p-adic Numbers

Why p-adic numbers matter

• Wiles's proof of Fermat's Last Theorem (1995). The modularity theorem identifies Galois representations of elliptic curves with Galois representations of modular forms — both realised in p-adic vector spaces. The deformation theory, p-adic L-functions, and p-adic Hodge theory used in the proof live in the p-adic universe; without ℚₚ, the proof has no language to be stated.
• Local-global principles. The Hasse-Minkowski theorem reduces solving quadratic equations over ℚ to solving them over ℝ and every ℚₚ — a finite check (only finitely many primes matter). Failure of local-global for cubic and higher forms is measured by Tate-Shafarevich groups, central in the Birch-Swinnerton-Dyer conjecture.
• Class field theory. Local class field theory describes abelian extensions of ℚₚ via the multiplicative group ℚₚ*. Global class field theory glues local data into a coherent picture. Artin reciprocity is stated in this language.
• Iwasawa theory. Studies how arithmetic invariants (class groups, units, L-values) behave in towers of number fields where Galois groups are p-adic Lie groups. Mazur-Wiles theorem settled deep questions about cyclotomic fields using p-adic methods.
• p-adic L-functions. Analytic objects living in ℚₚ that interpolate special values of complex L-functions. The Iwasawa main conjecture relates p-adic L-functions to algebraic invariants — proved for many cases, deeply influences modern arithmetic geometry.
• Cryptography and coding theory. p-adic algorithms (Hensel lifting) factorise polynomials over ℤ in polynomial time; the LLL algorithm and lattice-based cryptography use p-adic-related ideas. Some post-quantum proposals use p-adic reductions.
• Geometry of numbers and dynamical systems. p-adic dynamics, Berkovich spaces, and rigid analytic geometry have become core tools in arithmetic dynamics, Mazur's torsion theorem, and the Mochizuki framework.

Common misconceptions

• "p-adic numbers are just base-p numerals." Every p-adic integer can be written aₙp^n + aₙ₊₁p^(n+1) + … with digits 0 ≤ aᵢ < p, like a base-p numeral with infinitely many digits to the left. But ℚₚ contains more than this: the field structure, square roots, exponentials, and analytic functions are all genuinely new objects — not just a re-spelling of integers.
• "Ultrametric is weird and unnatural." Ultrametricity is the natural geometry for a valuation. Distances measured by "how divisible by p" inherit ultrametricity for free, and the resulting topology — totally disconnected, locally compact, ball-isoceles — is exactly what arithmetic asks for. Far from weird, it is the ambient geometry whenever divisibility is the relevant notion of "size".
• "p-adic numbers are only theoretical." The Berlekamp-Hensel polynomial factorisation algorithm — implemented in every CAS and used in coding theory and computer algebra — is purely p-adic. Lattice-based cryptography uses LLL with p-adic-flavoured ideas. p-adic numbers ship in real software.
• "ℚₚ is a subfield of ℂ." No. ℚₚ and ℂ are both completions of ℚ but for incompatible absolute values; there is no canonical embedding ℚₚ ↪ ℂ. The algebraic closures ℚ̄ₚ and ℂ̄ = ℂ are abstractly isomorphic as fields (both algebraically closed of characteristic zero with cardinality continuum) but no canonical map exists, and the natural topologies disagree.
• "Each ℚₚ is isomorphic to all the others." No. ℚ₂ and ℚ₃ are non-isomorphic — for instance ℚ₂ contains square roots that ℚ₃ does not. They are distinct local fields with distinct arithmetic.
• "p must be small." p ranges over all primes; ℚ_(2^256-189) is a perfectly good local field used in cryptography. Algorithms are typically polynomial in log p, so primes of cryptographic size are routine.

A surprising p-adic identity

In ℚ₅, the series 1 + 5 + 25 + 125 + … converges (to 1/(1-5) = -1/4) because the terms are increasingly divisible by 5 — 5-adic small. Numerically: partial sums 1, 6, 31, 156, 781 mod 5^k all agree mod 5^k with -1/4 = …4444444. Verify: 4 · (1 - 5^k)/(1 - 5) ≡ 1 (mod 5^k) for k = 1, 2, 3, … since (1 - 5^k) ≡ 1 (mod 5^k). The "divergent" geometric series of real analysis is a convergent identity in ℚ₅ — a tiny example of how 5-adic geometry rearranges intuition.

Ostrowski's theorem and the choice of completion

A natural question: how many essentially different absolute values does ℚ have? Ostrowski (1916) answered: every nontrivial absolute value on ℚ is equivalent to either the standard archimedean |·|_∞ or to one of the p-adic |·|ₚ for some prime p. So the completions of ℚ — up to topological equivalence — are exactly ℝ and the family of ℚₚ. The "places" of ℚ are the primes plus one archimedean place; the product formula ∏_v |x|_v = 1 over all places binds them together. p-adic numbers are not exotic; they are one of two natural kinds of completion of the rationals.