q 1 Q 5 , l ) s c ^ 4 M − l 2 s 2 .
For gcd ( Q 1 , Q 5 ) = 1 the index with fixed J and summed over J → 2 is [14,32]
d ˜ ( n , Q 1 , Q 5 , J ) = ( − 1 ) J + 1 ∑ r = − 2 2 ( − 1 ) r 4 2 − r F ( J − r ) ,
that is, ∑ J ( − 1 ) J + 1 d ˜ ( J ) ζ J = ( ζ 1 / 2 − ζ − 1 / 2 ) 4 ∑ l F ( l ) ζ l ; the factor ( ζ 1 / 2 − ζ − 1 / 2 ) 4 removes the fermion zero modes of the D1-D5 system that carry J 3 L .3 Sen’s index at fixed J and fixed J → L 2 = J 2 ( J 2 + 1 ) is [32]
d ( n , Q 1 , Q 5 , J ) = d ˜ ( n , Q 1 , Q 5 , J ) − d ˜ ( n , Q 1 , Q 5 , J + 2 ) , J ≥ 0 .

Results

Theorem A.
Let ( Q , P ) be charges with Δ > 0 and gcd ( ℓ 1 , ℓ 2 ) = 1 . Then d ( Q , P ) > 1 2 | c ^ ( Δ ) | > 0 .
Theorem B.
Let Q 1 , Q 5 , n ≥ 1 with gcd ( Q 1 , Q 5 ) = 1 , and J ∈ Z with 4 Q 1 Q 5 n − J 2 > 0 . Then d ˜ ( n , Q 1 , Q 5 , J ) > 0 .
Theorem C.
Under the hypotheses of Theorem B with J ≥ 0 , d ( n , Q 1 , Q 5 , J ) > 0 .
The three statements are exactly the predictions of [8] and [32]. They cover every charge for which an exact formula is known; the four-dimensional charges with gcd ( ℓ 1 , ℓ 2 ) > 1 remain outside, see Section 8.3.

What the proof has to overcome

The sign of c ^ ( D ) is ( − 1 ) D + 1 (Lemma 2), and Δ ≡ Q · P ( mod 2 ) because the lattice is even. So the term s = 1 of (2) is | c ^ ( Δ ) | . If s is odd then s 2 ≡ 1 ( mod 8 ) and Δ / s 2 ≡ Δ ( mod 2 ) , so every odd s contributes a positive term as well; this is the observation behind the proofs of [8]. An even s breaks it. Take 1 2 Q 2 = 1 2 P 2 = 2 and Q · P = 2 , for instance Q = e 1 + 2 f 1 and P = e 2 + 2 f 2 + 2 f 1 in two hyperbolic planes with e i · f i = 1 . Then ℓ 1 = 1 , ℓ 2 = 2 , Δ = 12 , and
d ( Q , P ) = − c ^ ( 12 ) + 2 c ^ ( 3 ) = 208 − 16 = 192 .
The term s = 2 is negative. Positivity is then a matter of size: the negative terms must be smaller than the positive ones, for every Δ and every ℓ 1 ℓ 2 . In five dimensions the same problem appears in three places: in the divisor sum; in the difference (5), where two coefficients of the same sign are subtracted from each other; and at the values of l with 4 M − l 2 < 0 , where the term ( 4 M − l 2 ) / s 2 = − 1 of (3) survives and can carry either sign.

Plan

The proof rests on one inequality, Proposition 1: the growth | c ^ ( D ) | − | c ^ ( D − 4 ) | of the coefficients within a residue class exceeds twice every divisor sum ∑ s ≥ 2 s 2 | c ^ ( E / s 2 ) | with E ≤ D . Section 2 collects the background and derives the facts about helicity traces, charge invariants and ϕ − 2 , 1 that the proof uses. Section 3 writes | c ^ ( D ) | through the numbers a ( n ) of partitions of n into parts of six colours; this gives monotonicity and an upper bound. Section 4 proves explicit bounds for a ( n ) from above and below. Section 5 proves the inequality, analytically for D ≥ 359 and by finite arithmetic below. Section 6 and Section 7 prove the three theorems and work out examples. Section 8 discusses sharpness, the U-duality invariant extension of [21] and the charges not covered by (2). Appendix A derives the elementary facts used along the way, and Appendix B describes the finite computation.

2. Background

2.1. Helicity Traces and Fermion Zero Modes

Let h be the helicity, the eigenvalue of the rotation generator J 3 in the rest frame of a massive particle in four dimensions. The helicity traces are
B 2 m = 1 ( 2 m ) ! Tr ( − 1 ) 2 h ( 2 h ) 2 m = 1 ( 2 m ) ! y d d y 2 m Z ( y ) | y = 1 , Z ( y ) = Tr ( − 1 ) 2 h y 2 h ,
the trace running over the states of fixed charge [3,4,20]. A BPS state that breaks 4 k of the supercharges carries 2 k pairs of fermion zero modes, the goldstinos of the broken supersymmetries. Each pair acts as a fermionic oscillator that raises h by 1 2 , so it multiplies Z ( y ) by y 2 h 0 ( 1 − y ) for some h 0 , or, after a shift of the helicity and a sign that do not affect the argument, by y 1 / 2 − y − 1 / 2 .4 Hence
Z ( y ) = y 1 / 2 − y − 1 / 2 2 k Z rest ( y ) ,
where Z rest is the trace over the remaining degrees of freedom. Since y 1 / 2 − y − 1 / 2 vanishes to first order at y = 1 and y d d y ( y 1 / 2 − y − 1 / 2 ) = 1 there, every B 2 m with m < k vanishes, and in ( y d / d y ) 2 k Z at y = 1 only the term in which each derivative falls on a different zero-mode factor survives. There are ( 2 k ) ! such terms, so
B 2 k = Z rest ( 1 ) = Tr rest ( − 1 ) 2 h .
For type II string theory on T 6 there are 32 supercharges, a quarter-BPS black hole preserves 4 of them and breaks 28, so 2 k = 14 and the relevant index is B 14 [26,30]. It is the first trace that does not vanish, it is protected, and d = − B 14 is the index (2). In five dimensions the same reasoning applied to the four zero modes that carry J 3 L gives the factor ( ζ 1 / 2 − ζ − 1 / 2 ) 4 of (4) [14].

2.2. Why the Sign Is Predicted

At strong coupling the states counted by Z rest live on the horizon of the black hole. Sen showed, by the quantum entropy function [28,29], that a single-centred BPS black hole has a near horizon geometry A d S 2 × S 2 whose S U ( 2 ) isometry forces the horizon degrees of freedom to carry J → 2 = 0 [14,31]. Then ( − 1 ) 2 h = 1 on all of them, and the trace equals the dimension of the space of horizon states, up to the overall sign fixed by the conventions. Chowdhury, Garavuso, Mondal and Sen tested this at weak coupling, finding that the microstates of D-brane systems with small charges all have zero angular momentum [9,10]. The prediction therefore reads d > 0 in four dimensions; in five dimensions, where the horizon of the rotating black hole carries J 3 L , it reads d ˜ > 0 for the index summed over J → 2 and d > 0 for the index at fixed J → L 2 [32]. The large charge behaviour, log | c ^ ( Δ ) | ∼ π & # x 00394 ; , agrees with the black hole entropy, including the logarithmic corrections [13,32].

2.3. Charges and Their Invariants

The charges ( Q , P ) transform under T-duality, the integral automorphisms g of the lattice, by ( Q , P ) ↦ ( g Q , g P ) , and under S-duality by ( Q , P ) ↦ ( a Q + b P , c Q + d P ) with a b c d ∈ S L ( 2 , Z ) ; both are part of the U-duality group [12,30].
Lemma 1.
Δ, ℓ 1 and ℓ 2 are invariant under T- and S-duality.
Proof. 
Put f ( x , y ) = 1 2 ( x Q + y P ) 2 = 1 2 Q 2 x 2 + ( Q · P ) x y + 1 2 P 2 y 2 , a binary quadratic form with integer coefficients. Its discriminant is ( Q · P ) 2 − Q 2 P 2 = − Δ , and ℓ 2 is the greatest common divisor of its coefficients. T-duality preserves inner products, so it fixes f. S-duality replaces f ( x , y ) by f ( a x + c y , b x + d y ) , an invertible integral change of variables; this preserves the discriminant, and the content of a form divides the content of any form obtained from it by an integral substitution, so applying the inverse substitution shows that the content is unchanged.
For ℓ 1 , let Q ∧ P be the vector of minors Q i P j − Q j P i . Under T-duality Q ∧ P ↦ ( Λ 2 g ) ( Q ∧ P ) , where Λ 2 g is an integral matrix with integral inverse Λ 2 ( g − 1 ) ; such a matrix preserves the greatest common divisor of the entries of a vector, by the same divisibility argument. Under S-duality Q ∧ P ↦ ( a d − b c ) Q ∧ P = Q ∧ P . □
Sen used these transformations to bring every charge with gcd ( ℓ 1 , ℓ 2 ) = 1 to a normal form in which the index can be read off from the formula of [21,33], and so arrived at (2) [30]. In the normal form one may take Q = ℓ 1 Q 0 , P = P 0 with Q 0 ∧ P 0 primitive; then 1 2 Q 2 = ℓ 1 2 · 1 2 Q 0 2 and Q · P = ℓ 1 Q 0 · P 0 . The examples in Section 6 are in this form.5

2.4. The Jacobi form ϕ − 2 , 1

The odd theta function and the Dedekind eta function are
ϑ 1 ( τ , z ) = − i ∑ r ∈ Z + 1 2 ( − 1 ) r − 1 2 q r 2 / 2 ζ r , η ( τ ) = q 1 / 24 ∏ n ≥ 1 ( 1 − q n ) .
By Jacobi’s triple product identity (Appendix A),
ϑ 1 ( τ , z ) = − i q 1 / 8 ζ 1 / 2 − ζ − 1 / 2 ∏ n ≥ 1 ( 1 − q n ) ( 1 − q n ζ ) ( 1 − q n ζ − 1 ) ,
and therefore
ϕ − 2 , 1 ( τ , z ) = − ϑ 1 ( τ , z ) 2 η ( τ ) 6 = ζ − 2 + ζ − 1 ∏ n ≥ 1 ( 1 − q n ζ ) 2 ( 1 − q n ζ − 1 ) 2 ( 1 − q n ) 4 .
This is a weak Jacobi form of weight − 2 and index 1 [16]. For index one, invariance under z ↦ z + λ τ + μ forces the coefficient of q k ζ l to depend only on D = 4 k − l 2 and on l mod 2 , which is determined by D; this is the reason for writing the coefficients as c ^ ( 4 k − l 2 ) . Lemma 2 below proves this directly. Grouping the terms by l mod 2 gives the theta decomposition
ϕ − 2 , 1 = h 0 ( τ ) θ 1 , 0 ( τ , z ) + h 1 ( τ ) θ 1 , 1 ( τ , z ) , θ 1 , μ = ∑ l ≡ μ ( 2 ) q l 2 / 4 ζ l , h μ = ∑ D ≡ − μ 2 ( 4 ) c ^ ( D ) q D / 4 .
Lemma 2 identifies the two components in closed form:
h 0 ( τ ) = − ϑ 2 ( 2 τ ) η ( τ ) 6 , h 1 ( τ ) = ϑ 3 ( 2 τ ) η ( τ ) 6 ,
with ϑ 2 ( 2 τ ) = ∑ t ∈ Z q ( t + 1 / 2 ) 2 and ϑ 3 ( 2 τ ) = ∑ m ∈ Z q m 2 . These are the functions through which Bringmann and Murthy express the indices [8].6

2.5. The Five-Dimensional Generating Function

The D1-D5 system wrapped on T 4 × S 1 is described at low energy by a two-dimensional conformal field theory, and the generating function of its BPS states is built from the elliptic genus of a symmetric product [15,21]. After the four fermion zero modes are removed, the coefficient of q n ζ l in the relevant generating function is F ( l ) of (3), and the index is (4) [14]. The relation between the four- and five-dimensional black holes [17] identifies J with the momentum along the circle of a Taub-NUT space, and the same coefficients c ^ ( D ) appear in both indices. Expanding the factor
ζ 1 / 2 − ζ − 1 / 2 4 = ζ 2 − 4 ζ + 6 − 4 ζ − 1 + ζ − 2 = ∑ r = − 2 2 ( − 1 ) r 4 2 − r ζ r
and comparing coefficients of ζ J gives (4). Since F ( − l ) = F ( l ) and the binomial coefficients are symmetric, d ˜ ( − J ) = d ˜ ( J ) .

3. The Coefficients

Let
P ( q ) = ∏ n ≥ 1 ( 1 − q n ) − 6 = ∑ n ≥ 0 a ( n ) q n = 1 + 6 q + 27 q 2 + 98 q 3 + 315 q 4 + … ,
so that a ( n ) counts the partitions of n into parts of six colours, and put a ( n ) = 0 for n < 0 . The values for n ≤ 90 are listed in Table A1.
Lemma 2.
For every D ≥ − 1 with D ≡ 0 , 3 ( mod 4 ) ,
c ^ ( D ) = ( − 1 ) D + 1 ∑ j ∈ Z , j ≡ D + 1 ( 2 ) j 2 ≤ D + 1 a D + 1 − j 2 4 .
In particular ( − 1 ) D + 1 c ^ ( D ) = | c ^ ( D ) | ≥ 1 , and
| c ^ ( 4 k − 1 ) | = ∑ m ∈ Z a ( k − m 2 ) , | c ^ ( 4 k ) | = 2 ∑ t ≥ 0 a k − t ( t + 1 ) .
Proof. 
By the definition of ϑ 1 ,
− ϑ 1 ( τ , z ) 2 = ∑ r ∈ Z + 1 2 ( − 1 ) r − 1 2 q r 2 / 2 ζ r 2 , η ( τ ) − 6 = q − 1 / 4 P ( q ) .
Hence the coefficient of q k ζ l in (1) is ∑ ( − 1 ) r + s − 1 a k + 1 4 − 1 2 ( r 2 + s 2 ) , summed over half-integers r , s with r + s = l . Put j = r − s . Then r 2 + s 2 = 1 2 ( l 2 + j 2 ) , the argument of a becomes 1 4 ( D + 1 − j 2 ) with D = 4 k − l 2 , the pairs ( r , s ) correspond to the integers j ≡ l + 1 ( mod 2 ) , and ( − 1 ) r + s − 1 = ( − 1 ) l − 1 . Since D ≡ l ( mod 2 ) , the sign is ( − 1 ) D + 1 and the parity condition is j ≡ D + 1 . This proves (9); the term j = 0 or j = ± 1 is at least a ( 0 ) = 1 . For D = 4 k − 1 the integers j are even, j = 2 m , and 1 4 ( D + 1 − j 2 ) = k − m 2 ; for D = 4 k they are odd, j = 2 t + 1 , and 1 4 ( D + 1 − j 2 ) = k − t ( t + 1 ) , the values t and − 1 − t giving the same term. □
In generating function form the lemma says ∑ k | c ^ ( 4 k − 1 ) | q k = ϑ 3 ( 2 τ ) P ( q ) and ∑ k | c ^ ( 4 k ) | q k = q − 1 / 4 ϑ 2 ( 2 τ ) P ( q ) , which is (8). Figure 1 shows the coefficients and their alternating sign.
Lemma 3.
The sequence a ( n ) is nondecreasing, and a ( n ) − a ( n − 1 ) ≥ n + 4 4 for n ≥ 1 ; also a ( n ) ≥ n + 5 5 . For D ≥ 3 with D ≡ 0 , 3 ( mod 4 ) put k = ⌊ ( D + 1 ) / 4 ⌋ and G ( D ) = | c ^ ( D ) | − | c ^ ( D − 4 ) | . Then
G ( D ) ≥ a ( k ) − a ( k − 1 ) , D = 4 k − 1 , 2 a ( k ) − a ( k − 1 ) , D = 4 k .
In particular | c ^ ( D ) | is strictly increasing on each of the classes D ≡ 3 and D ≡ 0 ( mod 4 ) , D ≥ − 1 . Moreover, for every E ≥ − 1 ,
| c ^ ( E ) | ≤ E + 1 + 1 a ⌊ ( E + 1 ) / 4 ⌋ .
Proof. 
The series ( 1 − q ) P ( q ) = ( 1 − q ) − 5 ∏ m ≥ 2 ( 1 − q m ) − 6 has nonnegative coefficients, and since the second factor has constant term 1, its coefficient of q n is at least the coefficient n + 4 4 of ( 1 − q ) − 5 ; hence a ( n ) − a ( n − 1 ) ≥ n + 4 4 > 0 for n ≥ 1 . In the same way P ( q ) = ( 1 − q ) − 6 ∏ m ≥ 2 ( 1 − q m ) − 6 gives a ( n ) ≥ n + 5 5 . For D = 4 k − 1 , Lemma 2 gives
G ( D ) = ∑ m 2 ≤ k a ( k − m 2 ) − a ( k − 1 − m 2 ) ,
a sum of nonnegative terms whose term m = 0 is a ( k ) − a ( k − 1 ) . For D = 4 k the same computation with t ( t + 1 ) in place of m 2 and the factor 2 gives the second case. In (9) the integers j of fixed parity with j 2 ≤ E + 1 lie in an interval of length 2 E + 1 with spacing 2, so there are at most E + 1 + 1 of them, and each argument of a is at most ⌊ ( E + 1 ) / 4 ⌋ ; since a is nondecreasing, (11) follows. □
Two arithmetic facts link the charges with these coefficients.
Lemma 4.
Let Q , P lie in an even lattice with integral bilinear form, and let ℓ 1 , ℓ 2 and Δ be as in Section 1. Then Δ ≡ Q · P ( mod 2 ) , Δ ≡ 0 , 3 ( mod 4 ) , and ℓ 1 2 ∣ Δ and ℓ 2 2 ∣ Δ . If gcd ( ℓ 1 , ℓ 2 ) = 1 , then s 2 ∣ Δ for every s ∣ ℓ 1 ℓ 2 .
Proof. 
Q 2 and P 2 are even, so Q 2 P 2 ≡ 0 ( mod 4 ) and Δ ≡ − ( Q · P ) 2 , which is ≡ Q · P ( mod 2 ) and ≡ 0 or 3 ( mod 4 ) because a square is 0 or 1 modulo 4. Writing Δ = 4 · Q 2 2 · P 2 2 − ( Q · P ) 2 shows ℓ 2 2 ∣ Δ . Let L be the Gram matrix in a basis of the lattice, so that Q · P = Q T L P . By Lagrange’s identity (Appendix A) applied to the vectors Q , P , L Q , L P ,
Δ = ( Q · L Q ) ( P · L P ) − ( Q · L P ) ( P · L Q ) = ∑ i < j ( Q ∧ P ) i j ( L Q ∧ L P ) i j ,
where the dot is now the Euclidean product of coordinate vectors. The vector L Q ∧ L P = ( Λ 2 L ) ( Q ∧ P ) has entries that are integral combinations of the entries of Q ∧ P . Both factors are divisible by ℓ 1 , so ℓ 1 2 ∣ Δ . If gcd ( ℓ 1 , ℓ 2 ) = 1 then ( ℓ 1 ℓ 2 ) 2 ∣ Δ , and every divisor s of ℓ 1 ℓ 2 has s 2 ∣ Δ . □

4. Partitions into Six Colours

The proof needs a ( n ) bounded above at about half the argument and below at the full argument, with explicit constants. Figure 2 shows the bounds of this section against the true values.
Lemma 5.
For every n ≥ 1 ,
a ( n ) ≤ 1.0001 e 2 π n 8 n 3 / 2 , and a ( n ) ≤ e 2 π n ( n ≥ 0 ) .
Proof. 
For t > 0 the transformation η ( i / t ) = t η ( i t ) [2] gives
P ( e − 2 π t ) = e − π t / 2 η ( i t ) − 6 = t 3 e π / ( 2 t ) − π t / 2 ∏ m ≥ 1 1 − e − 2 π m / t − 6 .
Indeed η ( i t ) − 6 = e π t / 2 P ( e − 2 π t ) by the definition of η , and η ( i t ) − 6 = t 3 η ( i / t ) − 6 = t 3 e π / ( 2 t ) ∏ m ( 1 − e − 2 π m / t ) − 6 . Since the coefficients of P are nonnegative, a ( n ) e − 2 π n t ≤ P ( e − 2 π t ) . Take t = 1 / ( 2 n ) : then 2 π n t = π / ( 2 t ) = π n , t 3 = 1 / ( 8 n 3 / 2 ) and e − π t / 2 < 1 . With x = e − 2 π / t = e − 4 π n ≤ e − 4 π < 3.5 · 10 − 6 ,
− log ∏ m ≥ 1 ( 1 − x m ) = ∑ m ≥ 1 ∑ r ≥ 1 x m r r ≤ ∑ m ≥ 1 x m 1 − x m ≤ x ( 1 − x ) 2 ,
so the product raised to the power − 6 is at most exp 6 x / ( 1 − x ) 2 < 1.0001 . The second bound is a ( 0 ) = 1 for n = 0 ; for n ≥ 1 , 1.0001 8 n 3 / 2 < 1 . Appendix A gives a proof of the second bound that does not use the transformation of η . □
Lemma 6.
Let b ( n ) be the coefficients of exp 6 q / ( 1 − q ) , that is,
b ( 0 ) = 1 , b ( n ) = ∑ j = 1 n 6 j j ! n − 1 j − 1 ( n ≥ 1 ) .
Then a ( n ) ≥ b ( n ) for all n ≥ 0 , and b ( n ) ≥ e 2 6 n − 6 / n for n ≥ 50 .
Proof. 
Logarithmic differentiation of P ( q ) = exp 6 ∑ m ≥ 1 ∑ r ≥ 1 q m r / r and of exp ( 6 ∑ i ≥ 1 q i ) gives
n a ( n ) = 6 ∑ i = 1 n σ ( i ) a ( n − i ) , n b ( n ) = 6 ∑ i = 1 n i b ( n − i ) ,
where σ ( i ) is the sum of the divisors of i; indeed q d d q ∑ m , r q m r / r = ∑ m , r m q m r = ∑ i σ ( i ) q i . Since σ ( i ) ≥ i and a ( 0 ) = b ( 0 ) = 1 , induction on n gives a ( n ) ≥ b ( n ) ≥ 0 . The closed form follows from exp ( 6 u ) = ∑ j 6 j u j / j ! with u = q / ( 1 − q ) and u j = ∑ n ≥ j n − 1 j − 1 q n .
For the lower bound let x = 6 n , j = ⌊ x ⌋ and θ = x − j ∈ [ 0 , 1 ) ; note 1 ≤ j ≤ n . Keeping one term,
b ( n ) ≥ 6 j j ! n − 1 j − 1 = 6 j j ! · j n n j = ( 6 n ) j j n ( j ! ) 2 ∏ i < j 1 − i n .
Robbins’ form of Stirling’s formula [24] is j ! ≤ 2 π j ( j / e ) j e 1 / ( 12 j ) . It gives ( j ! ) 2 ≤ 2 π j ( j / e ) 2 j e 1 / ( 6 j ) , and so
( 6 n ) j j ( j ! ) 2 ≥ e 2 j 2 π x j 2 j e − 1 / ( 6 j ) ≥ e 2 x 2 π e − 2 θ 2 / x − 1 / ( 6 j ) ,
because ( x / j ) 2 j = ( 1 − θ / x ) − 2 j ≥ e 2 j θ / x and therefore e 2 j ( x / j ) 2 j ≥ e 2 x − 2 θ + 2 ( x − θ ) θ / x = e 2 x − 2 θ 2 / x ≥ e 2 x − 2 / x . For 0 ≤ y ≤ 1 2 we have log ( 1 − y ) ≥ − y − y 2 (Appendix A), and i / n < j / n ≤ 6 / n < 1 2 . Since j ( j − 1 ) ≤ x ( x − 1 ) = 6 n − x and ∑ i < j i 2 ≤ j 3 / 3 ≤ x 3 / 3 = 2 n x ,
log ∏ i < j 1 − i n ≥ − j ( j − 1 ) 2 n − 1 n 2 ∑ i < j i 2 ≥ − 3 + x 2 n − 2 x n = − 3 − 3 x 2 n .
Altogether n b ( n ) e − 2 x ≥ exp − log ( 2 π ) − 3 − 2 / x − 3 x / ( 2 n ) − 1 / ( 6 j ) . The three last terms decrease in n; at n = 50 they are 0.116 , 0.520 and 0.010 , and log ( 2 π ) + 3 = 4.838 . The exponent is therefore at least − 5.49 > − 6 . □
Lemma 7.
For n ≥ 1 , a ( n ) − a ( n − 1 ) ≥ 5 n + 5 a ( n ) .
Proof. 
Let w ( i ) ≥ 0 be the coefficients of ∏ m ≥ 2 ( 1 − q m ) − 6 . Then P = ( 1 − q ) − 6 ∑ w ( i ) q i and ( 1 − q ) P = ( 1 − q ) − 5 ∑ w ( i ) q i , so
a ( n ) = ∑ i = 0 n w ( n − i ) i + 5 5 , a ( n ) − a ( n − 1 ) = ∑ i = 0 n w ( n − i ) i + 4 4 .
Since i + 4 4 = 5 i + 5 i + 5 5 ≥ 5 n + 5 i + 5 5 for i ≤ n , the claim follows. □
Combining (10) with Lemmas 6 and 7: for D ≥ 3 , D ≡ 0 , 3 ( mod 4 ) and k = ⌊ ( D + 1 ) / 4 ⌋ ≥ 50 ,
G ( D ) ≥ e Λ ( k ) , Λ ( k ) = 2 6 k − 6 − log k + log 5 k + 5 .

5. The Key Inequality

For E ≥ 1 put
W ( E ) = ∑ s ≥ 2 s 2 ∣ E s 2 c ^ ( E / s 2 ) , W * ( D ) = max 1 ≤ E ≤ D W ( E ) ,
and W ( E ) = 0 for E ≤ 0 . So W ( E ) bounds every divisor sum ∑ s ≥ 2 w ( s ) c ^ ( E / s 2 ) with | w ( s ) | ≤ s 2 .
Proposition 1.
For every D ≥ 3 with D ≡ 0 , 3 ( mod 4 ) ,
G ( D ) > 2 W * ( D ) ,
G ( D ) > W * ( D ) + 1 16 ( D + 4 ) ( D + 8 ) .
Proof. 
The range D ≥ 359 . Here k = ⌊ ( D + 1 ) / 4 ⌋ ≥ 90 and D ≤ 4 k . Let 1 ≤ E ≤ D . Every term of W ( E ) has E / s 2 ≥ 1 , and by (11) it is at most s 2 ( E / s 2 + 1 + 1 ) a ( ⌊ ( E / s 2 + 1 ) / 4 ⌋ ) , which increases with E. The term s = 2 is therefore at most
4 k + 1 + 1 a ⌊ ( k + 1 ) / 4 ⌋ ≤ S 2 ( k ) : = 1.0001 · 4 ( k + 1 + 1 ) ( k + 1 ) 3 / 2 e π k + 1 ,
by Lemma 5 and because e 2 π y y − 3 / 2 increases for y ≥ 1 (its logarithmic derivative is π / y − 3 / ( 2 y ) > 0 ), evaluated at y = ( k + 1 ) / 4 ≥ ⌊ ( k + 1 ) / 4 ⌋ . For s ≥ 3 the factors E / s 2 + 1 + 1 and a ( · ) are largest at s = 3 , and ∑ 3 ≤ s ≤ E s 2 ≤ ⌊ E ⌋ 3 ≤ E 3 / 2 ≤ ( 4 k ) 3 / 2 , since ∑ s = 1 N s 2 = 1 6 N ( N + 1 ) ( 2 N + 1 ) ≤ N 3 . With the second bound of Lemma 5,
∑ s ≥ 3 ≤ S 3 ( k ) : = ( 4 k ) 3 / 2 4 k / 9 + 1 + 1 e 2 π 3 k + 9 / 4 .
So W * ( D ) ≤ S 2 ( k ) + S 3 ( k ) , and 1 16 ( D + 4 ) ( D + 8 ) ≤ ( k + 1 ) ( k + 2 ) . Put S 4 ( k ) = 2 ( k + 1 ) ( k + 2 ) . If
h i ( k ) : = Λ ( k ) − log 4 − log S i ( k ) > 0 ( i = 2 , 3 , 4 ) ,
then by (12) G ( D ) > 4 max ( S 2 , S 3 ) ≥ 2 ( S 2 + S 3 ) ≥ 2 W * ( D ) , which is (13), and G ( D ) > 4 S 4 ( k ) , so that G ( D ) − W * ( D ) > 1 2 G ( D ) > 2 S 4 ( k ) > 1 16 ( D + 4 ) ( D + 8 ) , which is (14).
At k = 90 ,
Λ ( 90 ) = 33.03 , log 4 + log S 2 = 28.33 , log 4 + log S 3 = 32.33 , log 4 + log S 4 = 11.11 .
For the derivatives, Λ ′ ( k ) = 6 / k − 1 / k − 1 / ( k + 5 ) ≥ 6 / k − 2 / k . Differentiating the logarithms and discarding the negative terms,
( log S 2 ) ′ ≤ π 2 k + 1 2 k , ( log S 3 ) ′ ≤ π 3 k + 2 k , ( log S 4 ) ′ ≤ 2 k .
For S 2 the derivative of log ( k + 1 + 1 ) is 2 ( k + 1 + k + 1 ) − 1 < 1 2 k . For S 3 the derivative of 3 2 log ( 4 k ) is 3 2 k , and with y = 4 k / 9 + 1 the derivative of log ( y + 1 ) is 2 9 ( y + y ) < 2 9 y < 1 2 k . Hence
h 2 ′ ≥ 0.878 k − 2.5 k , h 3 ′ ≥ 1.402 k − 4 k , h 4 ′ ≥ 2.449 k − 4 k ,
all positive for k ≥ 9 . So (15) holds for every k ≥ 90 .7
The range 3 ≤ D ≤ 358 . Here every quantity in (13) and (14) is an integer determined by a ( 0 ) , … , a ( 89 ) through (9). Appendix B describes the computation, which compares integers only. Both inequalities hold; the largest value of 2 W * ( D ) / G ( D ) is 64 / 152 = 0.421 , at D = 12 , and the largest value of W * ( D ) + 1 16 ( D + 4 ) ( D + 8 ) / G ( D ) is 77 112 = 0.6875 , at D = 3 , where W * ( 3 ) = 0 and G ( 3 ) = 8 − 1 = 7 . Figure 3 shows both ratios. □
Figure 3. The two ratios of Proposition 1 for 3 ≤ D ≤ 359 . Both stay below 1, with the largest values at the smallest discriminants, and fall off like e − c D .
Figure 3. The two ratios of Proposition 1 for 3 ≤ D ≤ 359 . Both stay below 1, with the largest values at the smallest discriminants, and fall off like e − c D .
Preprints 243591 g003
Figure 4. The lower bound Λ ( k ) for log G ( D ) against the upper bounds log 4 + log S i ( k ) for the divisor sums. The analytic argument applies from k = 90 on.
Figure 4. The lower bound Λ ( k ) for log G ( D ) against the upper bounds log 4 + log S i ( k ) for the divisor sums. The analytic argument applies from k = 90 on.
Preprints 243591 g004
The inequality is used in the following form.
Corollary 1.
Let D ≥ 3 with D ≡ 0 , 3 ( mod 4 ) , and let w ( s ) , for the integers s ≥ 1 with s 2 ∣ D , be real numbers with w ( 1 ) = 1 and | w ( s ) | ≤ s 2 . Then
( − 1 ) D + 1 ∑ s 2 ∣ D w ( s ) c ^ ( D / s 2 ) ≥ | c ^ ( D ) | − W ( D ) > 1 2 | c ^ ( D ) | .
Proof. 
The term s = 1 is ( − 1 ) D + 1 c ^ ( D ) = | c ^ ( D ) | by Lemma 2, and the others add up to at most W ( D ) in absolute value. By (13), W ( D ) ≤ W * ( D ) < 1 2 G ( D ) ≤ 1 2 | c ^ ( D ) | . □

6. Four-Dimensional Dyons

Proof of Theorem A.
By Lemma 4, Δ ≡ 0 , 3 ( mod 4 ) , so Δ > 0 forces Δ ≥ 3 ; s 2 ∣ Δ for every s ∣ ℓ 1 ℓ 2 ; and ( − 1 ) Q · P + 1 = ( − 1 ) Δ + 1 . Hence
d ( Q , P ) = ( − 1 ) Δ + 1 ∑ s 2 ∣ Δ w ( s ) c ^ ( Δ / s 2 ) , w ( s ) = s , s ∣ ℓ 1 ℓ 2 , 0 , otherwise ,
and 0 ≤ w ( s ) ≤ s ≤ s 2 , w ( 1 ) = 1 . Corollary 1 gives d ( Q , P ) > 1 2 | c ^ ( Δ ) | . □
The proof uses neither the parity of s nor the sign of w ( s ) ; only | w ( s ) | ≤ s 2 enters, and the weights of (2) are smaller by a factor s. By Lemma 1 the index depends on the charges only through Δ , ℓ 1 ℓ 2 and the parity of Q · P , which is the parity of Δ .
Table 1 lists the index for small charges in Sen’s normal form of Section 2.3, realised as Q = ℓ 1 ( e 1 + n f 1 ) , P = e 2 + m f 2 + p f 1 in two hyperbolic planes, so that 1 2 Q 2 = ℓ 1 2 n , 1 2 P 2 = m , Q · P = ℓ 1 p , and Q 0 ∧ P 0 is primitive because its ( e 1 , e 2 ) entry is 1. The terms of (2) are listed with their signs. A negative term occurs exactly when s is even and Δ / s 2 has the parity opposite to Δ , as for Δ = 12 , 28 , 44 with s = 2 and Δ = 48 with s = 4 ; in every case it is a tiny fraction of the leading term. Figure 5 shows the growth of the index.8

7. Five-Dimensional Black Holes

Fix Q 1 , Q 5 , n ≥ 1 with gcd ( Q 1 , Q 5 ) = 1 , put M = Q 1 Q 5 n and g = gcd ( n , Q 1 Q 5 ) , and for l ∈ Z write D ( l ) = 4 M − l 2 and
ε ( l ) = ( − 1 ) l + 1 F ( l ) ,
with F as in (3). If s ∣ g then s ∣ n and s ∣ Q 1 Q 5 , so s 2 ∣ M ; if moreover s ∣ l , then s 2 ∣ D ( l ) . Since F ( − l ) = F ( l ) , ε is an even function of l. Note that D ( l ) ≡ l ( mod 2 ) and D ( l ) ≡ 0 , 3 ( mod 4 ) .
Lemma 8.
1.
If D ( l ) ≥ 3 , then ε ( l ) ≥ | c ^ ( D ( l ) ) | − W ( D ( l ) ) > 1 2 | c ^ ( D ( l ) ) | .
2.
If D ( l ) = 0 , then ε ( l ) = 2 σ gcd ( g , l ) ; if D ( l ) = − 1 , then ε ( l ) = 1 .
3.
If D ( l ) < − 1 , then ε ( l ) = ( − 1 ) l + 1 s 0 when D ( l ) = − s 0 2 with s 0 ≥ 2 and s 0 ∣ gcd ( g , l ) , and ε ( l ) = 0 otherwise. In the first case l ≡ s 0 ( mod 2 ) , so ε ( l ) = s 0 > 0 for odd l and ε ( l ) = − s 0 < 0 for even l.
Proof. 
Since ( − 1 ) l + 1 = ( − 1 ) D ( l ) + 1 , (1) is Corollary 1 with w ( s ) = s for s ∣ gcd ( g , l ) and w ( s ) = 0 otherwise. If D ( l ) = 0 then l is even and every s ∣ gcd ( g , l ) contributes s c ^ ( 0 ) = − 2 s , so ε ( l ) = 2 σ ( gcd ( g , l ) ) . If D ( l ) = − 1 , only s = 1 has s 2 ∣ D ( l ) , l is odd and ε ( l ) = c ^ ( − 1 ) = 1 . If D ( l ) < − 1 , a term s of (3) is nonzero only if D ( l ) / s 2 ≥ − 1 , that is s 2 ≥ − D ( l ) ; since s 2 divides D ( l ) , this means s 2 = − D ( l ) . So there is at most one nonzero term, s = s 0 = − D ( l ) , present when s 0 is an integer dividing gcd ( g , l ) , and it equals s 0 c ^ ( − 1 ) = s 0 . Finally l ≡ D ( l ) = − s 0 2 ≡ s 0 ( mod 2 ) . □
The values in (3) are the polar terms. They are nonzero only when g > 1 , and they are small, but they enter (4) and (5) with fixed signs and have to be controlled. The next lemma ties their size to the discriminant of a nearby l.
Lemma 9.
Let 0 ≤ l 1 < l 2 with D ( l 1 ) ≥ 3 , and suppose D ( l 2 ) = − s 0 2 with s 0 ≥ 2 and s 0 ∣ gcd ( g , l 2 ) . Put t = l 2 − l 1 and λ = l 2 / s 0 . Then λ is odd, λ ≥ 3 , and
D ( l 1 ) + t 2 = s 0 u , u = 2 t λ − s 0 ,
where u is a positive integer with u ≡ s 0 ( mod 2 ) . In particular s 0 ≤ D ( l 1 ) + t 2 , and s 0 ≤ 1 2 D ( l 1 ) + t 2 when s 0 is even.
Proof. 
As noted above, s 0 ∣ g gives s 0 2 ∣ M ; put μ = M / s 0 2 ≥ 1 . Dividing 4 M − l 2 2 = − s 0 2 by s 0 2 gives λ 2 = 4 μ + 1 ≥ 5 , so λ is odd and at least 3. Next,
D ( l 1 ) = 4 M − ( l 2 − t ) 2 = 4 M − l 2 2 + 2 t l 2 − t 2 = − s 0 2 + 2 t s 0 λ − t 2 ,
which is the identity. Its left side is positive, so u ≥ 1 , and u = 2 t λ − s 0 ≡ s 0 ( mod 2 ) . If s 0 is even then u ≥ 2 . □
For example Q 1 = 11 , Q 5 = 1 , n = 22 give M = 242 and g = 11 ; at l = 33 , D ( 33 ) = 968 − 1089 = − 121 and ε ( 33 ) = 11 . With l 1 = 31 we have D ( 31 ) = 7 , t = 2 , λ = 3 and 7 + 4 = 11 · 1 . Lemma 9 is sharp here. Figure 6 shows this charge.
Since ( − 1 ) J + 1 ( − 1 ) r = ( − 1 ) ( J − r ) + 1 , the definition (4) becomes
d ˜ ( n , Q 1 , Q 5 , J ) = ∑ r = − 2 2 4 2 − r ε ( J − r ) .
Proof of Theorem B.
Since d ˜ is even in J, we may assume J ≥ 0 . Put D = D ( J ) ; then D ≥ 3 , because D > 0 and D ≡ 0 , 3 ( mod 4 ) . The term r = 0 of (16) is 6 ε ( J ) > 3 | c ^ ( D ) | by Lemma 8(1). The other terms have l = J − r with r ≠ 0 . If D ( l ) ≥ − 1 the term is nonnegative by Lemma 8. If D ( l ) < − 1 then l 2 > J 2 , so | l | = J + t with t ∈ { 1 , 2 } ; the term is negative only if | l | is even and D ( l ) = − s 0 2 with s 0 even, and then, by Lemma 9 with l 1 = J and l 2 = | l | , it is at least − 4 2 − r · 1 2 ( D + 4 ) . For J ≥ 1 the values | l | > J are J + 1 and J + 2 , with coefficients 4 and 1; for J = 0 the even ones are l = ± 2 , with coefficients 1 and 1. In all cases
d ˜ ( n , Q 1 , Q 5 , J ) > 3 | c ^ ( D ) | − 5 2 ( D + 4 ) .
For D = 3 , 4 the right side is 24 − 17.5 and 36 − 20 . For D ≥ 7 put k = ⌊ ( D + 1 ) / 4 ⌋ ≥ 2 ; then D ≤ 4 k , and by Lemma 2 and Lemma 3, | c ^ ( D ) | ≥ a ( k ) ≥ k + 5 5 ≥ ( k + 5 ) ( k + 1 ) , since ( k + 4 ) ( k + 3 ) ( k + 2 ) ≥ 120 . So 3 | c ^ ( D ) | ≥ 3 k 2 + 18 k + 15 > 10 k + 10 ≥ 5 2 ( D + 4 ) . □
In the index at fixed J the terms no longer have the same sign, and the monotonicity of the coefficients has to absorb the difference.
Lemma 10.
Let 0 ≤ l 1 < l 2 with l 1 ≡ l 2 ( mod 2 ) and D ( l 1 ) ≥ 3 .
1.
If D ( l 2 ) ≥ − 1 , then ε ( l 1 ) > ε ( l 2 ) .
2.
If D ( l 2 ) < − 1 and l 2 − l 1 ≤ 6 , then ε ( l 1 ) > ε ( l 2 ) .
Proof. 
Put D 1 = D ( l 1 ) and D 2 = D ( l 2 ) . Then D 1 − D 2 = ( l 2 − l 1 ) ( l 2 + l 1 ) is a product of two even numbers, both at least 2, so D 2 ≤ D 1 − 4 and D 2 ≡ D 1 ( mod 4 ) . By Lemma 3, | c ^ ( D 1 ) | − | c ^ ( D 2 ) | ≥ G ( D 1 ) whenever D 2 ≥ − 1 .
(1) If D 2 ≥ 3 , Lemma 8 and (13) give
ε ( l 1 ) − ε ( l 2 ) ≥ | c ^ ( D 1 ) | − | c ^ ( D 2 ) | − W ( D 1 ) − W ( D 2 ) ≥ G ( D 1 ) − 2 W * ( D 1 ) > 0 .
If D 2 = − 1 , then ε ( l 2 ) = 1 = | c ^ ( − 1 ) | and ε ( l 1 ) − ε ( l 2 ) ≥ G ( D 1 ) − W * ( D 1 ) > 0 .
If D 2 = 0 , then l 2 2 = 4 M and ε ( l 2 ) = 2 σ ( m ) with m = gcd ( g , l 2 ) ≤ l 2 . Since l 1 ≤ l 2 − 2 , D 1 ≥ l 2 2 − ( l 2 − 2 ) 2 = 4 l 2 − 4 , so m ≤ l 2 ≤ 1 4 ( D 1 + 4 ) . With | c ^ ( 0 ) | = 2 and σ ( m ) ≤ 1 2 m ( m + 1 ) (Appendix A),
ε ( l 1 ) − ε ( l 2 ) ≥ | c ^ ( D 1 ) | − | c ^ ( 0 ) | − W ( D 1 ) − 2 σ ( m ) − 1 ≥ G ( D 1 ) − W * ( D 1 ) − m ( m + 1 ) ,
and m ( m + 1 ) ≤ 1 16 ( D 1 + 4 ) ( D 1 + 8 ) . The right side is positive by (14).
(2) By Lemma 8(3) we may assume ε ( l 2 ) > 0 , since otherwise ε ( l 2 ) ≤ 0 < ε ( l 1 ) . Then l 2 is odd, D ( l 2 ) = − s 0 2 with s 0 odd, s 0 ≥ 3 , and ε ( l 2 ) = s 0 . Also l 1 is odd, so D 1 is odd and D 1 ≡ 3 ( mod 4 ) ; and t = l 2 − l 1 ∈ { 2 , 4 , 6 } . Lemma 9 gives s 0 ≤ D 1 + t 2 ≤ D 1 + 36 .
Suppose D 1 ≥ 11 and write D 1 = 4 k − 1 with k ≥ 3 . For k = 3 , ε ( l 1 ) > 1 2 | c ^ ( 11 ) | = 76 > 47 . For k ≥ 4 , ε ( l 1 ) > 1 2 | c ^ ( D 1 ) | ≥ 1 2 a ( k ) ≥ 1 2 k + 5 5 ; this is 63 at k = 4 and grows by 1 2 k + 4 4 ≥ 35 with each step, while D 1 + 36 = 4 k + 35 is 51 at k = 4 and grows by 4. In both cases ε ( l 1 ) > D 1 + 36 ≥ s 0 .
Suppose D 1 ∈ { 3 , 7 } . By Lemma 9, s 0 u = D 1 + t 2 and s 0 + u = 2 t λ with λ odd and λ ≥ 3 , so s 0 + u is 4 λ , 8 λ or 12 λ for t = 2 , 4 , 6 . The products D 1 + t 2 are 7 , 19 , 39 for D 1 = 3 and 11 , 23 , 43 for D 1 = 7 ; their factorisations s 0 u have s 0 + u ∈ { 8 } , { 20 } , { 16 , 40 } and { 12 } , { 24 } , { 44 } . The only sums of the required form are 12 = 4 · 3 and 24 = 8 · 3 , both for D 1 = 7 , with s 0 = 11 and s 0 = 23 . Then ε ( l 1 ) ≥ | c ^ ( 7 ) | − W ( 7 ) = 39 > 23 ≥ s 0 , and for D 1 = 3 no polar value occurs. □
The case D 2 = 0 is the only place where a divisor sum enters with no small factor in front of it: all its terms carry the same sign as the leading term of ε ( l 2 ) , and they are subtracted. This is the reason for the second inequality of Proposition 1.
Proof of Theorem C.
By (5) and (16), collecting the coefficient of each ε ( J − u ) ,
d ( n , Q 1 , Q 5 , J ) = ε ( J − 2 ) + 4 ε ( J − 1 ) + 5 ε ( J ) − 5 ε ( J + 2 ) − 4 ε ( J + 3 ) − ε ( J + 4 ) ,
since the coefficients of ( ζ 1 / 2 − ζ − 1 / 2 ) 4 ( 1 − ζ − 2 ) are 1 , − 4 , 5 , 0 , − 5 , 4 , − 1 . Hence
d ( n , Q 1 , Q 5 , J ) = ∑ r = 0 2 c r ε ( | J − r | ) − ε ( J + 2 + r ) , ( c 0 , c 1 , c 2 ) = ( 5 , 4 , 1 ) ,
using that ε is even. Put l 1 = | J − r | and l 2 = J + 2 + r . Then l 1 < l 2 , the two numbers have the same parity, and l 2 − l 1 is 2 r + 2 ≤ 6 when J ≥ r and 2 J + 2 ≤ 4 when J < r . If J ≥ r then l 1 ≤ J and D ( l 1 ) ≥ D ( J ) ≥ 3 . If J < r then ( J , r ) is ( 0 , 1 ) , ( 0 , 2 ) or ( 1 , 2 ) and l 1 is 1, 2 or 1; here D ( 1 ) = 4 M − 1 ≥ 3 , and D ( 2 ) = 4 M − 4 ≥ 3 unless M = 1 . In all these cases Lemma 10 shows that the bracket is positive. In the remaining case M = 1 , J = 0 , r = 2 , the bracket is ε ( 2 ) − ε ( 4 ) with D ( 2 ) = 0 and D ( 4 ) = − 12 , which is not minus a square, so it equals 2 σ ( 1 ) − 0 = 2 . Every bracket is positive, and so is d. □
Table 2 lists the five-dimensional indices for small charges, together with three values for the charge M = 242 , g = 11 of Figure 6. At ( Q 1 , Q 5 , n ) = ( 2 , 1 , 4 ) and J = 4 the polar value ε ( 6 ) = − 2 enters (16) with coefficient 1 and lowers d ˜ ( 4 ) from 36446 to 36444; at M = 242 and J = 29 the polar value ε ( 33 ) = 11 is subtracted with coefficient 1 in (5), against ε ( 27 ) ≈ 1.9 · 10 16 . Figure 7 shows the index at fixed J for g = 1 ; at fixed M it decreases with J, in line with the pairing in the last proof.

8. Remarks

8.1. Sharpness

The divisor terms are exponentially smaller than the leading term for large discriminants: by Lemma 5 the term s = 2 of W ( D ) is at most e π 2 D ( 1 + o ( 1 ) ) , while log | c ^ ( D ) | = π D ( 1 + o ( 1 ) ) [30]. The content of Proposition 1 is that this separation already holds at the smallest discriminants, where it is tight: at D = 12 the term s = 2 takes 16 of the 208 units of the leading term in (2), and 2 W * ( 12 ) is 42 % of G ( 12 ) . The polar values are smaller still, at most linear in the discriminant by Lemma 9, but they are not negligible at small charges: the case D 1 = 7 , s 0 = 23 of Lemma 10 uses more than half of ε ( l 1 ) = 39 . The lower bound of Lemma 6 grows like e 2 6 n instead of the true e 2 π n , which is why the analytic argument starts only at k = 90 .

8.2. The U-Duality Invariant Extension

For charges with gcd ( Q 1 , Q 5 , n ) = 1 , Maldacena, Moore and Strominger proposed the U-duality invariant extension of (3) in which the divisor s carries the weight N ( s ) s , with N ( s ) the number of common divisors of n , Q 1 , Q 5 , s , n Q 1 / s , n Q 5 / s , Q 1 Q 5 / s , n Q 1 Q 5 / s 2 , the sum running over the s for which l / s and all these quotients are integers [21]. For such charges N ( s ) = 1 , since a common divisor of n , Q 1 , Q 5 is 1, so the weights again lie in { 0 , s } , are even in l, and are supported on divisors s of l with s 2 ∣ M . The proofs of Lemmas 8, 9 and 10 use nothing else: in Lemma 8(2) the value at D ( l ) = 0 becomes 2 ∑ s over a set of divisors of l, which is at most 2 σ ( l ) ≤ l ( l + 1 ) , and the bound l 2 ≤ 1 4 ( D 1 + 4 ) is unchanged. Hence, if the extension holds, the indices it predicts satisfy Theorems B and C for every charge with gcd ( Q 1 , Q 5 , n ) = 1 .

8.3. Charges of Higher Torsion

For four-dimensional charges with gcd ( ℓ 1 , ℓ 2 ) > 1 no formula for the index has been established [30], and the theorems say nothing about them. Corollary 1 reduces their positivity to a structural statement: any index of the form ( − 1 ) Δ + 1 ∑ s 2 ∣ Δ w ( s ) c ^ ( Δ / s 2 ) with w ( 1 ) = 1 and | w ( s ) | ≤ s 2 is positive, whatever the signs of the other weights. Any future formula of this shape, with the divisor weights bounded by s 2 , would therefore give positive indices at once.9

Funding

None.

Data Availability Statement

All data that support the findings of this study are contained in the article: Table A1 lists the numbers a ( n ) used in the finite part of the proof, and Appendix B describes the verification. The program that repeats the verification, computes the tables and draws the figures is archived at doi:10.5281/zenodo.24122289.

Conflicts of Interest

None.

Ethics Statement

This study involves no human participants, animals or personal data; no ethical approval was required.

Appendix A Elementary Facts

Appendix Lagrange’s Identity

For vectors a , b , c , e ∈ R N ,
( a · c ) ( b · e ) − ( a · e ) ( b · c ) = ∑ i < j ( a i b j − a j b i ) ( c i e j − c j e i ) .
Both sides are bilinear in ( a , b ) and in ( c , e ) , so it suffices to take basis vectors a = e p , b = e q , c = e r , e = e s . The left side is δ p r δ q s − δ p s δ q r . On the right, the term i < j is nonzero only if { p , q } = { i , j } = { r , s } with p ≠ q , and it is then + 1 if ( p , q ) = ( r , s ) and − 1 if ( p , q ) = ( s , r ) . The two sides agree. In Lemma 4 this is applied with a = Q , b = P , c = L Q , e = L P , using Q · L P = P · L Q because L is symmetric.

Appendix Two inequalities

For 0 ≤ y ≤ 1 2 , the function f ( y ) = log ( 1 − y ) + y + y 2 has f ( 0 ) = 0 and f ′ ( y ) = y ( 1 − 2 y ) 1 − y ≥ 0 , so log ( 1 − y ) ≥ − y − y 2 . For m ≥ 1 the divisors of m are distinct elements of { 1 , … , m } , so σ ( m ) ≤ 1 + 2 + … + m = 1 2 m ( m + 1 ) .

Appendix Jacobi’s triple product

For | q | < 1 and x ≠ 0 ,
∏ n ≥ 1 ( 1 − q 2 n ) ( 1 + x q 2 n − 1 ) ( 1 + x − 1 q 2 n − 1 ) = ∑ m ∈ Z x m q m 2
[1]. Replace q by q 1 / 2 and x by − ζ q 1 / 2 . With r = m + 1 2 the sum becomes
∑ m ( − 1 ) m ζ m q ( m 2 + m ) / 2 = ζ − 1 / 2 q − 1 / 8 ∑ r ∈ Z + 1 2 ( − 1 ) r − 1 2 q r 2 / 2 ζ r = i ζ − 1 / 2 q − 1 / 8 ϑ 1 ( τ , z ) .
The product becomes ∏ n ( 1 − q n ) ( 1 − ζ q n ) ( 1 − ζ − 1 q n − 1 ) = ( 1 − ζ − 1 ) ∏ n ( 1 − q n ) ( 1 − ζ q n ) ( 1 − ζ − 1 q n ) . Since ζ 1 / 2 ( 1 − ζ − 1 ) = ζ 1 / 2 − ζ − 1 / 2 , this gives (6), and squaring and dividing by η 6 = q 1 / 4 ∏ ( 1 − q n ) 6 gives (7).

Appendix A bound for a(n) without modular forms

For u > 0 ,
log P ( e − u ) = 6 ∑ m ≥ 1 ∑ r ≥ 1 e − u m r r = 6 ∑ r ≥ 1 1 r · e − u r 1 − e − u r ≤ 6 ∑ r ≥ 1 1 u r 2 = π 2 u ,
because e − v / ( 1 − e − v ) = 1 / ( e v − 1 ) ≤ 1 / v . Hence a ( n ) ≤ e n u P ( e − u ) ≤ e n u + π 2 / u , and the choice u = π / n gives a ( n ) ≤ e 2 π n , the second bound of Lemma 5.

Appendix B The Finite Range

The inequalities (13) and (14) for 3 ≤ D ≤ 358 involve integers only, and their verification consists of four steps.
(F1)
The numbers a ( n ) , 0 ≤ n ≤ 90 , are computed from the recursion n a ( n ) = 6 ∑ i = 1 n σ ( i ) a ( n − i ) of Lemma 6, or equivalently by multiplying out six copies of ∏ ( 1 − q n ) − 1 ; they are listed in Table A1.
(F2)
For − 1 ≤ E ≤ 358 , | c ^ ( E ) | is computed from (9).
(F3)
For 1 ≤ E ≤ 358 , W ( E ) is computed from its definition, and W * ( D ) as a running maximum.
(F4)
For each D in the range with D ≡ 0 , 3 ( mod 4 ) , the integers G ( D ) , 2 W * ( D ) and 16 G ( D ) − 16 W * ( D ) − ( D + 4 ) ( D + 8 ) are compared.
Table A2 lists the quantities for D ≤ 63 ; the margin grows rapidly, as Figure 3 shows. The same program checks, as controls that the proofs do not use: the formula (9) against the direct expansion of − ϑ 1 2 / η 6 up to q 30 and against the product (7) up to q 12 ; the bounds of Section 4 for n ≤ 605 ; the conclusions of Theorems A to C by evaluating (2), (4) and (5) directly on 1722 four-dimensional charges in Sen’s normal form and 5750 five-dimensional charges; all 868 five-dimensional charges with M ≤ 600 in which a polar value enters, together with the bound of Lemma 9; and 3195 charges for the extension of [21].
Table A1. The numbers a ( n ) of partitions of n into parts of six colours, 0 ≤ n ≤ 90 .
Table A1. The numbers a ( n ) of partitions of n into parts of six colours, 0 ≤ n ≤ 90 .
n a ( n ) n a ( n ) n a ( n )
0 1 31 30227301468 62 13799370667455450
1 6 32 49522366656 63 19845194082605218
2 27 33 80590617278 64 28465163415636075
3 98 34 130307480730 65 40724826208003224
4 315 35 209396014650 66 58118725494437321
5 918 36 334492740175 67 82738453811650308
6 2492 37 531278722086 68 117504789092547003
7 6372 38 839204245278 69 166487707267085750
8 15525 39 1318587244800 70 235346673673043703
9 36280 40 2061234013149 71 331935256651796730
10 81816 41 3206265612102 72 467129825087621531
11 178794 42 4963610011174 73 655963278944277444
12 380051 43 7648740634158 74 919173274601444100
13 788004 44 11733857681850 75 1285312714955089064
14 1597725 45 17923026108974 76 1793621662235878482
15 3174210 46 27262103713383 77 2497928656596353478
16 6190182 47 41299026921900 78 3471941462435113530
17 11867310 48 62316761884412 79 4816410159748817880
18 22395359 49 93670787579400 80 6668809352618868378
19 41650050 50 140276509484724 81 9216404411744168100
20 76413078 51 209311092717496 82 12713856676088969832
21 138421358 52 311221082489247 83 17506907536752046098
22 247783113 53 461164908282750 84 24064191719868828650
23 438616728 54 681072218736560 85 33019905786901987212
24 768291650 55 1002575819281092 86 45230951219325322689
25 1332444330 56 1471174898831406 87 61853350983484206818
26 2289213495 57 2152131323752182 88 84444293827166272965
27 3898064226 58 3138799331684016 89 115098208842781661400
28 6581591157 59 4564363803456240 90 156627966849287037846
29 11023247880 60 6618341988415724
30 18320981569 61 9569726912844018
Table A2. The quantities of Proposition 1 for 3 ≤ D ≤ 63 .
Table A2. The quantities of Proposition 1 for 3 ≤ D ≤ 63 .
D | c ^ ( D ) | G ( D ) W ( D ) W * ( D ) D | c ^ ( D ) | G ( D ) W ( D ) W * ( D )
3 8 7 0 0 35 69168 40269 0 224
4 12 10 0 0 36 85500 49412 108 224
7 39 31 0 0 39 159372 90204 0 224
8 56 44 0 0 40 195312 109812 0 224
11 152 113 0 0 43 355224 195852 0 224
12 208 152 32 32 44 431984 236672 608 608
15 513 361 0 32 47 768885 413661 0 608
16 684 476 48 48 48 928720 496736 960 960
19 1560 1047 0 48 51 1621296 852411 0 960
20 2032 1348 0 48 52 1946352 1017632 0 960
23 4382 2822 0 48 55 3339201 1717905 0 960
24 5616 3584 0 48 56 3986656 2040304 0 960
27 11552 7170 72 72 59 6732232 3393031 0 960
28 14592 8976 156 156 60 7997184 4010528 2052 2052
31 28899 17347 0 156 63 13311450 6579218 351 2052
32 36088 21496 224 224

References

  1. G. E. Andrews, The Theory of Partitions. In Encyclopedia of Mathematics and its Applications 2; Cambridge University Press: Cambridge, 1984. [CrossRef]
  2. Apostol, T. M. Modular Functions and Dirichlet Series in Number Theory. In Graduate Texts in Mathematics 41, 2nd ed.; Springer: New York, 1990. [Google Scholar] [CrossRef]
  3. Bachas, C.; Kiritsis, E. F4 terms in N=4 string vacua. Nucl. Phys. B Proc. Suppl. 1997, 55, 194–199. [Google Scholar] [CrossRef]
  4. Bachas, C.; Fabre, C.; Kiritsis, E.; Obers, N. A.; Vanhove, P. Heterotic/type-I duality and D-brane instantons. Nucl. Phys. B 1998, 509, 33–52. [Google Scholar] [CrossRef]
  5. Bhattacharjee, D. Generalization of quartic and quintic Calabi–Yau manifolds fibered by polarized K3 surfaces. Res. Sq. 2022. [Google Scholar] [CrossRef] [PubMed]
  6. Bhattacharjee, D.; Nandi, P. Enriques and Kummer surfaces arising from K3 involutions; Preprints, 2026. [Google Scholar] [CrossRef]
  7. Breckenridge, J. C.; Myers, R. C.; A. W. Peet, C. Vafa, D-branes and spinning black holes. Phys. Lett. B 1997, 391, 93–98. [Google Scholar] [CrossRef]
  8. Bringmann, K.; Murthy, S. On the positivity of black hole degeneracies in string theory. Commun. Number Theory Phys. 2013, 7, 15–56. [Google Scholar] [CrossRef]
  9. Chowdhury, A.; Garavuso, R. S.; Mondal, S.; Sen, A. BPS state counting in N=8 supersymmetric string theory for pure D-brane configurations. J. High Energy Phys. 2014, no. 10, 186. [Google Scholar] [CrossRef]
  10. Chowdhury, A.; Garavuso, R. S.; Mondal, S.; Sen, A. Do all BPS black hole microstates carry zero angular momentum? J. High Energy Phys. 2016, no. 4, 082. [Google Scholar] [CrossRef]
  11. Chowdhury, A.; Maji, S. Counting N=8 black holes as algebraic varieties. J. High Energy Phys. 2024, 2024(no. 5), 091. [Google Scholar] [CrossRef]
  12. Cremmer, E.; Julia, B. The SO(8) supergravity. Nucl. Phys. B 1979, 159, 141–212. [Google Scholar] [CrossRef]
  13. Dabholkar, A.; Gomes, J.; Murthy, S. Quantum black holes, localization and the topological string. J. High Energy Phys. 2011, no. 6, 019. [Google Scholar] [CrossRef]
  14. Dabholkar, A.; Gomes, J.; Murthy, S.; Sen, A. Supersymmetric index from black hole entropy. J. High Energy Phys. 2011, no. 4, 034. [Google Scholar] [CrossRef]
  15. Dijkgraaf, R.; Moore, G.; Verlinde, E.; Verlinde, H. Elliptic genera of symmetric products and second quantized strings. Commun. Math. Phys. 1997, 185, 197–209. [Google Scholar] [CrossRef]
  16. Eichler, M.; Zagier, D. The Theory of Jacobi Forms. In Progress in Mathematics 55; Birkhäuser, Boston, 1985. [Google Scholar] [CrossRef]
  17. Gaiotto, D.; Strominger, A.; Yin, X. New connections between 4D and 5D black holes. J. High Energy Phys. 2006, no. 2, 024. [Google Scholar] [CrossRef]
  18. Hardy, G. H.; Ramanujan, S. Asymptotic formulae in combinatory analysis. Proc. Lond. Math. Soc. (2) 1918, 17, 75–115. [Google Scholar] [CrossRef]
  19. Kallosh, R.; Kol, B. E7 symmetric area of the black hole horizon. Phys. Rev. D. 1996, 53, R5344–R5348. [Google Scholar] [CrossRef] [PubMed]
  20. Kiritsis, E.; Obers, N. A. Heterotic/type-I duality in D. J. High Energy Phys. 1997, no. 10, 004. [Google Scholar] [CrossRef]
  21. Maldacena, J.; Moore, G.; Strominger, A. Counting BPS black holes in toroidal type II string theory. arXiv 1999, arXiv:hep. [Google Scholar]
  22. Rademacher, H. On the expansion of the partition function in a series. Ann. Math. (2) 1943, 44, 416–422. [Google Scholar] [CrossRef]
  23. Rademacher, H. Topics in Analytic Number Theory. In Grundlehren der mathematischen Wissenschaften 169; Springer: Berlin, 1973. [Google Scholar] [CrossRef]
  24. Robbins, H. A remark on Stirling’s formula. Amer. Math. Mon. 1955, 62, 26–29. [Google Scholar] [CrossRef]
  25. Sadhu, R.; Bhattacharjee, D.; Bhattacharya, S.; Thakur, S. N. E8×E8 heterotic gravity and M-theory. SSRN 2026. [Google Scholar] [CrossRef]
  26. Sen, A. Black hole entropy function, attractors and precision counting of microstates. Gen. Relativ. Gravit. 2008, 40, 2249–2431. [Google Scholar] [CrossRef]
  27. Sen, A. N=8 dyon partition function and walls of marginal stability. J. High Energy Phys. 2008, no. 7, 118. [Google Scholar] [CrossRef]
  28. Sen, A. Quantum entropy function from AdS2/CFT1 correspondence. Int. J. Mod. Phys. A 2009, 24, 4225–4244. [Google Scholar] [CrossRef]
  29. Sen, A. Arithmetic of quantum entropy function. J. High Energy Phys. 2009, no. 8, 068. [Google Scholar] [CrossRef]
  30. Sen, A. Arithmetic of N=8 black holes. J. High Energy Phys. 2010, no. 2, 090. [Google Scholar] [CrossRef]
  31. Sen, A. How do black holes predict the sign of the Fourier coefficients of Siegel modular forms? Gen. Relativ. Gravit. 2011, 43, 2171–2183. [Google Scholar] [CrossRef]
  32. Sen, A. Logarithmic corrections to rotating extremal black hole entropy in four and five dimensions. Gen. Relativ. Gravit. 2012, 44, 1947–1991. [Google Scholar] [CrossRef]
  33. Shih, D.; Strominger, A.; Yin, X. Counting dyons in N=8 string theory. J. High Energy Phys. 2006, no. 6, 037. [Google Scholar] [CrossRef]
  34. Strominger, A.; Vafa, C. Microscopic origin of the Bekenstein–Hawking entropy. Phys. Lett. B 1996, 379, 99–104. [Google Scholar] [CrossRef]
1
For N = 4 string theory the corresponding prediction concerns Siegel modular forms, and positivity holds there only after wall crossing is taken into account [31]. The N = 8 case is simpler in that the index is a finite divisor sum, but it is not a single coefficient.
2
Sen derives (2) from the counting formula of Maldacena, Moore and Strominger [21] and of Shih, Strominger and Yin [33] by a duality transformation to a normal form [30]; the condition gcd ( ℓ 1 , ℓ 2 ) = 1 is what that transformation needs. The integer d is − B 14 , the fourteenth helicity trace with the contribution of the fermion zero modes removed (Section 2.1). Chowdhury, Garavuso, Mondal and Sen recovered the first coefficients of (2) by counting the solutions of the supersymmetric vacuum equations of D-brane systems [9], and Chowdhury and Maji extended the count [11].
3
Equation (4) is [32] with the complete divisor sum of [14], in which Q 1 is replaced by Q 1 Q 5 . We keep the overall sign of [32], in which the prediction reads d ˜ > 0 and d > 0 ; the index of [14], defined with the extra factor ( − 1 ) J of [21], is − d ˜ . The sum in (3) runs over all s, including those for which ( 4 M − l 2 ) / s 2 = − 1 ; these terms matter in Section 7.
4
The shift is the choice of the helicity of the Clifford vacuum. Each oscillator pair gives the two states of helicities h 0 and h 0 + 1 2 with opposite statistics, hence y 2 h 0 − y 2 h 0 + 1 .
5
The quartic invariant Δ is the Cremmer–Julia invariant of N = 8 supergravity [12,19]; the transformations used here form the subgroup S O ( 4 , 4 ; Z ) × S L ( 2 , Z ) of the U-duality group that acts within the D-brane charges considered [30].
6
The asymptotic size of c ^ ( D ) follows from (8) and the circle method of Hardy, Ramanujan and Rademacher [18,22,23]: | c ^ ( D ) | ∼ const · D − 2 e π D [14]. The proof below uses only explicit inequalities and no asymptotic expansion.
7
Lemma 6 requires k ≥ 50 , which holds. The inequalities (15) already hold for k ≥ 84 ; the choice 90 leaves a margin of 0.7 in the tightest one, see Figure 4.
8
The rows with ℓ 1 = 2 , 3 have the same Δ and ℓ 1 ℓ 2 as rows with ℓ 1 = 1 and the same index, as Lemma 1 and (2) require.
9
Sen remarks that the condition gcd ( ℓ 1 , ℓ 2 ) = 1 might be relaxed with the results of [33] for non-primitive charge vectors [30]; no such formula has been derived.
Figure 1. The coefficients c ^ ( 4 k − l 2 ) of ϕ − 2 , 1 for k ≤ 8 , on a logarithmic scale. The sign is ( − 1 ) D + 1 , and | c ^ | depends only on D = 4 k − l 2 .
Figure 1. The coefficients c ^ ( 4 k − l 2 ) of ϕ − 2 , 1 for k ≤ 8 , on a logarithmic scale. The sign is ( − 1 ) D + 1 , and | c ^ | depends only on D = 4 k − l 2 .
Preprints 243591 g001
Figure 2. From front to back: the upper bound of Lemma 5, the true values log a ( n ) , the comparison series log b ( n ) of Lemma 6, and its explicit lower bound 2 6 n − 6 − log n , for n ≤ 160 . The lower bound grows like 2 6 n instead of 2 π n ; Proposition 1 needs only that it beats the upper bound taken at a quarter of the argument.
Figure 2. From front to back: the upper bound of Lemma 5, the true values log a ( n ) , the comparison series log b ( n ) of Lemma 6, and its explicit lower bound 2 6 n − 6 − log n , for n ≤ 160 . The lower bound grows like 2 6 n instead of 2 π n ; Proposition 1 needs only that it beats the upper bound taken at a quarter of the argument.
Preprints 243591 g002
Figure 5. The index d ( Q , P ) for Q · P = 0 , ℓ 1 = 1 and 1 ≤ 1 2 Q 2 , 1 2 P 2 ≤ 7 , on a logarithmic scale; the colour shows ℓ 2 . The diagonal carries the largest ℓ 2 and hence the most divisor terms, without visible effect on the size.
Figure 5. The index d ( Q , P ) for Q · P = 0 , ℓ 1 = 1 and 1 ≤ 1 2 Q 2 , 1 2 P 2 ≤ 7 , on a logarithmic scale; the colour shows ℓ 2 . The diagonal carries the largest ℓ 2 and hence the most divisor terms, without visible effect on the size.
Preprints 243591 g005
Figure 6. The pairing in the proof of Theorem C for M = 242 , g = 11 , J = 29 . The green points l = J , J − 1 , J − 2 are paired with the red points l = J + 2 , J + 3 , J + 4 ; the arrows carry the weights c r . The point l = 33 lies at D = − 121 = − 11 2 and carries the polar value ε ( 33 ) = 11 ; against l 1 = 31 , where D = 7 and t = 2 , it attains the bound of Lemma 9.
Figure 6. The pairing in the proof of Theorem C for M = 242 , g = 11 , J = 29 . The green points l = J , J − 1 , J − 2 are paired with the red points l = J + 2 , J + 3 , J + 4 ; the arrows carry the weights c r . The point l = 33 lies at D = − 121 = − 11 2 and carries the polar value ε ( 33 ) = 11 ; against l 1 = 31 , where D = 7 and t = 2 , it attains the bound of Lemma 9.
Preprints 243591 g006
Figure 7. The index d ( n , Q 1 , Q 5 , J ) at fixed angular momentum for g = gcd ( n , Q 1 Q 5 ) = 1 , M = Q 1 Q 5 n ≤ 20 and all J ≥ 0 with 4 M − J 2 > 0 , on a logarithmic scale.
Figure 7. The index d ( n , Q 1 , Q 5 , J ) at fixed angular momentum for g = gcd ( n , Q 1 Q 5 ) = 1 , M = Q 1 Q 5 n ≤ 20 and all J ≥ 0 with 4 M − J 2 > 0 , on a logarithmic scale.
Preprints 243591 g007
Table 1. The index d ( Q , P ) of (2) for small charges with gcd ( ℓ 1 , ℓ 2 ) = 1 . The column “terms” lists ( − 1 ) Q · P + 1 s c ^ ( Δ / s 2 ) for the divisors s = 1 < s 2 < … of ℓ 1 ℓ 2 .
Table 1. The index d ( Q , P ) of (2) for small charges with gcd ( ℓ 1 , ℓ 2 ) = 1 . The column “terms” lists ( − 1 ) Q · P + 1 s c ^ ( Δ / s 2 ) for the divisors s = 1 < s 2 < … of ℓ 1 ℓ 2 .
1 2 Q 2 1 2 P 2 Q · P ℓ 1 ℓ 2 Δ terms d ( Q , P )
1 1 0 1 1 4 12 12
1 1 1 1 1 3 8 8
2 1 1 1 1 7 39 39
2 2 0 1 2 16 684 + 24 708
2 2 1 1 1 15 513 513
2 2 2 1 2 12 208 − 16 192
3 3 3 1 3 27 11552 + 24 11576
2 4 2 1 2 28 14592 − 78 14514
4 4 4 1 4 48 928720 + 416   − 32 929104
4 4 0 1 4 64 15740076 + 1368   + 48 15741492
6 6 6 1 6 108 10744220992 − 23104   + 624   − 48 10744198464
4 1 2 2 1 12 208 − 16 192
9 1 3 3 1 27 11552 + 24 11576
4 3 2 2 1 44 431984 − 304 431680
Table 2. The five-dimensional indices d ˜ ( J ) of (4) and d ( J ) of (5), with ε ( J ) , for small charges and J ≥ 0 , D ( J ) = 4 M − J 2 > 0 .
Table 2. The five-dimensional indices d ˜ ( J ) of (4) and d ( J ) of (5), with ε ( J ) , for small charges and J ≥ 0 , D ( J ) = 4 M − J 2 > 0 .
( Q 1 , Q 5 , n ) M g J D ( J ) ε ( J ) d ˜ ( J ) d ( J )
( 1 , 1 , 1 ) 1 1 0 4 12 140 84
( 1 , 1 , 1 ) 1 1 1 3 8 112 96
( 2 , 1 , 1 ) 2 1 0 8 56 672 384
( 2 , 1 , 1 ) 2 1 1 7 39 546 453
( 2 , 1 , 1 ) 2 1 2 4 12 288 272
( 1 , 1 , 3 ) 3 1 0 12 208 2576 1392
( 1 , 1 , 3 ) 3 1 1 11 152 2128 1704
( 1 , 1 , 3 ) 3 1 2 8 56 1184 1096
( 1 , 1 , 3 ) 3 1 3 3 8 424 416
( 2 , 1 , 2 ) 4 2 0 16 708 8736 4662
( 2 , 1 , 2 ) 4 2 1 15 513 7230 5691
( 2 , 1 , 2 ) 4 2 2 12 192 4074 3690
( 2 , 1 , 2 ) 4 2 3 7 39 1539 1476
( 3 , 1 , 3 ) 9 3 0 36 85536 1138736 511664
( 3 , 1 , 3 ) 9 3 1 35 69168 982248 690992
( 3 , 1 , 3 ) 9 3 2 32 36088 627072 531872
( 3 , 1 , 3 ) 9 3 3 27 11576 291256 270608
( 3 , 1 , 3 ) 9 3 4 20 2032 95200 92512
( 3 , 1 , 3 ) 9 3 5 11 152 20648 20464
( 2 , 1 , 4 ) 8 2 0 32 36200 477420 220304
( 2 , 1 , 4 ) 8 2 1 31 28899 409531 293413
( 2 , 1 , 4 ) 8 2 2 28 14514 257116 220672
( 2 , 1 , 4 ) 8 2 3 23 4382 116118 108678
( 2 , 1 , 4 ) 8 2 4 16 708 36444 35592
( 2 , 1 , 4 ) 8 2 5 7 39 7440 7409
( 4 , 1 , 4 ) 16 4 0 64 15741492 216932868 85682720
( 4 , 1 , 4 ) 16 4 1 63 13311450 191469951 122262875
( 4 , 1 , 4 ) 16 4 2 60 7996158 131250148 103670560
( 4 , 1 , 4 ) 16 4 3 55 3339201 69207076 61136658
( 4 , 1 , 4 ) 16 4 4 48 929104 27579588 25923846
( 4 , 1 , 4 ) 16 4 5 39 159372 8070418 7849856
( 4 , 1 , 4 ) 16 4 6 28 14514 1655742 1639092
( 4 , 1 , 4 ) 16 4 7 15 513 220562 219993
( 11 , 1 , 22 ) 242 11 29 127 124029778470 19462192309545443 19462068157996872
( 11 , 1 , 22 ) 242 11 30 68 30442464 82875898355028 82875867912364
( 11 , 1 , 22 ) 242 11 31 7 39 124151548571 124151548466
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.