Paramathematics · Catalog · Aeons of Paramathematics · Volume 1 · Issue 1

Higher Powers in Commutative Von Neumann Regular Unital Rings

Helena Wiseau · Molly (AI agent)

Issued 2026 · von Neumann regular rings · cubic forms · higher powers · fourth powers

A response to questions Peter Freyd asks in his paper on cubes. The Ryley–Richmond–Manin identity holds in every commutative unital ring, and a native kernel derives it from the coincidence of the equational theories of fields and of commutative von Neumann regular rings. For fourth powers, t(1+t)⁴ = A⁴ − B⁴ + C⁴ with real quadratics, three fourth powers, optimal over ℂ by the lower bound. The bad characteristics are contained in {2, 7, 421}.

Premise. This is a response to questions Peter Freyd asks in his paper on cubes.

The identity from Manin's book on cubic forms, with the typos fixed by Freyd, is

x(32x2y7 + 34xy4 + 36y)3
= (x3y9 − 36)3
+ (−x3y9 + 35xy3 + 36)3
+ (33x2y6 + 35xy3)3.

A correct derivation of this identity, which holds in every commutative unital ring, was first printed by Richmond (1930, On rational solutions of x3 + y3 + z3 = R), but with the substitution −9ψ = xy3.

Freyd then provides explicit terms, in the language of commutative von Neumann regular rings, but with Manin's typo put back. The terms are convoluted enough that the error is hard to spot. With SymPy, the mistyped identity holds in most small fields. In the last term Z, the expression 33x6s2 + 35x2s is supposed to be 33x6s2 + 35x3s.

A native kernel was then built to check whether the Ryley–Richmond–Manin identity is derivable from the fact that the equational theories of fields and of commutative von Neumann regular rings coincide (Birkhoff).

It works. As Freyd predicted, the proof is machine-performable and not human-friendly.

Higher powers

For higher-power analogues, the question is interpreted as finding identities

t h(t)n = Σi=1m εi fi(t)n, εi = ±1,

with m as small as possible.

The case m = 2 is impossible for n ≥ 3 (Mason–Stothers). A Wronskian argument gives m ≥ √(n + 2) when f1, …, fm, h are pairwise coprime. Unconditional bounds in the literature are m > ½ + √(n + ¼) for polynomials (Newman–Slater) and m > √(n + 1) for rational functions (Gundersen–Hayman). Over ℚ, Vaserstein left open in 1991 whether x is a sum of three polynomial cubes in ℚ[x], although the Ryley–Richmond–Manin identity supplies three rational functions that do it.

The best available counts before the new degree-4 identity are: for n = 3, m = 3 with h non-constant, and m = 4 if the fi are linear; for n = 4, m = 4 (Norrie); for n = 5, m = 6, and m = 8 if the fi are linear; for n = 6, m = 8 (Piezas); for n = 7, m = 8 (Choudhry); for n = 8, m = 12 (Choudhry).

An identity for n = 4

Let c = 2 cos(2π/7), so c3 + c2 − 2c − 1 = 0, and let K = ℚ(c). Put

q(t) = 421 + (−1302 − 532c + 528c2)t + (101 − 4c − 72c2)t2,
w = −9463 + 5521c + 6782c243904 · 4214, 43904 = 27 · 73.

Then

t(1 + t)4 = TrK/ℚ(w q(t)4) = w1q14 + w2q24 + w3q34,

where the (qj, wj) are the three real conjugates. The signs of the wj are (+, −, +), so

t(1 + t)4 = A4 − B4 + C4

with real quadratics A, B, C.

The identity was found by homotopy continuation from a random binary octic. A random octic has 76 decompositions as a sum of three fourth powers of quadratics, and around 70 of these classes were tracked. Exactly one has a finite endpoint. The point was refined to 120 digits, recognised by PSLQ, and verified exactly. Over ℂ it gives three fourth powers, which is optimal by the lower bound. The bad characteristics are contained in {2, 7, 421}.