Introduction
The partition rank reveals a remarkable coherence among several of Ramanujan’s third-order mock theta functions [27 , 5 ] . Replacing the underlying partitions by corners—plane partitions having one row and one column—leads to a closely parallel collection of generating functions.
2020 Mathematics Subject Classification. 11P81; 11P82; 11P83; 05A17; 05A19; 33D15.
Key words and phrases. corners, plane partitions, integer partitions, ranks, mock theta functions, partial theta functions, false theta functions, unimodal sequences, colored partitions, overpartitions, sign-reversing involutions, Ramanujan identities, congruences, rank differences, rank equidistribution, rank bias. We organize the paper according to this parallel, in the spirit of Ramanujan’s own index to the Lost Notebook. Analytically, the passage from a mock-theta rank series to the corresponding corner series keeps the finite rank kernel and replaces the quadratic power of q q q by the linear power that records the peak of the corner. Thus
n 2 ⟼ n , 2 n ( n + 1 ) ⟼ 2 n + 1. n^2 \longmapsto n,\qquad2n(n+1) \longmapsto2n+1. n 2 ⟼ n , 2 n ( n + 1 ) ⟼ 2 n + 1.
We use the descriptive phrase kernel-preserving linearization for this replacement. It is an organizing principle for the families considered here, not a claim that arbitrary analytic or modular properties are preserved.
We have chosen the title The Corners of Ramanujan because the analytic identities needed for the companion formulas are dispersed throughout Ramanujan’s Lost Notebook and acquire a unified corner interpretation here. The identities themselves are classical; their organization, the colored interpretations, and the new rank and arithmetic consequences are the focus of the present work.
Corners have appeared in a variety of mathematical contexts. The underlying stack models occur in work of Auluck [11 ] , Andrews [3 ] , Temperley [24 ] , and Wright [28 ] ; corners also arise as wedge characters in the work of Datta, Gaberdiel, Li, and Peng [18 ] , and as V-partitions in Stanley’s work [23 ] . The word “corner” appears to have been first used by Kreweras [21 ] ; his paper is in French, where the term is encoignure , and the OEIS uses the English translation “corner” [22 ] .
The ordinary corner numbers and their interpretation as unimodal sequences are classical. More importantly for the present paper, the two-variable series R c ( z ; q ) R_c(z;q) R c ( z ; q ) in (27) is, up to reversal of the rank sign, exactly the unimodal-rank generating function studied by Kim and Lovejoy [19 ] . They used a partial theta identity of Ramanujan to obtain rank differences and a congruence for a rank moment, and subsequently developed analogous identities for several special unimodal families [20 ] . Bridges and Bringmann studied Taylor coefficients of the same rank function through false Jacobi forms [13 ] ; Allen and Osburn obtained two-parameter Hecke–Appell expansions and mixed false-theta representations for generalized unimodal series [6 ] ; and Bringmann, Pandey, and van Ittersum developed related partial- and false-Eisenstein-type series and differential structures [17 ] . Burson gave a boxed 2-modular sign-reversing involution for one of Ramanujan’s false theta identities and a multivariate boxed m m m -modular extension [14 ] . We show that the instances ( m , r ) = ( 3 , 1 ) , ( 3 , 2 ) (m,r)=(3,1),(3,2) ( m , r ) = ( 3 , 1 ) , ( 3 , 2 ) and ( 2 , 1 ) (2,1) ( 2 , 1 ) give a common involutive realization of the rank-refined sparse corrections in the odd and three-dimensional corner identities. Root-of-unity specializations are a standard source of rank differences and cyclotomic equidistribution in partition theory; see, for example, Garvan [26 ] and Folsom, Males, and Rolen [25 ] . The specializations developed here concern the different tower-decorated series D ( z ; q ) \mathcal{D}(z;q) D ( z ; q ) introduced in Section 6.
For the odd family, Andrews’s convex compositions provide the earlier one-variable function [4 ] , while Bringmann and Lovejoy later introduced and analyzed the two-variable odd-unimodal rank series used below [16 Thm. 1.1] . Thus neither the definitions of the ordinary and odd unimodal-rank series nor the previously known rank identities are claimed as new here.
The paper has three layers. First, the known rank kernels and classical Lost Notebook identities are assembled into a single corner/mock-theta dictionary; this is primarily a synthesis and a change of viewpoint. Second, we give new colored-partition interpretations and exact two- and three-color relations whose errors are sparse theta series. Parallel positive decompositions separate the two-color companion into odd corners and staircase-decorated even corners, and the three-color companion into two copies of the restricted odd three-dimensional corners and a corresponding decorated even-corner family. Both errors are therefore differences of natural corner families. The two rank-refined correction terms are then placed in one boxed modular framework: Burson’s involutions cancel all nonfixed overpartition pairs, leaving exactly the exceptional quadratic families. Third, we develop new rank and arithmetic consequences: an oriented three-dimensional corner with a distinct tower, an exceptional parity specialization, quartic and cubic root-of-unity rank-difference formulas, exact rank equidistribution modulo 3 on 6 n + 5 6n+5 6 n + 5 , and an all-modulus classification of the resulting binary triangular support. This separation between recalled input, new interpretation, and new theorem will be maintained throughout the paper.
The arithmetic result has a particularly concise local formulation. With T j = j ( j + 1 ) / 2 T_j=j(j+1)/2 T j = j ( j + 1 ) /2 , define
R M = { T 2 a + 1 + T b ( m o d M ) : a , b ≥ 0 } . R_M=\{T_{2a+1}+T_b \pmod{M}:a,b\geq0\}. R M = { T 2 a + 1 + T b ( mod M ) : a , b ≥ 0 } .
Theorem 7.2 proves that a residue r r r is absent from R M R_M R M precisely when there is a prime power p α ∥ M p^\alpha\parallel M p α ∥ M with p ≡ 3 ( m o d 4 ) p\equiv3\pmod4 p ≡ 3 ( mod 4 ) for which v p ( 4 r + 1 ) v_p(4r+1) v p ( 4 r + 1 ) is odd and less than α \alpha α . This local description turns the sparse theta discrepancies into congruence families for the restricted three-dimensional corners and their colored companions. A separate cubic specialization of the side-rank series refines one of these divisibility statements: the three rank classes are exactly equidistributed at weights 6 n + 5 6n+5 6 n + 5 , while the rank difference on weights congruent to 2 modulo 6 is supported on a single quadratic sequence. This yields complete modulo 3 and modulo 12 dissections on the corresponding progressions.
The organization is as follows. § 2 reviews the two rank kernels, the relevant third-order mock theta functions, and the partial and false theta functions. § 3 introduces the parallel corner-rank series and displays the dictionary used in this paper. § 4 and § 5 treat ordinary and odd corners, respectively; the latter also gives the first positive corner decomposition and identifies the rank-refined sparse correction with two boxed 3-modular involutive series. § 6 introduces oriented three-dimensional corners, consisting of one row, one column, and one tower, gives the parallel three-color positive decomposition, and identifies the corresponding correction with the boxed 2-modular case. Section 7 states the support classification and the resulting congruence consequences. Sections 8, 9, and 10 give, separately and in the order of their statements, the proofs of the results in Sections 4, 5, and 6. Section 11 contains the proofs of the arithmetic results. The final section records further directions suggested by the construction.
Rank functions, mock theta functions, and Ramanujan's identities
For n ≥ 0 n \ge0 n ≥ 0 , write
( a ; q ) 0 = 1 , ( a ; q ) n = ∏ j = 0 n − 1 ( 1 − a q j ) , ( a ; q ) ∞ = ∏ j = 0 ∞ ( 1 − a q j ) . (a;q)_0=1,\qquad(a;q)_n=\prod_{j=0}^{n-1}(1-aq^j),\qquad(a;q)_\infty=\prod_{j=0}^{\infty}(1-aq^j). ( a ; q ) 0 = 1 , ( a ; q ) n = j = 0 ∏ n − 1 ( 1 − a q j ) , ( a ; q ) ∞ = j = 0 ∏ ∞ ( 1 − a q j ) .
and use the usual condensed notation for several parameters. For standard partition and q q q -series notation, see [2 ] . Throughout, ∣ q ∣ < 1 |q|<1 ∣ q ∣ < 1 when analytic convergence is relevant; otherwise all identities are formal power-series identities. When an intermediate calculation uses q 1 / 2 q^{1/2} q 1/2 , it is understood in the formal ring Z [ [ q 1 / 2 ] ] \mathbb{Z}[[q^{1/2}]] Z [[ q 1/2 ]] ; the final identities used below lie in Z [ [ q ] ] \mathbb{Z}[[q]] Z [[ q ]] . For a modulus M M M , a residue is called a square modulo M M M if it is congruent to x 2 x^2 x 2 for some integer x x x ; otherwise it is called a nonsquare modulo M M M .
Let N ( m , n ) N(m,n) N ( m , n ) denote the number of partitions of n n n with Dyson rank m m m [1 ] . The classical rank generating function is
R ( z ; q ) : = ∑ n ≥ 0 ∑ m ∈ Z N ( m , n ) z m q n = ∑ n ≥ 0 q n 2 ( z q ; q ) n ( z − 1 q ; q ) n . (1) R(z;q):=\sum_{n\geq0}\sum_{m\in\mathbb{Z}}N(m,n)z^m q^n=\sum_{n\geq0}\frac{q^{n^2}}{(zq;q)_n(z^{-1}q;q)_n}. \tag*{(1)} R ( z ; q ) := n ≥ 0 ∑ m ∈ Z ∑ N ( m , n ) z m q n = n ≥ 0 ∑ ( z q ; q ) n ( z − 1 q ; q ) n q n 2 . (1)
In particular,
R ( 1 ; q ) = 1 ( q ; q ) ∞ , R ( − 1 ; q ) = f 3 ( q ) , R ( i ; q ) = ϕ 3 ( q ) , (2) R(1;q)=\frac{1}{(q;q)_\infty},\qquad R(-1;q)=f_3(q),\qquad R(i;q)=\phi_3(q), \tag*{(2)} R ( 1 ; q ) = ( q ; q ) ∞ 1 , R ( − 1 ; q ) = f 3 ( q ) , R ( i ; q ) = ϕ 3 ( q ) , (2)
where
f 3 ( q ) : = ∑ n ≥ 0 q n 2 ( − q ; q ) n 2 , (3) f_3(q):=\sum_{n\geq0}\frac{q^{n^2}}{(-q;q)_n^2}, \tag*{(3)} f 3 ( q ) := n ≥ 0 ∑ ( − q ; q ) n 2 q n 2 , (3)
ϕ 3 ( q ) : = ∑ n ≥ 0 q n 2 ( − q 2 ; q 2 ) n . (4) \phi_3(q):=\sum_{n\geq0}\frac{q^{n^2}}{(-q^2;q^2)_n}. \tag*{(4)} ϕ 3 ( q ) := n ≥ 0 ∑ ( − q 2 ; q 2 ) n q n 2 . (4)
A second rank-type series is
S ( z ; q ) : = ∑ n ≥ 0 q 2 n 2 + 2 n ( z q ; q 2 ) n + 1 ( z − 1 q ; q 2 ) n + 1 . (5) S(z;q):=\sum_{n\geq0}\frac{q^{2n^2+2n}}{(zq;q^2)_{n+1}(z^{-1}q;q^2)_{n+1}}. \tag*{(5)} S ( z ; q ) := n ≥ 0 ∑ ( z q ; q 2 ) n + 1 ( z − 1 q ; q 2 ) n + 1 q 2 n 2 + 2 n . (5)
It gives
S ( 1 ; q ) = ω 3 ( q ) , S ( i ; q ) = ν 3 ( q 2 ) , (6) S(1;q)=\omega_3(q),\qquad S(i;q)=\nu_3(q^2), \tag*{(6)} S ( 1 ; q ) = ω 3 ( q ) , S ( i ; q ) = ν 3 ( q 2 ) , (6)
with
ω 3 ( q ) : = ∑ n ≥ 0 q 2 n ( n + 1 ) ( q ; q 2 ) n + 1 , (7) \omega_{3}(q) := \sum_{n\geq0} \frac{q^{2n(n+1)}}{(q;q^{2})_{n+1}}, \tag*{(7)} ω 3 ( q ) := n ≥ 0 ∑ ( q ; q 2 ) n + 1 q 2 n ( n + 1 ) , (7)
ν 3 ( q ) : = ∑ n ≥ 0 q n ( n + 1 ) ( − q ; q 2 ) n + 1 . (8) \nu_{3}(q) := \sum_{n\geq0} \frac{q^{n(n+1)}}{(-q;q^{2})_{n+1}}. \tag*{(8)} ν 3 ( q ) := n ≥ 0 ∑ ( − q ; q 2 ) n + 1 q n ( n + 1 ) . (8)
We shall also use
ψ 3 ( q ) : = ∑ n ≥ 1 q n 2 ( q ; q 2 ) n . (9) \psi_{3}(q) := \sum_{n\geq1} \frac{q^{n^{2}}}{(q;q^{2})_{n}}. \tag*{(9)} ψ 3 ( q ) := n ≥ 1 ∑ ( q ; q 2 ) n q n 2 . (9)
The terminology surrounding unilateral theta-type series is not completely uniform. We shall call
f Θ ( q ) : = ∑ n ≥ 0 ( − 1 ) n q n ( n + 1 ) / 2 (10) f\Theta(q) := \sum_{n\geq0} (-1)^{n}q^{n(n+1)/2} \tag*{(10)} f Θ ( q ) := n ≥ 0 ∑ ( − 1 ) n q n ( n + 1 ) /2 (10)
a partial theta series. We also define
f E ( q ) : = ∑ n ≥ 0 q n ( 3 n + 1 ) / 2 ( 1 − q 2 n + 1 ) , (11) f\mathrm{E}(q) := \sum_{n\geq0} q^{n(3n+1)/2}(1-q^{2n+1}), \tag*{(11)} f E ( q ) := n ≥ 0 ∑ q n ( 3 n + 1 ) /2 ( 1 − q 2 n + 1 ) , (11)
f O ( q ) : = ∑ n ≥ 0 q 3 n 2 + 2 n ( 1 − q 2 n + 1 ) . (12) f\mathrm{O}(q) := \sum_{n\geq0} q^{3n^{2}+2n}(1-q^{2n+1}). \tag*{(12)} f O ( q ) := n ≥ 0 ∑ q 3 n 2 + 2 n ( 1 − q 2 n + 1 ) . (12)
After reindexing the second terms, these have the signed bilateral forms
f E ( q ) = ∑ n ∈ Z sgn ( n + 1 2 ) q n ( 3 n + 1 ) / 2 , f\mathrm{E}(q) = \sum_{n\in\mathbb{Z}} \operatorname{sgn}\left(n+\frac{1}{2}\right)q^{n(3n+1)/2}, f E ( q ) = n ∈ Z ∑ sgn ( n + 2 1 ) q n ( 3 n + 1 ) /2 ,
f O ( q ) = ∑ n ∈ Z sgn ( n + 1 2 ) q n ( 3 n + 2 ) , f\mathrm{O}(q) = \sum_{n\in\mathbb{Z}} \operatorname{sgn}\left(n+\frac{1}{2}\right)q^{n(3n+2)}, f O ( q ) = n ∈ Z ∑ sgn ( n + 2 1 ) q n ( 3 n + 2 ) ,
so we refer to them as false theta series in the standard signed-theta sense. The subscripts suggest triangular, Euler-pentagonal, and odd-pentagonal support. We use “partial and false theta” collectively when discussing the three series.
We shall repeatedly use the q q q -binomial theorem
∑ n ≥ 0 ( a ; q ) n z n ( q ; q ) n = ( a z ; q ) ∞ ( z ; q ) ∞ , (13) \sum_{n\geq0} \frac{(a;q)_{n}z^{n}}{(q;q)_{n}} = \frac{(az;q)_{\infty}}{(z;q)_{\infty}}, \tag*{(13)} n ≥ 0 ∑ ( q ; q ) n ( a ; q ) n z n = ( z ; q ) ∞ ( a z ; q ) ∞ , (13)
Euler’s pentagonal number theorem
( q ; q ) ∞ = ∑ k = − ∞ ∞ ( − 1 ) k q k ( 3 k + 1 ) / 2 , (14) (q;q)_{\infty} = \sum_{k=-\infty}^{\infty} (-1)^{k}q^{k(3k+1)/2}, \tag*{(14)} ( q ; q ) ∞ = k = − ∞ ∑ ∞ ( − 1 ) k q k ( 3 k + 1 ) /2 , (14)
Gauss’s identity
G ( q ) : = ∑ n ≥ 0 q n ( n + 1 ) / 2 = ( q 2 ; q 2 ) ∞ ( q ; q 2 ) ∞ . (15) G(q) := \sum_{n\geq0}q^{n(n+1)/2} = \frac{(q^{2};q^{2})_{\infty}}{(q;q^{2})_{\infty}}. \tag*{(15)} G ( q ) := n ≥ 0 ∑ q n ( n + 1 ) /2 = ( q ; q 2 ) ∞ ( q 2 ; q 2 ) ∞ . (15)
Jacobi’s triple product
∑ n ∈ Z ( − 1 ) n x n q n ( n − 1 ) / 2 = ( x ; q ) ∞ ( q / x ; q ) ∞ ( q ; q ) ∞ , (16) \sum_{n\in\mathbb{Z}}(-1)^n x^n q^{n(n-1)/2}=(x;q)_\infty(q/x;q)_\infty(q;q)_\infty, \tag*{(16)} n ∈ Z ∑ ( − 1 ) n x n q n ( n − 1 ) /2 = ( x ; q ) ∞ ( q / x ; q ) ∞ ( q ; q ) ∞ , (16)
and Heine’s transformation
∑ n ≥ 0 ( a ; q ) n ( b ; q ) n z n ( q ; q ) n ( c ; q ) n = ( b ; q ) ∞ ( a z ; q ) ∞ ( c ; q ) ∞ ( z ; q ) ∞ ∑ n ≥ 0 ( c / b ; q ) n ( z ; q ) n b n ( q ; q ) n ( a z ; q ) n . (17) \sum_{n\geq0}\frac{(a;q)_n(b;q)_n z^n}{(q;q)_n(c;q)_n}
=\frac{(b;q)_\infty(az;q)_\infty}{(c;q)_\infty(z;q)_\infty}
\sum_{n\geq0}\frac{(c/b;q)_n(z;q)_n b^n}{(q;q)_n(az;q)_n}. \tag*{(17)} n ≥ 0 ∑ ( q ; q ) n ( c ; q ) n ( a ; q ) n ( b ; q ) n z n = ( c ; q ) ∞ ( z ; q ) ∞ ( b ; q ) ∞ ( a z ; q ) ∞ n ≥ 0 ∑ ( q ; q ) n ( a z ; q ) n ( c / b ; q ) n ( z ; q ) n b n . (17)
Heine’s transformation was also discovered independently by Ramanujan; see Berndt [12 Ch. 16, Entry 6] .
We shall also use the following corollary of the Rogers–Fine identity [7 p. 223, Eq. (9.1.1)] :
∑ n ≥ 0 ( a ; q ) n a n = ∑ n ≥ 0 ( − 1 ) n q n ( 3 n − 1 ) / 2 a 3 n ( 1 + a q n ) . (18) \sum_{n\geq0}(a;q)_n a^n=\sum_{n\geq0}(-1)^n q^{n(3n-1)/2}a^{3n}(1+aq^n). \tag*{(18)} n ≥ 0 ∑ ( a ; q ) n a n = n ≥ 0 ∑ ( − 1 ) n q n ( 3 n − 1 ) /2 a 3 n ( 1 + a q n ) . (18)
The same formula will be applied below with the base q q q replaced by q 4 q^4 q 4 .
Three identities from Part I of Ramanujan’s Lost Notebook are [7 Entries 9.3.3, 9.4.2, and 9.5.3]
∑ n ≥ 0 ( q ; q 2 ) n q n ( − q ; q 2 ) n + 1 = ∑ n ≥ 0 ( − 1 ) n q 2 n ( n + 1 ) = f Θ ( q 4 ) , (19) \sum_{n\geq0}\frac{(q;q^2)_n q^n}{(-q;q^2)_{n+1}}
=\sum_{n\geq0}(-1)^n q^{2n(n+1)}
=f\Theta(q^4), \tag*{(19)} n ≥ 0 ∑ ( − q ; q 2 ) n + 1 ( q ; q 2 ) n q n = n ≥ 0 ∑ ( − 1 ) n q 2 n ( n + 1 ) = f Θ ( q 4 ) , (19)
∑ n ≥ 0 ( − 1 ) n q n ( n + 1 ) / 2 ( − q ; q ) n = f E ( q ) , (20) \sum_{n\geq0}\frac{(-1)^n q^{n(n+1)/2}}{(-q;q)_n}
=fE(q), \tag*{(20)} n ≥ 0 ∑ ( − q ; q ) n ( − 1 ) n q n ( n + 1 ) /2 = f E ( q ) , (20)
and
∑ n ≥ 0 q n ( − q ; q 2 ) n + 1 = ∑ n ≥ 0 ( − 1 ) n q 6 n 2 + 4 n ( 1 + q 4 n + 2 ) = f O ( − q 2 ) . (21) \sum_{n\geq0}\frac{q^n}{(-q;q^2)_{n+1}}
=\sum_{n\geq0}(-1)^n q^{6n^2+4n}(1+q^{4n+2})
=fO(-q^2). \tag*{(21)} n ≥ 0 ∑ ( − q ; q 2 ) n + 1 q n = n ≥ 0 ∑ ( − 1 ) n q 6 n 2 + 4 n ( 1 + q 4 n + 2 ) = f O ( − q 2 ) . (21)
We also use Burson’s boxed modular refinement of (2.19) [14 Thm. 5.2] .
Recall that an overpartition is a partition in which the first occurrence of each distinct part may be overlined [15 ] . For integers m ≥ 2 m\geq2 m ≥ 2 and 1 ≤ r < m 1\leq r<m 1 ≤ r < m , let B m , r ( N ) \mathcal{B}_{m,r}(N) B m , r ( N ) be the set of pairs ( k , π ‾ ) (k,\overline{\pi}) ( k , π ) such that k ≥ 0 k\geq0 k ≥ 0 , π ‾ \overline{\pi} π is an overpartition into parts congruent to r r r modulo m m m and not exceeding m k + r mk+r mk + r , every overlined part is strictly less than m k + r mk+r mk + r , and
r k + ∣ π ‾ ∣ = N . rk+|\overline{\pi}|=N. r k + ∣ π ∣ = N .
Put ν ( π ‾ ) \nu(\overline{\pi}) ν ( π ) for the number of parts, counted with multiplicity, and declare B m , r ( N ) = ∅ \mathcal{B}_{m,r}(N)=\varnothing B m , r ( N ) = ∅ when N < 0 N<0 N < 0 . The corresponding signed bivariate series is
B m , r ( u , q ) : = ∑ N ≥ 0 ∑ ( k , π ‾ ) ∈ B m , r ( N ) ( − 1 ) ν ( π ‾ ) u k + ν ( π ‾ ) q N = ∑ k ≥ 0 ( u q r ; q m ) k u k q r k ( − u q r ; q m ) k + 1 = ∑ j ≥ 0 ( − 1 ) j u 2 j q j ( m j + 2 r ) . (22) \begin{aligned}
\mathcal{B}_{m,r}(u,q):=
\sum_{N\geq0}\sum_{(k,\overline{\pi})\in\mathcal{B}_{m,r}(N)}
(-1)^{\nu(\overline{\pi})}u^{k+\nu(\overline{\pi})}q^N \\
&=\sum_{k\geq0}\frac{(u q^r;q^m)_k u^k q^{rk}}{(-u q^r;q^m)_{k+1}} \\
&=\sum_{j\geq0}(-1)^j u^{2j}q^{j(mj+2r)}. \tag*{(22)}
\end{aligned} B m , r ( u , q ) := N ≥ 0 ∑ ( k , π ) ∈ B m , r ( N ) ∑ ( − 1 ) ν ( π ) u k + ν ( π ) q N = k ≥ 0 ∑ ( − u q r ; q m ) k + 1 ( u q r ; q m ) k u k q r k = j ≥ 0 ∑ ( − 1 ) j u 2 j q j ( mj + 2 r ) . (22)
For fixed k k k , the product has the stated overpartition interpretation: a nonmaximal permitted part s s s contributes ( 1 − u q s ) / ( 1 + u q s ) (1-uq^s)/(1+uq^s) ( 1 − u q s ) / ( 1 + u q s ) , the maximal part m k + r mk+r mk + r contributes 1 / ( 1 + u q m k + r ) 1/(1+uq^{mk+r}) 1/ ( 1 + u q mk + r ) , and the boxed row contributes u k q r k u^kq^{rk} u k q r k . Recall that an involution ι \iota ι on a signed set satisfies ι 2 = i d \iota^2=\mathrm{id} ι 2 = id . An object X X X is called a fixed point when ι ( X ) = X \iota(X)=X ι ( X ) = X . Every nonfixed object belongs to a two-cycle { X , ι ( X ) } \{X,\iota(X)\} { X , ι ( X )} ; if ι \iota ι reverses the sign, the two members of that cycle cancel in the signed generating function. Thus only the fixed points contribute to the final signed sum.
Burson’s boxed m m m -modular involution preserves both N N N and k + ν ( π ‾ ) k+\nu(\overline{\pi}) k + ν ( π ) and reverses the sign on every nonfixed pair. Its fixed points are the pairs
( j , ( m j + r ) j ) (23) (j,(mj+r)^j) \tag*{(23)} ( j , ( mj + r ) j ) (23)
where ( s ) j (s)^j ( s ) j denotes the partition consisting of j j j copies of the part s s s . These fixed points have weight ( − 1 ) j u 2 j q j ( m j + 2 r ) (-1)^j u^{2j}q^{j(mj+2r)} ( − 1 ) j u 2 j q j ( mj + 2 r ) .
The next two identities are from Part II [8 Ch. 6] . First,
∑ n ≥ 0 q n ( − a q ; q ) n ( − b q ; q ) n = 1 ( − a q ; q ) ∞ ( − b q ; q ) ∞ ∑ m ≥ 0 ( − 1 ) m + 1 a − m − 1 b m q ( m + 1 2 ) + ( 1 + 1 a ) ∑ m ≥ 0 ( − 1 ) m a − m b m q ( m + 1 2 ) ( − b q ; q ) m . (24) \sum_{n\geq0}\frac{q^n}{(-aq;q)_n(-bq;q)_n}
=
\frac{1}{(-aq;q)_\infty(-bq;q)_\infty}
\sum_{m\geq0}(-1)^{m+1}a^{-m-1}b^m q^{\binom{m+1}{2}}
+
\left(1+\frac{1}{a}\right)\sum_{m\geq0}\frac{(-1)^m a^{-m}b^m q^{\binom{m+1}{2}}}{(-bq;q)_m}. \tag*{(24)} n ≥ 0 ∑ ( − a q ; q ) n ( − b q ; q ) n q n = ( − a q ; q ) ∞ ( − b q ; q ) ∞ 1 m ≥ 0 ∑ ( − 1 ) m + 1 a − m − 1 b m q ( 2 m + 1 ) + ( 1 + a 1 ) m ≥ 0 ∑ ( − b q ; q ) m ( − 1 ) m a − m b m q ( 2 m + 1 ) . (24)
Second,
( 1 + 1 a ) ∑ n ≥ 0 ( − a ; q ) 2 n q 2 n + 1 ( a q ; q 2 ) n + 1 ( q / a ; q 2 ) n + 1 = 1 ( q ; q 2 ) ∞ ( a q ; q 2 ) ∞ ( q / a ; q 2 ) ∞ ∑ n ≥ 0 ( − a ) n q ( n + 1 2 ) − ∑ n ≥ 0 ( − a ) n q n 2 + n . (25) \left(1+\frac{1}{a}\right)\sum_{n\geq0}\frac{(-a;q)_{2n}q^{2n+1}}{(aq;q^2)_{n+1}(q/a;q^2)_{n+1}}
=
\frac{1}{(q;q^2)_\infty(aq;q^2)_\infty(q/a;q^2)_\infty}
\sum_{n\geq0}(-a)^nq^{\binom{n+1}{2}}
-
\sum_{n\geq0}(-a)^nq^{n^2+n}. \tag*{(25)} ( 1 + a 1 ) n ≥ 0 ∑ ( a q ; q 2 ) n + 1 ( q / a ; q 2 ) n + 1 ( − a ; q ) 2 n q 2 n + 1 = ( q ; q 2 ) ∞ ( a q ; q 2 ) ∞ ( q / a ; q 2 ) ∞ 1 n ≥ 0 ∑ ( − a ) n q ( 2 n + 1 ) − n ≥ 0 ∑ ( − a ) n q n 2 + n . (25)
We shall also need the following companion identity from Part III of Ramanujan’s Lost Notebook [9 p. 60, Eq. (8.25)] :
∑ n ≥ 1 q n ( q ; q 2 ) n = q ∑ n ≥ 0 q 2 n ( n + 1 ) ( q ; q 2 ) n + 1 2 . (26) \sum_{n\geq1}\frac{q^n}{(q;q^2)_n}
=
q\sum_{n\geq0}\frac{q^{2n(n+1)}}{(q;q^2)_{n+1}^2}. \tag*{(26)} n ≥ 1 ∑ ( q ; q 2 ) n q n = q n ≥ 0 ∑ ( q ; q 2 ) n + 1 2 q 2 n ( n + 1 ) . (26)
Corner ranks and the mock-theta dictionary
A corner is a plane partition whose Ferrers diagram has one row and one column. Equivalently, it consists of a central part m m m , a row partition, and a column partition, with every row and column part at most m m m . If the row has r r r parts and the column has s s s parts, the corner rank is r − s r-s r − s . Let C ( r , n ) C(r,n) C ( r , n ) denote the number of corners of weight n n n and corner rank r r r . Thus
R c ( z ; q ) : = ∑ n ≥ 0 ∑ r ∈ Z C ( r , n ) z r q n = ∑ m ≥ 0 q m ( z q ; q ) m ( z − 1 q ; q ) m . (27) R_c(z;q):=\sum_{n\geq0}\sum_{r\in\mathbb{Z}}C(r,n)z^r q^n=\sum_{m\geq0}\frac{q^m}{(zq;q)_m(z^{-1}q;q)_m}. \tag*{(27)} R c ( z ; q ) := n ≥ 0 ∑ r ∈ Z ∑ C ( r , n ) z r q n = m ≥ 0 ∑ ( z q ; q ) m ( z − 1 q ; q ) m q m . (27)
Up to reversing the sign convention for the rank, (27) is exactly the unimodal-rank generating function U ( z ; q ) U(z;q) U ( z ; q ) of Kim and Lovejoy [19 Eq. (2.2)] . The reversal is immaterial because the series is invariant under z ↦ z − 1 z\mapsto z^{-1} z ↦ z − 1 .
If every part is required to be odd, the corresponding series is
S c ( z ; q ) : = ∑ m ≥ 0 q 2 m + 1 ( z q ; q 2 ) m + 1 ( z − 1 q ; q 2 ) m + 1 . (28) S_c(z;q):=\sum_{m\geq0}\frac{q^{2m+1}}{(zq;q^2)_{m+1}(z^{-1}q;q^2)_{m+1}}. \tag*{(28)} S c ( z ; q ) := m ≥ 0 ∑ ( z q ; q 2 ) m + 1 ( z − 1 q ; q 2 ) m + 1 q 2 m + 1 . (28)
This is the odd-unimodal rank series of Bringmann and Lovejoy [16 Thm. 1.1] . The denominators in (27) and (28) are exactly those of (1) and (5) ; only the quadratic core weights have been linearized.
The resulting dictionary is summarized in Table 1 . The entries in the last column are developed in the following sections.
Table 1. Third-order rank specializations and their corner companions.
The table is a guide rather than a claim that all analytic properties of a mock theta function survive linearization. Its force is that one fixed rank kernel produces both a mock-theta specialization and a corner series, while Ramanujan’s identities evaluate the latter through partial and false theta functions.
The present dictionary does not include the remaining classical third-order functions χ 3 \chi_3 χ 3 and ρ 3 \rho_3 ρ 3 . No claim of a complete dictionary for every third-order mock theta function is therefore intended.
Ordinary corner companions
We state the results first and give their proofs together in § 8 .
Let c ( n ) c(n) c ( n ) denote the number of corners of n n n . Then
C ( q ) : = ∑ n ≥ 0 c ( n ) q n = R c ( 1 ; q ) = ∑ m ≥ 0 q m ( q ; q ) m 2 . (29) C(q):=\sum_{n\geq0}c(n)q^n=R_c(1;q)=\sum_{m\geq0}\frac{q^m}{(q;q)_m^2}. \tag*{(29)} C ( q ) := n ≥ 0 ∑ c ( n ) q n = R c ( 1 ; q ) = m ≥ 0 ∑ ( q ; q ) m 2 q m . (29)
For example, c ( 4 ) = 12 c(4)=12 c ( 4 ) = 12 ; the twelve objects are
4 , 31 , 3 1 , 22 , 2 2 , 211 , 2 1 1 , 2 1 , 1111 , 1 1 1 1 , 1 1 1 1 , 1 1 1 1 . 4,\ 31,\ \frac{3}{1},\ 22,\ \frac{2}{2},\ 211,\ \frac{2\ 1}{1},\ \frac{2}{1},\ 1111,\ \frac{1\ 1\ 1}{1},\ \frac{1\ 1}{1\ 1},\ \frac{1}{1\ 1\ 1}. 4 , 31 , 1 3 , 22 , 2 2 , 211 , 1 2 1 , 1 2 , 1111 , 1 1 1 1 , 1 1 1 1 , 1 1 1 1 .
The same objects are ordinary unimodal sequences with distinguished peak. The following is Ramanujan’s partial-theta identity in the form used by Kim and Lovejoy.
Proposition 4.1. We have
C ( q ) = f Θ ( q ) ( q ; q ) ∞ 2 (30) C(q)=\frac{f_{\Theta}(q)}{(q;q)_{\infty}^{2}} \tag*{(30)} C ( q ) = ( q ; q ) ∞ 2 f Θ ( q ) (30)
At the level of unimodal sequences, this is the specialization of the Ramanujan partial-theta identity used by Kim and Lovejoy [19 ] . We record it in corner language because it is the starting point for the new colored interpretation below. We are not aware of a comparably direct bijective proof of (30) .
Theorem 4.2. For n ≥ 0 n\geq0 n ≥ 0 , the following quantities are equal:
c ( n ) c(n) c ( n ) ;
the number of two-color partitions of n + 1 n+1 n + 1 in which the largest part occurs exactly once, and that occurrence is red;
the number of two-color partitions of n n n , with parts colored red and blue, in which the smallest positive integer absent as a blue part is odd.
Proposition 4.3. The number c ( n ) c(n) c ( n ) is odd if and only if n n n is a generalized pentagonal number, that is,
n = k ( 3 k ± 1 ) 2 n=\frac{k(3k\pm1)}{2} n = 2 k ( 3 k ± 1 ) for some $k\geq 0.
Corollary 4.4. Let M ≥ 2 M\geq2 M ≥ 2 and 0 ≤ r < M 0\leq r<M 0 ≤ r < M . If 24 r + 1 24r+1 24 r + 1 is a nonsquare modulo M M M , then
c ( M n + r ) ≡ 0 ( m o d 2 ) ( n ≥ 0 ) . c(Mn+r)\equiv0\pmod{2}\qquad(n\geq0). c ( M n + r ) ≡ 0 ( mod 2 ) ( n ≥ 0 ) .
The specialization z = − 1 z=-1 z = − 1 next supplies the corner companion of f 3 f_3 f 3 :
C f ( q ) : = R c ( − 1 ; q ) = ∑ m ≥ 0 q m ( − q ; q ) m 2 . (31) C_f(q):=R_c(-1;q)=\sum_{m\geq0}\frac{q^m}{(-q;q)_m^2}. \tag*{(31)} C f ( q ) := R c ( − 1 ; q ) = m ≥ 0 ∑ ( − q ; q ) m 2 q m . (31)
Its coefficient of q n q^n q n is the number of corners of n n n with even rank minus the number with odd rank.
Proposition 4.5. We have
C f ( q ) = − f Θ ( q ) ( − q ; q ) ∞ 2 + 2 f E ( q ) . (32) C_f(q)=-\frac{f_{\Theta}(q)}{(-q;q)_{\infty}^{2}}+2f_E(q). \tag*{(32)} C f ( q ) = − ( − q ; q ) ∞ 2 f Θ ( q ) + 2 f E ( q ) . (32)
Finally, for z = i z=i z = i we obtain the companion of ϕ 3 \phi_3 ϕ 3 :
C ϕ ( q ) : = R c ( i ; q ) = ∑ m ≥ 0 q m ( − q 2 ; q 2 ) m . (33) C_{\phi}(q) := R_c(i;q) = \sum_{m\geq0} \frac{q^m}{(-q^2;q^2)_m}. \tag*{(33)} C ϕ ( q ) := R c ( i ; q ) = m ≥ 0 ∑ ( − q 2 ; q 2 ) m q m . (33)
Proposition 4.6. We have
C ϕ ( q ) = ∑ m ≥ 0 ( q ; q 4 ) m q m + q 2 ∑ m ≥ 0 ( q 3 ; q 4 ) m q 3 m (34) C_{\phi}(q) = \sum_{m\geq0} (q;q^4)_m q^m + q^2 \sum_{m\geq0} (q^3;q^4)_m q^{3m} \tag*{(34)} C ϕ ( q ) = m ≥ 0 ∑ ( q ; q 4 ) m q m + q 2 m ≥ 0 ∑ ( q 3 ; q 4 ) m q 3 m (34) = f E ( − q ) . (35) = \mathrm{fE}(-q). \tag*{(35)} = fE ( − q ) . (35)
The two preceding propositions are direct evaluations of the indicated root-of-unity specializations by the classical identities in Section 2. Their role is to complete the dictionary; no novelty is claimed for the underlying q q q -series transformations.
Odd corner companions
We state the results first and give their proofs together in Section 9.
Let c o ( n ) c_o(n) c o ( n ) denote the number of nonempty corners of n n n for which the peak, row parts, and column parts are all odd. Then
C ω ( q ) : = ∑ n ≥ 0 c o ( n ) q n = S c ( 1 ; q ) = ∑ m ≥ 0 q 2 m + 1 ( q ; q 2 ) m + 1 2 . (36) C_{\omega}(q) := \sum_{n\geq0} c_o(n)q^n = S_c(1;q) = \sum_{m\geq0} \frac{q^{2m+1}}{(q;q^2)_{m+1}^2}. \tag*{(36)} C ω ( q ) := n ≥ 0 ∑ c o ( n ) q n = S c ( 1 ; q ) = m ≥ 0 ∑ ( q ; q 2 ) m + 1 2 q 2 m + 1 . (36)
For example, c o ( 5 ) = 9 c_o(5)=9 c o ( 5 ) = 9 :
5 , 311 , 3 1 , 3 1 1 , 11111 , 1 1 1 1 1 , 1 1 1 1 1 , 1 1 1 1 1 , 1 1 1 1 1 5,\ 311,\ \frac{3}{1},\ \frac{3}{1\,1},\ 11111,\ \begin{matrix}1&1&1&1\\1\end{matrix},\ \begin{matrix}1&1&1\\1\\1\end{matrix},\ \begin{matrix}1&1\\1\\1\\1\end{matrix},\ \begin{matrix}1\\1\\1\\1\\1\end{matrix} 5 , 311 , 1 3 , 1 1 3 , 11111 , 1 1 1 1 1 , 1 1 1 1 1 , 1 1 1 1 1 , 1 1 1 1 1
After increasing the peak by 2, this is Andrews’s one-variable odd convex-composition function [4 ] . Bringmann and Lovejoy later developed the two-variable odd-unimodal rank series
O ( z ; q ) = S c ( z ; q ) = ∑ m ≥ 0 q 2 m + 1 ( z q ; q 2 ) m + 1 ( z − 1 q ; q 2 ) m + 1 . (37) O(z;q) = S_c(z;q) = \sum_{m\geq0} \frac{q^{2m+1}}{(zq;q^2)_{m+1}(z^{-1}q;q^2)_{m+1}}. \tag*{(37)} O ( z ; q ) = S c ( z ; q ) = m ≥ 0 ∑ ( z q ; q 2 ) m + 1 ( z − 1 q ; q 2 ) m + 1 q 2 m + 1 . (37)
Their Theorem 1.1 gives
O ( z ; q ) = ∑ j ≥ 0 ( − 1 ) j + 1 z 3 j + 1 q 3 j 2 + 2 j ( 1 + z q 2 j + 1 ) + 1 ( z q ; q 2 ) ∞ ( z − 1 q ; q 2 ) ∞ ∑ j ≥ 0 ( − 1 ) j z 2 j + 1 q j ( j + 1 ) . (38) \begin{aligned}
O(z;q) &= \sum_{j\geq0} (-1)^{j+1}z^{3j+1}q^{3j^2+2j}(1+zq^{2j+1}) \\
&\quad+ \frac{1}{(zq;q^2)_{\infty}(z^{-1}q;q^2)_{\infty}}
\sum_{j\geq0}(-1)^j z^{2j+1}q^{j(j+1)}. \tag*{(38)}
\end{aligned} O ( z ; q ) = j ≥ 0 ∑ ( − 1 ) j + 1 z 3 j + 1 q 3 j 2 + 2 j ( 1 + z q 2 j + 1 ) + ( z q ; q 2 ) ∞ ( z − 1 q ; q 2 ) ∞ 1 j ≥ 0 ∑ ( − 1 ) j z 2 j + 1 q j ( j + 1 ) . (38)
The first line of (5.3) has a common boxed modular interpretation. The square reparametrization z = w 2 z=w^2 z = w 2 avoids fractional powers of the rank variable. The precise boxed 3-modular identity is as follows.
Proposition 5.1. For a formal variable w w w ,
∑ j ≥ 0 ( − 1 ) j + 1 w 6 j + 2 q 3 j 2 + 2 j ( 1 + w 2 q 2 j + 1 ) = − w 2 B 3 , 1 ( w 3 , q ) − w 4 q B 3 , 2 ( w 3 , q ) . (39) \sum_{j\geq0}(-1)^{j+1}w^{6j+2}q^{3j^2+2j}\left(1+w^2q^{2j+1}\right)=-w^2\mathcal{B}_{3,1}(w^3,q)-w^4q\mathcal{B}_{3,2}(w^3,q). \tag*{(39)} j ≥ 0 ∑ ( − 1 ) j + 1 w 6 j + 2 q 3 j 2 + 2 j ( 1 + w 2 q 2 j + 1 ) = − w 2 B 3 , 1 ( w 3 , q ) − w 4 q B 3 , 2 ( w 3 , q ) . (39) Thus the complete rank-refined sparse term in (38) , after z = w 2 z=w^2 z = w 2 , is governed by two instances of the single boxed modular involution (2.22).
From (5.1) and (5.2)–(5.3), applied with z = 1 z=1 z = 1 , we obtain
C ω ( q ) = − ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) + 1 ( q ; q 2 ) ∞ 2 ∑ j ≥ 0 ( − 1 ) j q j ( j + 1 ) . (40) C_{\omega}(q)=-\sum_{j\geq0}(-1)^jq^{3j^2+2j}\left(1+q^{2j+1}\right)+\frac{1}{(q;q^2)_\infty^2}\sum_{j\geq0}(-1)^jq^{j(j+1)}. \tag*{(40)} C ω ( q ) = − j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) + ( q ; q 2 ) ∞ 2 1 j ≥ 0 ∑ ( − 1 ) j q j ( j + 1 ) . (40)
Thus the same denominator that defines ω 3 ( q ) \omega_3(q) ω 3 ( q ) now carries the linear peak weight 2 m + 1 2m+1 2 m + 1 rather than the quadratic weight 2 m ( m + 1 ) 2m(m+1) 2 m ( m + 1 ) .
The simple largest-part interpretation survives in the odd setting.
Theorem 5.2. For n ≥ 0 n\geq0 n ≥ 0 , c o ( n ) c_o(n) c o ( n ) equals the number of two-color partitions of n + 2 n+2 n + 2 into odd parts in which the largest part occurs exactly once, and that occurrence is red.
We next define the smallest-part companion needed for the exact discrepancy. Let p 2 , d o ( n ) p_{2,do}(n) p 2 , d o ( n ) count two-color partitions of n n n with smallest part size s s s such that at least one smallest part is red, the red part of size s s s occurs once or twice and all larger red parts are distinct, while the blue parts are odd and at least 2 s − 1 2s-1 2 s − 1 . Put p 2 , d o ( 0 ) = 0 p_{2,do}(0)=0 p 2 , d o ( 0 ) = 0 . Then
∑ n ≥ 0 p 2 , d o ( n ) q n = ∑ s ≥ 1 q s ( − q s ; q ) ∞ ( q 2 s − 1 ; q 2 ) ∞ . (41) \sum_{n\geq0}p_{2,do}(n)q^n=\sum_{s\geq1}q^s\frac{(-q^s;q)_\infty}{(q^{2s-1};q^2)_\infty}. \tag*{(41)} n ≥ 0 ∑ p 2 , d o ( n ) q n = s ≥ 1 ∑ q s ( q 2 s − 1 ; q 2 ) ∞ ( − q s ; q ) ∞ . (41)
The definition is an analytic companion: fixing the smallest part s s s produces exactly the product needed in the transformation leading to Theorem 5.4.
For a ≥ 0 a\geq0 a ≥ 0 , let
δ a = ( a , a − 1 , … , 2 , 1 ) , ∣ δ a ∣ = T a = a ( a + 1 ) 2 , \delta_a=(a,a-1,\ldots,2,1),\qquad|\delta_a|=T_a=\frac{a(a+1)}{2}, δ a = ( a , a − 1 , … , 2 , 1 ) , ∣ δ a ∣ = T a = 2 a ( a + 1 ) ,
with δ 0 \delta_0 δ 0 empty. If Γ \Gamma Γ is an ordinary corner, write 2 Γ 2\Gamma 2Γ for the even corner obtained by doubling every entry. A staircase-decorated even corner is a triple
( 2 Γ , δ a , δ b ) (2\Gamma,\delta_a,\delta_b) ( 2Γ , δ a , δ b )
of weight 2 ∣ Γ ∣ + T a + T b 2|\Gamma|+T_a+T_b 2∣Γ∣ + T a + T b , with the two staircases regarded as independent decorations of the labeled row and column arms. Let e △ ( N ) e_{\triangle}(N) e △ ( N ) denote the number of such triples of weight N N N . By (29) and (15) ,
E △ ( q ) : = ∑ N ≥ 0 e △ ( N ) q N = ( ∑ a ≥ 0 q T a ) 2 ( ∑ k ≥ 0 q 2 k ( q 2 ; q 2 ) k 2 ) = G ( q ) 2 C ( q 2 ) . (42) E_{\triangle}(q) := \sum_{N\ge0} e_{\triangle}(N)q^N
= \left(\sum_{a\ge0} q^{T_a}\right)^2 \left(\sum_{k\ge0} \frac{q^{2k}}{(q^2;q^2)_k^2}\right)
= G(q)^2 C(q^2). \tag*{(42)} E △ ( q ) := N ≥ 0 ∑ e △ ( N ) q N = ( a ≥ 0 ∑ q T a ) 2 ( k ≥ 0 ∑ ( q 2 ; q 2 ) k 2 q 2 k ) = G ( q ) 2 C ( q 2 ) . (42)
The two-color companion admits the following positive decomposition into two corner families.
Theorem 5.3. We have
∑ N ≥ 0 p 2 , d o ( N + 1 ) q N = C ω ( q ) + E △ ( q ) . (43) \sum_{N\ge0} p_{2,do}(N+1)q^N = C_{\omega}(q) + E_{\triangle}(q). \tag*{(43)} N ≥ 0 ∑ p 2 , d o ( N + 1 ) q N = C ω ( q ) + E △ ( q ) . (43) Equivalently, for every N ≥ 0 N\ge0 N ≥ 0 ,
p 2 , d o ( N + 1 ) = c o ( N ) + e △ ( N ) . (44) p_{2,do}(N+1) = c_o(N) + e_{\triangle}(N). \tag*{(44)} p 2 , d o ( N + 1 ) = c o ( N ) + e △ ( N ) . (44)
Thus the comparison between the two-color companion and two copies of the odd corners can be rewritten as a comparison between two corner families. A direct sign-reversing involution on staircase-decorated even corners and odd corners would prove the sparse discrepancy combinatorially. The boxed modular model below supplies the required fixed points, but not yet such a direct corner-to-corner involution.
Theorem 5.4. For n ≥ 0 n\ge0 n ≥ 0 ,
p 2 , d o ( n + 1 ) − 2 c o ( n ) = { ( − 1 ) j , n = 3 j 2 + 2 j , ( − 1 ) j , n = 3 j 2 + 4 j + 1 , 0 , otherwise . (45) p_{2,do}(n+1)-2c_o(n)=
\begin{cases}
(-1)^j, & n=3j^2+2j,\\
(-1)^j, & n=3j^2+4j+1,\\
0, & \text{otherwise}.
\end{cases} \tag*{(45)} p 2 , d o ( n + 1 ) − 2 c o ( n ) = ⎩ ⎨ ⎧ ( − 1 ) j , ( − 1 ) j , 0 , n = 3 j 2 + 2 j , n = 3 j 2 + 4 j + 1 , otherwise . (45)
Combining the positive decomposition with Burson’s boxed modular identity gives the following corner-to-corner and signed-set descriptions.
Corollary 5.5. For every N ≥ 0 N\ge0 N ≥ 0 ,
e △ ( N ) − c o ( N ) = p 2 , d o ( N + 1 ) − 2 c o ( N ) = ∑ ( k , π ) ∈ B 3 , 1 ( N ) ( − 1 ) ν ( π ) + ∑ ( k , π ) ∈ B 3 , 2 ( N − 1 ) ( − 1 ) ν ( π ) . (46) \begin{aligned}
e_{\triangle}(N)-c_o(N) &= p_{2,do}(N+1)-2c_o(N)\\
&= \sum_{(k,\pi)\in B_{3,1}(N)}(-1)^{\nu(\pi)}\\
&\quad+ \sum_{(k,\pi)\in B_{3,2}(N-1)}(-1)^{\nu(\pi)}. \tag*{(46)}
\end{aligned} e △ ( N ) − c o ( N ) = p 2 , d o ( N + 1 ) − 2 c o ( N ) = ( k , π ) ∈ B 3 , 1 ( N ) ∑ ( − 1 ) ν ( π ) + ( k , π ) ∈ B 3 , 2 ( N − 1 ) ∑ ( − 1 ) ν ( π ) . (46)
The second sum is understood to be empty when N = 0 N=0 N = 0 . Under Burson’s involutions, the first signed sum is nonzero precisely when N = 3 j 2 + 2 j N=3j^2+2j N = 3 j 2 + 2 j , and the second is nonzero precisely when N = 3 j 2 + 4 j + 1 N=3j^2+4j+1 N = 3 j 2 + 4 j + 1 ; in either case the surviving value is ( − 1 ) j (-1)^j ( − 1 ) j .
Proposition 5.6. For n ≥ 0 n\ge0 n ≥ 0 , c o ( n ) c_o(n) c o ( n ) is odd if and only if 3 n + 1 3n+1 3 n + 1 is an even square.
Corollary 5.7. Let χ 3 \chi_3 χ 3 be the nontrivial character modulo 3. For n ≥ 0 n \ge0 n ≥ 0 ,
p 2 , d o ( n ) ≡ { χ 3 ( x ) , 3 n − 2 = x 2 for some x > 0 , 0 , 3 n − 2 is not a square ( m o d 4 ) . p_{2,\mathrm{do}}(n) \equiv\begin{cases}
\chi_3(x), & 3n-2=x^2\text{ for some }x>0,\\
0, & 3n-2\text{ is not a square}
\end{cases} \pmod{4}. p 2 , do ( n ) ≡ { χ 3 ( x ) , 0 , 3 n − 2 = x 2 for some x > 0 , 3 n − 2 is not a square ( mod 4 ) .
Corollary 5.8. Let M ≥ 2 M \ge2 M ≥ 2 and 0 ≤ r < M 0 \le r < M 0 ≤ r < M . If 3 r + 1 3r+1 3 r + 1 is a nonsquare modulo M M M , then c o ( M n + r ) ≡ 0 ( m o d 2 ) c_o(Mn+r) \equiv0 \pmod{2} c o ( M n + r ) ≡ 0 ( mod 2 ) . If 3 r − 2 3r-2 3 r − 2 is a nonsquare modulo M M M , then p 2 , d o ( M n + r ) ≡ 0 ( m o d 4 ) p_{2,\mathrm{do}}(Mn+r) \equiv0 \pmod{4} p 2 , do ( M n + r ) ≡ 0 ( mod 4 ) .
The z = i z=i z = i specialization gives the companion of ν 3 \nu_3 ν 3 . Put
C ν ( q ) : = ∑ m ≥ 0 q m ( − q ; q 2 ) m + 1 . (47) C_\nu(q) := \sum_{m\ge0}\frac{q^m}{(-q;q^2)_{m+1}}. \tag*{(47)} C ν ( q ) := m ≥ 0 ∑ ( − q ; q 2 ) m + 1 q m . (47)
Consequently, q C ν ( q 2 ) = S c ( i ; q ) qC_\nu(q^2)=S_c(i;q) q C ν ( q 2 ) = S c ( i ; q ) is the signed generating function for self-conjugate odd corners: a side partition is repeated on the two arms and contributes the sign ( − 1 ) ℓ ( λ ) (-1)^{\ell(\lambda)} ( − 1 ) ℓ ( λ ) .
Proposition 5.9. We have
C ν ( q ) = ∑ m ≥ 0 ( q 2 ; q 4 ) m q 2 m = f O ( − q 2 ) . (48) C_\nu(q)=\sum_{m\ge0}(q^2;q^4)_m q^{2m}=f_O(-q^2). \tag*{(48)} C ν ( q ) = m ≥ 0 ∑ ( q 2 ; q 4 ) m q 2 m = f O ( − q 2 ) . (48)
Finally, define
C ψ ( q ) : = ∑ m ≥ 1 q m ( q ; q 2 ) m . (49) C_\psi(q):=\sum_{m\ge1}\frac{q^m}{(q;q^2)_m}. \tag*{(49)} C ψ ( q ) := m ≥ 1 ∑ ( q ; q 2 ) m q m . (49)
This series is included because it has the same one-arm odd denominator as the ψ 3 \psi_3 ψ 3 kernel after linearizing the quadratic weight. It is the exceptional member of the dictionary.
Proposition 5.10. We have
C ψ ( q ) = q ω 3 ( q ) . (50) C_\psi(q)=q\omega_3(q). \tag*{(50)} C ψ ( q ) = q ω 3 ( q ) . (50)
Thus the expected partial- or false-theta companion of ψ 3 \psi_3 ψ 3 remains in the mock-theta world. The two preceding propositions are recorded to complete the dictionary; their analytic content follows from the classical identities in § 2 . This exception is one reason to regard the dictionary as a structured parallel rather than a formal operator on arbitrary q q q -series.
Oriented three-dimensional corners and a three-color companion
We state the results first and give their proofs together in § 10 .
The obvious extension of a corner to three dimensions has one row, one column, and one tower. To make this precise, and to avoid confusion with solid partitions in the usual order-ideal sense, we retain labels on the three coordinate arms.
Figure 1. Schematic labeled-arm diagrams. In (b), the first two arms are odd, the decorated arm is distinct and below the peak, the weight is 11, and the side rank is 2 − 1 = 1 2-1=1 2 − 1 = 1 .
Definition 6.1. An oriented three-dimensional corner is a quadruple
Γ = ( m ; λ , μ , ν ) , \Gamma=(m;\lambda,\mu,\nu), Γ = ( m ; λ , μ , ν ) , where m ≥ 0 m\ge0 m ≥ 0 and λ , μ , ν \lambda,\mu,\nu λ , μ , ν are ordinary partitions whose parts are at most m m m . Its weight is
∣ Γ ∣ = m + ∣ λ ∣ + ∣ μ ∣ + ∣ ν ∣ . |\Gamma|=m+|\lambda|+|\mu|+|\nu|. ∣Γ∣ = m + ∣ λ ∣ + ∣ μ ∣ + ∣ ν ∣. The three partitions occupy labeled coordinate arms; no quotient by a permutation or rotation is taken.
Only the following restricted family is used in this paper. Its row and column retain the odd-corner structure, while its tower is a distinct-part decoration.
Definition 6.2. A restricted odd three-dimensional corner is a tuple
( 2 m + 1 ; λ , μ , ν ) , m ≥ 0 , (2m+1;\lambda,\mu,\nu),\qquad m\ge0, ( 2 m + 1 ; λ , μ , ν ) , m ≥ 0 , in which every part of λ \lambda λ and μ \mu μ is odd and at most 2 m + 1 2m+1 2 m + 1 , while ν \nu ν has distinct parts in { 1 , 2 , … , 2 m } \{1,2,\ldots,2m\} { 1 , 2 , … , 2 m } . We write c 3 , o ( n ) c_{3,o}(n) c 3 , o ( n ) for the number of such objects of weight n n n and set c 3 , o ( 0 ) = 0 c_{3,o}(0)=0 c 3 , o ( 0 ) = 0 .
The restriction is asymmetric: the row and column form an odd-corner skeleton, whereas the tower is a bounded distinct-part decoration. The orientation is part of the structure. We select this family because its finite tower product is precisely the factor occurring in Ramanujan’s identity (25) ; no rotationally invariant solid-partition interpretation is intended.
For peak 2 m + 1 2m+1 2 m + 1 , the two odd arms contribute 1 / ( q ; q 2 ) m + 1 1/(q;q^2)_{m+1} 1/ ( q ; q 2 ) m + 1 , and the distinct tower contributes ( − q ; q ) 2 m (-q;q)_{2m} ( − q ; q ) 2 m . Hence
C 3 , o ( q ) : = ∑ n ≥ 0 c 3 , o ( n ) q n = ∑ m ≥ 0 ( − q ; q ) 2 m q 2 m + 1 ( q ; q 2 ) m + 1 . (51) C_{3,o}(q):=\sum_{n\ge0}c_{3,o}(n)q^n=\sum_{m\ge0}\frac{(-q;q)_{2m}q^{2m+1}}{(q;q^2)_{m+1}}. \tag*{(51)} C 3 , o ( q ) := n ≥ 0 ∑ c 3 , o ( n ) q n = m ≥ 0 ∑ ( q ; q 2 ) m + 1 ( − q ; q ) 2 m q 2 m + 1 . (51)
For a restricted odd three-dimensional corner Γ = ( 2 m + 1 ; λ , μ , ν ) \Gamma=(2m+1;\lambda,\mu,\nu) Γ = ( 2 m + 1 ; λ , μ , ν ) , define its side rank by
rank ( Γ ) = ℓ ( λ ) − ℓ ( μ ) . \operatorname{rank}(\Gamma)=\ell(\lambda)-\ell(\mu). rank ( Γ ) = ℓ ( λ ) − ℓ ( μ ) .
Let d ( r , n ) d(r,n) d ( r , n ) denote the number of such corners of weight n n n and side rank r r r . A rank variable on the two side arms then gives
D ( z ; q ) : = ∑ m ≥ 0 ( − q ; q ) 2 m q 2 m + 1 ( z q ; q 2 ) m + 1 ( z − 1 q ; q 2 ) m + 1 = ∑ n ≥ 0 ∑ r ∈ Z d ( r , n ) z r q n . (52) \mathcal{D}(z;q):=\sum_{m\geq0}\frac{(-q;q)_{2m}q^{2m+1}}{(zq;q^2)_{m+1}(z^{-1}q;q^2)_{m+1}}=\sum_{n\geq0}\sum_{r\in\mathbb{Z}}d(r,n)z^rq^n. \tag*{(52)} D ( z ; q ) := m ≥ 0 ∑ ( z q ; q 2 ) m + 1 ( z − 1 q ; q 2 ) m + 1 ( − q ; q ) 2 m q 2 m + 1 = n ≥ 0 ∑ r ∈ Z ∑ d ( r , n ) z r q n . (52)
For ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 and 0 ≤ a < ℓ 0\leq a<\ell 0 ≤ a < ℓ , let
d a ( ℓ ) ( n ) : = ∑ r ≡ a ( m o d ℓ ) d ( r , n ) . (53) d_a^{(\ell)}(n):=\sum_{\substack{r\equiv a\pmod{\ell}}}d(r,n). \tag*{(53)} d a ( ℓ ) ( n ) := r ≡ a ( mod ℓ ) ∑ d ( r , n ) . (53)
Interchanging the row and column fixes the class a = 0 a=0 a = 0 and gives
d a ( ℓ ) ( n ) = d ℓ − a ( ℓ ) ( n ) ( 1 ≤ a < ℓ ) . d_a^{(\ell)}(n)=d_{\ell-a}^{(\ell)}(n)\qquad(1\leq a<\ell). d a ( ℓ ) ( n ) = d ℓ − a ( ℓ ) ( n ) ( 1 ≤ a < ℓ ) .
Putting a = z a=z a = z in (25) yields
( 1 + z − 1 ) D ( z ; q ) = 1 ( q ; q 2 ) ∞ ( z q ; q 2 ) ∞ ( z − 1 q ; q 2 ) ∞ ∑ j ≥ 0 ( − z ) j q ( j + 1 2 ) − ∑ j ≥ 0 ( − z ) j q j ( j + 1 ) . (54) (1+z^{-1})\mathcal{D}(z;q)=\frac{1}{(q;q^2)_\infty(zq;q^2)_\infty(z^{-1}q;q^2)_\infty}\sum_{j\geq0}(-z)^jq^{\binom{j+1}{2}}
-\sum_{j\geq0}(-z)^jq^{j(j+1)}. \tag*{(54)} ( 1 + z − 1 ) D ( z ; q ) = ( q ; q 2 ) ∞ ( z q ; q 2 ) ∞ ( z − 1 q ; q 2 ) ∞ 1 j ≥ 0 ∑ ( − z ) j q ( 2 j + 1 ) − j ≥ 0 ∑ ( − z ) j q j ( j + 1 ) . (54)
The second term in (54) is the boxed 2-modular member of the same family. Its precise specialization is the following.
Proposition 6.3. For a formal variable w w w ,
∑ j ≥ 0 ( − w 2 ) j q j ( j + 1 ) = B 2 , 1 ( w , q 1 / 2 ) . (55) \sum_{j\geq0}(-w^2)^jq^{j(j+1)}=\mathcal{B}_{2,1}\left(w,q^{1/2}\right). \tag*{(55)} j ≥ 0 ∑ ( − w 2 ) j q j ( j + 1 ) = B 2 , 1 ( w , q 1/2 ) . (55)
Consequently, after z = w 2 z=w^2 z = w 2 , the rank-refined sparse correction in (54) is precisely the fixed-point contribution of Burson’s boxed 2-modular involution.
The point z = − 1 z=-1 z = − 1 is exceptional in (54) , because the factor 1 + z − 1 1+z^{-1} 1 + z − 1 vanishes. It nevertheless yields an exact rank identity. Retain the triangular-number series G ( q ) G(q) G ( q ) from (15) , and put
G 1 ( q ) : = ∑ j ≥ 0 j q j ( j + 1 ) / 2 . (56) G_1(q):=\sum_{j\geq0}jq^{j(j+1)/2}. \tag*{(56)} G 1 ( q ) := j ≥ 0 ∑ j q j ( j + 1 ) /2 . (56)
Theorem 6.4. We have
D ( − 1 ; q ) = G ( q 2 ) G ( q ) G 1 ( q ) − G 1 ( q 2 ) (57) \mathcal{D}(-1;q)=\frac{G(q^2)}{G(q)}G_1(q)-G_1(q^2) \tag*{(57)} D ( − 1 ; q ) = G ( q ) G ( q 2 ) G 1 ( q ) − G 1 ( q 2 ) (57) = ( q ; q 2 ) ∞ ( q 2 ; q 4 ) ∞ 2 G 1 ( q ) − G 1 ( q 2 ) . (58) =\frac{(q;q^2)_\infty}{(q^2;q^4)_\infty^2}G_1(q)-G_1(q^2). \tag*{(58)} = ( q 2 ; q 4 ) ∞ 2 ( q ; q 2 ) ∞ G 1 ( q ) − G 1 ( q 2 ) . (58)
By (53) ,
D ( − 1 ; q ) = ∑ n > 0 ( d 0 ( 2 ) ( n ) − d 1 ( 2 ) ( n ) ) q n . \mathcal{D}(-1;q)=\sum_{n>0}\left(d_0^{(2)}(n)-d_1^{(2)}(n)\right)q^n. D ( − 1 ; q ) = n > 0 ∑ ( d 0 ( 2 ) ( n ) − d 1 ( 2 ) ( n ) ) q n .
The next theorem gives a strict bias between the two parity classes.
Theorem 6.5. We have
− D ( − 1 ; − q ) = ∑ m ≥ 0 q 2 m + 1 ( − q 2 ; q 2 ) m ( q ; q 2 ) m ( 1 − q 2 m + 1 ) 2 . (59) -\mathcal{D}(-1;-q)=\sum_{m\geq0}\frac{q^{2m+1}(-q^2;q^2)_m}{(q;q^2)_m(1-q^{2m+1})^2}. \tag*{(59)} − D ( − 1 ; − q ) = m ≥ 0 ∑ ( q ; q 2 ) m ( 1 − q 2 m + 1 ) 2 q 2 m + 1 ( − q 2 ; q 2 ) m . (59) Consequently, for every n ≥ 1 n\geq1 n ≥ 1 ,
( − 1 ) n + 1 ( d 0 ( 2 ) ( n ) − d 1 ( 2 ) ( n ) ) > 0. (60) (-1)^{n+1}\left(d_0^{(2)}(n)-d_1^{(2)}(n)\right)>0. \tag*{(60)} ( − 1 ) n + 1 ( d 0 ( 2 ) ( n ) − d 1 ( 2 ) ( n ) ) > 0. (60) Thus even side ranks predominate when n n n is odd, whereas odd side ranks predominate when n n n is even.
The right-hand side of (6.9) has a direct positive partition interpretation. For fixed m m m , it records a distinguished part 2 m + 1 2m+1 2 m + 1 , arbitrary odd parts below the peak, distinct even parts not exceeding 2 m 2m 2 m , and two independent colors for additional copies of the peak.
The nonexceptional fourth and third roots of unity give two further exact rank differences. Write T j = j ( j + 1 ) / 2 T_j=j(j+1)/2 T j = j ( j + 1 ) /2 and define
Φ 4 ( q ) : = ∑ m ≥ 0 ( − 1 ) m q T 2 m = ∑ m ≥ 0 ( − 1 ) m q m ( 2 m + 1 ) , (61) \Phi_4(q):=\sum_{m\geq0}(-1)^m q^{T_{2m}}=\sum_{m\geq0}(-1)^m q^{m(2m+1)}, \tag*{(61)} Φ 4 ( q ) := m ≥ 0 ∑ ( − 1 ) m q T 2 m = m ≥ 0 ∑ ( − 1 ) m q m ( 2 m + 1 ) , (61)
Φ 3 ( q ) : = ∑ m ≥ 0 ( − 1 ) m ( q T 3 m + q T 3 m + 1 ) . (62) \Phi_3(q):=\sum_{m\geq0}(-1)^m\left(q^{T_{3m}}+q^{T_{3m+1}}\right). \tag*{(62)} Φ 3 ( q ) := m ≥ 0 ∑ ( − 1 ) m ( q T 3 m + q T 3 m + 1 ) . (62)
Theorem 6.6. Let ω = e 2 π i / 3 \omega=e^{2\pi i/3} ω = e 2 π i /3 . Then
D ( i ; q ) = ∑ n ≥ 0 ( d 0 ( 4 ) ( n ) − d 2 ( 4 ) ( n ) ) q n \mathcal{D}(i;q)=\sum_{n\geq0}\left(d_0^{(4)}(n)-d_2^{(4)}(n)\right)q^n D ( i ; q ) = n ≥ 0 ∑ ( d 0 ( 4 ) ( n ) − d 2 ( 4 ) ( n ) ) q n = Φ 4 ( q ) ( q ; q 2 ) ∞ ( − q 2 ; q 4 ) ∞ − Φ 4 ( q 2 ) , (63) =\frac{\Phi_4(q)}{(q;q^2)_\infty(-q^2;q^4)_\infty}-\Phi_4(q^2), \tag*{(63)} = ( q ; q 2 ) ∞ ( − q 2 ; q 4 ) ∞ Φ 4 ( q ) − Φ 4 ( q 2 ) , (63) and
D ( ω ; q ) = ∑ n ≥ 0 ( d 0 ( 3 ) ( n ) − d 1 ( 3 ) ( n ) ) q n \mathcal{D}(\omega;q)=\sum_{n\geq0}\left(d_0^{(3)}(n)-d_1^{(3)}(n)\right)q^n D ( ω ; q ) = n ≥ 0 ∑ ( d 0 ( 3 ) ( n ) − d 1 ( 3 ) ( n ) ) q n = Φ 3 ( q ) ( q 3 ; q 6 ) ∞ − Φ 3 ( q 2 ) . (64) =\frac{\Phi_3(q)}{(q^3;q^6)_\infty}-\Phi_3(q^2). \tag*{(64)} = ( q 3 ; q 6 ) ∞ Φ 3 ( q ) − Φ 3 ( q 2 ) . (64)
The cubic identity first gives exact rank equidistribution on the progression 6 n + 5 6n+5 6 n + 5 .
Corollary 6.7. For every n ≥ 0 n\geq0 n ≥ 0 ,
d 0 ( 3 ) ( 6 n + 5 ) = d 1 ( 3 ) ( 6 n + 5 ) = d 2 ( 3 ) ( 6 n + 5 ) = c 3 , o ( 6 n + 5 ) 3 . (65) d_0^{(3)}(6n+5)=d_1^{(3)}(6n+5)=d_2^{(3)}(6n+5)=\frac{c_{3,o}(6n+5)}{3}. \tag*{(65)} d 0 ( 3 ) ( 6 n + 5 ) = d 1 ( 3 ) ( 6 n + 5 ) = d 2 ( 3 ) ( 6 n + 5 ) = 3 c 3 , o ( 6 n + 5 ) . (65) Equivalently, the side-rank Laurent polynomial
∑ r ∈ Z d ( r , 6 n + 5 ) z r \sum_{r\in\mathbb{Z}}d(r,6n+5)z^r r ∈ Z ∑ d ( r , 6 n + 5 ) z r is divisible by 1 + z + z 2 1+z+z^2 1 + z + z 2 in Z [ z , z − 1 ] \mathbb{Z}[z,z^{-1}] Z [ z , z − 1 ] .
The same identity gives a sparse rank difference on the progression 6 n + 2 6n+2 6 n + 2 .
Corollary 6.8. If N ≡ 2 ( m o d 6 ) N \equiv2 \pmod{6} N ≡ 2 ( mod 6 ) , then
d 0 ( 3 ) ( N ) − d 1 ( 3 ) ( N ) = { ( − 1 ) m + 1 , N = ( 3 m + 1 ) ( 3 m + 2 ) , for some m ≥ 0 , 0 , otherwise. (66) d_0^{(3)}(N)-d_1^{(3)}(N)=
\begin{cases}
(-1)^{m+1}, & N=(3m+1)(3m+2),\text{ for some }m\ge0,\\
0, & \text{otherwise.}
\end{cases} \tag*{(66)} d 0 ( 3 ) ( N ) − d 1 ( 3 ) ( N ) = { ( − 1 ) m + 1 , 0 , N = ( 3 m + 1 ) ( 3 m + 2 ) , for some m ≥ 0 , otherwise. (66)
Let p 3 , d o ( n ) p_{3,\mathrm{do}}(n) p 3 , do ( n ) count three-color partitions of n n n , with colors red, blue, and yellow, and smallest part size s s s , subject to the following conditions: at least one smallest part is red; the red size s s s occurs once or twice and all larger red parts are distinct; the blue parts are distinct and at least s s s ; and the yellow parts are odd and at least 2 s − 1 2s-1 2 s − 1 . Set p 3 , d o ( 0 ) = 0 p_{3,\mathrm{do}}(0)=0 p 3 , do ( 0 ) = 0 . Then
∑ n ≥ 0 p 3 , d o ( n ) q n = ∑ s ≥ 1 q s ( − q s ; q 2 ) ∞ 2 ( q 2 s − 1 ; q 2 ) ∞ . (67) \sum_{n\ge0}p_{3,\mathrm{do}}(n)q^n=\sum_{s\ge1}q^s\frac{(-q^s;q^2)_\infty^2}{(q^{2s-1};q^2)_\infty}. \tag*{(67)} n ≥ 0 ∑ p 3 , do ( n ) q n = s ≥ 1 ∑ q s ( q 2 s − 1 ; q 2 ) ∞ ( − q s ; q 2 ) ∞ 2 . (67)
As in the two-color case, this is an analytic companion selected so that a fixed smallest part produces the product in the transformed corner series. We next give a parallel positive decomposition.
For k ≥ 0 k\ge0 k ≥ 0 , let Γ = ( k ; α , β ) \Gamma=(k;\alpha,\beta) Γ = ( k ; α , β ) be an ordinary corner and let 2 Γ = ( 2 k ; 2 α , 2 β ) 2\Gamma=(2k;2\alpha,2\beta) 2Γ = ( 2 k ; 2 α , 2 β ) be obtained by doubling every entry. For k ≥ 1 k\ge1 k ≥ 1 , a zero-augmented distinct tower below 2 k 2k 2 k is a subset
η ⊆ { 0 , 1 , … , 2 k − 1 } . \eta\subseteq\{0,1,\ldots,2k-1\}. η ⊆ { 0 , 1 , … , 2 k − 1 } .
For k = 0 k=0 k = 0 , the allowable tower set is understood to be empty and we take η = ∅ \eta=\varnothing η = ∅ , in agreement with ( − 1 ; q ) 0 = 1 (-1;q)_0=1 ( − 1 ; q ) 0 = 1 . In all cases, put wt ( η ) = ∑ t ∈ η t \operatorname{wt}(\eta)=\sum_{t\in\eta}t wt ( η ) = ∑ t ∈ η t ; the possible entry 0 0 0 is a mark, not a positive part. A staircase-and-tower-decorated even corner is a quadruple
( 2 Γ , δ a , δ b , η ) (2\Gamma,\delta_a,\delta_b,\eta) ( 2Γ , δ a , δ b , η )
of weight
2 ∣ Γ ∣ + T a + T b + wt ( η ) . 2|\Gamma|+T_a+T_b+\operatorname{wt}(\eta). 2∣Γ∣ + T a + T b + wt ( η ) .
Let e 3 , △ ( N ) e_{3,\triangle}(N) e 3 , △ ( N ) count these objects of weight N N N . Since the zero-augmented tower contributes ( − 1 ; q ) 2 k (-1;q)_{2k} ( − 1 ; q ) 2 k , their generating function is
E 3 , △ ( q ) : = ∑ N ≥ 0 e 3 , △ ( N ) q N = G ( q ) 2 ∑ k ≥ 0 ( − 1 ; q ) 2 k q 2 k ( q 2 ; q 2 ) k . (68) E_{3,\triangle}(q):=\sum_{N\ge0}e_{3,\triangle}(N)q^N
=G(q)^2\sum_{k\ge0}\frac{(-1;q)_{2k}q^{2k}}{(q^2;q^2)_k}. \tag*{(68)} E 3 , △ ( q ) := N ≥ 0 ∑ e 3 , △ ( N ) q N = G ( q ) 2 k ≥ 0 ∑ ( q 2 ; q 2 ) k ( − 1 ; q ) 2 k q 2 k . (68)
The three-color companion has a positive decomposition parallel to the two-color decomposition above.
Theorem 6.9. We have
∑ N ≥ 0 p 3 , d o ( N + 1 ) q N = 2 C 3 , o ( q ) + E 3 , △ ( q ) . (69) \sum_{N\ge0}p_{3,\mathrm{do}}(N+1)q^N=2C_{3,o}(q)+E_{3,\triangle}(q). \tag*{(69)} N ≥ 0 ∑ p 3 , do ( N + 1 ) q N = 2 C 3 , o ( q ) + E 3 , △ ( q ) . (69) Equivalently, for every N ≥ 0 N\ge0 N ≥ 0 ,
p 3 , d o ( N + 1 ) = 2 c 3 , o ( N ) + e 3 , △ ( N ) . (70) p_{3,\mathrm{do}}(N+1)=2c_{3,o}(N)+e_{3,\triangle}(N). \tag*{(70)} p 3 , do ( N + 1 ) = 2 c 3 , o ( N ) + e 3 , △ ( N ) . (70)
Thus the three-color companion splits positively into two labeled copies of the restricted odd three-dimensional corners and one decorated even-corner family. A direct bijection realizing this decomposition, and a direct sign-reversing involution between the latter family and two labeled copies of the restricted corners, remain to be found. Proposition 6.3 provides the fixed-point model for the sparse correction.
Theorem 6.10. For n ≥ 0 n \ge0 n ≥ 0 ,
p 3 , d o ( n + 1 ) − 4 c 3 , o ( n ) = { ( − 1 ) j , n = j ( j + 1 ) , 0 , otherwise . (71) p_{3,\mathrm{do}}(n+1)-4c_{3,\mathrm{o}}(n)=
\begin{cases}
(-1)^j, & n=j(j+1),\\
0, & \text{otherwise}.
\end{cases} \tag*{(71)} p 3 , do ( n + 1 ) − 4 c 3 , o ( n ) = { ( − 1 ) j , 0 , n = j ( j + 1 ) , otherwise . (71)
Combining the preceding decomposition with the boxed 2-modular identity gives the following two forms of the three-color discrepancy.
Corollary 6.11. For every N ≥ 0 N \ge0 N ≥ 0 ,
e 3 , △ ( N ) − 2 c 3 , o ( N ) = p 3 , d o ( N + 1 ) − 4 c 3 , o ( N ) = ∑ ( k , π ) ∈ B 2 , 1 ( 2 N ) ( − 1 ) ν ( π ) . (72) \begin{aligned}
e_{3,\triangle}(N)-2c_{3,\mathrm{o}}(N)&=p_{3,\mathrm{do}}(N+1)-4c_{3,\mathrm{o}}(N)\\
&=\sum_{(k,\pi)\in B_{2,1}(2N)}(-1)^{\nu(\pi)}. \tag*{(72)}
\end{aligned} e 3 , △ ( N ) − 2 c 3 , o ( N ) = p 3 , do ( N + 1 ) − 4 c 3 , o ( N ) = ( k , π ) ∈ B 2 , 1 ( 2 N ) ∑ ( − 1 ) ν ( π ) . (72)
Burson’s boxed 2-modular involution cancels all nonfixed pairs. A fixed pair occurs exactly when N = j ( j + 1 ) N=j(j+1) N = j ( j + 1 ) , and is then ( j , ( 2 j + 1 ) j ) (j,(2j+1)^j) ( j , ( 2 j + 1 ) j ) with sign ( − 1 ) j (-1)^j ( − 1 ) j .
Corollary 6.12. Let χ 4 \chi_4 χ 4 be the nontrivial character modulo 4. For n ≥ 0 n \ge0 n ≥ 0 ,
p 3 , d o ( n ) ≡ { χ 4 ( x ) , 4 n − 3 = x 2 for some x > 0 , 0 , 4 n − 3 is not a square ( m o d 4 ) . p_{3,\mathrm{do}}(n)\equiv
\begin{cases}
\chi_4(x), & 4n-3=x^2\text{ for some }x>0,\\
0, & 4n-3\text{ is not a square}
\end{cases}
\pmod{4}. p 3 , do ( n ) ≡ { χ 4 ( x ) , 0 , 4 n − 3 = x 2 for some x > 0 , 4 n − 3 is not a square ( mod 4 ) .
The cubic rank difference also gives the following congruence dissection.
Corollary 6.13. If N ≡ 2 ( m o d 6 ) N \equiv2 \pmod{6} N ≡ 2 ( mod 6 ) , then
c 3 , o ( N ) ≡ { ( − 1 ) m + 1 , N = ( 3 m + 1 ) ( 3 m + 2 ) for some m ≥ 0 , 0 , otherwise ( m o d 3 ) . (73) c_{3,\mathrm{o}}(N)\equiv
\begin{cases}
(-1)^{m+1}, & N=(3m+1)(3m+2)\text{ for some }m\ge0,\\
0, & \text{otherwise}
\end{cases}
\pmod{3}. \tag*{(73)} c 3 , o ( N ) ≡ { ( − 1 ) m + 1 , 0 , N = ( 3 m + 1 ) ( 3 m + 2 ) for some m ≥ 0 , otherwise ( mod 3 ) . (73) If N ≡ 3 ( m o d 6 ) N \equiv3 \pmod{6} N ≡ 3 ( mod 6 ) , then
p 3 , d o ( N ) ≡ { 5 ( − 1 ) m + 1 , N − 1 = ( 3 m + 1 ) ( 3 m + 2 ) for some m ≥ 0 , 0 , otherwise ( m o d 12 ) . (74) p_{3,\mathrm{do}}(N)\equiv
\begin{cases}
5(-1)^{m+1}, & N-1=(3m+1)(3m+2)\text{ for some }m\ge0,\\
0, & \text{otherwise}
\end{cases}
\pmod{12}. \tag*{(74)} p 3 , do ( N ) ≡ { 5 ( − 1 ) m + 1 , 0 , N − 1 = ( 3 m + 1 ) ( 3 m + 2 ) for some m ≥ 0 , otherwise ( mod 12 ) . (74)
For reference, Table 2 lists the initial values of the three less familiar counting functions and fixes the shifts used in the two colored identities.
Table 2. Initial values of the three less familiar counting functions.
Arithmetic of the sparse supports
The congruences in this section arise by excluding the support of the sparse theta corrections. We retain the notation
T j = j ( j + 1 ) 2 T_j = \frac{j(j+1)}{2} T j = 2 j ( j + 1 )
and, for M ≥ 2 M \ge2 M ≥ 2 , put
S M = { j ( j + 1 ) ( m o d M ) : j ≥ 0 } , S_M = \{j(j+1) \pmod M : j \ge0\}, S M = { j ( j + 1 ) ( mod M ) : j ≥ 0 } ,
R M = { T 2 a + 1 + T b ( m o d M ) : a , b ≥ 0 } . R_M = \{T_{2a+1} + T_b \pmod M : a,b \ge0\}. R M = { T 2 a + 1 + T b ( mod M ) : a , b ≥ 0 } .
The proofs are given in Section 11.
We first isolate the two support-exclusion mechanisms used below.
Theorem 7.1. Let M ≥ 2 M \ge2 M ≥ 2 and 0 ≤ r < M 0 \le r < M 0 ≤ r < M .
(1) If r ∉ R M r \notin R_M r ∈ / R M , then, for every n ≥ 0 n \ge0 n ≥ 0 ,
c 3 , o ( M n + r ) ≡ 0 ( m o d 2 ) , p 3 , d o ( M n + r + 1 ) ≡ 0 ( m o d 8 ) . c_{3,o}(Mn+r) \equiv0 \pmod2,\qquad p_{3,do}(Mn+r+1) \equiv0 \pmod8. c 3 , o ( M n + r ) ≡ 0 ( mod 2 ) , p 3 , d o ( M n + r + 1 ) ≡ 0 ( mod 8 ) . (2) If 3 ∣ M 3 \mid M 3 ∣ M , r ≡ 2 ( m o d 3 ) r \equiv2 \pmod3 r ≡ 2 ( mod 3 ) , and r ∉ S M r \notin S_M r ∈ / S M , then, for every n ≥ 0 n \ge0 n ≥ 0 ,
c 3 , o ( M n + r ) ≡ 0 ( m o d 3 ) , p 3 , d o ( M n + r + 1 ) ≡ 0 ( m o d 12 ) . c_{3,o}(Mn+r) \equiv0 \pmod3,\qquad p_{3,do}(Mn+r+1) \equiv0 \pmod{12}. c 3 , o ( M n + r ) ≡ 0 ( mod 3 ) , p 3 , d o ( M n + r + 1 ) ≡ 0 ( mod 12 ) . If both hypotheses hold, then, for every n ≥ 0 n \ge0 n ≥ 0 ,
c 3 , o ( M n + r ) ≡ 0 ( m o d 6 ) , p 3 , d o ( M n + r + 1 ) ≡ 0 ( m o d 24 ) . c_{3,o}(Mn+r) \equiv0 \pmod6,\qquad p_{3,do}(Mn+r+1) \equiv0 \pmod{24}. c 3 , o ( M n + r ) ≡ 0 ( mod 6 ) , p 3 , d o ( M n + r + 1 ) ≡ 0 ( mod 24 ) .
We now give the complete local description of the binary triangular support.
Theorem 7.2. Let M ≥ 2 M \ge2 M ≥ 2 and 0 ≤ r < M 0 \le r < M 0 ≤ r < M . Then
r ∉ R M ⟺ there exists p α ∣ M , p ≡ 3 ( m o d 4 ) , r \notin R_M \iff\text{there exists }p^\alpha\mid M,\ p \equiv3 \pmod4, r ∈ / R M ⟺ there exists p α ∣ M , p ≡ 3 ( mod 4 ) , such that v p ( 4 r + 1 ) is odd and v p ( 4 r + 1 ) < α . \text{such that }v_p(4r+1)\text{ is odd and }v_p(4r+1)<\alpha. such that v p ( 4 r + 1 ) is odd and v p ( 4 r + 1 ) < α . In particular, R M R_M R M is the full residue system if and only if every prime p ≡ 3 ( m o d 4 ) p \equiv3 \pmod4 p ≡ 3 ( mod 4 ) occurs in M M M with exponent at most 1.
The preceding two theorems give the following concrete prime-power family.
Corollary 7.3. Let p ≡ 3 ( m o d 4 ) p \equiv3 \pmod4 p ≡ 3 ( mod 4 ) be prime, α ≥ 2 \alpha\ge2 α ≥ 2 , and 0 ≤ r < p α 0 \le r < p^\alpha 0 ≤ r < p α . If v p ( 4 r + 1 ) v_p(4r+1) v p ( 4 r + 1 ) is odd and less than α \alpha α , then, for every n ≥ 0 n \ge0 n ≥ 0 ,
c 3 , o ( p α n + r ) ≡ 0 ( m o d 2 ) , p 3 , d o ( p α n + r + 1 ) ≡ 0 ( m o d 8 ) . c_{3,o}(p^\alpha n+r) \equiv0 \pmod2,\qquad p_{3,do}(p^\alpha n+r+1) \equiv0 \pmod8. c 3 , o ( p α n + r ) ≡ 0 ( mod 2 ) , p 3 , d o ( p α n + r + 1 ) ≡ 0 ( mod 8 ) . For p = 3 p=3 p = 3 , the moduli strengthen to 6 and 24, respectively.
For example, taking p = 3 p=3 p = 3 , α = 2 \alpha=2 α = 2 , and r ∈ { 5 , 8 } r \in\{5,8\} r ∈ { 5 , 8 } gives
c 3 , o ( 9 n + r ) ≡ 0 ( m o d 6 ) , p 3 , d o ( 9 n + r + 1 ) ≡ 0 ( m o d 24 ) ( n ≥ 0 ) . c_{3,o}(9n+r) \equiv0 \pmod6,\qquad p_{3,do}(9n+r+1) \equiv0 \pmod{24}\qquad(n \ge0). c 3 , o ( 9 n + r ) ≡ 0 ( mod 6 ) , p 3 , d o ( 9 n + r + 1 ) ≡ 0 ( mod 24 ) ( n ≥ 0 ) . All proofs for this section are collected in Section 11.
Proofs of the results in Section 4
We give the proofs in the order in which the results are stated.
Proof of Proposition 4.1. In Ramanujan’s identity (2.24), put a = b = − 1 a=b=-1 a = b = − 1 . The left-hand side then becomes
∑ n ≥ 0 q n ( q ; q ) n 2 = C ( q ) . \sum_{n\geq0}\frac{q^n}{(q;q)_n^2}=C(q). n ≥ 0 ∑ ( q ; q ) n 2 q n = C ( q ) .
Moreover, the coefficient 1 + 1 / a 1+1/a 1 + 1/ a of the second series is zero, so that series disappears. In the remaining series we have
( − 1 ) m + 1 a − m − 1 b m = ( − 1 ) m + 1 ( − 1 ) − m − 1 ( − 1 ) m = ( − 1 ) m , (-1)^{m+1}a^{-m-1}b^m=(-1)^{m+1}(-1)^{-m-1}(-1)^m=(-1)^m, ( − 1 ) m + 1 a − m − 1 b m = ( − 1 ) m + 1 ( − 1 ) − m − 1 ( − 1 ) m = ( − 1 ) m ,
and
( − a q ; q ) ∞ ( − b q ; q ) ∞ = ( q ; q ) ∞ 2 . (-aq;q)_\infty(-bq;q)_\infty=(q;q)_\infty^2. ( − a q ; q ) ∞ ( − b q ; q ) ∞ = ( q ; q ) ∞ 2 .
Consequently,
C ( q ) = 1 ( q ; q ) ∞ 2 ∑ m ≥ 0 ( − 1 ) m q ( m + 1 2 ) = f Θ ( q ) ( q ; q ) ∞ 2 . C(q)=\frac{1}{(q;q)_\infty^2}\sum_{m\geq0}(-1)^m q^{\binom{m+1}{2}}=\frac{\mathrm{f}\Theta(q)}{(q;q)_\infty^2}. C ( q ) = ( q ; q ) ∞ 2 1 m ≥ 0 ∑ ( − 1 ) m q ( 2 m + 1 ) = ( q ; q ) ∞ 2 f Θ ( q ) .
which is (4.2).
Proof of Theorem 4.2. Let L ( N ) L(N) L ( N ) denote the number of two-color partitions of N N N in which the largest part occurs exactly once and that occurrence is red. If the largest part is M M M , then it contributes q M q^M q M , while all remaining red and blue parts are at most M − 1 M-1 M − 1 . Hence
∑ N ≥ 1 L ( N ) q N = ∑ M ≥ 1 q M ( q ; q ) M − 1 2 = q ∑ m ≥ 0 q m ( q ; q ) m 2 = q C ( q ) . \sum_{N\geq1}L(N)q^N=\sum_{M\geq1}\frac{q^M}{(q;q)_{M-1}^2}
=q\sum_{m\geq0}\frac{q^m}{(q;q)_m^2}=qC(q). N ≥ 1 ∑ L ( N ) q N = M ≥ 1 ∑ ( q ; q ) M − 1 2 q M = q m ≥ 0 ∑ ( q ; q ) m 2 q m = q C ( q ) .
Comparison of coefficients gives L ( n + 1 ) = c ( n ) L(n+1)=c(n) L ( n + 1 ) = c ( n ) .
For the minimal-excludant interpretation, pair consecutive terms in the partial theta series:
f Θ ( q ) = ∑ j ≥ 0 ( q ( 2 j + 1 2 ) − q ( 2 j + 2 2 ) ) = ∑ j ≥ 0 q 1 + 2 + ⋯ + 2 j ( 1 − q 2 j + 1 ) . \begin{aligned}
\mathrm{f}\Theta(q)=\sum_{j\geq0}\left(q^{\binom{2j+1}{2}}-q^{\binom{2j+2}{2}}\right) \\
&=\sum_{j\geq0}q^{1+2+\cdots+2j}(1-q^{2j+1}).
\end{aligned} f Θ ( q ) = j ≥ 0 ∑ ( q ( 2 2 j + 1 ) − q ( 2 2 j + 2 ) ) = j ≥ 0 ∑ q 1 + 2 + ⋯ + 2 j ( 1 − q 2 j + 1 ) .
Using Proposition 4.1, we therefore obtain
C ( q ) = 1 ( q ; q ) ∞ ∑ j ≥ 0 q 1 + 2 + ⋯ + 2 j ( 1 − q 2 j + 1 ) ( q ; q ) ∞ = 1 ( q ; q ) ∞ ∑ j ≥ 0 ( ∏ k = 1 2 j q k 1 − q k ) ( ∏ k ≥ 2 j + 2 1 1 − q k ) . \begin{aligned}
C(q)=\frac{1}{(q;q)_\infty}\sum_{j\geq0}\frac{q^{1+2+\cdots+2j}(1-q^{2j+1})}{(q;q)_\infty} \\
&=\frac{1}{(q;q)_\infty}\sum_{j\geq0}\left(\prod_{k=1}^{2j}\frac{q^k}{1-q^k}\right)\left(\prod_{k\geq2j+2}\frac{1}{1-q^k}\right).
\end{aligned} C ( q ) = ( q ; q ) ∞ 1 j ≥ 0 ∑ ( q ; q ) ∞ q 1 + 2 + ⋯ + 2 j ( 1 − q 2 j + 1 ) = ( q ; q ) ∞ 1 j ≥ 0 ∑ ( k = 1 ∏ 2 j 1 − q k q k ) k ≥ 2 j + 2 ∏ 1 − q k 1 .
The first factor 1 / ( q ; q ) ∞ 1/(q;q)_\infty 1/ ( q ; q ) ∞ generates the red parts. For the blue parts, the summand indexed by j j j requires each of 1 , 2 , … , 2 j 1,2,\ldots,2j 1 , 2 , … , 2 j to occur, forbids 2 j + 1 2j+1 2 j + 1 , and leaves every larger part unrestricted. Thus the smallest positive integer absent as a blue part is 2 j + 1 2j+1 2 j + 1 , which is odd. This proves the third interpretation; see also Andrews and Newman [10 ] . □ \square □
Proof of Proposition 4.3. Work in the ring F 2 [ [ q ] ] \mathbb{F}_2[[q]] F 2 [[ q ]] . Since
( 1 − q j ) 2 ≡ 1 − q 2 j ( m o d 2 ) , (1-q^j)^2 \equiv1-q^{2j} \pmod{2}, ( 1 − q j ) 2 ≡ 1 − q 2 j ( mod 2 ) ,
we have ( q ; q ) m 2 ≡ ( q 2 ; q 2 ) m ( m o d 2 ) (q;q)_m^2 \equiv(q^2;q^2)_m \pmod{2} ( q ; q ) m 2 ≡ ( q 2 ; q 2 ) m ( mod 2 ) , and hence
C ( q ) = ∑ m ≥ 0 q m ( q ; q ) m 2 ≡ ∑ m ≥ 0 q m ( q 2 ; q 2 ) m ( m o d 2 ) . C(q)=\sum_{m\geq0}\frac{q^m}{(q;q)_m^2}\equiv\sum_{m\geq0}\frac{q^m}{(q^2;q^2)_m}\pmod{2}. C ( q ) = m ≥ 0 ∑ ( q ; q ) m 2 q m ≡ m ≥ 0 ∑ ( q 2 ; q 2 ) m q m ( mod 2 ) .
The q q q -binomial theorem (2.13), with base q 2 q^2 q 2 , a = 0 a=0 a = 0 , and z = q z=q z = q , gives
∑ m ≥ 0 q m ( q 2 ; q 2 ) m = 1 ( q ; q 2 ) ∞ . \sum_{m\geq0}\frac{q^m}{(q^2;q^2)_m}=\frac{1}{(q;q^2)_\infty}. m ≥ 0 ∑ ( q 2 ; q 2 ) m q m = ( q ; q 2 ) ∞ 1 .
Since
1 ( q ; q 2 ) ∞ = ( q 2 ; q 2 ) ∞ ( q ; q ) ∞ = ( − q ; q ) ∞ ≡ ( q ; q ) ∞ ( m o d 2 ) , \frac{1}{(q;q^2)_\infty}=\frac{(q^2;q^2)_\infty}{(q;q)_\infty}=(-q;q)_\infty\equiv(q;q)_\infty\pmod{2}, ( q ; q 2 ) ∞ 1 = ( q ; q ) ∞ ( q 2 ; q 2 ) ∞ = ( − q ; q ) ∞ ≡ ( q ; q ) ∞ ( mod 2 ) ,
Euler’s pentagonal number theorem (2.14) yields
C ( q ) ≡ ∑ k = − ∞ ∞ q k ( 3 k + 1 ) / 2 ( m o d 2 ) . C(q)\equiv\sum_{k=-\infty}^{\infty}q^{k(3k+1)/2}\pmod{2}. C ( q ) ≡ k = − ∞ ∑ ∞ q k ( 3 k + 1 ) /2 ( mod 2 ) .
The exponents in this bilateral sum are precisely the generalized pentagonal numbers k ( 3 k ± 1 ) / 2 k(3k\pm1)/2 k ( 3 k ± 1 ) /2 with k ≥ 0 k\geq0 k ≥ 0 . The result follows by comparing coefficients. □ \square □
Proof of Corollary 4.4. Suppose that c ( M n + r ) c(Mn+r) c ( M n + r ) is odd and put N = M n + r N=Mn+r N = M n + r . By Proposition 4.3, either N = k ( 3 k − 1 ) / 2 N=k(3k-1)/2 N = k ( 3 k − 1 ) /2 or N = k ( 3 k + 1 ) / 2 N=k(3k+1)/2 N = k ( 3 k + 1 ) /2 for some k ≥ 0 k\geq0 k ≥ 0 . Accordingly,
24 N + 1 = ( 6 k − 1 ) 2 or 24 N + 1 = ( 6 k + 1 ) 2 . 24N+1=(6k-1)^2 \qquad\text{or}\qquad24N+1=(6k+1)^2. 24 N + 1 = ( 6 k − 1 ) 2 or 24 N + 1 = ( 6 k + 1 ) 2 .
Reducing modulo M M M shows that 24 r + 1 24r+1 24 r + 1 must be a square modulo M M M . The contrapositive gives the asserted congruence. □ \square □
Proof of Proposition 4.5. Set a = b = 1 a=b=1 a = b = 1 in (2.24). The left-hand side is C f ( q ) C_f(q) C f ( q ) . The first series on the right becomes
− 1 ( − q ; q ) ∞ 2 ∑ m ≥ 0 ( − 1 ) m q ( m + 1 2 ) = − f Θ ( q ) ( − q ; q ) ∞ 2 . -\frac{1}{(-q;q)_\infty^2}\sum_{m\geq0}(-1)^m q^{\binom{m+1}{2}}=-\frac{f_{\Theta}(q)}{(-q;q)_\infty^2}. − ( − q ; q ) ∞ 2 1 m ≥ 0 ∑ ( − 1 ) m q ( 2 m + 1 ) = − ( − q ; q ) ∞ 2 f Θ ( q ) .
The factor 1 + 1 / a 1+1/a 1 + 1/ a in the second series is now 2, and the remaining series is exactly the left-hand side of (20) . Therefore
C f ( q ) = − f Θ ( q ) ( − q ; q ) ∞ 2 + 2 ∑ m ≥ 0 ( − 1 ) m q m ( m + 1 ) / 2 ( − q ; q ) m = − f Θ ( q ) ( − q ; q ) ∞ 2 + 2 f E ( q ) , \begin{aligned}
C_f(q)=-\frac{f\Theta(q)}{(-q;q)_{\infty}^{2}}+2\sum_{m\geq0}\frac{(-1)^m q^{m(m+1)/2}}{(-q;q)_m} \\
&=-\frac{f\Theta(q)}{(-q;q)_{\infty}^{2}}+2fE(q),
\end{aligned} C f ( q ) = − ( − q ; q ) ∞ 2 f Θ ( q ) + 2 m ≥ 0 ∑ ( − q ; q ) m ( − 1 ) m q m ( m + 1 ) /2 = − ( − q ; q ) ∞ 2 f Θ ( q ) + 2 f E ( q ) ,
which proves (4.4). In particular, the negative sign in the first term follows directly from the specialization. □
Proof of Proposition 4.6. Since
1 ( − q 2 ; q 2 ) n = ( q 2 ; q 2 ) n ( q 4 ; q 4 ) n , \frac{1}{(-q^2;q^2)_n}=\frac{(q^2;q^2)_n}{(q^4;q^4)_n}, ( − q 2 ; q 2 ) n 1 = ( q 4 ; q 4 ) n ( q 2 ; q 2 ) n ,
we may write
C ϕ ( q ) = ∑ n ≥ 0 ( q 2 ; q 2 ) n q n ( q 4 ; q 4 ) n = ( q 2 ; q 2 ) ∞ ∑ n ≥ 0 q n ( q 4 ; q 4 ) n ( q 2 n + 2 ; q 2 ) ∞ . \begin{aligned}
C_\phi(q)=\sum_{n\geq0}\frac{(q^2;q^2)_nq^n}{(q^4;q^4)_n} \\
&=(q^2;q^2)_\infty\sum_{n\geq0}\frac{q^n}{(q^4;q^4)_n(q^{2n+2};q^2)_\infty}.
\end{aligned} C ϕ ( q ) = n ≥ 0 ∑ ( q 4 ; q 4 ) n ( q 2 ; q 2 ) n q n = ( q 2 ; q 2 ) ∞ n ≥ 0 ∑ ( q 4 ; q 4 ) n ( q 2 n + 2 ; q 2 ) ∞ q n .
By the q q q -binomial theorem (13) with base q 2 q^2 q 2 ,
1 ( q 2 n + 2 ; q 2 ) ∞ = ∑ m ≥ 0 q ( 2 n + 2 ) m ( q 2 ; q 2 ) m . \frac{1}{(q^{2n+2};q^2)_\infty}=\sum_{m\geq0}\frac{q^{(2n+2)m}}{(q^2;q^2)_m}. ( q 2 n + 2 ; q 2 ) ∞ 1 = m ≥ 0 ∑ ( q 2 ; q 2 ) m q ( 2 n + 2 ) m .
Substitution and interchange of the two sums give
C ϕ ( q ) = ( q 2 ; q 2 ) ∞ ∑ m ≥ 0 q 2 m ( q 2 ; q 2 ) m ∑ n ≥ 0 q ( 2 m + 1 ) n ( q 4 ; q 4 ) n = ( q 2 ; q 2 ) ∞ ∑ m ≥ 0 q 2 m ( q 2 ; q 2 ) m ( q 2 m + 1 ; q 4 ) ∞ , \begin{aligned}
C_\phi(q)=(q^2;q^2)_\infty\sum_{m\geq0}\frac{q^{2m}}{(q^2;q^2)_m}\sum_{n\geq0}\frac{q^{(2m+1)n}}{(q^4;q^4)_n} \\
&=(q^2;q^2)_\infty\sum_{m\geq0}\frac{q^{2m}}{(q^2;q^2)_m(q^{2m+1};q^4)_\infty},
\end{aligned} C ϕ ( q ) = ( q 2 ; q 2 ) ∞ m ≥ 0 ∑ ( q 2 ; q 2 ) m q 2 m n ≥ 0 ∑ ( q 4 ; q 4 ) n q ( 2 m + 1 ) n = ( q 2 ; q 2 ) ∞ m ≥ 0 ∑ ( q 2 ; q 2 ) m ( q 2 m + 1 ; q 4 ) ∞ q 2 m ,
where the q q q -binomial theorem (13) with base q 4 q^4 q 4 was used once more.
We now separate the even and odd values of m m m . The product identities
( q 2 ; q 2 ) 2 r = ( q 2 ; q 4 ) r ( q 4 ; q 4 ) r , (q^2;q^2)_{2r}=(q^2;q^4)_r(q^4;q^4)_r, ( q 2 ; q 2 ) 2 r = ( q 2 ; q 4 ) r ( q 4 ; q 4 ) r ,
( q 2 ; q 2 ) 2 r + 1 = ( 1 − q 2 ) ( q 6 ; q 4 ) r ( q 4 ; q 4 ) r , (q^2;q^2)_{2r+1}=(1-q^2)(q^6;q^4)_r(q^4;q^4)_r, ( q 2 ; q 2 ) 2 r + 1 = ( 1 − q 2 ) ( q 6 ; q 4 ) r ( q 4 ; q 4 ) r ,
1 ( q 4 r + 1 ; q 4 ) ∞ = ( q ; q 4 ) r ( q ; q ) ∞ , \frac{1}{(q^{4r+1};q^4)_\infty}=\frac{(q;q^4)_r}{(q;q)_\infty}, ( q 4 r + 1 ; q 4 ) ∞ 1 = ( q ; q ) ∞ ( q ; q 4 ) r ,
1 ( q 4 r + 3 ; q 4 ) ∞ = ( q 3 ; q 4 ) r ( q 3 ; q 4 ) ∞ \frac{1}{(q^{4r+3};q^4)_\infty}=\frac{(q^3;q^4)_r}{(q^3;q^4)_\infty} ( q 4 r + 3 ; q 4 ) ∞ 1 = ( q 3 ; q 4 ) ∞ ( q 3 ; q 4 ) r
show that C ϕ ( q ) = E ( q ) + O ( q ) C_{\phi}(q)=E(q)+O(q) C ϕ ( q ) = E ( q ) + O ( q ) , where
E ( q ) = ( q 2 ; q 2 ) ∞ ( q ; q 4 ) ∞ ∑ r ≥ 0 ( q ; q 4 ) r q 4 r ( q 4 ; q 4 ) r ( q 2 ; q 4 ) r , E(q)=\frac{(q^{2};q^{2})_{\infty}}{(q;q^{4})_{\infty}}\sum_{r\geq0}\frac{(q;q^{4})_{r}q^{4r}}{(q^{4};q^{4})_{r}(q^{2};q^{4})_{r}}, E ( q ) = ( q ; q 4 ) ∞ ( q 2 ; q 2 ) ∞ r ≥ 0 ∑ ( q 4 ; q 4 ) r ( q 2 ; q 4 ) r ( q ; q 4 ) r q 4 r ,
O ( q ) = q 2 ( q 2 ; q 2 ) ∞ ( 1 − q 2 ) ( q 3 ; q 4 ) ∞ ∑ r ≥ 0 ( q 3 ; q 4 ) r q 4 r ( q 4 ; q 4 ) r ( q 6 ; q 4 ) r . O(q)=\frac{q^{2}(q^{2};q^{2})_{\infty}}{(1-q^{2})(q^{3};q^{4})_{\infty}}\sum_{r\geq0}\frac{(q^{3};q^{4})_{r}q^{4r}}{(q^{4};q^{4})_{r}(q^{6};q^{4})_{r}}. O ( q ) = ( 1 − q 2 ) ( q 3 ; q 4 ) ∞ q 2 ( q 2 ; q 2 ) ∞ r ≥ 0 ∑ ( q 4 ; q 4 ) r ( q 6 ; q 4 ) r ( q 3 ; q 4 ) r q 4 r .
Apply Heine’s transformation (2.17) with base q 4 q^{4} q 4 . Taking ( a , b , c , z ) = ( q , q 4 , 0 , q ) (a,b,c,z)=(q,q^{4},0,q) ( a , b , c , z ) = ( q , q 4 , 0 , q ) gives
E ( q ) = ∑ n ≥ 0 ( q ; q 4 ) n q n , E(q)=\sum_{n\geq0}(q;q^{4})_{n}q^{n}, E ( q ) = n ≥ 0 ∑ ( q ; q 4 ) n q n ,
while taking ( a , b , c , z ) = ( q 3 , q 4 , 0 , q 3 ) (a,b,c,z)=(q^{3},q^{4},0,q^{3}) ( a , b , c , z ) = ( q 3 , q 4 , 0 , q 3 ) gives
O ( q ) = q 2 ∑ n ≥ 0 ( q 3 ; q 4 ) n q 3 n . O(q)=q^{2}\sum_{n\geq0}(q^{3};q^{4})_{n}q^{3n}. O ( q ) = q 2 n ≥ 0 ∑ ( q 3 ; q 4 ) n q 3 n .
Here we have used
( q 2 ; q 2 ) ∞ = ( q 2 ; q 4 ) ∞ ( q 4 ; q 4 ) ∞ (q^{2};q^{2})_{\infty}=(q^{2};q^{4})_{\infty}(q^{4};q^{4})_{\infty} ( q 2 ; q 2 ) ∞ = ( q 2 ; q 4 ) ∞ ( q 4 ; q 4 ) ∞
and
( q 2 ; q 2 ) ∞ 1 − q 2 = ( q 4 ; q 4 ) ∞ ( q 6 ; q 4 ) ∞ . \frac{(q^{2};q^{2})_{\infty}}{1-q^{2}}=(q^{4};q^{4})_{\infty}(q^{6};q^{4})_{\infty}. 1 − q 2 ( q 2 ; q 2 ) ∞ = ( q 4 ; q 4 ) ∞ ( q 6 ; q 4 ) ∞ .
Thus
C ϕ ( q ) = ∑ n ≥ 0 ( q ; q 4 ) n q n + q 2 ∑ n ≥ 0 ( q 3 ; q 4 ) n q 3 n , C_{\phi}(q)=\sum_{n\geq0}(q;q^{4})_{n}q^{n}+q^{2}\sum_{n\geq0}(q^{3};q^{4})_{n}q^{3n}, C ϕ ( q ) = n ≥ 0 ∑ ( q ; q 4 ) n q n + q 2 n ≥ 0 ∑ ( q 3 ; q 4 ) n q 3 n ,
which proves (4.6).
It remains to identify this expression with fE(− q -q − q ). Applying the Rogers–Fine formula (2.18) first with base q 4 q^{4} q 4 and a = q a=q a = q , and then with base q 4 q^{4} q 4 and a = q 3 a=q^{3} a = q 3 , gives
∑ n ≥ 0 ( q ; q 4 ) n q n = ∑ r ≥ 0 ( − 1 ) r q 6 r 2 + r ( 1 + q 4 r + 1 ) , \sum_{n\geq0}(q;q^{4})_{n}q^{n}=\sum_{r\geq0}(-1)^{r}q^{6r^{2}+r}(1+q^{4r+1}), n ≥ 0 ∑ ( q ; q 4 ) n q n = r ≥ 0 ∑ ( − 1 ) r q 6 r 2 + r ( 1 + q 4 r + 1 ) ,
q 2 ∑ n ≥ 0 ( q 3 ; q 4 ) n q 3 n = ∑ r ≥ 0 ( − 1 ) r q 6 r 2 + 7 r + 2 ( 1 + q 4 r + 3 ) . q^{2}\sum_{n\geq0}(q^{3};q^{4})_{n}q^{3n}=\sum_{r\geq0}(-1)^{r}q^{6r^{2}+7r+2}(1+q^{4r+3}). q 2 n ≥ 0 ∑ ( q 3 ; q 4 ) n q 3 n = r ≥ 0 ∑ ( − 1 ) r q 6 r 2 + 7 r + 2 ( 1 + q 4 r + 3 ) .
On the other hand, split the defining series
f E ( − q ) = ∑ n ≥ 0 ( − q ) n ( 3 n + 1 ) / 2 ( 1 − ( − q ) 2 n + 1 ) \mathrm{fE}(-q)=\sum_{n\geq0}(-q)^{n(3n+1)/2}(1-(-q)^{2n+1}) fE ( − q ) = n ≥ 0 ∑ ( − q ) n ( 3 n + 1 ) /2 ( 1 − ( − q ) 2 n + 1 )
into its even and odd indices. For n = 2 r n=2r n = 2 r the summand is
( − 1 ) r q 6 r 2 + r ( 1 + q 4 r + 1 ) , (-1)^{r}q^{6r^{2}+r}(1+q^{4r+1}), ( − 1 ) r q 6 r 2 + r ( 1 + q 4 r + 1 ) ,
whereas for n = 2 r + 1 n=2r+1 n = 2 r + 1 it is
( − 1 ) r q 6 r 2 + 7 r + 2 ( 1 + q 4 r + 3 ) . (-1)^{r}q^{6r^{2}+7r+2}(1+q^{4r+3}). ( − 1 ) r q 6 r 2 + 7 r + 2 ( 1 + q 4 r + 3 ) .
These are exactly the two Rogers–Fine series displayed above. Hence C ϕ ( q ) = C_{\phi}(q)= C ϕ ( q ) = fE(− q -q − q ), proving (4.7). □ \square □
Proofs of the results in Section 5
We give the proofs in the order in which the results are stated. The proofs of Theorems 5.3 and 5.4 use the same parity splitting, which is isolated below together with the partial-theta evaluation needed for the discrepancy.
Proof of Proposition 5.1. By (2.22),
B 3 , 1 ( w 3 , q ) = ∑ j ≥ 0 ( − 1 ) j w 6 j q 3 j 2 + 2 j . \mathcal{B}_{3,1}(w^3,q)=\sum_{j\geq0}(-1)^j w^{6j}q^{3j^2+2j}. B 3 , 1 ( w 3 , q ) = j ≥ 0 ∑ ( − 1 ) j w 6 j q 3 j 2 + 2 j .
and
B 3 , 2 ( w 3 , q ) = ∑ j ≥ 0 ( − 1 ) j w 6 j q 3 j 2 + 4 j . \mathcal{B}_{3,2}(w^3,q)=\sum_{j\geq0}(-1)^j w^{6j}q^{3j^2+4j}. B 3 , 2 ( w 3 , q ) = j ≥ 0 ∑ ( − 1 ) j w 6 j q 3 j 2 + 4 j .
Multiplying the first identity by − w 2 -w^2 − w 2 and the second by − w 4 q -w^4q − w 4 q , and then adding, gives (5.4). □
Proof of Theorem 5.2. Let L o ( N ) L_o(N) L o ( N ) denote the number of two-color partitions of N N N into odd parts in which the largest part occurs exactly once, and that occurrence is red. If the largest part is 2 m + 3 2m+3 2 m + 3 , it contributes q 2 m + 3 q^{2m+3} q 2 m + 3 , while all remaining red and blue parts are odd and at most 2 m + 1 2m+1 2 m + 1 . Hence
∑ N ≥ 2 L o ( N ) q N = ∑ m ≥ 0 q 2 m + 3 ( q ; q 2 ) m + 1 2 = q 2 ∑ m ≥ 0 q 2 m + 1 ( q ; q 2 ) m + 1 2 = q 2 C ω ( q ) . \begin{aligned}
\sum_{N\geq2} L_o(N)q^N=\sum_{m\geq0}\frac{q^{2m+3}}{(q;q^2)_{m+1}^2} \\
&=q^2\sum_{m\geq0}\frac{q^{2m+1}}{(q;q^2)_{m+1}^2}=q^2C_\omega(q).
\end{aligned} N ≥ 2 ∑ L o ( N ) q N = m ≥ 0 ∑ ( q ; q 2 ) m + 1 2 q 2 m + 3 = q 2 m ≥ 0 ∑ ( q ; q 2 ) m + 1 2 q 2 m + 1 = q 2 C ω ( q ) .
Comparison of coefficients gives L o ( n + 2 ) = c o ( n ) L_o(n+2)=c_o(n) L o ( n + 2 ) = c o ( n ) for every $n\geq 0. □
To prove Theorems 5.3 and 5.4, we must separate the auxiliary series according to the parity of an intermediate index. The odd-index part will give the odd-corner series C ω ( q ) C_\omega(q) C ω ( q ) , while the even-index part will give the staircase-decorated series E △ ( q ) E_\triangle(q) E △ ( q ) . Their sum yields the positive decomposition and their difference yields the sparse correction. The next two lemmas carry out these two steps.
Lemma 9.1. Put
T 1 ( q ) = ( − q ; q 2 ) ∞ ∑ m ≥ 0 ( q ; q 2 ) m q m ( − q ; q ) m , T_1(q)=(-q;q^2)_\infty\sum_{m\geq0}\frac{(q;q^2)_m q^m}{(-q;q)_m}, T 1 ( q ) = ( − q ; q 2 ) ∞ m ≥ 0 ∑ ( − q ; q ) m ( q ; q 2 ) m q m , and
T 2 ( q ) = ( q 2 ; q 2 ) ∞ ∑ m ≥ 0 ( q ; q 2 ) m q m ( q ; q ) m . T_2(q)=(q^2;q^2)_\infty\sum_{m\geq0}\frac{(q;q^2)_m q^m}{(q;q)_m}. T 2 ( q ) = ( q 2 ; q 2 ) ∞ m ≥ 0 ∑ ( q ; q ) m ( q ; q 2 ) m q m . Then
T 1 ( q ) = C ω ( q ) + E △ ( q ) , (75) T_1(q)=C_\omega(q)+E_\triangle(q), \tag*{(75)} T 1 ( q ) = C ω ( q ) + E △ ( q ) , (75) and
T 2 ( q ) = E △ ( q ) − C ω ( q ) . (76) T_2(q)=E_\triangle(q)-C_\omega(q). \tag*{(76)} T 2 ( q ) = E △ ( q ) − C ω ( q ) . (76)
In particular,
C ω ( q ) = 1 2 T 1 ( q ) − 1 2 T 2 ( q ) . C_{\omega}(q)=\frac{1}{2}T_1(q)-\frac{1}{2}T_2(q). C ω ( q ) = 2 1 T 1 ( q ) − 2 1 T 2 ( q ) .
Proof. For r ≥ 0 r\ge0 r ≥ 0 , define
H r ( q ) : = ( q 2 ; q 2 ) ∞ ( q ; q 2 ) ∞ q r ( q ; q ) r ∑ m ≥ 0 ( q ; q 2 ) m q ( r + 1 ) m ( q 2 ; q 2 ) m . (77) H_r(q):=\frac{(q^2;q^2)_\infty}{(q;q^2)_\infty}\frac{q^r}{(q;q)_r}\sum_{m\ge0}\frac{(q;q^2)_m q^{(r+1)m}}{(q^2;q^2)_m}. \tag*{(77)} H r ( q ) := ( q ; q 2 ) ∞ ( q 2 ; q 2 ) ∞ ( q ; q ) r q r m ≥ 0 ∑ ( q 2 ; q 2 ) m ( q ; q 2 ) m q ( r + 1 ) m . (77)
Interchanging the two sums and using the q q q -binomial theorem (13) first with z = q m + 1 z=q^{m+1} z = q m + 1 and then with z = − q m + 1 z=-q^{m+1} z = − q m + 1 gives
T 1 ( q ) = ∑ r ≥ 0 H r ( q ) , T 2 ( q ) = ∑ r ≥ 0 ( − 1 ) r H r ( q ) . (78) T_1(q)=\sum_{r\ge0}H_r(q),\qquad T_2(q)=\sum_{r\ge0}(-1)^rH_r(q). \tag*{(78)} T 1 ( q ) = r ≥ 0 ∑ H r ( q ) , T 2 ( q ) = r ≥ 0 ∑ ( − 1 ) r H r ( q ) . (78)
Indeed, the two inner sums over r r r are
∑ r ≥ 0 q r ( m + 1 ) ( q ; q ) r = 1 ( q m + 1 ; q ) ∞ , ∑ r ≥ 0 ( − 1 ) r q r ( m + 1 ) ( q ; q ) r = 1 ( − q m + 1 ; q ) ∞ , \sum_{r\ge0}\frac{q^{r(m+1)}}{(q;q)_r}=\frac{1}{(q^{m+1};q)_\infty},\qquad
\sum_{r\ge0}\frac{(-1)^r q^{r(m+1)}}{(q;q)_r}=\frac{1}{(-q^{m+1};q)_\infty}, r ≥ 0 ∑ ( q ; q ) r q r ( m + 1 ) = ( q m + 1 ; q ) ∞ 1 , r ≥ 0 ∑ ( q ; q ) r ( − 1 ) r q r ( m + 1 ) = ( − q m + 1 ; q ) ∞ 1 ,
and the product identities
( q 2 ; q 2 ) m = ( q ; q ) m ( − q ; q ) m , ( − q ; q ) ∞ = 1 ( q ; q 2 ) ∞ (q^2;q^2)_m=(q;q)_m(-q;q)_m,\qquad(-q;q)_\infty=\frac{1}{(q;q^2)_\infty} ( q 2 ; q 2 ) m = ( q ; q ) m ( − q ; q ) m , ( − q ; q ) ∞ = ( q ; q 2 ) ∞ 1
reduce the resulting expressions to T 1 ( q ) T_1(q) T 1 ( q ) and T 2 ( q ) T_2(q) T 2 ( q ) .
A further application of (13) , now with base q 2 q^2 q 2 , yields
∑ m ≥ 0 ( q ; q 2 ) m q ( r + 1 ) m ( q 2 ; q 2 ) m = ( q r + 2 ; q 2 ) ∞ ( q r + 1 ; q 2 ) ∞ . \sum_{m\ge0}\frac{(q;q^2)_m q^{(r+1)m}}{(q^2;q^2)_m}
=\frac{(q^{r+2};q^2)_\infty}{(q^{r+1};q^2)_\infty}. m ≥ 0 ∑ ( q 2 ; q 2 ) m ( q ; q 2 ) m q ( r + 1 ) m = ( q r + 1 ; q 2 ) ∞ ( q r + 2 ; q 2 ) ∞ .
If r = 2 k + 1 r=2k+1 r = 2 k + 1 , then
H 2 k + 1 ( q ) = q 2 k + 1 ( q 2 ; q 2 ) k ( q ; q 2 ) k + 1 ( q ; q ) 2 k + 1 = q 2 k + 1 ( q ; q 2 ) k + 1 2 , H_{2k+1}(q)=q^{2k+1}\frac{(q^2;q^2)_k}{(q;q^2)_{k+1}(q;q)_{2k+1}}
=\frac{q^{2k+1}}{(q;q^2)_{k+1}^2}, H 2 k + 1 ( q ) = q 2 k + 1 ( q ; q 2 ) k + 1 ( q ; q ) 2 k + 1 ( q 2 ; q 2 ) k = ( q ; q 2 ) k + 1 2 q 2 k + 1 ,
because ( q ; q ) 2 k + 1 = ( q ; q 2 ) k + 1 ( q 2 ; q 2 ) k (q;q)_{2k+1}=(q;q^2)_{k+1}(q^2;q^2)_k ( q ; q ) 2 k + 1 = ( q ; q 2 ) k + 1 ( q 2 ; q 2 ) k . Thus
H 2 k + 1 ( q ) = q 2 k + 1 ( q ; q 2 ) k + 1 2 , (79) H_{2k+1}(q)=\frac{q^{2k+1}}{(q;q^2)_{k+1}^2}, \tag*{(79)} H 2 k + 1 ( q ) = ( q ; q 2 ) k + 1 2 q 2 k + 1 , (79)
which is the summand of C ω ( q ) C_{\omega}(q) C ω ( q ) with peak 2 k + 1 2k+1 2 k + 1 .
If r = 2 k r=2k r = 2 k , then
H 2 k ( q ) = q 2 k ( q 2 ; q 2 ) ∞ 2 ( q ; q 2 ) ∞ 2 ( q ; q 2 ) k ( q 2 ; q 2 ) k ( q ; q ) 2 k = G ( q ) 2 q 2 k ( q 2 ; q 2 ) k 2 , H_{2k}(q)=q^{2k}\frac{(q^2;q^2)_\infty^2}{(q;q^2)_\infty^2}\frac{(q;q^2)_k}{(q^2;q^2)_k(q;q)_{2k}}
=G(q)^2\frac{q^{2k}}{(q^2;q^2)_k^2}, H 2 k ( q ) = q 2 k ( q ; q 2 ) ∞ 2 ( q 2 ; q 2 ) ∞ 2 ( q 2 ; q 2 ) k ( q ; q ) 2 k ( q ; q 2 ) k = G ( q ) 2 ( q 2 ; q 2 ) k 2 q 2 k ,
where we used (15) and ( q ; q ) 2 k = ( q ; q 2 ) k ( q 2 ; q 2 ) k (q;q)_{2k}=(q;q^2)_k(q^2;q^2)_k ( q ; q ) 2 k = ( q ; q 2 ) k ( q 2 ; q 2 ) k . Therefore
H 2 k ( q ) = G ( q ) 2 q 2 k ( q 2 ; q 2 ) k 2 (80) H_{2k}(q)=G(q)^2\frac{q^{2k}}{(q^2;q^2)_k^2} \tag*{(80)} H 2 k ( q ) = G ( q ) 2 ( q 2 ; q 2 ) k 2 q 2 k (80)
which is the summand of G ( q ) 2 C ( q 2 ) = E △ ( q ) G(q)^2C(q^2)=E_{\triangle}(q) G ( q ) 2 C ( q 2 ) = E △ ( q ) with doubled peak 2 k 2k 2 k . Thus the odd-index part of (78) is C ω ( q ) C_{\omega}(q) C ω ( q ) and the even-index part is E △ ( q ) E_{\triangle}(q) E △ ( q ) . Adding and subtracting these two parts proves (75) and (76) . □
Lemma 9.2. We have
T 2 ( q ) = ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) . T_2(q)=\sum_{j\geq0}(-1)^j q^{3j^2+2j}\left(1+q^{2j+1}\right). T 2 ( q ) = j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) .
Proof. We use Heine’s transformation (17) , together with the formal half-power convention stated in § 2 . Since
( q ; q 2 ) m = ( q 1 / 2 ; q ) m ( − q 1 / 2 ; q ) m , (q;q^2)_m=(q^{1/2};q)_m(-q^{1/2};q)_m, ( q ; q 2 ) m = ( q 1/2 ; q ) m ( − q 1/2 ; q ) m ,
we apply (17) with
( a , b , c , z ) = ( − q 1 / 2 , q 1 / 2 , 0 , q ) . (a,b,c,z)=(-q^{1/2},q^{1/2},0,q). ( a , b , c , z ) = ( − q 1/2 , q 1/2 , 0 , q ) .
This gives
T 2 ( q ) = ( q 2 ; q 2 ) ∞ ∑ m ≥ 0 ( q ; q 2 ) m q m ( q ; q ) m = ( q 2 ; q 2 ) ∞ ( q 1 / 2 ; q ) ∞ ( − q 1 / 2 ; q ) ∞ ( q ; q ) ∞ ∑ n ≥ 0 q n / 2 ( − q 1 / 2 ; q ) n + 1 . \begin{aligned}
T_2(q)=(q^2;q^2)_\infty\sum_{m\geq0}\frac{(q;q^2)_m q^m}{(q;q)_m} \\
&=\frac{(q^2;q^2)_\infty(q^{1/2};q)_\infty(-q^{1/2};q)_\infty}{(q;q)_\infty}\sum_{n\geq0}\frac{q^{n/2}}{(-q^{1/2};q)_{n+1}}.
\end{aligned} T 2 ( q ) = ( q 2 ; q 2 ) ∞ m ≥ 0 ∑ ( q ; q ) m ( q ; q 2 ) m q m = ( q ; q ) ∞ ( q 2 ; q 2 ) ∞ ( q 1/2 ; q ) ∞ ( − q 1/2 ; q ) ∞ n ≥ 0 ∑ ( − q 1/2 ; q ) n + 1 q n /2 .
The prefactor is 1, because
( q 1 / 2 ; q ) ∞ ( − q 1 / 2 ; q ) ∞ = ( q ; q 2 ) ∞ (q^{1/2};q)_\infty(-q^{1/2};q)_\infty=(q;q^2)_\infty ( q 1/2 ; q ) ∞ ( − q 1/2 ; q ) ∞ = ( q ; q 2 ) ∞
and
( q ; q ) ∞ = ( q ; q 2 ) ∞ ( q 2 ; q 2 ) ∞ . (q;q)_\infty=(q;q^2)_\infty(q^2;q^2)_\infty. ( q ; q ) ∞ = ( q ; q 2 ) ∞ ( q 2 ; q 2 ) ∞ .
Replacing q q q by q 1 / 2 q^{1/2} q 1/2 in (21) now gives
∑ n ≥ 0 q n / 2 ( − q 1 / 2 ; q ) n + 1 = ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) , \sum_{n\geq0}\frac{q^{n/2}}{(-q^{1/2};q)_{n+1}}=\sum_{j\geq0}(-1)^j q^{3j^2+2j}\left(1+q^{2j+1}\right), n ≥ 0 ∑ ( − q 1/2 ; q ) n + 1 q n /2 = j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) ,
which proves the lemma. □
Proof of Theorem 5.3 . The generating function (41) gives
∑ n ≥ 0 p 2 , d o ( n ) q n = q ∑ m ≥ 0 ( − q m + 1 ; q ) ∞ ( q 2 m + 1 ; q 2 ) ∞ = q ( − q ; q ) ∞ ( q ; q 2 ) ∞ ∑ m ≥ 0 ( q ; q 2 ) m q m ( − q ; q ) m = q T 1 ( q ) . \begin{aligned}
\sum_{n\geq0}p_{2,do}(n)q^n=q\sum_{m\geq0}\frac{(-q^{m+1};q)_\infty}{(q^{2m+1};q^2)_\infty} \\
&=q\frac{(-q;q)_\infty}{(q;q^2)_\infty}\sum_{m\geq0}\frac{(q;q^2)_m q^m}{(-q;q)_m} \\
&=qT_1(q).
\end{aligned} n ≥ 0 ∑ p 2 , d o ( n ) q n = q m ≥ 0 ∑ ( q 2 m + 1 ; q 2 ) ∞ ( − q m + 1 ; q ) ∞ = q ( q ; q 2 ) ∞ ( − q ; q ) ∞ m ≥ 0 ∑ ( − q ; q ) m ( q ; q 2 ) m q m = q T 1 ( q ) .
Since p 2 , d o ( 0 ) = 0 p_{2,do}(0)=0 p 2 , d o ( 0 ) = 0 , division by q q q and (75) give
∑ N ≥ 0 p 2 , d o ( N + 1 ) q N = T 1 ( q ) = C ω ( q ) + E △ ( q ) . \sum_{N\geq0}p_{2,do}(N+1)q^N=T_1(q)=C_{\omega}(q)+E_{\triangle}(q). N ≥ 0 ∑ p 2 , d o ( N + 1 ) q N = T 1 ( q ) = C ω ( q ) + E △ ( q ) .
This proves (5.8); coefficient comparison gives (5.9). □
Proof of Theorem 5.4. The preceding calculation gives
∑ N ≥ 0 p 2 , d o ( N + 1 ) q N = T 1 ( q ) . \sum_{N\ge0} p_{2,\mathrm{do}}(N+1)q^N=T_1(q). N ≥ 0 ∑ p 2 , do ( N + 1 ) q N = T 1 ( q ) .
By (9.1) and (9.2), T 1 ( q ) − 2 C ω ( q ) = T 2 ( q ) T_1(q)-2C_\omega(q)=T_2(q) T 1 ( q ) − 2 C ω ( q ) = T 2 ( q ) . Lemma 9.2 therefore gives
∑ N ≥ 0 ( p 2 , d o ( N + 1 ) − 2 c o ( N ) ) q N = ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) . \sum_{N\ge0}\left(p_{2,\mathrm{do}}(N+1)-2c_o(N)\right)q^N=\sum_{j\ge0}(-1)^j q^{3j^2+2j}\left(1+q^{2j+1}\right). N ≥ 0 ∑ ( p 2 , do ( N + 1 ) − 2 c o ( N ) ) q N = j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) .
The theorem follows by comparing coefficients. □
Proof of Corollary 5.5. The positive decomposition (5.8) gives
E Δ ( q ) − C ω ( q ) = ∑ N ≥ 0 ( p 2 , d o ( N + 1 ) − 2 c o ( N ) ) q N . (81) E_\Delta(q)-C_\omega(q)=\sum_{N\ge0}\left(p_{2,\mathrm{do}}(N+1)-2c_o(N)\right)q^N. \tag*{(81)} E Δ ( q ) − C ω ( q ) = N ≥ 0 ∑ ( p 2 , do ( N + 1 ) − 2 c o ( N ) ) q N . (81)
By Theorem 5.4, the right-hand side of (9.7) is
∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 2 j + ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 4 j + 1 . (82) \sum_{j\ge0}(-1)^j q^{3j^2+2j}+\sum_{j\ge0}(-1)^j q^{3j^2+4j+1}. \tag*{(82)} j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 2 j + j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 4 j + 1 . (82)
We now identify the two series in (9.8) with the two boxed modular series. Setting u = 1 u=1 u = 1 in (2.22), first with ( m , r ) = ( 3 , 1 ) (m,r)=(3,1) ( m , r ) = ( 3 , 1 ) and then with ( m , r ) = ( 3 , 2 ) (m,r)=(3,2) ( m , r ) = ( 3 , 2 ) , gives
B 3 , 1 ( 1 , q ) = ∑ j ≥ 0 ( − 1 ) j q j ( 3 j + 2 ) = ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 2 j \mathcal{B}_{3,1}(1,q)=\sum_{j\ge0}(-1)^j q^{j(3j+2)}=\sum_{j\ge0}(-1)^j q^{3j^2+2j} B 3 , 1 ( 1 , q ) = j ≥ 0 ∑ ( − 1 ) j q j ( 3 j + 2 ) = j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 2 j
and
q B 3 , 2 ( 1 , q ) = q ∑ j ≥ 0 ( − 1 ) j q j ( 3 j + 4 ) = ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 4 j + 1 . q\mathcal{B}_{3,2}(1,q)=q\sum_{j\ge0}(-1)^j q^{j(3j+4)}=\sum_{j\ge0}(-1)^j q^{3j^2+4j+1}. q B 3 , 2 ( 1 , q ) = q j ≥ 0 ∑ ( − 1 ) j q j ( 3 j + 4 ) = j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 4 j + 1 .
Consequently,
E Δ ( q ) − C ω ( q ) = B 3 , 1 ( 1 , q ) + q B 3 , 2 ( 1 , q ) . (83) E_\Delta(q)-C_\omega(q)=\mathcal{B}_{3,1}(1,q)+q\mathcal{B}_{3,2}(1,q). \tag*{(83)} E Δ ( q ) − C ω ( q ) = B 3 , 1 ( 1 , q ) + q B 3 , 2 ( 1 , q ) . (83)
The first line of (2.22) shows that
[ q N ] B 3 , 1 ( 1 , q ) = ∑ ( k , π ˉ ) ∈ B 3 , 1 ( N ) ( − 1 ) ν ( π ˉ ) . [q^N]\mathcal{B}_{3,1}(1,q)=\sum_{(k,\bar{\pi})\in\mathcal{B}_{3,1}(N)}(-1)^{\nu(\bar{\pi})}. [ q N ] B 3 , 1 ( 1 , q ) = ( k , π ˉ ) ∈ B 3 , 1 ( N ) ∑ ( − 1 ) ν ( π ˉ ) .
Similarly,
[ q N ] q B 3 , 2 ( 1 , q ) = [ q N − 1 ] B 3 , 2 ( 1 , q ) = ∑ ( k , π ˉ ) ∈ B 3 , 2 ( N − 1 ) ( − 1 ) ν ( π ˉ ) , [q^N]q\mathcal{B}_{3,2}(1,q)=[q^{N-1}]\mathcal{B}_{3,2}(1,q)
=\sum_{(k,\bar{\pi})\in\mathcal{B}_{3,2}(N-1)}(-1)^{\nu(\bar{\pi})}, [ q N ] q B 3 , 2 ( 1 , q ) = [ q N − 1 ] B 3 , 2 ( 1 , q ) = ( k , π ˉ ) ∈ B 3 , 2 ( N − 1 ) ∑ ( − 1 ) ν ( π ˉ ) ,
where the last sum is empty for N = 0 N=0 N = 0 . Taking the coefficient of q N q^N q N in (9.9) proves (5.11).
It remains to explain the exceptional weights. For ( m , r ) = ( 3 , 1 ) (m,r)=(3,1) ( m , r ) = ( 3 , 1 ) , the fixed pair in (2.23) is ( j , ( 3 j + 1 ) j ) (j,(3j+1)^j) ( j , ( 3 j + 1 ) j ) . Its sign is ( − 1 ) j (-1)^j ( − 1 ) j , and its weight is
1 ⋅ j + j ( 3 j + 1 ) = 3 j 2 + 2 j . 1\cdot j+j(3j+1)=3j^2+2j. 1 ⋅ j + j ( 3 j + 1 ) = 3 j 2 + 2 j .
For ( m , r ) = ( 3 , 2 ) (m,r)=(3,2) ( m , r ) = ( 3 , 2 ) , the fixed pair is ( j , ( 3 j + 2 ) j ) (j,(3j+2)^j) ( j , ( 3 j + 2 ) j ) , with sign ( − 1 ) j (-1)^j ( − 1 ) j and weight
2 j + j ( 3 j + 2 ) = 3 j 2 + 4 j . 2j+j(3j+2)=3j^2+4j. 2 j + j ( 3 j + 2 ) = 3 j 2 + 4 j .
The additional factor q q q in (83) shifts this second weight to 3 j 2 + 4 j + 1 3j^2+4j+1 3 j 2 + 4 j + 1 . Burson’s involutions cancel all remaining pairs, which proves the final assertion. □
Proof of Proposition 5.6. From (5.1),
∑ n ≥ 0 c o ( n ) q n = ∑ m ≥ 0 q 2 m + 1 ( q ; q 2 ) m + 1 ( q 2 ; q 4 ) m + 1 ≡ q ∑ m ≥ 0 q 2 m ( q 2 ; q 4 ) m + 1 ( m o d 2 ) . \sum_{n\geq0}c_o(n)q^n=\sum_{m\geq0}\frac{q^{2m+1}}{(q;q^2)_{m+1}(q^2;q^4)_{m+1}}
\equiv q\sum_{m\geq0}\frac{q^{2m}}{(q^2;q^4)_{m+1}}\pmod{2}. n ≥ 0 ∑ c o ( n ) q n = m ≥ 0 ∑ ( q ; q 2 ) m + 1 ( q 2 ; q 4 ) m + 1 q 2 m + 1 ≡ q m ≥ 0 ∑ ( q 2 ; q 4 ) m + 1 q 2 m ( mod 2 ) .
Modulo 2, the signs in (2.21) do not matter. Replacing q q q there by q 2 q^2 q 2 therefore gives
∑ n ≥ 0 c o ( n ) q n ≡ ∑ j ≥ 0 q 12 j 2 + 8 j + 1 + ∑ j ≥ 0 q 12 j 2 + 16 j + 5 ( m o d 2 ) . \sum_{n\geq0}c_o(n)q^n\equiv\sum_{j\geq0}q^{12j^2+8j+1}+\sum_{j\geq0}q^{12j^2+16j+5}\pmod{2}. n ≥ 0 ∑ c o ( n ) q n ≡ j ≥ 0 ∑ q 12 j 2 + 8 j + 1 + j ≥ 0 ∑ q 12 j 2 + 16 j + 5 ( mod 2 ) .
The two sequences do not overlap. Moreover,
3 ( 12 j 2 + 8 j + 1 ) + 1 = ( 6 j + 2 ) 2 , 3(12j^2+8j+1)+1=(6j+2)^2, 3 ( 12 j 2 + 8 j + 1 ) + 1 = ( 6 j + 2 ) 2 ,
and
3 ( 12 j 2 + 16 j + 5 ) + 1 = ( 6 j + 4 ) 2 . 3(12j^2+16j+5)+1=(6j+4)^2. 3 ( 12 j 2 + 16 j + 5 ) + 1 = ( 6 j + 4 ) 2 .
The first sequence has positive square root 6 j + 2 6j+2 6 j + 2 , and the second has positive square root 6 j + 4 6j+4 6 j + 4 . Conversely, suppose that 3 n + 1 = x 2 3n+1=x^2 3 n + 1 = x 2 with x x x even. Since x 2 ≡ 1 ( m o d 3 ) x^2\equiv1\pmod{3} x 2 ≡ 1 ( mod 3 ) , the integer x x x is not divisible by 3; hence x ≡ 2 x\equiv2 x ≡ 2 or 4 ( m o d 6 ) 4\pmod{6} 4 ( mod 6 ) . Thus x = 6 j + 2 x=6j+2 x = 6 j + 2 or x = 6 j + 4 x=6j+4 x = 6 j + 4 for some j ≥ 0 j\geq0 j ≥ 0 , and n n n belongs to one of the two displayed exponent sequences. Therefore c o ( n ) c_o(n) c o ( n ) is odd exactly when 3 n + 1 3n+1 3 n + 1 is an even square. □
Proof of Corollary 5.7. The case N = 0 N=0 N = 0 is immediate from p 2 , d o ( 0 ) = 0 p_{2,\mathrm{do}}(0)=0 p 2 , do ( 0 ) = 0 . Let N ≥ 1 N\geq1 N ≥ 1 and put n = N − 1 n=N-1 n = N − 1 in (5.10). Then
p 2 , d o ( N ) = 2 c o ( N − 1 ) + s ( N − 1 ) , (84) p_{2,\mathrm{do}}(N)=2c_o(N-1)+s(N-1), \tag*{(84)} p 2 , do ( N ) = 2 c o ( N − 1 ) + s ( N − 1 ) , (84)
where
s ( n ) = [ q n ] ∑ j ≥ 0 ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) . s(n)=[q^n]\sum_{j\geq0}(-1)^jq^{3j^2+2j}(1+q^{2j+1}). s ( n ) = [ q n ] j ≥ 0 ∑ ( − 1 ) j q 3 j 2 + 2 j ( 1 + q 2 j + 1 ) .
Thus s ( n ) s(n) s ( n ) is nonzero only at the two families
A j = 3 j 2 + 2 j , B j = 3 j 2 + 4 j + 1 , A_j=3j^2+2j,\qquad B_j=3j^2+4j+1, A j = 3 j 2 + 2 j , B j = 3 j 2 + 4 j + 1 ,
and in either family its value is ( − 1 ) j (-1)^j ( − 1 ) j .
We first relate these two families to the square condition in the statement. A direct calculation gives
3 ( A j + 1 ) − 2 = ( 3 j + 1 ) 2 , 3 ( B j + 1 ) − 2 = ( 3 j + 2 ) 2 . (85) 3(A_j+1)-2=(3j+1)^2,\qquad3(B_j+1)-2=(3j+2)^2. \tag*{(85)} 3 ( A j + 1 ) − 2 = ( 3 j + 1 ) 2 , 3 ( B j + 1 ) − 2 = ( 3 j + 2 ) 2 . (85)
Conversely, suppose that 3 N − 2 = x 2 3N-2=x^2 3 N − 2 = x 2 for some x > 0 x>0 x > 0 . Since x 2 ≡ 1 ( m o d 3 ) x^2\equiv1\pmod{3} x 2 ≡ 1 ( mod 3 ) , one has 3 ∤ x 3\nmid x 3 ∤ x . Hence either x = 3 j + 1 x=3j+1 x = 3 j + 1 or x = 3 j + 2 x=3j+2 x = 3 j + 2 for a unique j ≥ 0 j\geq0 j ≥ 0 . The two alternatives in (85) then show that, respectively, N − 1 = A j N-1=A_j N − 1 = A j or N − 1 = B j N-1=B_j N − 1 = B j . Therefore the square condition is equivalent to N − 1 N-1 N − 1 belonging to one of the two supports of s s s .
Assume first that N − 1 = A j N-1=A_j N − 1 = A j . Proposition 5.6 applies to
3 A j + 1 = ( 3 j + 1 ) 2 3A_j+1=(3j+1)^2 3 A j + 1 = ( 3 j + 1 ) 2
The number 3 j + 1 3j+1 3 j + 1 is even exactly when j j j is odd. Hence
2 c o ( A j ) ≡ { 0 , j even , 2 , j odd ( m o d 4 ) . 2c_o(A_j)\equiv
\begin{cases}
0, & j\ \text{even},\\
2, & j\ \text{odd}
\end{cases}
\pmod{4}. 2 c o ( A j ) ≡ { 0 , 2 , j even , j odd ( mod 4 ) .
Using (84) and s ( A j ) = ( − 1 ) j s(A_j)=(-1)^j s ( A j ) = ( − 1 ) j , we obtain in both parity cases
p 2 , d o ( A j + 1 ) ≡ 1 ( m o d 4 ) . p_{2,do}(A_j+1)\equiv1\pmod{4}. p 2 , d o ( A j + 1 ) ≡ 1 ( mod 4 ) .
The positive square root in this case is x = 3 j + 1 x=3j+1 x = 3 j + 1 , and therefore χ 3 ( x ) = 1 \chi_3(x)=1 χ 3 ( x ) = 1 .
Now assume that N − 1 = B j N-1=B_j N − 1 = B j . Since
3 B j + 1 = ( 3 j + 2 ) 2 , 3B_j+1=(3j+2)^2, 3 B j + 1 = ( 3 j + 2 ) 2 ,
Proposition 5.6 shows that c o ( B j ) c_o(B_j) c o ( B j ) is odd exactly when 3 j + 2 3j+2 3 j + 2 is even, or equivalently when j j j is even. Thus
2 c o ( B j ) ≡ { 2 , j even , 0 , j odd ( m o d 4 ) . 2c_o(B_j)\equiv
\begin{cases}
2, & j\ \text{even},\\
0, & j\ \text{odd}
\end{cases}
\pmod{4}. 2 c o ( B j ) ≡ { 2 , 0 , j even , j odd ( mod 4 ) .
Together with s ( B j ) = ( − 1 ) j s(B_j)=(-1)^j s ( B j ) = ( − 1 ) j , equation (84) gives
p 2 , d o ( B j + 1 ) ≡ − 1 ( m o d 4 ) . p_{2,do}(B_j+1)\equiv-1\pmod{4}. p 2 , d o ( B j + 1 ) ≡ − 1 ( mod 4 ) .
Here the positive square root is x = 3 j + 2 x=3j+2 x = 3 j + 2 , so χ 3 ( x ) = − 1 \chi_3(x)=-1 χ 3 ( x ) = − 1 .
Finally, if 3 N − 2 3N-2 3 N − 2 is not a square, then N − 1 N-1 N − 1 is neither A j A_j A j nor B j B_j B j ; hence s ( N − 1 ) = 0 s(N-1)=0 s ( N − 1 ) = 0 . Proposition 5.6 also shows that c o ( N − 1 ) c_o(N-1) c o ( N − 1 ) is even, because 3 ( N − 1 ) + 1 = 3 N − 2 3(N-1)+1=3N-2 3 ( N − 1 ) + 1 = 3 N − 2 is not an even square. Equation (84) therefore gives p 2 , d o ( N ) ≡ 0 ( m o d 4 ) p_{2,do}(N)\equiv0\pmod{4} p 2 , d o ( N ) ≡ 0 ( mod 4 ) . This proves the corollary. □
Proof of Corollary 5.8. If N = M n + r N=Mn+r N = M n + r and 3 r + 1 3r+1 3 r + 1 is a nonsquare modulo M M M , then 3 N + 1 3N+1 3 N + 1 cannot be an integer square, so Proposition 5.6 gives the congruence for c o c_o c o . If 3 r − 2 3r-2 3 r − 2 is a nonsquare modulo M M M , then 3 N − 2 3N-2 3 N − 2 cannot be an integer square, so Corollary 5.7 gives the congruence for p 2 , d o p_{2,do} p 2 , d o . □
Proof of Proposition 5.9. Ramanujan’s identity (21) gives
C ν ( q ) = ∑ m ≥ 0 q m ( − q ; q 2 ) m + 1 = ∑ m ≥ 0 ( − 1 ) m q 6 m 2 + 4 m ( 1 + q 4 m + 2 ) = f O ( − q 2 ) . \begin{aligned}
C_\nu(q)=\sum_{m\ge0}\frac{q^m}{(-q;q^2)_{m+1}} \\
&=\sum_{m\ge0}(-1)^m q^{6m^2+4m}(1+q^{4m+2})=f\mathrm{O}(-q^2).
\end{aligned} C ν ( q ) = m ≥ 0 ∑ ( − q ; q 2 ) m + 1 q m = m ≥ 0 ∑ ( − 1 ) m q 6 m 2 + 4 m ( 1 + q 4 m + 2 ) = f O ( − q 2 ) .
It remains to identify the finite-product series with the same false theta function. Apply the Rogers–Fine formula (18) with base q 4 q^4 q 4 and a = q 2 a=q^2 a = q 2 . Then
∑ m ≥ 0 ( q 2 ; q 4 ) m q 2 m = ∑ m ≥ 0 ( − 1 ) m ( q 4 ) m ( 3 m − 1 ) / 2 q 6 m ( 1 + q 4 m + 2 ) = ∑ m ≥ 0 ( − 1 ) m q 6 m 2 + 4 m ( 1 + q 4 m + 2 ) = f O ( − q 2 ) . \begin{aligned}
\sum_{m\geq0}(q^2;q^4)_m q^{2m}=\sum_{m\geq0}(-1)^m(q^4)^{m(3m-1)/2}q^{6m}(1+q^{4m+2}) \\
&=\sum_{m\geq0}(-1)^m q^{6m^2+4m}(1+q^{4m+2})=\mathrm{fO}(-q^2).
\end{aligned} m ≥ 0 ∑ ( q 2 ; q 4 ) m q 2 m = m ≥ 0 ∑ ( − 1 ) m ( q 4 ) m ( 3 m − 1 ) /2 q 6 m ( 1 + q 4 m + 2 ) = m ≥ 0 ∑ ( − 1 ) m q 6 m 2 + 4 m ( 1 + q 4 m + 2 ) = fO ( − q 2 ) .
Thus both series in (5.13) are equal to f O ( − q 2 ) \mathrm{fO}(-q^2) fO ( − q 2 ) . □
Proof of Proposition 5.10. By (5.14), C ψ ( q ) C_{\psi}(q) C ψ ( q ) is the left-hand side of (2.26), while (2.7) identifies its right-hand side as q ω 3 ( q ) q\omega_3(q) q ω 3 ( q ) . This proves (5.15). □
Proofs of the results in Section 6
We first record the boxed modular application, then prove the exceptional parity results and the quartic and cubic root-of-unity identities. The proof of Theorem 6.10 is parallel to that of Theorem 5.4: we isolate an auxiliary series and evaluate it using (2.17) and (2.19). The final corollaries follow from the resulting sparse theta terms and the cubic rank difference.
Proof of Proposition 6.3. In (2.22), take ( m , r ) = ( 2 , 1 ) (m,r)=(2,1) ( m , r ) = ( 2 , 1 ) and u = w u=w u = w , and replace the base q q q by q 1 / 2 q^{1/2} q 1/2 . Its last member becomes
∑ j ≥ 0 ( − 1 ) j w 2 j ( q 1 / 2 ) j ( 2 j + 2 ) = ∑ j ≥ 0 ( − w 2 ) j q j ( j + 1 ) , \sum_{j\geq0}(-1)^j w^{2j}(q^{1/2})^{j(2j+2)}=\sum_{j\geq0}(-w^2)^j q^{j(j+1)}, j ≥ 0 ∑ ( − 1 ) j w 2 j ( q 1/2 ) j ( 2 j + 2 ) = j ≥ 0 ∑ ( − w 2 ) j q j ( j + 1 ) ,
which is (6.5). The fixed-point assertion is part of Burson’s construction recorded in (2.23). □
Proof of Theorem 6.4. Write
A ( z ; q ) = ∑ j ≥ 0 ( − z ) j q j ( j + 1 ) / 2 , B ( z ; q ) = ∑ j ≥ 0 ( − z ) j q j ( j + 1 ) , A(z;q)=\sum_{j\geq0}(-z)^j q^{j(j+1)/2},\qquad B(z;q)=\sum_{j\geq0}(-z)^j q^{j(j+1)}, A ( z ; q ) = j ≥ 0 ∑ ( − z ) j q j ( j + 1 ) /2 , B ( z ; q ) = j ≥ 0 ∑ ( − z ) j q j ( j + 1 ) ,
and
Q ( z ; q ) = ( z q ; q 2 ) ∞ ( z − 1 q ; q 2 ) ∞ . Q(z;q)=(zq;q^2)_\infty(z^{-1}q;q^2)_\infty. Q ( z ; q ) = ( z q ; q 2 ) ∞ ( z − 1 q ; q 2 ) ∞ .
Then (6.4) is
( 1 + z − 1 ) D ( z ; q ) = A ( z ; q ) ( q ; q 2 ) ∞ Q ( z ; q ) − B ( z ; q ) . (1+z^{-1})D(z;q)=\frac{A(z;q)}{(q;q^2)_\infty Q(z;q)}-B(z;q). ( 1 + z − 1 ) D ( z ; q ) = ( q ; q 2 ) ∞ Q ( z ; q ) A ( z ; q ) − B ( z ; q ) .
Both sides are analytic in a neighborhood of z = − 1 z=-1 z = − 1 . Differentiating and then setting z = − 1 z=-1 z = − 1 gives
− D ( − 1 ; q ) = d d z ( A ( z ; q ) ( q ; q 2 ) ∞ Q ( z ; q ) ) ∣ z = − 1 − B ′ ( − 1 ; q ) . (86) -D(-1;q)=\left.\frac{d}{dz}\left(\frac{A(z;q)}{(q;q^2)_\infty Q(z;q)}\right)\right|_{z=-1}-B'(-1;q). \tag*{(86)} − D ( − 1 ; q ) = d z d ( ( q ; q 2 ) ∞ Q ( z ; q ) A ( z ; q ) ) z = − 1 − B ′ ( − 1 ; q ) . (86)
Now
A ( − 1 ; q ) = G ( q ) , A ′ ( − 1 ; q ) = − G 1 ( q ) , B ′ ( − 1 ; q ) = − G 1 ( q 2 ) . A(-1;q)=G(q),\qquad A'(-1;q)=-G_1(q),\qquad B'(-1;q)=-G_1(q^2). A ( − 1 ; q ) = G ( q ) , A ′ ( − 1 ; q ) = − G 1 ( q ) , B ′ ( − 1 ; q ) = − G 1 ( q 2 ) .
Moreover, Q ( z ; q ) = Q ( z − 1 ; q ) Q(z;q)=Q(z^{-1};q) Q ( z ; q ) = Q ( z − 1 ; q ) , and therefore Q ′ ( − 1 ; q ) = 0 Q'(-1;q)=0 Q ′ ( − 1 ; q ) = 0 . Finally, setting z = − 1 z=-1 z = − 1 in (54) shows that
G ( q ) ( q ; q 2 ) ∞ Q ( − 1 ; q ) = G ( q 2 ) . \frac{G(q)}{(q;q^2)_\infty Q(-1;q)}=G(q^2). ( q ; q 2 ) ∞ Q ( − 1 ; q ) G ( q ) = G ( q 2 ) .
Substitution in (86) yields
− D ( − 1 ; q ) = − G ( q 2 ) G ( q ) G 1 ( q ) + G 1 ( q 2 ) . -\mathcal{D}(-1;q)=-\frac{G(q^2)}{G(q)}G_1(q)+G_1(q^2). − D ( − 1 ; q ) = − G ( q ) G ( q 2 ) G 1 ( q ) + G 1 ( q 2 ) .
which is (57) . Finally, Gauss’s identity (15) gives
G ( q 2 ) G ( q ) = ( q ; q 2 ) ∞ ( q 2 ; q 4 ) ∞ 2 , \frac{G(q^2)}{G(q)}=\frac{(q;q^2)_\infty}{(q^2;q^4)^2_\infty}, G ( q ) G ( q 2 ) = ( q 2 ; q 4 ) ∞ 2 ( q ; q 2 ) ∞ ,
and hence (58) .
Proof of Theorem 6.5 . Starting from (52) , set z = − 1 z=-1 z = − 1 and then replace q q q by − q -q − q . Since
( q ; − q ) 2 m = ( q ; q 2 ) m ( − q 2 ; q 2 ) m , (q;-q)_{2m}=(q;q^2)_m(-q^2;q^2)_m, ( q ; − q ) 2 m = ( q ; q 2 ) m ( − q 2 ; q 2 ) m ,
we obtain
− D ( − 1 ; − q ) = ∑ m ≥ 0 q 2 m + 1 ( q ; − q ) 2 m ( q ; q 2 ) m + 1 2 = ∑ m ≥ 0 q 2 m + 1 ( − q 2 ; q 2 ) m ( q ; q 2 ) m ( 1 − q 2 m + 1 ) 2 , \begin{aligned}
-\mathcal{D}(-1;-q)=\sum_{m\geq0}\frac{q^{2m+1}(q;-q)_{2m}}{(q;q^2)_{m+1}^2} \\
&=\sum_{m\geq0}\frac{q^{2m+1}(-q^2;q^2)_m}{(q;q^2)_m(1-q^{2m+1})^2},
\end{aligned} − D ( − 1 ; − q ) = m ≥ 0 ∑ ( q ; q 2 ) m + 1 2 q 2 m + 1 ( q ; − q ) 2 m = m ≥ 0 ∑ ( q ; q 2 ) m ( 1 − q 2 m + 1 ) 2 q 2 m + 1 ( − q 2 ; q 2 ) m ,
which proves (59) . For each m ≥ 0 m\geq0 m ≥ 0 , the polynomial ( − q 2 ; q 2 ) m (-q^2;q^2)_m ( − q 2 ; q 2 ) m has non-negative coefficients, and so do the reciprocal products 1 / ( q ; q 2 ) m 1/(q;q^2)_m 1/ ( q ; q 2 ) m and 1 / ( 1 − q 2 m + 1 ) 2 1/(1-q^{2m+1})^2 1/ ( 1 − q 2 m + 1 ) 2 . Hence every summand on the right-hand side of (59) has nonnegative coefficients. More precisely, the summand with m = 0 m=0 m = 0 is
q ( 1 − q ) 2 = ∑ n ≥ 1 n q n . \frac{q}{(1-q)^2}=\sum_{n\geq1}nq^n. ( 1 − q ) 2 q = n ≥ 1 ∑ n q n .
Therefore, for every n ≥ 1 n\geq1 n ≥ 1 , the coefficient of q n q^n q n in the full right-hand side is at least n n n , and is consequently strictly positive. On the other hand,
− D ( − 1 ; − q ) = ∑ n ≥ 0 ( − 1 ) n + 1 ( d 0 ( 2 ) ( n ) − d 1 ( 2 ) ( n ) ) q n . -\mathcal{D}(-1;-q)=\sum_{n\geq0}(-1)^{n+1}\left(d_0^{(2)}(n)-d_1^{(2)}(n)\right)q^n. − D ( − 1 ; − q ) = n ≥ 0 ∑ ( − 1 ) n + 1 ( d 0 ( 2 ) ( n ) − d 1 ( 2 ) ( n ) ) q n .
Comparing the coefficient of q n q^n q n for each n ≥ 1 n\geq1 n ≥ 1 proves (60) .
Proof of Theorem 6.6 . We first treat z = i z=i z = i . Put
H 4 ( q ) : = ( q ; q 2 ) ∞ ( − q 2 ; q 4 ) ∞ . H_4(q):=(q;q^2)_\infty(-q^2;q^4)_\infty. H 4 ( q ) := ( q ; q 2 ) ∞ ( − q 2 ; q 4 ) ∞ .
Since
( i q ; q 2 ) ∞ ( − i q ; q 2 ) ∞ = ( − q 2 ; q 4 ) ∞ , (iq;q^2)_\infty(-iq;q^2)_\infty=(-q^2;q^4)_\infty, ( i q ; q 2 ) ∞ ( − i q ; q 2 ) ∞ = ( − q 2 ; q 4 ) ∞ ,
identity (54) gives
D ( i ; q ) = 1 + i 2 ( A i ( q ) H 4 ( q ) − A i ( q 2 ) ) , A i ( q ) : = ∑ j ≥ 0 ( − i ) j q T j . (87) \mathcal{D}(i;q)=\frac{1+i}{2}\left(\frac{A_i(q)}{H_4(q)}-A_i(q^2)\right),\qquad A_i(q):=\sum_{j\geq0}(-i)^jq^{T_j}. \tag*{(87)} D ( i ; q ) = 2 1 + i ( H 4 ( q ) A i ( q ) − A i ( q 2 ) ) , A i ( q ) := j ≥ 0 ∑ ( − i ) j q T j . (87)
Define the bilateral theta series
Θ 4 ( q ) : = ∑ n ∈ Z ( − 1 ) n q n ( 2 n + 1 ) . \Theta_4(q) := \sum_{n\in\mathbb{Z}} (-1)^n q^{n(2n+1)}. Θ 4 ( q ) := n ∈ Z ∑ ( − 1 ) n q n ( 2 n + 1 ) .
Jacobi’s triple product (16) , with base q 4 q^4 q 4 and x = q 3 x=q^3 x = q 3 , gives
Θ 4 ( q ) = ( q ; q 2 ) ∞ ( q 4 ; q 4 ) ∞ . (88) \Theta_4(q)=(q;q^2)_\infty(q^4;q^4)_\infty. \tag*{(88)} Θ 4 ( q ) = ( q ; q 2 ) ∞ ( q 4 ; q 4 ) ∞ . (88)
Consequently,
Θ 4 ( q ) = H 4 ( q ) Θ 4 ( q 2 ) , (89) \Theta_4(q)=H_4(q)\Theta_4(q^2), \tag*{(89)} Θ 4 ( q ) = H 4 ( q ) Θ 4 ( q 2 ) , (89)
because
( − q 2 ; q 4 ) ∞ ( q 2 ; q 4 ) ∞ = ( q 4 ; q 8 ) ∞ . (-q^2;q^4)_\infty(q^2;q^4)_\infty=(q^4;q^8)_\infty. ( − q 2 ; q 4 ) ∞ ( q 2 ; q 4 ) ∞ = ( q 4 ; q 8 ) ∞ .
Writing the negative terms in the bilateral sum with nonnegative indices, we have
Θ 4 ( q ) = ∑ n ≥ 0 ( − 1 ) n ( q T 2 n − q T 2 n + 1 ) . \Theta_4(q)=\sum_{n\geq0}(-1)^n\left(q^{T_{2n}}-q^{T_{2n+1}}\right). Θ 4 ( q ) = n ≥ 0 ∑ ( − 1 ) n ( q T 2 n − q T 2 n + 1 ) .
On the other hand, separating the even and odd indices gives
A i ( q ) = ∑ n ≥ 0 ( − 1 ) n ( q T 2 n − i q T 2 n + 1 ) . A_i(q)=\sum_{n\geq0}(-1)^n\left(q^{T_{2n}}-iq^{T_{2n+1}}\right). A i ( q ) = n ≥ 0 ∑ ( − 1 ) n ( q T 2 n − i q T 2 n + 1 ) .
Combining these two expressions with the definition of Φ 4 ( q ) \Phi_4(q) Φ 4 ( q ) gives
1 + i 2 A i ( q ) = Φ 4 ( q ) + − 1 + i 2 Θ 4 ( q ) . (90) \frac{1+i}{2}A_i(q)=\Phi_4(q)+\frac{-1+i}{2}\Theta_4(q). \tag*{(90)} 2 1 + i A i ( q ) = Φ 4 ( q ) + 2 − 1 + i Θ 4 ( q ) . (90)
Substituting (90) into (87) and using (89) , the theta-product terms cancel, leaving (63) .
We next take z = ω z=\omega z = ω , where 1 + ω + ω 2 = 0 1+\omega+\omega^2=0 1 + ω + ω 2 = 0 . Since
( 1 − ω x ) ( 1 − ω 2 x ) = 1 + x + x 2 = 1 − x 3 1 − x , (1-\omega x)(1-\omega^2x)=1+x+x^2=\frac{1-x^3}{1-x}, ( 1 − ω x ) ( 1 − ω 2 x ) = 1 + x + x 2 = 1 − x 1 − x 3 ,
we have
( q ; q 2 ) ∞ ( ω q ; q 2 ) ∞ ( ω 2 q ; q 2 ) ∞ = ( q 3 ; q 6 ) ∞ . (91) (q;q^2)_\infty(\omega q;q^2)_\infty(\omega^2q;q^2)_\infty=(q^3;q^6)_\infty. \tag*{(91)} ( q ; q 2 ) ∞ ( ω q ; q 2 ) ∞ ( ω 2 q ; q 2 ) ∞ = ( q 3 ; q 6 ) ∞ . (91)
Thus (54) becomes
D ( ω ; q ) = − ω 2 ( A ω ( q ) ( q 3 ; q 6 ) ∞ − A ω ( q 2 ) ) , A ω ( q ) : = ∑ j ≥ 0 ( − ω ) j q T j . (92) \mathcal{D}(\omega;q)=-\omega^2\left(\frac{A_\omega(q)}{(q^3;q^6)_\infty}-A_\omega(q^2)\right),\qquad A_\omega(q):=\sum_{j\geq0}(-\omega)^j q^{T_j}. \tag*{(92)} D ( ω ; q ) = − ω 2 ( ( q 3 ; q 6 ) ∞ A ω ( q ) − A ω ( q 2 ) ) , A ω ( q ) := j ≥ 0 ∑ ( − ω ) j q T j . (92)
Introduce
Θ 3 ( q ) : = ∑ n ∈ Z ( − 1 ) n q T 3 n . \Theta_3(q):=\sum_{n\in\mathbb{Z}}(-1)^nq^{T_{3n}}. Θ 3 ( q ) := n ∈ Z ∑ ( − 1 ) n q T 3 n .
Jacobi’s triple product (16) , with base q 9 q^9 q 9 and x = q 6 x=q^6 x = q 6 , gives
Θ 3 ( q ) = ( q 3 ; q 9 ) ∞ = ( q 3 ; q 6 ) ∞ Θ 3 ( q 2 ) . (93) \Theta_3(q)=(q^3;q^9)_\infty=(q^3;q^6)_\infty\Theta_3(q^2). \tag*{(93)} Θ 3 ( q ) = ( q 3 ; q 9 ) ∞ = ( q 3 ; q 6 ) ∞ Θ 3 ( q 2 ) . (93)
Moreover,
Θ 3 ( q ) = ∑ n ≥ 0 ( − 1 ) n ( q T 3 n − q T 3 n + 2 ) . \Theta_3(q)=\sum_{n\geq0}(-1)^n\left(q^{T_{3n}}-q^{T_{3n+2}}\right). Θ 3 ( q ) = n ≥ 0 ∑ ( − 1 ) n ( q T 3 n − q T 3 n + 2 ) .
To compare A ω ( q ) A_{\omega}(q) A ω ( q ) with these series, write j = 3 m + r j=3m+r j = 3 m + r . Since ( − ω ) 3 = − 1 (-\omega)^3=-1 ( − ω ) 3 = − 1 , the three resulting coefficients are
− ω 2 ( − ω ) 3 m = − ω 2 ( − 1 ) m , − ω 2 ( − ω ) 3 m + 1 = ( − 1 ) m , -\omega^2(-\omega)^{3m}=-\omega^2(-1)^m,\qquad-\omega^2(-\omega)^{3m+1}=(-1)^m, − ω 2 ( − ω ) 3 m = − ω 2 ( − 1 ) m , − ω 2 ( − ω ) 3 m + 1 = ( − 1 ) m ,
and
− ω 2 ( − ω ) 3 m + 2 = − ω ( − 1 ) m . -\omega^2(-\omega)^{3m+2}=-\omega(-1)^m. − ω 2 ( − ω ) 3 m + 2 = − ω ( − 1 ) m .
Using − ω 2 = 1 + ω -\omega^2=1+\omega − ω 2 = 1 + ω , and then grouping the terms with T 3 m T_{3m} T 3 m , T 3 m + 1 T_{3m+1} T 3 m + 1 , and T 3 m + 2 T_{3m+2} T 3 m + 2 , gives
− ω 2 A ω ( q ) = Φ 3 ( q ) + ω Θ 3 ( q ) . (94) -\omega^2 A_{\omega}(q)=\Phi_3(q)+\omega\Theta_3(q). \tag*{(94)} − ω 2 A ω ( q ) = Φ 3 ( q ) + ω Θ 3 ( q ) . (94)
Substitution in (92) , followed by (93) , cancels the bilateral theta terms and proves (64) .
Finally, the symmetry d ( r , n ) = d ( − r , n ) d(r,n)=d(-r,n) d ( r , n ) = d ( − r , n ) gives
D ( i ; q ) = ∑ n ≥ 0 ( d 0 ( 4 ) ( n ) − d 2 ( 4 ) ( n ) ) q n . \mathcal{D}(i;q)=\sum_{n\geq0}\left(d_0^{(4)}(n)-d_2^{(4)}(n)\right)q^n. D ( i ; q ) = n ≥ 0 ∑ ( d 0 ( 4 ) ( n ) − d 2 ( 4 ) ( n ) ) q n .
and, since ω + ω 2 = − 1 \omega+\omega^2=-1 ω + ω 2 = − 1 ,
D ( ω ; q ) = ∑ n ≥ 0 ( d 0 ( 3 ) ( n ) − d 1 ( 3 ) ( n ) ) q n . \mathcal{D}(\omega;q)=\sum_{n\geq0}\left(d_0^{(3)}(n)-d_1^{(3)}(n)\right)q^n. D ( ω ; q ) = n ≥ 0 ∑ ( d 0 ( 3 ) ( n ) − d 1 ( 3 ) ( n ) ) q n .
Proof of Corollary 6.7. The reciprocal of ( q 3 ; q 6 ) ∞ (q^3;q^6)_\infty ( q 3 ; q 6 ) ∞ is supported on multiples of 3. The exponents T 3 m T_{3m} T 3 m are congruent to 0 or 3 modulo 6, and the exponents T 3 m + 1 T_{3m+1} T 3 m + 1 are congruent to 1 or 4 modulo 6. Hence the first term on the right of (64) is supported only on the residue classes 0, 1, 3, 4 modulo 6. The second term is supported only on 0 and 2 modulo 6. Therefore the coefficient of q 6 n + 5 q^{6n+5} q 6 n + 5 in D ( ω ; q ) \mathcal{D}(\omega;q) D ( ω ; q ) is zero. Since d 1 ( 3 ) ( 6 n + 5 ) = d 2 ( 3 ) ( 6 n + 5 ) d_1^{(3)}(6n+5)=d_2^{(3)}(6n+5) d 1 ( 3 ) ( 6 n + 5 ) = d 2 ( 3 ) ( 6 n + 5 ) , we obtain
d 0 ( 3 ) ( 6 n + 5 ) = d 1 ( 3 ) ( 6 n + 5 ) = d 2 ( 3 ) ( 6 n + 5 ) . d_0^{(3)}(6n+5)=d_1^{(3)}(6n+5)=d_2^{(3)}(6n+5). d 0 ( 3 ) ( 6 n + 5 ) = d 1 ( 3 ) ( 6 n + 5 ) = d 2 ( 3 ) ( 6 n + 5 ) .
Their sum is c 3 , o ( 6 n + 5 ) c_{3,o}(6n+5) c 3 , o ( 6 n + 5 ) , which proves (65) .
For the equivalent divisibility statement, put
P n ( z ) = ∑ r ∈ Z d ( r , 6 n + 5 ) z r . P_n(z)=\sum_{r\in\mathbb{Z}}d(r,6n+5)z^r. P n ( z ) = r ∈ Z ∑ d ( r , 6 n + 5 ) z r .
The preceding equality gives P n ( ω ) = 0 P_n(\omega)=0 P n ( ω ) = 0 . Since P n ( z ) P_n(z) P n ( z ) has integer coefficients and 1 + z + z 2 1+z+z^2 1 + z + z 2 is the minimal polynomial of ω \omega ω over Q \mathbb{Q} Q , the Laurent polynomial P n ( z ) P_n(z) P n ( z ) is divisible by 1 + z + z 2 1+z+z^2 1 + z + z 2 in Z [ z , z − 1 ] \mathbb{Z}[z,z^{-1}] Z [ z , z − 1 ] . □
Proof of Corollary 6.8. Let N ≡ 2 ( m o d 6 ) N\equiv2\pmod{6} N ≡ 2 ( mod 6 ) . As in the preceding proof, the first term on the right of (64) is supported only on the residue classes 0, 1, 3, 4 modulo 6, and hence contributes nothing to the coefficient of q N q^N q N . It remains to examine − Φ 3 ( q 2 ) -\Phi_3(q^2) − Φ 3 ( q 2 ) . By the definition of Φ 3 \Phi_3 Φ 3 ,
Φ 3 ( q 2 ) = ∑ m ≥ 0 ( − 1 ) m ( q 2 T 3 m + q 2 T 3 m + 1 ) . \Phi_3(q^2)=\sum_{m\geq0}(-1)^m\left(q^{2T_{3m}}+q^{2T_{3m+1}}\right). Φ 3 ( q 2 ) = m ≥ 0 ∑ ( − 1 ) m ( q 2 T 3 m + q 2 T 3 m + 1 ) .
The exponents 2 T 3 m = 3 m ( 3 m + 1 ) 2T_{3m}=3m(3m+1) 2 T 3 m = 3 m ( 3 m + 1 ) are congruent to 0 0 0 modulo 6 6 6 , whereas
2 T 3 m + 1 = ( 3 m + 1 ) ( 3 m + 2 ) ≡ 2 ( m o d 6 ) . 2T_{3m+1}=(3m+1)(3m+2)\equiv2 \pmod{6}. 2 T 3 m + 1 = ( 3 m + 1 ) ( 3 m + 2 ) ≡ 2 ( mod 6 ) .
Therefore a term of degree N ≡ 2 ( m o d 6 ) N\equiv2\pmod{6} N ≡ 2 ( mod 6 ) occurs in Φ 3 ( q 2 ) \Phi_3(q^2) Φ 3 ( q 2 ) precisely when
N = ( 3 m + 1 ) ( 3 m + 2 ) N=(3m+1)(3m+2) N = ( 3 m + 1 ) ( 3 m + 2 )
for some m ≥ 0 m\geq0 m ≥ 0 , and its coefficient is then ( − 1 ) m (-1)^m ( − 1 ) m . The minus sign in (64) changes this coefficient to ( − 1 ) m + 1 (-1)^{m+1} ( − 1 ) m + 1 . If no such m m m exists, the coefficient is zero. This proves (66) . □
The three-dimensional positive decomposition and sparse correction are governed by the following analogous parity splitting.
Lemma 10.1. Put
Y 1 ( q ) = ( − q ; q ) ∞ 3 ∑ m ≥ 0 ( q ; q 2 ) m q m ( − q ; q 2 ) m , Y_1(q)=(-q;q)_\infty^3\sum_{m\geq0}\frac{(q;q^2)_m q^m}{(-q;q^2)_m}, Y 1 ( q ) = ( − q ; q ) ∞ 3 m ≥ 0 ∑ ( − q ; q 2 ) m ( q ; q 2 ) m q m , and
Y 2 ( q ) = ( q 2 ; q 2 ) ∞ ( q ; q ) ∞ ∑ m ≥ 0 ( q ; q 2 ) m q m ( q ; q ) m 2 . Y_2(q)=(q^2;q^2)_\infty(q;q)_\infty\sum_{m\geq0}\frac{(q;q^2)_m q^m}{(q;q)_m^2}. Y 2 ( q ) = ( q 2 ; q 2 ) ∞ ( q ; q ) ∞ m ≥ 0 ∑ ( q ; q ) m 2 ( q ; q 2 ) m q m . Then
Y 1 ( q ) = E 3 , △ ( q ) + 2 C 3 , o ( q ) , (95) Y_1(q)=E_{3,\triangle}(q)+2C_{3,o}(q), \tag*{(95)} Y 1 ( q ) = E 3 , △ ( q ) + 2 C 3 , o ( q ) , (95) Y 2 ( q ) = E 3 , △ ( q ) − 2 C 3 , o ( q ) . (96) Y_2(q)=E_{3,\triangle}(q)-2C_{3,o}(q). \tag*{(96)} Y 2 ( q ) = E 3 , △ ( q ) − 2 C 3 , o ( q ) . (96) In particular,
2 C 3 , o ( q ) = 1 2 Y 1 ( q ) − 1 2 Y 2 ( q ) . 2C_{3,o}(q)=\frac{1}{2}Y_1(q)-\frac{1}{2}Y_2(q). 2 C 3 , o ( q ) = 2 1 Y 1 ( q ) − 2 1 Y 2 ( q ) .
Proof. Let H r ( q ) H_r(q) H r ( q ) be the series in (77) and put
J r ( q ) : = ( − 1 ; q ) r H r ( q ) . J_r(q):=(-1;q)_rH_r(q). J r ( q ) := ( − 1 ; q ) r H r ( q ) .
Interchanging the sums in (77) and applying the q q q -binomial theorem (13) with a = − 1 a=-1 a = − 1 gives
∑ r ≥ 0 ( − 1 ; q ) r q r ( m + 1 ) ( q ; q ) r = ( − q m + 1 ; q ) ∞ ( q m + 1 ; q ) ∞ , \sum_{r\geq0}\frac{(-1;q)_r q^{r(m+1)}}{(q;q)_r}
=\frac{(-q^{m+1};q)_\infty}{(q^{m+1};q)_\infty}, r ≥ 0 ∑ ( q ; q ) r ( − 1 ; q ) r q r ( m + 1 ) = ( q m + 1 ; q ) ∞ ( − q m + 1 ; q ) ∞ ,
whereas replacing q m + 1 q^{m+1} q m + 1 by − q m + 1 -q^{m+1} − q m + 1 gives
∑ r ≥ 0 ( − 1 ; q ) r ( − q m + 1 ) r ( q ; q ) r = ( q m + 1 ; q ) ∞ ( − q m + 1 ; q ) ∞ . \sum_{r\geq0}\frac{(-1;q)_r(-q^{m+1})^r}{(q;q)_r}
=\frac{(q^{m+1};q)_\infty}{(-q^{m+1};q)_\infty}. r ≥ 0 ∑ ( q ; q ) r ( − 1 ; q ) r ( − q m + 1 ) r = ( − q m + 1 ; q ) ∞ ( q m + 1 ; q ) ∞ .
Using
( q 2 ; q 2 ) m = ( q ; q ) m ( − q ; q ) m , ( − q ; q ) ∞ = 1 ( q ; q 2 ) ∞ , (q^2;q^2)_m=(q;q)_m(-q;q)_m,\qquad
(-q;q)_\infty=\frac{1}{(q;q^2)_\infty}, ( q 2 ; q 2 ) m = ( q ; q ) m ( − q ; q ) m , ( − q ; q ) ∞ = ( q ; q 2 ) ∞ 1 ,
the two resulting expressions simplify to
Y 1 ( q ) = ∑ r ≥ 0 J r ( q ) , Y 2 ( q ) = ∑ r ≥ 0 ( − 1 ) r J r ( q ) . (97) Y_1(q)=\sum_{r\geq0}J_r(q),\qquad
Y_2(q)=\sum_{r\geq0}(-1)^rJ_r(q). \tag*{(97)} Y 1 ( q ) = r ≥ 0 ∑ J r ( q ) , Y 2 ( q ) = r ≥ 0 ∑ ( − 1 ) r J r ( q ) . (97)
The formulas (79) and (80) now identify the odd and even parts. For r = 2 k + 1 r=2k+1 r = 2 k + 1 ,
J 2 k + 1 ( q ) = ( − 1 ; q ) 2 k + 1 q 2 k + 1 ( q ; q 2 ) k + 1 2 = 2 ( − q ; q ) 2 k q 2 k + 1 ( q ; q 2 ) k + 1 2 . J_{2k+1}(q)=(-1;q)_{2k+1}\frac{q^{2k+1}}{(q;q^2)_{k+1}^{2}}
=\;2(-q;q)_{2k}\frac{q^{2k+1}}{(q;q^2)_{k+1}^{2}}. J 2 k + 1 ( q ) = ( − 1 ; q ) 2 k + 1 ( q ; q 2 ) k + 1 2 q 2 k + 1 = 2 ( − q ; q ) 2 k ( q ; q 2 ) k + 1 2 q 2 k + 1 .
Thus
∑ k ≥ 0 J 2 k + 1 ( q ) = 2 C 3 , o ( q ) . (98) \sum_{k\geq0}J_{2k+1}(q)=2C_{3,o}(q). \tag*{(98)} k ≥ 0 ∑ J 2 k + 1 ( q ) = 2 C 3 , o ( q ) . (98)
For r = 2 k r=2k r = 2 k ,
J 2 k ( q ) = ( − 1 ; q ) 2 k G ( q ) 2 q 2 k ( q 2 ; q 2 ) k 2 , J_{2k}(q)=(-1;q)_{2k}G(q)^2\frac{q^{2k}}{(q^2;q^2)_k^2}, J 2 k ( q ) = ( − 1 ; q ) 2 k G ( q ) 2 ( q 2 ; q 2 ) k 2 q 2 k ,
and therefore, by (68) ,
∑ k ≥ 0 J 2 k ( q ) = E 3 , △ ( q ) . (99) \sum_{k\geq0}J_{2k}(q)=E_{3,\triangle}(q). \tag*{(99)} k ≥ 0 ∑ J 2 k ( q ) = E 3 , △ ( q ) . (99)
Separating the even and odd indices in (97) , and then using (98) and (99) , proves (95) and (96) . □
Lemma 10.2. We have
Y 2 ( q ) = ∑ j ≥ 0 ( − 1 ) j q j ( j + 1 ) . Y_2(q)=\sum_{j\geq0}(-1)^j q^{j(j+1)}. Y 2 ( q ) = j ≥ 0 ∑ ( − 1 ) j q j ( j + 1 ) .
Proof. Again use the formal half-power convention from § 2 . Since
( q ; q 2 ) n = ( q 1 / 2 ; q ) n ( − q 1 / 2 ; q ) n , (q;q^2)_n=(q^{1/2};q)_n(-q^{1/2};q)_n, ( q ; q 2 ) n = ( q 1/2 ; q ) n ( − q 1/2 ; q ) n ,
apply Heine’s transformation (17) with
( a , b , c , z ) = ( − q 1 / 2 , q 1 / 2 , q , q ) . (a,b,c,z)=(-q^{1/2},q^{1/2},q,q). ( a , b , c , z ) = ( − q 1/2 , q 1/2 , q , q ) .
Then
Y 2 ( q ) = ( q 2 ; q 2 ) ∞ ( q ; q ) ∞ ∑ n ≥ 0 ( q 2 ; q 2 ) n q n ( q ; q 2 ) n 2 = ∑ n ≥ 0 ( q 1 / 2 ; q ) n q n / 2 ( − q 1 / 2 ; q ) n + 1 . Y_2(q)=(q^2;q^2)_\infty(q;q)_\infty\sum_{n\geq0}\frac{(q^2;q^2)_nq^n}{(q;q^2)_n^2}
=\sum_{n\geq0}\frac{(q^{1/2};q)_nq^{n/2}}{(-q^{1/2};q)_{n+1}}. Y 2 ( q ) = ( q 2 ; q 2 ) ∞ ( q ; q ) ∞ n ≥ 0 ∑ ( q ; q 2 ) n 2 ( q 2 ; q 2 ) n q n = n ≥ 0 ∑ ( − q 1/2 ; q ) n + 1 ( q 1/2 ; q ) n q n /2 .
The prefactor cancels for the same reason as in Lemma 9.2 . Finally, replacing q q q by q 1 / 2 q^{1/2} q 1/2 in (19) gives
∑ n ≥ 0 ( q 1 / 2 ; q ) n q n / 2 ( − q 1 / 2 ; q ) n + 1 = ∑ j ≥ 0 ( − 1 ) j q j ( j + 1 ) . \sum_{n\geq0}\frac{(q^{1/2};q)_nq^{n/2}}{(-q^{1/2};q)_{n+1}}
=\sum_{j\geq0}(-1)^j q^{j(j+1)}. n ≥ 0 ∑ ( − q 1/2 ; q ) n + 1 ( q 1/2 ; q ) n q n /2 = j ≥ 0 ∑ ( − 1 ) j q j ( j + 1 ) .
Proof of Theorem 6.9 . From (67) ,
∑ n ≥ 0 p 3 , d o ( n ) q n = q ( − q ; q ) ∞ 3 ∑ m ≥ 0 ( q ; q 2 ) m q m ( − q ; q 2 ) m = q Y 1 ( q ) . \sum_{n\ge0} p_{3,\mathrm{do}}(n)q^n=q(-q;q)_{\infty}^{3}\sum_{m\ge0}\frac{(q;q^2)_m q^m}{(-q;q^2)_m}=qY_1(q). n ≥ 0 ∑ p 3 , do ( n ) q n = q ( − q ; q ) ∞ 3 m ≥ 0 ∑ ( − q ; q 2 ) m ( q ; q 2 ) m q m = q Y 1 ( q ) .
Since p 3 , d o ( 0 ) = 0 p_{3,\mathrm{do}}(0)=0 p 3 , do ( 0 ) = 0 , division by q q q gives
∑ N ≥ 0 p 3 , d o ( N + 1 ) q N = Y 1 ( q ) . \sum_{N\ge0}p_{3,\mathrm{do}}(N+1)q^N=Y_1(q). N ≥ 0 ∑ p 3 , do ( N + 1 ) q N = Y 1 ( q ) .
Equation (95) now yields
∑ N ≥ 0 p 3 , d o ( N + 1 ) q N = 2 C 3 , o ( q ) + E 3 , △ ( q ) , \sum_{N\ge0}p_{3,\mathrm{do}}(N+1)q^N=2C_{3,o}(q)+E_{3,\triangle}(q), N ≥ 0 ∑ p 3 , do ( N + 1 ) q N = 2 C 3 , o ( q ) + E 3 , △ ( q ) ,
which is (69) . Comparing coefficients gives (70) .
Proof of Theorem 6.10 . The preceding proof gives
∑ N ≥ 0 p 3 , d o ( N + 1 ) q N = Y 1 ( q ) . \sum_{N\ge0}p_{3,\mathrm{do}}(N+1)q^N=Y_1(q). N ≥ 0 ∑ p 3 , do ( N + 1 ) q N = Y 1 ( q ) .
By (95) and (96) , Y 1 ( q ) − 4 C 3 , o ( q ) = Y 2 ( q ) Y_1(q)-4C_{3,o}(q)=Y_2(q) Y 1 ( q ) − 4 C 3 , o ( q ) = Y 2 ( q ) . Lemma 10.2 therefore gives
∑ N ≥ 0 ( p 3 , d o ( N + 1 ) − 4 c 3 , o ( N ) ) q N = ∑ j ≥ 0 ( − 1 ) j q j ( j + 1 ) . \sum_{N\ge0}\left(p_{3,\mathrm{do}}(N+1)-4c_{3,o}(N)\right)q^N=\sum_{j\ge0}(-1)^j q^{j(j+1)}. N ≥ 0 ∑ ( p 3 , do ( N + 1 ) − 4 c 3 , o ( N ) ) q N = j ≥ 0 ∑ ( − 1 ) j q j ( j + 1 ) .
The theorem follows by comparing coefficients.
Proof of Corollary 6.11 . From the positive decomposition (70) ,
e 3 , △ ( N ) = p 3 , d o ( N + 1 ) − 2 c 3 , o ( N ) . e_{3,\triangle}(N)=p_{3,\mathrm{do}}(N+1)-2c_{3,o}(N). e 3 , △ ( N ) = p 3 , do ( N + 1 ) − 2 c 3 , o ( N ) .
Subtracting another 2 c 3 , o ( N ) 2c_{3,o}(N) 2 c 3 , o ( N ) from both sides gives
e 3 , △ ( N ) − 2 c 3 , o ( N ) = p 3 , d o ( N + 1 ) − 4 c 3 , o ( N ) . (100) e_{3,\triangle}(N)-2c_{3,o}(N)=p_{3,\mathrm{do}}(N+1)-4c_{3,o}(N). \tag*{(100)} e 3 , △ ( N ) − 2 c 3 , o ( N ) = p 3 , do ( N + 1 ) − 4 c 3 , o ( N ) . (100)
Next specialize Burson’s identity (22) by taking ( m , r ) = ( 2 , 1 ) (m,r)=(2,1) ( m , r ) = ( 2 , 1 ) and u = 1 u=1 u = 1 . We obtain
∑ M ≥ 0 ( ∑ ( k , π ) ∈ B 2 , 1 ( M ) ( − 1 ) ν ( π ‾ ) ) q M = ∑ j ≥ 0 ( − 1 ) j q j ( 2 j + 2 ) = ∑ j ≥ 0 ( − 1 ) j q 2 j ( j + 1 ) . \sum_{M\ge0}\left(\sum_{(k,\pi)\in B_{2,1}(M)}(-1)^{\nu(\overline{\pi})}\right)q^M
=\sum_{j\ge0}(-1)^j q^{j(2j+2)}
=\sum_{j\ge0}(-1)^j q^{2j(j+1)}. M ≥ 0 ∑ ( k , π ) ∈ B 2 , 1 ( M ) ∑ ( − 1 ) ν ( π ) q M = j ≥ 0 ∑ ( − 1 ) j q j ( 2 j + 2 ) = j ≥ 0 ∑ ( − 1 ) j q 2 j ( j + 1 ) .
Taking the coefficient of q 2 N q^{2N} q 2 N gives
∑ ( k , π ) ∈ B 2 , 1 ( 2 N ) ( − 1 ) ν ( π ‾ ) = [ q 2 N ] ∑ j ≥ 0 ( − 1 ) j q 2 j ( j + 1 ) = [ q N ] ∑ j ≥ 0 ( − 1 ) j q j ( j + 1 ) . \sum_{(k,\pi)\in B_{2,1}(2N)}(-1)^{\nu(\overline{\pi})}
=[q^{2N}]\sum_{j\ge0}(-1)^j q^{2j(j+1)}
=[q^N]\sum_{j\ge0}(-1)^j q^{j(j+1)}. ( k , π ) ∈ B 2 , 1 ( 2 N ) ∑ ( − 1 ) ν ( π ) = [ q 2 N ] j ≥ 0 ∑ ( − 1 ) j q 2 j ( j + 1 ) = [ q N ] j ≥ 0 ∑ ( − 1 ) j q j ( j + 1 ) .
By Theorem 6.10 , the last coefficient is p 3 , d o ( N + 1 ) − 4 c 3 , o ( N ) p_{3,\mathrm{do}}(N+1)-4c_{3,o}(N) p 3 , do ( N + 1 ) − 4 c 3 , o ( N ) . Together with (100) , this proves all equalities in (72) .
For completeness, Burson’s involution pairs every nonfixed object with an object of the same weight and opposite sign. Its fixed point indexed by j j j is the pair ( j , ( 2 j + 1 ) j ) (j,(2j+1)^j) ( j , ( 2 j + 1 ) j ) . In the present specialization its total boxed weight is
j + j ( 2 j + 1 ) = 2 j ( j + 1 ) , j+j(2j+1)=2j(j+1), j + j ( 2 j + 1 ) = 2 j ( j + 1 ) ,
and its sign is ( − 1 ) j (-1)^j ( − 1 ) j . Hence a fixed point contributes to the coefficient of q 2 N q^{2N} q 2 N exactly when N = j ( j + 1 ) N=j(j+1) N = j ( j + 1 ) , in agreement with the sparse correction in Theorem 6.10 . □
Proof of Corollary 6.12. The case N = 0 N=0 N = 0 is immediate from the convention p 3 , d o ( 0 ) = 0 p_{3,do}(0)=0 p 3 , d o ( 0 ) = 0 . Now let N ≥ 1 N\geq1 N ≥ 1 . Replacing n n n by N − 1 N-1 N − 1 in (71) gives
p 3 , d o ( N ) = 4 c 3 , o ( N − 1 ) + ε ( N − 1 ) , (101) p_{3,do}(N)=4c_{3,o}(N-1)+\varepsilon(N-1), \tag*{(101)} p 3 , d o ( N ) = 4 c 3 , o ( N − 1 ) + ε ( N − 1 ) , (101)
where
ε ( K ) = { ( − 1 ) j , K = j ( j + 1 ) for some j ≥ 0 , 0 , otherwise . \varepsilon(K)=
\begin{cases}
(-1)^j, & K=j(j+1)\text{ for some }j\geq0,\\
0, & \text{otherwise}.
\end{cases} ε ( K ) = { ( − 1 ) j , 0 , K = j ( j + 1 ) for some j ≥ 0 , otherwise .
Reducing (101) modulo 4 eliminates the first term and yields
p 3 , d o ( N ) ≡ ε ( N − 1 ) ( m o d 4 ) . p_{3,do}(N)\equiv\varepsilon(N-1)\pmod{4}. p 3 , d o ( N ) ≡ ε ( N − 1 ) ( mod 4 ) .
The condition N − 1 = j ( j + 1 ) N-1=j(j+1) N − 1 = j ( j + 1 ) is equivalent to
4 N − 3 = 4 j ( j + 1 ) + 1 = ( 2 j + 1 ) 2 . 4N-3=4j(j+1)+1=(2j+1)^2. 4 N − 3 = 4 j ( j + 1 ) + 1 = ( 2 j + 1 ) 2 .
Conversely, if 4 N − 3 = x 2 4N-3=x^2 4 N − 3 = x 2 for a positive integer x x x , then x x x is odd. Writing x = 2 j + 1 x=2j+1 x = 2 j + 1 with j ≥ 0 j\geq0 j ≥ 0 gives
N − 1 = x 2 − 1 4 = j ( j + 1 ) . N-1=\frac{x^2-1}{4}=j(j+1). N − 1 = 4 x 2 − 1 = j ( j + 1 ) .
Thus the sparse correction is nonzero exactly when 4 N − 3 4N-3 4 N − 3 is a positive square.
In that case, with x = 2 j + 1 x=2j+1 x = 2 j + 1 , one has x ≡ 1 ( m o d 4 ) x\equiv1\pmod{4} x ≡ 1 ( mod 4 ) when j j j is even and x ≡ 3 ( m o d 4 ) x\equiv3\pmod{4} x ≡ 3 ( mod 4 ) when j j j is odd. Therefore
χ 4 ( x ) = ( − 1 ) j = ε ( N − 1 ) , \chi_4(x)=(-1)^j=\varepsilon(N-1), χ 4 ( x ) = ( − 1 ) j = ε ( N − 1 ) ,
and hence p 3 , d o ( N ) ≡ χ 4 ( x ) ( m o d 4 ) p_{3,do}(N)\equiv\chi_4(x)\pmod{4} p 3 , d o ( N ) ≡ χ 4 ( x ) ( mod 4 ) . If 4 N − 3 4N-3 4 N − 3 is not a square, then ε ( N − 1 ) = 0 \varepsilon(N-1)=0 ε ( N − 1 ) = 0 , so (101) gives p 3 , d o ( N ) ≡ 0 ( m o d 4 ) p_{3,do}(N)\equiv0\pmod{4} p 3 , d o ( N ) ≡ 0 ( mod 4 ) . This proves both cases of the corollary. □
Proof of Corollary 6.13. We first prove (73) . By the symmetry obtained by interchanging the two side arms,
d 1 ( 3 ) ( N ) = d 2 ( 3 ) ( N ) . d_1^{(3)}(N)=d_2^{(3)}(N). d 1 ( 3 ) ( N ) = d 2 ( 3 ) ( N ) .
Since the three rank classes exhaust all restricted odd three-dimensional corners of weight N N N , it follows that
c 3 , o ( N ) = d 0 ( 3 ) ( N ) + d 1 ( 3 ) ( N ) + d 2 ( 3 ) ( N ) = d 0 ( 3 ) ( N ) + 2 d 1 ( 3 ) ( N ) ≡ d 0 ( 3 ) ( N ) − d 1 ( 3 ) ( N ) ( m o d 3 ) . \begin{aligned}
c_{3,o}(N)&=d_0^{(3)}(N)+d_1^{(3)}(N)+d_2^{(3)}(N)\\
&=d_0^{(3)}(N)+2d_1^{(3)}(N)\\
&\equiv d_0^{(3)}(N)-d_1^{(3)}(N)\pmod{3}.
\end{aligned} c 3 , o ( N ) = d 0 ( 3 ) ( N ) + d 1 ( 3 ) ( N ) + d 2 ( 3 ) ( N ) = d 0 ( 3 ) ( N ) + 2 d 1 ( 3 ) ( N ) ≡ d 0 ( 3 ) ( N ) − d 1 ( 3 ) ( N ) ( mod 3 ) .
For N ≡ 2 ( m o d 6 ) N \equiv2 \pmod{6} N ≡ 2 ( mod 6 ) , Corollary 6.8 therefore gives
c 3 , o ( N ) ≡ { ( − 1 ) m + 1 , N = ( 3 m + 1 ) ( 3 m + 2 ) for some m ≥ 0 , 0 , otherwise ( m o d 3 ) , c_{3,o}(N) \equiv\begin{cases}
(-1)^{m+1}, & N=(3m+1)(3m+2)\text{ for some }m\ge0,\\
0, & \text{otherwise}
\end{cases} \pmod{3}, c 3 , o ( N ) ≡ { ( − 1 ) m + 1 , 0 , N = ( 3 m + 1 ) ( 3 m + 2 ) for some m ≥ 0 , otherwise ( mod 3 ) ,
which is (73) .
We next prove (74) . Let N ≡ 3 ( m o d 6 ) N \equiv3 \pmod{6} N ≡ 3 ( mod 6 ) and put K = N − 1 K=N-1 K = N − 1 , so that K ≡ 2 ( m o d 6 ) K \equiv2 \pmod{6} K ≡ 2 ( mod 6 ) . Applying (71) with K K K in place of n n n gives
p 3 , d o ( N ) = 4 c 3 , o ( K ) + ε ( K ) , (102) p_{3,do}(N)=4c_{3,o}(K)+\varepsilon(K), \tag*{(102)} p 3 , d o ( N ) = 4 c 3 , o ( K ) + ε ( K ) , (102)
where ε ( K ) = ( − 1 ) j \varepsilon(K)=(-1)^j ε ( K ) = ( − 1 ) j if K = j ( j + 1 ) K=j(j+1) K = j ( j + 1 ) and is zero otherwise.
Suppose first that
K = ( 3 m + 1 ) ( 3 m + 2 ) K=(3m+1)(3m+2) K = ( 3 m + 1 ) ( 3 m + 2 )
for some m ≥ 0 m\ge0 m ≥ 0 . Then K = j ( j + 1 ) K=j(j+1) K = j ( j + 1 ) with j = 3 m + 1 j=3m+1 j = 3 m + 1 , and therefore
ε ( K ) = ( − 1 ) 3 m + 1 = ( − 1 ) m + 1 . \varepsilon(K)=(-1)^{3m+1}=(-1)^{m+1}. ε ( K ) = ( − 1 ) 3 m + 1 = ( − 1 ) m + 1 .
The first part of the proof also gives
c 3 , o ( K ) ≡ ( − 1 ) m + 1 ( m o d 3 ) . c_{3,o}(K)\equiv(-1)^{m+1}\pmod{3}. c 3 , o ( K ) ≡ ( − 1 ) m + 1 ( mod 3 ) .
Thus there is an integer t t t such that
c 3 , o ( K ) = ( − 1 ) m + 1 + 3 t . c_{3,o}(K)=(-1)^{m+1}+3t. c 3 , o ( K ) = ( − 1 ) m + 1 + 3 t .
Substitution in (102) yields
p 3 , d o ( N ) = 5 ( − 1 ) m + 1 + 12 t , p_{3,do}(N)=5(-1)^{m+1}+12t, p 3 , d o ( N ) = 5 ( − 1 ) m + 1 + 12 t ,
and hence
p 3 , d o ( N ) ≡ 5 ( − 1 ) m + 1 ( m o d 12 ) . p_{3,do}(N)\equiv5(-1)^{m+1}\pmod{12}. p 3 , d o ( N ) ≡ 5 ( − 1 ) m + 1 ( mod 12 ) .
Now suppose that K K K is not of the displayed form. Because K ≡ 2 ( m o d 6 ) K\equiv2\pmod{6} K ≡ 2 ( mod 6 ) , any representation K = j ( j + 1 ) K=j(j+1) K = j ( j + 1 ) would force j ( j + 1 ) ≡ 2 ( m o d 3 ) j(j+1)\equiv2\pmod{3} j ( j + 1 ) ≡ 2 ( mod 3 ) , and hence j ≡ 1 ( m o d 3 ) j\equiv1\pmod{3} j ≡ 1 ( mod 3 ) . Thus one would have j = 3 m + 1 j=3m+1 j = 3 m + 1 and consequently K = ( 3 m + 1 ) ( 3 m + 2 ) K=(3m+1)(3m+2) K = ( 3 m + 1 ) ( 3 m + 2 ) , contrary to the assumption. Therefore ε ( K ) = 0 \varepsilon(K)=0 ε ( K ) = 0 . The first part of the proof gives c 3 , o ( K ) ≡ 0 ( m o d 3 ) c_{3,o}(K)\equiv0\pmod{3} c 3 , o ( K ) ≡ 0 ( mod 3 ) , so 4 c 3 , o ( K ) 4c_{3,o}(K) 4 c 3 , o ( K ) is divisible by 12. Equation (102) now gives p 3 , d o ( N ) ≡ 0 ( m o d 12 ) p_{3,do}(N)\equiv0\pmod{12} p 3 , d o ( N ) ≡ 0 ( mod 12 ) . This completes the proof. □
The proofs also explain the coefficients in the two discrepancy formulas. The factor 2 in Theorem 5.4 comes from the odd-index projector ( 1 − ( − 1 ) r ) / 2 (1-(-1)^r)/2 ( 1 − ( − 1 ) r ) /2 in Lemma 9.1 . In the three-dimensional case, the same projector is accompanied by the additional factor 2 in
( − 1 ; q ) 2 m + 1 = 2 ( − q ; q ) 2 m , (-1;q)_{2m+1}=2(-q;q)_{2m}, ( − 1 ; q ) 2 m + 1 = 2 ( − q ; q ) 2 m ,
which produces the coefficient 4 in Theorem 6.10 . Thus these coefficients arise from the parity-splitting transformations and are not arbitrary normalizations.
Proofs of the results in Section 7
We give the proofs in the order in which the principal results are stated. The first auxiliary lemma shows that absence from the binary support also excludes the unary correction support.
Lemma 11.1. For every M ≥ 2 M \ge2 M ≥ 2 , one has S M ⊆ R M S_M \subseteq R_M S M ⊆ R M .
Proof. Fix j j j . There is an odd nonnegative integer t t t satisfying t ≡ j t \equiv j t ≡ j or t ≡ − j − 1 ( m o d M ) t \equiv-j-1 \pmod{M} t ≡ − j − 1 ( mod M ) . If M M M is odd, adding M M M changes parity; if M M M is even, the two residue classes have opposite parity. Write t = 2 a + 1 t=2a+1 t = 2 a + 1 and take b = t b=t b = t . Then
T 2 a + 1 + T b = 2 T t = t ( t + 1 ) ≡ j ( j + 1 ) ( m o d M ) . T_{2a+1}+T_b=2T_t=t(t+1)\equiv j(j+1)\pmod{M}. T 2 a + 1 + T b = 2 T t = t ( t + 1 ) ≡ j ( j + 1 ) ( mod M ) .
Thus every residue represented by j ( j + 1 ) j(j+1) j ( j + 1 ) is also represented by T 2 a + 1 + T b T_{2a+1}+T_b T 2 a + 1 + T b . □
Proof of Theorem 7.1. Set
C 3 , o ( q ) = ∑ N ≥ 0 c 3 , o ( N ) q N , P ( q ) = ( q ; q 2 ) ∞ , C_{3,o}(q)=\sum_{N\geq0}c_{3,o}(N)q^N,\qquad P(q)=(q;q^2)_\infty, C 3 , o ( q ) = N ≥ 0 ∑ c 3 , o ( N ) q N , P ( q ) = ( q ; q 2 ) ∞ ,
and
A ( q ) = ∑ j ≥ 0 ( − 1 ) j q j ( j + 1 ) / 2 , B ( q ) = ∑ j ≥ 0 ( − 1 ) j q j ( j + 1 ) . A(q)=\sum_{j\geq0}(-1)^j q^{j(j+1)/2},\qquad B(q)=\sum_{j\geq0}(-1)^j q^{j(j+1)}. A ( q ) = j ≥ 0 ∑ ( − 1 ) j q j ( j + 1 ) /2 , B ( q ) = j ≥ 0 ∑ ( − 1 ) j q j ( j + 1 ) .
Putting a = 1 a=1 a = 1 in (25) and using (51) gives
2 C 3 , o ( q ) = A ( q ) P ( q ) 3 − B ( q ) . (103) 2C_{3,o}(q)=\frac{A(q)}{P(q)^3}-B(q). \tag*{(103)} 2 C 3 , o ( q ) = P ( q ) 3 A ( q ) − B ( q ) . (103)
We first prove the congruences modulo 3. Since
P ( q ) 3 = ( q ; q 2 ) ∞ 3 ≡ ( q 3 ; q 6 ) ∞ ( m o d 3 ) , P(q)^3=(q;q^2)_\infty^3\equiv(q^3;q^6)_\infty\pmod{3}, P ( q ) 3 = ( q ; q 2 ) ∞ 3 ≡ ( q 3 ; q 6 ) ∞ ( mod 3 ) ,
we have
A ( q ) P ( q ) 3 ≡ A ( q ) ( q 3 ; q 6 ) ∞ ( m o d 3 ) . \frac{A(q)}{P(q)^3}\equiv\frac{A(q)}{(q^3;q^6)_\infty}\pmod{3}. P ( q ) 3 A ( q ) ≡ ( q 3 ; q 6 ) ∞ A ( q ) ( mod 3 ) .
The reciprocal of ( q 3 ; q 6 ) ∞ (q^3;q^6)_\infty ( q 3 ; q 6 ) ∞ contributes only powers divisible by 3, while
j ( j + 1 ) 2 ≡ 0 , 1 ( m o d 3 ) . \frac{j(j+1)}{2}\equiv0,1\pmod{3}. 2 j ( j + 1 ) ≡ 0 , 1 ( mod 3 ) .
Hence A ( q ) / P ( q ) 3 A(q)/P(q)^3 A ( q ) / P ( q ) 3 has zero coefficient modulo 3 in every degree congruent to 2 modulo 3. Suppose now that 3 ∣ M 3\mid M 3 ∣ M , r ≡ 2 ( m o d 3 ) r\equiv2\pmod{3} r ≡ 2 ( mod 3 ) , and r ∉ S M r\notin S_M r ∈ / S M . Every integer M n + r Mn+r M n + r is congruent to 2 modulo 3, and B ( q ) B(q) B ( q ) has no term in any degree congruent to r r r modulo M M M . Taking the coefficient of q M n + r q^{Mn+r} q M n + r in (103) gives
2 c 3 , o ( M n + r ) ≡ 0 ( m o d 3 ) , 2c_{3,o}(Mn+r)\equiv0\pmod{3}, 2 c 3 , o ( M n + r ) ≡ 0 ( mod 3 ) ,
and therefore
c 3 , o ( M n + r ) ≡ 0 ( m o d 3 ) . c_{3,o}(Mn+r)\equiv0\pmod{3}. c 3 , o ( M n + r ) ≡ 0 ( mod 3 ) .
We next prove the congruence modulo 2. Let
U ( q ) = ∑ j ≥ 0 q j ( j + 1 ) / 2 U(q)=\sum_{j\geq0}q^{j(j+1)/2} U ( q ) = j ≥ 0 ∑ q j ( j + 1 ) /2
and let
V ( q ) = ∑ j ≥ 0 q ( 2 j + 1 ) ( 2 j + 2 ) / 2 V(q)=\sum_{j\geq0}q^{(2j+1)(2j+2)/2} V ( q ) = j ≥ 0 ∑ q ( 2 j + 1 ) ( 2 j + 2 ) /2
be its odd-index subseries. Then A ( q ) = U ( q ) − 2 V ( q ) A(q)=U(q)-2V(q) A ( q ) = U ( q ) − 2 V ( q ) . By (15) ,
U ( q ) = ( q 2 ; q 2 ) ∞ P ( q ) , U(q)=\frac{(q^2;q^2)_\infty}{P(q)}, U ( q ) = P ( q ) ( q 2 ; q 2 ) ∞ ,
so
U ( q ) P ( q ) 3 = ( q 2 ; q 2 ) ∞ P ( q ) 4 . \frac{U(q)}{P(q)^3}=\frac{(q^2;q^2)_\infty}{P(q)^4}. P ( q ) 3 U ( q ) = P ( q ) 4 ( q 2 ; q 2 ) ∞ .
Modulo 4,
P ( q ) 4 ≡ ( q 2 ; q 4 ) ∞ 2 , P(q)^4\equiv(q^2;q^4)_\infty^2, P ( q ) 4 ≡ ( q 2 ; q 4 ) ∞ 2 ,
and hence
U ( q ) P ( q ) 3 ≡ ( q 2 ; q 2 ) ∞ ( q 2 ; q 4 ) ∞ 2 = ( q 4 ; q 4 ) ∞ ( q 4 ; q 4 ) ∞ = U ( q 2 ) ( m o d 4 ) . \frac{U(q)}{P(q)^3}\equiv\frac{(q^2;q^2)_\infty}{(q^2;q^4)_\infty^2}=\frac{(q^4;q^4)_\infty}{(q^4;q^4)_\infty}=U(q^2)\pmod{4}. P ( q ) 3 U ( q ) ≡ ( q 2 ; q 4 ) ∞ 2 ( q 2 ; q 2 ) ∞ = ( q 4 ; q 4 ) ∞ ( q 4 ; q 4 ) ∞ = U ( q 2 ) ( mod 4 ) .
Moreover,
1 P ( q ) 3 ≡ U ( q ) ( m o d 2 ) . \frac{1}{P(q)^3}\equiv U(q)\pmod{2}. P ( q ) 3 1 ≡ U ( q ) ( mod 2 ) .
Indeed, by (15) this is equivalent to
1 ( q ; q 2 ) ∞ 2 ≡ ( q 2 ; q 2 ) ∞ ( m o d 2 ) , \frac{1}{(q;q^2)_\infty^2}\equiv(q^2;q^2)_\infty\pmod{2}, ( q ; q 2 ) ∞ 2 1 ≡ ( q 2 ; q 2 ) ∞ ( mod 2 ) ,
which follows from
( q ; q 2 ) ∞ 2 ≡ ( q 2 ; q 4 ) ∞ ( m o d 2 ) (q;q^2)_\infty^2\equiv(q^2;q^4)_\infty\pmod{2} ( q ; q 2 ) ∞ 2 ≡ ( q 2 ; q 4 ) ∞ ( mod 2 )
and
1 ( q 2 ; q 4 ) ∞ ≡ ( q 2 ; q 2 ) ∞ ( m o d 2 ) . \frac{1}{(q^2;q^4)_\infty}\equiv(q^2;q^2)_\infty\pmod{2}. ( q 2 ; q 4 ) ∞ 1 ≡ ( q 2 ; q 2 ) ∞ ( mod 2 ) .
Consequently,
A ( q ) P ( q ) 3 ≡ U ( q 2 ) − 2 V ( q ) U ( q ) ( m o d 4 ) . (104) \frac{A(q)}{P(q)^3}\equiv U(q^2)-2V(q)U(q)\pmod{4}. \tag*{(104)} P ( q ) 3 A ( q ) ≡ U ( q 2 ) − 2 V ( q ) U ( q ) ( mod 4 ) . (104)
The support modulo M M M of U ( q 2 ) U(q^2) U ( q 2 ) is S M S_M S M , and that of V ( q ) U ( q ) V(q)U(q) V ( q ) U ( q ) is R M R_M R M . If r ∉ R M r\notin R_M r ∈ / R M , Lemma 11.1 also gives r ∉ S M r\notin S_M r ∈ / S M . Thus neither term on the right of (104) , nor the series B ( q ) B(q) B ( q ) in (103) , contributes in a degree congruent to r r r modulo M M M . It follows that
2 c 3 , o ( M n + r ) ≡ 0 ( m o d 4 ) , 2c_{3,o}(Mn+r)\equiv0\pmod{4}, 2 c 3 , o ( M n + r ) ≡ 0 ( mod 4 ) ,
and therefore
c 3 , o ( M n + r ) ≡ 0 ( m o d 2 ) . c_{3,o}(Mn+r)\equiv0\pmod{2}. c 3 , o ( M n + r ) ≡ 0 ( mod 2 ) .
It remains to transfer these conclusions to the three-color companion. By Theorem 6.10 ,
p 3 , d o ( M n + r + 1 ) = 4 c 3 , o ( M n + r ) + ε ( M n + r ) , (105) p_{3,do}(Mn+r+1)=4c_{3,o}(Mn+r)+\varepsilon(Mn+r), \tag*{(105)} p 3 , d o ( M n + r + 1 ) = 4 c 3 , o ( M n + r ) + ε ( M n + r ) , (105)
where ε ( K ) \varepsilon(K) ε ( K ) is zero unless K = j ( j + 1 ) K=j(j+1) K = j ( j + 1 ) for some j ≥ 0 j\ge0 j ≥ 0 . If r ∉ R M r\notin R_M r ∈ / R M , then r ∉ S M r\notin S_M r ∈ / S M by Lemma 11.1 ; hence the correction in (105) vanishes and the parity congruence for c 3 , o ( M n + r ) c_{3,o}(M n+r) c 3 , o ( M n + r ) gives
p 3 , d o ( M n + r + 1 ) ≡ 0 ( m o d 8 ) . p_{3,do}(M n+r+1)\equiv0\pmod{8}. p 3 , d o ( M n + r + 1 ) ≡ 0 ( mod 8 ) .
If 3 ∣ M 3\mid M 3 ∣ M , r ≡ 2 ( m o d 3 ) r\equiv2\pmod{3} r ≡ 2 ( mod 3 ) , and r ∉ S M r\notin S_M r ∈ / S M , the correction again vanishes and the congruence modulo 3 3 3 gives
p 3 , d o ( M n + r + 1 ) ≡ 0 ( m o d 12 ) . p_{3,do}(M n+r+1)\equiv0\pmod{12}. p 3 , d o ( M n + r + 1 ) ≡ 0 ( mod 12 ) .
When both hypotheses hold, the corresponding congruences combine to moduli 6 6 6 and 24 24 24 . This proves the theorem. □
We now prove Theorem 7.2 by reducing the support problem to prime powers. The next three lemmas give the local–global reduction, the power-of-two case, and the required criterion at odd prime powers.
Lemma 11.2. Let M 1 , … , M k M_1,\ldots,M_k M 1 , … , M k be pairwise coprime positive integers and M = M 1 ⋯ M k M=M_1\cdots M_k M = M 1 ⋯ M k . A residue r r r belongs to R M R_M R M if and only if its reduction belongs to R M i R_{M_i} R M i for every i i i .
Proof. The forward implication is immediate. Conversely, choose a i , b i a_i,b_i a i , b i such that
T 2 a i + 1 + T b i ≡ r ( m o d M i ) . T_{2a_i+1}+T_{b_i}\equiv r\pmod{M_i}. T 2 a i + 1 + T b i ≡ r ( mod M i ) .
Since
T 2 a + 1 = 2 a 2 + 3 a + 1 T_{2a+1}=2a^2+3a+1 T 2 a + 1 = 2 a 2 + 3 a + 1
is an integral polynomial in a a a , the ordinary Chinese remainder theorem provides a a a whose reduction modulo each M i M_i M i is a i a_i a i . The triangular number T b T_b T b has period 2 M i 2M_i 2 M i modulo M i M_i M i . For every odd M i M_i M i , replacing b i b_i b i by b i + M i b_i+M_i b i + M i preserves T b i T_{b_i} T b i modulo M i M_i M i and reverses the parity of b i b_i b i . We may therefore choose the representatives b i b_i b i to have one common parity. The generalized Chinese remainder theorem then solves
b ≡ b i ( m o d 2 M i ) . b\equiv b_i\pmod{2M_i}. b ≡ b i ( mod 2 M i ) .
simultaneously, because the pairwise greatest common divisors of the moduli 2 M i 2M_i 2 M i are 2 2 2 and the residues have the same parity. This a , b a,b a , b represents r r r modulo every M i M_i M i , hence modulo M M M . □
Lemma 11.3. For every α ≥ 1 \alpha\ge1 α ≥ 1 , the set R 2 α R_{2^\alpha} R 2 α is the full residue system modulo 2 α 2^\alpha 2 α .
Proof. We first recall that triangular numbers are surjective modulo 2 α 2^\alpha 2 α . For a prescribed residue r r r , the integer 8 r + 1 8r+1 8 r + 1 is congruent to 1 1 1 modulo 8 8 8 . Every integer congruent to 1 1 1 modulo 8 8 8 is an odd square modulo every higher power of 2 2 2 : starting with a solution modulo 8 8 8 , a solution x x x modulo 2 k 2^k 2 k lifts to either x x x or x + 2 k − 1 x+2^{k-1} x + 2 k − 1 modulo 2 k + 1 2^{k+1} 2 k + 1 , since these two squares differ by 2 k 2^k 2 k modulo 2 k + 1 2^{k+1} 2 k + 1 . Hence there is an odd x x x with
x 2 ≡ 8 r + 1 ( m o d 2 α + 3 ) . x^2\equiv8r+1\pmod{2^{\alpha+3}}. x 2 ≡ 8 r + 1 ( mod 2 α + 3 ) .
Writing x = 2 b + 1 x = 2b + 1 x = 2 b + 1 and dividing by 8 8 8 gives T b ≡ r ( m o d 2 α ) T_b \equiv r \pmod{2^\alpha} T b ≡ r ( mod 2 α ) . Now take a = 2 α − 1 a = 2^\alpha- 1 a = 2 α − 1 . Then
T 2 a + 1 = T 2 α + 1 − 1 = 2 α ( 2 α + 1 − 1 ) ≡ 0 ( m o d 2 α ) , T_{2a+1} = T_{2^{\alpha+1}-1} = 2^\alpha(2^{\alpha+1}-1) \equiv0 \pmod{2^\alpha}, T 2 a + 1 = T 2 α + 1 − 1 = 2 α ( 2 α + 1 − 1 ) ≡ 0 ( mod 2 α ) ,
so every residue belongs to R 2 α R_{2^\alpha} R 2 α .
Lemma 11.4. Let p p p be an odd prime and α ≥ 1 \alpha\ge1 α ≥ 1 . If p ≡ 1 ( m o d 4 ) p \equiv1 \pmod4 p ≡ 1 ( mod 4 ) , then every residue modulo p α p^\alpha p α is a sum of two squares. If p ≡ 3 ( m o d 4 ) p \equiv3 \pmod4 p ≡ 3 ( mod 4 ) and t = v p ( A ) < α t = v_p(A) < \alpha t = v p ( A ) < α , then A A A is a sum of two squares modulo p α p^\alpha p α if and only if t t t is even; residues divisible by p α p^\alpha p α are also sums of two squares.
Proof. Suppose first that p ≡ 1 ( m o d 4 ) p \equiv1 \pmod4 p ≡ 1 ( mod 4 ) . Then − 1 -1 − 1 has a square root i i i modulo p p p , and Hensel lifting gives i 2 ≡ − 1 ( m o d p α ) i^2 \equiv-1 \pmod{p^\alpha} i 2 ≡ − 1 ( mod p α ) . For any residue A A A , put
x = A + 1 2 , y = A − 1 2 i ( m o d p α ) . x = \frac{A+1}{2}, \qquad y = \frac{A-1}{2i} \pmod{p^\alpha}. x = 2 A + 1 , y = 2 i A − 1 ( mod p α ) .
Then x 2 + y 2 ≡ A ( m o d p α ) x^2+y^2 \equiv A \pmod{p^\alpha} x 2 + y 2 ≡ A ( mod p α ) .
Now suppose p ≡ 3 ( m o d 4 ) p \equiv3 \pmod4 p ≡ 3 ( mod 4 ) . Every unit U U U modulo p β p^\beta p β is a sum of two squares. Modulo p p p , the two sets { x 2 } \{x^2\} { x 2 } and { U − y 2 } \{U-y^2\} { U − y 2 } each have ( p + 1 ) / 2 (p+1)/2 ( p + 1 ) /2 elements and hence intersect. The resulting solution has at least one nonzero partial derivative, so it lifts to every p β p^\beta p β .
If A = p 2 k U A = p^{2k}U A = p 2 k U with U U U a unit and 2 k < α 2k < \alpha 2 k < α , represent U U U modulo p α − 2 k p^{\alpha-2k} p α − 2 k as X 2 + Y 2 X^2+Y^2 X 2 + Y 2 and multiply by p 2 k p^{2k} p 2 k . This proves representability when v p ( A ) v_p(A) v p ( A ) is even, while A ≡ 0 ( m o d p α ) A \equiv0 \pmod{p^\alpha} A ≡ 0 ( mod p α ) is represented by 0 2 + 0 2 0^2+0^2 0 2 + 0 2 .
Conversely, suppose x 2 + y 2 ≡ A ( m o d p α ) x^2+y^2 \equiv A \pmod{p^\alpha} x 2 + y 2 ≡ A ( mod p α ) and t = v p ( A ) < α t = v_p(A) < \alpha t = v p ( A ) < α . Let k = min ( v p ( x ) , v p ( y ) ) k = \min(v_p(x),v_p(y)) k = min ( v p ( x ) , v p ( y )) . After dividing by p k p^k p k , at least one of the two new variables is nonzero modulo p p p , and their sum of squares is nonzero modulo p p p , since − 1 -1 − 1 is a nonsquare modulo p p p . Hence v p ( x 2 + y 2 ) = 2 k v_p(x^2+y^2)=2k v p ( x 2 + y 2 ) = 2 k . Since x 2 + y 2 ≡ A ( m o d p α ) x^2+y^2 \equiv A \pmod{p^\alpha} x 2 + y 2 ≡ A ( mod p α ) and t < α t < \alpha t < α , we have t = 2 k t=2k t = 2 k , which is even. □
Proof of Theorem 7.2 . By Lemma 11.2 , it is enough to classify the image modulo prime powers. Lemma 11.3 shows that the power of 2 2 2 creates no obstruction.
Let p p p be odd. Since 2 2 2 and 4 4 4 are invertible modulo p α p^\alpha p α , one has
r ∈ R p α ⟺ 8 r + 2 is a sum of two squares modulo p α . r \in R_{p^\alpha} \Longleftrightarrow8r+2\text{ is a sum of two squares modulo }p^\alpha. r ∈ R p α ⟺ 8 r + 2 is a sum of two squares modulo p α .
Indeed,
8 T 2 u + 1 + 1 = ( 4 u + 3 ) 2 , 8 T v + 1 = ( 2 v + 1 ) 2 , 8T_{2u+1}+1=(4u+3)^2,\qquad8T_v+1=(2v+1)^2, 8 T 2 u + 1 + 1 = ( 4 u + 3 ) 2 , 8 T v + 1 = ( 2 v + 1 ) 2 ,
and the residues 4 u + 3 4u+3 4 u + 3 and 2 v + 1 2v+1 2 v + 1 run through all classes modulo p α p^\alpha p α .
If p ≡ 1 ( m o d 4 ) p \equiv1 \pmod4 p ≡ 1 ( mod 4 ) , Lemma 11.4 shows that R p α R_{p^\alpha} R p α is full. If p ≡ 3 ( m o d 4 ) p \equiv3 \pmod4 p ≡ 3 ( mod 4 ) , the same lemma shows that 8 r + 2 = 2 ( 4 r + 1 ) 8r+2=2(4r+1) 8 r + 2 = 2 ( 4 r + 1 ) fails to be a sum of two squares exactly when v p ( 4 r + 1 ) v_p(4r+1) v p ( 4 r + 1 ) is odd and less than α \alpha α . Combining the local conditions by Lemma 11.2 proves the equivalence.
If every prime p ≡ 3 ( m o d 4 ) p \equiv3 \pmod4 p ≡ 3 ( mod 4 ) has exponent at most 1 1 1 , no local obstruction is possible. Conversely, if p α ∥ M p^\alpha\mathrel{\Vert} M p α ∥ M with p ≡ 3 ( m o d 4 ) p \equiv3 \pmod4 p ≡ 3 ( mod 4 ) and α ≥ 2 \alpha\ge2 α ≥ 2 , choose r r r modulo p α p^\alpha p α so that 4 r + 1 ≡ p ( m o d p α ) 4r+1 \equiv p \pmod{p^\alpha} 4 r + 1 ≡ p ( mod p α ) and choose its residues arbitrarily at the other prime-power factors. The Chinese remainder theorem then gives a class modulo M M M with v p ( 4 r + 1 ) = 1 < α v_p(4r+1)=1<\alpha v p ( 4 r + 1 ) = 1 < α , so R M R_M R M is not full. □
Proof of Corollary 7.3. The valuation hypothesis and Theorem 7.2 give r ∉ R p α r \notin R_{p^\alpha} r ∈ / R p α . Part (1) of Theorem 7.1 therefore gives the congruences modulo 2 and 8.
If p = 3 p=3 p = 3 , the divisibility 3 ∣ 4 r + 1 3 \mid4r+1 3 ∣ 4 r + 1 implies r ≡ 2 ( m o d 3 ) r \equiv2 \pmod{3} r ≡ 2 ( mod 3 ) . Moreover, r ∉ R 3 α r \notin R_{3^\alpha} r ∈ / R 3 α and Lemma 11.1 imply r ∉ S 3 α r \notin S_{3^\alpha} r ∈ / S 3 α . Part (2) of Theorem 7.1 then gives the congruences modulo 3 and 12. Combining the two pairs yields the stated moduli 6 and 24. □
Concluding remarks
The corner viewpoint gives a parallel world for a substantial part of the third-order mock-theta family. The parallel is exact at the level of the finite rank kernels: the quadratic core of the mock-theta series is replaced by a linear corner peak, and Ramanujan’s Lost Notebook supplies the resulting partial- and false-theta evaluations. The exceptional identity C ψ ( q ) = q ω 3 ( q ) C_{\psi}(q)=q\omega_{3}(q) C ψ ( q ) = q ω 3 ( q ) shows that the passage does not invariably leave the mock-theta world.
The positive decompositions in Theorems 5.3 and 6.9 rewrite the two colored companions as
C ω ( q ) + E △ ( q ) and 2 C 3 , o ( q ) + E 3 , △ ( q ) , C_{\omega}(q)+E_{\triangle}(q) \qquad\text{and} \qquad2C_{3,o}(q)+E_{3,\triangle}(q), C ω ( q ) + E △ ( q ) and 2 C 3 , o ( q ) + E 3 , △ ( q ) ,
respectively. Consequently, both sparse corrections are signed differences between natural corner families. The correction terms in Theorems 5.4 and 6.10 are also governed by one boxed modular mechanism. Burson’s involutions cancel all nonfixed overpartition pairs, and the fixed pairs (23) account for the two quadratic families in the two-color identity and the single quadratic family in the three-color identity. Corollaries 5.5 and 6.11 therefore give auxiliary involutive models for the sparse errors. The remaining direct problems are now parallel: one seeks an involution between staircase-decorated even corners and odd corners in the two-color case, and an involution between staircase-and-tower-decorated even corners and two labeled copies of the restricted odd three-dimensional corners in the three-color case. Such constructions would lift the boxed fixed-point models back to the original corner objects and explain the factors 2 and 4 without passing through generating functions.
The exact specialization at z = − 1 z=-1 z = − 1 and the quartic and cubic identities in Theorems 6.4 , 6.5 , and 6.6 show that the two-variable three-dimensional series contains exact information not visible from its specialization at z = 1 z=1 z = 1 . In particular, the cubic specialization produces both exact rank equidistribution and a sparse rank difference. Several questions remain. Can the two corner-to-corner differences in Corollaries 5.5 and 6.11 be proved by direct sign-reversing involutions, and can the resulting positive decompositions be realized by explicit bijections from the original colored companions? Can kernel-preserving linearization be realized by a common Bailey-pair, Hecke–Appell, or Appell–Lerch mechanism for the complete identities rather than only their correction terms? What further rank dissections follow from higher roots of unity, and which congruences arise from cancellation inside R M R_M R M rather than merely from support exclusion? Finally, the corresponding one-arm linearization leads naturally to fifth-order mock theta functions. That direction appears sufficiently rich to merit a separate treatment.
The Pennsylvania State University, University Park, Pennsylvania 16802 Email address: andrews@math.psu.edu
Department of Mathematical Sciences, United Arab Emirates University, P.O. Box 15551, Al Ain, United Arab Emirates Email address: melbachraoui@uaeu.ac.ae