The Yamagishi Gravitational Multipole Aggregation Theory — fourth edition: the licence is the exponent two and four pi does not survive into the answer, but that licence reaches only as far as the shell theorem.
A distant spherically symmetric mass distribution can be reproduced, as far as its external gravitational field is concerned, by just two numbers, the total mass and the centre of mass, with zero error. This is Newton’s shell theorem, and this paper pins down the single source of that aggregation licence. The first edition attributed it to four pi, and more precisely to the Yamagishi patch that sets the gravitational constant to four pi (Paper 35). The second edition corrected this: the licence is granted by the exponent two, that is by flux conservation, and not by the value of the constant. The third edition carried that correction into the theorems of the body and narrowed the central claim to one line — four pi does not survive into the answer of the shell theorem. The four pi in the flux integral through a sphere and the four pi on the right-hand side of Gauss’s law are the same object, the total solid angle, appearing twice, and they cancel completely (verified symbolically). The four pi remaining in the answer once the patch is substituted is on the side of convention; the geometric one has already gone. What the first edition pointed at and called geometry was a convention.
The fourth edition adds two corrections. First, it measures the reach. Through the third edition the abstract also called the shell theorem the ground on which the Barnes-Hut and fast multipole methods compress the cost of N-body computation. That line was too strong. The shell theorem is a zero-error statement about spherically symmetric shells, whereas both methods handle the asymmetric clumps that fall into one cell of an octree, and their error is never zero. Indeed Corollary 4.3 of this very paper already stated that the error of an individual interaction is of order theta squared: the fact the fourth edition points out was not imported from outside but was already inside the paper, and only the abstract still claimed zero error. The attribution reaches the shell theorem and stops there.
FMM’s licence is low-rank separability. What lowers the cost is that the interaction block between a well-separated source box and target box is numerically low-rank, not the exponent of the kernel. The same compression works for kernels whose exponent is not two — the two-dimensional logarithm, Yukawa, Helmholtz, the inverse fourth power, a Gaussian, the inverse square root — and the existence of kernel-independent FMM, which never expands the kernel explicitly, is itself the counterexample (Paper 124). FMM is moreover linear in the particle count, while N log N is Barnes-Hut; Theorem 7.1 of this paper already drew that distinction correctly in the body, but the abstract wrote the two together.
Second, there is a reinforcement in the opposite direction. The claim that four pi does not survive into the answer has gained a second leg. Four pi is an overall factor of the kernel, and scaling a kernel by a constant scales every singular value by the same constant while leaving their ratios unchanged, so the numerical rank at a relative threshold is unchanged and the complexity of a hierarchical low-rank method does not depend on any overall factor. Hence four pi cannot enter the complexity. This is a proof from computational cost, independent of the geometric cancellation on the two sides of the shell theorem (Paper 124). The third edition’s claim is not weakened but given a second leg.
Where the attribution moved across the editions is this. The first edition saw the patched constant as a geometric structure serving as the licence. The second moved the licence to the exponent two and demoted four pi to a solid angle that is not the licence. The third showed absence: four pi never appears in the conclusion of the shell theorem at all, and while a demotion is a matter of degree, an absence is a matter of presence or not. The fourth shows that the licence granted by the exponent reaches only as far as the shell theorem, and adds a second, complexity-theoretic witness to the absence of four pi.
IMPORTANT (scope): a structural interpretation and self-audit of established standard facts (Layer 2). The shell theorem, the multipole expansion, causal finitisation, Barnes-Hut and FMM are all established content in physics and computer science, and no new physical law is claimed. What is not claimed: any new physical law (there is none) / a solution by Yukawa-type exponential screening (withdrawn in the second edition; gravity admits no screening) / that the licence for aggregation is the patched constant (corrected in the second edition) / that four pi survives into the answer of the shell theorem (the third edition showed the cancellation) / that the shell theorem is the ground of FMM (withdrawn in the fourth edition; the error is never zero) / that four pi affects the complexity (being an overall factor, it cannot). Note also that the explosion of cost described in Section 1 is not an open problem in physics: standard field theory nowhere performs a sum over all pairs, and each point responds only to a local field. The paradox arises from the premise that the universe is an N-body simulator, and outside that premise it does not arise. What remains: what licenses aggregation is the exponent — but how far the exponent licenses must be measured separately.
On the making of this work: The ideas and content of this work stem from the author’s own considerations. Assistance from an AI (a large language model) was used for structuring, English translation, and checking the algebra. Any remaining errors or misinterpretations are solely the author’s. Feedback and corrections are sincerely appreciated.
Reference
Yamagishi, Y. (2026). The Yamagishi Gravitational Multipole Aggregation Theory [Paper 51]. Zenodo.
DOI: 10.5281/zenodo.21602935
ORCID: 0009-0006-8587-3450