Under the condition that one at least of the leading coefficients amn, a0n differs from zero, the equation

has as solution a series convergent for all x greater (or all x less) than a fixed number. The coefficients of the various terms in the series are expressed in terms of the arbitrary values of the solution and its first n derivatives in an initial interval of appropriate length.

This paper was assisted in publication by a grant from the Carnegie Trust for the Universities of Scotland.

A set of variables is assumed to depend upon a number of common factors and specifics. Formulae are then derived for the sampling variances and covariances of the residual covariances obtained by removing the effect of the factors. The variances and covariances of the set of estimated loadings are also found. It must, however, be noted that the results obtained are valid only when an efficient method of estimation is used.

The title I have chosen for this lecture contains an implication which perhaps will not be generally accepted; there was a time when I would not have accepted it myself. The implication is that a particular kind of philosophy is possible which may be called scientific in contrast with other kinds which cannot be so called. I would go further and identify this scientific philosophy with what is generally called science, and this implies that the distinction that is often assumed to exist between science and philosophy is a false one. For this view I believe there is historical evidence. Science, as a separate, self-contained study, dates from the seventeenth century. Before that time, such consideration as was given to the subject-matter of present-day science was given it by philosophers and regarded as a part of their philosophising, and when in the seventeenth century a new kind of procedure was introduced, it was looked upon by its pioneers not as an attack on a new problem but as a new attack on an old problem. The science of that time was the “new philosophy”, faintly adumbrated by some mediæval philosophers, struggling for expression in Francis Bacon, and coming to full recognition in Galileo. Only later, when it had made such progress in certain limited fields of study that a new body of investigators was called into being who confined themselves to those fields, was the new philosophy transformed into a non-philosophy and called generally by the name “science”.

In classical mechanics the mass of a system of gravitating particles can be denned to be the mass of an equivalent particle which gives the same field at great distances, or alternatively the mass can be defined by means of Gauss' Theorem. Reference to the former procedure was made by Eddington and Clark (1938) in a discussion on the problem of n bodies. The relativistic extension of Gauss' Theorem has been investigated by Whittaker (1935) for a particular form of the line-element and for more general fields by Ruse (1935). The latter, treating the problem from a purely geometrical point of view, expressed the integral of the normal component of the gravitational force as the sum of two volume integrals. The physical significance of one of these integrals was quite obvious but the meaning of the other was far from clear. In this paper the terms in Ruse's result are examined as far as the order m2 in the case of a fundamental observer at rest and the 1938 discussion modified to bring the two investigations into line. It is concluded that the surface integral of the normal component of the gravitational force taken over an infinite sphere is –4π × the energy of the system.

Throughout the preceding discussion we have considered only those contributions to the stresses which are due to either the motion of the body or the presence of other bodies. That is, we have not considered the stresses which occur in the systems discussed by Whittaker, and we have, in effect, assumed that the gravitational mass of an isolated body at rest is the same as its invariant mass at least as terms of order m2 are concerned. In this appendix we complete the investigation by demonstrating the validity of this assumption.

The gravitational field of a system of particles was investigated by de Sitter as far back as 1916. A minor alteration to the analysis was made by Eddington and Clark in 1938. The amended value of the potential g44 is the same as that derived by Einstein, Infeld and Hoffmann without making use of the energy-tensor; this agreement suggests that the revised de Sitter argument is correct. In this paper we show that this is not the case, for the de Sitter analysis completely overlooked any possible interaction terms in the stress components of the energy-tensor. We find the value of these terms, pmn, and show that the agreement mentioned above is due to the fact that the volume integral of pu vanishes.

Little progress has been made in the development of a relativity theory of elasticity, although it has been realised that no disturbance can be propagated with a velocity greater than that of light. In 1917 Lorentz (1) gave a relativistic formulation of the laws of elasticity in the case of small strain and, applying the theory to the problem of a rotating, incompressible, homogeneous disc, he claimed that the radius as measured by an observer at rest on the disc undergoes a contraction. His result was accepted by Eddington (4) but was attacked by others. A great deal has been written on the subject, but it has never been pointed out that both Lorentz and Eddington were considering material in which the waves of dilatation travel with an infinite velocity. In this paper we define “incompressible” matter as that in which these waves are propagated with the velocity of light and Poisson's ratio tends to the value ½. This gives an upper limit to the modulus of compression k, which in this case is the elastic constant λ, and as a result the expansion determined by the ordinary classical theory has to be taken into account. It is found that the “relativity contraction” is exactly cancelled by the “classical expansion”. Throughout the discussion on the rotating disc the analysis is restricted to the case of small strain.

The equations of equilibrium of a continuous static distribution of matter are also investigated in the case of weak fields for which the fourth power of the density may be neglected.

The systems of “partitive numbers” introduced in this paper differ from ordinary number systems in being subject to non-associative addition. They are intended primarily to serve as the indices of powers in algebraic systems having non-associative multiplication, or as the coefficients of multiples in systems with non-associative addition, but are defined more generally than is probably necessary for these purposes. They are essentially the same as root-trees (Setzbäume) with non-branching knots other than terminal knots ignored, with operations of addition and multiplication defined.

Partitive numbers are of two kinds, partitioned cardinals and partitioned serials, defined respectively as the partition-types of repeatedly partitioned classes and series. For each kind, multiplication is binary (i.e. any ordered pair has a unique product) and associative. Addition is in general a free operation (i.e. the summands are not limited to two, and indeed, assuming the multiplicative axiom, may form an infinite class or series); but it is non-associative, which means that for example a + b + c (involving one operation of addition) is distinguished from (a + b) + c and a + (b + c) (involving two operations). A one-sided distributive law is obeyed:

Partitioned cardinals are commutative in addition.

The structure of commutative associative linear algebras is well known and is usually derived from more general results concerning non-commutative algebras (Cartan, Frobenius). The novelty of the present treatment is that while it avoids the complexities of the non-commutative case, it exhibits the essential relationship between the theory of commuting matrices and that of commutative algebras.

While theorems 1 and 2 of this paper are implicit in the writings of Voss (1889), Taber (1890), and Plemelj (1901), it has been considered worth while to recapitulate these results in the explicit form required for the discussion of commutative algebras. In doing so, some new facts emerge.

I. Throughout this paper k1, …, k3 will denote s ≥ I fixed distinct positive integers. Some years ago Pillai (1936) found an asymptotic formula, with error term O(x/log x), for the number of positive integers n ≤ x such that n + k1, …, n + k3 are all square-free. I recently considered (Mirsky, 1947) the corresponding problem for r-free integers (i.e. integers not divisible by the rth power of any prime), and was able, in particular, to reduce the error term in Pillai's formula.

Our present object is to discuss various generalizations and extensions of Pillai's problem. In all investigations below we shall be concerned with a set A of integers. This is any given, finite or infinite, set of integers greater than 1 and subject to certain additional restrictions which will be stated later. The elements of A will be called a-numbers, and the letter a will be reserved for them. A number which is not divisible by any a-number will be called A-free, and our main concern will be with the study of A-free numbers. Their additive properties have recently been investigated elsewhere (Mirsky, 1948), and some estimates obtained in that investigation will be quoted in the present paper.

It has been acknowledged for a long time that current quantum theory is incomplete. The difficulties and unanswered problems which have gradually become apparent during the development of the theory will not be discussed here, and only one aspect of the situation will be mentioned, that there seems to exist a large number of particles with different rest-masses, the numerical values of which demand a theoretical explanation. The experimental material has recently been greatly increased by the discovery of several kinds of mesons with different rest-masses.