A proof of the Thompson Moonshine Conjecture

In this paper we prove the existence of an infinite dimensional graded super-module for the finite sporadic Thompson group $Th$ whose McKay-Thompson series are weakly holomorphic modular forms of weight $\frac 12$ satisfying properties conjectured by Harvey and Rayhaun.

Thompson further observed that a similar phenomenon occurs when one considers combinations of χ j (g) for other g ∈ M. Based on theses observations, he conjectured in [42] that there should exist an infinite dimensional graded M-module that reflects these combinations. Conway and Norton [16] made this more precise (The so-called Monstrous Moonshine conjecture), conjecturing that for each conjugacy class of M there is an explicit associated genus 0 subgroup of SL 2 (R) whose normalized Hauptmodul coincides with the so-called McKay-Thompson series (see Section 2.3 for a definition of this term) of the conjugacy class with respect to the module. An abstract proof of this conjecture (i.e. one whose construction of the module is done only implicitly in terms of the McKay-Thompson series) was announced by Atkin, Fong and Smith [22,40]. Their proof was based on an idea of Thompson strace Wm (g)q m is a specifically given weakly holomorphic modular form (see Section 2) of weight 1 2 in the Kohnen plus space.
• if |g| is odd, then the only other pole of order 3 4 is at the cusp 1 2|g| , otherwise there is only the pole at ∞. It vanishes at all other cusps. If |g| = 36, it suffices to assume that the Fourier expansion is of the form 2q −3 + χ 2 (g) + O(q 4 ).
The proof of Theorem 1.2, like the proofs of the Mathieu and Umbral Moonshine Conjectures, relies on the following idea. For each n ≥ 0, the function defined by where we write The rest of the paper is organized as follows. In Section 2, we recall some relevant definitions on supermodules, harmonic Maaß forms, and the construction of the (tentative) McKay-Thompson series in [28]. In Section 3, we show that these series are in fact all weakly holomorphic modular forms (instead of harmonic weak Maaß forms) with integer Fourier coefficients and that all the multiplicities m j in (1.1) are integers. Section 4 is concerned with the proof of the positivity of these multiplicities, which finishes the proof of Theorem 1.2. Finally, in Section 5, we give some interesting observations connecting the McKay-Thompson series to replicable functions.
Definition 2.1. A vector space V is called a superspace, if it is equipped with a Z/2Z-grading V = V (0) ⊕V (1) , where V (0) is called the even and V (1) is called the odd part of V . For an endomorphism α of V respecting this grading, i.e. α(V (i) ) ⊆ V (i) , we define its supertrace to be Now let G be a finite group and (V, ρ) a representation of G. If the G-module V admits a decomposition into an even and odd part as above which is compatible with the G-action, we call V a G-supermodule. For a G-subsupermodule W of V and g ∈ G we then write strace W (g) := strace (ρ| W (g)) .
Note that strace(g) only depends on the conjugacy class of g, which we denote by [g].

2.2.
Harmonic Maaß forms. Harmonic Maaß forms are an important generalization of classical, elliptic modular forms. In the weight 1/2 case, they are intimately related to the mock theta functions, a term coined by Ramanujan in his famous 1920 deathbed letter to Hardy. It took until the first decade of the 21 st century before work by Zwegers [45], Bruinier-Funke [9] and Bringmann-Ono [6,8] established the "right" framework for these enigmatic functions of Ramanujan's, namely that of harmonic Maaß forms. Since then, there have been many applications of harmonic Maaß forms both in various fields of pure mathematics, see, for instance, [1,5,11,18] among many others, and mathematical physics, especially in regards to quantum black holes and wall crossing [17] as well as Mathieu and Umbral Moonshine [14,15,20,25]. For a general overview on the subject we refer the reader to [34,44].
Recall the definition of the congruence subgroup of the full modular group SL 2 (Z).
Definition 2.2. We call a smooth function f : H → C a harmonic (weak) 1 Maaß form of weight k ∈ 1 2 Z of level N with multiplier system ψ, if the following conditions are satisfied: and where we assume 4|N if k / ∈ Z. 1 We usually omit the word "weak" from now on (2) The function f is annihilated by the weight k hyperbolic Laplacian, for some c > 0 as v → ∞. Analogous conditions are required at all cusps of Γ 0 (N). We denote the space of harmonic Maaß forms of weight k, level N and multiplier ψ is denoted by H k (N, ψ), where we omit the multiplier if it is trivial.
(1) Obviously, the weight k hyperbolic Laplacian annihilates holomorphic functions, so that the space H k (N, ψ) contains the spaces S k (N, ψ) of cusp forms (holomorphic modular forms vanishing at all cusps), M k (N, ψ) of holomorphic modular forms, and M ! k (N, ψ) of weakly holomorphic modular forms (holomorphic functions on H transforming like modular forms with possible poles at cusps).
(2) It should be pointed out that the definition of modular forms resp. harmonic Maaß forms with multiplier is slightly different in [28], where the multiplier is included into the definition of the slash operator f | k γ, so that multipliers here are always the inverse of the multipliers there.
It is not hard to see from the definition that harmonic Maaß forms naturally split into a holomorphic part and a non-holomorphic part (see for example equations (3.2a) and (3.2b) in [9]). Lemma 2.4. Let f ∈ H k (N, ψ) be a harmonic Maaß form of weight k = 1 such that ψ(( 1 1 0 1 )) = 1. Then there is a canonical splitting where for some m 0 ∈ Z we have the Fourier expansions where Γ(α; x) denotes the usual incomplete Gamma-function.
In the theory of harmonic Maaß forms, there is a very important differential operator that associates a weakly holomorphic modular form to a harmonic Maaß form [9, Proposition 3.2 and Theorem 3.7], often referred to as its shadow 2 .
Proposition 2.5. The operator is well-defined and surjective with kernel M ! k (N, ν). Moreover, we have that and we call this cusp form the shadow of f .
The ξ-operator can also be used to define the Bruinier-Funke pairing where ·, · denotes the Petersson inner product on the space of cusp forms. We will later use the following result [9, Proposition 3.5].
Proposition 2.6. For f = f + + f − as in Lemma 2.4 and g = ∞ n=1 b n q n ∈ S 2−k (N, ψ) such that f grows exponentially only at the cusp ∞ and is bounded at all other cusps of Γ 0 (N), we have that If f has poles at other cusps, the pairing is given by summing the corresponding terms using the q-series expansions for f and g at each such cusp.

Rademacher sums and McKay-Thompson series.
Here we recall a few basic facts about Poincaré series, Rademacher sums, and Rademacher series. For further details, the reader is referred to [12,13,19] and the references therein. An important way to construct modular forms of a given weight and multiplier is through Poincaré series. If one assumes absolute and locally uniform convergence, then the function P where Γ ∞ = ±T with T = ( 1 1 0 1 ) denotes the stabilizer of the cusp ∞ in Γ 0 (N) and µ ∈ log(ψ(T )) 2πi + Z, transforms like a modular form of weight k with multiplier ψ under the action of Γ 0 (N) and is holomorphic on H. In fact it is known that we have absolute and locally uniform convergence for weights k > 2 and in those cases, P [µ] ψ,k is a weakly holomorphic modular form, which is holomorphic if µ ≥ 0 and cuspidal if µ > 0.
For certain groups and multiplier systems, one can obtain conditionally, locally uniformly convergent series, now called Rademacher sums, for weights k ≥ 1, by fixing the order of summation as follows. Let for a positive integer K Γ K,K 2 (N) := ( a b c d ) ∈ Γ 0 (N) : |c| < K and |d| < K 2 . One can then define the Rademacher sum in [13]. He originally applied this to obtain an exact formula for the coefficients of the modular j-function.
In this paper, we especially need to look at Rademacher sums of weights 1 2 and 3 2 for Γ 0 (4N) with multiplier (4N) and v, h are integers with h| gcd(4N, 96). First we establish convergence of these series. (τ ) and R [3] ψ, 3 2 converge locally uniformly on H and therefore define holomorphic functions on H.
Proof. By following the steps outlined in [12,Section 5] to establish the convergence of weight 1 2 Rademacher series with a slightly different multiplier system (related to that of the Dedekind eta function) mutatis mutandis, we find that the Rademacher sums we are interested in converge, assuming the convergence at s = 3 4 of the Kloosterman zeta function where the * at the sum indicates that it runs over primitive residue classes modulo c, d denotes the multiplicative inverse of d modulo c, and e(α) := exp(2πiα) as usual. We omit the subscript if ψ = 1. In order to establish positivity of the multiplicities of irreducible characters in Section 4, we will show not only convergence of this series, but even explicit estimates for its value, which will complete the proof.
Since the Rademacher sum R (τ ) is 1-periodic by construction, it has a Fourier expansion, which can (at least formally) be established by standard methods. Projecting this function to the Kohnen plus space then yields the function where A N,ψ is given by (2.5) (1 + δ odd (Nc)) K ψ (−3, 0, 4Nc) (4Nc) 3 2 , (1 + δ odd (Nc)) K ψ (−3, n, 4Nc) 4Nc I 1 2 π √ 3n Nc . For each conjugacy class [g] of the Thompson group T h, we associate integers v g and h g (where h g |96) as specified in Table A.5 and the character ψ [g] := ψ |g|,vg,hg , where |g| denotes the order of g in T h, as well as a finite sequence of rational numbers κ m,g which are also given in Table A.5 and define the function This is going to be the explicitly given weakly holomorphic modular form (see Proposition 3.1) mentioned in Conjecture 1.1, meaning that we have for all conjugacy classes [g] of T h. We now prove and recall some important facts about Rademacher sums that we shall use later on. As in [12, Propositions 7.1 and 7.2], one sees the following. (τ ) with ψ as in (2.2) is a mock modular form of weight 1 2 whose shadow is a cusp form with the conjugate multiplier ψ, which is a constant multiple of the Rademacher sum R Next we establish the behaviour of Rademacher sums at cusps. Here we have to take into account that the sums we look at are projected into the Kohnen plus space which might affect the behaviour at cusps. For a function f ∈ M ! k+ 1 2 (Γ 0 (4N)), where k is an integer and N is odd, the projection of f to the plus space is defined by Using this projection operator, one sees that the following is true. The proof is similar in nature to that of Proposition 3 in [31] and is carried out in some detail (for the special case where N = p is an odd prime) in [27, Section 2]. Lemma 2.9. Let N be odd and f ∈ H k+ 1 2 (Γ 0 (4N)) for some k ∈ N 0 , such that f + (τ ) = q −m + ∞ n=0 a n q n for some m > 0 with −m ≡ 0, (−1) k (mod 4) has a non-vanishing principal part only at the cusp ∞ and is bounded at the other cusps of Γ 0 (4N). Then the projection f | pr of f to the plus space has a pole of order m at ∞ and has a pole of order m and is bounded at all other cusps.
Proof. In order to compute the expansion of f | pr at a given cusp a = a c , we compute Multiplying out the matrices, we see that this is (up to a constant factor) equal to By assumption, this function can only have a pole at ∞ if the denominator of the fraction (4+4N v)a+c 16N va+4c in lowest terms (where we allow the denominator to be 0 which we interpret as ∞) is divisible by 4N, which is easily seen to imply that N|c. Since there are only three inequivalent cusps of Γ 0 (4N) whose denominator is divisible by N, represented by ∞, 1 N , 1 2N , we can restrict ourselves to and C ∈ C is a constant that a priori depends on v, but by working out the corresponding automorphy factors, one sees with easy, elementary methods that it does indeed not. Note that it does however depend on N. Furthermore it is not hard to see that the difference δ v − 4β v runs through all residue classes modulo 16 that are congruent to N modulo 4. This implies that in only powers of q 1 16 survive whose exponent is divisible by 4. Since by assumption we have that f |σ 1 N = O(1) as τ → ∞, we therefore see that f | pr has a pole of order m 4 at the cusp 1 N if and only if m is divisible by 4. For the cusp 1 2N , the argumentation is analogous. One finds that The dependence on v is so that in the summation only powers of q 1 4 with exponents ≡ (−1) k (mod 4) survive, which implies our Lemma.
For even N, it turns out that the Rademacher series are automatically in the plus space. This follows immediately from the next lemma.
Lemma 2.10. Let m, n ∈ Z such that m ≡ n (mod 4) and c ∈ N be divisible by 8.
Proof. We write c = 2 ℓ c ′ with ℓ ≥ 3 and c ′ odd. By the Chinese Remainder Theorem one easily sees the following multiplicative property of the Kloosterman sum, is a Salié sum and c ′ denotes the inverse of c ′ modulo 2 ℓ and 2 ℓ denotes the inverse of 2 ℓ modulo c ′ . Therefore, it suffices to show the lemma for c = 2 ℓ with ℓ ≥ 3. The case where ℓ = 3 can be checked directly, so assume ℓ ≥ 4 from now on. In this case, it is straightforward to see that This yields that for m ≡ n (mod 2) we have that for all odd d ∈ {1, ..., 2 ℓ−1 − 1}, so that the summands in the Kloosterman sum pair up with opposite signs, making the sum 0 as claimed.
If m and n have the same parity, but are not congruent modulo 4, a similar pairing also works. In this case we find through similar reasoning that for ℓ ≥ 5 we have Again, we can pair summands with opposite signs, proving the lemma.
From the preceding two lemmas we immediately find that the following is true.
Proposition 2.11. For any g ∈ T h, the function Z |g|,ψ [g] is a mock modular form which has a pole of order 3 at ∞, a pole of order 3 4 at 1 2N if N is odd, and vanishes at all other cusps.
Proof. As described in Appendix E of [12] we see that the Rademacher sums R have only a pole of order 3 at ∞ and grow at most polynomially at all other cusps. By Lemmas 2.9 and 2.10 we see that the poles are as described in the proposition. The vanishing at all remaining cusps follows as in [10, Theorem 3.3].

Identifying the McKay-Thompson series as modular forms
In this section we want to establish that the multiplicities of each irreducible character are integers. To this end, we first establish the exact modularity and integrality properties of the conjectured McKay-Thompson series F [g] (τ ), which are stated without proof in [28].
Proof. As we know from Proposition 2.8, we have The space S + which is directly verifiable using the built-in functions for spaces of modular forms in for example Magma [3]. Furthermore, we have the Bruinier-Funke pairing (see Proposition 2.6) combined with Proposition 2.11 which tell us that and then use the same reasoning as in the proof of Lemma 2.9 to see that projection to the plus space only alters this value by a multiplicative non-zero constant, since the additional pole at the cusp 1 2|g| (if |g| is odd) is directly forced by the plus space condition.
From (3.1) we can now deduce, because the Petersson inner product is positive definite on the space of cusp forms, that the shadow of Z |g|,ψ [g] must be 0 if every f ∈ S + For the remaining 18 conjugacy classes, one can use the same arguments as above, but with the refinement that instead of looking at the full space S + (4|g|, ψ [g] ). Since computing bases for these spaces is not something that a standard computer algebra system can do without any further work, we describe how to go about doing this. Let f ∈ S + 3 2 (4|g|, ψ [g] ) for some conjugacy class [g]. Then f · ϑ is a modular form of weight 2 with the same multiplier and level (respectively trivial multiplier and level N [g] ). Using programs 3 written by Rouse and Webb [37] one can verify that the algebra of modular forms of level A Magma script computing dimensions and bases of these spaces can be obtained from the second author's homepage. Using this script we find that dim S + and for all remaining conjugacy classes [g], we find that every f ∈ S +  Proof. We have established in Proposition 3.1 that the Rademacher series Z |g|,ψ [g] (τ ) are all weakly holomorphic modular forms of weight 1 2 for Γ 0 (N [g] ) in the plus space. The given theta corrections are holomorphic modular forms of the same weight and level, hence so is their sum. Furthermore, theta functions don't have poles, so that all poles of F [g] come from the Rademacher series which has a pole of order 3 only at the cusps of Γ 0 (N [g] ) lying above the cusp ∞ on the modular curve X 0 (4|g|). Hence the function is a cusp form with integer Fourier coefficients, is a weight 2k holomorphic modular form with trivial multiplier under the group Γ 0 (N [g]  In order to compute the necessary Fourier coefficients exactly without relying on the rather slow convergence of the Fourier coefficients of the Rademacher series, one can construct linear combinations of weight 1 2 weakly holomorphic eta quotients again using the programs 4 written by Rouse and Webb [37] which have the same principal part at ∞ (and the related cusp 1 2N ) as the Rademacher series (see Proposition 2.11) and the same constant terms as the theta corrections if there are any, wherefore their difference must be a weight 1 2 holomorphic cusp form for Γ 0 (N [g] ) with trivial multiplier, which by the Serre-Stark basis theorem [39] is easily seen to be 0 5 . The largest bound up to which coefficients need to be checked turns out to be 384 for [g] = 24CD. [g] = 12AB that Harvey and Rayhaun [28, Table 5] give transforms with a different multiplier than the Rademacher series Z 12,ψ 12AB (τ ): As one computes directly from the fact that −ϑ(τ ) + 3ϑ(9τ ) transforms with the multiplier ψ 3,1,3 under the group Γ 0 (12), f (τ ) transforms under the group Γ 0 (48) with the multiplier ψ 12,1,3 , while Z 12,ψ 12AB (τ ) transforms with the multiplier ψ 12,7,12 . But since both multipliers become trivial on the group Γ 0 (144), the proposition remains valid.
We can now establish the uniqueness claim in Theorem 1.2 very easily. • its Fourier expansion is of the form 2q −3 + χ 2 (g) + O(q 4 ) and all its Fourier coefficients are integers. • if |g| is odd, then the only other pole of order 3 4 is at the cusp 1 2|g| , otherwise there is only the pole at ∞. It vanishes at all other cusps. For |g| = 36, F [g] (τ ) is uniquely determined by additionally fixing the coefficient of q 4 to be χ 4 (g) + χ 5 (g).
Proof. As we have used already, the function 2Z |g|,ψ [g] (τ ) has the right behaviour at the cusps so that F [g] (τ ) − 2Z |g|,ψ [g] (τ ) is a holomorphic weight 1 2 modular form. As it turns out, in all cases but the one where |g| = 36, this space is at most twodimensional, which can be seen by the Serre-Stark basis theorem if ψ [g] is trivial or through a computation similar to the one described in the proof of Proposition 3.1 if the multiplier is not trivial. Hence prescribing the constant and first term in the Fourier expansion determines the form uniquely. If |g| = 36, the space of weight 1 2 modular forms turns out to be 3-dimensional, so that fixing one further Fourier coefficient suffices to determine the form uniquely.
We ultimately want to study the multiplicities of the irreducible characters of T h. To this end, we now consider the functions with m j (n) as in (1.1), the generating functions of the multiplicities. We want to show that all those numbers m j (n) are integers. A natural approach for this would be to view F χ j as a weakly holomorphic modular form of weight 1 2 and level N χ j := lcm{N [g] : χ j (g) = 0} and then use a Sturm bound type argument as in the proof of Proposition 3.2. However, these levels turn out to be infeasibly large in most cases. For example we have that N χ 1 = 2 778 572 160, so one would have to compute at least a few 100 million Fourier coefficients of F χ 1 to make such an argument work, which is entirely infeasible. This bound can be reduced substantially however by breaking the problem into many smaller problems involving simpler congruences, each of which requires far fewer coefficients to prove.
We proceed by a linear algebra argument. Let C be the coefficient matrix containing the coefficients of the alleged McKay-Thompson series for each conjugacy class. In theory we have that C is a 48 × ∞ matrix. In practice, we take C to be a 48 × B matrix with B large. Let X be the 48 × 48 matrix with columns indexed by conjugacy classes of T h and rows indexed by irreducible characters, whose (χ i , [g])-th entry is Using the first Schur orthogonality relation for characters, we have that the rows of the matrix m := XC which are indexed by the characters χ i are exactly the multiplicity values under consideration for the given character. The matrix C does not have full rank. Besides the duplicated series (such as T [12A] = T [12B] ), we have additional linear relations given in Appendix B.1. As some of these relations involve the theta functions used as correction terms in the construction, let us define C + to be the matrix extending C to include the coefficients of the theta functions ϑ(n 2 τ ) , for n = 1, 2, 3, 6, 9. Then there are matrices N, N * of dimensions 48 × 35 and 35 × 53 respectively, so that N * C + has full rank and m = XC = XNN * C + .
We construct the matrix N * by taking a 53 × 53 identity matrix indexed by conjugacy classes and removing the rows corresponding to one of each duplicate series and also the highest level conjugacy class (as ordered for instance by Table A.5) appearing in each of the linear relations. The matrix N may be constructed starting with a 35 × 35 identity matrix, adding in columns to reconstitute the removed conjugacy classes and removing the columns corresponding to the theta series.
The rows of N * C + exhibit additional congruence as listed in Appendix B.2. For each prime p dividing the order of the Thompson group, we construct a matrix M p to reduce by these congruences. We start as before with a 35 × 35 identity matrix. Then for each congruence listed we replace the row of the matrix with index given by the highest weight conjugacy class appearing in the congruence with a new row constructed to reduce by that congruence. Given the congruence where the a g ∈ Z, the new row will be given by Assuming for the moment the validity of these congruences, we have that M p N * C + is an integer matrix. Moreover, by construction M p is invertible (depending on ordering, we have that M p is lower-triangular with non-vanishing main diagonal).
In each case we have computationally verified that the matrix is the product of two p-integral matrices, we have that every multiplicity must also be p-integral. The congruences listed in Appendix B.2 were found computationally by reducing the matrix (N * C + ) (mod p) and computing the left kernel. After multiplying by a matrix constructed similar to M p above so as to reduce by the congruences found, the process was repeated. The list of congruences given represents a complete list, in the sense that the matrix (M p N * C + ) both is integral and has full rank modulo p.
Many of the congruences can be easily proven using standard trace arguments for spaces of modular forms of level pN to level N. For uniformity we will instead rely on Sturm's theorem following the argument described above. The worst case falls with any congruence involving the conjugacy class 24CD. These occur for both primes p = 2 and 3. The nature of the congruences, however, do not require us to increase the level beyond the corresponding level N 24CD = 1152. There is a unique normalized cusp form of weight 19/2 and level 4 in the plus space. This form vanishes to order 3 at the cusp ∞ and to order 3/4 at the cusp 1/2. This is sufficient so that multiplying by this cusp form moves these potential congruences into spaces of holomorphic modular forms of weight 10, level 1152. The Sturm bound for this space falls just shy of 2 000 coefficients. This bound could certainly be reduced by more careful analysis, but this is sufficient for our needs. The congruences were observed up to 10 000 coefficients. These computations were completed using Sage mathematical software [38].

Remark 3.5.
A similar process can be used in the case of Monstrous Moonshine to prove the integrality of the Monster character multiplicities. This gives a (probably 6 ) alternate proof the theorem of Atkin-Fong-Smith [22,40]. As in the case of Thompson moonshine, we have calculated a list of congruences for each prime dividing the order of the Monster, proven by means of Sturm's theorem. This is list complete in the sense that once we have reduced by the congruences for a given prime, the resulting forms have full rank modulo that prime. The Monster congruences may be of independent interest and are available upon request to the authors.

Positivity of the multiplicities
To establish the positivity of the multiplicities of the irreducible representations we follow Gannon's work mutatis mutandis. First we notice that by the first Schur orthogonality relation for characters and the triangle inequality, for each irreducible representation ρ with corresponding character χ of T h we have the estimate Here C(g) denotes the centralizer of g in T h and the summation runs over all conjugacy classes of T h. Thus to prove positivity, we show that | strace W k (1)| always dominates all the others. To this end, we use the description of the Fourier coefficients of F [g] (τ ) in terms of Maaß-Poincaré series, see (2.4) and (2.6). We use the following elementary (and rather crude) estimates, and set δ c = 1 + δ odd (Nc) which has the obvious bounds 1 ≤ δ c ≤ 2.
The convergence and bounds of the coefficients A N,ψ rely on the convergence of the modified Selberg-Kloosterman zeta function The zeta function only converges conditionally at 3/4. The bounds we obtain are crude and very large, but they do not grow with n.
If we set then using the triangle inequality we find and D is the dominant term coming from the c = 1 term in the expression for A N,ψ . We may estimate |D| by N .
Here we have used the second estimate for the Bessel function. We will only be interested in a lower bound for |D| when [g] = [1A]. In this case, we will just use the exact expression If we set L := π N √ 3n, the sum for 2 ≤ c ≤ L in R can be estimated as follows: 2N .
For the terms of R with c ≥ L we can use the first estimate on the Bessel function.
where ζ(s) denotes the Riemann zeta function. We now need only estimate Z * ψ (m, n; 3 4 ). It turns out that we can use Gannon's estimates almost directly once we write the modified zeta function in a form sufficiently similar to the expressions he uses in his estimates. However we will need to slightly modify some of Harvey and Rayhaun's notation for v and h. Letĥ = h (h,4) andv = 4ν (h,4) (modĥ) so that h | (N, 24) and ψ(4Nc, d) = exp −2πiv cd h .
We note that in every case given we have thatv ≡ ±1 (modĥ).
With this notation we have that .
Using a result by Kohnen [31,Proposition 5], we can write our Kloosterman sum as a sum over a more sparse set. Kohnen shows that Here [α, β, γ] is a positive definite integral binary quadratic form, in this case with discriminant mn, and χ ∆ is the genus character defined as follows on integral binary quadratic forms with discriminant divisible by ∆ by We can write equation (4.2) in this form if we replace n withñ = n − 4vc 2 · N h . Unfortunately, this makes the sum over a set of quadratic forms with discriminant mñ which depends on c. This is not ideal for approximating the zeta function. To fix this, notice that if the quadratic form Q = [Nc, β, γ] has discriminant mñ, then the form Q ′ = [Nc, β, γ ′ /ĥ] with γ ′ = β 2 −mn 4N c/ĥ is a positive definite binary quadratic form, with discriminant mn and γ ′ ≡ mvc (modĥ). This relation defines a bijection between such forms.
Let Q N ;ĥ,mv (D) denote the set of quadratic forms Q = [Nc, β, γ/ĥ] of discriminant D with c, β, γ ∈ Z and γ ≡ mvc (modĥ), and let Q N (d) denote the set of quadratic forms Q = [Nc, β, γ] of discriminant d. Then we have the bijection We may drop the subscript of ϕ as it will generally be clear from context. The set Q N ;ĥ,mv (D) is acted upon by a certain matrix group which we denote by Γ 0 (N; h, mv). This group consists of matrices a b/h Nc d of determinant 1 where each letter is an integer satisfying the relations a ≡ ℓd (modĥ) and b ≡ ℓmvc (modĥ).
Proposition 4.2. Assume the notation above, and let Q 1 , Q 2 ∈ Q N ;ĥ,∆v (D) where ∆, D are discriminants, with ∆ fundamental and D divisible by ∆. If Q 1 and Q 2 are related by the action of some M ∈ Γ 0 (N; h, mv), then Proof. Since ∆ is a discriminant, the definition of χ ∆ (Q) as a Kronecker symbol allows us to reduce the coefficients of the quadratic form Q modulo ∆. It is also multiplicative. If ∆ = ∆ 1 ∆ 2 is a factorization into discriminants, then We will want a factorization of ∆ into discriminants ∆ = ∆ ′ ∆ h where (∆ ′ ,ĥ) = 1 and ∆ h is divisible only by primes dividingĥ. Since ∆ is a fundamental discriminant andĥ divides 24, this means that |∆ h | also divides 24.
By construction, ϕ Q 1 ≡ Q 1 (mod ∆ ′ ). As a determinant 1 matrix will not alter the integers represented by a quadratic form, we have that Suppose Q 1 = [Nα, β, γ/ĥ] and Q 2 = Q 1 |M where M is the matrix a b/ĥ Nc d with a ≡ ℓd (modĥ). If we set γ ′ = (γ − mvα)/ĥ, then a short calculation show that Since M has determinant 1, we have that a and d are coprime to ∆. Since ∆ h | 24, we have that a 2 ≡ d 2 ≡ 1 (mod ∆ h ). Moreover, by considering the discriminant we see that β is even if ∆ h is, and 4 | β if 8 | ∆ h . In either case, we find that ∆ h divides βN. Therefore we may further reduce to The ℓ does not change the possible numbers represented, and so we have that concluding the proof of proposition 4.2.
Here Q = [Nα, β, γ/ĥ], ω Q is the order of the stabilizer of Q in Γ 0 (N;ĥ, mv), and c(Q, r, Ns) = Q(r,N s) 4N . This equation is analogous to Equation (4.26) of [25], but differs in four main points: First, we have normalized the zeta function slightly differently. Second, the bijection ϕĥ ,mv and proposition 4.2 give a more general version of Gannon's Lemma 5(b) allowing us to sum over Q N ;ĥ,v (mn)/Γ 0 (N;ĥ, mv) rather than Q N ;ĥ (mn)/Γ 0 (N;ĥ). Third, Gannon's case was restricted to discriminants where the stabilizer could only be {±I}, and so he replaces the ω Q term with a 2 in his equation. We will use this as a lower bound for ω Q . Fourth, his sum contains a power of −1 while ours contains a genus character. In either case, the sign is constant for a given representative quadratic form Q.
Gannon estimates the inner sums in absolute value and the outer sum by bounding the number of classes of quadratic forms. His bounds for the size of Q N ;ĥ (mn)/Γ 0 (N;ĥ) are crude enough to also hold for the number of classes of Q N ;ĥ,v (mn)/Γ 0 (N;ĥ, mv). Proposition 4.1 follows from using Gannon's bounds modified only to account for our differences in normalization. Combining these estimates as described above, we find that each multiplicity of the irreducible components of W n must alway be positive for n ≥ 375. Explicit calculations up to n = 375 show that these multiplicities are always positive. The worst cases for the estimates with n ≤ 375 arise from the trivial character or from estimating for Selberg-Kloosterman zeta function for the 24CD conjugacy class. These calculations were performed using Sage mathematical software [38].

Replicability
One important property of the Hauptmoduln occurring in Monstrous Moonshine is that they are replicable.
Definition 5.1. Let f (τ ) = q −1 + ∞ n=0 H n q n be a (formal) power series with integer coefficients and consider the function where τ 1 , τ 2 ∈ H are two independent variables and q j = e 2πiτ j , j = 1, 2. We call f replicable, if we have that H a,b = H c,d whenever ab = cd and gcd(a, b) = gcd(c, d).
This property of the Hauptmoduln involved in Monstrous Moonshine in a sense reflects the algebra structure of the Monstrous-Moonshine module, see [16,33].
An important, but not immediately obvious fact is that any replicable function is determined by its first 23 Fourier coefficients, [23,33].
Theorem 5.2. Let f (τ ) = q −1 + ∞ n=1 a n q n be a replicable function. Then one can compute the coefficient a n for any n ∈ N constructively out of the coefficients a 2 , a 3 , a 4 , a 5 , a 7 , a 8 , a 9 , a 11 , a 17 , a 19 , a 23 }.
A Maple procedure to perform this computation is printed at the end of [23]. In [20], there is also an analogous notion of replicability in the mock modular sense, which requires that the Fourier coefficients satisfy a certain type of recurrence. This is a special phenomenon occuring for mock theta functions, i.e., mock modular forms whose shadow is a unary theta function, satisfying certain growth conditions at cusps, see [29,32].
It is now natural to ask about replicability properties of the McKay-Thompson series in the case of Thompson Moonshine. Let g be any element of the Thompson group and T [g] (τ ) = F [g] (τ ) be the corresponding McKay-Thompson series as in (2.6). As we have seen, this is a weekly holomorphic modular form of weight 1 2 living in the Kohnen plus space. To relate these to the Hauptmoduln and other replicable functions discussed in [23], we split into an even and an odd part in the notation of Conjecture 1.1. Letting we define the weight 0 modular functions , (j = 0, 1) [g] (τ ). Note that these functions don't have poles in H.
As it turns out through direct inspection, these weight 0 functions are often replicable functions or univariate rational functions therein. We used the list of replicable functions given in [23] as a reference and found the identities given in Tables A.6 -A.8, which are all identities of the given form using the aforementioned table of replicable functions at the end of [23] and allowing the degree of the denominator of the rational function to be as large as 40.