The Transfer Principle

The logical engine of the NSA spine. Transfer is the precise statement that $\mathbb{R}$ and ${}^{\ast }\mathbb{R}$ satisfy exactly the same first-order truths. It is what turns infinitesimal computations into proofs about the reals: reason in ${}^{\ast }\mathbb{R}$ with infinitesimals (where it is easy), then transfer the first-order conclusion back to $\mathbb{R}$. This page upgrades the sentence-form corollary of Łoś's theorem (ultrapower.md §5) into a working tool and draws the crucial internal/external boundary that says what transfer may and may not touch.

References: Goldblatt, Lectures on the Hyperreals, ch. 4 & 11–14; Robinson, Non-standard Analysis, ch. 2–3; Keisler, Foundations of Infinitesimal Calculus.

1. The superstructure and the $\ast$-map

To transfer statements about functions and sets, not just numbers, one works in the superstructure $V(\mathbb{R})$ — $\mathbb{R}$ together with all sets, relations, and functions built over it by iterated power sets. Each object $A$ (a number, a subset of $\mathbb{R}$, a function $\mathbb{R}\to \mathbb{R}$, a relation…) has a $\ast$-transform ${}^{\ast }A$ in the corresponding nonstandard superstructure $V{(}^{\ast }\mathbb{R})$:

Concretely, in the ultrapower, ${}^{\ast }f([a_n])=[f(a_n)]$ and $[a_n]\in {}^{\ast }A\iff \{n:a_n\in A\}\in \mathcal{U}$. A standard function and its $\ast$-transform agree on standard inputs: ${}^{\ast }f(r)=f(r)$ for $r\in \mathbb{R}$, so it is customary to drop the star on standard functions and write $\sin$, $\exp$ for ${}^{\ast }\sin$, ${}^{\ast }\exp$.

2. The transfer principle

Transfer principle. Let $\phi$ be a first-order sentence in the language of the superstructure, with all constants, functions, relations, and bounded quantifiers replaced by their $\ast$-transforms. Then $$\mathbb{R}\text{-world}\models \phi ⟺>{}^{\ast }\mathbb{R}\text{-world}\models {}^{\ast }\phi .$$

For statements purely about $(\mathbb{R},+,\cdot ,<,\dots )$ this is exactly the Łoś corollary of ultrapower.md §5; the superstructure version extends it to quantification over ${}^{\ast }$-sets and ${}^{\ast }$-functions, provided quantifiers are bounded by $\ast$-transformed sets (this restriction is the seed of the internal/external distinction, §4).

The rule of use is mechanical:

To transfer: write the true real theorem as a first-order sentence with all quantifiers ranging over named sets; star every set, function, and relation; the result is a true theorem of ${}^{\ast }\mathbb{R}$ — and vice versa.

3. Transfer at work

Each example illustrates the "star everything, keep the logical form" rule.

• Ordered field axioms. "$\forall x\forall y(x+y=y+x)$" transfers verbatim: ${}^{\ast }\mathbb{R}$ is a commutative ordered field (ultrapower.md §4) — for free, no coordinate computation.
• Roots. "$\forall x(x\ge 0\to \exists y{y}^2=x)$" transfers: every nonnegative hyperreal has a hyperreal square root, so ${}^{\ast }\mathbb{R}$ is real closed. Applied to an infinitesimal $\epsilon >0$, it has an infinitesimal square root $\sqrt{\epsilon }$.
• The function $\sin$ is bounded. "$\forall x(-1\le \sin x\le 1)$" transfers to ${}^{\ast }\sin$ on all of ${}^{\ast }\mathbb{R}$, including infinite arguments — a fact one uses constantly in nonstandard calculus.
• Discreteness of $\mathbb{N}$. "$\forall n\in \mathbb{N}\neg \exists m\in \mathbb{N}(n<m<n+1)$" transfers: ${}^{\ast }\mathbb{N}$ has no elements strictly between consecutive integers — so the unlimited hypernaturals (internal-sets.md) still sit in a discrete order of integer-spaced blocks.
• Archimedean property does not transfer as "no infinitesimals." The real Archimedean property reads "$\forall x\exists n\in \mathbb{N}(x<n)$." It does transfer — but with ${}^{\ast }\mathbb{N}$: every hyperreal is below some hypernatural. It says nothing about ordinary $\mathbb{N}$, which is why ${}^{\ast }\mathbb{R}$ can be non-Archimedean over $\mathbb{R}$ while transferring the Archimedean sentence. Watch the quantifier domains — this is the single most common transfer pitfall.

4. Internal vs. external — the boundary of transfer

Transfer only applies to internal objects: those in the range of the $\ast$-map (hyperextensions ${}^{\ast }A$, hyperfunctions ${}^{\ast }f$) and, more generally, the sets definable with internal parameters (internal-sets.md). Sets assembled "from outside" — by picking out elements according to their standardness — are external, and transferring a property to them is illegitimate.

The canonical external sets:

• $\mathbb{R}$ itself, inside ${}^{\ast }\mathbb{R}$ is external. If it were internal, transfer of "every nonempty bounded-above subset has a least upper bound" (the completeness of $\mathbb{R}$) would apply to it — but $\mathbb{R}\subset {}^{\ast }\mathbb{R}$ is bounded above (by any infinite hyperreal) with no least upper bound in ${}^{\ast }\mathbb{R}$. Contradiction, so $\mathbb{R}$ is external.
• The infinitesimals, the finite hyperreals, $\mathbb{N}\subset {}^{\ast }\mathbb{N}$ are all external, by similar least-upper-bound / least-element arguments (hyperreals.md, internal-sets.md).

Why this matters. The completeness axiom and the well-ordering of $\mathbb{N}$ are first-order over the internal subsets and transfer perfectly to ${}^{\ast }$-internal subsets — but they fail for the external sets that carve out "the infinitesimals" or "the standard reals." Every apparent paradox of NSA (a bounded set with no sup, a nonempty set of naturals with no least element) is an external set illegitimately treated as internal. Keeping the boundary in view is the whole discipline.

5. The three practical principles

In use, transfer is almost always deployed in one of three shapes:

1. Downward (real ⇒ hyper): a true real theorem, starred, is a true hyper theorem — imports all of classical algebra/analysis into ${}^{\ast }\mathbb{R}$.
2. Upward (hyper ⇒ real): a first-order property proved in ${}^{\ast }\mathbb{R}$ (typically the easy part, using infinitesimals) transfers back to a real theorem — this is how NSA proves things.
3. Definitional replacement: an $\epsilon$–$\delta$ definition (continuity, convergence, integrability) is shown equivalent to an infinitesimal one (infinitesimal-calculus.md); the equivalence is a transfer argument, after which one may compute infinitesimally and read off the classical conclusion.

Where this page is used

• Every nonstandard proof in infinitesimal-calculus.md, nonstandard-integration.md, and sequences-topology.md is a transfer + standard-part argument.
• The internal/external boundary is developed concretely in internal-sets.md.
• Rests on Łoś's theorem from ultrapower.md.

Next: The Hyperreal Numbers — infinitesimals, the halo, and the standard part.