\infty

Infinity

Notation: structural shorthand

In plain English

Not a number but a direction of travel. $n \to \infty$ means "as n grows without bound". $L_\infty$ in a scaling law is the value loss would approach as model size grows forever: an asymptote, a horizon you extrapolate toward and never observe. Physicists use "infinite-N" or "infinite-width" limits as idealisations that make calculations tractable.

As code

// watch a sequence approach its limit
for (const n of [1, 10, 100, 1e4, 1e6]) console.log(n, 1 + 1 / n); // → 1

Where the analogy bends: Code can only show the trend at finite n. The limit is the value the trend approaches.

See it

Terms added51 → 1.00000000

\infty

is not a number you reach.

\sum_{n=1}^{\infty} 1/2^n = 1

means the partial sums get as close to 1 as you like.

What it means, field by field

Mathematics

Limits, unbounded intervals, infinite sums and integrals.

Physics

Thermodynamic limit (N → ∞ particles); the large-N limit in AdS/CFT (Maldacena 1997).

AI / ML

Infinite-width limits (neural tangent kernel); irreducible loss L_∞; the infinite-data limit.

Epistemics

An asymptote fitted from finite data is an extrapolation. Treat it as one.

In a research paper

Illustrative form of a neural scaling law with an irreducible-loss term (cf. Kaplan et al. 2020; Hoffmann et al. 2022 fit a similar form).

L(N) = L_\infty + A\,N^{-\alpha}

Loss approaches a floor L∞ as model size N grows; the second term shrinks toward zero but never reaches it at any finite N.

Common misreadings

Treating ∞ as a big number you can do ordinary arithmetic with (∞ − ∞ is undefined).
Believing an estimated L∞ was measured. It is inferred by extrapolating a fit.

Connected symbols

\sum

Summation (capital sigma)

\int

Integral

\approx

Approximately equal

\propto

Proportional to

Where it appears on this site

∂ An illustrative neural scaling law § 1. Learning to Read Equations § 5. Complexity and Systems Thinking § 7. The Physics–Biology–AI Connection Lab