General linear group

In mathematics, the general linear group of degree n is the group of n × n invertible matrices with entries in a given field K (or more generally, a commutative ring with identity), under the operation of matrix multiplication. This group, denoted $\mathrm{GL}_n(K)$ or $\mathrm{GL}(n, K)$, consists of all such matrices with nonzero determinant and forms a basic example of a Lie group when K is the real or complex numbers. It plays a central role in linear algebra, representation theory, and geometry.^[1]

Definition and basic properties

Vector space formulation

The general linear group of a vector space $V$, denoted $\mathrm{GL}(V)$, consists of all invertible linear transformations from $V$ to itself, which are also known as the automorphisms of $V$. These transformations preserve the vector space structure, mapping $V$ bijectively onto itself while maintaining addition and scalar multiplication. The group operation is defined by composition of these linear maps, where the composition of two invertible transformations is again invertible and linear.^[2] For a finite-dimensional vector space $V$ of dimension $n \geq 1$, $\mathrm{GL}(V)$ is non-abelian when $n \geq 2$, meaning that the composition of transformations does not generally commute. The identity element of the group is the identity map on $V$, which leaves every vector unchanged, and every element $T \in \mathrm{GL}(V)$ has an inverse $T^{-1}$ that is also a linear automorphism satisfying $T \circ T^{-1} = T^{-1} \circ T =$ identity.^[2][3] When $V$ is a vector space of dimension $n$ over a field $K$, the group is commonly denoted $\mathrm{GL}(n, K)$. This notation emphasizes the dependence on both the dimension and the underlying field, and it abstracts the structure away from specific bases. In this setting, matrix representations can be used to coordinatize elements of $\mathrm{GL}(n, K)$ once a basis is chosen, though the group itself is defined independently of coordinates.^[2] To verify that $\mathrm{GL}(V)$ forms a group, note that it is closed under composition: if $T_1, T_2 \in \mathrm{GL}(V)$, then $T_1 \circ T_2$ is invertible with inverse $T_2^{-1} \circ T_1^{-1}$. Associativity follows directly from the associativity of function composition on the set of all maps from $V$ to $V$. The identity map serves as the neutral element, and inverses exist by definition for each element.^[2] As an algebraic variety over an algebraically closed field, $\mathrm{GL}(n, K)$ has dimension $n^2$, reflecting its embedding as an open subset of the space of all $n \times n$ matrices.^[4]

Matrix representation and invertibility

The general linear group $ \mathrm{GL}(V) $ of a finite-dimensional vector space $ V $ of dimension $ n $ over a field $ K $ is isomorphic to the matrix group $ \mathrm{GL}(n, K) $, consisting of all $ n \times n $ invertible matrices with entries in $ K $. This isomorphism arises from choosing a basis for $ V $, which allows linear endomorphisms of $ V $ to be represented by matrices in $ M_n(K) $, the ring of all $ n \times n $ matrices over $ K $; the invertible ones correspond precisely to automorphisms of $ V $.^[5][6] The group operation in $ \mathrm{GL}(n, K) $ is matrix multiplication, which mirrors the composition of linear maps: for matrices $ A, B \in \mathrm{GL}(n, K) $, the product $ AB $ represents the composition of the corresponding automorphisms and is also invertible.^[7][8] A matrix $ A \in M_n(K) $ belongs to $ \mathrm{GL}(n, K) $ if and only if it admits a two-sided multiplicative inverse $ A^{-1} \in M_n(K) $ such that $ A A^{-1} = A^{-1} A = I_n $, the identity matrix; this condition is equivalent to the linear map represented by $ A $ being bijective.^[7][9] Matrices without inverses, such as the zero matrix or any singular matrix with linearly dependent rows or columns, are excluded from $ \mathrm{GL}(n, K) $.^[8] For $ n=1 $, $ \mathrm{GL}(1, K) $ is isomorphic to the multiplicative group $ K^\times $ of nonzero elements in $ K $, since 1×1 invertible matrices are simply the nonzero scalars. Over the real numbers $ \mathbb{R} $, the group $ \mathrm{GL}(2, \mathbb{R}) $ includes matrices representing rotations (like those in the special orthogonal subgroup) combined with scalings, such as

$$\begin{pmatrix} a & -b \\ b & a \end{pmatrix}$$

for $ a^2 + b^2 \neq 0 $, which scale by $ \sqrt{a^2 + b^2} $ and rotate by $ \tan^{-1}(b/a) $.^[10][9]

Determinant characterization

The general linear group $\mathrm{GL}(n, K)$ consists of all $n \times n$ matrices $A$ with entries in the field $K$ such that $\det(A) \neq 0$. The non-vanishing of the determinant is the precise condition for invertibility over $K$, as a matrix is invertible if and only if its determinant is non-zero. This characterization provides an explicit algebraic test for membership in the group and is independent of the choice of basis.^[1]

Generalizations over rings and fields

Over finite fields

The general linear group over a finite field $\mathbb{F}_q$ with $q$ elements, denoted $GL(n, q)$, consists of all invertible $n \times n$ matrices with entries in $\mathbb{F}_q$. The order of this group is given by the formula

$$|GL(n, q)| = \prod_{k=0}^{n-1} (q^n - q^k).$$

This product arises from counting the number of ordered bases for the vector space $\mathbb{F}_q^n$, where the first basis vector can be any of the $q^n - 1$ nonzero vectors, the second any vector not in the span of the first (totaling $q^n - q$ choices), and so on for subsequent vectors.^[11] This combinatorial interpretation aligns with the matrix perspective: each invertible matrix corresponds to a linear transformation that maps the standard basis to another ordered basis of $\mathbb{F}_q^n$. For small cases, explicit computations yield $|GL(2, 2)| = 6$, and this group is isomorphic to the symmetric group $S_3$ via its action permuting the three nonzero vectors in $\mathbb{F}_2^2$. Similarly, $|GL(2, 3)| = 48$.^[11][12][13] The group $GL(n, q)$ acts on $\mathbb{F}_q^n$ by matrix multiplication, and this action is transitive on the set of nonzero vectors: for any two nonzero vectors $u, v \in \mathbb{F}_q^n$, there exists $g \in GL(n, q)$ such that $gu = v$, as one can extend $\{u\}$ to a basis and map it to a basis containing $v$. The action also induces transitivity on the set of lines (1-dimensional subspaces) in $\mathbb{F}_q^n$, with the projective space $\mathbb{P}^{n-1}(\mathbb{F}_q)$ serving as the orbit space.^[14][15] In coding theory, elements of $GL(n, q)$ generate linear error-correcting codes through their action on code subspaces or via hyperplane sections, providing constructions with controlled minimum distance for reliable data transmission over noisy channels. For instance, the hyperplanes in the variety of $GL(n, q)$ yield codes whose parameters relate directly to the group's order and stabilizer sizes.^[16]

Over arbitrary commutative rings

The general linear group over an arbitrary commutative ring $ R $ with identity, denoted $ \mathrm{GL}_n(R) $, consists of all $ n \times n $ matrices with entries in $ R $ whose determinants lie in the multiplicative group of units $ R^\times $, under the operation of matrix multiplication.^[17] This definition captures the automorphisms of the free $ R $-module $ R^n $, as a matrix $ A \in M_n(R) $ with $ \det(A) \in R^\times $ admits an inverse in $ M_n(R) $, ensuring it induces a module isomorphism.^[18] Over such rings, the determinant condition generalizes the field case but introduces subtleties due to the potential lack of unique factorization or division algorithms in $ R $. For principal ideal domains like the ring of integers $ \mathbb{Z} $, the group $ \mathrm{GL}_n(\mathbb{Z}) $ comprises precisely those integer matrices with determinant $ \pm 1 $. The units of $ \mathbb{Z} $ are $ { \pm 1 } $, so this condition ensures invertibility over $ \mathbb{Z} $. The kernel of the determinant map $ \det: \mathrm{GL}_n(\mathbb{Z}) \to { \pm 1 } $ is the special linear group $ \mathrm{SL}_n(\mathbb{Z}) $, consisting of matrices with determinant exactly 1, which plays a central role in arithmetic geometry and modular forms. The structure of $ \mathrm{GL}_n(R) $ depends profoundly on the Bass stable rank of $ R $, defined as the smallest integer $ d $ such that every unimodular row of length greater than $ d $ over $ R $ reduces to a shorter unimodular row via elementary operations. For a ring $ R $ with $ \mathrm{sr}(R) = d $, it holds that $ \mathrm{GL}_n(R) = \mathrm{E}_n(R) $ for all $ n > d $, where $ \mathrm{E}_n(R) $ is the subgroup generated by all elementary matrices (those obtained by adding multiples of one row/column to another). This equality reflects stabilization in algebraic K-theory, as $ \mathrm{E}_n(R) $ exhausts the stable general linear group $ \mathrm{GL}(R) = \varinjlim_n \mathrm{GL}_n(R) $, and the stable rank quantifies how quickly this stabilization occurs.^[19] Rings like Euclidean domains have stable rank 2, while polynomial rings in sufficiently many variables can have higher ranks, complicating the generation of $ \mathrm{GL}_n(R) $ for small $ n $. A representative example arises over polynomial rings $ R = k[x] $, where $ k $ is a field. The units $ k[x]^\times $ are the nonzero constant polynomials, isomorphic to $ k^\times $, so $ \mathrm{GL}_1(k[x]) \cong k^\times $.^[20] For $ n \geq 2 $, $ \mathrm{GL}_n(k[x]) $ includes matrices with constant nonzero determinants; since k[x] has stable rank 2, the elementary subgroup E_n(k[x]) equals SL_n(k[x]) for n ≥ 2, and thus generates the special linear part. Suslin's stability theorem confirms that this holds more generally for polynomial rings in multiple variables, with generation by elementary matrices for sufficiently large n. Unlike over fields, where all finitely generated projective modules are free, over general commutative rings there can exist non-free finitely generated projective modules (e.g., over Dedekind domains that are not principal ideal domains, such as the ring of integers in quadratic number fields with class number greater than 1, where non-principal ideals give rank-1 projectives). The group GL_n(R) describes the automorphisms of the free module R^n, while the endomorphism rings and automorphism groups of non-free projectives are studied using more advanced tools in algebraic K-theory.^[19]

Historical development

The concept of the general linear group emerged in the mid-19th century alongside the development of matrix theory. In 1858, Arthur Cayley published his seminal memoir on the theory of matrices, where he formalized the operations of matrix addition, multiplication, and the role of the determinant in characterizing nonsingular matrices, thereby laying the groundwork for the group of invertible linear transformations.^[21] Cayley and James Joseph Sylvester collaborated closely on invariant theory, and the term "general linear group" emerged in the late 19th century to describe the collection of all invertible linear substitutions acting on vector spaces. By the late 19th century, attention turned to linear groups over finite fields. Eliakim Hastings Moore, in the 1890s, explored the structure of finite vector spaces and the actions of linear transformations upon them, establishing foundational results on the classification of such groups and linking them to Galois fields.^[22] Building on this, Leonard Eugene Dickson in 1901 provided the first systematic treatment of linear groups over arbitrary fields in his monograph Linear Groups, with an Exposition of the Galois Field Theory, where he derived the cardinality of GL(n, q) as the product ∏_{k=0}^{n-1} (q^n - q^k).^[23] Parallel developments in continuous symmetry arose through Sophus Lie's work in the late 1880s, who conceptualized the general linear group GL(n, ℝ) as a continuous transformation group acting on Euclidean space, integrating it into his theory of differential equations and symmetries.^[24] In the early 1900s, Élie Cartan advanced this framework by developing the associated Lie algebra 𝔤𝔩(n), the tangent space at the identity comprising all n × n matrices under the commutator bracket, which facilitated the classification of semisimple Lie algebras.^[25] The mid-20th century saw the abstraction of the general linear group within modern algebra. In the 1930s, Emil Artin reformulated Galois theory using the action of Galois groups on vector spaces, incorporating GL(n, K) as the group of automorphisms of separable extensions, thereby embedding linear groups into the study of field extensions without reliance on primitive elements.^[26] Jean Dieudonné, in the 1940s as part of the Bourbaki collective, generalized the structure of GL(n) to division rings via the Dieudonné determinant, establishing it as a foundational object in the theory of algebraic groups over arbitrary fields.^[27] In the postwar era, applications proliferated across number theory and topology. Robert Langlands initiated his program in the 1960s, positing deep correspondences between n-dimensional Galois representations and automorphic representations of GL(n, ℚ\ℝ × ∏ ℚ_p), unifying disparate areas like class field theory and modular forms.^[28] Concurrently, in the 1970s, Daniel Quillen connected GL(n, k) to stable homotopy theory through algebraic K-theory, showing that the higher K-groups K_*(k) are the homotopy groups of the classifying space BGL(k)^+, linking linear algebra to the stable stems of spheres.^[29]

Key subgroups

Special linear group

The special linear group $\mathrm{SL}(n, K)$, where $K$ is a field, is defined as the kernel of the determinant homomorphism $\det: \mathrm{GL}(n, K) \to K^\times$, consisting of all $n \times n$ invertible matrices over $K$ with determinant equal to 1.^[30][7] This makes $\mathrm{SL}(n, K)$ a normal subgroup of $\mathrm{GL}(n, K)$, and since the determinant map is surjective, the index of $\mathrm{SL}(n, K)$ in $\mathrm{GL}(n, K)$ equals the cardinality of $K^\times$, which is infinite when $K$ is an infinite field.^[31] For $n \geq 2$ and fields $K$ with more than three elements, $\mathrm{SL}(n, K)$ is a perfect group, meaning it equals its own commutator subgroup $[ \mathrm{SL}(n, K), \mathrm{SL}(n, K) ]$.^[30] This property highlights its simple structure in group-theoretic terms, excluding small exceptional cases like $\mathrm{SL}(2, \mathbb{F}_2)$ and $\mathrm{SL}(2, \mathbb{F}_3)$. Moreover, $\mathrm{SL}(n, K)$ is generated by elementary transvections, specifically the matrices $E_{ij}(\lambda) = I + \lambda e_{ij}$ for $i \neq j$ and $\lambda \in K$, where $e_{ij}$ denotes the standard matrix unit with a 1 in the $(i,j)$-entry and zeros elsewhere.^[30] These generators reflect the group's close relation to the structure of vector spaces over $K$. Notable examples include $\mathrm{SL}(2, \mathbb{R})$, which acts on the upper half-plane $\mathcal{H} = \{ z \in \mathbb{C} \mid \operatorname{Im}(z) > 0 \}$ via Möbius transformations $z \mapsto \frac{az + b}{cz + d}$ for $\begin{pmatrix} a & b \\ c & d \end{pmatrix} \in \mathrm{SL}(2, \mathbb{R})$.^[32] This action preserves the hyperbolic metric and underlies much of the geometry of the hyperbolic plane. Similarly, $\mathrm{SL}(2, \mathbb{Z})$ is known as the modular group, acting on $\mathcal{H}$ with a fundamental domain given by the region $\{ z \in \mathcal{H} \mid |\operatorname{Re}(z)| \leq 1/2, |z| \geq 1 \}$, which tiles $\mathcal{H}$ under the group action.^[33] The associated Lie algebra $\mathfrak{sl}(n, K)$ comprises all $n \times n$ matrices $X$ over $K$ with $\operatorname{tr}(X) = 0$, forming a Lie subalgebra of $\mathfrak{gl}(n, K)$ under the commutator bracket.^[34] On $\mathfrak{sl}(n, K)$, the trace form $\operatorname{tr}(XY)$ serves as an analog to the Killing form, being proportional to it via the relation $B(X,Y) = 2n \operatorname{tr}(XY)$, where $B$ is the Killing form; this non-degenerate bilinear form underscores the semisimple nature of $\mathfrak{sl}(n, K)$ for $n \geq 2$.^[35]

Diagonal and unitary subgroups

The diagonal subgroup $D(n, K)$ of the general linear group $\mathrm{GL}(n, K)$ over a field $K$ consists of all invertible diagonal matrices, i.e., those with nonzero entries on the diagonal. This subgroup is isomorphic to the direct product $(K^\times)^n$, where $K^\times$ is the multiplicative group of the field, reflecting the independence of the diagonal entries.^[36] It serves as a maximal torus in $\mathrm{GL}(n, K)$, meaning it is a torus (a connected abelian algebraic group) that is maximal among such subgroups, and all maximal tori in $\mathrm{GL}(n, K)$ are conjugate under the group action.^[36] Over the complex numbers, the unitary group $U(n)$ is the compact subgroup of $\mathrm{GL}(n, \mathbb{C})$ defined by

$$U(n) = \{ A \in \mathrm{GL}(n, \mathbb{C}) \mid A^* A = I \},$$

where $A^*$ denotes the conjugate transpose of $A$, and $I$ is the identity matrix. This condition ensures that elements of $U(n)$ preserve the standard Hermitian inner product on $\mathbb{C}^n$. As a Lie group, $U(n)$ is compact and has real dimension $n^2$.^[37] Similarly, over the reals, the orthogonal group $O(n)$ is the compact subgroup of $\mathrm{GL}(n, \mathbb{R})$ given by

$$O(n) = \{ A \in \mathrm{GL}(n, \mathbb{R}) \mid A^T A = I \},$$

where $A^T$ is the transpose; it preserves the standard Euclidean inner product on $\mathbb{R}^n$. The group $O(n)$ has two connected components, distinguished by the determinant ($\det A = \pm 1$), with the special orthogonal group $\mathrm{SO}(n)$ comprising the component where $\det A = 1$; both have Lie algebra dimension $n(n-1)/2$.^[38] The Cartan decomposition provides a key structural insight into $\mathrm{GL}(n, \mathbb{R})$, expressing it as $\mathrm{GL}(n, \mathbb{R}) = [O(n)](/page/Orthogonal_group) A [O(n)](/page/Orthogonal_group)$, where $A$ is the subgroup of positive diagonal matrices (a Cartan subgroup). This decomposition arises from the polar form of matrices and highlights the role of the orthogonal group as a maximal compact subgroup.^{[39] [40]} Associated with the maximal torus (such as the diagonal subgroup), the Weyl group of $\mathrm{GL}(n, K)$ is the quotient of the normalizer of the torus by the torus itself, yielding the symmetric group $S_n$. It acts on the torus by permuting the diagonal entries, generated by reflections corresponding to adjacent transpositions, and is realized via monomial matrices (permutation matrices with nonzero entries on the permuted positions).^{[40] [41]}

Related groups and extensions

Projective linear group

The projective linear group $\mathrm{PGL}(n, R)$, also known as the projective general linear group, is the quotient of the general linear group $\mathrm{GL}(n, R)$ by its center, the scalar matrices $R^\times I_n$. This quotient captures the action of $\mathrm{GL}(n, R)$ on projective space, where scalar multiples act trivially.^[42]

Affine and semilinear groups

The affine group $\mathrm{Aff}(n, k)$, or general affine group over a field $k$, is the semidirect product $\mathrm{GL}(n, k) \ltimes k^n$, combining linear transformations with translations. It consists of all invertible affine transformations $x \mapsto Ax + b$ where $A \in \mathrm{GL}(n, k)$ and $b \in k^n$.^[43] The general semilinear group $\Gamma \mathrm{L}(n, k)$ extends $\mathrm{GL}(n, k)$ by incorporating field automorphisms, forming the semidirect product $\mathrm{GL}(n, k) \rtimes \mathrm{Aut}(k)$. Its elements are semilinear transformations $T(v) = \sigma(Av)$ for $\sigma \in \mathrm{Aut}(k)$, $A \in \mathrm{GL}(n, k)$.^[44]

Infinite-dimensional general linear group

The infinite-dimensional general linear group arises in two primary settings: the algebraic context over vector spaces and the analytic context over Hilbert spaces. In the analytic setting, for a complex Hilbert space $H$, the group $\mathrm{GL}(H)$ consists of all bounded linear operators on $H$ that admit bounded inverses, forming a group under composition. This group is equipped with the operator norm topology inherited from the Banach algebra $B(H)$ of all bounded operators on $H$. $\mathrm{GL}(H)$ is an open subset of $B(H)$, since for any invertible $T \in B(H)$, the set of $S \in B(H)$ satisfying $\|T^{-1}(T - S)\| < 1$ consists of invertible operators, with inverses given by the convergent Neumann series. The Calkin algebra, defined as the quotient $B(H)/K(H)$ where $K(H)$ is the ideal of compact operators, has $\mathrm{GL}(H)$ mapping onto its group of invertible elements; the preimage under this quotient map yields the Fredholm operators, whose index is analyzed via the Atiyah–Singer index theorem in geometric applications, associating the index to topological invariants like the Euler characteristic. Algebraically, for an infinite-dimensional vector space $V$ over a field $k$, $\mathrm{GL}(V)$ denotes the group of all invertible linear endomorphisms of $V$. However, due to the lack of finite bases, stability considerations lead to the infinite general linear group $\mathrm{GL}(\infty, k) = \varinjlim_n \mathrm{GL}(n, k)$, the direct (inductive) limit under the stabilization embeddings $\mathrm{GL}(n, k) \hookrightarrow \mathrm{GL}(n+1, k)$ that pad matrices with a $1$ in the bottom-right corner. This colimit embeds all finite-dimensional $\mathrm{GL}(n, k)$ densely and plays a central role in algebraic K-theory, where $K_1(k) = \mathrm{GL}(\infty, k)^{\mathrm{ab}}$ is the abelianization. Unlike its finite-dimensional counterparts, $\mathrm{GL}(H)$ for infinite-dimensional separable $H$ is not a Lie group in the classical finite-dimensional sense, though it forms an open subgroup of the Banach Lie group $B(H)$. A key topological property is its contractibility in the norm topology, established by Kuiper's theorem, implying that $\mathrm{GL}(H)$ is simply connected and homotopy-equivalent to a point. This contrasts with the finite-dimensional case, where $\mathrm{GL}(n, \mathbb{C})$ has nontrivial homotopy groups. Examples illustrate these structures: for the separable Hilbert space $H = \ell^2(\mathbb{Z})$, the bilateral shift operator $S e_n = e_{n+1}$ (where $\{e_n\}$ is the standard basis) is unitary, hence lies in $\mathrm{GL}(H)$ with norm $1$. The unilateral shift on $H = \ell^2(\mathbb{N})$ is a Fredholm operator in the preimage of the invertibles in the Calkin algebra, with Fredholm index $-1$, demonstrating the connection to index theory; more generally, the additivity of the index under composition reflects the group structure near the identity component.

Linear monoids

The full linear monoid over a commutative ring $R$, denoted $M_n(R)$, consists of all $n \times n$ matrices with entries in $R$, equipped with the operation of matrix multiplication, which is associative and has the identity matrix as the unit element.^[45] This structure forms a monoid, and the general linear group $GL(n, R)$ arises as the submonoid of invertible (unit) elements within it.^[45] In contrast to the group of invertibles, $M_n(R)$ includes non-invertible matrices, making it a richer algebraic object that captures all linear endomorphisms of the free $R$-module $R^n$. As a semigroup under multiplication, $M_n(R)$ exhibits notable structural features beyond those of groups. Idempotents in $M_n(R)$ are matrices $E$ satisfying $E^2 = E$, which, over integral domains, correspond to projections onto direct summands of $R^n$. Zero divisors abound, consisting of non-zero matrices $A$ and $B$ such that $AB = 0$, reflecting the presence of non-trivial kernels and cokernels in the associated linear maps. A key invariant is the rank of a matrix, defined as the minimal number of generators of the image submodule, which satisfies $\operatorname{rank}(AB) \leq \min(\operatorname{rank}(A), \operatorname{rank}(B))$ and provides a partial order on idempotents via comparability of their images. When $R = K$ is a field, $M_n(K)$ is isomorphic to the ring of $K$-linear endomorphisms $\operatorname{End}_K(K^n)$ under composition, forming a simple Artinian algebra.^[46] By the Wedderburn-Artin theorem, its unique decomposition as a semisimple algebra is $M_n(K) \cong M_n(K)$ itself, highlighting its indecomposability into smaller matrix components over the division ring $K$.^[46] In the infinite-dimensional setting, the full linear monoid analogue is $B(H)$, the set of all bounded linear operators on a complex Hilbert space $H$, forming a monoid under operator composition with the identity operator as unit.^[47] This structure generalizes finite-dimensional endomorphisms while incorporating norm considerations absent in the finite case. Matrix semigroups like subsemigroups of $M_n(R)$ have applications in automata theory, where they model state transformations in weighted or rational automata to analyze language recognition and growth rates.^[48] In control theory, they arise in the study of linear dynamical systems, aiding the determination of controllability and reachability via semigroup-generated trajectories.^[49]