Applications of Non-Standard Analysis

The payoff page of the NSA spine. Having rebuilt calculus with infinitesimals, we step back to the method — transfer down, compute infinitesimally, take the standard part, transfer up — and survey where it earns its keep: clean proofs of classical theorems, genuinely new mathematics that was first found nonstandardly, and the foundational vindication of the physicist's calculus that motivated the whole spine (physics-use-of-calculus.md §4.1).

References: Goldblatt, Lectures on the Hyperreals, ch. 16–18; Albeverio et al., Nonstandard Methods in Stochastic Analysis and Mathematical Physics; Arkeryd, Nonstandard Analysis (Boltzmann equation); Robinson, Non-standard Analysis.

1. The method

Every nonstandard proof follows one arc:

1. Transfer down. Import the relevant real theorems into ${}^{\ast }\mathbb{R}$ by transfer; everything classical is now available for hyperreal objects.
2. Work infinitesimally. Replace $\epsilon$–$\delta$ / $\epsilon$–$N$ quantifier alternations by direct statements about halos, infinitesimals, and hyperfinite objects (hyperreals.md, internal-sets.md). Quantifier complexity collapses: "for all $\epsilon$ there is $\delta$…" becomes a single implication between infinitely-close points.
3. Take standard parts. Round finite hyperreals back to reals (hyperreals.md §4); this is where the real conclusion crystallizes.
4. Transfer up. A first-order conclusion established in ${}^{\ast }\mathbb{R}$ is a theorem of $\mathbb{R}$.

The recurring discipline is internal/external hygiene (internal-sets.md §2): standard parts and halos are external, so they may be used but never fed to transfer.

2. Clean proofs of classical theorems

The spine already contains several proofs that are markedly shorter than their $\epsilon$–$\delta$ originals:

• Extreme & intermediate value theorems via a hyperfinite grid and standard part (infinitesimal-calculus.md §5).
• Heine–Cantor (continuous on compact $\Rightarrow$ uniformly continuous) as a one-liner from near-standardness (sequences-topology.md §4).
• Bolzano–Weierstrass: a bounded sequence's term $a_H$ at any unlimited $H$ has a shadow, which is a subsequential limit (sequences-topology.md §3).
• The Fundamental Theorem of Calculus: divide an infinitesimal area by an infinitesimal width (nonstandard-integration.md §4).

2.1 Peano existence for ODEs

A representative "new proof of an old theorem." Consider $y^{\prime }=f(x,y)$, $y(x_0)=y_0$, with $f$ continuous (no Lipschitz condition — this is Peano, not Picard–Lindelöf).

• Nonstandard construction. Take an unlimited $H$ and run Euler's method with infinitesimal step $\Delta x=1/H$ across a hyperfinite grid: $Y_{i+1}=Y_i+{}^{\ast }f(x_i,Y_i)\Delta x$. By transfer this internal recursion produces an internal hyperfinite polygonal approximant $Y$.
• Standard part. On a bounded region $f$ is finite, so the $Y_i$ stay finite; set $y(x)=\mathrm{st}(Y_i)$ for $x_i\approx x$. Microcontinuity of $f$ (infinitesimal-calculus.md §1) shows the infinitesimal Euler increments integrate to $y(x)=y_0+{\int }_{x_0}^{x}f(t,y(t))dt$, so $y$ solves the ODE.

The subtle compactness step in the classical Arzelà–Ascoli proof (extracting a uniformly convergent subsequence of Euler polygons) is replaced by a single standard part — the hyperfinite polygon already exists; one only shadows it.

3. Results found first — or most naturally — nonstandardly

NSA is not only a proof-shortening device; some theorems were discovered through it:

• Bernstein–Robinson theorem (1966). Every polynomially compact operator on a Hilbert space has a nontrivial invariant subspace — proved by Robinson with nonstandard hulls, then translated to a standard proof by Halmos. A landmark demonstration that NSA proves genuinely hard, new results.
• Loeb measure (1975). Standard-part of an internal hyperfinite counting measure yields a bona fide countably additive measure (nonstandard-integration.md §6); this gives slick constructions of Lebesgue measure, Brownian motion, and Itô integration, and is now a standard tool in stochastic analysis and mathematical physics.
• Nonstandard methods in PDE / kinetic theory. Existence results for the Boltzmann equation (Arkeryd) and other equations exploit hyperfinite discretizations that are exact rather than approximate.

4. The physics payoff

This spine was motivated by Why the Physicist's Calculus Is Legitimate. NSA discharges the promissory note of that remark's §4.1:

• "$dx$ is a length." In ${}^{\ast }\mathbb{R}$ it is a nonzero infinitesimal, a genuine element you may divide by and multiply (hyperreals.md).
• "$\int fdx$ = sum of slices." Literally a hyperfinite sum of infinitesimal slices, and its standard part is the Riemann integral (nonstandard-integration.md §2–3).
• "$\frac{dy}{dx}$ is a fraction; cancel the $du$." An honest quotient of hyperreals; the chain-rule cancellation is exact (infinitesimal-calculus.md §4).
• "Drop higher-order terms." The rigorous operation $\mathrm{st}$ (hyperreals.md §4.1), not hand-waving.
• Transfer guarantees the answers so obtained are true real theorems — which is exactly why the physicist's fast method is reliable in its domain of smooth functions.

The moral. The physicist computing with infinitesimals is, whether or not they know it, working in ${}^{\ast }\mathbb{R}$ and taking a standard part at the end. NSA does not tell physicists to change what they do; it certifies that what they already do is sound. The one caveat NSA cannot lift is the same one classical analysis flags — interchanging limits, divergent series, distributions — where the issue is convergence, not infinitesimals (physics-use-of-calculus.md §7).

5. Costs and caveats

• Non-constructive. The free ultrafilter rests on BPI (ultrapower.md §2), so NSA is not a constructive theory; you cannot exhibit a specific infinitesimal. Contrast the nilpotent, intuitionistic approach of smooth-infinitesimal.md.
• Internal/external vigilance. The price of the powerful transfer principle is constant care about which sets are internal (internal-sets.md §2); most "paradoxes" are external sets used illegitimately.
• No new theorems in principle. By transfer, anything NSA proves about $\mathbb{R}$ is already a first-order consequence of the reals — NSA is a conservative extension. Its value is methodological: shorter proofs, better intuition, and constructions (Loeb measure) that are awkward to phrase standardly.

Where this page is used

• Synthesizes the whole NSA spine into a reusable method.
• Redeems the forward references from physics-use-of-calculus.md §4.1.

Next: Smooth Infinitesimal Analysis (a Contrast) — the other rigorous home for infinitesimals.