Non-Standard Analysis: Motivation and History

The opening page of the non-standard analysis (NSA) spine. Before building the machinery, it is worth being clear about what problem NSA solves: it makes the infinitesimal — a quantity greater than $0$ yet smaller than every positive real — into a rigorous object, and thereby vindicates two and a half centuries of calculus that was computed with infinitesimals long before anyone could justify them. This is the promised rigorous backing for the claim in Why the Physicist's Calculus Is Legitimate §4.1 that "$dx$ is a real tiny thing" can be made literally true.

References: Robinson, Non-standard Analysis (1966), ch. 1; Goldblatt, Lectures on the Hyperreals, ch. 1; Dauben, Abraham Robinson; Keisler, Elementary Calculus, epilogue.

1. The calculus of infinitesimals (Leibniz, Euler)

When Newton and Leibniz created the calculus, the derivative was a genuine quotient of infinitesimals. Leibniz wrote $\frac{dy}{dx}$ meaning a ratio of infinitely small increments, and the integral $\int ydx$ as an infinite sum of infinitely thin rectangles of area $ydx$. Euler computed with infinitesimals and infinite numbers with spectacular success — infinite products, series, the $\Gamma$ function — treating an infinitesimal $\epsilon$ as a number one could add, multiply, and (crucially) discard when negligible: $x+\epsilon \approx x$, and $(dx{)}^2$ dropped against $dx$.

This calculus worked — it produced correct physics and correct analysis — but its foundations were incoherent as stated. Berkeley's famous 1734 polemic The Analyst skewered the infinitesimal as a "ghost of a departed quantity": in the same computation $dx$ was treated as nonzero (you divide by it) and then as zero (you drop it). No consistent number system of the day contained such objects.

2. The rigor crisis and the $\epsilon$–$\delta$ exorcism

The 19th century resolved the incoherence not by justifying infinitesimals but by eliminating them. Cauchy, and decisively Weierstrass, recast every infinitesimal statement as a statement about limits quantified with $\epsilon$ and $\delta$ over ordinary real numbers (continuity.md, differentiation.md). "$dy/dx$" became ${\lim }_{h\to 0}\frac{f(x+h)-f(x)}{h}$ — a single indivisible symbol, not a quotient; "$\int fdx$" became a limit of Riemann sums (riemann-integral.md), not a sum of infinitesimals. Dedekind and Cantor built $\mathbb{R}$ itself rigorously (real-numbers.md), and the Archimedean property — no real is infinitely small — became a theorem. The infinitesimal was formally banished.

The banishment was a genuine advance in rigor, but it left a gap between practice and theory: physicists and applied mathematicians never stopped reasoning with infinitesimals, because that reasoning is faster and more intuitive. The $\epsilon$–$\delta$ apparatus explained what the answers meant but not why the infinitesimal method reliably found them.

3. Robinson's resolution (1966)

In 1961 Abraham Robinson closed the gap using 20th-century mathematical logic. His insight: the tools of model theory — the compactness theorem and the ultraproduct construction — build a proper ordered-field extension ${}^{\ast }\mathbb{R}⊋\mathbb{R}$, the hyperreals, that genuinely contains infinitesimals and infinite numbers, and satisfies exactly the same first-order laws as $\mathbb{R}$. The bridge back to classical mathematics is the transfer principle (transfer-principle.md): any first-order statement true of $\mathbb{R}$ is true of ${}^{\ast }\mathbb{R}$ and conversely. Infinitesimal computations, followed by taking standard parts (hyperreals.md), therefore prove genuine theorems about the reals.

The model theory page already records the seed of this idea: adjoining a constant $c$ with axioms $c>1,c>2,c>3,\dots$ is finitely satisfiable, so by compactness there is a model of $\mathrm{Th}(\mathbb{R})$ containing an infinite element — and its reciprocal is a nonzero infinitesimal. NSA is the systematic development of that observation into a usable calculus.

Robinson's thesis. The infinitesimal reasoning of Leibniz and Euler was never wrong — it was incompletely founded. NSA supplies the missing foundation and shows the old computations were, all along, shorthand for transfer + standard part. Berkeley's paradox dissolves: $dx$ is nonzero (an honest infinitesimal, so you may divide by it) and "dropping it" is the separate, well-defined operation $\mathrm{st}(\cdot )$ of rounding a finite hyperreal to its nearest real.

4. Two routes: ultrapower vs. axiomatic (IST)

There are two standard ways to get the hyperreals, and this spine uses the first:

• The ultrapower / model-theoretic route (Robinson, Luxemburg, Goldblatt). Construct ${}^{\ast }\mathbb{R}$ concretely as an ultrapower ${\mathbb{R}}^{\mathbb{N}}/\mathcal{U}$ and derive transfer from Łoś's theorem. Concrete, honest about the choice principle it consumes, and directly reuses model theory.
• The axiomatic route: Internal Set Theory (Nelson, 1977). Instead of building a bigger field, adjoin to ordinary set theory a predicate "standard" and three axioms (Idealization, Standardization, Transfer). Infinitesimals then live inside the ordinary reals as the nonstandard elements. Elegant for practice but further from the analysis and model-theory already in these notes; we mention it as an alternative (applications.md) and do not develop it.

A third tradition — smooth infinitesimal analysis / synthetic differential geometry — takes infinitesimals to be nilpotent (${\epsilon }^2=0$) and is not a variant of Robinson's approach; it lives in intuitionistic logic and is treated as a deliberate contrast in smooth-infinitesimal.md.

5. What this spine does

The pages that follow build the ultrapower route end to end and then rebuild calculus on it:

1. Construction — ultrapower.md builds ${}^{\ast }\mathbb{R}$; transfer-principle.md proves transfer.
2. The number system — hyperreals.md develops infinitesimals and the standard part; internal-sets.md supplies hyperfinite sets.
3. Calculus — infinitesimal-calculus.md and nonstandard-integration.md recover the derivative and integral; sequences-topology.md does limits and compactness.
4. Payoff — applications.md and the contrast page smooth-infinitesimal.md.

Where this page is used

• Sets the agenda for the whole NSA spine.
• Supplies the historical and philosophical backing referenced by physics-use-of-calculus.md §4.1.

Next: The Ultrapower Construction of the Hyperreals — build ${}^{\ast }\mathbb{R}$ concretely.