---
name: unit-distance-golod-shafarevich
type: reference
title: "The Golod-Shafarevich Inequality"
description: "The central mechanism for proving infinite unramified towers: relation rank, Frattini quotient, and the capacity constraint that shapes all constructions."
tags: [golod-shafarevich, class-field-towers, frattini-quotient, relation-rank, algebraic-number-theory]
timestamp: 2026-07-20
---

# The Golod-Shafarevich Inequality

## Overview

The Golod-Shafarevich (GS) inequality is the central mathematical mechanism in the unit distance proof. It determines whether an unramified class field tower is infinite, which is the fundamental existence result required to construct point sets with many unit distances.

The inequality relates three quantities:

$$r < \frac{d^2}{4}$$

where:
- $d$ = rank of the Frattini quotient (related to the class group of the base field)
- $r$ = relation rank (constraints imposed by prime splitting)

If this inequality is satisfied, the tower is infinite, and the construction can produce infinitely many unit-distance pairs.

## 1. Mathematical Background

### 1.1 Class Field Towers

Given a number field $F$, the **Hilbert class field** $H(F)$ is the maximal unramified abelian extension of $F$. The class field tower is the sequence:

$$F = F_0 \subset F_1 \subset F_2 \subset \cdots$$

where $F_{i+1}$ is the Hilbert class field of $F_i$. The tower is **finite** if $F_n = F_{n+1}$ for some $n$ (i.e., the class group eventually becomes trivial), and **infinite** otherwise.

### 1.2 The Pro-p Tower

For the unit distance construction, we are interested in the **pro-p tower** — the maximal unramified pro-$p$ extension tower, where $p$ is a prime (typically $p = 2$ or $p = 3$).

The pro-$p$ tower over $F$ is:

$$F = F_0 \subset F_1 \subset F_2 \subset \cdots$$

where $F_{i+1}$ is the maximal unramified pro-$p$ extension of $F_i$.

### 1.3 Why Infinite Towers Matter

An infinite tower provides:
1. **Infinitely many layers** of field extensions
2. **Increasingly rich sets** of algebraic numbers
3. **More unit-norm elements** via Lemma 2.2
4. **Larger point configurations** with unit distances

Without an infinite tower, the construction would be limited to finitely many layers, yielding only finitely many unit-distance pairs per point.

## 2. The Frattini Quotient

### 2.1 Definition

For a pro-$p$ group $G$, the **Frattini subgroup** $\Phi(G)$ is the intersection of all maximal open subgroups. The **Frattini quotient** is:

$$\overline{G} = G / \Phi(G)$$

For a pro-$p$ group, $\overline{G}$ is an elementary abelian $p$-group (a vector space over $\mathbb{F}_p$).

### 2.2 Rank and Dimension

The **rank** $d$ of $\overline{G}$ is its dimension as an $\mathbb{F}_p$-vector space:

$$d = \dim_{\mathbb{F}_p} \overline{G}$$

This is also called the **generator rank** of $G$, since $d$ is the minimum number of generators needed for $G$.

### 2.3 Relation to Class Groups

For the pro-$p$ tower over $F$:
- The Frattini quotient of $\text{Gal}(F_1/F)$ is isomorphic to the $p$-part of the class group of $F$ (modulo Frattini subgroup)
- The rank $d$ equals the $p$-rank of the class group of $F$
- For $p = 2$, this is the **2-class rank** of $F$

## 3. The Relation Rank

### 3.1 Definition

The **relation rank** $r$ is the dimension of the space of relations among the generators of $\overline{G}$. If $\overline{G}$ is generated by $d$ elements, then $r$ is the number of independent relations they satisfy.

### 3.2 Where Relations Come From

In the unit distance construction, relations come from **killing Frobenius elements** of split rational primes.

When a rational prime $q$ splits completely in $F$, it factors into $[F:\mathbb{Q}]$ prime ideals. In the pro-$p$ tower, each such prime ideal contributes a Frobenius element that must be killed (set to the identity) to ensure the prime splits completely in the tower.

### 3.3 Relation Count

For a base field $F$ with Galois group $\text{Gal}(F/\mathbb{Q})$:

- Each rational split prime $q$ contributes $|\text{Gal}(F/\mathbb{Q})| / p^{v_p(|\text{Gal}(F/\mathbb{Q})|)}$ relations
- For $p = 2$ and $\text{Gal}(F/\mathbb{Q}) \cong (\mathbb{Z}/2\mathbb{Z})^N$: each prime contributes 2 relations (one per conjugate pair under complex conjugation)
- For $t$ split primes: total relations from splitting = $2t$

### 3.4 Total Relation Rank

The total relation rank is:

$$r = r_{\text{class}} + r_{\text{split}}$$

where:
- $r_{\text{class}}$ = intrinsic relations from the class group structure
- $r_{\text{split}} = 2t$ = relations from killing Frobenius elements

The intrinsic relation rank is bounded by:

$$r_{\text{class}} \le d + r_1 + r_2 - 1$$

where $r_1$ = number of real places, $r_2$ = number of complex places of $F$.

## 4. The Inequality

### 4.1 Statement

The Golod-Shafarevich inequality states:

> If $G$ is a finite $p$-group with generator rank $d$ and relation rank $r$, then:
>
> $$r \ge \frac{d^2}{4}$$

**Contrapositive:** If $r < d^2/4$, then $G$ cannot be a finite $p$-group — it must be infinite.

### 4.2 Application to Class Field Towers

For the pro-$p$ tower over $F$:
1. Compute the generator rank $d$ (2-class rank of $F$)
2. Compute the relation rank $r$ (intrinsic + $2t$ from split primes)
3. Check whether $r < d^2/4$

If the inequality holds, the tower is infinite, and the construction proceeds.

### 4.3 The Capacity Constraint

The GS inequality creates a **capacity constraint** on the number of split primes:

$$r_{\text{class}} + 2t < \frac{d^2}{4}$$

$$2t < \frac{d^2}{4} - r_{\text{class}}$$

$$t < \frac{d^2}{8} - \frac{r_{\text{class}}}{2}$$

This bounds the maximum number of split primes that can be accommodated while keeping the tower infinite.

## 5. Capacity Analysis

### 5.1 Capacity Formula

The **GS capacity** is the maximum number of split primes:

$$t_{\max} = \left\lfloor \frac{d^2/4 - r_{\text{class}} - 1}{2} \right\rfloor$$

(The $-1$ ensures strict inequality.)

### 5.2 Capacity for Different Base Fields

| Base Field | $d$ | $r_{\text{class}}$ | $d^2/4$ | $t_{\max}$ |
|---|---|---|---|---|
| $\mathbb{Q}(\sqrt{-D})$ (imaginary quadratic) | $\ell - 1$ | $\ell - 1$ | $(\ell-1)^2/4$ | $\lfloor ((\ell-1)^2/4 - \ell)/2 \rfloor$ |
| $\mathbb{Q}(\sqrt{p_1}, \ldots, \sqrt{p_N})$ (totally real) | $2^N - 1$ | $2^N - 1$ | $(2^N-1)^2/4$ | $\lfloor ((2^N-1)^2/4 - 2^N)/2 \rfloor$ |
| $\mathbb{Q}(\sqrt{-2}, \sqrt{3}, \sqrt{5}, \sqrt{7})$ (H16) | 15 | 22 | 56.25 | $\lfloor (56.25 - 22 - 1)/2 \rfloor = 16.6 \to 16$ |

Wait — in H16, $t = 17$ is used with $r = 22$, giving $r + 2t = 56 < 56.25$. The capacity calculation is:

$$t_{\max} = \left\lfloor \frac{56.25 - 22 - \epsilon}{2} \right\rfloor = \left\lfloor \frac{34.25 - \epsilon}{2} \right\rfloor = 17$$

for sufficiently small $\epsilon > 0$.

### 5.3 The Quadratic Ceiling

For imaginary quadratic fields, the 2-class rank grows linearly: $d = \ell - 1$. The GS capacity scales as $d^2/4 \sim \ell^2/4$, but the discriminant penalty also grows as $\log H \sim \frac{1}{2} \ell \log \ell$.

The exponent $\delta$ scales roughly as:

$$\delta \sim \frac{t \log 2 - \log H}{4 \log Q} \sim \frac{\ell^2/8 - \ell \log \ell/2}{\ell^3} \sim \frac{1}{\ell}$$

This creates a ceiling near $\delta \approx 0.014$ for imaginary quadratic fields (as achieved by H15).

### 5.4 Breaking the Ceiling

H16 breaks the quadratic ceiling by using a multi-quadratic field where $d = 2^N - 1$ grows exponentially. The capacity scales as $d^2/4 \sim 2^{2N}/4$, while the discriminant penalty grows as $\log H \sim \frac{1}{2} \sum \log p_i$ (linear in $N$).

## 6. The GS Margin

### 6.1 Definition

The **GS margin** is the gap between the actual relation rank and the GS bound:

$$\text{Margin} = \frac{d^2}{4} - r_{\text{total}}$$

where $r_{\text{total}} = r_{\text{class}} + 2t$.

### 6.2 Margin Analysis

| Construction | $d^2/4$ | $r_{\text{total}}$ | Margin |
|---|---|---|---|
| H15 (imaginary quadratic) | 5625 | 5475 | 150 |
| H16 (multi-quadratic D16) | 56.25 | 56 | **0.25** |

H16 uses the construction at **maximum capacity** — the margin is only 0.25. This means:
- Adding one more split prime would violate the bound
- The construction is optimal for this base field
- Further improvement requires a larger base field (higher $d$)

### 6.3 Margin and Exponent

The GS margin directly affects the exponent. A larger margin allows more split primes, which increases the entropy numerator. However, more split primes also increase the denominator (via $\log Q$). The optimal balance is achieved when the margin is small but positive.

## 7. Worked Example: H16

### Step 1: Compute $d$

$$d = 2^4 - 1 = 15$$

### Step 2: Compute $r_{\text{class}}$

$$r_{\text{class}} = d + r_1 + r_2 - 1 = 15 + 0 + 8 - 1 = 22$$

### Step 3: Determine $t_{\max}$

$$t_{\max} = \left\lfloor \frac{d^2/4 - r_{\text{class}} - \epsilon}{2} \right\rfloor = \left\lfloor \frac{56.25 - 22 - \epsilon}{2} \right\rfloor = 17$$

### Step 4: Verify GS Inequality

$$r_{\text{total}} = r_{\text{class}} + 2t = 22 + 34 = 56$$

$$56 < 56.25 \quad \checkmark$$

### Step 5: Conclude

The tower is infinite. The construction is valid.

## 8. Historical Context

### 8.1 The Original Results

- **Golod (1964):** Proved that certain class field towers are infinite using a pigeonhole argument
- **Shafarevich (1964):** Independently proved the result and established the precise inequality
- **The inequality $r \ge d^2/4$** is sometimes called the **Golod-Shafarevich theorem** or the **Golod bound**

### 8.2 Improvements

The bound $r \ge d^2/4$ has been improved in various contexts:

- For $p$-groups: $r \ge d^2/4$ (original)
- For specific group structures: tighter bounds may apply
- For the unit distance problem: the original bound is sufficient

### 8.3 Open Questions

- Can the GS bound be improved to $r \ge d^2/4 + c$ for some constant $c > 0$?
- Are there constructions that achieve $r = d^2/4 - \epsilon$ for arbitrarily small $\epsilon$?
- How does the GS margin relate to the depth of the tower?

## 9. Connection to the Unit Distance Problem

### 9.1 The Chain of Logic

$$\text{GS inequality} \implies \text{infinite tower} \implies \text{infinitely many unit-norm elements} \implies u(n) = \Omega(n^{1+\delta})$$

### 9.2 Why GS is Necessary

Without the GS guarantee:
- The tower might be finite, yielding only finitely many layers
- The construction would produce only finitely many unit-distance pairs
- The lower bound would be $u(n) = O(n)$, not $u(n) = \Omega(n^{1+\delta})$

### 9.3 Why GS is Sufficient

If the GS inequality holds:
- The tower is provably infinite
- Each layer contributes new unit-norm elements
- The entropy accumulates across layers
- The exponent $\delta$ is positive

## References

- Golod, "On nilpotent groups of finite exponent" (1964)
- Shafarevich, "On p-extensions" (1964)
- Koch, *Galois theory of p-extensions* (1970)
- Washington, *Introduction to Cyclotomic Fields* (1982)
- Neukirch, *Algebraic Number Theory* (1999)
- OpenAI, `unit-distance-proof.pdf`, `unit-distance-remarks.pdf`