An automorphism is an isomorphism from a mathematical object to itself. It rearranges the object’s elements or components without changing the structure under consideration. For a group, it preserves multiplication; for a graph, adjacency; for a vector space, addition and scalar multiplication. The automorphisms of an object form a group under composition, called its automorphism group.
Automorphisms make the idea of symmetry precise across mathematics. They identify which features of an object are intrinsic and which depend on labels, coordinates, or other choices of presentation. They are central to the study of polynomial equations, algebraic structures, geometric spaces, networks, and logical models. The same underlying set can have very different automorphism groups depending on the structure imposed on it, so specifying what an automorphism must preserve is essential.
Definition and structural preservation
For a mathematical structure , an automorphism is a map that is invertible and preserves all the designated structure. In elementary algebraic settings, this usually means a bijection satisfying equations that express preservation of operations. For a group , for example, the defining condition is
Bijectivity then ensures that the inverse map also preserves multiplication. For structures involving relations or topology, the requirements need more careful formulation: both a map and its inverse must preserve the relevant structure.
Consider a set equipped with a binary relation . An automorphism must satisfy
For a simple graph, is adjacency; for an ordered set, it is the order relation. The biconditional matters. Merely sending related pairs to related pairs is not, in general, enough to make a bijection an automorphism. Likewise, a continuous bijection of a topological space onto itself need not have a continuous inverse, so bijectivity and continuity alone do not supply the full definition of a topological automorphism.
The set of automorphisms of , denoted , is a group under composition. Its identity element is the identity map, composition is associative, and the inverse of every automorphism is again an automorphism. When has an underlying set, this group acts on by evaluation:
This group action organizes elements into orbits: two elements belong to the same orbit precisely when some automorphism takes one to the other. A structure whose only automorphism is the identity is called rigid.
Automorphisms are relative to a choice of structure. A finite set with elements and no additional structure has the full symmetric group as its automorphism group. Putting a linear order on that set reduces the group to the identity. Designating constants also restricts automorphisms: an automorphism of a structure with a named element must fix that element.
The real numbers provide a particularly instructive comparison. As a field, has no nonidentity automorphisms. As an additive group, however, it has many. Multiplication by any nonzero real number is an additive automorphism, although it generally does not preserve multiplication or the multiplicative identity. Thus an automorphism is never merely a property of a set or a formula; it is a symmetry of a specified mathematical structure.
Automorphisms of groups
In group theory, an automorphism preserves the multiplication table while possibly relabeling its entries. It necessarily fixes the identity, respects inverses, and preserves element orders, conjugacy, and subgroup relationships. Nevertheless, these necessary conditions do not by themselves characterize automorphisms: a permutation of a group can preserve element orders without respecting products.
An important family comes from conjugation. For each , define
This is an inner automorphism. The assignment is a group homomorphism whose kernel is the center , so the group of inner automorphisms satisfies
Inner automorphisms form a normal subgroup of , because
for every automorphism . The quotient
is the outer automorphism group. It measures automorphisms modulo those arising from conjugation within the group.
For an abelian group, every inner automorphism is the identity, but the full automorphism group can still be large and nonabelian. The additive group of integers has exactly two automorphisms, and , because an automorphism must send a generator to a generator. For the cyclic group , an automorphism sends a chosen generator to its -th power, where is relatively prime to . Consequently,
and its order is , where is Euler’s totient function. In particular, has four elements and is isomorphic to , not to .
A subgroup is called characteristic if every automorphism of the ambient group preserves it. This is stronger than being normal, which requires preservation only under inner automorphisms. The center and the subgroup generated by all commutators are characteristic. Characteristic subgroups help make group constructions independent of the particular names assigned to elements.
Automorphisms also specify how groups can be combined. A homomorphism from a group into determines a semidirect product . Its multiplication incorporates the chosen action of on . For example, a dihedral group arises by letting an element of order two act on a cyclic rotation group by inversion. Automorphisms therefore do more than describe existing symmetries: they provide data for constructing new groups.
Linear spaces, modules, and rings
For a vector space over a field , an automorphism is an invertible -linear map . The group of all such maps is the general linear group, written . After choosing a basis of a finite-dimensional space of dimension , it becomes the matrix group . A change of basis changes the matrices representing the automorphisms, but not the underlying transformations.
A square matrix represents an automorphism precisely when its determinant is nonzero. Over the finite field with elements, counting possible ordered bases gives
The first basis vector can be any nonzero vector. Each subsequent vector must lie outside the span of those already chosen. This calculation shows directly how the automorphism group encodes the freedom to choose coordinates.
Additional structure reduces the allowed transformations. If a real vector space carries an inner product, its structure-preserving automorphisms form an orthogonal group; their matrices satisfy . For a complex inner-product space, the corresponding condition is , giving a unitary group. More generally, preserving a specified quadratic form or bilinear form defines other classical groups. The difference between general linear and orthogonal automorphisms is the difference between preserving linear structure alone and preserving lengths and angles as well.
For a module over a ring , the analogous group is . These automorphisms are exactly the invertible elements of the ring of -linear endomorphisms of . In the case of the free abelian group , this gives . An integer matrix belongs to this group exactly when its determinant is or : being invertible over alone is insufficient because the inverse must again have integer entries.
A ring automorphism must preserve both addition and multiplication, and under the usual unital convention it preserves . The ring has only the identity automorphism, since fixing forces every integer to be fixed. For the polynomial algebra , every automorphism that fixes pointwise has the form
If substitution by a polynomial has a polynomial inverse , then , forcing both degrees to be one. In several variables, polynomial automorphisms are much more complicated.
For full matrix algebras, conjugation gives algebra automorphisms. The Skolem–Noether theorem implies that every -algebra automorphism of has this form. Transposition, by contrast, reverses multiplication:
It is therefore an anti-automorphism rather than an algebra automorphism when .
Field automorphisms and Galois theory
An automorphism of a field preserves addition, multiplication, and the identity. It consequently fixes the field’s prime subfield: a copy of in characteristic zero or of in characteristic . For a field extension , the notation denotes automorphisms of fixing every element of .
Such automorphisms permute roots of polynomials with coefficients in the fixed field. If , , and , then
This connects the algebra of equations to permutation groups. For example, has two automorphisms fixing , determined by
Its automorphism group is cyclic of order two.
Not every algebraic extension has as many automorphisms as its degree. The real field has degree three over but only the identity automorphism. The other two roots of are nonreal and do not belong to that field. Adjoining all three roots gives its splitting field, whose automorphisms form a group isomorphic to . The distinction illustrates the role of normality in field extensions.
For a finite Galois extension , Galois theory establishes an inclusion-reversing correspondence between intermediate fields and subgroups of
A subgroup corresponds to its fixed field
Moreover, the group has order . This theory grew from Évariste Galois’s work on equations in the early 1830s and explains why solvability by radicals is governed by a structural property of a permutation group.
For a finite field , the Frobenius map
is an automorphism. Its iterates generate the entire automorphism group over , which is cyclic of order . More generally, is cyclic of order , generated by .
Familiar infinite fields display contrasting behavior. Every field automorphism of preserves positivity because positive elements are exactly the nonzero squares. It therefore preserves order; since it fixes every rational number, the density of the rationals forces it to fix every real number. By contrast, has many field automorphisms, although its only continuous field automorphisms are the identity and complex conjugation. Continuity is a substantial extra requirement, not an automatic consequence of preserving field operations.
Graphs, finite structures, and computation
In graph theory, an automorphism of a simple graph is a permutation of its vertices preserving adjacency in both directions. If the graph has adjacency matrix , a permutation represented by a matrix is an automorphism precisely when
The formulation extends to directed graphs, colored graphs, and weighted graphs, provided the additional data are included among the structures that must be preserved. A symmetry of an uncolored graph may cease to be an automorphism once particular vertices receive distinct colors.
The complete graph has automorphism group , since every vertex permutation preserves adjacency. The cycle graph , for , has the dihedral group of order , consisting of cyclic rotations and reflections. A path with at least two vertices has only two automorphisms: the identity and reversal. These examples concern combinatorial structure rather than a particular drawing. A cycle drawn as an irregular polygon retains its graph automorphisms even if the drawing has no corresponding Euclidean symmetry.
Automorphism orbits distinguish different kinds of regularity. A graph is vertex-transitive if its automorphism group acts transitively on vertices, and edge-transitive if it acts transitively on edges. These properties are stronger than equal vertex degrees. For example, a disjoint union of a triangle and a square is regular of degree two, but its vertices cannot all lie in one orbit because an automorphism cannot map one connected component to a nonisomorphic component. Highly symmetric examples, such as the cube graph, support the study of regular networks and combinatorial designs.
Graph automorphisms also show how broadly groups can occur as symmetries. Frucht’s theorem, first proved in 1939, states that every finite group is isomorphic to the automorphism group of a finite undirected graph. A stronger version allows the graph to be chosen connected and cubic. Thus graph automorphism groups are not confined to the familiar symmetry groups of polygons or polyhedra.
Computationally, testing a proposed permutation is straightforward: adjacency can be checked directly. Finding all automorphisms is harder because a graph on vertices has possible permutations. Practical algorithms avoid exhaustive search by refining partitions of vertices according to structural information, branching on unresolved choices, and using discovered symmetries to prune the search. Systems such as nauty and Traces use these ideas to compute automorphism groups and canonical labelings.
Graph automorphism is closely related to graph isomorphism, but it asks a different question: symmetries within one graph rather than equivalences between two graphs. László Babai’s quasipolynomial-time graph-isomorphism result, announced in 2015, was a major theoretical advance in this area. In applications, automorphism computations support chemical structure comparison, enumeration of finite objects, and symmetry reduction in search problems, where equivalent configurations need not be explored separately.
Topological, geometric, and analytic automorphisms
In topology, the automorphisms of a space are its self-homeomorphisms: continuous bijections with continuous inverses. They form the group . These transformations preserve topological properties such as connectedness and compactness but need not preserve distances, angles, or differentiability. A substantial deformation of a sphere can be a homeomorphism without being an isometry.
For a smooth manifold, the appropriate automorphisms are diffeomorphisms, smooth bijections with smooth inverses. A Riemannian manifold has a more restrictive automorphism group when its metric is included in the structure: its transformations must be isometries. For Euclidean space, every bijective distance-preserving transformation has the form
where is orthogonal and is a translation vector. Requiring a distinguished origin to remain fixed removes the translations.
In complex geometry, automorphisms are biholomorphic self-maps. They are often much more rigid than homeomorphisms because holomorphicity imposes strong analytic constraints. Every automorphism of the complex plane has the form , with . Every automorphism of the Riemann sphere is a Möbius transformation,
Multiplying all four coefficients by the same nonzero scalar leaves the transformation unchanged.
For the open unit disk, the automorphisms are exactly
Their form follows from the Schwarz lemma after moving a selected point to the origin. They preserve the disk’s hyperbolic metric even though they generally do not preserve ordinary Euclidean distance.
The genus of a compact Riemann surface strongly influences its automorphisms. A genus-zero surface is the Riemann sphere and has an infinite automorphism group. A complex torus of genus one has infinitely many translation automorphisms, while the subgroup fixing its origin is finite. For genus , the automorphism group is finite, and Hurwitz’s bound gives
The Klein quartic, a genus-three surface, attains the bound with 168 automorphisms.
These examples illustrate the viewpoint associated with Felix Klein’s Erlangen program of 1872: geometries can be studied through transformation groups and their invariants. Topological, smooth, conformal, and metric structures impose different restrictions on allowable symmetries. Even for the same underlying space, the resulting automorphism groups can differ drastically in size, algebraic structure, and continuity properties.
Automorphisms in logic and model theory
In model theory, a structure consists of a domain together with interpretations of specified function symbols, relation symbols, and constants. An automorphism is a bijection of the domain preserving all these interpretations. If the language contains a binary operation , a binary relation , and a constant , an automorphism must satisfy
This framework places groups, fields, graphs, and ordered sets within one general definition.
Automorphisms preserve more than the operations and relations explicitly named in the language. They preserve the truth of every first-order formula. If is such a formula, then
This follows by induction on formulas, using bijectivity to handle quantifiers. Consequently, every set definable without parameters is invariant under every automorphism.
Parameters produce a relative version. For , write for the automorphisms fixing every element of . Every set definable using parameters from is preserved by this subgroup. If a formula with parameters from uniquely specifies an element , then every member of fixes . This links automorphisms to definable closure, the collection of elements uniquely definable from a given parameter set.
The converse requires hypotheses. In an arbitrary structure, being fixed by every automorphism does not necessarily imply definability. The ordered field is rigid, but in its countable language only countably many elements can be individually defined without parameters, whereas is uncountable. In suitably saturated and homogeneous ambient models, automorphism criteria characterize definable closure and algebraic closure. This is one reason model theory often studies a structure inside a larger, highly homogeneous model.
The ordered rationals provide a concrete contrast. As a field, is rigid. As a pure ordered set, it has many automorphisms; for example, preserves order. More generally, every order-preserving bijection between two finite subsets extends to an automorphism of . A back-and-forth construction proves this by successively extending a finite partial map while ensuring that every rational eventually enters its domain and range.
Automorphism groups therefore express the extent to which a structure can distinguish its own elements. Elements in different automorphism orbits are distinguishable by global structure, although not necessarily by a single formula in an arbitrary model. Under suitable countability and categoricity conditions, orbit structure and first-order definability align much more closely, making permutation groups an effective tool for classifying logical structures.
Categorical formulation and transport of symmetry
Category theory gives the most general standard definition: an automorphism of an object is an invertible morphism . If denotes all endomorphisms of , composition makes this collection a monoid, and
is its group of invertible elements. This formulation does not require objects to be sets or morphisms to be functions between underlying sets.
The category determines the relevant symmetry. In the category of sets, automorphisms are permutations. In the category of groups, they are bijective homomorphisms. In the category of topological spaces, they are homeomorphisms. A space equipped with several kinds of structure may be regarded as an object in several different categories, each selecting a different automorphism group. This makes precise the distinction between, for example, a complex manifold’s biholomorphic automorphisms and the homeomorphisms of its underlying topological space.
Invertibility must be understood within the category, not merely as bijectivity of an underlying function. The map
is continuous and bijective, but its inverse is not continuous at . It is therefore not an isomorphism of topological spaces. Conversely, in categories whose morphisms are not ordinary functions, set-theoretic injectivity and surjectivity may not be meaningful at all. The categorical definition avoids both difficulties by requiring an actual two-sided inverse morphism.
An isomorphism transports automorphisms by conjugation:
This is a group isomorphism. Hence isomorphic objects have isomorphic automorphism groups, although the particular group isomorphism generally depends on the chosen . Automorphism groups are useful invariants, but they are not complete invariants: nonisomorphic objects can have isomorphic automorphism groups. The rings and , for example, both have trivial automorphism groups.
The assignment does not generally extend to a functor on all morphisms. A homomorphism need not provide any way to transfer every automorphism of to . For a quotient group , an automorphism of descends to the quotient precisely when it preserves . If is characteristic, all automorphisms descend, yielding a homomorphism
but this homomorphism need not be injective or surjective.
A related distinction appears in algebraic topology. A self-homeomorphism of a pointed space induces an automorphism of its fundamental group. Without a fixed basepoint, choices of connecting paths change the induced automorphism by an inner automorphism. The naturally associated invariant is therefore often an outer automorphism rather than a specific automorphism. Such constructions show why the passage from geometric symmetry to algebraic symmetry depends carefully on which data are fixed and which identifications are allowed.