Dirichlet's Unit Theorem

One-line statement: For a number field

K

with

r_{1}

real embeddings and

r_{2}

pairs of complex embeddings, the unit group of its ring of integers is finitely generated of rank

r_{1} + r_{2} - 1

, modulo the finite cyclic group of roots of unity. Key result in words: The units of

O_{K}

form a group isomorphic to

μ_{K} \times Z^{r_{1} + r_{2} - 1}

, where

μ_{K}

is the (finite) group of roots of unity in

K

. Where it applies: Algebraic number theory, the structure of rings of integers, Pell's equation, class number formulas, regulators,

L

-functions. Who/when: Johann Peter Gustav Lejeune Dirichlet, announced 1846.

Dirichlet's Unit Theorem describes the multiplicative structure of the group of units $O_{K}^{\times}$ of the ring of integers $O_{K}$ of an algebraic number field $K$ . While additively $O_{K}$ is a free $Z$ -module of rank $n = [K : Q]$ , its units form a more subtle object: a finitely generated abelian group whose free rank is determined entirely by the signature $(r_{1}, r_{2})$ of the field, namely $r = r_{1} + r_{2} - 1$ . The torsion part is exactly the finite cyclic group $μ_{K}$ of roots of unity lying in $K$ . The surprising fact is that the rank depends only on how $K$ embeds into $R$ and $C$ — purely archimedean data — and not at all on the arithmetic of primes inside $O_{K}$ . The proof is a beautiful application of the geometry of numbers, converting a multiplicative problem into a question about lattices in Euclidean space.

The Core Insight

The map $u \mapsto (lo g ∣ σ_{1} (u) ∣, \dots)$ — taking logarithms of the absolute values of all the archimedean embeddings of a unit — turns multiplication into addition and converts the multiplicative unit group into an additive lattice. Because units have norm $\pm 1$ , their image lands inside a hyperplane (the trace-zero hyperplane). The two key facts are: (1) the kernel of this logarithmic embedding is exactly the finite group of roots of unity, and (2) its image is a full-rank lattice in that hyperplane. The hyperplane has dimension $r_{1} + r_{2} - 1$ , and that number is the rank.

Formal Statement

Number field, embeddings, signature

Let $K$ be a number field: a finite extension of $Q$ of degree $n = [K : Q]$ . There are exactly $n$ field embeddings $K ↪ C$ . Of these:

$r_{1}$ are real embeddings $σ_{1}, \dots, σ_{r_{1}} : K ↪ R$ ;
$2 r_{2}$ are complex (non-real) embeddings, occurring in conjugate pairs $τ_{j}, \overset{τ}{ˉ}_{j}$ for $j = 1, \dots, r_{2}$ .

The pair $(r_{1}, r_{2})$ is the signature of $K$ , and

n = r_{1} + 2 r_{2} .

⟳Complex embeddings come in conjugate pairs (compose with complex conjugation), so they contribute

2 r_{2}

; the rest are real.

The ring of integers $O_{K}$ is the integral closure of $Z$ in $K$ ; its unit group $O_{K}^{\times}$ consists of elements $u$ with $u^{- 1} \in O_{K}$ , equivalently the elements of norm $N_{K / Q} (u) = \pm 1$ . The set $μ_{K}$ of roots of unity in $K$ is a finite cyclic subgroup.

The logarithmic embedding

Fix a labeling of the archimedean places: $r_{1}$ real embeddings $σ_{i}$ and $r_{2}$ complex pairs (one per pair) $τ_{j}$ . Define $ℓ : K^{\times} \to R^{r_{1} + r_{2}}$ by

ℓ (x) = (lo g ∣ σ_{1} (x) ∣, \dots, lo g ∣ σ_{r_{1}} (x) ∣, 2 lo g ∣ τ_{1} (x) ∣, \dots, 2 lo g ∣ τ_{r_{2}} (x) ∣) .

⟳The factor

2

on complex places reflects that

∣ z ∣^{2} = z \overset{z}{ˉ}

accounts for both members of a conjugate pair; it makes the product formula and norm clean.

For a unit $u$ , the absolute norm satisfies $∣ N_{K / Q} (u) ∣ = \prod_{i} ∣ σ_{i} (u) ∣ \cdot \prod_{j} ∣ τ_{j} (u) ∣^{2} = 1$ , so

k = 1 \sum r_{1} + r_{2} ℓ (u)_{k} = 0.

⟳Take

lo g

∣ N (u) ∣ = 1

, which is the product of all archimedean absolute values weighted by local degrees (

1

for real,

2

for complex).

Thus $ℓ (O_{K}^{\times})$ lies in the trace-zero hyperplane

H = {(x_{1}, \dots, x_{r_{1} + r_{2}}) : \sum_{k} x_{k} = 0} \subset R^{r_{1} + r_{2}}, dim H = r_{1} + r_{2} - 1.

Dirichlet's Unit Theorem

Let $K$ be a number field with signature $(r_{1}, r_{2})$ and set $r = r_{1} + r_{2} - 1$ . Then the unit group is finitely generated abelian:

O_{K}^{\times} ≅ μ_{K} \times Z^{r}, r = r_{1} + r_{2} - 1.

⟳Combine that

ker ℓ ∣_{O_{K}^{\times}} = μ_{K}

(torsion) with the fact that

ℓ (O_{K}^{\times})

is a full lattice in

H

(free part of rank

dim H = r

Equivalently, there exist fundamental units $ε_{1}, \dots, ε_{r} \in O_{K}^{\times}$ such that every unit $u$ is uniquely

u = ζ ε_{1}^{m_{1}} ε_{2}^{m_{2}} \dots ε_{r}^{m_{r}}, ζ \in μ_{K}, m_{i} \in Z .

⟳A basis

ℓ (ε_{i})

of the lattice

ℓ (O_{K}^{\times})

pulls back to generators of the free part;

ζ

accounts for the kernel.

Why It Holds / Derivation

Proof via the geometry of numbers

We prove $O_{K}^{\times} ≅ μ_{K} \times Z^{r}$ in four steps.

Step 1 — The kernel of $ℓ$ is $μ_{K}$ . Let

u \in O_{K}^{\times}

with

ℓ (u) = 0

. Then

∣ σ (u) ∣ = 1

for every archimedean embedding

σ

. Hence

u

and all its conjugates lie on the unit circle, so the set

{u, u^{2}, u^{3}, \dots}

has all conjugates bounded by

1

. The minimal polynomials of

u^{m}

are monic integer polynomials whose roots are bounded, so their coefficients are bounded; only finitely many such polynomials exist, forcing the powers

u^{m}

to take finitely many values. Therefore

u^{a} = u^{b}

for some

a > b

, so

u^{a - b} = 1

and

u

is a root of unity. Conversely roots of unity have

∣ σ (ζ) ∣ = 1

everywhere, so

ℓ (ζ) = 0

. Thus

ker (ℓ ∣_{O_{K}^{\times}}) = μ_{K}, a finite cyclic group.

⟳Kronecker's theorem — an algebraic integer all of whose conjugates have absolute value

1

is a root of unity; finiteness comes from boundedness of conjugates plus integrality.

Step 2 — $Γ := ℓ (O_{K}^{\times})$ is a discrete subgroup (a lattice) of $H$ . For any bound

B

, the units

u

with

∣ σ (u) ∣ \leq e^{B}

for all

σ

form a finite set: such

u

are algebraic integers with bounded conjugates, hence have minimal polynomials with bounded integer coefficients. Intersecting

Γ

with any bounded ball thus gives finitely many points, so

Γ

is discrete. A discrete subgroup of a finite-dimensional real vector space is a free abelian group whose rank equals the dimension of the

R

-span. Hence

Γ

is a lattice of some rank

s \leq dim H = r

⟳Discreteness ⇒ free of rank = dimension of

R

-span; finiteness of bounded units gives discreteness.

Step 3 — Structure of $O_{K}^{\times}$ . The short exact sequence

1 \to μ_{K} \to O_{K}^{\times} ℓ Γ \to 0

has

Γ ≅ Z^{s}

free, so it splits:

O_{K}^{\times} ≅ μ_{K} \times Z^{s} .

⟳An extension of a free abelian group by any group splits, because

Z^{s}

is projective.

It remains to show $s = r$ , i.e. $Γ$ has full rank in $H$ .

Step 4 — Full rank via Minkowski's lattice-point theorem. This is the crux. We must produce enough units so that

ℓ (O_{K}^{\times})

spans

H

. Equivalently: for each coordinate index

k

, there is a unit whose

ℓ

-image is not nonpositive in coordinate

k

alone — concretely, it suffices to show that for every nonzero linear functional vanishing on... we instead argue directly. Embed

O_{K}

as a full lattice

Λ

in the Minkowski space

R^{r_{1}} \times C^{r_{2}} ≅ R^{n}

via

x \mapsto (σ_{i} (x); τ_{j} (x))

; its covolume is

2^{- r_{2}} ∣ d_{K} ∣

where

d_{K}

is the discriminant.

Fix a target index $m$ . Choose positive reals $c_{1}, \dots, c_{r_{1} + r_{2}}$ with $\prod_{k} c_{k}^{e_{k}} > (2/ π)^{r_{2}} ∣ d_{K} ∣$ (here $e_{k} = 1$ for real, $2$ for complex places). By Minkowski's theorem, the symmetric convex region ${∣ σ_{i} (x) ∣ \leq c_{i}, ∣ τ_{j} (x) ∣ \leq c_{r_{1} + j}}$ contains a nonzero $x \in O_{K}$ . Its norm satisfies $1 \leq ∣ N (x) ∣ \leq \prod_{k} c_{k}^{e_{k}} =: C$ (bounded independently of which $c_{k}$ we shrink/grow). By varying the $c_{k}$ so that one chosen coordinate is forced small while their product stays $\geq$ the Minkowski bound, one generates infinitely many algebraic integers $x_{i}$ with $∣ N (x_{i}) ∣ \leq C$ . Since only finitely many ideals have norm $\leq C$ , infinitely many of the principal ideals $(x_{i})$ coincide; then ratios $x_{i} / x_{j}$ are units, and the construction can be arranged so that these units' logarithmic images point "in a chosen direction." Doing this for each direction shows $Γ$ is not contained in any proper subspace of $H$ , hence

rank_{Z} Γ = dim H = r_{1} + r_{2} - 1 = r .

⟳Minkowski guarantees lattice points in large symmetric convex bodies; pigeonholing by ideal norm produces units, and steering the convex body's shape steers their log-images to span

H

Combining Steps 3 and 4 gives $O_{K}^{\times} ≅ μ_{K} \times Z^{r}$ . $■$

The logarithmic embedding sends units to a lattice in the trace-zero hyperplane. Shown: a real quadratic field, $r_1=2$, $r_2=0$, $r=1$.

Johann Peter Gustav Lejeune Dirichlet | 1805–1859

Dirichlet was a German mathematician whose work bridged the analytic and arithmetic worlds. He is renowned for Dirichlet's theorem on primes in arithmetic progressions (introducing $L$ -functions and characters), for sharpening the notion of a function, and for the pigeonhole principle (the "Dirichlet box principle") — which is exactly the combinatorial heart of the unit theorem's finiteness arguments. He announced the unit theorem in 1846, reportedly working out key ideas during a concert. He succeeded Gauss at Göttingen in 1855. His blend of careful analysis with number theory set the template for analytic number theory; the unit theorem is a jewel of his geometric-arithmetic intuition, later cast cleanly in Minkowski's geometry of numbers.

Building Intuition

Think of a unit $u$ as a single object viewed through $r_{1} + r_{2}$ "windows" — one per archimedean place. Each window reports the size $lo g ∣ σ (u) ∣$ . Multiplying units adds these sizes window-by-window, so the unit group becomes an additive grid (lattice). Because the total "size budget" must be zero (the norm is $\pm 1$ ), all reports lie on a balance-beam: a hyperplane of dimension one less than the number of windows. That subtracted dimension is the " $- 1$ " in $r_{1} + r_{2} - 1$ .

Why are roots of unity invisible? They have size $0$ in every window — they sit at the origin of the lattice and contribute nothing to the free directions. They are the "torsion" the theorem packages separately.

The hard part of the proof is showing the lattice is full — that units genuinely fill out the whole hyperplane and aren't squeezed into a lower-dimensional sliver. This is exactly where you need to produce units, not just describe them, and that is Minkowski's geometry of numbers: large symmetric boxes must contain lattice points (integers of $O_{K}$ ), and pigeonholing those by ideal norm manufactures units pointing in every direction.

Worked Examples

Real quadratic field: $K = \mathbb{Q}(\sqrt{2})$

Given:

K = Q (2)

, with

O_{K} = Z [2]

. Step 1 — Signature.

2

is real, so both embeddings (

2 \mapsto + 2

and

\mapsto - 2

) are real:

r_{1} = 2

r_{2} = 0

. Then

r = r_{1} + r_{2} - 1 = 2 + 0 - 1 = 1.

O_{K}^{\times} ≅ μ_{K} \times Z

. Step 2 — Roots of unity. The only roots of unity in a real field are

\pm 1

, so

μ_{K} = {\pm 1} ≅ Z /2

. Step 3 — Find the fundamental unit. We need

a + b 2

with norm

a^{2} - 2 b^{2} = \pm 1

. Smallest nontrivial solution:

a = 1, b = 1

gives

1^{2} - 2 \cdot 1^{2} = - 1

. Check

ε = 1 + 2

is a unit:

(1 + 2) (- 1 + 2) = 1

, so

ε^{- 1} = 2 - 1

. ✓ Step 4 — Structure. Every unit is

\pm (1 + 2)^{m}

m \in Z

O_{Q (2)}^{\times} = {\pm 1} \times ⟨ 1 + 2 ⟩ ≅ Z /2 \times Z, ε = 1 + 2

(Powers of $ε$ encode all solutions to the Pell-type equation $a^{2} - 2 b^{2} = \pm 1$ .)

Imaginary quadratic vs. cubic

Part A — $K = Q (i)$ . Embeddings:

i \mapsto \pm i

, a conjugate pair, so

r_{1} = 0

r_{2} = 1

. Then

r = 0 + 1 - 1 = 0

: no fundamental units at all. The unit group is purely torsion:

O_{Q (i)}^{\times} = μ_{K} = {1, i, - 1, - i} ≅ Z /4

This matches the classical fact that the Gaussian integers have only

\pm 1, \pm i

as units. Part B — totally real cubic, $K = Q (θ)$ , $θ^{3} - 3 θ - 1 = 0$ . The cubic

x^{3} - 3 x - 1

has discriminant

81 = 9^{2} > 0

, hence three real roots:

r_{1} = 3

r_{2} = 0

. Then

r = 3 + 0 - 1 = 2.

O_{K}^{\times} ≅ {\pm 1} \times Z^{2}

— two independent fundamental units. (One may take

θ

and

θ + 1

as a system of fundamental units; both have norm

\pm 1

rank = 2, O_{K}^{\times} ≅ Z /2 \times Z^{2}

Common Misconceptions

Pitfall 1 — "Rank equals degree." The free rank is

r_{1} + r_{2} - 1

, not

n = r_{1} + 2 r_{2}

. For

Q (i)

the degree is

2

but the rank is

0

. Pitfall 2 — "Fundamental units are unique." The system

{ε_{1}, \dots, ε_{r}}

is determined only up to multiplication by roots of unity and up to a

GL_{r} (Z)

change of basis. What is canonical is the rank

r

, the group

μ_{K}

, and the regulator

R_{K}

(the covolume of the unit lattice

Γ

H

). Pitfall 3 — "Roots of unity are always $\pm 1$ ." Only for fields with a real embedding. Imaginary quadratic fields can have

μ_{K} = Z /4

(

Q (i)

) or

Z /6

(

Q (- 3)

), and cyclotomic fields

Q (ζ_{n})

have

μ_{K} \supseteq ⟨ ζ_{n} ⟩

. Pitfall 4 — "Finding fundamental units is easy." Determining a fundamental unit can require exponentially large solutions; for

Q (d)

the fundamental unit's size can be astronomically larger than

d

(related to the regulator and continued fraction period).

Why the "$-1$" cannot be dropped

One might naively guess the unit lattice fills all of $R^{r_{1} + r_{2}}$ , giving rank $r_{1} + r_{2}$ . It does not. Consider any totally real quadratic field, say $Q (5)$ : here $r_{1} + r_{2} = 2$ , but the unit rank is $1$ , not $2$ . The fundamental unit is $ε = \frac{1 + 5}{2}$ (the golden ratio), with $ℓ (ε) = (lo g ε, lo g ∣ \overset{ε}{ˉ} ∣) = (lo g ε, - lo g ε)$ — necessarily lying on the line $x_{1} + x_{2} = 0$ . No unit can escape this line because $∣ N (u) ∣ = 1$ forces $\sum ℓ (u)_{k} = 0$ . The norm constraint is precisely what removes one dimension.

Connections

Dirichlet's Unit Theorem sits at a crossroads of several theories:

It is one half of the finiteness duo of algebraic number theory, the other being finiteness of the Ideal Class Group. Together they describe $O_{K}$ 's deviation from being a PID (class group) and the redundancy among its generators (units).
The covolume of the unit lattice is the regulator $R_{K}$ , which appears in the Analytic Class Number Formula alongside $h_{K}$ , $∣ d_{K} ∣$ , $∣ μ_{K} ∣$ as the residue of the Dedekind Zeta Function at $s = 1$ .
It is the function-field-and- $S$ -units prototype: replacing all primes by a finite set $S$ gives the $S$ -unit theorem ( $O_{K, S}^{\times}$ has rank $∣ S ∣ - 1 + r_{1} + r_{2} - 1$ when $S$ contains the archimedean places), foundational for Diophantine Equations and the unit equation.
The proof technique is the canonical application of Minkowski's Theorem / the geometry of numbers, the same machinery proving finiteness of class number.
For abelian fields it interacts with cyclotomic units and circular units, whose index in the full unit group is governed by the class number (Iwasawa Theory).

Real-World Applications

Pell's equation $x^{2} - d y^{2} = 1$ . All solutions are powers of the fundamental unit of $Q (d)$ ; the theorem (rank $1$ ) explains why solutions form a single geometric progression. Used historically and in cryptographic key-size estimates.
Class number computations. The analytic class number formula needs the regulator $R_{K}$ — defined directly from a system of fundamental units — making unit computation essential in computer algebra systems (PARI/GP, Magma, SageMath).
Integer factorization and number field sieve. The structure of units and $S$ -units in chosen number fields underpins relation-collection in the general number field sieve, the fastest known classical factoring algorithm.
Solving norm-form and Thue equations. $S$ -unit equations reduce many Diophantine problems to finitely many cases; effective bounds (Baker's theory) make them algorithmically solvable.
Cryptography over number fields. Ideal-lattice cryptography (Ring-LWE, NTRU) relies on the geometry of $O_{K}$ and its units; short-unit / log-unit lattice attacks (the "log-unit lattice") exploit exactly the embedding $ℓ$ of this theorem.
Arithmetic of elliptic curves and regulators. The analogous Mordell–Weil regulator in the Birch–Swinnerton-Dyer conjecture is modeled on the Dirichlet regulator; the conceptual template is identical.
Stark's conjectures. Refined leading-term formulas for Artin $L$ -functions predict that special $L$ -values are (logarithms of) explicit units (Stark units), generalizing the unit theorem's role in $ζ_{K}$ .

Historical Development

1770 — Lagrange & Euler. Complete theory of Pell's equation $x^{2} - d y^{2} = 1$ via continued fractions — implicitly the rank-1 unit theorem for real quadratic fields.
1801 — Gauss. Disquisitiones Arithmeticae develops binary quadratic forms; the "automorphs" of forms encode units of quadratic orders.
1846 — Dirichlet. Announces and proves the general unit theorem for rings of integers of number fields, famously reported to have conceived the key idea during a musical concert.
1871 — Dedekind. Introduces ideals and the modern notion of $O_{K}$ ; recasts unit theory in ring-theoretic terms in his supplements to Dirichlet's Vorlesungen über Zahlentheorie.
1896–1910 — Minkowski. Geometrie der Zahlen provides the lattice-point theorem giving the clean geometric proof of both unit and class-number finiteness used today.
1897 — Hilbert. Zahlbericht systematizes algebraic number theory; the regulator and unit lattice take their canonical place.
1920s–30s — Hasse, Chevalley. Idelic/adelic reformulations; the unit theorem becomes a statement about $S$ -units and the compactness of norm-one ideles ( $A_{K}^{1} / K^{\times}$ compact).
1966 — A. Baker. Effective linear forms in logarithms make computations of units and solutions of unit equations effective, with explicit bounds.
1970s–80s — Stark, Tate. Stark's conjectures conjecturally identify special $L$ -values with units, a deep refinement still driving research.

Subtleties & Limits

Effective and explicit unit computation. While the theorem guarantees a fundamental system exists, finding one is hard. The regulator

R_{K}

can be exponentially large in

∣ d_{K} ∣

; under GRH there are subexponential algorithms, but unconditional polynomial-time computation of fundamental units (equivalently the regulator) for real quadratic fields is open and closely tied to the difficulty exploited in some cryptographic proposals. Class number 1 problem and units. Disentangling the regulator from the class number in the class number formula is delicate; questions like infinitude of real quadratic fields with class number 1 (Gauss's conjecture) remain open and interact with unit sizes. Stark's conjecture. The precise statement that leading coefficients of Artin

L

-functions are determined by explicit "Stark units" is proven only in special cases (rank-one abelian over

Q

or imaginary quadratic via elliptic units). The general rank-

\geq 2

and non-abelian cases are wide open. Quantitative log-unit lattice. The geometry (shortest vectors, covering radius) of the log-unit lattice

ℓ (O_{K}^{\times}) \subset H

is poorly understood for general

K

and is exactly what governs the security of ideal-lattice cryptosystems; bounding its basis quality is an active, unresolved area.

Dirichlet's Unit Theorem

The Core Insight

Formal Statement

Number field, embeddings, signature

The logarithmic embedding

Dirichlet's Unit Theorem

Why It Holds / Derivation

Proof via the geometry of numbers

The logarithmic embedding sends units to a lattice in the trace-zero hyperplane. Shown: a real quadratic field, $r_1=2$, $r_2=0$, $r=1$.

Johann Peter Gustav Lejeune Dirichlet | 1805–1859

Building Intuition

Worked Examples

Real quadratic field: $K = \mathbb{Q}(\sqrt{2})$

Imaginary quadratic vs. cubic

Common Misconceptions

Why the "$-1$" cannot be dropped

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also

Before this

Where this leads