Local Fields and Ramification Theory

⚡ Key Takeaways

Local fields are complete discretely-valued fields with finite (or perfect) residue field — namely finite extensions of $\mathbb{Q}_p$ , $\mathbb{F}_q((t))$ , and the archimedean $\mathbb{R}, \mathbb{C}$ . Ramification theory measures how prime ideals split, inert, or ramify in extensions, organized by the ramification filtration of the Galois group. Key result: every finite separable extension factors as a tower of unramified, tamely ramified, and wildly ramified pieces, and the higher ramification groups $G_i$ control the different, conductor, and local reciprocity. Foundational to algebraic number theory, class field theory, and the Langlands program. Developed by Hensel, Hasse, Krull, and Hilbert (1890s–1930s).

💡 The Core Insight

Extensions of local fields are governed by one invariant filtration. Given a finite Galois extension $L/K$ with group $G$ , the ramification filtration $G = G_{-1} \supseteq G_0 \supseteq G_1 \supseteq \cdots$ measures, for each $\sigma \in G$ , how close $\sigma$ is to the identity in the $\mathfrak{m}$ -adic topology. The quotient $G/G_0$ recovers the residue-field extension (unramified part), $G_0/G_1$ is cyclic of order prime to $p$ (tame part), and $G_1$ is a $p$ -group (wild part). All finer arithmetic — the different exponent, the conductor, the behavior under class field theory — is read off from this single filtration via the Hasse–Arf theorem and the upper/lower numbering translation.

DefinitionDiscrete Valuation and Local Field

A discrete valuation on a field $K$ is a surjective homomorphism $v: K^\times \to \mathbb{Z}$ satisfying $v(x+y) \ge \min(v(x), v(y))$ , extended by $v(0) = +\infty$ . The valuation ring is

1

\quad \mathcal{O}_K = \{x \in K : v(x) \ge 0\}, \qquad \mathfrak{m} = \{x \in K : v(x) > 0\},

Derivation hint: $\mathcal{O}_K$ is closed under addition by the ultrametric inequality and under multiplication since $v$ is additive; $\mathfrak{m}$ is its unique maximal ideal because any $x$ with $v(x)=0$ is a unit. A uniformizer $\pi$ is any element with $v(\pi) = 1$ , so $\mathfrak{m} = (\pi)$ . The residue field is $k = \mathcal{O}_K/\mathfrak{m}$ . A local field is a field complete for $v$ with finite residue field $k = \mathbb{F}_q$ , $q = p^f$ . Every $x \in K^\times$ has a unique representation

2

\quad x = \pi^{v(x)} u, \qquad u \in \mathcal{O}_K^\times,

Derivation hint: set $u = x\pi^{-v(x)}$ ; then $v(u) = v(x) - v(x) = 0$ so $u$ is a unit.

DefinitionRamification Index and Residue Degree

Let $L/K$ be a finite extension of local fields with valuations $w \mid v$ . The ramification index $e$ and residue degree $f$ are

3

\quad e = e(L/K) = [w(L^\times) : v(K^\times)], \qquad f = f(L/K) = [\ell : k],

Derivation hint: $w$ restricted to $K^\times$ is a positive multiple of $v$ ; the index of the value groups is $e$ , and $f$ is the degree of the induced residue-field extension $\ell = \mathcal{O}_L/\mathfrak{m}_L$ over $k$ . The fundamental identity for complete fields is

4

\quad [L:K] = e\,f,

Derivation hint: $\mathcal{O}_L$ is a free $\mathcal{O}_K$ -module of rank $[L:K]$ ; choosing $f$ residue representatives and multiplying by $1, \pi_L, \dots, \pi_L^{e-1}$ gives an $\mathcal{O}_K$ -basis, so the rank is $ef$ . The extension is unramified if $e=1$ and $\ell/k$ is separable; totally ramified if $f=1$ ; tamely ramified if $p \nmid e$ ; wildly ramified if $p \mid e$ .

TheoremStructure of Finite Extensions

Let $L/K$ be a finite separable extension of local fields with residue characteristic $p$ . Then there is a unique maximal unramified subextension $K \subseteq K^{nr} \subseteq L$ , and: 1. $K^{nr}/K$ corresponds bijectively to the residue extension $\ell/k$ ; for each $f$ there is exactly one unramified extension of degree $f$ , namely $K(\zeta_{q^f - 1})$ , with $\mathrm{Gal}(K^{nr}/K) \cong \mathrm{Gal}(\ell/k) \cong \mathbb{Z}/f$ generated by Frobenius. 2. $L/K^{nr}$ is totally ramified. 3. (Tame extensions) A totally tamely ramified extension of degree $e$ has the form $K(\pi^{1/e})$ for some uniformizer $\pi$ , and is cyclic.

DefinitionHigher Ramification Groups — Lower Numbering

Let $L/K$ be finite Galois with group $G$ , $\mathcal{O}_L$ the valuation ring with uniformizer $\pi_L$ and valuation $w$ (normalized $w(\pi_L)=1$ ). For a real number $i \ge -1$ define

5

\quad G_i = \{\sigma \in G : w(\sigma(x) - x) \ge i+1 \text{ for all } x \in \mathcal{O}_L\}.

Derivation hint: equivalently $G_i = \{\sigma : w(\sigma(\pi_L) - \pi_L) \ge i+1\}$ when $\mathcal{O}_L = \mathcal{O}_K[\pi_L]$ , since $\sigma$ acts on a polynomial in $\pi_L$ and the minimum is achieved on the generator. These form a decreasing filtration $G = G_{-1} \supseteq G_0 \supseteq G_1 \supseteq \cdots$ with:

$G_{-1} = G$ ,
$G_0 =$ the inertia group $I(L/K)$ , so $G/G_0 \cong \mathrm{Gal}(\ell/k)$ ,
$G_1 =$ the wild inertia group, the $p$ -Sylow subgroup of $G_0$ ,
$G_0/G_1$ is cyclic of order prime to $p$ (the tame quotient).

TheoremQuotient Embeddings and the Different

For $i \ge 0$ there are canonical injective homomorphisms

6

\quad G_0/G_1 \hookrightarrow \ell^\times, \qquad G_i/G_{i+1} \hookrightarrow (\mathfrak{m}_L^i/\mathfrak{m}_L^{i+1}, +) \cong (\ell, +) \text{ for } i \ge 1,

Derivation hint: send $\sigma \mapsto \sigma(\pi_L)/\pi_L \bmod \mathfrak{m}_L$ for $i=0$ ; for $i \ge 1$ send $\sigma \mapsto (\sigma(\pi_L) - \pi_L)/\pi_L^{i+1} \bmod \mathfrak{m}_L$ . Each is a homomorphism modulo the next group. The different $\mathfrak{d}_{L/K} = \mathfrak{m}_L^{d}$ has exponent given by

7

\quad d = \sum_{i=0}^{\infty} (|G_i| - 1),

Derivation hint: the different is generated by $g'(\pi_L)$ for the minimal polynomial $g$ of $\pi_L$ ; expanding $g'(\pi_L) = \prod_{\sigma \neq 1}(\pi_L - \sigma\pi_L)$ and counting $w(\pi_L - \sigma\pi_L) = i+1$ for $\sigma \in G_i \setminus G_{i+1}$ gives the sum.

ProofThe Tame Quotient is Cyclic of Order Prime to p

Step 1. Let $\sigma \in G_0$ (inertia). Define $\theta(\sigma) = \sigma(\pi_L)/\pi_L \bmod \mathfrak{m}_L \in \ell^\times$ . Since $\sigma$ fixes the residue field pointwise (it lies in inertia) and permutes uniformizers, $\sigma(\pi_L) = \pi_L \cdot u_\sigma$ with $u_\sigma \in \mathcal{O}_L^\times$ , and $\theta(\sigma) = \bar u_\sigma$ .

Step 2. Check $\theta$ is a homomorphism: $\sigma\tau(\pi_L) = \sigma(\pi_L u_\tau) = \pi_L u_\sigma \cdot \sigma(u_\tau)$ . Since $\sigma$ acts trivially on residues, $\overline{\sigma(u_\tau)} = \bar u_\tau$ , so $\theta(\sigma\tau) = \theta(\sigma)\theta(\tau)$ .

Step 3. The kernel of $\theta$ is exactly $G_1$ by definition [5]: $\theta(\sigma) = 1 \iff \sigma(\pi_L)/\pi_L \equiv 1 \iff w(\sigma\pi_L - \pi_L) \ge 2 \iff \sigma \in G_1$ .

Step 4. Thus $G_0/G_1 \hookrightarrow \ell^\times$ . Since $\ell^\times$ is cyclic of order $q^f - 1$ (finite multiplicative group of a finite field), $G_0/G_1$ is cyclic of order dividing $q^f - 1$ , hence prime to $p$ .

Step 5. For $i \ge 1$ the analogous map lands in $(\ell, +)$ , an elementary abelian $p$ -group, forcing each $G_i/G_{i+1}$ to be an elementary abelian $p$ -group. Since the filtration is finite and intersects trivially, $G_1$ is a $p$ -group — the unique $p$ -Sylow of $G_0$ . $\blacksquare$

∎

ProofThe Different and Discriminant Conductor — Upper Numbering

Step 1. Define the Herbrand function $\psi$ via its inverse $\varphi$ :

8

\quad \varphi_{L/K}(u) = \int_0^u \frac{dt}{[G_0 : G_t]}, \qquad \psi_{L/K} = \varphi_{L/K}^{-1},

Derivation hint: $\varphi$ is a piecewise-linear increasing concave function with $\varphi(0)=0$ ; on $[i, i+1]$ its slope is $1/[G_0:G_i]$ , encoding the "speed" of the filtration.

Step 2. Define the upper numbering $G^v = G_{\psi(v)}$ . Herbrand's theorem states that upper numbering is compatible with quotients: for a normal subgroup $H \triangleleft G$ corresponding to $L/K \supseteq M/K$ ,

9

\quad (G/H)^v = G^v H / H,

Derivation hint: lower numbering is compatible with subgroups ( $H_i = G_i \cap H$ ) while upper numbering is compatible with quotients; the Herbrand function mediates the change of normalization via $\psi_{L/K} = \psi_{M/K} \circ \psi_{L/M}$ .

Step 3. This makes upper numbering well-defined on the infinite Galois group $\mathrm{Gal}(K^{sep}/K)$ as a filtration $G^v$ , since it survives inverse limits.

Step 4. (Hasse–Arf) For abelian $L/K$ , the jumps in the upper-numbering filtration occur at integers: if $G^v \neq G^{v+\epsilon}$ then $v \in \mathbb{Z}$ . This integrality is exactly what is needed for the conductor of an abelian character to be a genuine ideal $\mathfrak{m}_K^{c}$ with $c \in \mathbb{Z}_{\ge 0}$ . $\blacksquare$

∎

mathematician · David Hilbert | 1862–1943

Hilbert systematized ramification in his monumental Zahlbericht (1897), introducing the decomposition group, inertia group, and the higher ramification groups now bearing names like the "Hilbert ramification filtration." His program for explicit class field theory — finding generators of abelian extensions — framed the questions that local field theory would answer. Hilbert's 9th and 12th problems directly concern reciprocity laws and explicit construction of abelian extensions, both resolved (locally) through the machinery described here. He insisted on conceptual clarity over computation, and the abstract filtration approach is in his spirit, though the completion-theoretic foundations are due to Kurt Hensel (who introduced $p$ -adic numbers, 1897) and the systematic local theory to Helmut Hasse and Wolfgang Krull in the 1920s–30s.

🧠 Intuition

Think of a uniformizer $\pi$ as a "microscope coordinate" — measuring distance to the closed point of $\mathrm{Spec}\,\mathcal{O}_K$ . An element of $\mathcal{O}_K$ is a Taylor series in $\pi$ (literally, in $\mathbb{F}_q((t))$ ). A Galois automorphism $\sigma$ either:

moves the residue field (changes the "constant term" structure) — this is the unramified/Frobenius part, detected by $G/G_0$ ;
fixes residues but rotates $\pi$ by a root of unity $\zeta$ , i.e. $\sigma(\pi) \approx \zeta\pi$ — tame ramification, detected by $G_0/G_1 \hookrightarrow \ell^\times$ ;
fixes residues and the leading rotation, perturbing only at higher order $\sigma(\pi) = \pi(1 + a\pi^i + \cdots)$ — wild ramification, detected by $G_i/G_{i+1}$ .

The lower numbering counts $\pi$ -adic order of the perturbation ( $\sigma$ sits in $G_i$ if it disturbs $\pi$ only past order $i$ ). The upper numbering rescales so the filtration behaves well under passing to quotients — like switching from "arc length on $L$ " to "arc length on $K$ ." The Hasse–Arf integrality says: from the base field's viewpoint, the abelian jumps land exactly on integer ticks, which is why conductors are honest ideals.

Figure

The ramification filtration of

G = \mathrm{Gal}(L/K)

and its quotients

ExampleRamification in

\mathbb{Q}_p(\zeta_p)/\mathbb{Q}_p

Given. Let $p$ be an odd prime, $K = \mathbb{Q}_p$ , $L = \mathbb{Q}_p(\zeta_p)$ where $\zeta_p$ is a primitive $p$ -th root of unity. Compute $e, f$ , the different exponent, and the ramification filtration.

Step 1 — degree. $[L:K] = \deg \Phi_p = p-1$ , where $\Phi_p(x) = (x^p-1)/(x-1)$ .

Step 2 — uniformizer. Set $\pi = \zeta_p - 1$ . Then $\Phi_p(1 + \pi) = 0$ and evaluating $\Phi_p(x+1)$ shows its Newton polygon gives $v_p(\pi) = 1/(p-1)$ , so $w(\pi) = 1$ in $L$ . Hence the value group expands by factor $p-1$ : $e = p-1$ .

Step 3 — residue degree. By [4], $ef = p-1$ , and since $e = p-1$ we get $f = 1$ . The extension is totally (and wildly) ramified.

Step 4 — Galois group. $G \cong (\mathbb{Z}/p)^\times \cong \mathbb{Z}/(p-1)$ , with $\sigma_a(\zeta_p) = \zeta_p^a$ .

Step 5 — filtration. Compute $w(\sigma_a(\pi) - \pi) = w(\zeta_p^a - \zeta_p)$ . One finds $G_0 = G$ (totally ramified), and $G_1 = \cdots = G_{?}$ . Since $G_0$ is cyclic of order $p-1$ prime to $p$ , the extension is actually tame! So $G_1 = \{1\}$ .

Wait — reconcile: $e = p-1$ is prime to $p$ , so this is tame, not wild. The filtration is $G_0 = G$ , $G_1 = \{1\}$ .

Step 6 — different. By [7], $d = (|G_0| - 1) = (p-1) - 1 = p-2$ .

\boxed{e = p-1,\quad f=1,\quad \mathfrak{d}_{L/K} = \mathfrak{m}_L^{\,p-2}, \quad G_0=G,\ G_1=1 \ (\text{tame})}

ExampleWild ramification in

\mathbb{Q}_2(\zeta_4)/\mathbb{Q}_2

Given. $K = \mathbb{Q}_2$ , $L = \mathbb{Q}_2(i) = \mathbb{Q}_2(\zeta_4)$ . Here $p = 2$ , so the tame story above fails.

Step 1 — degree. $[L:K] = 2$ , minimal polynomial $x^2 + 1$ .

Step 2 — ramification. The discriminant of $\mathbb{Z}[i]$ is $-4$ , divisible by $2$ , so $2$ ramifies: $e = 2$ , $f = 1$ , totally ramified. A uniformizer is $\pi = 1 + i$ since $\pi^2 = 2i$ and $w(\pi) = 1$ .

Step 3 — wildness. Since $p = 2 \mid e = 2$ , this is wildly ramified: $G_1 \neq \{1\}$ .

Step 4 — filtration. $G = \{1, \sigma\}$ with $\sigma(i) = -i$ , so $\sigma(\pi) = 1 - i = \pi - 2i$ . Compute

w(\sigma(\pi) - \pi) = w(-2i) = w(2) = 2.

By definition [5], $\sigma \in G_i \iff i + 1 \le 2 \iff i \le 1$ . Hence $G_0 = G_1 = G$ and $G_2 = \{1\}$ .

Step 5 — different. By [7]:

d = (|G_0| - 1) + (|G_1| - 1) = (2-1) + (2-1) = 2.

Step 6 — discriminant check. $v_2(\mathrm{disc}_{L/K}) = f \cdot d = 1 \cdot 2 = 2$ , matching $\mathrm{disc} = -4 = -2^2$ . ✓

\boxed{e=2,\ f=1 \text{ (wild)};\quad G_0=G_1=G,\ G_2=1;\quad \mathfrak{d}_{L/K}=\mathfrak{m}_L^2}

⚠️ Common mistake

"Unramified means trivial Galois action." False. Unramified extensions have nontrivial Galois groups (cyclic, generated by Frobenius) — they just don't expand the value group. The Galois action is fully captured by the residue extension. Triviality of higher ramification ( $G_1 = 1$ ) is tameness, not unramifiedness.

"The lower numbering passes to quotients." It does not. Lower numbering ( $G_i$ ) is compatible with subgroups ( $H_i = G_i \cap H$ for $H \le G$ ), while upper numbering ( $G^v$ ) is compatible with quotients. Confusing these is the single most common error in ramification computations. The Herbrand function $\psi/\varphi$ exists precisely to translate.

" $e f = [L:K]$ always." This requires completeness (or at least Henselian local rings, or the extension of the absolute value being unique). For a general extension of global fields, $\sum_{w \mid v} e_w f_w = [L:K]$ — the sum over multiple primes above $v$ . Locality collapses the sum to one term.

Counterexample

Hasse–Arf fails for non-abelian groups. Consider a totally ramified extension with Galois group $G = S_3$ (or the quaternion group $Q_8$ ) over a local field. The upper-numbering jumps need not be integers. Concretely, one can construct $L/\mathbb{Q}_2$ with $\mathrm{Gal}(L/\mathbb{Q}_2) = Q_8$ whose ramification jump in upper numbering occurs at a non-integer value such as $v = 3/2$ . This is not a contradiction — Hasse–Arf is asserted only for abelian extensions, and the integrality is exactly the abelian phenomenon underpinning class field theory's conductor. For non-abelian $G$ , the conductor of a higher-dimensional representation (Artin conductor) is still an integer, but this requires the deeper Artin/Swan conductor formalism, not the naive jump.

connection

Local fields sit at the crossroads of several theories:

Local Class Field Theory establishes the local reciprocity isomorphism $K^\times \xrightarrow{\sim} W(K^{ab}/K)$ (Weil group) and translates the ramification filtration $G^v$ on the Galois side into the unit filtration $1 + \mathfrak{m}_K^n$ on $K^\times$ — the conductor-discriminant dictionary.
Global Class Field Theory assembles local data over all places via the idele class group $\mathbb{A}_F^\times / F^\times$ ; Hasse's local-global principle says global reciprocity is the "sum" of local reciprocities.
Galois Representations and the Langlands Program: the local Langlands correspondence matches representations of $\mathrm{Gal}(\bar K/K)$ with those of $\mathrm{GL}_n(K)$ , with ramification controlling conductor and depth.
Algebraic Number Theory: the different and discriminant from [7] feed directly into the conductor-discriminant formula and the analytic theory of Dedekind Zeta Functions.
Étale Cohomology: ramification groups are the arithmetic analogue of the monodromy/vanishing-cycle filtration in geometry; wild ramification mirrors irregular singularities of connections.

🔬 Where this appears

⏳ History

- 1897 — Kurt Hensel introduces the

p

-adic numbers, motivated by analogy with power series; the local viewpoint is born.

- 1897 — Hilbert's Zahlbericht codifies decomposition, inertia, and higher ramification groups for number fields; the "Hilbert filtration."

- 1910s — Hensel and Hasse develop the analytic and algebraic theory of

\mathbb{Q}_p

; Hensel's Lemma becomes a workhorse.

- 1920s — Emil Artin and Hasse formulate the reciprocity law; the Artin map links Galois groups to ideal class data.

- 1930 — Hasse proves the local-global (Hasse) principle for central simple algebras with Brauer and Noether; local class field theory takes shape.

- 1931 — Hasse and Arf-era work; Cahit Arf proves the Hasse–Arf theorem (1939) on integrality of abelian upper-numbering jumps.

- 1936 — Claude Chevalley introduces ideles and gives a purely local, algebraic foundation for class field theory, freeing it from analysis.

- 1950s–60s — Jean-Pierre Serre's Corps Locaux (Local Fields, 1962) gives the definitive modern treatment of the ramification filtration, Herbrand's theorem, and upper numbering.

- 1980s–2010s — Abbes–Saito extend ramification theory to imperfect residue fields and non-Galois settings; Scholze's perfectoid spaces (2012) reframe deeply ramified extensions.

open-problem

Ramification with imperfect residue fields. The classical filtration assumes a perfect (here finite) residue field. For local fields with imperfect residue field — arising in higher-dimensional local fields and in geometry (function fields of varieties) — the naive definition [5] breaks down: there is no canonical Herbrand function, and the different does not determine the filtration. The Abbes–Saito ramification theory (2002) builds a non-logarithmic and logarithmic filtration using rigid geometry (tubular neighborhoods of immersions), but key questions remain:

A clean Hasse–Arf theorem in the imperfect-residue, higher-dimensional setting is subtle; integrality results (Abbes–Saito, Saito) hold under hypotheses but a fully general statement is open.
The relationship between wild ramification and Euler characteristics (the Grothendieck–Ogg–Shafarevich formula and its higher-dimensional generalizations via Saito's characteristic cycle and Beilinson's singular support) is an active frontier.
For $p$ -adic differential equations and the $p$ -adic Langlands program, the precise dictionary between ramification slopes (Newton polygon of the Swan conductor) and irregularity of connections is still being mapped (work of Kedlaya, Xiao on Hasse–Arf in the differential setting).
Explicit wild ramification (computing $G_i$ for a given polynomial) remains computationally hard; there is no general closed form analogous to the tame case $K(\pi^{1/e})$ .

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also