Papers · Preprint · v2

Why Mathematics Even Works

Why imaginary numbers help with real problems.

In plain words

Imaginary numbers are not real. Infinitesimals are not real scalars. Roots of a polynomial need not lie in the base field. Yet all of them solve problems stated entirely in the lower layer.

The paper separates the lower layer, where the problem lives, from a promoted layer with extra objects. It bundles a useful calculation into a certificate: a hidden object with no lower counterpart, and a selected expression that returns an answer the lower layer accepts under explicit checks. The certificate licenses the answer, never the hidden object itself. For root problems, non descent is forced: no real linear map squares to reversal, so the quarter turn whose square returns reversal cannot be a real map.

Two layers. Below, the real numbers: the value alone cannot give a derivative, since t − x and 0 both have value 0 at x but slopes 1 and 0. An arrow takes f(t) = t³ at x = 2 up to a layer with ε ≠ 0 and ε² = 0, where (2 + ε)³ = 8 + 12ε, the terms in ε² and ε³ struck out. Two arrows bring back the standard part, 8 = f(2), the value, and the ε coefficient, 12 = f′(2), the derivative. ε stays in the upper layer.
The paper's dual-number witness, with t³ at 2 as an editorial instance. The value of a function at a point cannot tell you its slope there. Borrow a number ε that is not zero but squares to zero, evaluate at 2 + ε, and the slope appears as the coefficient of ε. The answer, 12, is an ordinary real number the lower layer can check; ε itself never comes back down.

Four witnesses are worked in full: imaginary numbers, dual numbers, Galois roots and graph cohomology. They carry four distinct labels, and a shared shape transfers obligations, not evidence. The paper also marks a limit: an audit can make a hidden step essential even when that step descends, so essential use alone does not force non descent.

What it shows

  • The Strict Audited Utility certificate, and exactly what it licenses.
  • Forced non descent for root and composition returns, with examples showing the hypotheses are needed.
  • Boundaries and replay in computations, and an example separating essential use from non descent.
  • Four standard examples worked exactly, with four distinct labels; the claim that they are canonical or minimal is not established.

What it does not claim

It is about calculations that pass through a promoted layer and return an answer the lower layer accepts, not mathematical practice in general. The general theorem of its first version, that every such calculation forces non descent, is not established. It does not claim that every essential use of a hidden object forces non descent, or that a promoted layer is ever computationally necessary.

Cite

Tsiokos, I. (2026). Why Mathematics Even Works. Zenodo. https://doi.org/10.5281/zenodo.23183149