---
title: Minkowski Linear Independence
url: https://www.emergentmind.com/topics/minkowski-linear-independence
type: topic
---

# Minkowski Linear Independence

Minkowski linear independence usually denotes the classical assertion that, in a multiquadratic extension generated by independent square-free radicands, the natural radical monomials form a $\mathbb{Q}$-basis. In the standard form, if $d_1,\dots,d_r\in\mathbb{Z}$ are nonzero, square-free, and independent in $\mathbb{Q}^\times/(\mathbb{Q}^\times)^2$, then the $2^r$ elements
\[
\left\{\prod_{j\in I}\sqrt{d_j}: I\subset\{1,\dots,r\}\right\}
\]
are $\mathbb{Q}$-linearly independent; in particular,
\[
[\mathbb{Q}(\sqrt{d_1},\dots,\sqrt{d_r}):\mathbb{Q}]=2^r,
\]
and the displayed set is a $\mathbb{Q}$-basis [2006.16674]. In contemporary work, the same phenomenon is extended to mixed radical towers under explicit reducedness constraints, while related geometry-of-numbers criteria use Minkowski’s convex body theorem to deduce lower bounds for $\mathbb{Q}$-rank from the existence of small linear forms.

## 1. Classical multiquadratic independence

The classical theorem concerns multiquadratic fields. Let $d_1,\dots,d_r$ be distinct square-free integers and write
\[
K=\mathbb{Q}(\sqrt{d_1},\dots,\sqrt{d_r}).
\]
A standard form of Minkowski’s theorem asserts that the “radical basis” is linearly independent over $\mathbb{Q}$. The relevant independence hypothesis is that the $d_j$ are independent in $\mathbb{Q}^\times/(\mathbb{Q}^\times)^2$; equivalently, no nontrivial product $\prod d_j^{e_j}$ is a square in $\mathbb{Q}^\times$ with $e_j\in\{0,1\}$ [2006.16674].

A common special case is the pairwise independence of square roots: if $n_1,\dots,n_k$ are distinct square-free integers and $q_i\in\mathbb{Q}$, then
\[
\sum_{i=1}^k q_i\sqrt{n_i}=0 \;\Rightarrow\; q_1=\cdots=q_k=0.
\]
This is the form most often used in elementary manipulations of radicals. It is also the simplest manifestation of the basis theorem for multiquadratic extensions.

The significance of the theorem is twofold. First, it identifies an explicit basis for the field. Second, it computes the degree exactly: the extension has the maximal possible degree $2^r$ once the parity classes of the radicands are independent. This makes the theorem a foundational statement about basis structure in radical extensions.

## 2. Character-theoretic proof and the classical mechanism

The classical proof uses the full Galois group of the multiquadratic field. For $\varepsilon=(\varepsilon_1,\dots,\varepsilon_r)\in\{0,1\}^r$, define the automorphism $\sigma_\varepsilon\in\mathrm{Gal}(K/\mathbb{Q})$ by
\[
\sigma_\varepsilon(\sqrt{d_j}) = (-1)^{\varepsilon_j}\sqrt{d_j}.
\]
If one assumes a relation
\[
\sum_I c_I \prod_{j\in I}\sqrt{d_j}=0,
\]
then applying all $\sigma_\varepsilon$ produces a $2^r\times 2^r$ linear system with matrix
\[
M_{\varepsilon,I}=(-1)^{\sum_{j\in I}\varepsilon_j}.
\]
The columns are orthogonal characters of the group $(\mathbb{Z}/2\mathbb{Z})^r$, hence independent; therefore all $c_I=0$ [2006.16674].

This proof is notable for its symmetry. Modern expositions emphasize that embeddings of multiquadratic fields into $\mathbb{R}$ and the orthogonality of sign characters separate the coefficients immediately. In that sense, classical Minkowski linear independence is a Galois-theoretic statement in which the basis vectors are indexed by subsets of $\{1,\dots,r\}$ and the coefficient-extraction mechanism is character orthogonality.

The same source also frames this as the quadratic prototype of a wider pattern. A plausible implication is that the theorem is best understood not as an isolated identity about square roots, but as the $n=2$ case of a general radical-basis phenomenon.

## 3. Reduced radicals and the mixed-exponent generalization

A generalization beyond square roots is obtained by imposing explicit “reduced” constraints and working with positive rational radicands to keep real branches consistent. For $b\in\mathbb{Q}^+$ and $m\ge 2$, $\sqrt[m]{b}$ is called reduced if it cannot be written as $e\sqrt[n]{d}$ with $n<m$ and $e,d\in\mathbb{Q}^+$. Equivalently, the minimal polynomial of $\sqrt[m]{b}$ over $\mathbb{Q}$ is $X^m-b$. A finite set
\[
S=\{\alpha_i=\sqrt[m_i]{b_i}:1\le i\le k\}
\]
is reduced if each $\alpha_i$ is reduced and there is no nontrivial tuple of exponents $(E_1,\dots,E_k)$ with $0\le E_i\le m_i-1$, not all zero, such that $\prod \alpha_i^{E_i}\in\mathbb{Q}$ [2006.16674].

Under these constraints, distinct radical monomials are $\mathbb{Q}$-linearly independent. More precisely, if
\[
M_j=\prod_{i=1}^k \alpha_i^{e_{i,j}},\qquad 0\le e_{i,j}\le m_i-1,
\]
are distinct monomials, then
\[
\sum_{j=1}^N q_j M_j=0 \text{ with } q_j\in\mathbb{Q}
\quad\Rightarrow\quad
q_1=\cdots=q_N=0.
\]
Equivalently, the monomials
\[
\left\{\prod \alpha_i^{e_i}:0\le e_i\le m_i-1\right\}
\]
are $\mathbb{Q}$-linearly independent, and
\[
[\mathbb{Q}(\alpha_1,\dots,\alpha_k):\mathbb{Q}]=\prod_{i=1}^k m_i.
\]

The same paper states a one-radical-at-a-time version. Fix an index $\ell$ and set $K_\ell=\mathbb{Q}(S\setminus\{\alpha_\ell\})$. If $t<m_\ell$ and $y_0,\dots,y_t\in K_\ell$ are not all zero, then
\[
y_0+y_1\alpha_\ell+\cdots+y_t\alpha_\ell^t
\]
is irrational, in particular nonzero. Consequently, $\alpha_\ell$ has minimal polynomial $X^{m_\ell}-b_\ell$ over $K_\ell$, and adjoining $\alpha_\ell$ multiplies the field degree by $m_\ell$. This theorem makes the induction underlying the degree formula completely explicit.

The reduced-set condition is described as the natural generalization of “square-free radicands” and independence of their parity vectors in the quadratic case. That formulation isolates the precise obstruction: dependence can only occur when a bounded product of radicals collapses to a rational number.

## 4. Elementary proof via minimal polynomials, gcd, and trace

The alternative proof avoids explicit use of Minkowski’s sign-character orthogonality and instead relies on three elementary ingredients: minimal polynomials of reduced irrationals, a gcd argument between two polynomials, and a trace-from-$K$-to-$\mathbb{Q}$ argument in a carefully chosen basis [2006.16674].

The first step is minimal-polynomial control. If $\alpha=\sqrt[m]{b}$ is reduced, then its minimal polynomial over $\mathbb{Q}$ is $X^m-b$. More generally, any finite product of reduced radicals
\[
c=\prod \sqrt[m_i]{b_i}^{\,e_i}
\]
has a minimal polynomial of the form $X^s-k$ for some $s,k$, determined by the least positive $s$ with $c^s\in\mathbb{Q}$. A lemma in the same paper states: if $\alpha=\sqrt[m]{b}$ is reduced, $t<m$, and $l_0,\dots,l_t\in\mathbb{Q}$ are not all zero, then
\[
l_0+l_1\alpha+\cdots+l_t\alpha^t
\]
is irrational. The proof uses
\[
u(X)=q l_t X^t+\cdots+q l_0-p,\qquad v(X)=X^m-b\in\mathbb{Q}[X],
\]
and studies $\gcd(u,v)$.

The second step transfers this gcd argument into a tower. For a reduced set $S$ and fixed $\ell$, let
\[
f(X)=y_tX^t+\cdots+y_0\in K_\ell[X],\qquad t<m_\ell,
\]
and suppose $f(\alpha_\ell)=0$. Let
\[
g(X)=X^{m_\ell}-b_\ell,\qquad h(X)=\gcd(f,g)\in K_\ell[X].
\]
Since all zeros of $g$ have the form $\omega\alpha_\ell$ with $\omega$ an $m_\ell$-th root of unity, the product of zeros of $h$ gives
\[
|h_0|=b_\ell^{\deg(h)/m_\ell}\in K_\ell.
\]

The third step is the trace obstruction. Let $B$ be a $\mathbb{Q}$-basis of $K_\ell$, and pick $B^\ast\in B$ with
\[
\mathrm{Tr}_{K_\ell/\mathbb{Q}}\!\left(b_\ell^{\deg(h)/m_\ell}B^\ast\right)\neq 0.
\]
The minimal polynomial over $\mathbb{Q}$ of $b_\ell^{\deg(h)/m_\ell}B^\ast$ must then be linear, forcing
\[
b_\ell^{\deg(h)/m_\ell}B^\ast\in\mathbb{Q}.
\]
Because $B^\ast\in K_\ell$ and $\deg(h)<m_\ell$, this contradicts the reduced-set property. Therefore $f(\alpha_\ell)\neq 0$.

In multiquadratic fields, this replaces character orthogonality with a structural obstruction: if a lower-degree polynomial relation existed, it would force a forbidden rational power of a radicand, or of a product of radicands, to lie in the smaller field. For higher roots, the same source states that the approach parallels Kummer theory: multiplicative independence modulo $n$-th powers ensures distinct conjugates and degree multiplication, and the reduced-set proof captures this without invoking $\zeta_n$.

## 5. Degree formulas, examples, counterexamples, and arithmetic uses

The degree and basis structure are explicit in both the multiquadratic and mixed-radical settings. If $d_1,\dots,d_r$ are square-free and independent in $\mathbb{Q}^\times/(\mathbb{Q}^\times)^2$, then
\[
[\mathbb{Q}(\sqrt{d_1},\dots,\sqrt{d_r}):\mathbb{Q}]=2^r,
\]
with $\mathbb{Q}$-basis
\[
\left\{\prod_{j\in I}\sqrt{d_j}: I\subset\{1,\dots,r\}\right\}.
\]
If
\[
S=\{\alpha_i=\sqrt[m_i]{b_i}\}
\]
is reduced, then inductively
\[
[\mathbb{Q}(\alpha_1,\dots,\alpha_k):\mathbb{Q}]=\prod_{i=1}^k m_i,
\]
and the monomials $\prod \alpha_i^{e_i}$ with $0\le e_i\le m_i-1$ form a $\mathbb{Q}$-basis [2006.16674].

The examples are concrete. In the quadratic case, $\{\sqrt{2},\sqrt{3}\}$ are $\mathbb{Q}$-linearly independent; so are $\{\sqrt{2},\sqrt{3},\sqrt{6}\}$. For $d_1=2$, $d_2=3$,
\[
[\mathbb{Q}(\sqrt{2},\sqrt{3}):\mathbb{Q}]=4,
\]
with $\mathbb{Q}$-basis $\{1,\sqrt{2},\sqrt{3},\sqrt{6}\}$. In the cubic case, $\{2^{1/3},3^{1/3}\}$ are $\mathbb{Q}$-linearly independent. More generally, if $a_1,\dots,a_m$ are cube-free and multiplicatively independent modulo cubes, the $3^m$ monomials in $a_i^{1/3}$ are $\mathbb{Q}$-independent.

The counterexamples identify exactly how the hypotheses can fail. $\sqrt{8}=2\sqrt{2}$ shows dependence when radicands are not square-free. $16^{1/3}=2\cdot 2^{1/3}$ shows dependence when radicands are not $3$rd-power-free. If $a_2=a_1\cdot c^n$ with $c\in\mathbb{Q}^\times$, then $a_2^{1/n}\in\mathbb{Q}(a_1^{1/n})$, and independence of the two radicals fails.

The same framework underlies several arithmetic applications. Independence guarantees that simplified radical expressions are unique: no hidden cancellations occur unless they are forced by trivial algebraic identities such as extracting $n$th powers. The explicit basis descriptions are crucial for computations of norms, traces, and integral bases in radical extensions. The criteria also underpin algorithms that determine when sums of radicals can vanish or represent rational numbers, preventing spurious simplifications and enabling certified equality testing.

## 6. Geometry-of-numbers criteria and related meanings of “Minkowski linear independence”

In a different but related line of work, Minkowski’s convex body theorem is used to deduce lower bounds on rational rank from the existence of small integer-coefficient linear forms. Fischler’s vector-valued generalization of Nesterenko’s criterion states: let $1\le k\le p-1$, let $e_1,\dots,e_k\in\mathbb{R}^p$, let $\tau_1,\dots,\tau_k>0$ be pairwise distinct, and let $(Q_n)$ satisfy $Q_{n+1}=Q_n^{1+o(1)}$. If there exist integer-coefficient linear forms
\[
L_n=\ell_{1,n}X_1+\cdots+\ell_{p,n}X_p
\]
with
\[
|L_n(e_j)|=Q_n^{-\tau_j+o(1)},\qquad \max_i |\ell_{i,n}|\le Q_n^{1+o(1)},
\]
then any subspace $F$ of $\mathbb{R}^p$ defined over $\mathbb{Q}$ which contains $e_1,\dots,e_k$ satisfies
\[
\dim F\ge k+\tau_1+\cdots+\tau_k.
\]
The proof is based on geometry of numbers, namely Minkowski’s theorem on convex bodies [1202.2279].

The geometric mechanism is explicit. One constructs a convex, compact, symmetric set $C(\varepsilon,Q)$ around the directions $e_j$ and their orthogonal complement, proves that $C(\varepsilon,Q)\cap\mathbb{Z}^p=\{0\}$, and then applies Minkowski’s theorem to the restriction of this body to any $\mathbb{Q}$-defined subspace $F$. Volume–determinant comparison forces the rank inequality. In this usage, “Minkowski linear independence” refers not to radicals but to a geometry-of-numbers engine for proving $\mathbb{Q}$-linear independence.

A further development gives almost-everywhere converses. For $m>n>0$, a weighted analogue of the Khintchine–Groshev theorem in the absolute-value setting yields the exact criterion
\[
\mathcal{L}^{mn}(\mathcal{W}_0(m,n;\Psi))=
\begin{cases}
0,& \sum_{r=1}^{\infty}\left(\prod_{j=1}^n \psi_j(r)\right) r^{m-n-1}<\infty,\\[4pt]
1,& \sum_{r=1}^{\infty}\left(\prod_{j=1}^n \psi_j(r)\right) r^{m-n-1}=\infty,
\end{cases}
\]
with monotonicity needed only in the special case $(m,n)=(2,1)$. This metric theorem is then combined with Minkowski’s successive minima to show that Siegel/Nesterenko-type lower bounds are optimal almost everywhere: generically, the bound
\[
\dim F\ge n+T_1+\cdots+T_n
\]
cannot be improved, except up to logarithmic factors [1302.1952].

There is also a terminological ambiguity outside geometry of numbers and radical extensions. In the result that “norms as a function of $p$ are linearly independent in finite dimensions,” “Minkowski” refers to the $L^p$ norm structure. The theorem states that if
\[
\sum_{k=1}^n c_k\|v_k\|_p=0
\]
for all $p$ in a real interval with nonempty interior, then the dependence is trivial: there are no cancellations across distinct equivalence classes of vectors, where the equivalence relation is generated by adding zero coordinates, permuting coordinates, negating coordinates, and rescaling by a nonzero scalar. That result is proved by complex analytic continuation and is explicitly independent of Minkowski’s theorems in the geometry of numbers [1102.5026].

A common misconception is therefore terminological. “Minkowski linear independence” may refer to the radical-basis theorem in multiquadratic and radical extensions, or to geometry-of-numbers criteria derived from Minkowski’s convex body theorem, while in the $p$-norm literature the name “Minkowski” has a different origin. The three settings share the language of linear independence, but they concern different objects: radical monomials, small linear forms and rational rank, and norm profiles as functions of $p$.

Source: https://www.emergentmind.com/topics/minkowski-linear-independence