Differentiation as multiplication by zero
The derivative is taught as a limit — creep h toward zero and hope the ratio settles. But we no longer have a zero that things creep toward; we have a zero you can multiply by and later divide back out. With that zero, the derivative stops being a limit and becomes an act of arithmetic: you multiply by zero and read a number off a shelf.
Recall the shelves from the multiplication chapter: every value carries a grade, the power of zero it rides on, and multiplying adds grades. Ordinary numbers sit at grade 0; one factor of zero drops you to grade 1; a factor of ω lifts you back up.
Now offset the input by a single graded zero — write x + 0, where that 0 is the invertible grade-1 zero, not a vanishing quantity — and feed it through a function. Because the offset carries a grade, the function's response sorts itself onto the shelves automatically:
This is the ordinary Taylor expansion, with one difference that changes everything: the powers of 0 are not small, they are graded. Nothing is being neglected, and nothing is being approached. The information that a limit works so hard to extract was never hiding in a process; it was sitting in plain sight, filed one grade down.
The first derivative is now simply the coefficient on grade 1. To read it, you don't take a limit — you do two moves you already own. Lift the whole expansion up one grade by multiplying by ω (which turns grade 1 into grade 0), then discard whatever is left off the ground floor. Call that discard the projection P — keep grade 0, drop the rest:
Multiply by zero to spread the function across the grades; multiply by ω to slide the grade you want down to the floor; project to read it. No h, no "arbitrarily close," no epsilon. The derivative is an exact, reversible operation — the same three moves will hand you the second derivative from grade 2, and the whole jet if you want it.
Take f(x) = x² + 3x − 4. Offset the input by the graded zero and multiply it out — ordinary algebra, nothing dropped and nothing approximated:
Now file each term on its shelf by how many factors of zero it carries:
The entire local behaviour of f is already here, sorted. The slope is sitting on grade 1 as its coefficient, 2x + 3 — we only have to bring it down to the floor and read it. Multiply by ω, which turns grade 1 into grade 0 (recall 0·ω = 1), then project:
So f′(x) = 2x + 3 — the derivative of x² + 3x − 4, obtained with no limit, no h → 0, only multiplication and a projection. And nothing extra had to be done to find the rest: the same expansion already carries the second derivative on grade 2, whose coefficient is 1, so f″ = 2 — exactly right.
There is a second way to ask "how is f changing," and the theory insists on treating it as an equal. The one above measures change by subtraction — it strips the value at grade 0 and normalizes. Its twin measures change by division: form the ratio f(x·(1+ε)) / f(x) and you get the logarithmic derivative, x f′/f — the natural rate when the thing you compare against is 1, not 0.
These are not two tricks that happen to rhyme. They are the same construction — strip the ground term, normalize by the increment — carried out with the two group operations we have been living in all book: additive, anchored at 0, and multiplicative, anchored at 1. And there are exactly two, because there are exactly two one-parameter groups on the line, and exponentiation is the single bridge between them: b⁰ = 1 carries the additive identity to the multiplicative one, and log walks it back.
The two differentials do not commute, and that is the point, not a nuisance. Writing D for the additive one and θ = x·D for the multiplicative one, their disagreement is exact and minimal:
If they commuted, one would be redundant and there would be nothing new here. They don't, so the pair genuinely carries independent information — and that single non-commutation is the seam the later chapters pry open, where the residue lives and, eventually, where time comes from.
Running the derivative backward — antidifferentiation — works cleanly everywhere except one address. At grade −1, the pole, the reconstruction hands off to a logarithm and the tidy bookkeeping stops being polynomial. That is not a flaw in the method; it is the method telling you, honestly, where the writable world ends. We will return to that boundary deliberately; for now it is enough to know the structural differential is exact everywhere the antiderivative doesn't send you through the pole.
The derivative has changed category. It is no longer a limit we take on faith that it converges; it is multiplication by a graded zero, followed by a reading. Every rule you know — the power rule, the product rule, the chain rule — is now grade bookkeeping rather than an epsilon argument, and it is total and reversible like the rest of the arithmetic. What we have not yet said is that this one operation, read through different charts, is several familiar derivatives wearing one face. That is the next thing to pull apart.