On the irreducibility of a truncated binomial expansion (Q2477975)

From the text: For positive integers \(k\) and \(n\) with \(k\leq n-1\), define \(P_{n,k}(x) =\sum_{j=0}^k \binom nj x^j\). In the case that \(k = n-1\), the polynomial \(P_{n,k}(x)\) takes the form \(P_{n,n-1}(x) = (x+1)^n -x^n\). If \(n\) is not a prime, \(P_{n,n-1}(x)\) is reducible over \(\mathbb Q\). If \(n=p\) is prime, the polynomial \(P_{n,n-1}(x) = P_{p,p-1}(x)\) is irreducible as Eisenstein's criterion applies to the reciprocal polynomial \(x^{p-1}P_{p,p-1}(1/x)\). This note concerns the irreducibility of \(P_{n,k}(x)\) in the case where \(1\leq k\leq n-2\). Computations for \(n\leq 100\) suggest that in this case \(P_{n,k}(x)\) is always irreducible. We will not be able to establish this but instead give some results which give further evidence that these polynomials are irreducible. The problem arose during the 2004 MSRI program on ``Topological aspects of real algebraic geometry'' in the context of work of Inna Scherbak in her investigations of the Schubert calculus in Grassmannians. She had observed that the roots of any given \(P_{n,k}(x)\) are simple. This follows from the identity \[ P_{n,k}(x) -(x+1)\frac{P'_{n,k}(x)}{n}= \binom{n-1}{k} x^k. \] She then asked whether, for a fixed positive integer \(n\), the various \(n(n-1)/2\) roots of \(P_{n,k}(x)\) for \(1\leq k\leq n-1\) are distinct. We will not resolve this problem but our methods imply that for each positive integer \(n\), almost all of the roots are distinct. In other words, the number of distinct roots is \(\sim n^2/2\) as \(n\) tends to infinity. We note that since the initial writing of this paper, \textit{I. Scherbak} [Intersections of Schubert varieties and highest weight vectors in tensor products \(sl_{N+1}\)-representations, \url{arXiv: math.RT/0409329}] has resolved the original question concerning the roots of these polynomials as a consequence of her work in representation theory. We also mention that these same polynomials have recently arisen in the context of work by \textit{I. V. Ostrovskii} [Comput. Methods Funct. Theory 4, No. 2, 275--282 (2004; Zbl 1075.30001)]. In particular, he finds a solution to a problem posed by Alexandre Eremenko on the distribution of the zeros of \(P_{n,k}(x)\) as \(k\) and \(n\) tend to infinity with \(k/n\) approaching a limit \(\alpha\in (0, 1)\). For our results, we consider \[ F_{n,k}(x) =\sum_{j=0}^k a_jc_jx_j, \tag{2} \] where \(a_0, a_1, \dots, a_k\) denote integers, each having all of its prime factors \(\leq k\). In particular, none of the \(a_j\) are zero. Our first result is that the number of possible reducible polynomials \(F_{n,k}(x)\) with \(n\leq N\) and \(1\leq k\leq n-2\) is small by comparison. More precisely, we show the following. Theorem 1. Let \(\varepsilon > 0\), and let \(N\) be a positive integer. For each integral pair \((n, k)\) with \(1\leq n\leq N\) and \(1\leq k\leq n-2\), consider the set \(S(n, k)\) of all polynomials of the form (2) where \(a_0, a_1, \dots, a_k\) denote arbitrary integers, each having all of its prime factors \(\leq k\). The number of such pairs \((n, k)\) for which there exists a polynomial \(f(x)\in S(n, k)\) that is reducible is \(O(N^{23/18+\varepsilon})\). Our second result is an explicit criterion for the irreducibility of \(F_{n,k}(x)\). Theorem 2. If there is a prime \(p>k\) that exactly divides \(n(n-k)\), then \(F_{n,k}(x)\) is irreducible for every choice of integers \(a_0, a_1, \dots, a_k\) with each having all of its prime factors \(\leq k\). Theorem 2 has a simple proof based on Eisenstein's criterion. It implies, in particular, that if \(n\) is a prime, then \(P_{n,k}(x)\) is irreducible for every \(k \in \{1, 2, \dots, n-1\}\). This then resolves the problem of Scherbak in the case that \(n\) is a prime. Our third and final result concerning the irreducibility of \(F_{n,k}(x)\) is as follows. Theorem 3. Let \(k\) be a fixed integer \(\geq 3\). There is an \(n_0 = n_0(k)\) such that if \(n\geq n_0\), then \(F_{n,k}(x)\) is irreducible for every choice of integers \(a_0, a_1, \dots, a_k\) with each having all of its prime factors \(\leq k\). The value of \(n_0(k)\) in this last result, being based on the solutions to certain Thue equations, can be effectively determined.