---
title: Primitive Vectors in Mathematics
url: https://www.emergentmind.com/topics/primitive-vectors
type: topic
---

# Primitive Vectors in Mathematics

Primitive vectors are vectors singled out by a context-dependent notion of irreducibility, maximality, or highest-weight behavior. In the integer lattice \(\mathbb{Z}^n\), a vector is primitive when its coordinates are coprime; in \(\mathbb{Z}^2\), sequences of primitive vectors support discrete formulas for rotation numbers; in arithmetic dynamics, primitive vectors on spheres and in \(S\)-arithmetic lattices are the basic objects in equidistribution and counting theorems; in the combinatorics of the hypercube, a primitive vector is a nonnegative integer normal vector whose prescribed numerical rectangles are all realizable; in multiple-recursive matrix methods, the paper does not isolate “primitive vector” as a separate term, but every nonzero state vector lies on a single maximal cycle when the transition matrix is primitive; and in supergroup representation theory, primitive vectors are highest-weight vectors for the action of a Borel or of the even subgroup [1211.2716] [1308.0877] [2009.14275] [1604.06753] [1608.08989].

## 1. Primitive lattice vectors in \(\mathbb{Z}^n\)

In the number-theoretic and lattice-theoretic sense, an integer vector \(v=(v_1,\dots,v_n)\in\mathbb{Z}^n\) is primitive if it cannot be written as a nontrivial integer multiple of another integer vector, equivalently if \(\gcd(v_1,\dots,v_n)=1\) [1211.2716]. Geometrically, such a vector is the first lattice point on its ray from the origin, and lattice-theoretically it is precisely a vector that can be extended to a \(\mathbb{Z}\)-basis of \(\mathbb{Z}^n\) [1211.2716]. The same notion is used in the matrix-completion literature: a row vector is primitive exactly when it is a \(1\times n\) primitive matrix, and such a vector can be completed to an \(n\times n\) unimodular matrix over \(\mathbb{Z}\) [2105.05383].

Two structural characterizations are emphasized in the matrix setting. First, a \(k\times n\) integer matrix is primitive if and only if for every prime \(p\) its reduction modulo \(p\) has full row rank; for a single row this reduces to the usual coprimality test [2105.05383]. Second, when \(k<n\), primitiveness is equivalent to the row Hermite normal form condition
\[
\operatorname{HNF}(A^T)=\begin{pmatrix}I_k\\0\end{pmatrix},
\]
so primitive submodules are exactly direct summands of \(\mathbb{Z}^n\) [2105.05383]. This identifies primitive vectors as the atomic one-dimensional cases of primitive matrices.

This lattice notion also governs asymptotic counting. For nonsingular integer \(n\times n\) matrices of determinant \(k\) and Euclidean norm at most \(T\), the subset with primitive rows has the asymptotic
\[
N'_{n,k}(T)=c'_{n,k}\,T^{n(n-1)}+O_\varepsilon\big(T^{n(n-1)-1/(2n)+\varepsilon}\big),
\]
while the density inside all determinant-\(k\) matrices is
\[
D_n(k)=\lim_{T\to\infty}\frac{N'_{n,k}(T)}{N_{n,k}(T)}=\frac{c'_{n,k}}{c_{n,k}},
\]
a multiplicative function of \(k\) [1211.2716]. For fixed \(n\ge 3\), \(D_n(k)\) lies strictly between \(1/\zeta(n-1)^n\) and \(1\), tends to \(1\) along rough sequences, and tends to \(1/\zeta(n-1)^n\) along totally divisible sequences [1211.2716]. This suggests that primitiveness is arithmetically local—encoded prime-by-prime—while remaining globally compatible with the geometry of large-norm counting.

## 2. Primitive vectors in planar lattice geometry and rotation theory

For \(v=(a,b)\in\mathbb{Z}^2\), primitiveness again means \(\gcd(a,b)=1\) [1308.0877]. A sequence of primitive vectors
\[
v_1,\dots,v_d\in\mathbb{Z}^2
\]
is assumed nondegenerate when
\[
\varepsilon_i:=\det(v_i,v_{i+1})\neq 0,\qquad i=1,\dots,d,
\]
with cyclic indexing \(v_{d+1}=v_1\) [1308.0877]. The associated closed polygonal line \(v_1\to\cdots\to v_d\to v_1\) has rotation number
\[
\mathrm{rot}(v_1,\dots,v_d)=\frac{1}{2\pi}\sum_{i=1}^d\int_{L_i}\frac{-y\,dx+x\,dy}{x^2+y^2},
\]
where \(L_i\) is the segment from \(v_i\) to \(v_{i+1}\) [1308.0877].

A special case is the unimodular case, where \(|\varepsilon_i|=1\) for all \(i\). Then Higashitani–Masuda’s formula expresses the rotation number purely combinatorially:
\[
\mathrm{rot}(v_1,\dots,v_d)=\frac{1}{12}\sum_{i=1}^d(3\varepsilon_i+a_i),
\]
with
\[
a_i=\varepsilon_{i-1}^{-1}\varepsilon_i^{-1}\det(v_{i+1},v_{i-1})
\]
and
\[
\varepsilon_{i-1}^{-1}v_{i-1}+\varepsilon_i^{-1}v_{i+1}+a_iv_i=0
\]
[1308.0877]. For general primitive sequences, the local failure of unimodularity is measured by integers \(x_i,y_i,l_i\) and Hirzebruch–Jung continued fractions of \(|\varepsilon_i|/x_i\), leading to the formula
\[
\mathrm{rot}(v_1,\dots,v_d)=\frac{1}{12}\sum_{i=1}^d\left(\Bigl(3(l_i+1)-\sum_{j=1}^{l_i}n_j^{(i)}\Bigr)\frac{\varepsilon_i}{|\varepsilon_i|}+a_i-\frac{x_i+y_i}{\varepsilon_i}\right)
\]
[1308.0877].

In this setting, primitiveness is the condition that each vector represents a primitive lattice direction, so that each cone \(\mathbb{R}_{\ge0}v_i+\mathbb{R}_{\ge0}v_{i+1}\) can be analyzed by lattice index and then resolved into unimodular cones by inserting additional primitive rays [1308.0877]. The paper makes explicit that the arithmetic of primitive vectors is the same arithmetic that governs the resolution of cyclic quotient singularities in toric geometry. A plausible implication is that primitiveness here is not merely a gcd condition: it is the discrete regularity assumption that makes a closed polygonal loop amenable to exact topological and toric formulas.

## 3. Arithmetic counting, spheres, and equidistribution

Primitive vectors are central in the study of lattice points on spheres and in \(S\)-arithmetic homogeneous dynamics. In \(\mathbb{Z}^d\), the primitive points on the sphere of squared radius \(D\) are
\[
S^{d-1}(D)=\{v\in\mathbb{Z}^d_{\mathrm{prim}}:\|v\|^2=D\},
\]
and each such \(v\) determines both a direction \(v/\sqrt D\in S^{d-1}\) and an orthogonal lattice
\[
\Lambda_v=\mathbb{Z}^d\cap v^\perp
\]
of covolume \(\sqrt D\) [1411.1272]. After normalization, the shape of \(\Lambda_v\) defines a point in
\[
X_{d-1}=SO_{d-1}(\mathbb{R})\backslash SL_{d-1}(\mathbb{R})/SL_{d-1}(\mathbb{Z}),
\]
and the paper proves joint equidistribution of directions and orthogonal grids in dimensions \(d>5\), and in dimensions \(4,5\) under the congruence restriction \(p\nmid D\) for a fixed odd prime \(p\) [1411.1272]. The limiting measure is the product \(m_{S^{d-1}}\otimes m_{Y_{d-1}}\) on the sphere and the space of orthogonal grids [1411.1272].

A finer invariant is the shortest solution \(w_v\) to the gcd equation
\[
a_1x_1+\cdots+a_nx_n=1
\]
for a primitive vector \(v=(a_1,\dots,a_n)\in\mathbb{Z}^n\), together with the covering radius \(\rho_v\) of the orthogonal lattice \(\Lambda_v\) [1903.01560]. The normalized lengths \(\|w_v\|/\rho_v\) equidistribute in \([0,1]\) with respect to a measure \(\nu_n\); this measure is Lebesgue only when \(n=2\), and non-Lebesgue otherwise [1903.01560]. By contrast, for \(n\ge3\) the naive normalization \(\|w_v\|/\|v\|\) collapses to zero along a full density set of primitive vectors, so no nontrivial equidistribution exists in that scale [1903.01560]. This shows that the geometry of the orthogonal lattice, rather than the norm of \(v\) itself, is the correct normalizing datum for the gcd equation in higher dimension.

The \(p\)-adic analogue considers primitive vectors in \(\mathbb{Z}^2\) as points of the \(p\)-adic unit sphere
\[
\mathbb{S}^1_p=\{v\in\mathbb{Q}_p^2:\|v\|_p=1\},
\qquad
\|(a,b)\|_p=\max\{|a|_p,|b|_p\},
\]
and proves joint equidistribution of
\[
\left(\frac{v}{\|v\|},v\right)\in \mathbb{S}^1\times \mathbb{S}^1_p
\]
with respect to \(\mathrm{Leb}\times \mu_p^2|_{\mathbb{S}^1_p}\) as \(\|v\|\to\infty\), with rate \(O(R^{-2\tau_p+\varepsilon})\) and \(\tau_p=\frac{1}{28}\) [2103.10889]. Here the real norm orders the primitive vectors, while the \(p\)-adic component records a local direction. The limiting product measure suggests asymptotic independence between the real and \(p\)-adic directional data.

At the level of \(S\)-arithmetic lattices, primitive vectors in \(\mathbb{Z}_S^d\) are defined by
\[
P(\mathbb{Z}_S^d)=\mathrm{SL}_d(\mathbb{Z}_S)\cdot \mathbf e_1,
\]
equivalently by the condition \(S\gcd(\mathbf v)=1\) [2310.03459]. For the primitive \(S\)-Siegel transform
\[
\widehat f(g\Gamma_d)=\sum_{\mathbf v\in P(\mathbb{Z}_S^d)} f(g\mathbf v),
\]
the mean value formula is
\[
\int_{G_d/\Gamma_d}\widehat f(g\Gamma_d)\,d\mu_d(g)
=\frac{1}{\zeta_S(d)}\int_{\mathbb{Q}_S^d}f(\mathbf x)\,d\mathbf x,
\]
and for \(d\ge3\) the primitive second moment has the Rogers-type form
\[
\frac{1}{\zeta_S(d)^2}\iint F(\mathbf x,\mathbf y)\,d\mathbf x\,d\mathbf y
+\frac{1}{\zeta_S(d)}\sum_{k\in\mathbb{Z}_S^\times}\int F(\mathbf x,k\mathbf x)\,d\mathbf x
\]
[2310.03459]. These formulas yield quantitative counting of primitive \(S\)-arithmetic lattice points, Schmidt-type asymptotics for primitive integer vectors with congruence conditions, quantitative Khintchine–Groshev theorems over primitive sets, and an \(S\)-arithmetic logarithm law for unipotent flows [2310.03459]. In all of these results, primitiveness is the condition that removes scalar redundancy and exposes the genuinely geometric distribution.

## 4. Primitive vectors as rows of primitive matrices

A \(k\times n\) integer matrix \(A\) is primitive if
\[
x=yA\in\mathbb{Z}^n,\ y\in\mathbb{Q}^k \implies y\in\mathbb{Z}^k,
\]
so its row span is a direct summand of \(\mathbb{Z}^n\) [2105.05383]. When \(k=1\), this recovers the primitive-vector condition. The paper emphasizes that a square primitive matrix is unimodular and that a \(k\times n\) primitive matrix can always be extended to an \(n\times n\) unimodular matrix over \(\mathbb{Z}\) [2105.05383].

The probabilistic extension problem starts with a fixed primitive matrix \(A\in\mathbb{Z}^{k\times n}\), \(\|A\|\le \lambda\), and appends \(n-k-s-1\) random rows with entries chosen independently and uniformly from \(\{0,1,\dots,\lambda-1\}\) [2105.05383]. If \(B\) is the resulting \((n-s-1)\times n\) matrix, then
\[
\Pr(B\text{ primitive})
\ge
1-\frac{4\zeta(2)}{(\lambda-1)\lambda^{s+1}}
-\frac{1}{(\lambda-1)\lambda^{s+1}\sum_{i=1}^{n-k-s-1}\frac{1}{(n-s-i+1)^2}},
\]
and a simpler lower bound is
\[
\Pr(B\text{ primitive})
\ge
1-\frac{4\zeta(2)}{\lambda^{s+1}}-\frac{2(n-s)^2}{(\lambda-1)\lambda^{s+1}}
\]
[2105.05383]. For \(s=3\) and \(\lambda\ge 3(n-3)^2/5\), this lower bound is at least \(0.2\) [2105.05383]. Specializing to \(k=1\), this gives a quantitative statement on how often a fixed primitive vector can be extended by random rows to a larger primitive family.

This probabilistic fact is converted into an algorithmic completion theorem. The paper proves that there exists a fast Las Vegas algorithm that completes a \(k\times n\) primitive matrix to an \(n\times n\) unimodular matrix within expected \(\tilde O(n^\omega\log\|A\|)\) bit operations, where \(\omega\) is the exponent of matrix multiplication [2105.05383]. For primitive vectors, this means that extension to a unimodular basis is not only possible in principle but can be performed with nearly optimal linear-algebraic complexity.

Placed next to the asymptotic counting theorem for determinant-\(k\) matrices with primitive rows [1211.2716], this suggests a coherent picture: the primitive-vector condition is simultaneously a local congruence condition, a direct-summand condition, and an algorithmically stable property under random completion.

## 5. Combinatorial and pseudorandom meanings

In the oriented-matroid study of the real affine cube over
\[
C^n=\{-1,1\}^n,
\]
a primitive vector is a nonnegative integer vector \(\mathbf h\in\mathbb{N}_0^n\) such that all of its 3- and 4-numerical rectangles are realizable [2009.14275]. For
\[
|\mathbf h|=\sum_{i=1}^n h_i,
\]
the \(a\)-level is
\[
S_a(\mathbf h)=\{\mathbf v\in C^n:\mathbf h\cdot \mathbf v=|\mathbf h|-2a\},
\]
and an \(\mathbf h\)-rectangle is a quadruple \((a\le b\le c\le d)\) with \(d=b+c-a\) [2009.14275]. Realizability means that there is an actual signed geometric rectangle in the affine cube with vertices lying in the prescribed levels. The paper proves that for a primitive vector \(\mathbf h\), every level \(S_a\) is nonempty, every non-extremal level contains at least two elements, and if \(0\le g\le |\mathbf h|/2+1\), then
\[
\mathbf g=(\mathbf h,g)\in\mathbb{N}_0^{n+1}
\]
is again primitive [2009.14275]. Primitive vectors are then used to define primitive hyperplanes whose cocircuits are forced in any oriented cube; this yields a proof that for \(n\le 7\), the real affine cube is uniquely determined by its signed rectangles and its signed cocircuits complementary of its facets and skew-facets [2009.14275]. In this context, primitiveness is neither gcd-based nor dynamical: it is a complete rectangular realizability condition across level sets.

A different usage occurs in pseudorandom generation by the multiple-recursive matrix method (MRMM). Here one works over \(\mathbb{F}_{q^m}\) and considers the recurrence
\[
S_{i+n}=C_0S_i+C_1S_{i+1}+\cdots+C_{n-1}S_{i+n-1},
\]
with transition matrix \(T\in M_{mn}(\mathbb{F}_q)\) [1604.06753]. An MRMM is primitive if every nonzero initial state has period exactly
\[
q^{mn}-1,
\]
equivalently if \(T\) has multiplicative order \(q^{mn}-1\), equivalently if the characteristic polynomial \(\det(M(X))\) is primitive of degree \(mn\) over \(\mathbb{F}_q\) [1604.06753]. The paper does not introduce “primitive vector” as a separate term, but it states that the nonzero vectors in the state space \(\mathbb{F}_q^{mn}\) form a single orbit under \(T\), so every nonzero state vector is primitive in the sense of lying on the unique long cycle of a Singer cycle [1604.06753]. This suggests a distinct, orbit-theoretic meaning of primitive vector: not indivisible in a lattice, but dynamically generating the full nonzero state space under iteration.

## 6. Primitive vectors in supergroup representation theory

For the general linear supergroup
\[
G=GL(m|n)
\]
over an algebraically closed field, with even subgroup
\[
G_{ev}\cong GL(m)\times GL(n),
\]
primitive vectors are highest-weight vectors for a Borel action [1309.3801]. A vector \(v\) in a rational \(G\)-supermodule is primitive if the line \(Kv\) is stabilized by the Borel subsupergroup \(B\); a vector is \(G_{ev}\)-primitive if \(Kv\) is stabilized by the even Borel \(B_{ev}\), equivalently if it is a weight vector annihilated by the positive even root spaces [1309.3801]. The relevant induced supermodule is
\[
H_G^0(\lambda)=\operatorname{ind}_B^G K_\lambda,
\]
and as a \(G_{ev}\)-module it decomposes as
\[
H_G^0(\lambda)\cong H_{G_{ev}}^0(\lambda)\otimes \Lambda(Y),
\qquad
Y=V_m^*\otimes V_n
\]
[1608.08989].

The first floor
\[
F_1=H_{G_{ev}}^0(\lambda)\otimes Y
\]
admits explicit \(G_{ev}\)-primitive vectors \(T_{ij}\) of weight
\[
\lambda_{ij}=\lambda-\varepsilon_i+\delta_j
\]
under the usual dominance conditions on \(\lambda\) [1309.3801]. In characteristic \(0\), these \(T_{ij}\) form a complete set of primitive vectors in \(F_1\), and \(F_1\) is a direct sum of induced \(G_{ev}\)-modules with highest weights \(\lambda_{ij}\) [1309.3801]. For higher floors
\[
F_k=H_{G_{ev}}^0(\lambda)\otimes \Lambda^kY,
\]
the paper constructs vectors \(T_{I|J}\) indexed by admissible multiindices \((I|J)\), and under suitable robustness conditions these form bases of the spaces of \(G_{ev}\)-primitive vectors of given weight [1309.3801].

The later paper refines this classification in the polynomial case. It constructs explicit even-primitive vectors in the largest polynomial subsupermodule \(\nabla(\lambda)\subset H_G^0(\lambda)\) and in the corresponding costandard supermodule for the Schur superalgebra \(S(m|n)\) [1608.08989]. The basis is indexed by marked tableaux, and for each marked tableau \(T^+\) one obtains an even-primitive vector
\[
v(T^+)=V_{I|J}p(TT^+),
\]
a linear combination of the basic \(T_{K|L}\) with prescribed content [1608.08989]. This yields a basis of all \(G_{ev}\)-primitive vectors of a given polynomial weight and realizes Littlewood–Richardson multiplicities as dimensions of explicit highest-weight spaces [1608.08989]. In this representation-theoretic context, primitive vectors are not indivisible lattice points but generators of highest-weight submodules.

Across these settings, “primitive vector” does not denote a single invariant notion. It denotes, depending on the ambient structure, a coprime lattice point, a primitive lattice direction in a polygonal fan, a generic state on a maximal cycle, a hyperplane normal with full rectangle realizability, or a highest-weight vector. What persists is the role of primitiveness as a condition that removes redundancy and exposes the irreducible geometric, arithmetic, combinatorial, or representation-theoretic content of the vector under study.

Source: https://www.emergentmind.com/topics/primitive-vectors