Introduction
Since Ramanujan’s famous partition congruences p ( 5 n + 4 ) ≡ 0 ( m o d 5 ) p(5n+4) \equiv0 \pmod{5} p ( 5 n + 4 ) ≡ 0 ( mod 5 ) , p ( 7 n + 5 ) ≡ 0 ( m o d 7 ) p(7n+5) \equiv0 \pmod{7} p ( 7 n + 5 ) ≡ 0 ( mod 7 ) , and p ( 11 n + 6 ) ≡ 0 ( m o d 11 ) p(11n+6) \equiv0 \pmod{11} p ( 11 n + 6 ) ≡ 0 ( mod 11 ) , the arithmetic properties of restricted partition families have been a central focus in number theory. While traditional methods often rely on generating functions, modular forms, or combinatorial statistics like Dyson’s rank and the Andrews–Garvan crank, looking at k k k -tuples (multipartitions) reveals deeper structures whose properties depend heavily on the p p p -adic valuation of k k k .
A prominent family in this area is partitions with designated summands, PD ( n ) \operatorname{PD}(n) PD ( n ) , introduced by Andrews, Lewis, and Lovejoy, where exactly one part of each size is marked. Their generating function is given by ∑ PD ( n ) q n = f 6 / ( f 1 f 2 f 3 ) \sum\operatorname{PD}(n)q^n = f_6/(f_1f_2f_3) ∑ PD ( n ) q n = f 6 / ( f 1 f 2 f 3 ) , where f k : = ∏ n = 1 ∞ ( 1 − q k n ) f_k := \prod_{n=1}^{\infty}(1-q^{kn}) f k := ∏ n = 1 ∞ ( 1 − q k n ) denotes the standard eta-quotient. Figure 1 illustrates this combinatorial structure through Young diagrams for n = 4 n=4 n = 4 , where every block of equal rows contains a designated marked cell. Over time, these objects have been extensively studied using 3-dissections, cranks, and various congruence families. Similarly, restricting parts to be odd gives PDO ( n ) \operatorname{PDO}(n) PDO ( n ) , explored by Baruah and Ojah and later refined by Fu and Sellers.
Figure 1. The ten partitions with designated summands of n = 4 n = 4 n = 4 (PD(4) = 10) represented via Young diagrams with designated dotted cells.
For bipartitions with designated summands, A 2 ( n ) = PD − 2 ( n ) A_2(n)=\operatorname{PD}_{-2}(n) A 2 ( n ) = PD − 2 ( n ) , Naika and Shivashankar along with Hao and Shen proved several deep congruences modulo powers of 2 and 3. Motivated by these works, we study the families A k ( n ) = PD − k ( n ) A_k(n)=\operatorname{PD}_{-k}(n) A k ( n ) = PD − k ( n ) and B k ( n ) = PDO − k ( n ) B_k(n)=\operatorname{PDO}_{-k}(n) B k ( n ) = PDO − k ( n ) , which count ordered k k k -tuples of designated partitions and odd-part designated partitions. Our approach starts with the 3-dissection of f 1 f 2 f_1f_2 f 1 f 2 using Hirschhorn’s classical methods, writing the main generating function as F = f 6 / ( f 1 f 2 f 3 ) = Θ P ( u ) F=f_6/(f_1f_2f_3)=\Theta P(u) F = f 6 / ( f 1 f 2 f 3 ) = Θ P ( u ) with a quartic polynomial P P P and explicit eta-quotients Θ \Theta Θ and u u u . This turns questions about F k ( m o d 3 e ) F^k \pmod{3^e} F k ( mod 3 e ) into finite checks on the coefficients of P ( u ) k P(u)^k P ( u ) k .
Throughout this paper, we give a complete classification for every e ≥ 1 e \ge1 e ≥ 1 of the parameters k k k where A k ( 3 n + 2 ) ≡ 0 ( m o d 3 e ) A_k(3n+2) \equiv0 \pmod{3^e} A k ( 3 n + 2 ) ≡ 0 ( mod 3 e ) for all n n n , and we provide level-wise recursions. We also prove exact formulas and new congruences for k = 2 k=2 k = 2 that improve earlier results, such as $A_2(24n+23) \equiv 0 (mod 64) and A 2 ( 18 n + 15 ) ≡ 0 A_2(18n+15)\equiv0 A 2 ( 18 n + 15 ) ≡ 0 (mod 32), alongside general congruences for A k ( 4 n + 3 ) A_k(4n+3) A k ( 4 n + 3 ) and the exact 3-adic valuation v 3 ( k ) + 1 v_3(k)+1 v 3 ( k ) + 1 for B k ( 3 n + 2 ) B_k(3n+2) B k ( 3 n + 2 ) when 3 ∣ k 3\mid k 3 ∣ k . The rest of the paper explores combinatorial interpretations using Young diagrams, links to overpartitions and sums of squares, group-action proofs, and computational conjectures, all using elementary tools like Frobenius congruences and Lucas’ theorem.
Preliminaries and the structure of F F F
In this section, we review basic notation. Let f m = ( q m ; q m ) ∞ = ∏ j ≥ 1 ( 1 − q m j ) f_m=(q^m;q^m)_\infty=\prod_{j\geq1}(1-q^{mj}) f m = ( q m ; q m ) ∞ = ∏ j ≥ 1 ( 1 − q mj ) for m ≥ 1 m\geq1 m ≥ 1 . Products of the form ∏ δ ∣ d f δ r δ \prod_{\delta\mid d}f_\delta^{r_\delta} ∏ δ ∣ d f δ r δ are called eta-quotients, which we use mainly for notation and dissection identities. We also frequently use standard product relations such as ( − q ; q ) ∞ = f 2 / f 1 (-q;q)_\infty=f_2/f_1 ( − q ; q ) ∞ = f 2 / f 1 , ( q ; q 2 ) ∞ = f 1 / f 2 (q;q^2)_\infty=f_1/f_2 ( q ; q 2 ) ∞ = f 1 / f 2 , and ( − q ; q 2 ) ∞ = f 2 2 / ( f 1 f 4 ) (-q;q^2)_\infty=f_2^2/(f_1f_4) ( − q ; q 2 ) ∞ = f 2 2 / ( f 1 f 4 ) .
Ramanujan’s theta functions are defined by φ ( q ) = ∑ m ∈ Z q m 2 = f 5 2 / ( f 2 f 4 2 ) \varphi(q)=\sum_{m\in\mathbb{Z}}q^{m^2}=f_5^2/(f_2f_4^2) φ ( q ) = ∑ m ∈ Z q m 2 = f 5 2 / ( f 2 f 4 2 ) and ψ ( q ) = ∑ m ≥ 0 q m ( m + 1 ) / 2 = f 2 2 / f 1 \psi(q)=\sum_{m\geq0}q^{m(m+1)/2}=f_2^2/f_1 ψ ( q ) = ∑ m ≥ 0 q m ( m + 1 ) /2 = f 2 2 / f 1 . We also use the Borwein cubic theta functions a ( q ) = ∑ m , n q m 2 + m n + n 2 a(q)=\sum_{m,n}q^{m^2+mn+n^2} a ( q ) = ∑ m , n q m 2 + mn + n 2 and b ( q ) = f 1 3 / f 3 = 1 2 ( 3 a ( q 3 ) − a ( q ) ) b(q)=f_1^3/f_3=\frac{1}{2}(3a(q^3)-a(q)) b ( q ) = f 1 3 / f 3 = 2 1 ( 3 a ( q 3 ) − a ( q )) [2 ] . For comparison, we consider overpartitions p ‾ ( n ) \overline{p}(n) p ( n ) with generating function f 2 / f 1 2 = 1 / φ ( − q ) f_2/f_1^2=1/\varphi(-q) f 2 / f 1 2 = 1/ φ ( − q ) , and cubic partitions a ( n ) a(n) a ( n ) with generating function 1 / ( f 1 f 2 ) 1/(f_1f_2) 1/ ( f 1 f 2 ) . For any partition family X X X , we write X − k ( n ) X_{-k}(n) X − k ( n ) to denote the number of k k k -tuples of total size n n n , whose generating function is the k k k -th power of the original one.
Recall that a partition with designated summands is defined by marking exactly one part of each size, which corresponds on its Young diagram to having exactly one row marked with a dot in each block of equal rows (Figure 1 ). Since a block of m m m equal rows can be marked in m m m ways, the generating function is given by
∑ n ≥ 0 PD ( n ) q n = ∏ t ≥ 1 ( 1 + ∑ m ≥ 1 m q t m ) = ∏ t ≥ 1 ( 1 + q t ( 1 − q t ) 2 ) = f 6 f 1 f 2 f 3 . \sum_{n\geq0}\operatorname{PD}(n)q^n=\prod_{t\geq1}\left(1+\sum_{m\geq1}mq^{tm}\right)=\prod_{t\geq1}\left(1+\frac{q^t}{(1-q^t)^2}\right)=\frac{f_6}{f_1f_2f_3}. n ≥ 0 ∑ PD ( n ) q n = t ≥ 1 ∏ ( 1 + m ≥ 1 ∑ m q t m ) = t ≥ 1 ∏ ( 1 + ( 1 − q t ) 2 q t ) = f 1 f 2 f 3 f 6 .
Similarly, when all parts are restricted to be odd, we obtain PDO ( n ) \operatorname{PDO}(n) PDO ( n ) , whose generating function is ∑ n ≥ 0 PDO ( n ) q n = f 4 f 2 2 / ( f 1 f 3 f 12 ) \sum_{n\geq0}\operatorname{PDO}(n)q^n=f_4f_2^2/(f_1f_3f_{12}) ∑ n ≥ 0 PDO ( n ) q n = f 4 f 2 2 / ( f 1 f 3 f 12 ) .
Definition 2.1. For k ≥ 1 k\geq1 k ≥ 1 , let A k ( n ) = PD − k ( n ) A_k(n)=\operatorname{PD}_{-k}(n) A k ( n ) = PD − k ( n ) and B k ( n ) = PDO − k ( n ) B_k(n)=\operatorname{PDO}_{-k}(n) B k ( n ) = PDO − k ( n ) , defined by their generating functions:
∑ n ≥ 0 A k ( n ) q n = F k , F = f 6 f 1 f 2 f 3 = 1 + q + 3 q 2 + 5 q 3 + 10 q 4 + ⋯ . \sum_{n\geq0}A_k(n)q^n=F^k,\qquad F=\frac{f_6}{f_1f_2f_3}=1+q+3q^2+5q^3+10q^4+\cdots. n ≥ 0 ∑ A k ( n ) q n = F k , F = f 1 f 2 f 3 f 6 = 1 + q + 3 q 2 + 5 q 3 + 10 q 4 + ⋯ .
∑ n ≥ 0 B k ( n ) q n = G k , G = f 4 f 6 2 f 1 f 3 f 12 = 1 + q + 2 q 2 + 4 q 3 + 5 q 4 + ⋯ . \sum_{n\geq0}B_k(n)q^n=G^k,\qquad G=\frac{f_4f_6^2}{f_1f_3f_{12}}=1+q+2q^2+4q^3+5q^4+\cdots. n ≥ 0 ∑ B k ( n ) q n = G k , G = f 1 f 3 f 12 f 4 f 6 2 = 1 + q + 2 q 2 + 4 q 3 + 5 q 4 + ⋯ .
The m m m -dissection of a series ∑ c n q n \sum c_nq^n ∑ c n q n splits it into ∑ r = 0 m − 1 q r D r ( q m ) \sum_{r=0}^{m-1}q^rD_r(q^m) ∑ r = 0 m − 1 q r D r ( q m ) , where D r ( q m ) = ∑ n ≡ r ( m ) c n q n − r D_r(q^m)=\sum_{n\equiv r\ (m)}c_nq^{n-r} D r ( q m ) = ∑ n ≡ r ( m ) c n q n − r . For m = 2 m=2 m = 2 , this gives the even and odd parts. We use the classical 2- and 3-dissections of eta-quotients due to Hirschhorn [3 ] , which we cite as needed.
In addition to dissections, we rely on two standard tools: Lucas’ theorem, which states that for prime p p p with base-p p p expansions n = ∑ n i p i n=\sum n_ip^i n = ∑ n i p i and m = ∑ m i p i m=\sum m_ip^i m = ∑ m i p i , we have ( n m ) ≡ ∏ i ( n i m i ) ( m o d p ) \binom{n}{m}\equiv\prod_i\binom{n_i}{m_i}\pmod p ( m n ) ≡ ∏ i ( m i n i ) ( mod p ) ; and the standard Frobenius congruence.
Lemma 2.2 ([3 ] ). For any prime p p p , integer j ≥ 1 j\geq1 j ≥ 1 , and series H H H with integer coefficients,
H ( q ) p j ≡ H ( q p ) p j − 1 ( m o d p j ) . H(q)^{p^j}\equiv H(q^p)^{p^{j-1}}\pmod{p^j}. H ( q ) p j ≡ H ( q p ) p j − 1 ( mod p j ) .
A first direct consequence is that the divisibility properties of terms in the arithmetic progressions 3 n + r 3n+r 3 n + r are periodic with respect to k k k .
Lemma 2.3. Let r ∈ { 0 , 1 , 2 } r\in\{0,1,2\} r ∈ { 0 , 1 , 2 } and e ≥ 1 e\geq1 e ≥ 1 . Then A k ( 3 n + r ) ≡ 0 ( m o d 3 e ) A_k(3n+r)\equiv0\pmod{3^e} A k ( 3 n + r ) ≡ 0 ( mod 3 e ) for all n n n if and only if A k + 3 e ( 3 n + r ) ≡ 0 ( m o d 3 e ) A_{k+3^e}(3n+r)\equiv0\pmod{3^e} A k + 3 e ( 3 n + r ) ≡ 0 ( mod 3 e ) for all n n n . The same property holds for B k B_k B k .
Proof. By Lemma 2.2, we have F 3 e ≡ U ( q 3 ) ( m o d 3 e ) F^{3^e}\equiv U(q^3)\pmod{3^e} F 3 e ≡ U ( q 3 ) ( mod 3 e ) , where U ( q 3 ) = F ( q 3 ) 3 e − 1 U(q^3)=F(q^3)^{3^{e-1}} U ( q 3 ) = F ( q 3 ) 3 e − 1 with U ( 0 ) = 1 U(0)=1 U ( 0 ) = 1 . This gives F k + 3 e ≡ U ( q 3 ) F k F^{k+3^e}\equiv U(q^3)F^k F k + 3 e ≡ U ( q 3 ) F k . Writing U ( q 3 ) = ∑ i u i q 3 i U(q^3)=\sum_i u_iq^{3i} U ( q 3 ) = ∑ i u i q 3 i , it follows that
A k + 3 e ( 3 n + r ) ≡ ∑ i ≤ n u i A k ( 3 ( n − i ) + r ) ( m o d 3 e ) . A_{k+3^e}(3n+r)\equiv\sum_{i\leq n}u_iA_k(3(n-i)+r)\pmod{3^e}. A k + 3 e ( 3 n + r ) ≡ i ≤ n ∑ u i A k ( 3 ( n − i ) + r ) ( mod 3 e ) .
If all A k ( 3 m + r ) A_k(3m+r) A k ( 3 m + r ) are divisible by 3 e 3^e 3 e , the left side is clearly divisible as well. Conversely, since u 0 = 1 u_0=1 u 0 = 1 , induction on n n n shows that A k ( 3 n + r ) A_k(3n+r) A k ( 3 n + r ) is also divisible by 3 e 3^e 3 e . □
Our main tool for all subsequent results is an explicit description of the series F F F .
Proposition 2.4. Let
c = f 6 f 9 4 f 3 f 18 2 , ρ = f 3 f 18 3 f 6 f 9 3 , u = q ρ , Λ = f 18 2 f 9 4 . c=\frac{f_6f_9^4}{f_3f_{18}^2},\qquad\rho=\frac{f_3f_{18}^3}{f_6f_9^3},\qquad u=q\rho,\qquad\Lambda=\frac{f_{18}^2}{f_9^4}. c = f 3 f 18 2 f 6 f 9 4 , ρ = f 6 f 9 3 f 3 f 18 3 , u = q ρ , Λ = f 9 4 f 18 2 . Then c , ρ , Λ ∈ Z [ [ q 3 ] ] c,\rho,\Lambda\in\mathbb{Z}[[q^3]] c , ρ , Λ ∈ Z [[ q 3 ]] have constant term 1, and
f 1 f 2 = c ( 1 − u − 2 u 2 ) , F = Λ 1 − u − 2 u 2 = Θ P ( u ) , f_1f_2=c(1-u-2u^2),\qquad F=\frac{\Lambda}{1-u-2u^2}=\Theta P(u), f 1 f 2 = c ( 1 − u − 2 u 2 ) , F = 1 − u − 2 u 2 Λ = Θ P ( u ) , where P ( u ) = 1 + u + 3 u 2 − 2 u 3 + 4 u 4 = ( 1 − u + u 2 ) ( 1 + 2 u + 4 u 2 ) P(u)=1+u+3u^2-2u^3+4u^4=(1-u+u^2)(1+2u+4u^2) P ( u ) = 1 + u + 3 u 2 − 2 u 3 + 4 u 4 = ( 1 − u + u 2 ) ( 1 + 2 u + 4 u 2 ) and Θ = Λ / ( ( 1 + u 3 ) ( 1 − 8 u 3 ) ) ∈ Z [ [ q 3 ] ] \Theta=\Lambda/((1+u^3)(1-8u^3))\in\mathbb{Z}[[q^3]] Θ = Λ/ (( 1 + u 3 ) ( 1 − 8 u 3 )) ∈ Z [[ q 3 ]] has constant term 1.
Proof. By Hirschhorn’s 3-dissection (see [11 ] , Lemma 2.6), we have
f 1 f 2 = f 6 f 9 4 f 3 f 18 2 − q f 9 f 18 − 2 q 2 f 3 f 18 4 f 6 f 9 2 . f_1f_2=\frac{f_6f_9^4}{f_3f_{18}^2}-qf_9f_{18}-2q^2\frac{f_3f_{18}^4}{f_6f_9^2}. f 1 f 2 = f 3 f 18 2 f 6 f 9 4 − q f 9 f 18 − 2 q 2 f 6 f 9 2 f 3 f 18 4 .
Direct cancellation gives f 9 f 18 = c ρ f_9f_{18}=c\rho f 9 f 18 = c ρ and f 3 f 18 4 / ( f 6 f 9 2 ) = c ρ 2 f_3f_{18}^4/(f_6f_9^2)=c\rho^2 f 3 f 18 4 / ( f 6 f 9 2 ) = c ρ 2 , which leads to f 1 f 2 = c ( 1 − u − 2 u 2 ) f_1f_2=c(1-u-2u^2) f 1 f 2 = c ( 1 − u − 2 u 2 ) . Since f 6 / ( f 3 c ) = f 18 2 / f 9 4 = Λ f_6/(f_3c)=f_{18}^2/f_9^4=\Lambda f 6 / ( f 3 c ) = f 18 2 / f 9 4 = Λ , we get F = f 6 / ( f 3 f 1 f 2 ) = Λ / ( 1 − u − 2 u 2 ) F=f_6/(f_3f_1f_2)=\Lambda/(1-u-2u^2) F = f 6 / ( f 3 f 1 f 2 ) = Λ/ ( 1 − u − 2 u 2 ) . Finally, since 1 − u − 2 u 2 = ( 1 + u ) ( 1 − 2 u ) 1-u-2u^2=(1+u)(1-2u) 1 − u − 2 u 2 = ( 1 + u ) ( 1 − 2 u ) , multiplying both the numerator and denominator by ( 1 − u + u 2 ) ( 1 + 2 u + 4 u 2 ) (1-u+u^2)(1+2u+4u^2) ( 1 − u + u 2 ) ( 1 + 2 u + 4 u 2 ) turns the denominator into ( 1 + u 3 ) ( 1 − 8 u 3 ) (1+u^3)(1-8u^3) ( 1 + u 3 ) ( 1 − 8 u 3 ) . □
While the formula F = Λ / ( 1 − u − 2 u 2 ) F=\Lambda/(1-u-2u^2) F = Λ/ ( 1 − u − 2 u 2 ) is equivalent to the 3-dissection from [5 ] , our new polynomial form F = Θ P ( u ) F=\Theta P(u) F = Θ P ( u ) is the key insight: it reduces any question about F k F^k F k modulo 3 e 3^e 3 e to a finite coefficient check (Lemma 2.5).
Lemma 2.5. Let c j ( k ) = [ u j ] P ( u ) k c_j(k)=[u^j]P(u)^k c j ( k ) = [ u j ] P ( u ) k . For any r ∈ { 0 , 1 , 2 } r\in\{0,1,2\} r ∈ { 0 , 1 , 2 } and M ≥ 1 M\geq1 M ≥ 1 ,
A k ( 3 n + r ) ≡ 0 ( m o d M ) for all n ⟺ c j ( k ) ≡ 0 ( m o d M ) for all j ≡ r ( m o d 3 ) . A_k(3n+r)\equiv0\pmod M\ \text{for all }n\Longleftrightarrow c_j(k)\equiv0\pmod M\ \text{for all }j\equiv r\pmod3. A k ( 3 n + r ) ≡ 0 ( mod M ) for all n ⟺ c j ( k ) ≡ 0 ( mod M ) for all j ≡ r ( mod 3 ) .
Proof. We can write F k = Θ k ∑ j c j ( k ) q j ρ j F^k=\Theta^k\sum_j c_j(k)q^j\rho^j F k = Θ k ∑ j c j ( k ) q j ρ j . Since Θ \Theta Θ and ρ \rho ρ belong to Z [ [ q 3 ] ] \mathbb{Z}[[q^3]] Z [[ q 3 ]] , the terms of F k F^k F k with exponents congruent to r ( m o d 3 ) r\pmod3 r ( mod 3 ) are given by q r Θ k ρ r ∑ i ≥ 0 c r + 3 i ( k ) ( q 3 ρ 3 ) i q^r\Theta^k\rho^r\sum_{i\ge0}c_{r+3i}(k)(q^3\rho^3)^i q r Θ k ρ r ∑ i ≥ 0 c r + 3 i ( k ) ( q 3 ρ 3 ) i . Here, Θ k ρ r \Theta^k\rho^r Θ k ρ r is a unit in Z [ [ q 3 ] ] \mathbb{Z}[[q^3]] Z [[ q 3 ]] , and ( q 3 ρ 3 ) i (q^3\rho^3)^i ( q 3 ρ 3 ) i has exact order 3 i 3i 3 i with leading coefficient 1. Thus, all coefficients of q 3 n + r q^{3n+r} q 3 n + r vanish modulo M M M if and only if every c r + 3 i ( k ) ≡ 0 ( m o d M ) c_{r+3i}(k)\equiv0\pmod M c r + 3 i ( k ) ≡ 0 ( mod M ) (which follows by inspecting the smallest i i i where c r + 3 i c_{r+3i} c r + 3 i is non-zero).
We will also need the explicit eta-quotient form of Θ \Theta Θ and a factorization of ( 1 + u 3 ) ( 1 − 8 u 3 ) (1+u^3)(1-8u^3) ( 1 + u 3 ) ( 1 − 8 u 3 ) .
Lemma 2.6. Let w = u 3 ∣ q 3 → x = x f 1 3 f 6 9 / ( f 2 3 f 3 9 ) w=u^3|_{q^3\to x}=xf_1^3f_6^9/(f_2^3f_3^9) w = u 3 ∣ q 3 → x = x f 1 3 f 6 9 / ( f 2 3 f 3 9 ) , where f m = f m ( x ) f_m=f_m(x) f m = f m ( x ) . Then,
1 + w = f 2 5 f 6 f 1 f 3 5 , 1 − 8 w = f 1 8 f 4 6 f 2 4 f 3 8 , and consequently Θ = f 9 9 f 7 3 f 6 f 18 3 . 1+w=\frac{f_2^5f_6}{f_1f_3^5},\qquad1-8w=\frac{f_1^8f_4^6}{f_2^4f_3^8},\qquad\text{and consequently}\qquad\Theta=\frac{f_9^9}{f_7^3f_6f_{18}^3}. 1 + w = f 1 f 3 5 f 2 5 f 6 , 1 − 8 w = f 2 4 f 3 8 f 1 8 f 4 6 , and consequently Θ = f 7 3 f 6 f 18 3 f 9 9 .
Proof. Both sides of each identity are modular functions of weight 0 on Γ 0 ( 6 ) \Gamma_0(6) Γ 0 ( 6 ) and are analytic (holomorphic) on the upper half-plane. By the valence formula (since the index of Γ 0 ( 6 ) \Gamma_0(6) Γ 0 ( 6 ) is 12), the difference between the two sides is uniquely determined by a finite number of coefficients in its q q q -expansion, which we can check to be equal.
To prove the final expression for Θ \Theta Θ , we multiply the first two identities to get ( 1 + w ) ( 1 − 8 w ) = f 1 7 f 2 5 f 6 5 / f 3 13 (1+w)(1-8w)=f_1^7f_2^5f_6^5/f_3^{13} ( 1 + w ) ( 1 − 8 w ) = f 1 7 f 2 5 f 6 5 / f 3 13 . In terms of q q q , this becomes ( 1 + u 3 ) ( 1 − 8 u 3 ) = f 7 7 f 6 5 / f 9 13 (1+u^3)(1-8u^3)=f_7^7f_6^5/f_9^{13} ( 1 + u 3 ) ( 1 − 8 u 3 ) = f 7 7 f 6 5 / f 9 13 . (Note: This product can also be proved elementarily without using modularity by starting from f 1 f 2 = f 3 ( 1 − u − 2 u 2 ) f_1f_2=f_3(1-u-2u^2) f 1 f 2 = f 3 ( 1 − u − 2 u 2 ) and multiplying over the three cube roots of unity q ↦ ω j q q\mapsto\omega^jq q ↦ ω j q .) Finally, dividing Λ = f 2 18 / f 9 4 \Lambda=f_2^{18}/f_9^4 Λ = f 2 18 / f 9 4 by this product yields the formula for Θ \Theta Θ .
Congruences for A k A_k A k
This section contains our main results. We begin by examining the 3-adic behavior of A k A_k A k : we completely classify the arithmetic progression 3 n + 2 3n+2 3 n + 2 and describe the classes 3 b ( 3 n + s ) 3^b(3n+s) 3 b ( 3 n + s ) modulo 3. We then turn to powers of 2 and the bipartition case k = 2 k=2 k = 2 , where an exact formula for the class 3 n + 1 3n+1 3 n + 1 lifts a known congruence modulo 9 to moduli 27 and 81. We first state the elementary divisibility that follows directly from Lemma 2.2.
Theorem 3.1. If p p p is a prime, p j ∣ k p^j\mid k p j ∣ k , and 1 ≤ r ≤ p − 1 1\le r\le p-1 1 ≤ r ≤ p − 1 , then A k ( p n + r ) ≡ B k ( p n + r ) ≡ 0 ( m o d p j ) A_k(pn+r)\equiv B_k(pn+r)\equiv0\pmod{p^j} A k ( p n + r ) ≡ B k ( p n + r ) ≡ 0 ( mod p j ) .
Proof. By Lemma 2.2, F k = ( F p j ) k / p j ≡ ( F ( q p ) p j − 1 ) k / p j ( m o d p j ) F^k=(F^{p^j})^{k/p^j}\equiv(F(q^p)^{p^{j-1}})^{k/p^j}\pmod{p^j} F k = ( F p j ) k / p j ≡ ( F ( q p ) p j − 1 ) k / p j ( mod p j ) , which is a series in q p q^p q p .
This represents the standard argument for k k k -tuples. The more interesting progressions are those it does not reach, which form the focus of the rest of the paper.
The progression 3 n + 2 3n+2 3 n + 2 is more delicate and is governed by the negative power F − 5 F^{-5} F − 5 . We write A − 5 ( n ) = [ q n ] F − 5 = [ q n ] ( f 1 f 2 f 3 ) 5 / f 6 5 A_{-5}(n)=[q^n]F^{-5}=[q^n](f_1f_2f_3)^5/f_6^5 A − 5 ( n ) = [ q n ] F − 5 = [ q n ] ( f 1 f 2 f 3 ) 5 / f 6 5 , even though this coefficient does not count combinatorial objects.
Proposition 3.2.
∑ n ≥ 0 A − 5 ( 3 n + 2 ) x n = − 81 x f 1 5 f 3 5 f 6 5 f 2 5 . \sum_{n\ge0}A_{-5}(3n+2)x^n=-81x\frac{f_1^5f_3^5f_6^5}{f_2^5}. n ≥ 0 ∑ A − 5 ( 3 n + 2 ) x n = − 81 x f 2 5 f 1 5 f 3 5 f 6 5 .
Proof. By Proposition 2.4, F − 5 = Λ − 5 ( 1 − u − 2 u 2 ) 5 F^{-5}=\Lambda^{-5}(1-u-2u^2)^5 F − 5 = Λ − 5 ( 1 − u − 2 u 2 ) 5 , and
( 1 − u − 2 u 2 ) 5 = 1 − 5 u + 30 u 3 − 15 u 4 − 81 u 5 + 30 u 6 + 120 u 7 − 80 u 9 − 32 u 10 . (1-u-2u^2)^5=1-5u+30u^3-15u^4-81u^5+30u^6+120u^7-80u^9-32u^{10}. ( 1 − u − 2 u 2 ) 5 = 1 − 5 u + 30 u 3 − 15 u 4 − 81 u 5 + 30 u 6 + 120 u 7 − 80 u 9 − 32 u 10 .
The only term with an exponent congruent to 2 (mod 3) is − 81 u 5 -81u^5 − 81 u 5 (since the coefficients of u 2 u^2 u 2 and u 8 u^8 u 8 vanish). Because Λ \Lambda Λ and ρ \rho ρ belong to Z [ [ q 3 ] ] \mathbb{Z}[[q^3]] Z [[ q 3 ]] , the part of F − 5 F^{-5} F − 5 with exponents congruent to 2 (mod 3) is − 81 q 5 ρ 5 Λ − 5 -81q^5\rho^5\Lambda^{-5} − 81 q 5 ρ 5 Λ − 5 . Dividing by q 2 q^2 q 2 , setting x = q 3 x=q^3 x = q 3 , and using the substitutions ρ ↦ f 1 3 f 6 3 / ( f 2 f 3 3 ) \rho\mapsto f_1^3f_6^3/(f_2f_3^3) ρ ↦ f 1 3 f 6 3 / ( f 2 f 3 3 ) and Λ ↦ f 6 2 / f 3 4 \Lambda\mapsto f_6^2/f_3^4 Λ ↦ f 6 2 / f 3 4 yields the result.
Theorem 3.3. Let k , e ≥ 1 k,e \ge1 k , e ≥ 1 . Then A k ( 3 n + 2 ) ≡ 0 ( m o d 3 e ) A_k(3n+2) \equiv0 \pmod{3^e} A k ( 3 n + 2 ) ≡ 0 ( mod 3 e ) for all n ≥ 0 n \ge0 n ≥ 0 if and only if
{ k ≡ 0 or k ≡ − 5 ( m o d 3 e ) , if e ≤ 4 , k ≡ 0 ( m o d 3 e ) , if e ≥ 5. \begin{cases}
k \equiv0 \text{ or } k \equiv-5 \pmod{3^e}, & \text{if } e \le4,\\
k \equiv0 \pmod{3^e}, & \text{if } e \ge5.
\end{cases} { k ≡ 0 or k ≡ − 5 ( mod 3 e ) , k ≡ 0 ( mod 3 e ) , if e ≤ 4 , if e ≥ 5. Proof. Since A k ( 2 ) = 3 k + ( k 2 ) = k ( k + 5 ) / 2 A_k(2)=3k+\binom{k}{2}=k(k+5)/2 A k ( 2 ) = 3 k + ( 2 k ) = k ( k + 5 ) /2 and gcd ( k , k + 5 ) ∣ 5 \gcd(k,k+5)\mid5 g cd( k , k + 5 ) ∣ 5 , a necessary condition is that 3 e 3^e 3 e divides either k k k or k + 5 k+5 k + 5 . If 3 e ∣ k 3^e\mid k 3 e ∣ k , the congruence holds by Theorem 3.1 . Let k = 3 e m − 5 k=3^e m-5 k = 3 e m − 5 with m ≥ 1 m\ge1 m ≥ 1 . Then F k = ( F 3 e ) m F − 5 F^k=(F^{3^e})^mF^{-5} F k = ( F 3 e ) m F − 5 , and by Lemma 2.2 , ( F 3 e ) m ≡ V ( q 3 ) ( m o d 3 e ) (F^{3^e})^m\equiv V(q^3)\pmod{3^e} ( F 3 e ) m ≡ V ( q 3 ) ( mod 3 e ) for some V ∈ Z [ [ q 3 ] ] V\in\mathbb{Z}[[q^3]] V ∈ Z [[ q 3 ]] with V ( 0 ) = 1 V(0)=1 V ( 0 ) = 1 . Thus, by Proposition 3.2,
∑ n A k ( 3 n + 2 ) x n = − 81 x V ( x ) f 1 5 f 3 5 f 6 5 f 2 5 ( m o d 3 e ) . \sum_{n} A_k(3n+2)x^n=-81xV(x)\frac{f_1^5f_3^5f_6^5}{f_2^5}\pmod{3^e}. n ∑ A k ( 3 n + 2 ) x n = − 81 x V ( x ) f 2 5 f 1 5 f 3 5 f 6 5 ( mod 3 e ) . For e ≤ 4 e\le4 e ≤ 4 , the right-hand side vanishes modulo 3 e 3^e 3 e . For e ≥ 5 e\ge5 e ≥ 5 , the coefficient of x x x is − 81 ≠ 0 ( m o d 3 5 ) -81\ne0\pmod{3^5} − 81 = 0 ( mod 3 5 ) , meaning 3 e ∤ A k ( 5 ) 3^e\nmid A_k(5) 3 e ∤ A k ( 5 ) .
Remark 3.4. For e = 1 , 2 e=1,2 e = 1 , 2 , the theorem states that
A k ( 3 n + 2 ) ≡ 0 ( m o d 3 ) ⟺ k ≡ 0 , 1 ( m o d 3 ) , A_k(3n+2)\equiv0\pmod{3}\Longleftrightarrow k\equiv0,1\pmod{3}, A k ( 3 n + 2 ) ≡ 0 ( mod 3 ) ⟺ k ≡ 0 , 1 ( mod 3 ) ,
A k ( 3 n + 2 ) ≡ 0 ( m o d 9 ) ⟺ k ≡ 0 , 4 ( m o d 9 ) . A_k(3n+2)\equiv0\pmod{9}\Longleftrightarrow k\equiv0,4\pmod{9}. A k ( 3 n + 2 ) ≡ 0 ( mod 9 ) ⟺ k ≡ 0 , 4 ( mod 9 ) .
For e = 3 , 4 e=3,4 e = 3 , 4 , the exceptional residues are k ≡ 22 ( m o d 27 ) k\equiv22\pmod{27} k ≡ 22 ( mod 27 ) and k ≡ 76 ( m o d 81 ) k\equiv76\pmod{81} k ≡ 76 ( mod 81 ) . In particular, modulo 3 5 3^5 3 5 , the progression 3 n + 2 3n+2 3 n + 2 yields no additional congruences beyond 3 5 ∣ k 3^5\mid k 3 5 ∣ k . For every k ≡ − 5 ( m o d 81 ) k\equiv-5\pmod{81} k ≡ − 5 ( mod 81 ) , the highest power of 3 3 3 dividing all A k ( 3 n + 2 ) A_k(3n+2) A k ( 3 n + 2 ) is exactly 3 4 3^4 3 4 : if 243 ∣ k + 5 243\mid k+5 243 ∣ k + 5 , then A k ( 5 ) ≡ − 81 ( m o d 243 ) A_k(5)\equiv-81\pmod{243} A k ( 5 ) ≡ − 81 ( mod 243 ) ; otherwise, v 3 ( A k ( 2 ) ) = 4 v_3(A_k(2))=4 v 3 ( A k ( 2 )) = 4 . The case k = 1 , e = 1 k=1,e=1 k = 1 , e = 1 corresponds to the classical congruence of Andrews, Lewis, and Lovejoy [1 ] .
For e ≤ 2 e\le2 e ≤ 2 , there is an alternative proof that avoids Hirschhorn’s dissection of f 1 f 2 f_1f_2 f 1 f 2 and explicitly connects the numbers A k ( n ) A_k(n) A k ( n ) to overpartitions, sums of squares, and the divisor function of the hexagonal lattice. Let p ‾ − m ( n ) \overline{p}_{-m}(n) p − m ( n ) denote the number of m m m -tuples of overpartitions of total size n n n , such that the generating function is given by ∑ p ‾ − m ( n ) q n = φ ( − q ) − m \sum\overline{p}_{-m}(n)q^n=\varphi(-q)^{-m} ∑ p − m ( n ) q n = φ ( − q ) − m .
Proposition 3.5. We have F ≡ φ ( − q ) − 2 F\equiv\varphi(-q)^{-2} F ≡ φ ( − q ) − 2 and G ≡ φ ( q ) / φ ( − q ) ( m o d 3 ) G\equiv\varphi(q)/\varphi(-q)\pmod{3} G ≡ φ ( q ) / φ ( − q ) ( mod 3 ) . Consequently, for all k ≥ 1 k\ge1 k ≥ 1 and n ≥ 0 n\ge0 n ≥ 0 ,
A k ( n ) ≡ p ‾ − 2 k ( n ) ( m o d 3 ) , B k ( n ) ≡ ∑ j r k ( j ) p ‾ − k ( n − j ) ( m o d 3 ) , A_k(n)\equiv\overline{p}_{-2k}(n)\pmod{3},\qquad B_k(n)\equiv\sum_j r_k(j)\overline{p}_{-k}(n-j)\pmod{3}, A k ( n ) ≡ p − 2 k ( n ) ( mod 3 ) , B k ( n ) ≡ j ∑ r k ( j ) p − k ( n − j ) ( mod 3 ) , where r k ( j ) r_k(j) r k ( j ) denotes the number of representations of j j j as an ordered sum of k k k squares.
Proof. Working modulo 3 3 3 , we use the standard basic relations f 3 ≡ f 1 3 f_3\equiv f_1^3 f 3 ≡ f 1 3 , f 6 ≡ f 2 3 f_6\equiv f_2^3 f 6 ≡ f 2 3 , and f 12 ≡ f 4 3 f_{12}\equiv f_4^3 f 12 ≡ f 4 3 . First, substituting these relations into the definition of F F F gives:
F = f 6 f 1 f 2 f 3 ≡ f 2 3 f 1 2 f 2 f 1 3 = f 2 2 f 1 4 = ( f 2 f 1 2 ) 2 ( m o d 3 ) . F=\frac{f_6}{f_1f_2f_3}\equiv\frac{f_2^3}{f_1^2f_2f_1^3}=\frac{f_2^2}{f_1^4}=\left(\frac{f_2}{f_1^2}\right)^2\pmod{3}. F = f 1 f 2 f 3 f 6 ≡ f 1 2 f 2 f 1 3 f 2 3 = f 1 4 f 2 2 = ( f 1 2 f 2 ) 2 ( mod 3 ) .
Since the generating function for overpartitions satisfies ∑ p ‾ ( n ) q n = f 2 / f 1 2 = φ ( − q ) − 1 \sum\overline{p}(n)q^n=f_2/f_1^2=\varphi(-q)^{-1} ∑ p ( n ) q n = f 2 / f 1 2 = φ ( − q ) − 1 , it immediately follows that F ≡ φ ( − q ) − 2 ( m o d 3 ) F\equiv\varphi(-q)^{-2}\pmod{3} F ≡ φ ( − q ) − 2 ( mod 3 ) .
Second, for G G G , we start with the classical theta-function identity
φ ( q ) φ ( − q ) = f 6 2 f 1 4 f 2 2 . \frac{\varphi(q)}{\varphi(-q)}=\frac{f_6^2}{f_1^4f_2^2}. φ ( − q ) φ ( q ) = f 1 4 f 2 2 f 6 2 .
To prove that G ≡ f 6 2 / ( f 1 4 f 2 2 ) ( m o d 3 ) G\equiv f_6^2/(f_1^4f_2^2)\pmod{3} G ≡ f 6 2 / ( f 1 4 f 2 2 ) ( mod 3 ) , we multiply G G G by the reciprocal expression and simplify modulo 3 3 3 :
G ⋅ f 1 4 f 2 2 f 6 2 = ( f 1 f 4 f 6 f 1 f 3 f 12 ) ⋅ f 1 4 f 2 2 f 6 2 = f 1 3 f 4 3 f 6 2 f 1 3 f 4 3 f 2 6 . G\cdot\frac{f_1^4f_2^2}{f_6^2}=\left(\frac{f_1f_4f_6}{f_1f_3f_{12}}\right)\cdot\frac{f_1^4f_2^2}{f_6^2}=\frac{f_1^3f_4^3f_6^2}{f_1^3f_4^3f_2^6}. G ⋅ f 6 2 f 1 4 f 2 2 = ( f 1 f 3 f 12 f 1 f 4 f 6 ) ⋅ f 6 2 f 1 4 f 2 2 = f 1 3 f 4 3 f 2 6 f 1 3 f 4 3 f 6 2 .
Using f 6 ≡ f 3 2 ( m o d 3 ) f_6 \equiv f_3^2 \pmod{3} f 6 ≡ f 3 2 ( mod 3 ) , the numerator becomes f 1 3 f 3 4 ( f 3 2 ) 2 = f 1 3 f 3 4 f 6 2 f_1^3 f_3^4 (f_3^2)^2 = f_1^3 f_3^4 f_6^2 f 1 3 f 3 4 ( f 3 2 ) 2 = f 1 3 f 3 4 f 6 2 , which exactly matches the denominator. Thus, the product is congruent to 1 ( m o d 3 ) 1 \pmod{3} 1 ( mod 3 ) , proving that G ≡ f 6 2 f 1 2 f 4 2 = φ ( q ) φ ( − q ) ( m o d 3 ) G \equiv\frac{f_6^2}{f_1^2 f_4^2} = \frac{\varphi(q)}{\varphi(-q)} \pmod{3} G ≡ f 1 2 f 4 2 f 6 2 = φ ( − q ) φ ( q ) ( mod 3 ) .
Finally, raising these congruences to the power k k k yields the desired coefficient relations for A k ( n ) = [ q n ] F k A_k(n)=[q^n]F^k A k ( n ) = [ q n ] F k and B k ( n ) = [ q n ] G k B_k(n)=[q^n]G^k B k ( n ) = [ q n ] G k in terms of overpartition tuples and sums of squares. □
Corollary 3.6. For j ≥ 0 j \ge0 j ≥ 0 , F 3 j + 1 ≡ φ ( − q ) φ ( − q 3 ) − 2 j − 1 F^{3j+1} \equiv\varphi(-q)\varphi(-q^3)^{-2j-1} F 3 j + 1 ≡ φ ( − q ) φ ( − q 3 ) − 2 j − 1 and F 3 j + 2 ≡ φ ( − q ) 2 φ ( − q 3 ) − 2 j − 2 ( m o d 3 ) F^{3j+2} \equiv\varphi(-q)^2\varphi(-q^3)^{-2j-2} \pmod{3} F 3 j + 2 ≡ φ ( − q ) 2 φ ( − q 3 ) − 2 j − 2 ( mod 3 ) . This clearly shows the dependence on k k k mod 3: φ ( − q ) \varphi(-q) φ ( − q ) has no exponents congruent to 2 ( m o d 3 ) \pmod{3} ( mod 3 ) , while φ ( − q ) 2 \varphi(-q)^2 φ ( − q ) 2 has coefficient 4 at q 2 q^2 q 2 .
Proof. This follows directly from φ ( − q ) − 3 ≡ φ ( − q 3 ) − 1 \varphi(-q)^{-3} \equiv\varphi(-q^3)^{-1} φ ( − q ) − 3 ≡ φ ( − q 3 ) − 1 (Lemma 2.2 ). □
Proposition 3.7. Let T ( q ) = ∑ n ≥ 1 S ( n ) q n − ∑ n ≥ 1 S ( n ) q 2 n T(q)=\sum_{n\ge1}S(n)q^n-\sum_{n\ge1}S(n)q^{2n} T ( q ) = ∑ n ≥ 1 S ( n ) q n − ∑ n ≥ 1 S ( n ) q 2 n . Then
F = φ ( − q ) − 2 b ( q ) b ( q 2 ) , F k ≡ φ ( − q ) − 2 k ( 1 − 3 k T ( q ) ) ( m o d 9 ) . F=\varphi(-q)^{-2}\frac{b(q)}{b(q^2)}, \qquad F^k\equiv\varphi(-q)^{-2k}(1-3kT(q)) \pmod{9}. F = φ ( − q ) − 2 b ( q 2 ) b ( q ) , F k ≡ φ ( − q ) − 2 k ( 1 − 3 k T ( q )) ( mod 9 ) . Equivalently, A k ( n ) = p ‾ − 2 k ( n ) − 3 k ∑ j t ( j ) p ‾ − 2 k ( n − j ) ( m o d 9 ) A_k(n)=\overline{p}_{-2k}(n)-3k\sum_j t(j)\overline{p}_{-2k}(n-j)\pmod{9} A k ( n ) = p − 2 k ( n ) − 3 k ∑ j t ( j ) p − 2 k ( n − j ) ( mod 9 ) , where t ( j ) = S ( j ) − S ( j / 2 ) t(j)=S(j)-S(j/2) t ( j ) = S ( j ) − S ( j /2 ) and S ( j / 2 ) = 0 S(j/2)=0 S ( j /2 ) = 0 for odd j j j .
Proof. We start with the exact identity F φ ( − q ) 2 = f 1 3 f 6 / ( f 2 3 f 3 ) = b ( q ) / b ( q 2 ) F\varphi(-q)^2=f_1^3f_6/(f_2^3f_3)=b(q)/b(q^2) F φ ( − q ) 2 = f 1 3 f 6 / ( f 2 3 f 3 ) = b ( q ) / b ( q 2 ) . Since b ( q ) ≡ 1 − 3 ∑ S ( n ) q n ( m o d 9 ) b(q)\equiv1-3\sum S(n)q^n\pmod{9} b ( q ) ≡ 1 − 3 ∑ S ( n ) q n ( mod 9 ) , working modulo 9 gives:
b ( q ) b ( q 2 ) ≡ 1 − 3 ∑ S ( n ) q n 1 − 3 ∑ S ( n ) q 2 n ≡ ( 1 − 3 ∑ S ( n ) q n ) ( 1 + 3 ∑ S ( n ) q 2 n ) = 1 − 3 T ( q ) ( m o d 9 ) . \frac{b(q)}{b(q^2)}\equiv\frac{1-3\sum S(n)q^n}{1-3\sum S(n)q^{2n}}\equiv\left(1-3\sum S(n)q^n\right)\left(1+3\sum S(n)q^{2n}\right)=1-3T(q)\pmod{9}. b ( q 2 ) b ( q ) ≡ 1 − 3 ∑ S ( n ) q 2 n 1 − 3 ∑ S ( n ) q n ≡ ( 1 − 3 ∑ S ( n ) q n ) ( 1 + 3 ∑ S ( n ) q 2 n ) = 1 − 3 T ( q ) ( mod 9 ) . Raising both sides to the power k k k yields the desired congruence for F k F^k F k . □
In other words, modulo 9, the numbers A k ( n ) A_k(n) A k ( n ) are determined by overpartition tuples combined with the divisor function of the hexagonal lattice.
**Second proof of Theorem 3.3 for e ≤ 2. ∗ ∗ F o r e\le2.** For e ≤ 2. ∗ ∗ F or e=1$, the statement follows from Corollary 3.6 together with the coefficient 4 at q 2 q^2 q 2 of φ ( − q ) 2 \varphi(-q)^2 φ ( − q ) 2 when k ≡ 2 ( m o d 3 ) k\equiv2\pmod{3} k ≡ 2 ( mod 3 ) . For e = 2 e=2 e = 2 , necessity follows from A k ( 2 ) = k ( k + 5 ) / 2 A_k(2)=k(k+5)/2 A k ( 2 ) = k ( k + 5 ) /2 , while the case $9\mid k$ is covered by Theorem 3.1 . By Lemma 2.3 , it remains to check k = 4 k=4 k = 4 . Writing φ = φ ( − q ) \varphi=\varphi(-q) φ = φ ( − q ) and ψ = φ ( − q 3 ) \psi=\varphi(-q^3) ψ = φ ( − q 3 ) , this proof relies entirely on classical theta-function dissections.
By Proposition 3.7 , F 4 ≡ φ − 8 ( 1 − 12 T ) ≡ φ − 8 ( 1 − 3 T ) ( m o d 9 ) F^4\equiv\varphi^{-8}(1-12T)\equiv\varphi^{-8}(1-3T)\pmod{9} F 4 ≡ φ − 8 ( 1 − 12 T ) ≡ φ − 8 ( 1 − 3 T ) ( mod 9 ) . Since φ 9 ≡ ψ 3 ( m o d 9 ) \varphi^9\equiv\psi^3\pmod{9} φ 9 ≡ ψ 3 ( mod 9 ) (Lemma 2.2 ), we have φ − 8 = φ ⋅ φ − 9 ≡ φ ψ − 3 \varphi^{-8}=\varphi\cdot\varphi^{-9}\equiv\varphi\psi^{-3} φ − 8 = φ ⋅ φ − 9 ≡ φ ψ − 3 , which leads to
F 4 ≡ ψ − 3 φ ( 1 − 3 T ) ( m o d 9 ) . F^4\equiv\psi^{-3}\varphi(1-3T)\pmod{9}. F 4 ≡ ψ − 3 φ ( 1 − 3 T ) ( mod 9 ) . where ψ − 3 \psi^{-3} ψ − 3 is a unit series in q 3 q^3 q 3 . Therefore, it suffices to show that all coefficients of q 3 n + 2 q^{3n+2} q 3 n + 2 in φ ( 1 − 3 T ) \varphi(1-3T) φ ( 1 − 3 T ) are divisible by 9. Since φ = ∑ ( − 1 ) m q m 2 \varphi=\sum(-1)^m q^{m^2} φ = ∑ ( − 1 ) m q m 2 has no exponents congruent to 2 ( m o d 3 ) \pmod{3} ( mod 3 ) , this reduces to showing that
[ q 3 n + 2 ] φ T ≡ 0 ( m o d 3 ) . [q^{3n+2}]\varphi T\equiv0\pmod{3}. [ q 3 n + 2 ] φT ≡ 0 ( mod 3 ) . We apply a 3-dissection to the components. Let S ( q ) = ∑ n ≥ 1 S ( n ) q n S(q)=\sum_{n\ge1}S(n)q^n S ( q ) = ∑ n ≥ 1 S ( n ) q n . Using Borwein’s dissections a ( q ) = a ( q 3 ) + 6 q f 9 3 / f 3 a(q)=a(q^3)+6qf_9^3/f_3 a ( q ) = a ( q 3 ) + 6 q f 9 3 / f 3 and a = 1 + 6 S a=1+6S a = 1 + 6 S , we obtain
S ( q ) = S ( q 3 ) + q f 9 3 f 3 , S(q)=S(q^3)+q\frac{f_9^3}{f_3}, S ( q ) = S ( q 3 ) + q f 3 f 9 3 , implying that S ( 3 n + 2 ) = 0 S(3n+2)=0 S ( 3 n + 2 ) = 0 .
Replacing q q q by q 2 q^2 q 2 and splitting according to n n n mod 3 gives
S ( q 2 ) = S ( q 6 ) + q 2 f 18 3 f 6 . S(q^2)=S(q^6)+q^2\frac{f_{18}^3}{f_6}. S ( q 2 ) = S ( q 6 ) + q 2 f 6 f 18 3 . Consequently, T = [ S ( q 3 ) − S ( q 6 ) ] + q f 9 3 / f 3 − q 2 f 18 3 / f 6 T=[S(q^3)-S(q^6)]+qf_9^3/f_3-q^2f_{18}^3/f_6 T = [ S ( q 3 ) − S ( q 6 )] + q f 9 3 / f 3 − q 2 f 18 3 / f 6 , with the three summands falling into classes 0, 1, and 2 modulo 3, respectively. From Hirschhorn’s dissection $\varphi(-q)=\varphi(-q^9)-
2 q f 3 f 18 2 / ( f 6 f 9 ) 2q f_{3}f_{18}^{2}/(f_{6}f_{9}) 2 q f 3 f 18 2 / ( f 6 f 9 ) , the class-0 part is f 9 2 / f 18 f_{9}^{2}/f_{18} f 9 2 / f 18 and the class-1 part is − 2 q f 3 f 18 2 / ( f 6 f 9 ) -2q f_{3}f_{18}^{2}/(f_{6}f_{9}) − 2 q f 3 f 18 2 / ( f 6 f 9 ) . The class-2 part of φ T \varphi T φT is therefore given by
− f 9 2 f 18 ⋅ f 18 3 f 6 − 2 q f 3 f 18 2 f 6 f 9 ⋅ q f 9 3 f 3 = − 3 q 2 f 9 2 f 18 2 f 6 ≡ 0 ( m o d 3 ) . -\frac{f_{9}^{2}}{f_{18}}\cdot\frac{f_{18}^{3}}{f_{6}}-2q\frac{f_{3}f_{18}^{2}}{f_{6}f_{9}}\cdot q\frac{f_{9}^{3}}{f_{3}}=-3q^{2}\frac{f_{9}^{2}f_{18}^{2}}{f_{6}}\equiv0\pmod{3}. − f 18 f 9 2 ⋅ f 6 f 18 3 − 2 q f 6 f 9 f 3 f 18 2 ⋅ q f 3 f 9 3 = − 3 q 2 f 6 f 9 2 f 18 2 ≡ 0 ( mod 3 ) .
This completes the proof for k ≡ 4 ( m o d 9 ) k\equiv4\pmod{9} k ≡ 4 ( mod 9 ) . □
We next record the classes 9 n + 3 9n+3 9 n + 3 and 9 n + 6 9n+6 9 n + 6 and, more generally, 3 b ( 3 n + s ) 3^{b}(3n+s) 3 b ( 3 n + s ) modulo 3 3 3 .
Theorem 3.8. For every k ≥ 0 k\geq0 k ≥ 0 :
(a) A k ( 9 n + 3 ) ≡ 0 ( m o d 3 ) A_k(9n+3)\equiv0\pmod{3} A k ( 9 n + 3 ) ≡ 0 ( mod 3 ) for all n n n if and only if k ≡ 0 , 4 , 8 ( m o d 9 ) k\equiv0,4,8\pmod{9} k ≡ 0 , 4 , 8 ( mod 9 ) ;
(b) A k ( 9 n + 6 ) ≡ 0 ( m o d 3 ) A_k(9n+6)\equiv0\pmod{3} A k ( 9 n + 6 ) ≡ 0 ( mod 3 ) for all n n n if and only if k ≡ 0 , 2 , 3 , 4 , 7 , 8 ( m o d 9 ) k\equiv0,2,3,4,7,8\pmod{9} k ≡ 0 , 2 , 3 , 4 , 7 , 8 ( mod 9 ) .
Proof. All congruences take place in F 3 [ [ x ] ] \mathbb{F}_{3}[[x]] F 3 [[ x ]] . Let c , ρ , u = x ρ , Λ c,\rho,u=x\rho,\Lambda c , ρ , u = x ρ , Λ be as in Proposition 2.4, with q q q renamed x x x .
Step 1: F F F modulo 3 3 3 . Since − 2 ≡ 1 ( m o d 3 ) -2\equiv1\pmod{3} − 2 ≡ 1 ( mod 3 ) we have 1 − u − 2 u 2 ≡ ( 1 + u ) 2 1-u-2u^{2}\equiv(1+u)^{2} 1 − u − 2 u 2 ≡ ( 1 + u ) 2 , and therefore
f 1 f 2 ≡ c ( 1 + u ) 2 , F = Λ 1 − u − 2 u 2 ≡ Λ ( 1 + u ) 2 = Λ ( 1 + u ) ( 1 + u ) 3 ≡ κ ( 1 + u ) , f_{1}f_{2}\equiv c(1+u)^{2},\qquad F=\frac{\Lambda}{1-u-2u^{2}}\equiv\frac{\Lambda}{(1+u)^{2}}=\frac{\Lambda(1+u)}{(1+u)^{3}}\equiv\kappa(1+u), f 1 f 2 ≡ c ( 1 + u ) 2 , F = 1 − u − 2 u 2 Λ ≡ ( 1 + u ) 2 Λ = ( 1 + u ) 3 Λ ( 1 + u ) ≡ κ ( 1 + u ) , where κ = f 18 2 / ( f 9 4 ( 1 + u 3 ) ) ∈ F 3 [ [ x 3 ] ] \kappa=f_{18}^{2}/(f_{9}^{4}(1+u^{3}))\in\mathbb{F}_{3}[[x^{3}]] κ = f 18 2 / ( f 9 4 ( 1 + u 3 )) ∈ F 3 [[ x 3 ]] .
Step 2: exponent classes. Put R = F 3 [ [ x 3 ] ] R=\mathbb{F}_{3}[[x^{3}]] R = F 3 [[ x 3 ]] . Since ρ ∈ R \rho\in R ρ ∈ R is a unit and the orders of the series u 3 i u^{3i} u 3 i are strictly increasing, a series ∑ i c i u 3 i + s \sum_i c_i u^{3i+s} ∑ i c i u 3 i + s with c i ∈ F 3 c_i\in\mathbb{F}_{3} c i ∈ F 3 vanishes if and only if all c i c_i c i do. In this sense the classes of exponents of x x x modulo 3 3 3 correspond to the classes of exponents of u u u modulo 3 3 3 .
Step 3: the part 3 m 3m 3 m . Write k = 3 t + r k=3t+r k = 3 t + r with 0 ≤ r ≤ 2 0\leq r\leq2 0 ≤ r ≤ 2 . Then F k ≡ κ k ∑ j ( k j ) u j F^k\equiv\kappa^k\sum_j\binom{k}{j}u^j F k ≡ κ k ∑ j ( j k ) u j , and by Lucas’ theorem ( k 3 i ) = ( t i ) \binom{k}{3i}=\binom{t}{i} ( 3 i k ) = ( i t ) . So the part of F k F^k F k with exponents divisible by 3 3 3 is κ k ∑ i ( t i ) u 3 i \kappa^k\sum_i\binom{t}{i}u^{3i} κ k ∑ i ( i t ) u 3 i . Writing y = x 3 y=x^3 y = x 3 and using u ( x ) 3 ≡ u ( y ) u(x)^3\equiv u(y) u ( x ) 3 ≡ u ( y ) we obtain
∑ m ≥ 0 A k ( 3 m ) y m ≡ ( f 6 2 f 3 4 ) k ( 1 + u ) t − k , u = u ( y ) = y ρ ( y ) , \sum_{m\geq0}A_k(3m)y^m\equiv\left(\frac{f_{6}^{2}}{f_{3}^{4}}\right)^k(1+u)^{t-k},\qquad u=u(y)=y\rho(y), m ≥ 0 ∑ A k ( 3 m ) y m ≡ ( f 3 4 f 6 2 ) k ( 1 + u ) t − k , u = u ( y ) = y ρ ( y ) , where now f j = f j ( y ) f_j=f_j(y) f j = f j ( y ) and f 6 2 / f 3 4 f_{6}^{2}/f_{3}^{4} f 6 2 / f 3 4 is a unit of F 3 [ [ y 3 ] ] \mathbb{F}_{3}[[y^{3}]] F 3 [[ y 3 ]] .
Step 4: the vanishing criterion. Choose e e e with N = 3 e + t − k ≥ 0 N=3e+t-k\geq0 N = 3 e + t − k ≥ 0 . Then ( 1 + u ) t − k = ( 1 + u ) N / ( 1 + u 3 ) e (1+u)^{t-k}=(1+u)^N/(1+u^3)^e ( 1 + u ) t − k = ( 1 + u ) N / ( 1 + u 3 ) e , and ( 1 + u 3 ) e (1+u^3)^e ( 1 + u 3 ) e is a unit of F 3 [ [ y 3 ] ] \mathbb{F}_{3}[[y^{3}]] F 3 [[ y 3 ]] . By Step 2, the coefficients of y 3 n + s y^{3n+s} y 3 n + s (s = 1 , 2 s=1,2 s = 1 , 2 ) vanish for all n n n if and only if ( N 3 i + s ) = 0 \binom{N}{3i+s}=0 ( 3 i + s N ) = 0 for all i i i . By Lucas’ theorem
( N 3 i + s ) ≡ ( N 0 s ) ( ⌊ N / 3 ⌋ i ) , N 0 = N m o d 3 , \binom{N}{3i+s}\equiv\binom{N_0}{s}\binom{\lfloor N/3\rfloor}{i},\qquad N_0=N\bmod3, ( 3 i + s N ) ≡ ( s N 0 ) ( i ⌊ N /3 ⌋ ) , N 0 = N mod 3 , so the condition is N 0 < s N_0<s N 0 < s , i.e. ( t − k ) m o d 3 < s (t-k)\bmod3<s ( t − k ) mod 3 < s . As t − k = − 2 t − r ≡ t − r ( m o d 3 ) t-k=-2t-r\equiv t-r\pmod{3} t − k = − 2 t − r ≡ t − r ( mod 3 ) , this reads ( t − r ) m o d 3 < s (t-r)\bmod3<s ( t − r ) mod 3 < s .
Step 5: the residues. Running through the nine residues k m o d 9 = 3 ( t m o d 3 ) + r k\bmod9=3(t\bmod3)+r k mod 9 = 3 ( t mod 3 ) + r , the condition ( t − r ) m o d 3 = 0 (t-r)\bmod3=0 ( t − r ) mod 3 = 0 (for s = 1 s=1 s = 1 ) holds exactly for k ≡ 0 , 4 , 8 ( m o d 9 ) k\equiv0,4,8\pmod{9} k ≡ 0 , 4 , 8 ( mod 9 ) , and the condition ( t − r ) m o d 3 ∈ { 0 , 1 } (t-r)\bmod3\in\{0,1\} ( t − r ) mod 3 ∈ { 0 , 1 } (for s = 2 s=2 s = 2 ) holds exactly for k ≡ 0 , 2 , 3 , 4 , 7 , 8 ( m o d 9 ) k\equiv0,2,3,4,7,8\pmod{9} k ≡ 0 , 2 , 3 , 4 , 7 , 8 ( mod 9 ) . Finally A k ( 9 n + 3 ) = A k ( 3 ( 3 n + 1 ) ) A_k(9n+3)=A_k(3(3n+1)) A k ( 9 n + 3 ) = A k ( 3 ( 3 n + 1 )) and A k ( 9 n + 6 ) = A k ( 3 ( 3 n + 2 ) ) A_k(9n+6)=A_k(3(3n+2)) A k ( 9 n + 6 ) = A k ( 3 ( 3 n + 2 )) .
Theorem 3.9. Let integers λ a \lambda_a λ a be defined by λ 0 = 0 \lambda_0=0 λ 0 = 0 , λ 1 = − 4 \lambda_1=-4 λ 1 = − 4 , and λ a + 1 = − λ a + 4 λ a − 1 \lambda_{a+1}=-\lambda_a+4\lambda_{a-1} λ a + 1 = − λ a + 4 λ a − 1 (giving λ a = − 4 , 4 , − 20 , 36 , − 116 , … \lambda_a=-4,4,-20,36,-116,\ldots λ a = − 4 , 4 , − 20 , 36 , − 116 , … ), and for k ≥ 0 k\geq0 k ≥ 0 , let M 1 = ⌊ k / 3 ⌋ − k M_1=\lfloor k/3\rfloor-k M 1 = ⌊ k /3 ⌋ − k and M a + 1 = ⌊ M a / 3 ⌋ + 2 λ a k M_{a+1}=\lfloor M_a/3\rfloor+2\lambda_a k M a + 1 = ⌊ M a /3 ⌋ + 2 λ a k . Then, for all b ≥ 1 b\geq1 b ≥ 1 , s ∈ { 1 , 2 } s\in\{1,2\} s ∈ { 1 , 2 } , and k ≥ 0 k\geq0 k ≥ 0 ,
A k ( 3 b ( 3 n + s ) ) ≡ 0 ( m o d 3 ) for all n ≥ 0 ⟺ M b ( k ) m o d 3 < s . A_k\left(3^b(3n+s)\right)\equiv0\pmod{3}\quad\text{for all }n\geq0\Longleftrightarrow M_b(k)\bmod3<s. A k ( 3 b ( 3 n + s ) ) ≡ 0 ( mod 3 ) for all n ≥ 0 ⟺ M b ( k ) mod 3 < s .
Proof. The argument iterates the method of Theorem 3.8 using the notation c , ρ , u , R c,\rho,u,R c , ρ , u , R . Let
Φ a ( z ) = ∑ m ≥ 0 A k ( 3 a m ) z m ∈ F 3 [ [ z ] ] . \Phi_a(z)=\sum_{m\geq0}A_k(3^am)z^m\in\mathbb{F}_3[[z]]. Φ a ( z ) = m ≥ 0 ∑ A k ( 3 a m ) z m ∈ F 3 [[ z ]] .
We first establish by induction that for every a ≥ 1 a\geq1 a ≥ 1 ,
Φ a ( z ) = W a ( z ) ( 1 + u ( z ) ) M a , \Phi_a(z)=W_a(z)(1+u(z))^{M_a}, Φ a ( z ) = W a ( z ) ( 1 + u ( z ) ) M a ,
where W a = f 3 p 3 f 6 p 6 f 9 p 9 f 18 p 18 ∈ F 3 [ [ z 3 ] ] W_a=f_3^{p_3}f_6^{p_6}f_9^{p_9}f_{18}^{p_{18}}\in\mathbb{F}_3[[z^3]] W a = f 3 p 3 f 6 p 6 f 9 p 9 f 18 p 18 ∈ F 3 [[ z 3 ]] with exponent vector p ( a ) = ( p 3 , p 6 , p 9 , p 18 ) p^{(a)}=(p_3,p_6,p_9,p_{18}) p ( a ) = ( p 3 , p 6 , p 9 , p 18 ) . For a = 1 a=1 a = 1 , this holds by Theorem 3.8 with p ( 1 ) = k ( − 4 , 2 , 0 , 0 ) p^{(1)}=k(-4,2,0,0) p ( 1 ) = k ( − 4 , 2 , 0 , 0 ) . Assuming the relation holds for some a a a in terms of the variable y y y , we choose e e e such that N = − M a + 3 e ≥ 0 N=-M_a+3e\geq0 N = − M a + 3 e ≥ 0 , allowing us to write ( 1 + u ( y ) ) M a = ( 1 + u ( y ) ) N ( 1 + u ( y ) 3 ) − e (1+u(y))^{M_a}=(1+u(y))^N(1+u(y)^3)^{-e} ( 1 + u ( y ) ) M a = ( 1 + u ( y ) ) N ( 1 + u ( y ) 3 ) − e . By Lucas’ theorem, the part of ( 1 + u ) N (1+u)^N ( 1 + u ) N with exponents congruent to 0 ( m o d 3 ) 0\pmod3 0 ( mod 3 ) is ( 1 + u 3 ) ⌊ N / 3 ⌋ (1+u^3)^{\lfloor N/3\rfloor} ( 1 + u 3 ) ⌊ N /3 ⌋ . Since W a ( y ) ∈ F 3 [ [ y 3 ] ] W_a(y)\in\mathbb{F}_3[[y^3]] W a ( y ) ∈ F 3 [[ y 3 ]] and u ( y ) 3 = u ( y 3 ) u(y)^3=u(y^3) u ( y ) 3 = u ( y 3 ) , setting z = y 3 z=y^3 z = y 3 yields
Φ a + 1 ( z ) = W a ( y ) ( 1 + u ( z ) ) ⌊ M a / 3 ⌋ . \Phi_{a+1}(z)=W_a(y)(1+u(z))^{\lfloor M_a/3\rfloor}. Φ a + 1 ( z ) = W a ( y ) ( 1 + u ( z ) ) ⌊ M a /3 ⌋ .
Using f 3 j ( y ) = f j ( z ) f_{3j}(y)=f_j(z) f 3 j ( y ) = f j ( z ) , W a ( y ) W_a(y) W a ( y ) evaluates to f 1 p 3 f 2 p 6 f 3 p 9 f 6 p 18 f_1^{p_3}f_2^{p_6}f_3^{p_9}f_6^{p_{18}} f 1 p 3 f 2 p 6 f 3 p 9 f 6 p 18 at z z z . By induction, p 6 = − p 3 / 2 p_6=-p_3/2 p 6 = − p 3 /2 (which holds for a = 1 a=1 a = 1 and is preserved by the recursion), and since all p 3 p_3 p 3 are even, p 6 − p 3 = − 3 2 p 3 p_6-p_3=-\frac{3}{2}p_3 p 6 − p 3 = − 2 3 p 3 is divisible by 3 3 3 . Applying the relation
f 1 p 3 f 2 p 6 = ( f 1 f 2 ) p 3 f 2 p 6 − p 3 ≡ c p 3 ( 1 + u ) 2 p 3 f 6 ( p 6 − p 3 ) / 3 ( m o d 3 ) (1) f_1^{p_3}f_2^{p_6}=(f_1f_2)^{p_3}f_2^{p_6-p_3}\equiv c^{p_3}(1+u)^{2p_3}f_6^{(p_6-p_3)/3}\pmod3 \tag*{(1)} f 1 p 3 f 2 p 6 = ( f 1 f 2 ) p 3 f 2 p 6 − p 3 ≡ c p 3 ( 1 + u ) 2 p 3 f 6 ( p 6 − p 3 ) /3 ( mod 3 ) (1)
gives the formula for Φ a + 1 \Phi_{a+1} Φ a + 1 with the updated exponent vector
p ( a + 1 ) = ( − p 3 + p 9 , 2 p 3 + p 6 3 + p 18 , 4 p 3 , − 2 p 3 ) p^{(a+1)}=\left(-p_3+p_9,\frac{2p_3+p_6}{3}+p_{18},4p_3,-2p_3\right) p ( a + 1 ) = ( − p 3 + p 9 , 3 2 p 3 + p 6 + p 18 , 4 p 3 , − 2 p 3 )
and M a + 1 = ⌊ M a / 3 ⌋ + 2 p 3 ( a ) M_{a+1}=\lfloor M_a/3\rfloor+2p_3^{(a)} M a + 1 = ⌊ M a /3 ⌋ + 2 p 3 ( a ) . Noting that p 3 ( a ) = λ a k p_3^{(a)}=\lambda_a k p 3 ( a ) = λ a k (since p 3 ( a + 1 ) = − p 3 ( a ) + p 9 ( a ) = − p 3 ( a ) + 4 p 3 ( a − 1 ) p_3^{(a+1)}=-p_3^{(a)}+p_9^{(a)}=-p_3^{(a)}+4p_3^{(a-1)} p 3 ( a + 1 ) = − p 3 ( a ) + p 9 ( a ) = − p 3 ( a ) + 4 p 3 ( a − 1 ) ), we obtain M a + 1 = ⌊ M a / 3 ⌋ + 2 λ a k M_{a+1}=\lfloor M_a/3\rfloor+2\lambda_a k M a + 1 = ⌊ M a /3 ⌋ + 2 λ a k .
Finally, A k ( 3 b ( 3 n + s ) ) A_k(3^b(3n+s)) A k ( 3 b ( 3 n + s )) is the coefficient of z 3 n + s z^{3n+s} z 3 n + s in Φ b \Phi_b Φ b . As before, this coefficient vanishes for all n n n if and only if ( N 3 i + s ) ≡ 0 \binom{N}{3i+s}\equiv0 ( 3 i + s N ) ≡ 0 for all i i i , which is equivalent to the condition M b m o d 3 < s M_b\bmod3<s M b mod 3 < s .
Remark 3.10. For b = 1 b=1 b = 1 this is Theorem 3.8 . For b = 2 b=2 b = 2 the sets of k m o d 27 k\bmod27 k mod 27 are { 0 , 1 , 4 , 5 , 8 , 12 , 16 , 20 , 24 } \{0,1,4,5,8,12,16,20,24\} { 0 , 1 , 4 , 5 , 8 , 12 , 16 , 20 , 24 } for s = 1 s=1 s = 1 and { 0 , 1 , 2 , 4 , 5 , 6 , 8 , 9 , 10 , 12 , 13 , 14 , 16 , 17 , 20 , 21 , 24 , 25 } \{0,1,2,4,5,6,8,9,10,12,13,14,16,17,20,21,24,25\} { 0 , 1 , 2 , 4 , 5 , 6 , 8 , 9 , 10 , 12 , 13 , 14 , 16 , 17 , 20 , 21 , 24 , 25 } for s = 2 s=2 s = 2 . For k = 2 k=2 k = 2 and s = 2 s=2 s = 2 the recursion gives A 2 ( 3 b ( 3 n + 2 ) ) ≡ 0 ( m o d 3 ) A_2(3^b(3n+2))\equiv0\pmod3 A 2 ( 3 b ( 3 n + 2 )) ≡ 0 ( mod 3 ) for every b ≥ 1 b\geq1 b ≥ 1 , which contains the first family of Naika and Shivashankar [10 ] (there b = α + 3 b=\alpha+3 b = α + 3 ); for k = 1 k=1 k = 1 it gives vanishing for even b b b and both values of s s s , and no vanishing for odd b b b .
We now turn to powers of 2 2 2 and to the case k = 2 k=2 k = 2 . The odd parts of F 2 F^2 F 2 , G G G and G 2 G^2 G 2 all come from one classical 2 2 2 -dissection.
Lemma 3.11. Hirschhorn’s 2 2 2 -dissection [3 ] reads 1 / ( f 1 f 3 ) = X ( q 2 ) + q Y ( q 2 ) 1/(f_1f_3)=X(q^2)+qY(q^2) 1/ ( f 1 f 3 ) = X ( q 2 ) + q Y ( q 2 ) with
X = f 2 4 f 5 6 f 1 2 f 2 f 3 f 4 f 12 , Y = f 5 2 f 12 2 f 1 4 f 2 2 f 4 f 6 2 , X Y = f 2 4 f 4 4 f 1 6 f 3 6 . X=\frac{f_2^4f_5^6}{f_1^2f_2f_3f_4f_{12}},\qquad
Y=\frac{f_5^2f_{12}^2}{f_1^4f_2^2f_4f_6^2},\qquad
XY=\frac{f_2^4f_4^4}{f_1^6f_3^6}. X = f 1 2 f 2 f 3 f 4 f 12 f 2 4 f 5 6 , Y = f 1 4 f 2 2 f 4 f 6 2 f 5 2 f 12 2 , X Y = f 1 6 f 3 6 f 2 4 f 4 4 . Consequently
∑ m ≥ 0 A 2 ( 2 m + 1 ) q m = 2 f 2 4 f 4 4 f 8 8 f 3 4 , ∑ m ≥ 0 B 1 ( 2 m + 1 ) q m = f 6 6 f 12 2 f 1 4 f 2 4 f 6 2 , ∑ m ≥ 0 B 2 ( 2 m + 1 ) q m = 2 f 6 6 f 2 6 f 1 6 f 3 2 . \sum_{m\geq0}A_2(2m+1)q^m=\frac{2f_2^4f_4^4}{f_8^8f_3^4},
\qquad
\sum_{m\geq0}B_1(2m+1)q^m=\frac{f_6^6f_{12}^2}{f_1^4f_2^4f_6^2},
\qquad
\sum_{m\geq0}B_2(2m+1)q^m=\frac{2f_6^6f_2^6}{f_1^6f_3^2}. m ≥ 0 ∑ A 2 ( 2 m + 1 ) q m = f 8 8 f 3 4 2 f 2 4 f 4 4 , m ≥ 0 ∑ B 1 ( 2 m + 1 ) q m = f 1 4 f 2 4 f 6 2 f 6 6 f 12 2 , m ≥ 0 ∑ B 2 ( 2 m + 1 ) q m = f 1 6 f 3 2 2 f 6 6 f 2 6 . Proof. F 2 = ( f 6 2 / f 2 2 ) ( 1 / ( f 1 f 3 ) ) 2 F^2=(f_6^2/f_2^2)(1/(f_1f_3))^2 F 2 = ( f 6 2 / f 2 2 ) ( 1/ ( f 1 f 3 ) ) 2 , and the odd part of ( X ( q 2 ) + q Y ( q 2 ) ) 2 (X(q^2)+qY(q^2))^2 ( X ( q 2 ) + q Y ( q 2 ) ) 2 is 2 q X ( q 2 ) Y ( q 2 ) 2qX(q^2)Y(q^2) 2 q X ( q 2 ) Y ( q 2 ) ; the prefactor is a series in q 2 q^2 q 2 . Replace q 2 q^2 q 2 by q q q . Likewise G = ( f 4 f 6 / f 12 ) / ( f 1 f 3 ) G=(f_4f_6/f_{12})/(f_1f_3) G = ( f 4 f 6 / f 12 ) / ( f 1 f 3 ) has odd part q Y ( q 2 ) f 4 f 6 / f 12 qY(q^2)f_4f_6/f_{12} q Y ( q 2 ) f 4 f 6 / f 12 , and G 2 G^2 G 2 has odd part 2 q X ( q 2 ) Y ( q 2 ) ( f 4 f 6 / f 12 ) 2 2qX(q^2)Y(q^2)(f_4f_6/f_{12})^2 2 q X ( q 2 ) Y ( q 2 ) ( f 4 f 6 / f 12 ) 2 .
Theorem 3.12. Let k ≥ 1 k\geq1 k ≥ 1 . Then the following congruences hold for all n ≥ 0 n\geq0 n ≥ 0 :
If k ≡ 3 ( m o d 4 ) k\equiv3\pmod4 k ≡ 3 ( mod 4 ) , then A k ( 4 n + 3 ) ≡ 0 ( m o d 2 ) A_k(4n+3)\equiv0\pmod2 A k ( 4 n + 3 ) ≡ 0 ( mod 2 ) .
If k ≡ 2 ( m o d 4 ) k\equiv2\pmod4 k ≡ 2 ( mod 4 ) , then A k ( 4 n + 3 ) ≡ 0 ( m o d 4 ) A_k(4n+3)\equiv0\pmod4 A k ( 4 n + 3 ) ≡ 0 ( mod 4 ) .
If k ≡ 2 ( m o d 8 ) k\equiv2\pmod8 k ≡ 2 ( mod 8 ) , then A k ( 4 n + 3 ) ≡ 0 ( m o d 8 ) A_k(4n+3)\equiv0\pmod8 A k ( 4 n + 3 ) ≡ 0 ( mod 8 ) .
Proof. (a) Working modulo 2, we have f 2 ≡ f 1 2 f_{2} \equiv f_{1}^{2} f 2 ≡ f 1 2 and f 6 ≡ f 3 2 f_{6} \equiv f_{3}^{2} f 6 ≡ f 3 2 (mod 2), which gives:
F ≡ f 3 f 1 3 ( m o d 2 ) . F \equiv\frac{f_{3}}{f_{1}^{3}} \pmod{2}. F ≡ f 1 3 f 3 ( mod 2 ) .
Using Hirschhorn’s 2-dissection [3 ] ,
f 3 f 1 3 = f 4 6 f 6 3 f 2 9 f 12 3 + 3 q f 4 2 f 6 f 12 2 f 2 7 , \frac{f_{3}}{f_{1}^{3}}=\frac{f_{4}^{6}f_{6}^{3}}{f_{2}^{9}f_{12}^{3}}+3q\frac{f_{4}^{2}f_{6}f_{12}^{2}}{f_{2}^{7}}, f 1 3 f 3 = f 2 9 f 12 3 f 4 6 f 6 3 + 3 q f 2 7 f 4 2 f 6 f 12 2 ,
we can express F ≡ α ( q 2 ) + q β ( q 2 ) F \equiv\alpha(q^{2})+q\beta(q^{2}) F ≡ α ( q 2 ) + q β ( q 2 ) (mod 2), where β ( q ) ≡ f 3 5 / f 1 3 \beta(q) \equiv f_{3}^{5}/f_{1}^{3} β ( q ) ≡ f 3 5 / f 1 3 (mod 2). Write k = 2 j + 1 k=2j+1 k = 2 j + 1 . Since F 2 ≡ F ( q 2 ) F^{2}\equiv F(q^{2}) F 2 ≡ F ( q 2 ) (mod 2), raising F F F to the power k k k yields:
F k ≡ F ( q 2 ) j ( α ( q 2 ) + q β ( q 2 ) ) ( m o d 2 ) , F^{k}\equiv F(q^{2})^{j}\left(\alpha(q^{2})+q\beta(q^{2})\right) \pmod{2}, F k ≡ F ( q 2 ) j ( α ( q 2 ) + q β ( q 2 ) ) ( mod 2 ) ,
which implies that A k ( 2 m + 1 ) ≡ [ x m ] β ( x ) F ( x ) j A_{k}(2m+1)\equiv[x^{m}]\beta(x)F(x)^{j} A k ( 2 m + 1 ) ≡ [ x m ] β ( x ) F ( x ) j (mod 2). When k ≡ 3 k\equiv3 k ≡ 3 (mod 4), j j j is odd. In this case, β F ≡ f 3 6 / f 1 6 = ( f 3 3 / f 1 3 ) 2 \beta F\equiv f_{3}^{6}/f_{1}^{6}=(f_{3}^{3}/f_{1}^{3})^{2} β F ≡ f 3 6 / f 1 6 = ( f 3 3 / f 1 3 ) 2 (mod 2) is a square and thus a series in x 2 x^{2} x 2 (mod 2). Likewise, F j − 1 ≡ F ( x 2 ) ( j − 1 ) / 2 F^{j-1}\equiv F(x^{2})^{(j-1)/2} F j − 1 ≡ F ( x 2 ) ( j − 1 ) /2 (mod 2) is a series in x 2 x^{2} x 2 . Their product contains no odd powers of x x x , so A k ( 4 n + 3 ) A_{k}(4n+3) A k ( 4 n + 3 ) , corresponding to m = 2 n + 1 m=2n+1 m = 2 n + 1 , is always even.
(b) and (c) By Lemma 3.11, F 2 = E ( q 2 ) + 2 q W ( q 2 ) F^{2}=E(q^{2})+2qW(q^{2}) F 2 = E ( q 2 ) + 2 q W ( q 2 ) , where W = f 2 4 f 4 6 / ( f 1 8 f 3 4 ) W=f_{2}^{4}f_{4}^{6}/(f_{1}^{8}f_{3}^{4}) W = f 2 4 f 4 6 / ( f 1 8 f 3 4 ) . Since f 1 2 ≡ f 2 f_{1}^{2}\equiv f_{2} f 1 2 ≡ f 2 (mod 2), we have f 1 4 ≡ f 2 2 f_{1}^{4}\equiv f_{2}^{2} f 1 4 ≡ f 2 2 (mod 4) and f 3 4 ≡ f 6 2 f_{3}^{4}\equiv f_{6}^{2} f 3 4 ≡ f 6 2 (mod 4). This gives:
W ≡ f 6 2 ( m o d 4 ) , W\equiv f_{6}^{2}\pmod{4}, W ≡ f 6 2 ( mod 4 ) ,
which is a series in x 2 x^{2} x 2 (mod 4). Therefore, all odd-indexed coefficients of W W W are divisible by 4.
Now, let k = 2 m k=2m k = 2 m with m m m odd. Expanding F k = ( E + 2 q W ) m F^{k}=(E+2qW)^{m} F k = ( E + 2 q W ) m , the odd part of F k F^{k} F k modulo 8 is:
2 m q W ( q 2 ) E ( q 2 ) m − 1 ( m o d 8 ) . 2m qW(q^{2})E(q^{2})^{m-1}\pmod{8}. 2 m q W ( q 2 ) E ( q 2 ) m − 1 ( mod 8 ) .
Thus, A k ( 4 n + 3 ) ≡ 2 m c n A_{k}(4n+3)\equiv2m c_{n} A k ( 4 n + 3 ) ≡ 2 m c n (mod 8), where c n = [ x 2 n + 1 ] ( W E m − 1 ) c_{n}=[x^{2n+1}](WE^{m-1}) c n = [ x 2 n + 1 ] ( W E m − 1 ) . Because m − 1 m-1 m − 1 is even, E m − 1 = ( E 2 ) ( m − 1 ) / 2 ≡ E ( x 2 ) ( m − 1 ) / 2 E^{m-1}=(E^{2})^{(m-1)/2}\equiv E(x^{2})^{(m-1)/2} E m − 1 = ( E 2 ) ( m − 1 ) /2 ≡ E ( x 2 ) ( m − 1 ) /2 (mod 2) is a series in x 2 x^{2} x 2 (mod 2). Consequently, the odd coefficients of W E m − 1 WE^{m-1} W E m − 1 originate from the odd coefficients of W W W , which are divisible by 4. Hence c n c_{n} c n is even, proving that A k ( 4 n + 3 ) ≡ 0 A_{k}(4n+3)\equiv0 A k ( 4 n + 3 ) ≡ 0 (mod 4), which proves (b).
For (c), suppose further that m ≡ 1 m\equiv1 m ≡ 1 (mod 4), so m − 1 = 4 s m-1=4s m − 1 = 4 s . Since E 2 = E ( x 2 ) + 2 Q E^{2}=E(x^{2})+2Q E 2 = E ( x 2 ) + 2 Q for some integer series Q Q Q , we have E 4 ≡ E ( x 2 ) 2 E^{4}\equiv E(x^{2})^{2} E 4 ≡ E ( x 2 ) 2 (mod 4), which is a series in x 2 x^{2} x 2 (mod 4). It follows that c n ≡ [ x 2 n + 1 ] W ⋅ ( series in x 2 ) ≡ 0 c_{n}\equiv[x^{2n+1}]W\cdot(\text{series in }x^{2})\equiv0 c n ≡ [ x 2 n + 1 ] W ⋅ ( series in x 2 ) ≡ 0 (mod 4), giving 2 m c n ≡ 0 2m c_{n}\equiv0 2 m c n ≡ 0 (mod 8). □
Proposition 3.13. We have A k ( 3 ) = k ( k + 1 ) ( k + 14 ) / 6 A_{k}(3)=k(k+1)(k+14)/6 A k ( 3 ) = k ( k + 1 ) ( k + 14 ) /6 . If k ≡ 6 k\equiv6 k ≡ 6 (mod 8), then 2 2 2^{2} 2 2 is the exact power of 2 dividing all A k ( 4 n + 3 ) A_{k}(4n+3) A k ( 4 n + 3 ) .
Proof. Using the expansion F = 1 + q + 3 q 2 + 5 q 3 + ⋯ F=1+q+3q^{2}+5q^{3}+\cdots F = 1 + q + 3 q 2 + 5 q 3 + ⋯ , the coefficient of q 3 q^{3} q 3 in F k F^{k} F k is given by A k ( 3 ) = 5 k + 3 k ( k − 1 ) + ( k 3 ) = k ( k + 1 ) ( k + 14 ) / 6 A_{k}(3)=5k+3k(k-1)+\binom{k}{3}=k(k+1)(k+14)/6 A k ( 3 ) = 5 k + 3 k ( k − 1 ) + ( 3 k ) = k ( k + 1 ) ( k + 14 ) /6 . For k ≡ 6 k\equiv6 k ≡ 6 (mod 8), the 2-adic valuations are v 2 ( k ) = 1 v_{2}(k)=1 v 2 ( k ) = 1 , v 2 ( k + 1 ) = 0 v_{2}(k+1)=0 v 2 ( k + 1 ) = 0 , and v 2 ( k + 14 ) = 2 v_{2}(k+14)=2 v 2 ( k + 14 ) = 2 , which gives v 2 ( A k ( 3 ) ) = 1 + 0 + 2 − 1 = 2 v_{2}(A_{k}(3))=1+0+2-1=2 v 2 ( A k ( 3 )) = 1 + 0 + 2 − 1 = 2 . Combined with Theorem 3.12 (b)—which guarantees divisibility by 4—this proves that 4 is the exact power of 2 dividing A k ( 4 n + 3 ) A_{k}(4n+3) A k ( 4 n + 3 ) . □
Theorem 3.14. For all n ≥ 0 n\ge0 n ≥ 0 , we have A 2 ( 24 n + 23 ) ≡ 0 A_{2}(24n+23)\equiv0 A 2 ( 24 n + 23 ) ≡ 0 (mod 64) and A 2 ( 24 n + 11 ) ≡ 0 A_{2}(24n+11)\equiv0 A 2 ( 24 n + 11 ) ≡ 0 (mod 32).
Proof. We begin by establishing an exact formula. By Lemma 3.11, the generating function satisfies ∑ A 2 ( 2 m + 1 ) x m = 2 f 2 4 f 6 4 / ( f 1 8 f 3 4 ) \sum A_{2}(2m+1)x^{m}=2f_{2}^{4}f_{6}^{4}/(f_{1}^{8}f_{3}^{4}) ∑ A 2 ( 2 m + 1 ) x m = 2 f 2 4 f 6 4 / ( f 1 8 f 3 4 ) . Defining two series in q 2 q^{2} q 2 as
a = f 1 4 f 2 14 f 8 4 and b = f 2 4 f 8 4 f 1 10 , a=\frac{f_{1}^{4}}{f_{2}^{14}f_{8}^{4}}
\qquad\text{and}\qquad
b=\frac{f_{2}^{4}f_{8}^{4}}{f_{1}^{10}}, a = f 2 14 f 8 4 f 1 4 and b = f 1 10 f 2 4 f 8 4 ,
we can express 1 / f 1 4 = a + 4 q b 1/f_{1}^{4}=a+4qb 1/ f 1 4 = a + 4 q b and 1 / f 3 4 = a ( q 3 ) + 4 q 3 b ( q 3 ) 1/f_{3}^{4}=a(q^{3})+4q^{3}b(q^{3}) 1/ f 3 4 = a ( q 3 ) + 4 q 3 b ( q 3 ) . Substituting these into the product yields
1 f 1 8 f 3 4 = ( a 2 + 8 q a b + 16 q 2 b 2 ) ( a ( q 3 ) + 4 q 3 b ( q 3 ) ) . \frac{1}{f_{1}^{8}f_{3}^{4}}=(a^{2}+8qab+16q^{2}b^{2})(a(q^{3})+4q^{3}b(q^{3})). f 1 8 f 3 4 1 = ( a 2 + 8 q ab + 16 q 2 b 2 ) ( a ( q 3 ) + 4 q 3 b ( q 3 )) .
Extracting the odd part of this expression, multiplying by the q 2 q^2 q 2 -series 2 f 2 4 f 6 4 2f_2^4f_6^4 2 f 2 4 f 6 4 , and writing y = x 2 y=x^2 y = x 2 along with a ~ = f 2 14 / ( f 1 14 f 4 4 ) \tilde{a}=f_2^{14}/(f_1^{14}f_4^4) a ~ = f 2 14 / ( f 1 14 f 4 4 ) , b ~ = f 2 2 f 4 4 / f 1 10 \tilde{b}=f_2^2f_4^4/f_1^{10} b ~ = f 2 2 f 4 4 / f 1 10 (with a ~ 3 = a ~ ( y 3 ) \tilde{a}_3=\tilde{a}(y^3) a ~ 3 = a ~ ( y 3 ) and b ~ 3 = b ~ ( y 3 ) \tilde{b}_3=\tilde{b}(y^3) b ~ 3 = b ~ ( y 3 ) ), we obtain:
∑ n ≥ 0 A 2 ( 4 n + 3 ) y n = 16 U 1 + 8 U 2 + 128 U 3 , \sum_{n\geq0} A_2(4n+3)y^n=16U_1+8U_2+128U_3, n ≥ 0 ∑ A 2 ( 4 n + 3 ) y n = 16 U 1 + 8 U 2 + 128 U 3 ,
where
U 1 = f 1 4 f 3 4 a ~ b ~ 3 = f 2 16 f 1 14 f 1 20 f 3 10 f 4 4 , U 2 = y f 1 4 f 3 4 b ~ 3 a ~ 3 2 = y f 2 28 f 2 2 f 4 12 f 1 24 f 3 6 f 4 8 , U_1=f_1^4f_3^4\tilde{a}\tilde{b}_3=\frac{f_2^{16}f_1^{14}}{f_1^{20}f_3^{10}f_4^4},\qquad U_2=yf_1^4f_3^4\tilde{b}_3\tilde{a}_3^2=\frac{yf_2^{28}f_2^2f_4^{12}}{f_1^{24}f_3^6f_4^8}, U 1 = f 1 4 f 3 4 a ~ b ~ 3 = f 1 20 f 3 10 f 4 4 f 2 16 f 1 14 , U 2 = y f 1 4 f 3 4 b ~ 3 a ~ 3 2 = f 1 24 f 3 6 f 4 8 y f 2 28 f 2 2 f 4 12 ,
and U 3 ∈ Z [ [ y ] ] U_3\in\mathbb{Z}[[y]] U 3 ∈ Z [[ y ]] is an auxiliary series whose explicit form is not needed. To reduce these terms, we use the congruences f 1 4 ≡ f 2 2 ( m o d 4 ) f_1^4\equiv f_2^2\pmod{4} f 1 4 ≡ f 2 2 ( mod 4 ) , f 1 8 ≡ f 2 4 ( m o d 8 ) f_1^8\equiv f_2^4\pmod{8} f 1 8 ≡ f 2 4 ( mod 8 ) , and their analogs for other indices to find:
U 1 ≡ f 2 6 f 6 2 f 3 2 ( m o d 4 ) and U 2 ≡ y f 3 2 f 12 4 f 6 2 ( m o d 8 ) . U_1\equiv\frac{f_2^6f_6^2}{f_3^2}\pmod{4}\qquad\text{and}\qquad U_2\equiv y\frac{f_3^2f_{12}^4}{f_6^2}\pmod{8}. U 1 ≡ f 3 2 f 2 6 f 6 2 ( mod 4 ) and U 2 ≡ y f 6 2 f 3 2 f 12 4 ( mod 8 ) .
Because f 3 f_3 f 3 , f 6 f_6 f 6 , and f 12 f_{12} f 12 depend only on powers of y 3 y^3 y 3 , the series U 2 ( m o d 8 ) U_2\pmod{8} U 2 ( mod 8 ) contains only exponents congruent to 1 ( m o d 3 ) 1\pmod{3} 1 ( mod 3 ) . Since 24 n + 23 = 4 ( 6 n + 5 ) + 3 24n+23=4(6n+5)+3 24 n + 23 = 4 ( 6 n + 5 ) + 3 and 24 n + 11 = 4 ( 6 n + 2 ) + 3 24n+11=4(6n+2)+3 24 n + 11 = 4 ( 6 n + 2 ) + 3 , we must examine coefficients of y N y^N y N where N ≡ 5 N\equiv5 N ≡ 5 and N ≡ 2 ( m o d 6 ) N\equiv2\pmod{6} N ≡ 2 ( mod 6 ) , respectively; both conditions imply N ≡ 2 ( m o d 3 ) N\equiv2\pmod{3} N ≡ 2 ( mod 3 ) . For such values of N N N , the term 8 U 2 8U_2 8 U 2 vanishes modulo 64 64 64 , leaving:
A 2 ( 4 N + 3 ) ≡ 16 [ y N ] U 1 ( m o d 64 ) . A_2(4N+3)\equiv16[y^N]U_1\pmod{64}. A 2 ( 4 N + 3 ) ≡ 16 [ y N ] U 1 ( mod 64 ) .
Thus, it suffices to prove that [ y N ] U 1 [y^N]U_1 [ y N ] U 1 is even when N ≡ 2 ( m o d 3 ) N\equiv2\pmod{3} N ≡ 2 ( mod 3 ) and divisible by 4 4 4 when N ≡ 5 ( m o d 6 ) N\equiv5\pmod{6} N ≡ 5 ( mod 6 ) (i.e., N ≡ 2 ( m o d 3 ) N\equiv2\pmod{3} N ≡ 2 ( mod 3 ) overall).
Finally, we analyze the coefficients. We write f 2 6 f 6 2 / f 3 2 = ( ∑ M c M y M ) ( ∑ j h j y 3 j ) f_2^6f_6^2/f_3^2=(\sum_M c_My^M)(\sum_jh_jy^{3j}) f 2 6 f 6 2 / f 3 2 = ( ∑ M c M y M ) ( ∑ j h j y 3 j ) , where c M = [ y M ] f 2 6 c_M=[y^M]f_2^6 c M = [ y M ] f 2 6 and ∑ j h j w j = f 2 2 / f 1 2 \sum_jh_jw^j=f_2^2/f_1^2 ∑ j h j w j = f 2 2 / f 1 2 with w = y 3 w=y^3 w = y 3 . Clearly, c M = 0 c_M=0 c M = 0 for odd M M M . For even M = 2 i M=2i M = 2 i with 2 i ≡ 2 ( m o d 3 ) 2i\equiv2\pmod{3} 2 i ≡ 2 ( mod 3 ) (meaning i ≡ 1 ( m o d 3 ) i\equiv1\pmod{3} i ≡ 1 ( mod 3 ) ), the coefficient c M c_M c M is even. Indeed, by Jacobi’s identity, modulo 2 2 2 we have:
f 1 6 ≡ f 2 3 ≡ ( ∑ m ≥ 0 z m ( m + 1 ) / 2 ) 2 ≡ ∑ m ≥ 0 z m ( m + 1 ) ( m o d 2 ) , f_1^6\equiv f_2^3\equiv\left(\sum_{m\geq0}z^{m(m+1)/2}\right)^2\equiv\sum_{m\geq0}z^{m(m+1)}\pmod{2}, f 1 6 ≡ f 2 3 ≡ ( m ≥ 0 ∑ z m ( m + 1 ) /2 ) 2 ≡ m ≥ 0 ∑ z m ( m + 1 ) ( mod 2 ) ,
which only generates exponents congruent to 0 0 0 and 2 ( m o d 3 ) 2\pmod{3} 2 ( mod 3 ) , never 1 ( m o d 3 ) 1\pmod{3} 1 ( mod 3 ) . Consequently, [ y N ] U 1 [y^N]U_1 [ y N ] U 1 is even for N ≡ 2 ( m o d 3 ) N\equiv2\pmod{3} N ≡ 2 ( mod 3 ) , which immediately implies A 2 ( 24 n + 11 ) ≡ 0 ( m o d 32 ) A_2(24n+11)\equiv0\pmod{32} A 2 ( 24 n + 11 ) ≡ 0 ( mod 32 ) .
For N ≡ 5 ( m o d 6 ) N\equiv5\pmod{6} N ≡ 5 ( mod 6 ) , we write N = M + 3 j N=M+3j N = M + 3 j with M M M even, which forces j j j to be odd. Since f 2 2 / f 1 2 ≡ f 1 2 ≡ f 2 ( m o d 2 ) f_2^2/f_1^2\equiv f_1^2\equiv f_2\pmod{2} f 2 2 / f 1 2 ≡ f 1 2 ≡ f 2 ( mod 2 ) is a series in w 2 w^2 w 2 , the coefficient h j h_j h j is even for every odd j j j . Thus, each product c M h j c_Mh_j c M h j is divisible by 4 4 4 , ensuring that [ y N ] U 1 ≡ 0 ( m o d 4 ) [y^N]U_1\equiv0\pmod{4} [ y N ] U 1 ≡ 0 ( mod 4 ) and completing the proof for A 2 ( 24 n + 23 ) ≡ 0 ( m o d 64 ) A_2(24n+23)\equiv0\pmod{64} A 2 ( 24 n + 23 ) ≡ 0 ( mod 64 ) .
Theorem 3.15. For all n ≥ 0 n\geq0 n ≥ 0 , we have A 2 ( 18 n + 15 ) ≡ 0 ( m o d 32 ) A_2(18n+15)\equiv0\pmod{32} A 2 ( 18 n + 15 ) ≡ 0 ( mod 32 ) .
Proof. By Lemma 3.11, the odd-indexed coefficients satisfy A 2 ( 2 m + 1 ) = 2 [ x m ] E ( x ) A_2(2m+1)=2[x^m]E(x) A 2 ( 2 m + 1 ) = 2 [ x m ] E ( x ) , where E = f 2 4 f 6 / ( f 1 8 f 3 4 ) E=f_2^4f_6/(f_1^8f_3^4) E = f 2 4 f 6 / ( f 1 8 f 3 4 ) . Writing 18 n + 15 = 2 ( 9 n + 7 ) + 1 18n+15=2(9n+7)+1 18 n + 15 = 2 ( 9 n + 7 ) + 1 , the problem reduces to proving that [ x m ] E ≡ 0 ( m o d 16 ) [x^m]E\equiv0\pmod{16} [ x m ] E ≡ 0 ( mod 16 ) whenever m ≡ 7 ( m o d 9 ) m\equiv7\pmod{9} m ≡ 7 ( mod 9 ) .
Using Lemma 2.2 with prime power 2 4 2^4 2 4 , we have f 1 16 ≡ f 2 8 ( m o d 16 ) f_1^{16}\equiv f_2^8\pmod{16} f 1 16 ≡ f 2 8 ( mod 16 ) and f 3 16 ≡ f 6 8 ( m o d 16 ) f_3^{16}\equiv f_6^8\pmod{16} f 3 16 ≡ f 6 8 ( mod 16 ) . This allows us to simplify E E E modulo 16 16 16 :
E ≡ f 1 8 f 3 12 f 2 4 f 6 4 = φ ( − x ) 4 ⋅ φ ( − x 3 ) 4 f 3 4 ( m o d 16 ) , E\equiv\frac{f_1^8f_3^{12}}{f_2^4f_6^4}=\varphi(-x)^4\cdot\varphi(-x^3)^4f_3^4\pmod{16}, E ≡ f 2 4 f 6 4 f 1 8 f 3 12 = φ ( − x ) 4 ⋅ φ ( − x 3 ) 4 f 3 4 ( mod 16 ) ,
where φ ( − x ) = f 1 2 / f 2 \varphi(-x)=f_1^2/f_2 φ ( − x ) = f 1 2 / f 2 . Note that the second factor φ ( − x 3 ) 4 f 3 4 \varphi(-x^3)^4f_3^4 φ ( − x 3 ) 4 f 3 4 is a series purely in x 3 x^3 x 3 . Next, we extract the terms of φ ( − x ) 4 \varphi(-x)^4 φ ( − x ) 4 whose exponents are congruent to 1 ( m o d 3 ) 1\pmod{3} 1 ( mod 3 ) . Using Hirschhorn’s 3-dissection [3 ] , we write φ ( − x ) = X 0 + X 1 \varphi(-x)=X_0+X_1 φ ( − x ) = X 0 + X 1 , where X 0 = φ ( − x 9 ) = f 9 2 / f 18 X_0=\varphi(-x^9)=f_9^2/f_{18} X 0 = φ ( − x 9 ) = f 9 2 / f 18 and X 1 = − 2 x g X_1=-2xg X 1 = − 2 xg with g = f 3 f 18 2 / ( f 6 f 9 ) ∈ Z [ [ x 3 ] ] g=f_3f_{18}^2/(f_6f_9)\in\mathbb{Z}[[x^3]] g = f 3 f 18 2 / ( f 6 f 9 ) ∈ Z [[ x 3 ]] . The part of ( X 0 + X 1 ) 4 (X_0+X_1)^4 ( X 0 + X 1 ) 4 with exponents congruent to 1 (mod 3) is
4 X 0 3 X 1 + X 1 4 = − 8 x g ( X 0 3 − 2 x 3 g 3 ) . 4X_0^3X_1+X_1^4=-8xg\left(X_0^3-2x^3g^3\right). 4 X 0 3 X 1 + X 1 4 = − 8 xg ( X 0 3 − 2 x 3 g 3 ) .
Setting z = x 3 z=x^3 z = x 3 and defining h ( z ) = φ ( − z ) 4 f 1 ( z ) 4 = f 1 12 / f 2 4 h(z)=\varphi(-z)^4f_1(z)^4=f_1^{12}/f_2^4 h ( z ) = φ ( − z ) 4 f 1 ( z ) 4 = f 1 12 / f 2 4 in the variable z z z , we obtain
∑ N ≥ 0 ( [ x 3 N + 1 ] E ) z N ≡ − 8 B ( z ) ( m o d 16 ) , where B = g ( X 0 3 − 2 z g 3 ) h , \sum_{N\geq0}\left([x^{3N+1}]E\right)z^N\equiv-8B(z)\pmod{16},
\qquad\text{where }B=g\left(X_0^3-2zg^3\right)h, N ≥ 0 ∑ ( [ x 3 N + 1 ] E ) z N ≡ − 8 B ( z ) ( mod 16 ) , where B = g ( X 0 3 − 2 z g 3 ) h ,
with g = f 1 f 2 2 / ( f 2 f 3 ) g=f_1f_2^2/(f_2f_3) g = f 1 f 2 2 / ( f 2 f 3 ) and X 0 = f 3 2 / f 6 X_0=f_3^2/f_6 X 0 = f 3 2 / f 6 expressed in z z z . The index condition m = 3 N + 1 ≡ 7 m=3N+1\equiv7 m = 3 N + 1 ≡ 7 (mod 9) translates to N ≡ 2 N\equiv2 N ≡ 2 (mod 3). Thus, we only need to show that [ z N ] B [z^N]B [ z N ] B is even whenever N ≡ 2 N\equiv2 N ≡ 2 (mod 3).
Working modulo 2, we have f 2 ≡ f 1 2 f_2\equiv f_1^2 f 2 ≡ f 1 2 and f 6 ≡ f 3 2 f_6\equiv f_3^2 f 6 ≡ f 3 2 , which simplifies B B B to
B ≡ g X 0 3 h = f 1 13 f 3 5 f 2 5 f 6 ≡ f 1 3 f 3 3 ( m o d 2 ) . B\equiv gX_0^3h=\frac{f_1^{13}f_3^5}{f_2^5f_6}\equiv f_1^3f_3^3\pmod{2}. B ≡ g X 0 3 h = f 2 5 f 6 f 1 13 f 3 5 ≡ f 1 3 f 3 3 ( mod 2 ) .
By Jacobi’s identity, f 1 3 = ∑ n ≥ 0 ( − 1 ) n ( 2 n + 1 ) z n ( n + 1 ) / 2 ≡ ∑ n ≥ 0 z n ( n + 1 ) / 2 ( m o d 2 ) f_1^3=\sum_{n\geq0}(-1)^n(2n+1)z^{n(n+1)/2}\equiv\sum_{n\geq0}z^{n(n+1)/2}\pmod{2} f 1 3 = ∑ n ≥ 0 ( − 1 ) n ( 2 n + 1 ) z n ( n + 1 ) /2 ≡ ∑ n ≥ 0 z n ( n + 1 ) /2 ( mod 2 ) , while f 3 3 f_3^3 f 3 3 is a series in z 3 z^3 z 3 . Because triangular numbers n ( n + 1 ) / 2 n(n+1)/2 n ( n + 1 ) /2 are never congruent to 2 (mod 3), no terms in f 1 3 f 3 3 f_1^3f_3^3 f 1 3 f 3 3 have exponents congruent to 2 (mod 3). Consequently, [ z N ] B [z^N]B [ z N ] B is even for N ≡ 2 N\equiv2 N ≡ 2 (mod 3), completing the proof. □
Remark 3.16. Naika and Shivashankar [10 ] established that A 2 ( 24 n + 23 ) ≡ 0 ( m o d 32 ) A_2(24n+23)\equiv0\pmod{32} A 2 ( 24 n + 23 ) ≡ 0 ( mod 32 ) and A 2 ( 18 n + 15 ) ≡ 0 ( m o d 48 ) A_2(18n+15)\equiv0\pmod{48} A 2 ( 18 n + 15 ) ≡ 0 ( mod 48 ) , while Hao and Shen [9 ] showed that A 2 ( 18 n + 15 ) ≡ 0 ( m o d 9 ) A_2(18n+15)\equiv0\pmod{9} A 2 ( 18 n + 15 ) ≡ 0 ( mod 9 ) . Theorems 3.14 and 3.15 strengthen the 2-adic divisibility bounds to 64 and 32, respectively. Combining Theorem 3.15 with the result of [9 ] yields the overall congruence
A 2 ( 18 n + 15 ) ≡ 0 ( m o d 288 ) . A_2(18n+15)\equiv0\pmod{288}. A 2 ( 18 n + 15 ) ≡ 0 ( mod 288 ) .
We note that while A 2 ( 24 n + 11 ) ≡ 0 ( m o d 32 ) A_2(24n+11)\equiv0\pmod{32} A 2 ( 24 n + 11 ) ≡ 0 ( mod 32 ) was not highlighted in the abstract of [10 ] , it complements the known congruences for 24 n + 15 24n+15 24 n + 15 and 24 n + 23 24n+23 24 n + 23 .
Theorem 3.17. (a) We have the exact identity
∑ j ≥ 0 A 2 ( 3 j + 1 ) x j = 2 f 3 2 f 6 2 f 1 6 f 2 2 + 27 x f 3 6 f 6 6 f 1 10 f 2 6 . \sum_{j\geq0}A_2(3j+1)x^j=2\frac{f_3^2f_6^2}{f_1^6f_2^2}+27x\frac{f_3^6f_6^6}{f_1^{10}f_2^6}. j ≥ 0 ∑ A 2 ( 3 j + 1 ) x j = 2 f 1 6 f 2 2 f 3 2 f 6 2 + 27 x f 1 10 f 2 6 f 3 6 f 6 6 . (b) ∑ j ≥ 0 A 2 ( 12 j + 4 ) q j ≡ 12 f 1 8 ( m o d 27 ) \sum_{j\geq0}A_2(12j+4)q^j\equiv12f_1^8\pmod{27} ∑ j ≥ 0 A 2 ( 12 j + 4 ) q j ≡ 12 f 1 8 ( mod 27 ) , which implies ≡ 3 f 1 8 ≡ 3 f 3 4 / f 1 4 ( m o d 9 ) \equiv3f_1^8\equiv3f_3^4/f_1^4\pmod{9} ≡ 3 f 1 8 ≡ 3 f 3 4 / f 1 4 ( mod 9 ) .
(c) For all n ≥ 0 n\geq0 n ≥ 0 , A 2 ( 12 n + 10 ) ≡ 0 ( m o d 27 ) A_2(12n+10)\equiv0\pmod{27} A 2 ( 12 n + 10 ) ≡ 0 ( mod 27 ) .
(d) ∑ n ≥ 0 A 2 ( 12 n + 10 ) q n ≡ 27 f 1 4 f 3 3 f 2 2 ( m o d 81 ) \sum_{n\geq0}A_2(12n+10)q^n\equiv27\frac{f_1^4f_3^3}{f_2^2}\pmod{81} ∑ n ≥ 0 A 2 ( 12 n + 10 ) q n ≡ 27 f 2 2 f 1 4 f 3 3 ( mod 81 ) .
Proof. (a) By Proposition 2.4 , we have F 2 = Θ 2 P ( u ) 2 F^2=\Theta^2P(u)^2 F 2 = Θ 2 P ( u ) 2 . The polynomial P ( u ) 2 P(u)^2 P ( u ) 2 has coefficients 2, 13, − 16 -16 − 16 at u u u , u 4 u^4 u 4 , u 7 u^7 u 7 (the terms with exponents congruent to 1 (mod 3)). Since Θ , ρ ∈ Z [ [ q 3 ] ] \Theta,\rho\in\mathbb{Z}[[q^3]] Θ , ρ ∈ Z [[ q 3 ]] and u 3 = q 3 ρ 3 u^3=q^3\rho^3 u 3 = q 3 ρ 3 , picking out the terms with exponents ≡ 1 \equiv1 ≡ 1 (mod 3) and setting x = q 3 x=q^3 x = q 3 gives
∑ j ≥ 0 A 2 ( 3 j + 1 ) x j = f 3 15 f 1 13 f 2 3 f 6 3 ( 2 + 13 w − 16 w 2 ) , \sum_{j\geq0}A_2(3j+1)x^j=\frac{f_3^{15}}{f_1^{13}f_2^3f_6^3}\left(2+13w-16w^2\right), j ≥ 0 ∑ A 2 ( 3 j + 1 ) x j = f 1 13 f 2 3 f 6 3 f 3 15 ( 2 + 13 w − 16 w 2 ) ,
where Θ 2 ρ \Theta^2\rho Θ 2 ρ becomes f 3 15 / ( f 1 13 f 2 3 f 6 3 ) f_3^{15}/(f_1^{13}f_2^3f_6^3) f 3 15 / ( f 1 13 f 2 3 f 6 3 ) by Lemma 2.6 , and u 3 u^3 u 3 becomes w w w . Rewriting the polynomial as 2 + 13 w − 16 w 2 = 2 ( 1 + w ) ( 1 − 8 w ) + 27 w 2+13w-16w^2=2(1+w)(1-8w)+27w 2 + 13 w − 16 w 2 = 2 ( 1 + w ) ( 1 − 8 w ) + 27 w and substituting ( 1 + w ) ( 1 − 8 w ) = f 1 7 f 2 f 6 5 / f 3 5 (1+w)(1-8w)=f_1^7f_2f_6^5/f_3^5 ( 1 + w ) ( 1 − 8 w ) = f 1 7 f 2 f 6 5 / f 3 5 from Lemma 2.6 directly yields identity (a). (b) and (c) Let S 1 = ∑ A 2 ( 3 j + 1 ) x j S_1=\sum A_2(3j+1)x^j S 1 = ∑ A 2 ( 3 j + 1 ) x j . From part (a), we have S 1 ≡ 2 X 2 f 6 2 / f 2 2 ( m o d 27 ) S_1\equiv2X^2f_6^2/f_2^2\pmod{27} S 1 ≡ 2 X 2 f 6 2 / f 2 2 ( mod 27 ) , where X = f 3 / f 1 3 X=f_3/f_1^3 X = f 3 / f 1 3 . Hirschhorn’s 2-dissection [3 ] gives X = a + x b X=a+xb X = a + x b , where a = f 4 6 f 3 6 / ( f 2 9 f 12 2 ) a=f_4^6f_3^6/(f_2^9f_{12}^2) a = f 4 6 f 3 6 / ( f 2 9 f 12 2 ) and b = 3 f 4 2 f 6 f 12 2 / f 2 7 b=3f_4^2f_6f_{12}^2/f_2^7 b = 3 f 4 2 f 6 f 12 2 / f 2 7 . Since a a a and b b b are series in even powers of x x x , the odd part of X 2 X^2 X 2 is
2 x a b = 6 x f 4 8 f 6 4 f 2 16 . 2xab=6x\frac{f_4^8f_6^4}{f_2^{16}}. 2 x ab = 6 x f 2 16 f 4 8 f 6 4 .
The odd part of S 1 S_1 S 1 is therefore 12 x f 4 8 f 6 6 / f 2 18 ( m o d 27 ) 12x f_4^8f_6^6/f_2^{18}\pmod{27} 12 x f 4 8 f 6 6 / f 2 18 ( mod 27 ) . Matching coefficients of x 2 i + 1 x^{2i+1} x 2 i + 1 corresponds to A 2 ( 6 i + 4 ) A_2(6i+4) A 2 ( 6 i + 4 ) . Substituting y = x 2 y=x^2 y = x 2 gives
H ( y ) : = ∑ i A 2 ( 6 i + 4 ) y i ≡ 12 f 2 8 f 3 6 f 1 18 ( m o d 27 ) . H(y):=\sum_i A_2(6i+4)y^i\equiv12\frac{f_2^8f_3^6}{f_1^{18}}\pmod{27}. H ( y ) := i ∑ A 2 ( 6 i + 4 ) y i ≡ 12 f 1 18 f 2 8 f 3 6 ( mod 27 ) .
By Lemma 2.2, f 1 9 ≡ f 3 3 ( m o d 9 ) f_1^9\equiv f_3^3\pmod9 f 1 9 ≡ f 3 3 ( mod 9 ) , which implies f 1 18 ≡ f 3 6 ( m o d 9 ) f_1^{18}\equiv f_3^6\pmod9 f 1 18 ≡ f 3 6 ( mod 9 ) and f 3 6 / f 1 18 ≡ 1 ( m o d 9 ) f_3^6/f_1^{18}\equiv1\pmod9 f 3 6 / f 1 18 ≡ 1 ( mod 9 ) . Multiplying by 12 gives 12 ⋅ ( f 3 6 / f 1 18 − 1 ) ≡ 0 ( m o d 27 ) 12\cdot(f_3^6/f_1^{18}-1)\equiv0\pmod{27} 12 ⋅ ( f 3 6 / f 1 18 − 1 ) ≡ 0 ( mod 27 ) , so H ( y ) H(y) H ( y ) simplifies to
H ( y ) ≡ 12 f 2 ( y ) 8 ( m o d 27 ) . H(y)\equiv12f_2(y)^8\pmod{27}. H ( y ) ≡ 12 f 2 ( y ) 8 ( mod 27 ) .
Because the right-hand side contains only even powers of y y y , any coefficient A 2 ( 6 i + 4 ) A_2(6i+4) A 2 ( 6 i + 4 ) with i i i odd must vanish modulo 27. Setting i = 2 n + 1 i=2n+1 i = 2 n + 1 gives A 2 ( 12 n + 10 ) ≡ 0 ( m o d 27 ) A_2(12n+10)\equiv0\pmod{27} A 2 ( 12 n + 10 ) ≡ 0 ( mod 27 ) , which proves (c). Setting i = 2 j i=2j i = 2 j gives A 2 ( 12 j + 4 ) ≡ 12 [ z j ] f 1 ( z ) 8 ( m o d 27 ) A_2(12j+4)\equiv12[z^j]f_1(z)^8\pmod{27} A 2 ( 12 j + 4 ) ≡ 12 [ z j ] f 1 ( z ) 8 ( mod 27 ) , which proves (b). The congruence modulo 3 follows from f 1 8 ≡ f 3 3 / f 1 ≡ f 4 3 / f 4 1 ( m o d 3 ) f_1^8\equiv f_3^3/f_1\equiv f_4^3/f_4^1\pmod3 f 1 8 ≡ f 3 3 / f 1 ≡ f 4 3 / f 4 1 ( mod 3 ) .
(d) We now refine the calculation of H ( y ) H(y) H ( y ) modulo 27 to modulus 81. From part (a), the full series is
S 1 = 2 X 2 f 6 2 f 2 2 + 27 x f 3 6 f 6 6 f 1 10 f 2 6 . S_1=2X^2\frac{f_6^2}{f_2^2}+27x\frac{f_3^6f_6^6}{f_1^{10}f_2^6}. S 1 = 2 X 2 f 2 2 f 6 2 + 27 x f 1 10 f 2 6 f 3 6 f 6 6 .
The odd part of the first term remains 12 x f 4 8 f 6 6 / f 2 18 12x f_4^8f_6^6/f_2^{18} 12 x f 4 8 f 6 6 / f 2 18 . For the second term, evaluating its eta-quotient modulo 3 gives
f 3 6 f 6 6 f 1 10 f 2 6 ≡ f 1 8 f 2 12 ≡ f 2 12 f 3 3 f 1 ( m o d 3 ) . \frac{f_3^6f_6^6}{f_1^{10}f_2^6}\equiv f_1^8f_2^{12}\equiv f_2^{12}\frac{f_3^3}{f_1}\pmod3. f 1 10 f 2 6 f 3 6 f 6 6 ≡ f 1 8 f 2 12 ≡ f 2 12 f 1 f 3 3 ( mod 3 ) .
Using Hirschhorn’s identity
f 3 3 f 1 = f 3 4 f 6 2 f 2 2 f 12 + x f 3 12 f 4 , \frac{f_3^3}{f_1}=\frac{f_3^4f_6^2}{f_2^2f_{12}}+x\frac{f_3^{12}}{f_4}, f 1 f 3 3 = f 2 2 f 12 f 3 4 f 6 2 + x f 4 f 3 12 ,
the odd part of the second term becomes 27 x f 2 10 f 4 3 f 6 2 / f 12 27x f_2^{10}f_4^3f_6^2/f_{12} 27 x f 2 10 f 4 3 f 6 2 / f 12 .
Setting y = x 2 y=x^2 y = x 2 , the odd part of S 1 S_1 S 1 yields the congruence modulo 81:
H ( y ) ≡ 12 K ( y ) + 27 g ( y ) , H(y)\equiv12K(y)+27g(y), H ( y ) ≡ 12 K ( y ) + 27 g ( y ) ,
where
K = f 2 8 f 3 6 f 1 18 and g = f 1 10 f 2 3 f 3 2 f 6 . K=\frac{f_2^8f_3^6}{f_1^{18}}\quad\text{and}\quad g=\frac{f_1^{10}f_2^3f_3^2}{f_6}. K = f 1 18 f 2 8 f 3 6 and g = f 6 f 1 10 f 2 3 f 3 2 .
We isolate the coefficient of y 2 n + 1 y^{2n+1} y 2 n + 1 to determine A 2 ( 12 n + 10 ) A_2(12n+10) A 2 ( 12 n + 10 ) . For the term K K K , since f 1 27 ≡ f 3 9 ( m o d 27 ) f_1^{27}\equiv f_3^9\pmod{27} f 1 27 ≡ f 3 9 ( mod 27 ) , we write K ≡ f 2 8 b 3 ( m o d 27 ) K\equiv f_2^8b^3\pmod{27} K ≡ f 2 8 b 3 ( mod 27 ) with b = f 1 3 / f 3 b=f_1^3/f_3 b = f 1 3 / f 3 . Hirschhorn’s 2-dissection gives b = c − 3 y e b=c-3ye b = c − 3 y e , where
c = f 4 3 f 12 and e = f 2 2 f 12 3 f 4 f 6 2 c=\frac{f_4^3}{f_{12}}\quad\text{and}\quad e=\frac{f_2^2f_{12}^3}{f_4f_6^2} c = f 12 f 4 3 and e = f 4 f 6 2 f 2 2 f 12 3
are series in even powers of y y y . Expanding gives b 3 ≡ c 3 − 9 y c 2 e ( m o d 27 ) b^3\equiv c^3-9yc^2e\pmod{27} b 3 ≡ c 3 − 9 y c 2 e ( mod 27 ) , so the odd part of K K K is
Odd ( K ) ≡ − 9 y f 2 10 f 5 5 f 12 f 6 2 ( m o d 27 ) . \operatorname{Odd}(K)\equiv-9y\frac{f_2^{10}f_5^5f_{12}}{f_6^2}\pmod{27}. Odd ( K ) ≡ − 9 y f 6 2 f 2 10 f 5 5 f 12 ( mod 27 ) .
Multiplying by 12 yields
12 Odd ( K ) ≡ − 27 y f 2 10 f 4 5 f 12 f 6 2 ( m o d 81 ) . 12\operatorname{Odd}(K) \equiv-27y\frac{f_2^{10}f_4^5f_{12}}{f_6^2}\pmod{81}. 12 Odd ( K ) ≡ − 27 y f 6 2 f 2 10 f 4 5 f 12 ( mod 81 ) .
In terms of z = y 2 z=y^2 z = y 2 , its contribution to ∑ n A 2 ( 12 n + 10 ) z n \sum_n A_2(12n+10)z^n ∑ n A 2 ( 12 n + 10 ) z n is − 27 h -27h − 27 h , where
h = f 1 10 f 2 5 f 6 f 3 2 ≡ f 1 4 f 2 8 ( m o d 3 ) . h=\frac{f_1^{10}f_2^5f_6}{f_3^2}\equiv f_1^4f_2^8\pmod{3}. h = f 3 2 f 1 10 f 2 5 f 6 ≡ f 1 4 f 2 8 ( mod 3 ) .
For the term g g g , using f 1 2 = φ ( − y ) f 2 f_1^2=\varphi(-y)f_2 f 1 2 = φ ( − y ) f 2 and f 3 2 = φ ( − y 3 ) f 6 f_3^2=\varphi(-y^3)f_6 f 3 2 = φ ( − y 3 ) f 6 , we have
g = f 2 8 φ ( − y ) 5 φ ( − y 3 ) ≡ f 2 8 φ ( − y ) 8 ( m o d 3 ) , g=f_2^8\varphi(-y)^5\varphi(-y^3)\equiv f_2^8\varphi(-y)^8\pmod{3}, g = f 2 8 φ ( − y ) 5 φ ( − y 3 ) ≡ f 2 8 φ ( − y ) 8 ( mod 3 ) ,
since φ ( − y 3 ) ≡ φ ( − y ) 3 ( m o d 3 ) \varphi(-y^3)\equiv\varphi(-y)^3\pmod{3} φ ( − y 3 ) ≡ φ ( − y ) 3 ( mod 3 ) . From the theta identity φ ( − y ) 2 ≡ P − y Q ( m o d 3 ) \varphi(-y)^2\equiv P-yQ\pmod{3} φ ( − y ) 2 ≡ P − y Q ( mod 3 ) with P = φ ( y 2 ) 2 P=\varphi(y^2)^2 P = φ ( y 2 ) 2 and Q = ψ ( y 4 ) 2 Q=\psi(y^4)^2 Q = ψ ( y 4 ) 2 , we expand
φ ( − y ) 8 ≡ ( P − y Q ) 4 ( m o d 3 ) . \varphi(-y)^8\equiv(P-yQ)^4\pmod{3}. φ ( − y ) 8 ≡ ( P − y Q ) 4 ( mod 3 ) .
Its odd part in y y y is − y P Q ( P 2 + y 2 Q 2 ) ( m o d 3 ) -yPQ(P^2+y^2Q^2)\pmod{3} − y P Q ( P 2 + y 2 Q 2 ) ( mod 3 ) . Changing variables to z = y 2 z=y^2 z = y 2 gives P = φ ( z ) 2 P=\varphi(z)^2 P = φ ( z ) 2 , Q = ψ ( z 2 ) 2 Q=\psi(z^2)^2 Q = ψ ( z 2 ) 2 , and P Q = f 2 8 / f 1 4 PQ=f_2^8/f_1^4 P Q = f 2 8 / f 1 4 , which yields
Odd ( g ) ∣ z = − f 1 4 f 2 8 ( φ ( z ) 4 + z ψ ( z 2 ) 4 ) ( m o d 3 ) . \left.\operatorname{Odd}(g)\right|_z=-f_1^4f_2^8\left(\varphi(z)^4+z\psi(z^2)^4\right)\pmod{3}. Odd ( g ) ∣ z = − f 1 4 f 2 8 ( φ ( z ) 4 + z ψ ( z 2 ) 4 ) ( mod 3 ) .
By Jacobi’s four-square theorem and Legendre’s identity, φ ( z ) 4 + z ψ ( z 2 ) 4 ≡ 1 ( m o d 3 ) \varphi(z)^4+z\psi(z^2)^4\equiv1\pmod{3} φ ( z ) 4 + z ψ ( z 2 ) 4 ≡ 1 ( mod 3 ) because the odd-indexed terms cancel out modulo 3 3 3 .
Combining both terms gives
∑ n ≥ 0 A 2 ( 12 n + 10 ) q n ≡ 27 ( − h + Odd ( g ) ) ≡ 27 ( − f 1 4 f 2 8 − f 1 4 f 2 8 ) ≡ 27 f 1 4 f 2 8 ( m o d 81 ) . \sum_{n\geq0}A_2(12n+10)q^n\equiv27(-h+\operatorname{Odd}(g))\equiv27(-f_1^4f_2^8-f_1^4f_2^8)\equiv27f_1^4f_2^8\pmod{81}. n ≥ 0 ∑ A 2 ( 12 n + 10 ) q n ≡ 27 ( − h + Odd ( g )) ≡ 27 ( − f 1 4 f 2 8 − f 1 4 f 2 8 ) ≡ 27 f 1 4 f 2 8 ( mod 81 ) .
Finally, substituting f 2 8 ≡ f 3 6 / f 2 ( m o d 3 ) f_2^8\equiv f_3^6/f_2\pmod{3} f 2 8 ≡ f 3 6 / f 2 ( mod 3 ) completes the proof of (d). □
Remark 3.18. Hao and Shen [9 ] showed A 2 ( 12 n + 10 ) ≡ 0 ( m o d 9 ) A_2(12n+10)\equiv0\pmod{9} A 2 ( 12 n + 10 ) ≡ 0 ( mod 9 ) via ∑ A 2 ( 12 n + 10 ) q n ≡ 36 f 1 4 f 3 6 / f 2 ( m o d 9 ) \sum A_2(12n+10)q^n\equiv36f_1^4f_3^6/f_2\pmod{9} ∑ A 2 ( 12 n + 10 ) q n ≡ 36 f 1 4 f 3 6 / f 2 ( mod 9 ) , and Naika and Shivashankar [10 ] proved A 2 ( 6 n + 4 ) ≡ 0 ( m o d 3 ) A_2(6n+4)\equiv0\pmod{3} A 2 ( 6 n + 4 ) ≡ 0 ( mod 3 ) . Part (c) strengthens the modulus of the former from 9 to 27, and part (d) provides the next 3-adic digit. Part (b) characterizes the even part of 6 n + 4 6n+4 6 n + 4 ; its reduction modulo 9 is implicit in [9 ] , while the modulo 27 congruence appears to be new.
The odd-part version B k B_k B k
This section is the analogue of the previous one for designated partitions into odd parts. The 3-adic picture is simpler here, because the progression 3 n + 2 3n+2 3 n + 2 is controlled entirely by v 3 ( k ) v_3(k) v 3 ( k ) (Theorems 4.1 and 4.2); for powers of 2 we obtain congruences for B k ( 4 n + 3 ) B_k(4n+3) B k ( 4 n + 3 ) and their exactness for k ≡ 6 ( m o d 8 ) k\equiv6\pmod{8} k ≡ 6 ( mod 8 ) . For k = 2 k=2 k = 2 everything in this section is a statement about PDO( 2 n ) (2n) ( 2 n ) , because ∑ n B 2 ( n ) q n = G 2 = ∑ n PDO ( 2 n ) q n \sum_n B_2(n)q^n=G^2=\sum_n\operatorname{PDO}(2n)q^n ∑ n B 2 ( n ) q n = G 2 = ∑ n PDO ( 2 n ) q n , i.e. B 2 ( n ) = B 1 ( 2 n ) B_2(n)=B_1(2n) B 2 ( n ) = B 1 ( 2 n ) . This identity is implicit in [1 ] , was stated in [4 ] and has been refined in [14 ] ; it is proved by the 2-dissection of G G G that also underlies Lemma 3.11. Write G = f 4 f 6 2 / ( f 1 f 3 f 12 ) = 1 + q + 2 q 2 + 4 q 3 + ⋯ G=f_4f_6^2/(f_1f_3f_{12})=1+q+2q^2+4q^3+\cdots G = f 4 f 6 2 / ( f 1 f 3 f 12 ) = 1 + q + 2 q 2 + 4 q 3 + ⋯ , so that B k ( 2 ) = 2 k + ( k + 3 2 ) = k ( k + 3 ) / 2 B_k(2)=2k+\binom{k+3}{2}=k(k+3)/2 B k ( 2 ) = 2 k + ( 2 k + 3 ) = k ( k + 3 ) /2 .
Theorem 4.1. (a) B k ( 3 n + 2 ) ≡ 0 ( m o d 3 ) B_k(3n+2)\equiv0\pmod{3} B k ( 3 n + 2 ) ≡ 0 ( mod 3 ) for all n n n if and only if 3 ∣ k 3\mid k 3 ∣ k .
(b) B k ( 3 n + 2 ) ≡ 0 ( m o d 9 ) B_k(3n+2)\equiv0\pmod{9} B k ( 3 n + 2 ) ≡ 0 ( mod 9 ) for all n n n if and only if 3 ∣ k 3\mid k 3 ∣ k .
(c) If k ≡ 3 ( m o d 9 ) k\equiv3\pmod{9} k ≡ 3 ( mod 9 ) then B k ( 3 n + 2 ) ≢ 0 ( m o d 27 ) B_k(3n+2)\not\equiv0\pmod{27} B k ( 3 n + 2 ) ≡ 0 ( mod 27 ) for n = 0 n=0 n = 0 , so 9 is the exact power of 3 dividing all B k ( 3 n + 2 ) B_k(3n+2) B k ( 3 n + 2 ) .
Proof. Necessity. B k ( 2 ) = k ( k + 3 ) / 2 B_k(2)=k(k+3)/2 B k ( 2 ) = k ( k + 3 ) /2 . If 3 ∤ k 3\nmid k 3 ∤ k then gcd ( k , k + 3 ) = 1 \gcd(k,k+3)=1 g cd( k , k + 3 ) = 1 and 3 ∤ k ( k + 3 ) 3\nmid k(k+3) 3 ∤ k ( k + 3 ) , so B k ( 2 ) ≢ 0 ( m o d 3 ) B_k(2)\not\equiv0 \pmod{3} B k ( 2 ) ≡ 0 ( mod 3 ) . This gives the “only if” parts of (a) and (b). Part (c) follows from B k ( 2 ) = 9 t ( t + 1 ) / 2 B_k(2)=9t(t+1)/2 B k ( 2 ) = 9 t ( t + 1 ) /2 for k = 3 t k=3t k = 3 t , t ≡ 1 ( m o d 3 ) t\equiv1\pmod{3} t ≡ 1 ( mod 3 ) : then 3 ∤ t ( t + 1 ) / 2 3\nmid t(t+1)/2 3 ∤ t ( t + 1 ) /2 , so 27 ∤ B k ( 2 ) 27\nmid B_k(2) 27 ∤ B k ( 2 ) .
Sufficiency. If 3 ∣ k 3\mid k 3 ∣ k , (a) is Theorem 3.1. For (b) write the 3-dissection G = G 0 ( q 3 ) + q G 1 ( q 3 ) + q 2 G 2 ( q 3 ) G=G_0(q^3)+qG_1(q^3)+q^2G_2(q^3) G = G 0 ( q 3 ) + q G 1 ( q 3 ) + q 2 G 2 ( q 3 ) . We use Hirschhorn’s 3-dissection
f 4 f 1 = f 12 f 18 4 f 3 3 f 36 2 + q f 6 2 f 9 f 36 f 3 4 f 18 + 2 q 2 f 6 f 18 f 36 f 3 3 \frac{f_4}{f_1}=\frac{f_{12}f_{18}^4}{f_3^3f_{36}^2}+q\frac{f_6^2f_9f_{36}}{f_3^4f_{18}}+2q^2\frac{f_6f_{18}f_{36}}{f_3^3} f 1 f 4 = f 3 3 f 36 2 f 12 f 18 4 + q f 3 4 f 18 f 6 2 f 9 f 36 + 2 q 2 f 3 3 f 6 f 18 f 36
and the factor f 6 2 / ( f 3 f 12 ) f_6^2/(f_3f_{12}) f 6 2 / ( f 3 f 12 ) , which is a series in q 3 q^3 q 3 . After replacing q 3 q^3 q 3 by q q q ,
G 0 = f 2 2 f 4 6 f 1 4 f 12 2 , G 1 = f 4 2 f 3 3 f 12 f 1 5 f 4 2 , G 2 = 2 f 3 2 f 6 f 12 f 1 4 f 4 . G_0=\frac{f_2^2f_4^6}{f_1^4f_{12}^2},\qquad G_1=\frac{f_4^2f_3^3f_{12}}{f_1^5f_4^2},\qquad G_2=\frac{2f_3^2f_6f_{12}}{f_1^4f_4}. G 0 = f 1 4 f 12 2 f 2 2 f 4 6 , G 1 = f 1 5 f 4 2 f 4 2 f 3 3 f 12 , G 2 = f 1 4 f 4 2 f 3 2 f 6 f 12 .
Expanding G 3 G^3 G 3 and keeping the exponents ≡ 2 ( m o d 3 ) \equiv2\pmod{3} ≡ 2 ( mod 3 ) (the multinomial coefficients that occur are 3; the term 6 G 0 G 1 G 2 6G_0G_1G_2 6 G 0 G 1 G 2 has exponent ≡ 0 \equiv0 ≡ 0 ) gives the exact identity
∑ n ≥ 0 B 3 ( 3 n + 2 ) q n = 3 ( G 0 G 1 2 + G 0 2 G 2 + q G 1 G 2 2 ) = 3 G 0 G 1 2 ( 1 + r 2 + r 3 ) . \sum_{n\ge0}B_3(3n+2)q^n=3(G_0G_1^2+G_0^2G_2+qG_1G_2^2)=3G_0G_1^2(1+r_2+r_3). n ≥ 0 ∑ B 3 ( 3 n + 2 ) q n = 3 ( G 0 G 1 2 + G 0 2 G 2 + q G 1 G 2 2 ) = 3 G 0 G 1 2 ( 1 + r 2 + r 3 ) .
with
r 2 = G 0 G 2 G 1 2 = 2 f 1 2 f 4 f 6 9 f 3 2 f 6 3 f 12 2 , r 3 = q G 2 2 G 0 G 1 = 4 q f 1 f 3 3 f 3 3 f 4 . r_2=\frac{G_0G_2}{G_1^2}=\frac{2f_1^2f_4f_6^9}{f_3^2f_6^3f_{12}^2},\qquad r_3=\frac{qG_2^2}{G_0G_1}=\frac{4qf_1f_3^3}{f_3^3f_4}. r 2 = G 1 2 G 0 G 2 = f 3 2 f 6 3 f 12 2 2 f 1 2 f 4 f 6 9 , r 3 = G 0 G 1 q G 2 2 = f 3 3 f 4 4 q f 1 f 3 3 .
Modulo 3, using f 1 3 ≡ f 3 f_1^3\equiv f_3 f 1 3 ≡ f 3 , f 2 3 ≡ f 6 f_2^3\equiv f_6 f 2 3 ≡ f 6 , f 4 3 ≡ f 12 f_4^3\equiv f_{12} f 4 3 ≡ f 12 throughout, we get, with α = f 12 2 / ( f 1 8 f 4 4 ) \alpha=f_{12}^2/(f_1^8f_4^4) α = f 12 2 / ( f 1 8 f 4 4 ) and β = f 4 4 / f 1 4 \beta=f_4^4/f_1^4 β = f 4 4 / f 1 4 ,
r 2 ≡ 2 α 2 , r 3 ≡ q β 2 ( m o d 3 ) . r_2\equiv2\alpha^2,\qquad r_3\equiv q\beta^2\pmod{3}. r 2 ≡ 2 α 2 , r 3 ≡ q β 2 ( mod 3 ) .
In terms of Ramanujan’s functions, α = φ ( q ) 2 / φ ( − q ) 2 \alpha=\varphi(q)^2/\varphi(-q)^2 α = φ ( q ) 2 / φ ( − q ) 2 and β 2 = ψ ( q 2 ) 4 / φ ( − q ) 4 \beta^2=\psi(q^2)^4/\varphi(-q)^4 β 2 = ψ ( q 2 ) 4 / φ ( − q ) 4 , using φ ( q ) = f 2 5 / ( f 1 2 f 4 2 ) \varphi(q)=f_2^5/(f_1^2f_4^2) φ ( q ) = f 2 5 / ( f 1 2 f 4 2 ) , φ ( − q ) = f 1 2 / f 2 \varphi(-q)=f_1^2/f_2 φ ( − q ) = f 1 2 / f 2 , ψ ( q ) = f 2 2 / f 1 \psi(q)=f_2^2/f_1 ψ ( q ) = f 2 2 / f 1 and φ ( q ) φ ( − q ) = φ ( − q 2 ) \varphi(q)\varphi(-q)=\varphi(-q^2) φ ( q ) φ ( − q ) = φ ( − q 2 ) . Jacobi’s identity φ ( q ) 4 − φ ( − q ) 4 = 16 q ψ ( q 2 ) 4 \varphi(q)^4-\varphi(-q)^4=16q\psi(q^2)^4 φ ( q ) 4 − φ ( − q ) 4 = 16 q ψ ( q 2 ) 4 then yields α 2 = 1 + 16 q β 2 \alpha^2=1+16q\beta^2 α 2 = 1 + 16 q β 2 , and therefore
1 + r 2 + r 3 ≡ 1 + 2 ( 1 + 16 q β 2 ) + q β 2 ≡ 3 + 33 q β 2 ≡ 0 ( m o d 3 ) . 1+r_2+r_3\equiv1+2(1+16q\beta^2)+q\beta^2\equiv3+33q\beta^2\equiv0\pmod{3}. 1 + r 2 + r 3 ≡ 1 + 2 ( 1 + 16 q β 2 ) + q β 2 ≡ 3 + 33 q β 2 ≡ 0 ( mod 3 ) .
Hence B 3 ( 3 n + 2 ) ≡ 0 ( m o d 9 ) B_3(3n+2)\equiv0\pmod{9} B 3 ( 3 n + 2 ) ≡ 0 ( mod 9 ) ; more precisely the series G 3 = P 0 ( q 3 ) + q P 1 ( q 3 ) + q 2 P 2 ( q 3 ) G^3=P_0(q^3)+qP_1(q^3)+q^2P_2(q^3) G 3 = P 0 ( q 3 ) + q P 1 ( q 3 ) + q 2 P 2 ( q 3 ) has P 2 ≡ 0 ( m o d 9 ) P_2\equiv0\pmod{9} P 2 ≡ 0 ( mod 9 ) , and P 1 ≡ 0 ( m o d 3 ) P_1\equiv0\pmod{3} P 1 ≡ 0 ( mod 3 ) because G 3 ≡ G ( q 3 ) ( m o d 3 ) G^3\equiv G(q^3)\pmod{3} G 3 ≡ G ( q 3 ) ( mod 3 ) . For k = 3 t k=3t k = 3 t we get G k = ( P 0 + q P 1 + q 2 P 2 ) t ≡ P 0 t + t q P 1 P 0 t − 1 ( m o d 9 ) G^k=(P_0+qP_1+q^2P_2)^t\equiv P_0^t+tqP_1P_0^{t-1}\pmod{9} G k = ( P 0 + q P 1 + q 2 P 2 ) t ≡ P 0 t + tq P 1 P 0 t − 1 ( mod 9 ) , because P 1 ≡ 0 ( m o d 9 ) P_1\equiv0\pmod{9} P 1 ≡ 0 ( mod 9 ) ; this contains only exponents ≡ 0 , 1 ( m o d 3 ) \equiv0,1\pmod{3} ≡ 0 , 1 ( mod 3 ) , so B k ( 3 n + 2 ) ≡ 0 ( m o d 9 ) B_k(3n+2)\equiv0\pmod{9} B k ( 3 n + 2 ) ≡ 0 ( mod 9 ) .
Theorem 4.2
Theorem 4.2. Let 3 ∣ k 3\mid k 3 ∣ k and set a = v 3 ( k ) ≥ 1 a=v_3(k)\ge1 a = v 3 ( k ) ≥ 1 . Then B k ( 3 n + 2 ) ≡ 0 ( m o d 3 a + 1 ) B_k(3n+2)\equiv0\pmod{3^{a+1}} B k ( 3 n + 2 ) ≡ 0 ( mod 3 a + 1 ) for all n ≥ 0 n\ge0 n ≥ 0 . Furthermore, 3 a + 1 3^{a+1} 3 a + 1 is the exact power of 3 dividing all coefficients B k ( 3 n + 2 ) B_k(3n+2) B k ( 3 n + 2 ) .
Proof. First, we establish the lower bound by induction. Write G 3 a = Q 0 + q Q 1 + q 2 Q 2 G^{3^a}=Q_0+qQ_1+q^2Q_2 G 3 a = Q 0 + q Q 1 + q 2 Q 2 , where each Q i Q_i Q i is a series in q 3 q^3 q 3 . We show that all coefficients of Q 1 Q_1 Q 1 are divisible by 3 a 3^a 3 a and all coefficients of Q 2 Q_2 Q 2 are divisible by 3 a + 1 3^{a+1} 3 a + 1 .
For a = 1 a=1 a = 1 , this condition holds directly from the proof of Theorem 4.1, which establishes P 1 ≡ 0 ( m o d 3 ) P_1\equiv0\pmod{3} P 1 ≡ 0 ( mod 3 ) and P 2 ≡ 0 ( m o d 9 ) P_2\equiv0\pmod{9} P 2 ≡ 0 ( mod 9 ) . Assuming the assertion holds for some a ≥ 1 a\ge1 a ≥ 1 , we cube the expression Q 0 + q Q 1 + q 2 Q 2 Q_0+qQ_1+q^2Q_2 Q 0 + q Q 1 + q 2 Q 2 . The terms with exponents congruent to 1 modulo 3 are
3 Q 0 2 Q 1 q , 3 Q 0 Q 2 2 q 4 , 3 Q 1 2 Q 2 q 4 , 3Q_0^2Q_1q,\qquad3Q_0Q_2^2q^4,\qquad3Q_1^2Q_2q^4, 3 Q 0 2 Q 1 q , 3 Q 0 Q 2 2 q 4 , 3 Q 1 2 Q 2 q 4 ,
all of which are divisible by 3 a + 1 3^{a+1} 3 a + 1 . The terms with exponents congruent to 2 modulo 3 are
3 Q 0 Q 1 2 q 2 , 3 Q 0 2 Q 2 q 2 , 3 Q 1 Q 2 2 q 5 , 3Q_0Q_1^2q^2,\qquad3Q_0^2Q_2q^2,\qquad3Q_1Q_2^2q^5, 3 Q 0 Q 1 2 q 2 , 3 Q 0 2 Q 2 q 2 , 3 Q 1 Q 2 2 q 5 ,
which are divisible by 3 min ( 2 a + 1 , a + 2 , 3 a + 3 ) = 3 a + 2 3^{\min(2a+1,a+2,3a+3)}=3^{a+2} 3 m i n ( 2 a + 1 , a + 2 , 3 a + 3 ) = 3 a + 2 because a ≥ 1 a\ge1 a ≥ 1 . This confirms the claim for a + 1 a+1 a + 1 .
Now write k = 3 a m k=3^am k = 3 a m and expand G k = ( Q 0 + q Q 1 + q 2 Q 2 ) m G^k=(Q_0+qQ_1+q^2Q_2)^m G k = ( Q 0 + q Q 1 + q 2 Q 2 ) m . Any term of the form
Q 0 m − j − l ( q Q 1 ) j ( q 2 Q 2 ) l Q_0^{m-j-l}(qQ_1)^j(q^2Q_2)^l Q 0 m − j − l ( q Q 1 ) j ( q 2 Q 2 ) l
satisfying j + 2 l ≡ 2 ( m o d 3 ) j+2l\equiv2\pmod{3} j + 2 l ≡ 2 ( mod 3 ) requires either l ≥ 1 l\ge1 l ≥ 1 or j ≥ 2 j\ge2 j ≥ 2 . Its 3-adic valuation is at least j a + l ( a + 1 ) ≥ a + 1 ja+l(a+1)\ge a+1 j a + l ( a + 1 ) ≥ a + 1 , which proves that B k ( 3 n + 2 ) ≡ 0 ( m o d 3 a + 1 ) B_k(3n+2)\equiv0\pmod{3^{a+1}} B k ( 3 n + 2 ) ≡ 0 ( mod 3 a + 1 ) .
Second, we verify exactness. Evaluated at n = 0 n=0 n = 0 , we have
B k ( 2 ) = k ( k + 3 ) 2 = 3 a + 1 m ( 3 a − 1 m + 1 ) 2 . B_k(2)=\frac{k(k+3)}{2}=\frac{3^{a+1}m(3^{a-1}m+1)}{2}. B k ( 2 ) = 2 k ( k + 3 ) = 2 3 a + 1 m ( 3 a − 1 m + 1 ) .
If a ≥ 2 a\ge2 a ≥ 2 , or if a = 1 a=1 a = 1 with m ≢ 2 ( m o d 3 ) m\not\equiv2\pmod{3} m ≡ 2 ( mod 3 ) , the factor m ( 3 a − 1 m + 1 ) m(3^{a-1}m+1) m ( 3 a − 1 m + 1 ) is not divisible by 3, which yields v 3 ( B k ( 2 ) ) = a + 1 v_3(B_k(2))=a+1 v 3 ( B k ( 2 )) = a + 1 .
It remains to handle a = 1 a=1 a = 1 with m ≡ 2 ( m o d 3 ) m\equiv2\pmod{3} m ≡ 2 ( mod 3 ) , which corresponds to k = 6 , 15 , 24 ( m o d 27 ) k=6,15,24\pmod{27} k = 6 , 15 , 24 ( mod 27 ) . By Lemma 2.3 with e = 3 e=3 e = 3 , it is enough to show a single non-divisible coefficient for each of these three values. Direct computation shows that B 6 ( 5 ) = 1044 B_6(5)=1044 B 6 ( 5 ) = 1044 , B 15 ( 5 ) = 27693 B_{15}(5)=27693 B 15 ( 5 ) = 27693 , and B 24 ( 5 ) = 183456 B_{24}(5)=183456 B 24 ( 5 ) = 183456 . In all three cases, B k ( 5 ) ≡ 18 ( m o d 27 ) B_k(5)\equiv18\pmod{27} B k ( 5 ) ≡ 18 ( mod 27 ) , which is not divisible by 27. This completes the proof of exactness.
We now turn our attention to powers of 2. The initial results rely solely on the 2-dissection from Lemma 3.11.
Theorem 4.3. For all n ≥ 0 n\ge0 n ≥ 0 , the following congruences hold:
(a) B 1 ( 4 n + 3 ) ≡ 0 ( m o d 4 ) B_1(4n+3)\equiv0\pmod{4} B 1 ( 4 n + 3 ) ≡ 0 ( mod 4 ) .
(b) If k ≡ 1 ( m o d 4 ) k\equiv1\pmod{4} k ≡ 1 ( mod 4 ) , then B k ( 4 n + 3 ) ≡ 0 ( m o d 2 ) B_k(4n+3)\equiv0\pmod{2} B k ( 4 n + 3 ) ≡ 0 ( mod 2 ) .
(c) If k ≡ 2 ( m o d 4 ) k\equiv2\pmod{4} k ≡ 2 ( mod 4 ) , then B k ( 4 n + 3 ) ≡ 0 ( m o d 4 ) B_k(4n+3)\equiv0\pmod{4} B k ( 4 n + 3 ) ≡ 0 ( mod 4 ) .
Proof. (a) By Lemma 3.11, we can write G = E 1 ( q 2 ) + q O 1 ( q 2 ) G=E_1(q^2)+qO_1(q^2) G = E 1 ( q 2 ) + q O 1 ( q 2 ) with
O 1 = f 2 6 f 12 2 f 1 4 f 4 2 f 6 2 . O_1=\frac{f_2^6f_{12}^2}{f_1^4f_4^2f_6^2}. O 1 = f 1 4 f 4 2 f 6 2 f 2 6 f 12 2 .
Working modulo 4, we have f 1 4 ≡ f 2 2 f_1^4\equiv f_2^2 f 1 4 ≡ f 2 2 and f 4 2 ≡ f 4 2 f_4^2\equiv f_4^2 f 4 2 ≡ f 4 2 , which simplifies O 1 O_1 O 1 to
O 1 ≡ f 12 2 f 6 2 ( m o d 4 ) . O_1\equiv\frac{f_{12}^2}{f_6^2}\pmod{4}. O 1 ≡ f 6 2 f 12 2 ( mod 4 ) .
Since this is a series in x 2 x^2 x 2 , all of its odd-indexed coefficients are divisible by 4, which proves (a).
(b) Let k = 2 j + 1 k=2j+1 k = 2 j + 1 with j j j even. Then G k ≡ G ⋅ G ( q 2 ) j ( m o d 2 ) G^k\equiv G\cdot G(q^2)^j\pmod{2} G k ≡ G ⋅ G ( q 2 ) j ( mod 2 ) , which implies that
B k ( 2 m + 1 ) ≡ [ x m ] O 1 ( x ) G ( x ) j ( m o d 2 ) . B_k(2m+1)\equiv[x^m]O_1(x)G(x)^j\pmod{2}. B k ( 2 m + 1 ) ≡ [ x m ] O 1 ( x ) G ( x ) j ( mod 2 ) .
Because G j ≡ G ( x 2 ) j / 2 ( m o d 2 ) G^j\equiv G(x^2)^{j/2}\pmod{2} G j ≡ G ( x 2 ) j /2 ( mod 2 ) is a series in x 2 x^2 x 2 and the odd coefficients of O 1 O_1 O 1 are even, the coefficient vanishes modulo 2 when m = 2 n + 1 m=2n+1 m = 2 n + 1 , yielding B k ( 4 n + 3 ) ≡ 0 ( m o d 2 ) B_k(4n+3)\equiv0\pmod{2} B k ( 4 n + 3 ) ≡ 0 ( mod 2 ) .
(c) Write G 2 = E 2 ( q 2 ) + 2 q W 2 ( q 2 ) G^2=E_2(q^2)+2qW_2(q^2) G 2 = E 2 ( q 2 ) + 2 q W 2 ( q 2 ) , where
W 2 = f 2 6 f 6 2 f 1 6 f 3 2 ≡ f 2 3 f 6 ( m o d 2 ) W_2=\frac{f_2^6f_6^2}{f_1^6f_3^2}\equiv f_2^3f_6\pmod{2} W 2 = f 1 6 f 3 2 f 2 6 f 6 2 ≡ f 2 3 f 6 ( mod 2 )
is a series in x 2 x^2 x 2 modulo 2. For k = 2 m k=2m k = 2 m with m m m odd, the odd part of ( E 2 + 2 q W 2 ) m (E_2+2qW_2)^m ( E 2 + 2 q W 2 ) m is given by
2 m q W 2 E 2 m − 1 ( m o d 4 ) . 2m qW_2E_2^{m-1}\pmod{4}. 2 m q W 2 E 2 m − 1 ( mod 4 ) .
Since E 2 m − 1 ≡ E 2 ( x 2 ) ( m − 1 ) / 2 ( m o d 2 ) E_2^{m-1}\equiv E_2(x^2)^{(m-1)/2}\pmod{2} E 2 m − 1 ≡ E 2 ( x 2 ) ( m − 1 ) /2 ( mod 2 ) , the coefficient [ x 2 n + 1 ] ( W 2 E 2 m − 1 ) [x^{2n+1}](W_2E_2^{m-1}) [ x 2 n + 1 ] ( W 2 E 2 m − 1 ) is even. Consequently, B k ( 4 n + 3 ) ≡ 0 ( m o d 4 ) B_k(4n+3)\equiv0\pmod{4} B k ( 4 n + 3 ) ≡ 0 ( mod 4 ) .
Proposition 4.4.
∑ n ≥ 0 B 1 ( 4 n + 3 ) q n = 4 f 4 4 f 6 2 f 1 4 f 3 2 . \sum_{n\ge0}B_1(4n+3)q^n=4\frac{f_4^4f_6^2}{f_1^4f_3^2}. n ≥ 0 ∑ B 1 ( 4 n + 3 ) q n = 4 f 1 4 f 3 2 f 4 4 f 6 2 . In particular B 1 ( 4 n + 3 ) ≡ 0 ( m o d 4 ) B_1(4n+3)\equiv0\pmod{4} B 1 ( 4 n + 3 ) ≡ 0 ( mod 4 ) and B 1 ( 8 n + 7 ) ≡ 0 ( m o d 8 ) B_1(8n+7)\equiv0\pmod{8} B 1 ( 8 n + 7 ) ≡ 0 ( mod 8 ) , congruences which are due to Baruah and Ojah [4 ] .
Proof. By Lemma 3.11, ∑ B 1 ( 2 t + 1 ) x t = O 1 ( x ) = f 6 2 f 12 2 f 4 2 f 6 2 ⋅ 1 f 1 4 \sum B_1(2t+1)x^t=O_1(x)=\frac{f_6^2f_{12}^2}{f_4^2f_6^2}\cdot\frac{1}{f_1^4} ∑ B 1 ( 2 t + 1 ) x t = O 1 ( x ) = f 4 2 f 6 2 f 6 2 f 12 2 ⋅ f 1 4 1 . We use the classical 2-dissection [3 ]
1 f 1 4 = f 4 14 f 12 4 f 4 4 + 4 q f 4 2 f 8 4 f 12 10 . \frac{1}{f_1^4}=\frac{f_4^{14}}{f_{12}^{4}f_4^4}+4q\frac{f_4^2f_8^4}{f_{12}^{10}}. f 1 4 1 = f 12 4 f 4 4 f 4 14 + 4 q f 12 10 f 4 2 f 8 4 .
The prefactor is a series in x 2 x^2 x 2 , so the odd part of O 1 O_1 O 1 is 4 x f 8 4 f 12 2 f 4 2 f 6 2 4x\frac{f_8^4f_{12}^2}{f_4^2f_6^2} 4 x f 4 2 f 6 2 f 8 4 f 12 2 . Taking the coefficients of x 2 n + 1 x^{2n+1} x 2 n + 1 and replacing x 2 x^2 x 2 by q q q gives the formula. Modulo 2 we have f 4 4 f 6 2 f 1 4 f 3 2 ≡ f 1 2 f 3 2 = ( f 1 6 f 3 ) 2 \frac{f_4^4f_6^2}{f_1^4f_3^2}\equiv f_1^2f_3^2=(f_1^6f_3)^2 f 1 4 f 3 2 f 4 4 f 6 2 ≡ f 1 2 f 3 2 = ( f 1 6 f 3 ) 2 , a square, hence a series in q 2 q^2 q 2 modulo 2, so its coefficients of odd index are even; this gives the congruence modulo 8 for 8 n + 7 8n+7 8 n + 7 .
Theorem 4.5
Theorem 4.5. Let k ≥ 1 k\ge1 k ≥ 1 . Then the following congruences hold for all n ≥ 0 n\ge0 n ≥ 0 :
(a) If k ≡ 1 ( m o d 8 ) k\equiv1\pmod{8} k ≡ 1 ( mod 8 ) , then B k ( 4 n + 3 ) ≡ 0 ( m o d 4 ) B_k(4n+3)\equiv0\pmod{4} B k ( 4 n + 3 ) ≡ 0 ( mod 4 ) .
(b) If k ≡ 6 ( m o d 8 ) k\equiv6\pmod{8} k ≡ 6 ( mod 8 ) , then B k ( 4 n + 3 ) ≡ 0 ( m o d 8 ) B_k(4n+3)\equiv0\pmod{8} B k ( 4 n + 3 ) ≡ 0 ( mod 8 ) .
(c) If k ≡ 3 ( m o d 4 ) k\equiv3\pmod{4} k ≡ 3 ( mod 4 ) , then B k ( 3 ) B_k(3) B k ( 3 ) is odd, implying that B k ( 4 n + 3 ) ≢ 0 ( m o d 2 ) B_k(4n+3)\not\equiv0\pmod{2} B k ( 4 n + 3 ) ≡ 0 ( mod 2 ) for some n n n .
Proof. We write G = E 1 ( q 2 ) + q O 1 ( q 2 ) G=E_1(q^2)+qO_1(q^2) G = E 1 ( q 2 ) + q O 1 ( q 2 ) and G 2 = E ( q 2 ) + 2 q W ( q 2 ) G^2=E(q^2)+2qW(q^2) G 2 = E ( q 2 ) + 2 q W ( q 2 ) , where W = f 2 6 f 2 2 f 1 6 f 3 2 W=\frac{f_2^6f_2^2}{f_1^6f_3^2} W = f 1 6 f 3 2 f 2 6 f 2 2 (Lemma 3.11). Since G 2 ≡ G ( q 2 ) ( m o d 2 ) G^2\equiv G(q^2)\pmod{2} G 2 ≡ G ( q 2 ) ( mod 2 ) , we have E ≡ G ≡ f 1 3 f 3 ( m o d 2 ) E\equiv G\equiv\frac{f_1^3}{f_3}\pmod{2} E ≡ G ≡ f 3 f 1 3 ( mod 2 ) . Consequently, E 2 = E ( x 2 ) + 2 x W ( x 2 ) ( m o d 4 ) E^2=E(x^2)+2xW(x^2)\pmod{4} E 2 = E ( x 2 ) + 2 x W ( x 2 ) ( mod 4 ) , which leads to E 4 ≡ E ( x 2 ) 2 ( m o d 4 ) E^4\equiv E(x^2)^2\pmod{4} E 4 ≡ E ( x 2 ) 2 ( mod 4 ) , representing a series in x 2 x^2 x 2 modulo 4.
(a) Let k = 8 s + 1 k=8s+1 k = 8 s + 1 . Because ( E + 2 q W ) 4 s ≡ E 4 s ( m o d 8 ) (E+2qW)^{4s}\equiv E^{4s}\pmod{8} ( E + 2 q W ) 4 s ≡ E 4 s ( mod 8 ) , the odd part of G k G^k G k is congruent to q O 1 ( q 2 ) E 4 s ( q 2 ) ( m o d 4 ) qO_1(q^2)E^{4s}(q^2)\pmod{4} q O 1 ( q 2 ) E 4 s ( q 2 ) ( mod 4 ) . Here, E 4 s ≡ ( E 4 ) s E^{4s}\equiv(E^4)^s E 4 s ≡ ( E 4 ) s is even modulo 4, and the odd coefficients of O 1 O_1 O 1 are divisible by 4 (as established in the proof of Theorem 4.3 (a)). Thus, the coefficients of q 4 n + 3 q^{4n+3} q 4 n + 3 vanish modulo 4.
(b) Let k = 2 m k=2m k = 2 m with m = 4 s + 3 m=4s+3 m = 4 s + 3 . The odd part of ( E + 2 q W ) m (E+2qW)^m ( E + 2 q W ) m is congruent to 2 m q W E m − 1 ( m o d 8 ) 2mqWE^{m-1}\pmod{8} 2 m q W E m − 1 ( mod 8 ) (since terms with j ≥ 3 j\ge3 j ≥ 3 carry a factor of 2 3 2^3 2 3 ). This gives B k ( 2 t + 1 ) ≡ 2 m [ x t ] ( W E m − 1 ) ( m o d 8 ) B_k(2t+1)\equiv2m[x^t](WE^{m-1})\pmod{8} B k ( 2 t + 1 ) ≡ 2 m [ x t ] ( W E m − 1 ) ( mod 8 ) . For odd t t t , it suffices to show that [ x t ] ( W E m − 1 ) ≡ 0 ( m o d 4 ) [x^t](WE^{m-1})\equiv0\pmod{4} [ x t ] ( W E m − 1 ) ≡ 0 ( mod 4 ) . Since E m − 1 = E 2 ( E 4 ) s E^{m-1}=E^2(E^4)^s E m − 1 = E 2 ( E 4 ) s with E 4 E^4 E 4 even modulo 4, we only need the odd part of W E 2 WE^2 W E 2 to be congruent to 0 modulo 4. Using E 2 = E ( x 2 ) + 2 x W ( x 2 ) E^2=E(x^2)+2xW(x^2) E 2 = E ( x 2 ) + 2 x W ( x 2 ) , we obtain
odd ( W E 2 ) = W o d d E ( x 2 ) + 2 x W ( x 2 ) W e v e n ( m o d 4 ) . \operatorname{odd}(WE^2)=W_{\mathrm{odd}}E(x^2)+2xW(x^2)W_{\mathrm{even}}\pmod{4}. odd ( W E 2 ) = W odd E ( x 2 ) + 2 x W ( x 2 ) W even ( mod 4 ) .
Modulo 4, f 1 4 ≡ f 2 2 f_1^4\equiv f_2^2 f 1 4 ≡ f 2 2 implies W ≡ ( f 1 f 2 f 3 ) 2 W\equiv(f_1f_2f_3)^2 W ≡ ( f 1 f 2 f 3 ) 2 . Applying Hirschhorn’s dissection
f 1 f 3 = f 2 2 f 8 f 12 2 f 4 2 f 6 f 24 − q f 4 4 f 6 f 24 f 2 2 f 8 2 f 12 2 , f_1f_3=\frac{f_2^2f_8f_{12}^2}{f_4^2f_6f_{24}}-q\frac{f_4^4f_6f_{24}}{f_2^2f_8^2f_{12}^2}, f 1 f 3 = f 4 2 f 6 f 24 f 2 2 f 8 f 12 2 − q f 2 2 f 8 2 f 12 2 f 4 4 f 6 f 24 ,
we find W o d d ≡ − 2 x f 2 2 f 4 2 f 12 2 ( m o d 4 ) W_{\mathrm{odd}}\equiv-2xf_2^2f_4^2f_{12}^2\pmod{4} W odd ≡ − 2 x f 2 2 f 4 2 f 12 2 ( mod 4 ) , while W ≡ f 3 2 f 6 ( m o d 2 ) W\equiv f_3^2f_6\pmod{2} W ≡ f 3 2 f 6 ( mod 2 ) is even. In terms of y = x 2 y=x^2 y = x 2 , the odd part of W E 2 WE^2 W E 2 takes the form 2 x [ f 1 2 f 2 2 f 6 2 E ( y ) + W ( y ) W e ( y ) ] ( m o d 4 ) 2x[f_1^2f_2^2f_6^2E(y)+W(y)W_{\mathrm{e}}(y)]\pmod{4} 2 x [ f 1 2 f 2 2 f 6 2 E ( y ) + W ( y ) W e ( y )] ( mod 4 ) , where W e ( y ) ≡ f 1 3 f 3 W_{\mathrm{e}}(y)\equiv f_1^3f_3 W e ( y ) ≡ f 1 3 f 3 , W ( y ) = W e ( y ) 2 W(y)=W_{\mathrm{e}}(y)^2 W ( y ) = W e ( y ) 2 , and E ( y ) ≡ f 1 3 / f 3 ( m o d 2 ) E(y)\equiv f_1^3/f_3\pmod{2} E ( y ) ≡ f 1 3 / f 3 ( mod 2 ) . The bracketed expression is congruent modulo 2 to f 1 6 f 3 4 ⋅ ( f 1 3 / f 3 ) + f 1 3 f 3 3 ≡ 0 ( m o d 2 ) f_1^6f_3^4\cdot(f_1^3/f_3)+f_1^3f_3^3\equiv0\pmod{2} f 1 6 f 3 4 ⋅ ( f 1 3 / f 3 ) + f 1 3 f 3 3 ≡ 0 ( mod 2 ) .
(c) Evaluating at n = 0 n=0 n = 0 , we have B k ( 3 ) = 4 k + 2 k ( k − 1 ) + ( k 3 ) = k ( k + 2 ) ( k + 7 ) / 6 B_k(3)=4k+2k(k-1)+\binom{k}{3}=k(k+2)(k+7)/6 B k ( 3 ) = 4 k + 2 k ( k − 1 ) + ( 3 k ) = k ( k + 2 ) ( k + 7 ) /6 . For k ≡ 3 ( m o d 4 ) k\equiv3\pmod{4} k ≡ 3 ( mod 4 ) , the binomial coefficient ( k 3 ) \binom{k}{3} ( 3 k ) is odd while the other terms are even, making B k ( 3 ) B_k(3) B k ( 3 ) odd and ensuring that B k ( 4 n + 3 ) B_k(4n+3) B k ( 4 n + 3 ) is not identically divisible by 2 for all n n n .
Theorem 4.6
Theorem 4.6. For k ≡ 6 ( m o d 8 ) k\equiv6\pmod{8} k ≡ 6 ( mod 8 ) , 8 is the exact power of 2 that divides all coefficients B k ( 4 n + 3 ) B_k(4n+3) B k ( 4 n + 3 ) .
Proof. Divisibility by 8 is guaranteed by Theorem 4.5 (b). Therefore, it remains to show that for each k ≡ 6 ( m o d 8 ) k\equiv6\pmod{8} k ≡ 6 ( mod 8 ) , there is at least one coefficient B k ( 4 n + 3 ) B_k(4n+3) B k ( 4 n + 3 ) that is not divisible by 16. Writing k = 8 s + 6 k=8s+6 k = 8 s + 6 , we recall the expression for the first relevant coefficient:
B k ( 3 ) = k ( k + 2 ) ( k + 7 ) 6 . B_k(3)=\frac{k(k+2)(k+7)}{6}. B k ( 3 ) = 6 k ( k + 2 ) ( k + 7 ) .
Case 1: s s s is even. Here, v 2 ( k ) = 1 v_{2}(k)=1 v 2 ( k ) = 1 . Since k + 2 = 8 ( s + 1 ) k+2=8(s+1) k + 2 = 8 ( s + 1 ) with s + 1 s+1 s + 1 being odd, we have v 2 ( k + 2 ) = 3 v_{2}(k+2)=3 v 2 ( k + 2 ) = 3 , while k + 7 k+7 k + 7 is odd. Consequently, the 2-adic valuation is
v 2 ( B k ( 3 ) ) = 1 + 3 − 1 = 3 , v_{2}(B_{k}(3))=1+3-1=3, v 2 ( B k ( 3 )) = 1 + 3 − 1 = 3 ,
which means B k ( 3 ) B_{k}(3) B k ( 3 ) is not divisible by 16.
Case 2: s s s is odd (meaning k ≡ 14 ( m o d 16 ) k\equiv14\pmod{16} k ≡ 14 ( mod 16 ) ). In this case, we examine the next coefficient, B k ( 7 ) B_{k}(7) B k ( 7 ) . We express B k ( n ) B_{k}(n) B k ( n ) as a polynomial in ( k i ) \binom{k}{i} ( i k ) :
B k ( n ) = ∑ i = 0 n c i ( n ) ( k i ) , where c i ( n ) = [ q n ] ( G − 1 ) i . B_{k}(n)=\sum_{i=0}^{n}c_{i}(n)\binom{k}{i},
\qquad\text{where }c_{i}(n)=[q^{n}](G-1)^{i}. B k ( n ) = i = 0 ∑ n c i ( n ) ( i k ) , where c i ( n ) = [ q n ] ( G − 1 ) i .
For i ≤ 7 i\leq7 i ≤ 7 and 1 ≤ j ≤ 7 1\leq j\leq7 1 ≤ j ≤ 7 , we have v 2 ( ( 64 j ) ) = 6 − v 2 ( j ) ≥ 4 v_{2}\left(\binom{64}{j}\right)=6-v_{2}(j)\geq4 v 2 ( ( j 64 ) ) = 6 − v 2 ( j ) ≥ 4 . Thus, by Vandermonde’s identity,
( k + 64 i ) = ∑ j ( 64 j ) ( k i − j ) ≡ ( k i ) ( m o d 16 ) . \binom{k+64}{i}=\sum_{j}\binom{64}{j}\binom{k}{i-j}\equiv\binom{k}{i}\pmod{16}. ( i k + 64 ) = j ∑ ( j 64 ) ( i − j k ) ≡ ( i k ) ( mod 16 ) .
This implies that B k ( 7 ) ( m o d 16 ) B_{k}(7)\pmod{16} B k ( 7 ) ( mod 16 ) depends only on k ( m o d 64 ) k\pmod{64} k ( mod 64 ) . The possible residues modulo 64 corresponding to k ≡ 14 ( m o d 16 ) k\equiv14\pmod{16} k ≡ 14 ( mod 16 ) are 14, 30, 46, and 62. For each of these residues, B k ( 7 ) ≡ 8 ( m o d 16 ) B_{k}(7)\equiv8\pmod{16} B k ( 7 ) ≡ 8 ( mod 16 ) , meaning that v 2 ( B k ( 7 ) ) = 3 v_{2}(B_{k}(7))=3 v 2 ( B k ( 7 )) = 3 for every such k k k . This completes the proof. □
Combinatorial interpretations and other partition functions
Young diagrams enter the paper in two ways. They give the definition of the objects we count (Figure 1 ), and they allow some of the divisibilities above to be proved by arguments about tuples of diagrams rather than by generating functions. Let T k ( n ) T_{k}(n) T k ( n ) denote the set of ordered k k k -tuples of designated Young diagrams of total size n n n (the arguments below work for any family of weighted objects). The divisibility of Theorem 3.1 then has a purely bijective proof.
Proposition 5.1. If p j ∣ k p^{j}\mid k p j ∣ k and p ∤ n p\nmid n p ∤ n , then ∣ T k ( n ) ∣ ≡ 0 ( m o d p j ) \lvert T_{k}(n)\rvert\equiv0\pmod{p^{j}} ∣ T k ( n )∣ ≡ 0 ( mod p j ) .
Proof. Let C = Z / p j C=\mathbb{Z}/p^{j} C = Z / p j act on T k ( n ) T_{k}(n) T k ( n ) by the cyclic shift of the tuple by k / p j k/p^{j} k / p j positions. If a tuple has a nontrivial stabiliser, it contains the subgroup of order p p p , i.e. the tuple is invariant under the shift by k / p k/p k / p ; then it consists of p p p copies of a block of k / p k/p k / p objects, so n = p ⋅ ( weight of the block ) n=p\cdot(\text{weight of the block}) n = p ⋅ ( weight of the block ) , contradicting p ∤ n p\nmid n p ∤ n . So all orbits are regular and have size p j p^{j} p j . □
For k = 1 k=1 k = 1 the congruence PD ( 3 n + 2 ) ≡ 0 ( m o d 3 ) \operatorname{PD}(3n+2)\equiv0\pmod{3} PD ( 3 n + 2 ) ≡ 0 ( mod 3 ) already has combinatorial explanations. Chen, Ji, Jin and Shen [5 Section 3] construct a bijection Δ \Delta Δ from the designated partitions of n n n to the pairs ( α , β ) (\alpha,\beta) ( α , β ) in which α \alpha α is an ordinary partition of n 1 n_{1} n 1 and β \beta β is a partition of n 2 = n − n 1 n_{2}=n-n_{1} n 2 = n − n 1 into parts not congruent to ± 1 \pm1 ± 1 modulo 6; this realises the factorisation F = 1 f 1 ⋅ f 6 f 2 f 3 F=\frac{1}{f_{1}}\cdot\frac{f_{6}}{f_{2}f_{3}} F = f 1 1 ⋅ f 2 f 3 f 6 . Let ℓ e \ell_{e} ℓ e denote the number of even parts. The pd-rank r d ( λ ) = ℓ e ( α ) − ℓ e ( β ) r_{d}(\lambda)=\ell_{e}(\alpha)-\ell_{e}(\beta) r d ( λ ) = ℓ e ( α ) − ℓ e ( β ) then splits the designated partitions of 3 n + 2 3n+2 3 n + 2 into three classes of equal size, and the equidistribution is proved there by generating functions. Hao, Shen and Zang [6 ] extended this to overline designated summands, Hao [7 ] gave a bijective proof by orbits of three with distinct V R V_{R} V R -rank, and Shen [8 ] defined a crank. All of these concern k = 1 k=1 k = 1 or bipartitions; for k ≥ 3 k\geq3 k ≥ 3 no combinatorial statistic is known to us.
Proposition 5.2. Let r d r_{d} r d be the pd-rank and put
R ( z , q ) = ∑ λ z r d ( λ ) q ∣ λ ∣ = ∏ n o d d 1 1 − q n ∏ n e v e n 1 ( 1 − z q n ) ( 1 − z − 1 q n ) ∏ n ≡ 3 ( 6 ) 1 1 − q n . R(z,q)=\sum_{\lambda}z^{r_{d}(\lambda)}q^{\lvert\lambda\rvert}
=\prod_{n\ \mathrm{odd}}\frac{1}{1-q^{n}}
\prod_{n\ \mathrm{even}}\frac{1}{(1-zq^{n})(1-z^{-1}q^{n})}
\prod_{n\equiv3\ (6)}\frac{1}{1-q^{n}}. R ( z , q ) = λ ∑ z r d ( λ ) q ∣ λ ∣ = n odd ∏ 1 − q n 1 n even ∏ ( 1 − z q n ) ( 1 − z − 1 q n ) 1 n ≡ 3 ( 6 ) ∏ 1 − q n 1 .
Then R ( 1 , q ) = F R(1,q)=F R ( 1 , q ) = F and, for ζ = e 2 π i / 3 \zeta=e^{2\pi i/3} ζ = e 2 π i /3 ,
R ( ζ , q ) = ψ ( q ) f 3 , ψ ( q ) = ∑ m ≥ 0 q m ( m + 1 ) / 2 = f 2 2 f 1 . R(\zeta,q)=\frac{\psi(q)}{f_{3}},
\qquad
\psi(q)=\sum_{m\geq0}q^{m(m+1)/2}=\frac{f_{2}^{2}}{f_{1}}. R ( ζ , q ) = f 3 ψ ( q ) , ψ ( q ) = m ≥ 0 ∑ q m ( m + 1 ) /2 = f 1 f 2 2 .
Consequently F ≡ ψ ( q ) / f 3 ( m o d 3 ) F \equiv\psi(q)/f_3 \pmod{3} F ≡ ψ ( q ) / f 3 ( mod 3 ) , and for every k ≥ 1 k \ge1 k ≥ 1
A k ( n ) ≡ ∑ j ≥ 0 t k ( n − 3 j ) p − k ( j ) ( m o d 3 ) , A_k(n) \equiv\sum_{j\ge0} t_k(n-3j)p_{-k}(j) \pmod{3}, A k ( n ) ≡ j ≥ 0 ∑ t k ( n − 3 j ) p − k ( j ) ( mod 3 ) ,
where t k ( N ) t_k(N) t k ( N ) is the number of k k k -tuples of staircase partitions ( m , m − 1 , … , 1 ) (m,m-1,\ldots,1) ( m , m − 1 , … , 1 ) , m ≥ 0 m \ge0 m ≥ 0 , with N N N cells in total, and p − k ( j ) p_{-k}(j) p − k ( j ) the number of k k k -tuples of ordinary partitions of total size j j j .
Proof. The product for R R R is the pair ( α , β ) (\alpha,\beta) ( α , β ) of [5 ] (the evaluation at ζ \zeta ζ is implicit in their proof of the equidistribution; the consequence for k k k -tuples is what we add here) with z z z marking ℓ e ( α ) \ell_e(\alpha) ℓ e ( α ) and z − 1 z^{-1} z − 1 marking ℓ e ( β ) \ell_e(\beta) ℓ e ( β ) ; at z = 1 z=1 z = 1 it is F F F . For z = ζ z=\zeta z = ζ , ( 1 − ζ q n ) ( 1 − ζ − 1 q n ) = ( 1 − q 3 n ) / ( 1 − q n ) (1-\zeta q^n)(1-\zeta^{-1}q^n)=(1-q^{3n})/(1-q^n) ( 1 − ζ q n ) ( 1 − ζ − 1 q n ) = ( 1 − q 3 n ) / ( 1 − q n ) , so the even-part factors give ( q 2 ; q 2 ) ∞ / ( q 6 ; q 6 ) ∞ (q^2;q^2)_\infty/(q^6;q^6)_\infty ( q 2 ; q 2 ) ∞ / ( q 6 ; q 6 ) ∞ and
R ( ζ , q ) = ( q 2 ; q 2 ) ∞ ( q ; q 2 ) ∞ ( q 3 ; q 6 ) ∞ ( q 6 ; q 6 ) ∞ = ( q 2 ; q 2 ) ∞ ( q ; q 2 ) ∞ ⋅ 1 ( q 3 ; q 3 ) ∞ . R(\zeta,q)=\frac{(q^2;q^2)_\infty}{(q;q^2)_\infty(q^3;q^6)_\infty(q^6;q^6)_\infty}
=\frac{(q^2;q^2)_\infty}{(q;q^2)_\infty}\cdot\frac{1}{(q^3;q^3)_\infty}. R ( ζ , q ) = ( q ; q 2 ) ∞ ( q 3 ; q 6 ) ∞ ( q 6 ; q 6 ) ∞ ( q 2 ; q 2 ) ∞ = ( q ; q 2 ) ∞ ( q 2 ; q 2 ) ∞ ⋅ ( q 3 ; q 3 ) ∞ 1 .
using ( q 3 ; q 6 ) ∞ ( q 6 ; q 6 ) ∞ = ( q 3 ; q 3 ) ∞ (q^3;q^6)_\infty(q^6;q^6)_\infty=(q^3;q^3)_\infty ( q 3 ; q 6 ) ∞ ( q 6 ; q 6 ) ∞ = ( q 3 ; q 3 ) ∞ . Gauss’s identity gives ( q 2 ; q 2 ) ∞ / ( q ; q 2 ) ∞ = ψ ( q ) (q^2;q^2)_\infty/(q;q^2)_\infty=\psi(q) ( q 2 ; q 2 ) ∞ / ( q ; q 2 ) ∞ = ψ ( q ) . Since N 0 + N 1 ζ + N 2 ζ 2 ≡ N 0 + N 1 + N 2 ( m o d 1 − ζ ) N_0+N_1\zeta+N_2\zeta^2\equiv N_0+N_1+N_2\pmod{1-\zeta} N 0 + N 1 ζ + N 2 ζ 2 ≡ N 0 + N 1 + N 2 ( mod 1 − ζ ) and ( 1 − ζ ) ∩ Z = 3 Z (1-\zeta)\cap\mathbb{Z}=3\mathbb{Z} ( 1 − ζ ) ∩ Z = 3 Z , we get F ≡ R ( ζ , q ) ( m o d 3 ) F\equiv R(\zeta,q)\pmod{3} F ≡ R ( ζ , q ) ( mod 3 ) , and the same for k k k -th powers. □ \square □
The proposition says that, modulo 3, a designated partition is a staircase diagram together with an ordinary Young diagram in which every cell has been replaced by three. For k = 1 k=1 k = 1 the vanishing for n ≡ 2 ( m o d 3 ) n\equiv2\pmod{3} n ≡ 2 ( mod 3 ) is then the fact that triangular numbers are ≡ 0 , 1 ( m o d 3 ) \equiv0,1\pmod{3} ≡ 0 , 1 ( mod 3 ) . For k ≡ 1 ( m o d 3 ) k\equiv1\pmod{3} k ≡ 1 ( mod 3 ) the same conclusion follows from ψ ( q ) 3 ≡ ψ ( q 3 ) ( m o d 3 ) \psi(q)^3\equiv\psi(q^3)\pmod{3} ψ ( q ) 3 ≡ ψ ( q 3 ) ( mod 3 ) , which gives ψ k ≡ ψ ⋅ ψ ( q 3 ) ( k − 1 ) / 3 \psi^k\equiv\psi\cdot\psi(q^3)^{(k-1)/3} ψ k ≡ ψ ⋅ ψ ( q 3 ) ( k − 1 ) /3 . This identity, however, is not realised by a statistic on k k k -tuples, and a combinatorial explanation of Theorem 3.3 for k ≡ 1 ( m o d 3 ) k\equiv1\pmod{3} k ≡ 1 ( mod 3 ) , k ≥ 4 k\ge4 k ≥ 4 , would require a new idea; this is Problem 6.5.
Remark 5.3. Proposition 5.1 is a bijective argument for the families p j ∣ k p^j\mid k p j ∣ k and works in every k k k -tuple model. The other divisibility results for the classes 3 n + r 3n+r 3 n + r and 4 n + 3 4n+3 4 n + 3 rest on dissections. By Propositions 3.5 and 5.2 the ones modulo 3 are equivalent to statements about overpartition tuples (sums of squares) or staircase tuples (sums of triangular numbers).
Proposition 5.4. Let T r ( N ) T_r(N) T r ( N ) be the set of r r r -tuples of weighted objects of total weight N N N (e.g. designated Young diagrams). For k = 3 j + r k=3j+r k = 3 j + r with 0 ≤ r ≤ 2 0\le r\le2 0 ≤ r ≤ 2 ,
∣ T k ( N ) ∣ ≡ # { ( μ 1 , … , μ j ; ν ) : μ i ∈ T 1 , ν ∈ T r , 3 ∑ ∣ μ i ∣ + ∣ ν ∣ = N } ( m o d 3 ) . |T_k(N)|\equiv\#\{(\mu_1,\ldots,\mu_j;\nu):\ \mu_i\in T_1,\ \nu\in T_r,\ 3\sum|\mu_i|+|\nu|=N\}\pmod{3}. ∣ T k ( N ) ∣ ≡ # {( μ 1 , … , μ j ; ν ) : μ i ∈ T 1 , ν ∈ T r , 3 ∑ ∣ μ i ∣ + ∣ ν ∣ = N } ( mod 3 ) .
Proof. Let Z / 3 \mathbb{Z}/3 Z /3 act on k k k -tuples by cyclically rotating the first 3 coordinates, the next 3, and so on (j j j blocks), fixing the last r r r coordinates. Non-fixed orbits have size 3. A fixed tuple has equal entries inside each block, so it is determined by ( μ 1 , … , μ j ; ν ) (\mu_1,\ldots,\mu_j;\nu) ( μ 1 , … , μ j ; ν ) with weight 3 ∑ ∣ μ i ∣ + ∣ ν ∣ 3\sum|\mu_i|+|\nu| 3 ∑ ∣ μ i ∣ + ∣ ν ∣ . □ \square □
Corollary 5.5. If k ≡ 0 , 1 ( m o d 3 ) k\equiv0,1\pmod{3} k ≡ 0 , 1 ( mod 3 ) then A k ( 3 n + 2 ) ≡ 0 ( m o d 3 ) A_k(3n+2)\equiv0\pmod{3} A k ( 3 n + 2 ) ≡ 0 ( mod 3 ) .
This is the case e = 1 e=1 e = 1 of Theorem 3.3 ; the point here is the bijective argument, which reduces the case k ≡ 1 ( m o d 3 ) k\equiv1\pmod{3} k ≡ 1 ( mod 3 ) to k = 1 k=1 k = 1 .
Proof. For k = 3 j k=3j k = 3 j the fixed tuples have weight ≡ 0 ( m o d 3 ) \equiv0\pmod{3} ≡ 0 ( mod 3 ) , so there are none of weight 3 n + 2 3n+2 3 n + 2 . For k = 3 j + 1 k=3j+1 k = 3 j + 1 a fixed tuple of weight 3 n + 2 3n+2 3 n + 2 consists of μ 1 , … , μ j \mu_1,\ldots,\mu_j μ 1 , … , μ j and a single ν \nu ν with ∣ ν ∣ ≡ 2 ( m o d 3 ) |\nu|\equiv2\pmod{3} ∣ ν ∣ ≡ 2 ( mod 3 ) ; for fixed μ \mu μ 's the number of such ν \nu ν is A 1 ( 3 m + 2 ) A_1(3m+2) A 1 ( 3 m + 2 ) , which is divisible by 3 because, by Chen–Ji–Jin–Shen, the three pd-rank classes of the partitions of 3 m + 2 3m+2 3 m + 2 have equal size. □ \square □
Iterating Proposition 5.4 on the base-3 digits of k k k gives F k ≡ ∏ i F ( q 3 i ) k i ( m o d 3 ) F^k\equiv\prod_i F(q^{3^i})^{k_i}\pmod{3} F k ≡ ∏ i F ( q 3 i ) k i ( mod 3 ) , a bijective form of Lemma 2.2 for p = 3 p=3 p = 3 . The same orbit argument modulo 9 would require tracking orbits of size 3 versus 9 and does not give Theorem 3.3 ; its combinatorial proof is open.
Proposition 5.6. Let α \alpha α run over partitions into distinct parts, β \beta β over partitions into parts ≡ 2 ( m o d 4 ) \equiv2 \pmod{4} ≡ 2 ( mod 4 ) , and γ \gamma γ over partitions into distinct parts ≡ 3 ( m o d 6 ) \equiv3 \pmod{6} ≡ 3 ( mod 6 ) . Then PDO ( n ) \operatorname{PDO}(n) PDO ( n ) is the number of triples ( α , β , γ ) (\alpha,\beta,\gamma) ( α , β , γ ) with ∣ α ∣ + ∣ β ∣ + ∣ γ ∣ = n |\alpha|+|\beta|+|\gamma|=n ∣ α ∣ + ∣ β ∣ + ∣ γ ∣ = n ; that is,
∑ n ≥ 0 PDO ( n ) q n = f 4 f 6 2 f 1 f 3 f 12 = ( − q ; q ) ∞ 1 ( q 2 ; q 4 ) ∞ ( − q 3 ; q 6 ) ∞ . \sum_{n\geq0}\operatorname{PDO}(n)q^n=\frac{f_4f_6^2}{f_1f_3f_{12}}=(-q;q)_\infty\frac{1}{(q^2;q^4)_\infty}(-q^3;q^6)_\infty. n ≥ 0 ∑ PDO ( n ) q n = f 1 f 3 f 12 f 4 f 6 2 = ( − q ; q ) ∞ ( q 2 ; q 4 ) ∞ 1 ( − q 3 ; q 6 ) ∞ . Proof. By Section 2, the generating function is ∏ n o d d ( 1 + q 3 n ) / ( ( 1 − q n ) ( 1 − q 2 n ) ) \prod_{n\ \mathrm{odd}}(1+q^{3n})/((1-q^n)(1-q^{2n})) ∏ n odd ( 1 + q 3 n ) / (( 1 − q n ) ( 1 − q 2 n )) . Now ∏ n o d d ( 1 − q n ) − 1 = ( − q ; q ) ∞ \prod_{n\ \mathrm{odd}}(1-q^n)^{-1}=(-q;q)_\infty ∏ n odd ( 1 − q n ) − 1 = ( − q ; q ) ∞ , ∏ n o d d ( 1 − q 2 n ) − 1 = 1 / ( q 2 ; q 4 ) ∞ \prod_{n\ \mathrm{odd}}(1-q^{2n})^{-1}=1/(q^2;q^4)_\infty ∏ n odd ( 1 − q 2 n ) − 1 = 1/ ( q 2 ; q 4 ) ∞ , and ∏ n o d d ( 1 + q 3 n ) = ( − q 3 ; q 6 ) ∞ \prod_{n\ \mathrm{odd}}(1+q^{3n})=(-q^3;q^6)_\infty ∏ n odd ( 1 + q 3 n ) = ( − q 3 ; q 6 ) ∞ .
This is the analogue for PDO \operatorname{PDO} PDO of the pair ( α , β ) (\alpha,\beta) ( α , β ) of [5 ] . An explicit bijection from dotted Young diagrams with odd rows to such triples (the analogue of Δ \Delta Δ ) is not constructed here, and we do not know a pd-rank for it.
Conjectures
Computations suggest that the divisibility properties established in this paper are part of a broader, regular pattern. Specifically, our first two conjectures state that the divisibility condition derived from the initial coefficient—namely, A k ( 3 ) = k ( k + 1 ) ( k + 14 ) / 6 A_k(3)=k(k+1)(k+14)/6 A k ( 3 ) = k ( k + 1 ) ( k + 14 ) /6 and B k ( 3 ) = k ( k + 2 ) ( k + 7 ) / 6 B_k(3)=k(k+2)(k+7)/6 B k ( 3 ) = k ( k + 2 ) ( k + 7 ) /6 —is essentially also sufficient for the progression 4 n + 3 4n+3 4 n + 3 .
Conjecture 6.1. For every k ≥ 1 k\geq1 k ≥ 1 , the largest power of 2 2 2 dividing all A k ( 4 n + 3 ) A_k(4n+3) A k ( 4 n + 3 ) is 2 v 2 ( A k ( 3 ) ) 2^{v_2(A_k(3))} 2 v 2 ( A k ( 3 )) if k ≢ 2 ( m o d 16 ) k\not\equiv2\pmod{16} k ≡ 2 ( mod 16 ) , and exactly 2 3 2^3 2 3 if k ≡ 2 ( m o d 16 ) k\equiv2\pmod{16} k ≡ 2 ( mod 16 ) .
Conjecture 6.2. For every k ≥ 1 k\geq1 k ≥ 1 , the largest power of 2 2 2 dividing all B k ( 4 n + 3 ) B_k(4n+3) B k ( 4 n + 3 ) is 2 v 2 ( B k ( 3 ) ) 2^{v_2(B_k(3))} 2 v 2 ( B k ( 3 )) , except when k ≡ 1 ( m o d 8 ) k\equiv1\pmod{8} k ≡ 1 ( mod 8 ) (where it is exactly 2 2 2^2 2 2 ) or k ≡ 6 ( m o d 8 ) k\equiv6\pmod{8} k ≡ 6 ( mod 8 ) (where it is exactly 2 3 2^3 2 3 , as established in Theorem 4.6 ).
Both statements have been verified for k ≤ 96 k\leq96 k ≤ 96 and the first few hundred values of n n n . Theorem 3.12 and Proposition 3.13 establish only the lower bounds for certain residue classes of k k k . The exceptional cases are particularly striking: for k ≡ 2 ( m o d 16 ) k\equiv2\pmod{16} k ≡ 2 ( mod 16 ) , A k ( 3 ) A_k(3) A k ( 3 ) is divisible by 2 4 2^4 2 4 or higher, yet the exact divisibility bound is precisely 2 3 2^3 2 3 . Theorem 3.17 shows that for k = 2 k=2 k = 2 , the progression 12 n + 10 12n+10 12 n + 10 exhibits unusually deep divisibility. Among 3 ≤ k ≤ 27 3\leq k\leq27 3 ≤ k ≤ 27 , this phenomenon appears again only for k = 6 k=6 k = 6 , where the divisibility experiences a sharp jump.
Conjecture 6.3. For all n ≥ 0 n\geq0 n ≥ 0 , A 6 ( 12 n + 10 ) ≡ 0 ( m o d 3 5 ) A_6(12n+10)\equiv0\pmod{3^5} A 6 ( 12 n + 10 ) ≡ 0 ( mod 3 5 ) , and 3 5 3^5 3 5 is the exact power of 3 3 3 dividing all of them.
For k = 6 k=6 k = 6 , the generating function for the progression 3 n + 1 3n+1 3 n + 1 can be expressed exactly as
∑ A 6 ( 3 j + 1 ) x j = 3 Θ 6 ρ R ( w ) , \sum A_6(3j+1)x^j=3\Theta^6\rho R(w), ∑ A 6 ( 3 j + 1 ) x j = 3 Θ 6 ρR ( w ) ,
where R ( w ) R(w) R ( w ) is a polynomial of degree 7 7 7 in w = x f 3 3 f 9 6 / ( f 3 2 f 9 3 ) w=xf_3^3f_9^6/(f_3^2f_9^3) w = x f 3 3 f 9 6 / ( f 3 2 f 9 3 ) satisfying R ( w ) ≡ E 6 ( 5 E − 3 ) ( m o d 81 ) R(w)\equiv E^6(5E-3)\pmod{81} R ( w ) ≡ E 6 ( 5 E − 3 ) ( mod 81 ) with E = 1 + w E=1+w E = 1 + w . This conjecture is equivalent to showing that certain coefficients vanish modulo 81 81 81 in the resulting eta-quotient expansion, a fact we have not yet been able to prove. We also observe unexpectedly high divisibility in several other progressions:
Conjecture 6.4. (a) A 3 ( 12 n + 7 ) ≡ A 3 ( 12 n + 11 ) ≡ 0 ( m o d 3 4 ) A_3(12n+7)\equiv A_3(12n+11)\equiv0\pmod{3^4} A 3 ( 12 n + 7 ) ≡ A 3 ( 12 n + 11 ) ≡ 0 ( mod 3 4 ) .
(b) A 4 ( 6 n + 5 ) ≡ 0 ( m o d 3 3 ) A_4(6n+5)\equiv0\pmod{3^3} A 4 ( 6 n + 5 ) ≡ 0 ( mod 3 3 ) and A 7 ( 12 n + 11 ) ≡ 0 ( m o d 3 3 ) A_7(12n+11)\equiv0\pmod{3^3} A 7 ( 12 n + 11 ) ≡ 0 ( mod 3 3 ) .
(c) A 8 ( 18 n + 15 ) ≡ 0 ( m o d 3 5 ) A_8(18n+15)\equiv0\pmod{3^5} A 8 ( 18 n + 15 ) ≡ 0 ( mod 3 5 ) .
For k = 3 k=3 k = 3 , Theorem 3.3 yields only 9 ∣ A 3 ( 3 n + 2 ) 9\mid A_3(3n+2) 9 ∣ A 3 ( 3 n + 2 ) , whereas part (a) states that divisibility jumps to 81 81 81 along both 12 n + 11 ⊂ 3 n + 2 12n+11\subset3n+2 12 n + 11 ⊂ 3 n + 2 and 12 n + 7 ⊂ 3 n + 1 12n+7\subset3n+1 12 n + 7 ⊂ 3 n + 1 . We conclude with an open question rather than a conjecture. By Corollary 3.6 and Proposition 5.2, the divisibility A k ( 3 n + 2 ) ≡ 0 ( m o d 3 ) A_k(3n+2)\equiv0\pmod{3} A k ( 3 n + 2 ) ≡ 0 ( mod 3 ) for k ≡ 1 ( m o d 3 ) k\equiv1\pmod{3} k ≡ 1 ( mod 3 ) is equivalent to certain properties of overpartition tuples and staircase tuples; however, a combinatorial explanation remains unknown. While the pd-rank introduced in [5 ] provides such an explanation for k = 1 k = 1 k = 1 , simply summing the component pd-ranks weighted by powers of ζ = e 2 π i / 3 \zeta= e^{2\pi i/3} ζ = e 2 π i /3 fails for k ≥ 2 k \ge2 k ≥ 2 .
Problem 6.5. For k ≡ 1 ( m o d 3 ) k \equiv1 \pmod{3} k ≡ 1 ( mod 3 ) with k ≥ 4 k \ge4 k ≥ 4 , find a natural statistic on k k k -tuples of designated partitions of size 3 n + 2 3n + 2 3 n + 2 whose values modulo 3 3 3 divide them into three classes of equal size, preferably through an explicit bijection on Young diagrams.
Solving this for k = 4 k = 4 k = 4 would already be of significant interest, as k = 4 k = 4 k = 4 is the first instance of the 3 3 3 -adic phenomenon in Theorem 3.3 where k k k is not divisible by 3 3 3 . Taken together, our results indicate that the arithmetic of k k k -tuples of designated partitions is governed by two key components: the 3 3 3 -adic structure F = Θ P ( u ) F = \Theta P(u) F = Θ P ( u ) , which explains the progression 3 n + 2 3n + 2 3 n + 2 and the classes 3 b ( 3 n + s ) 3^b(3n + s) 3 b ( 3 n + s ) for all k k k , and a 2 2 2 -adic structure along 4 n + 3 4n + 3 4 n + 3 that is currently understood only partially, as reflected in Conjectures 6.1 and 6.2. The conjectures presented here outline natural directions for future work. Ultimately, finding a crank for k k k -tuples and a combinatorial analogue of the bijection Δ \Delta Δ for odd parts would provide the combinatorial counterparts to these two algebraic structures.
References
Department of Computer Science, University of Torbat-e Jam, Torbat-e Jam, Iran
Email address: yaqubi@tjamaaс.ac.ir, daniel.yaqubi@yahoo.es
Department of Pure Mathematics, Ferdowsi University of Mashhad, Mashhad, Iran
Email address: mirzavaziri@um.ac.ir, mirzavaziri@gmail.com