The purpose of this paper is to give a short proof of an important recent theorem of Bombieri [2] on the mean value of the remainder term in the prime number theorem for arithmetic progressions. Applications of the theorem have been made by Bombieri and Davenport [3], Rodriques [9], and Elliott and Halberstam [5]. For earlier versions of the theorem and a survey of other applications, see Barban [1], and Halberstam and Roth [7, Chapter 4].

In a paper [1] published in 1956, David Gale introduced the idea of representing a convex polytope by a diagram (now called a Gale diagram). Later work by Gale, T. S. Motzkin, and more recently by M. A. Perles and B. Grünbaum, has shown the importance of this idea. Using it, a large number of results which were formerly inaccessible have been proved. For an account of Gale diagrams and their applications, the reader is referred to Grünbaum's recent book [2; §5.4 and §6.3] and we shall largely follow the notations used there.

The Integral Equation. Consider a periodic wave moving with constant velocity c from right to left on the surface of an inviscid, incompressible fluid which is at rest at infinity. The motion is assumed to be irrotational and two-dimensional. The bottom is horizontal, and the depth of the undisturbed fluid is h.

Christoffel's problem, in its classical form, asks for the determination of necessary and sufficient conditions on a function φ, defined over the unit spherical surface Ω, in order that there exist a convex body K for which φ (u) is the sum of the principal radii of curvature at that boundary point of K where the outer unit normal is u. The figures Ω and K are in Euclidean n-dimensional space (n ≥ 3). It is assumed that φ is continuously differentiable and that K is of sufficient smoothness. A solution of Christoffe's problem was given in [6]. Yet that treatment is rather unsatisfactory in that the smoothness restrictions are set by the method rather than the problem, cf. [5; p. 60]. The present paper overcomes this defect. To do this it is first necessary to generalize the original problem so as to seek conditions on a measure M, defined over the Borel sets of Ω, in order that M be a first order area function for a convex body K. When K has sufficient smoothness, then φ is the Radon-Nikodym derivative of M with respect to surface area measure on Ω. It is this generalized Christoffel problem which is solved in what follows.

The integral equation

occurs in the theory of the motion of an electrically conducting fluid, occupying the space between two electrodes at different potentials, in the presence of a strong magnetic fluid [Hunt and Malcolm 1]. In this theory a solution is required in which

and one of the principal properties of the solution to be determined is

In recent years, functions of n variables whose partial derivatives are measures, have been found to retain the properties of functions of bounded variation of one variable to a remarkable degree [e.g., G1, G2, G3, K, Z, and especially the announcement F].

In the nonoscillation theory of ordinary differential equations the asymptotic behaviour of the solutions has often been described by exhibiting asymptotic expansions for these solutions. A fundamental illustration of this technique may be found in Hille [6] wherein the linearly independent solutions of a second order homogeneous differential equation were described by single termed asymptotic expansions. For the dominant solution, this result was successively extended by Waltman in [8] for a second order equation and in [9] for an nth order equation. A further generalization of these results appears in [3] where a complete nth order nonhomogeneous nonlinear differential equation was considered; again, asymptotic representations were given to describe the behaviour of the solutions of the differential equation. Moore and Nehari [7], Wong [10], [11], and Hale and Onuchic [2], also use asymptotic representations in discussing the behaviour of the solutions of certain differential equations. All of the above results are essentially perturbation problems with the unperturbed linear differential equation having the form y(n) = h(t) for some n and h(t).

Let E be a compact subset of R2, of Hausdorff dimension s > 0 and for each real number t let Ft be the linear set {x1 + tx2: (x1x2) ∈ E}. In this note we shall prove

THEOREM. If s ≤ 1 then Ft has dimension ≥ s, excepting numbers t in a set of dimension ≤ s. If s > 1 then Ft, has positive Lebesgue measure, excepting numbers t in a set of Lebesgue measure 0.

The problem of estimating accurately the order of magnitude of the least primitive root g(p) to a large prime modulus p is as yet unsolved. The first non-trivial estimate was obtained by I. M. Vinogradov (see [5]) who in about 1919 showed that

Let G be any group and G′ its derived, then G/G′—the group G made abelian—will be denoted by Ga. Over any ring R, denote by E2(R) the group generated by the matrices as x ranges over R; the structure of E2(R)a has been described in a recent theorem [2; Th. 9.3] for certain rings R, the “quasi-free rings for GE2” (cf. §2 below). Now over a commutative Euclidean domain, E2(R) is just the special linear group SL2(R); this suggests applying the theorem to the ring of integers in a Euclidean number field. However, the only number fields whose rings were shown to be quasi-free for GE2 in [2] were the non-Euclidean imaginary quadratic fields. In fact that leaves the application of Th. 9.3 of [2] to the ring of Gaussian integers unjustified (I am indebted to J.-P. Serre for drawing this oversight to my attention). In order to justify this application one would have to show either (a) that the Gaussian integers are quasi-free for GE2., or (b) that Th. 9.3 of [2] holds under weaker hypotheses which are satisfied by the Gaussian integers. Our object in this note is to establish (b)–indeed our only course, since the Gaussian integers turn out to be not quasi-free.

Let D be an integral domain with identity having quotient field K. A non-zero fractional ideal F of D is said to be divisorial if F is an intersection of principal fractional ideals of D[4; 2]. Equivalently, F is divisorial if there is a non-zero fractional ideal E of D such that

Divisorial ideals arose in the investigations of Van der Waerden, Artin, and Krull in the 1930's and were called v-ideals by Krull [9; 118]. The concept has played an important role in the development of multiplicative ideal theory.

A graph G (finite, undirected, and without loops or multiple lines) is n-connected if the removal of fewer than n points from G neither disconnects it nor reduces it to the trivial graph consisting of a single point. We present in this note a sufficient set of conditions on the degrees (valences) of the points of a graph G so that G is n-connected.

Let V = V2n be the Segre product variety of two n–dimensional complex projective spaces. Then any Cremona transformation of Sn into (regarded as an irreducible algebraic system of ∞n ordered pairs of points) is represented on V by an irreducible n-dimensional subvariety H which satisfies (on V) the algebraic equivalence

where Si, j is a subvariety of V and m1, …, mn–1 are positive integers. We call m1, …, mn–1 the characters of T noting that, numerically,

Helly's theorem asserts that if W is a family of compact convex sets in a j-dimensional linear space, and if any j + 1 members of ℱ have a non-empty intersection, then there is a point common to all members of ℱ. If one attempts to generalise this result to the case when ℱ consists of sets which are expressible as the union of at most n disjoint compact convex sets then, in general, one finds that there is no number h(n, j) such that if any h(n, j) members of ℱ have a non-empty intersection, then there is a point common to all members of ℱ. The difficulty lies in the fact that, in general, the intersections of members of ℱ are more complicated in structure than are the members of ℱ.

The object of this paper is first to generalize the basic inequality of the large sieve method to exponential sums in many variables, and then to deduce results for algebraic number fields that are analogous to known results for the rational field.

Let Ω be a compact metric space of non-σ-finite Λh-measure, where Λh denotes the Hausdorff measure corresponding to some continuous increasing real function h(t), denned for t ≥ 0 and with h(t) > 0 and h(t) > 0 for t > 0. C. A. Rogers has drawn attention to two problems that have remained unsolved for some years, and which seem to present considerable difficulty:

I. Does Ω necessarily have subsets of finite positive Λh-measure?

II. Does Ω. necessarily contain a system ofdisjoint closed subsets each of non-σ-finite Λh-measure ?

The answers are affirmative when Ω is a subset of a Euclidean space (see [1], [2], [3], [4], [5]); recently D. G. Larman ([8], [9]) was able to extend each of these results to the case of the “finite-dimensional” compact metric spaces that he introduced in [7]. Both of his proofs depend on the detailed structure of such spaces; the aim of the present paper is to establish a general theorem, from which it follows in particular that Larman's second result is a consequence of his first, and that an affirmative answer to I would imply an affirmative answer to II. (Since it seems likely that I must be answered negatively in general, and II affirmatively, the present theorem may well be of only ephemeral significance; the nature of the proof reinforces this expectation.)

In this paper, we consider a class of spaces for which the convolutions with any set of regularizes converge in the topology of the space. We have already dealt with this matter in [2], but the conditions on the topology were unnecessarily restrictive and the proof somewhat unnatural. The present theorem is not only substantially more general, but is also more satisfying in that the argument reveals an unexpected connection between two topics; namely, the approximation of Lebesgue integrals by means of Riemann sums, and the uniqueness of certain types of locally convex topologies in vector lattices.

Let T(n, k) denote the number of trees n with n labelled nodes of which exactly k have degree two. We shall derive a formula for T(n, k) and then determine the asymptotic behaviour of T(n,0); this will enable us to calculate the limiting distribution of Xn the number of nodes of degree two in a random tree n. Rényi [5] has treated the corresponding problem for nodes of degree one in random trees.

Let E be a continuum and n ≥ 4 a given positive integer. A system of points z1,…, zn єE that maximizes

is called a system of Fekete points. Such a system may be not unique.

Let Cn be the n-dimensional complex number space of the complex variables z1,…, zn and be the unit hypersphere. Further, let G be the group of all holomorphic automorphisms of K, then G is a n(n + 2)-dimensional real Lie group. In [3] the author has proved that for any P∈∂K (the boundary of K) there exists a decomposition of G in the form , where G0 is the group of all analytic rotations about the origin and is a 2n-dimensional real Lie group, whose underlying topological space is K, which acts transitively on K and P is a fixed point of all elements of .