Worked solution: Hilbert's existence proof for Waring's problem (1909)

Frequently asked questions

Is there a simpler, self-contained way to see why odd exponents work too, without the full induction?

There is a genuinely different, much more elementary route for odd exponents specifically: a finite-difference argument (going back to work summarized by Paul Pollack, building on E.M. Wright 1934) shows directly, without any sphere-averaging identity, that every integer is a sum of a bounded number of $k$ -th powers whenever $k$ is odd — but this is a separate elementary technique, not part of Hilbert's own unified argument, which handles all $k$ through one mechanism.

Source

David Hilbert, 1909

Beweis für die Darstellbarkeit der ganzen Zahlen durch eine feste Anzahl n-ter Potenzen (Waringsches Problem)

Condensed summary

Step 5 of 7: Reaching every exponent, odd and even, by induction

Previous step Next step

In plain words

Step 4 supplied a power identity by matching $k$ (Hilbert's exponent) with $r=4$ squares, but the displayed relation does not by itself represent every integer as a sum of $2k$ -th powers. To reach odd exponents too, Hilbert's actual 1909 argument proceeds by strong induction on the target exponent: assuming Waring's problem is already settled for every smaller exponent, the same identity machinery — used with a different, larger number of variables $r$ tailored to the exponent at hand, rather than always $r=4$ — closes the gap for the exponent currently under consideration, odd or even.

This condensed account summarizes what is, in Hilbert's original paper, the most technical bookkeeping: matching the sphere-averaging identity's parameters $(k,r)$ to whatever exponent is being attacked next, and invoking the induction hypothesis (that all smaller exponents already have a finite $g$ ) to supply the raw material the identity needs. The details are intricate combinatorics rather than new ideas — the conceptual heart of the proof is entirely contained in Steps 3–4.

g(k) < \infty \ \text{for every integer } k \ge 2

Detailed analysis

This step is deliberately condensed: Hilbert's full 1909 argument (Mathematische Annalen 67) establishes $g(k)<\infty$ for every integer $k\ge2$ , not just even $k$ , by strong induction on $k$ using the same sphere-averaging identity with the number of variables $r$ chosen appropriately at each stage, combined with the inductive hypothesis that every smaller exponent already has a finite $g$ -value (Theorem 3.6 of Nathanson's textbook treatment, 'The Hilbert–Waring Theorem: the set of non-negative $k$ -th powers is a basis of finite order for every positive integer $k$ '). We summarize this technical induction here, following Ellison's 1971 survey ('Waring's Problem', American Mathematical Monthly 78), rather than reproduce its case-by-case bookkeeping in full.

The important structural point, made explicit by Ellison and by later expositions (e.g. Rademacher–Toeplitz), is that no genuinely new idea is needed beyond the sphere-averaging identity of Step 3: what varies across the induction is only which auxiliary integers Lagrange's (or a similar, more general) representation theorem supplies, and how many variables $r$ the identity is applied with. The mathematical content of the whole proof is captured by Steps 2–4; this step is the technical scaffolding that turns 'works for even exponents' into 'works for every exponent'.

Having established $g(k)<\infty$ for every $k\ge2$ , the Hilbert–Waring theorem is proved. Hilbert's proof is entirely qualitative: it gives no explicit numerical bound on $g(k)$ whatsoever, only its finiteness — a point Hilbert's contemporaries found remarkable, since existence proofs without explicit bounds were still controversial in some corners of the mathematical community around 1900.

Terms in this step

Strong induction: A form of mathematical induction in which the inductive step for a value $k$ is allowed to use the truth of the statement for every smaller value, not just $k-1$ — appropriate here since the identity for exponent $k$ draws on representation theorems whose complexity varies with $k$ in a way not tied to a single previous case.