The restricted partition and \(q\)-partial fractions (Q6095447)

The restricted partition function \(p_N(n)\), which counts the number of partitions of \(n\) into parts no larger than \(N\), has generating function \[F_N(x) = \sum_{n=0}^\infty p_N(n) x^n = \prod_{k=1}^N (1-x^k)^{-1}.\] The function itself can be written as a sum of components, \(p_N(n) = W_k(n;N)\). The \(W_k(n;N)\), called the \(k\)-th \emph{Sylvester waves}, are \(k\)-periodic quasi-polynomials in \(n\). If one writes the \(q\)-partial fraction decomposition \begin{align*} F_N(x) &= \sum_{l=1}^N \frac{g_{1,l}^{(N)}(x)}{(1-x)^l} + \sum_{l=1}^{\lfloor N/2 \rfloor} \frac{g_{2,l}^{(N)}(x)}{(1+x)^l} +\sum_{k=3}^N \sum_{l=1}^{\lfloor N/k \rfloor} \frac{g_{k,l}^{(N)}(x)}{(1-x^k)^l} \\ &= \sum_{l=1}^N \frac{\Gamma_{0,1,l}(N)}{(1-x)^l} + \sum_{l=1}^{\lfloor N/2 \rfloor} \frac{\Gamma_{0,2,l}(N)}{(1+x)^l} +\sum_{k=3}^N \sum_{l=1}^{\lfloor N/k \rfloor} \sum_{h=0}^{k-1} \frac{\Gamma_{h,k,l}(N) x^h}{(1-x^k)^l}, \end{align*} then one can express the waves as sums in these coefficients, e.g. \[W_1(n;N) = \sum_{l=1}^N \binom{n+l-1}{n}\Gamma_{0,1,l} \quad \text{ and } \quad W_2(n;N) = (-1)^n \sum_{l=1}^{\lfloor N/2 \rfloor} \binom{n+l-1}{n} \Gamma_{0,2,l}.\] The waves can be used to provide good approximations to the unrestricted partition function \(p(n)\) (for instance, for \(n \leq 1480\) the first wave is a good approximation) and can be expressed in terms of higher order Bernoulli numbers. The goal of the paper is to develop formulas for the waves amenable to good estimations. The author uses \(q\)-partial fraction decomposition, aiming for the cyclotomic polynomials, and employing the extended cover-up method. We need the following notation. Define \[\sigma_k(t;s) = \sum_{\xi \in \Delta_k} \frac{\xi^{-t}}{(\xi)_s}\] with \(\Delta_k\) the set of all \(k\)-th primitive roots of unity, \((q)_n = \prod_{i=1}^n (1-q^i)\) the \(q\)-Pochhammer symbol (with \((q)_0 = 1\)). Let the degenerate Bernoulli numbers \(\beta_k(\lambda)\) be defined by the relation \(\tilde{\beta}_k(m) = m^k \beta_k(1/m)\) and \[\frac{m(1-x)}{1-x^m} = \sum_{k=0}^\infty (-1)^k \frac{\tilde{\beta}_k(m)}{k!}(1-x)^k,\] and the degenerate Euler numbers \(\epsilon_k(\lambda)\) be defined by the relation \(\tilde{\epsilon}_k(m) = m^k \epsilon_k(1/m)\) and \[\frac{2}{1+x^m} = \sum_{k=0}^\infty \frac{(-1)^k \tilde{\epsilon}_k(m)}{k!}(1-x)^k.\] The main theorems of the paper then give the following identities for the \(\Gamma_{h,k,l}\): Theorem 2.0.1. We have \(\Gamma_{0,1,N}(N) = 1/N!\) and for \(1 \leq j < N\), \[\Gamma_{0,1,N-j}(N) = \frac{(-1)^j}{N!} \sum_{j_2 + \dots + j_N = j} \frac{\tilde{\beta}_{j_2}(2)\dots \tilde{\beta}_{j_N}(N)}{j_2! \dots j_N!}.\] Theorem 2.0.2. We have \(\Gamma_{0,2,\lfloor N/2 \rfloor} = \frac{1}{2^N \lfloor N/2 \rfloor !}\), and for \(1 \leq j < \lfloor N/2 \rfloor\), \[\Gamma_{0,2,\lfloor N/2 \rfloor -j}(N) = \frac{(-1)^j}{2^N \lfloor N/2 \rfloor !} \sum_{j_1 + \dots + j_N = j} \frac{\tilde{\gamma}_{j_1}(1)\dots \tilde{\gamma}_{j_N}(N)}{j_1! \dots j_N!}.\] Here \(\tilde{\gamma}_k(n)\) is \(\tilde{\epsilon}_k(n)\) for \(n\) odd, and \(\tilde{\beta}_k(n)\) for \(n\) even. Furthermore, \(g_{2,l}^{(N)} (x) = \Gamma_{0,2,l} (N)\). Theorem 2.0.3. Denote \(R_{k,s}(x) = \sum_{t=0}^{k-1} \sigma(t;s) x^t\). For \(j \geq 3\), \[\Gamma_{j,k,\lfloor N/k \rfloor}(N) = \frac{1}{k^{\lfloor N/k \rfloor+1} \lfloor N/k \rfloor !} \sigma_k(j;N \% k)\] where \(\%\) denotes \(N\) mod \(k\), and \[g_{k, \lfloor N/k \rfloor}(x) = \frac{1}{k^{\lfloor N/k \rfloor+1} \lfloor N/k \rfloor !} R_{k,N \% k} (x).\]