arXiv:1712.00608v1 [math.NT] 2 Dec 2017 ftefr [ form the of of expansions generalized and variants several 1.1. n fteform the of and ucin aaadproduct. Hadamard function; ATRZTO HOESFRHDMR PRODUCTS HADAMARD FOR THEOREMS FACTORIZATION N IHRODRDRVTVSO ABR SERIES LAMBERT OF DERIVATIVES HIGHER-ORDER AND 2010 Date e od n phrases. and words Key abr eisfcoiaintheorems. factorization series Lambert usa 5 Tuesday : ahmtc ujc Classification. Subject Mathematics dniyfrtenme fdsic rmsdividing primes distinct fun of zeta number Riemann the the for for identity series new several functions, ating ftececet fteinteger-order pr the results of coefficients new the our summation-b of new of functions, Applications explicit multiplicative involving the partitions. sums to of an functions theory expansi multiplicative article, the characteristic special the the over connect within ries they expansions that their in to requi significant to give corre enough we coefficients important are attention series cases The function and generating series. genertwo series series Lambert Lambert Lambert of of of derivatives products expansions Hadamard cha the for which of factorizations theorems cases factorization articl primary original this cific these Within to theorems. analogs factorization series Lambert Abstract. 3 , 5 , th 6 efis umrz on oko eea rlmnr canonic preliminary several on work joint summarize first We , eebr 2017. December, 9 n n X X ] ERI NTTT FTECHNOLOGY OF INSTITUTE GEORGIA ≥ ≥ EEAIGFUNCTIONS GENERATING 1 1 abr eis atrzto hoe;mti factorizat matrix theorem; factorization series; Lambert 1 1 a a ¯ − − n n [email protected] COLO MATHEMATICS OF SCHOOL q q 8 HRYSRE NW STREET CHERRY 686 q q n n [email protected] n n 1 KLSBUILDING SKILES 117 AI .SCHMIDT D. MAXIE TAT,G 30332 GA ATLANTA, 1. = = Introduction ( ( q q ; ; 1 1 q q 12;1P1 51;05A19. 05A17; 11P81; 11A25; USA ) ) ∞ ∞ j 1 th n n X X ≥ ≥ eiaie fLmetsre gener- series Lambert of derivatives 1 1 abr eisfcoiaintheorems factorization series Lambert k X k X =1 =1 n n nterfrne ehv proved have we references the In s s e n,k n,k n . e a a k k yadtv aueof nature additive ly · · sdinterpretations ased n fLmetse- Lambert of ons eetbihnew establish we e to,ada exact an and ction, q q pnigt these to sponding atrz w spe- two racterize eteseildue special the re n n vd e exotic new ovide oevr are moreover, d , , o higher-order for tn functions: ating o;partition ion; al 2 M. D. SCHMIDT for an anda ¯n depending on an arbitrary arithmetic function an and where the n k k lower-triangular sequence sn,k := [q ](q; q) q /(1 − q ), which we typically require ∞ to beb independent of the an, is the difference of two partition functions counting the number of k’s in their respective odd (even) distinct partitions. In the concluding remarks to [3] we gave several specific examples of other con- structions of related Lambert series factorization theorems. In the reference, we also proved a few new properties of the factorizations of Lambert series generating functions over the convolution of two arithmetic functions, f ∗ g expanded by n (f ∗ g)(n)qn 1 = s (g)f(k) · qn, 1 − qn (q; q) n,k n 1 ∞ n 1 k=1 X≥ X≥ X which we cite in this article. The Lambert series factorizationse of these two forms and their variations, two to three of which we consider in this article, also imply matrix factorizations of these expansions which are dictated by the corresponding typically invertible matrix of sn,k or sn,k(g). The matrix-based interpretation of these factorization theorems is perhaps the most intuitive way to explore how these expansions “factor” into distinct matricese applied to vectors of special sequences.
1.2. Significance of our results. Our new results are rare and important because they provide a mechanism that effectively translates the divisor sums of the coef- ficients of a Lambert series generating function into ordinary sums which similarly generate the same prescribed arithmetic function, say an. Moreover, these fac- torization theorems connect the special functions in multiplicative number theory which are typically tied to a particular Lambert series expansion with the additive theory of partitions and partition functions in unusual and unexpected new ways. We are one of the first authors to examine such relations between the additive and multiplicative in detail (see also [4, 3, 5]). We note that we are not the first to consider the derivatives of Lambert series generating functions [8], though our perspective on the connections afforded by these factorizations is distinctly new.
1.3. Focus within this article. Within this article we explore the expansions of factorization theorems for two primary additional special case variants which are distinctive and important enough in their applications to require special atten- tion here: Hadamard products of two Lambert series generating functions and the higher-order integer derivatives of Lambert series generating functions. Section 2 proves several new properties of the first case, where the results proved in Section 3 consider the second case in detail. The significance of these two particular fac- torizations is that they have a broad range of applications to expanding special and classical arithmetic functions from multiplicative number theory. In Section 4 we tie up loose ends by offering two other related variants of the Lambert series factorizations. Namely, we prove factorization theorems for generating function convolutions and provide a purely matrix-based proof of a new formula for the coefficients enumerated by a Lambert series generating function.
New results and characterizations. The Hadamard product generating function cases lead to several forms of new so-termed “exotic” sums for classical special functions as illustrated in the explicit corollaries given in Example 2.4 of the next section. For example, if we form the Hadamard product of generating functions of FACTORIZATION THEOREMS FOR HADAMARD PRODUCTS AND DERIVATIVES 3 the two Lambert series over Euler’s totient function, φ(n), we obtain the following more exotic-looking sum for our multiplicative function of interest:
√24k 23 b −6 − n p(d − k) j k j(3j + b) 2 φ(n)= µ(n/d) k2 + (−1)j k − . d 2 k=1 d n b= 1 j=1 X X| X± X The importance of obtaining new formulas and identities for the derivatives of a se- ries whose coefficients we are interested in studying should be obvious, though our factorization results also provide new alternate characterizations of these deriva- tives apart from typical ODEs which we can form involving the derivatives of any sequence ordinary generating functions. Our results for the factorization theorems for derivatives of Lambert series generating functions in Section 3 provides two particular factorizations which characterize these expansions.
2. Factorization Theorems for Hadamard Products 2.1. Hadamard products of generating functions. The Hadamard product of two ordinary generating functions F (q) and G(q), respectively enumerating the sequences of {fn}n 0 and {gn}n 0 is defined by ≥ ≥ n (F ◦ G)(q) := fngn · q , for |q| < σF σG, n 0 X≥ where σF and σG denote the radii of convergence of each respective generating function. Analytically, we have an integral formula and corresponding coefficient extraction formula for the Hadamard product of two generating functions when F (q) is expandable in a fractional series respectively given by [1, §1.12(V); Ex. 1.30, p. 85] [11, §6.3] π 2 1 ıt ıt (F ◦ G)(q )= F (qe )G(qe− )dt 2π 0 Z q (F ◦ G)(q) = [x0]F G(x). x In the context of the factorization theorems we consider in this article and in the ref- erences, we consider the Hadamard products of two Lambert series generating func- tions for special arithmetic functions fn and gn which we define coefficient-wise to enumerate the product of the divisor sums over each sequence corresponding to the coefficients of the individual Lambert series over the two functions. This subtlety is discussed shortly in Definition 2.1. As we prove below, it turns out that we can formulate analogous factorization theorems for the cases of these Hadamard prod- ucts as well. The next definition makes the expansion of the particular Hadamard product functions we consider in this section more precise. Definition 2.1 (Hadamard Products for Lambert Series Generating Functions). For any fixed arithmetic functions f and g, we define the Hadamard product of the two Lambert series over f and g to be the auxiliary Lambert series generating function over the composite function afg(n) whose coefficients are given by
a (m)qm a (d) = [qn] fg := f g , fg 1 − qm d d d n m 1 d n d n X| X≥ X| X| :=fg(n) | {z } 4 M. D. SCHMIDT so that by M¨obius inversion we have that
afg(n)= fg(d)µ(n/d) = (fg ∗µ)(n). d n X| We note that this definition of our Hadamard product generating functions is al- ready slightly different than the one given above in that we define the Hadamard product of two Lambert series generating functions by the expansion of a third composite Lambert series which corresponds to the particular expansion of the factorization in (1) below. 2.2. Main results and applications. The next theorems in this section define ( 1) the key matrix sequences, sn,k(f) and sn,k− (f), in terms of the next factorization of the Lambert series over afg(n) from the definition above in the form of n a (n)qn 1 fg = s (f)g · qn, (1) 1 − qn (q; q) n,k k n 1 ∞ n 1 k=1 X≥ X≥ X where sn,k(f) is independent of the function g, which is equivalent to defining the factorization expansion by the inverse matrix sequences as
n n ( 1) k afg(n)q gn = s − · [q ] (q; q) × . (2) n,k ∞ 1 − qn k=1 n 1 X X≥ We also define the following function to expand divisor sums over arithmetic func- tions as ordinary sums for any integers 1 ≤ k ≤ n: 1, if k|n; Tdiv(n, k) := (0, otherwise. Theorem 2.2. For all integers 1 ≤ k ≤ n, we have the following definition of the factorization matrix sequence defining the expansion on the right-hand-side of (1) where we adopt the notation f(n) := d n fd: | sn,k(f)= Tdiv(n, k)f(n) e P √24(n k)+1 b − − 6e j(3j + b) j(3j + b) + (−1)j T n − , k · f n − . div 2 2 b= 1 j=1 X± X e Proof. By the factorization in (1) and the definition of afg(n) given above, we have that for f(n)= d n fd | n e P s (f) = [g ] f × g T (n, d) n,k k d d div d n d=1 X| X n n = [q ](q; q) × Tdiv(n, k)f(n)q , ∞ n 1 X≥ which equals the stated expansion of the sequence by Euler’se pentagonal number theorem which provides that j j(3j 1)/2 j(3j+1)/2 (q; q) =1+ (−1) q − + q . ∞ j 1 X≥ FACTORIZATION THEOREMS FOR HADAMARD PRODUCTS AND DERIVATIVES 5
Theorem 2.3 (Inverse Sequences). For all integers 1 ≤ k ≤ n, we have the next definition of the inverse factorization matrix sequence which equivalently defines the expansion on the right-hand-side of (1).
( 1) p(d − k) sn,k− (f)= µ(n/d), d n f(d) X| Proof. We expand the right-hand-side of the factorization in (1) for the sequence e ( 1) 1 gn := sn,r− (f), i.e., the exact inverse sequence, for some fixed r ≥ 1 as follows : n j ( 1) ( 1) f(n) · sd,r− (f)= sj,ksk,r− · p(n − j) d n j=0 k=1 X| X X e n = [j = r]δ p(n − j)= p(n − r). j=0 X Then the last equation implies that
( 1) p(n − r) sd,r− (f)= , d n f(n) X| which by M¨obius inversion implies our stated result. e Example 2.4 (Applications of the Theorem). For the arithmetic function pairs t s t (f,g) := (n ,n ), (φ(n), Λ(n)), (n ,Jt(n)), respectively, and some constants s,t ∈ C where σα(n) denotes the generalized sum-of-divisors function, Λ(n) is von Mangoldt’s function, φ(n) is Euler’s totient function, and Jt(n) is the Jordan totient function, we employ the equivalent expan- sions of the factorization result in (2) to formulate the following “exotic” sums as consequences of the theorems above: n s p(d − k) n = µ(n/d) σt(k)σs(k) (3) σt(d) k=1 d n " X X| √24k+1 b 6 − j k j(3j + b) j(3j + b) + (−1)jσ k − σ k − t 2 s 2 b= 1 j=1 # X± X n p(d − k) Λ(n)= µ(n/d) k log(k) d k=1 d n " X X| √24k 23 b −6 − j k j(3j + b) j(3j + b) + (−1)j k − log k − 2 2 b= 1 j=1 # X± X √24k 23 b n −6 − p(d − k) j k j(3j + b) 2t J (n)= µ(n/d) k2t + (−1)j k − . t dt 2 k=1 d n b= 1 j=1 X X| X± X 1 Notation: Iverson’s convention compactly specifies boolean-valued conditions and is equiva- lent to the Kronecker delta function, δi,j ,as[n = k]δ ≡ δn,k. Similarly, [cond = True]δ ≡ δcond,True in the remainder of the article. 6 M. D. SCHMIDT
By forming a second divisor sum over the divisors of n on both sides of the first equation above, the first more exotic-looking sum for the sum-of-divisors functions leads to an expression for σs(n) as a sum over the paired product functions, σt(n) · σs(n). We do not know of another such identity relating the generalized sum-of- divisors functions existing in the literature or in the references which we cite in this article. However, the following relations between the multiplicative generalized sum-of-divisors functions and special additive partition functions are known where pk(n) denotes the number of partitions of n into at most k parts and pp(n) denotes the number of plane, or planar, partitions of n [10, A000219] [7, §26.9, §26.12]: n 1 − n · p(n)= p(k)σ1(n) k=0 Xn n · pk(n)= pk(n − t) j, k ≥ 1 t=1 j t X Xj | k ≤ n n · pp(n)= pp(n − j)σ2(j). j=1 X Corollary 2.5 (New Series for the Riemann Zeta Function). For fixed s,t ∈ C such that ℜ(s) > 1 we have the following infinite sum representations of the Riemann zeta function where we denote the sequence of interleaved pentagonal numbers, ω(±n), 1 j 3j+1 by Gj = 2 2 2 for j ≥ 0 [10, A001318]: n p(d − k) j/2 σt(k − Gj )σs(k − Gj ) ζ(s)= µ(n/d) × (−1)⌈ ⌉ s σt(d) (k − Gj ) n 1 k=1 d n j:Gj j/2 Gj (q; q) = (−1)⌈ ⌉q . ∞ j 0 X≥ The convergence of these infinite series is guaranteed by our hypothesis that ℜ(s) > 1. We compare the results in the previous theorem to the known Dirichlet gener- ating functions for the sum-of-divisors functions which are expanded by [2, Thm. 291] [7, §27.4] σ (n) ζ(s)ζ(s − α)= α , ℜ(s) > 1, α +1 ns n 1 X≥ ζ(s)ζ(s − α)ζ(s − β)ζ(s − α − β) σ (n)σ (n) = α β , ℜ(s) > 1, α +1,β +1. ζ(2s − α − β) ns n 1 X≥ FACTORIZATION THEOREMS FOR HADAMARD PRODUCTS AND DERIVATIVES 7 s In particular, we note that while the series n σs(n)/n , and similarly for the second series, are divergent, our sums given in Corollary 2.5 do indeed converge for ℜ(s) > 1. P 3. Factorization Theorems for Derivatives of Lambert Series 1000000000 00 1200000000 00 0 −230000000 00 −1 2 −34000000 00 −2 −4 −3 −4500000 00 −2 2 6 −4 −56000 0 00 −2 −4 −6 0 −5 −6700 0 00 −1 2 −3 8 0 −6 −780 0 00 0 −2 9 −4 0 0 −7 −89 0 0 0 1 2 −6 −8 15 0 0 −8 −9 10 0 0 2 0 −3 4 −10 6 0 0 −9 −10 11 0 3 2 12 12 −5 12 7 0 0 −10 −11 12 (i) s1,n,k 100000000000 − 1 1 2 2 0000000000 − 1 1 1 3 3 3 000000000 1 1 1 4 0 4 4 00000000 3 2 1 1 0 5 5 5 5 0000000 1 1 1 1 1 0 6 3 6 6 000000 4 5 3 2 1 1 7 1 7 7 7 7 7 0 0 0 0 0 9 7 5 3 3 1 1 1 8 8 8 8 8 4 8 8 0 0 0 0 16 4 8 7 5 1 2 1 1 9 3 9 9 9 3 9 9 9 0 0 0 5 11 11 9 1 1 3 1 1 1 2 10 10 10 2 2 10 5 10 10 0 0 31 30 15 7 5 3 2 1 1 11 11 2 11 1 11 11 11 11 11 11 0 13 8 7 5 13 3 7 5 1 1 1 1 4 3 4 4 12 4 12 12 4 6 12 12 (−1) (ii) s1,n,k Figure 3.1. Factorization matrix sequences for the first-order deriva- tives of an arbitrary Lambert series generating function when 1 ≤ n, k ≤ 12 3.1. Derivatives of Lambert series generating functions. We seek analogous factorization theorems for the higher-order tth derivatives for all integers t ≥ 1 of 8 M. D. SCHMIDT an arbitrary Lambert series in the form of n a qn 1 qt · D n = s a · qn. (4) t 1 − qn (q; q) t,n,k k n t ∞ n t k=t X≥ X≥ X We note that we consider these sums over n ≥ t to produce an invertible matrix of the factorization sequences st,n,k. At first computation, the corresponding matrix sequences for these higher-order derivatives do not suggest any immediate intuitions ( 1) to exact formulas for st,n,k and st,n,k− as in the factorizations expanded above. The listings given in Figure 3.1 provide the first several rows of these sequences in the first-order derivative case where t := 1 for comparison with our intuition. However, may still prove using the method invoked in the proof of Theorem 2.3 to show that in the first-order case we have that2 ( 1) p(d − k) s − = µ(n/d), 1,n,k d d n X| which, in light of the construction of Corollary 2.5 and its notation, leads to the following further convergent series infinite expansion of the Riemann zeta function for ℜ(s) > 1: n p(d − k) j/2 σs(k − Gj ) ζ(s)= µ(n/d) × (−1)⌈ ⌉ s 1 . d (k − Gj ) − n 1 k=1 d n j:Gj 3.2. Main results. We will next require the statements of the next two results proved in [8] to state and prove the main results in this section. We refer the reader to the proofs of these two lemmas given in the reference. Lemma 3.1 (Modified Lambert Series Coefficients). For any fixed arithmetic func- tion an and integers m,t ≥ 1, n, k ≥ 0, we have the following identity for the series 2 More generally, if we expand the next mixed series of jth derivatives and initial terms annihilated by the differential operator as n j−1 n j anq i 1 n q · Dj + (a ∗ 1)(i)q = sj,n,kak · q , X 1 − qn X (q; q) X X n≥1 i=1 ∞ n≥t k=t n k k for integers j ≥ 2, we can easily prove that sj,n,k = sn,k = [q ]q /(1 − q )(q; q)∞ for 1 ≤ n