Matrix Groups: GL, SL, O, and U
What symmetries preserve a shape, a length, or a volume? This simple question opens the door to some of the richest infinite groups in all of mathematics. Once we know the Binary Operations and the Group Axioms that define an abstract group, the natural next step is clear. We should build large families of concrete examples from linear algebra. Matrices under multiplication supply exactly the right raw material. Matrix groups matter because they are simultaneously algebraic and geometric. A matrix is a symbol we can multiply, invert, and manipulate by hand. But it is also a linear transformation acting on a vector space. It stretches, rotates, or reflects that space. This double life makes matrix groups the bridge between pure group theory and geometry, physics, and computation. Every rotation of a rigid body is secretly an element of one of these groups. So is every quantum-mechanical symmetry, every linear code, and every computer graphics transformation. The big picture is a tower of increasingly restrictive conditions on matrices. We start with the general linear group , all invertible matrices. We impose a determinant condition to get the special linear group . We impose a length- or inner-product-preserving condition to get the orthogonal group . Its complex analogue is the unitary group . Each restriction cuts down a large group to a smaller, more rigid subgroup. This foreshadows the general theory of Subgroups and the Subgroup Criterion. This article defines all four families precisely. It proves that each is indeed a group under matrix multiplication. It explains the geometric meaning of each condition. It also works out concrete numerical examples in dimension 2 and 3. We also examine what breaks when hypotheses are dropped. This is the surest way to understand why each hypothesis is there in the first place.
- is the group of all invertible matrices over a field , under matrix multiplication.
- is the subgroup of matrices with determinant .
- is the group of real matrices preserving the standard dot product; equivalently .
- is the complex analogue, preserving the standard Hermitian inner product, satisfying .
- Determinants of orthogonal and unitary matrices have absolute value , giving natural subgroups and .
- These are all infinite (for , resp.) non-abelian (for ) groups, foreshadowing Group Homomorphisms via the determinant map.
- Key idea: invertibility gives a group under multiplication; extra geometric conditions (determinant, length-preservation) carve out subgroups that are still groups because those conditions are compatible with multiplication and inversion.
Where this lives today
The general linear group appears as the group of change-of-basis matrices in every linear algebra course and as the "ambient" group whose subgroups organize representation theory. The special linear group acts on lattices and modular forms, and is central to number theory (e.g., acting on the upper half plane defines the modular group underlying elliptic curves and modular forms). The orthogonal group and its subgroup are literally the group of rotations (and rotoreflections) of three-dimensional space, used throughout robotics, computer graphics, computer vision, and crystallography to encode rigid body orientation. The unitary group , especially and , is the mathematical backbone of the Standard Model of particle physics, where encodes the gauge symmetries of the strong, weak, and electromagnetic forces; also governs the evolution of quantum states in quantum computing, since unitary matrices are exactly the operations allowed on qubit states to preserve total probability. In cryptography, matrix groups over finite fields (finite analogues of ) underlie certain public-key schemes and error-correcting codes, and in statistics, orthogonal matrices arise as the change-of-basis matrices diagonalizing covariance matrices in principal component analysis.
Ask the tutor: Why does the determinant map from to being a homomorphism guarantee is a subgroup, and how does this generalize to other matrix groups?
Let denote a field (for concreteness, take or unless otherwise stated), and let be an integer.
General linear group.
the set of matrices over with nonzero determinant, under matrix multiplication.
Special linear group.
Orthogonal group (real case, ). Let denote the transpose of and the identity matrix.
Unitary group (complex case, ). Let denote the conjugate transpose.
In each case the binary operation is ordinary matrix multiplication, and we will show below that each set is closed under this operation, contains the identity matrix, contains inverses of its elements, and inherits associativity from .
Intuition
Think of an invertible matrix as a machine that takes column vectors in and outputs new column vectors, . The condition says exactly that this machine is reversible: no information about is destroyed, because has a matrix inverse undoing the transformation. This is the entire content of : it is the group of all reversible linear transformations of , full stop, with no further restriction. The special linear group asks for more: not just reversibility, but volume preservation. The determinant of measures the signed factor by which scales volumes (areas in dimension 2, volumes in dimension 3). Demanding means preserves both volume and orientation exactly, with no stretching or shrinking at all — think of shear transformations and rotations, but not dilations. The orthogonal and unitary groups impose a different, stronger geometric condition: preservation of length and angle, measured via an inner product. In , the standard inner product is , and exactly when for all — rotations and reflections, the rigid motions fixing the origin. In , the analogous Hermitian inner product is preserved exactly by unitary matrices. Since preserving all lengths and angles is a much stronger demand than preserving volume alone, we will see strictly and, in fact, that every orthogonal matrix has determinant , so sits inside the volume-computation picture too, but as a much smaller, "rigid" subgroup.
Let be a field and an integer. Then: 1. is a group under matrix multiplication, with identity and inverses given by matrix inversion. 2. is a subgroup of : it is closed under multiplication, contains , and is closed under inversion. 3. , defined for , is a subgroup of . 4. , defined for , is a subgroup of . Moreover, every element of has determinant , and every element of has determinant of absolute value ; consequently and genuinely (both matrices with the required determinant condition satisfied automatically), so and are indeed subsets of -adjacent territory but not contained in in general (only their determinant- subgroups are).
We use the subgroup criterion technique: to show a nonempty subset of a group is a subgroup, it suffices to show is closed under the group operation and under taking inverses (closure under the operation, together with inverses, forces the identity to lie in as well, since for any , by assumption and then by closure). We apply this after first establishing itself is a group, since all other claims are subgroup claims relative to it.
Step 1: under multiplication is a monoid, and is its group of units. Matrix multiplication on is associative: for , the entry of and of both equal
by direct expansion of the double sum defining matrix multiplication (this is a finite algebraic identity in the field , using only commutativity and associativity of in and the distributive law, and holds entrywise). The identity matrix satisfies for all , by direct computation of matrix entries (, so , and symmetrically on the other side). Thus is a monoid with identity . Now recall the basic linear-algebra fact (Cramer/adjugate theory over a field): a matrix has a two-sided multiplicative inverse in if and only if , and in that case the inverse is unique and is given by
where is the adjugate (transpose of the cofactor matrix), an expression built from -polynomial combinations of the entries of , hence itself lying in . We take this classical fact from linear algebra as known (it does not depend on group theory, so invoking it here is not circular). Consequently , defined as , is precisely the set of units of the monoid . It is a general fact, verified directly, that the set of units of any monoid forms a group under the monoid operation: it is closed under multiplication (if are units with inverses , then satisfies and similarly on the other side, so is a unit with inverse ), contains the identity ( is its own inverse), and is closed under inversion (if is a unit with inverse , then is a unit with inverse ). Associativity is inherited from . Hence is a group. This proves claim (1).
Step 2: is multiplicative, so is a subgroup. The determinant satisfies for all ; this is a standard identity of linear algebra (provable via the Leibniz expansion of the determinant as a signed sum over permutations, or via elementary row operations), independent of group theory, and we invoke it as known. Since , we have , and clearly so ; thus is nonempty. Closure under multiplication: if , then , so , and since , also . Hence . Closure under inversion: if , then , so exists in , and from we get ; since , this forces . Hence . By the subgroup criterion, is a subgroup of , proving claim (2).
Step 3: is a subgroup of . First we check : if , then taking determinants of both sides (using , another standard fact of linear algebra) gives , so , hence . This simultaneously proves the claimed fact that every element of has determinant . Nonemptiness: , so . Closure under multiplication: suppose , so and . Using (a standard identity: transpose reverses order of a product, provable entrywise, ), we compute
Hence . Closure under inversion: let . Since by the above, exists. From , multiply both sides on the left by and on the right appropriately: actually more directly, means is a left inverse of ; since is invertible with unique two-sided inverse , a left inverse must equal the two-sided inverse, so . Then
using . Hence . By the subgroup criterion, is a subgroup of , proving claim (3).
Step 4: is a subgroup of , by an identical argument with conjugate transpose. The conjugate transpose satisfies (since is transpose composed with entrywise complex conjugation, and both transpose-reversal and conjugation of a product being the product of conjugates in either order are standard facts: entrywise since complex conjugation is a ring homomorphism of , combined with the transpose-reversal identity above) and . If , then taking determinants: . Since (conjugating every entry of before taking the determinant, a polynomial in the entries with integer coefficients, conjugates the determinant, as complex conjugation is a field automorphism of fixing ), we get , so ; in particular . This proves the claimed fact that every unitary matrix has determinant of absolute value . Nonemptiness: , so . Closure under multiplication: for ,
so . Closure under inversion: as before, with invertible forces (a left inverse of an invertible matrix is the two-sided inverse), and then
so . By the subgroup criterion, is a subgroup of , proving claim (4). All four claims are established, and associativity throughout is inherited automatically from the ambient monoid , so no separate verification of associativity is needed for the subgroups.
∎Worked examples
Example 1: An explicit element of , and its inverse, showing the group operation concretely. Let
Compute . So in fact and ; let's instead perturb it slightly to get a genuine example and also illustrate inversion. Take
Since , , but since , . By the adjugate formula for matrices, , so
Check: . Row 1: , and . Row 2: , and . So indeed , confirming with the exhibited inverse, consistent with Theorem: Each of , , , is a group. Note also , matching the general fact used in the proof.
Example 2: A genuinely asymmetric element of and verification that it is orthogonal, plus its determinant. We build a rotation by angle with (a –– triangle angle, not a "nice" angle like ) about the -axis, combined with a reflection in the third coordinate to make the example asymmetric and to illustrate an orthogonal matrix with determinant :
We verify directly. First,
Compute entry by entry. Entry : . Entry : . Entry : . Entry : . Entry : . Entry : . By symmetry of the computation ( is automatically symmetric since ), the remaining entries equal their transposed counterparts, already computed as . So , confirming . Now compute . Expanding along the third row (which has entries ):
This matches the theorem: every element of has , and here we exhibit the case explicitly, so . This shows is strictly larger than , and that the "reflection part" genuinely changes the determinant sign while orthogonality is preserved — a good check that the two conditions (determinant , versus ) are logically independent.
The theorem crucially uses that consists only of matrices with . Suppose instead we tried to form a "group" of all matrices (dropping the invertibility restriction) under multiplication, claiming it is a group. This fails, and the failure is instructive. Take , , and
Then , so . We claim has no multiplicative inverse in : if satisfied , then
which can never equal since the entry of is always . So has no inverse in , and fails the inverse axiom of a group. Even worse, note under multiplication is not even a group's worth of "cancellative" structure: taking , we get , the zero matrix, illustrating that has zero divisors, a phenomenon fundamentally incompatible with group structure (in a group, with combinations behaving like zero divisors cannot occur, since every nonidentity element still has a well-defined inverse and cancellation always holds). This shows precisely why the determinant condition is not an artificial restriction but the exact condition needed to guarantee invertibility, and hence group structure.
Scope, edge cases & common traps
| Situation | What happens | Why |
|---|---|---|
| (nonzero scalars), , , , the circle group | Determinant of a matrix is just its entry | |
| a finite field | is finite, e.g. | Orthogonal/unitary analogues over finite fields exist but require care with the bilinear/Hermitian form definition |
| Non-abelian for | are all non-abelian once (with the trivial exception ) | Matrix multiplication generally does not commute in dimension |
| vs | has two connected pieces (det and det ); only is path-connected | Determinant is continuous and takes only values on , so it separates into two disjoint pieces |
| Assuming | False in general | allows ; only the determinant-1 elements of lie in |
The mathematician behind it
Origins · 1 minCamille Jordan and how this was discovered
Camille Jordan was a French mathematician whose 1870 treatise Traité des substitutions et des équations algébriques systematized the study of permutation groups and linear substitutions, laying groundwork that would later crystallize into the modern theory of matrix groups such as and . Jordan's work bridged Galois's theory of equations with the emerging language of linear transformations, and the Jordan canonical form — a cornerstone tool for understanding conjugacy inside — bears his name to this day.
Is abelian?
No, except in trivial cases. For , is abelian because multiplication in the field is commutative. But for , is non-abelian: for instance in , take and ; direct computation gives while , which are different matrices. This non-commutativity is one reason matrix groups are such a rich source of examples in group theory: they exhibit genuinely non-abelian behavior even in low dimensions, unlike simpler examples such as cyclic groups.
Why do we require to be a field in these definitions?
The definitions of rely on the determinant formula for the inverse, , which requires dividing by ; this division only makes sense when has a multiplicative inverse, guaranteed for nonzero elements precisely when the scalars form a field. Over a general commutative ring , one instead defines as matrices whose determinant is a unit of (not merely nonzero), and the same group-theoretic proof goes through verbatim with "unit" replacing "nonzero." Over , for instance, consists of integer matrices with determinant exactly , since those are the units of .
Are and finite or infinite groups?
Both are infinite for (for ) and effectively for (for , e.g. is the only finite case). For , contains a rotation by every possible angle , giving uncountably many distinct elements, and similarly contains uncountably many diagonal matrices . These groups are also naturally topological spaces (compact manifolds), which is part of why they are studied as Lie groups, though that structure is extra data layered on top of the pure group structure discussed here.
How exactly does relate to , and why is it called "special"?
is defined as , i.e., the orthogonal matrices with determinant exactly (as opposed to ). Since the determinant map is a group homomorphism (a fact foreshadowing Group Homomorphisms, since ) onto the two-element group , and is exactly its kernel, is automatically a subgroup of of index (assuming so that some orthogonal matrix has determinant , e.g. a reflection). Geometrically, consists precisely of the rotations, while consists of orientation-reversing maps like reflections.
Does every matrix in preserve some geometric structure, the way preserves length?
Not necessarily in the strong "isometry" sense, but every does preserve linear independence and hence dimension: since is invertible, are linearly independent in whenever are. That is the entire content of invertibility — it is a purely algebraic/linear condition, not a metric one. additionally preserves volume (up to sign in odd characteristic considerations, but exactly for ), while and preserve the much finer metric structure of lengths and angles. So the four groups form a genuine hierarchy of "how much structure is preserved," from purely linear () to volume () to full metric rigidity (, ).
Can I check membership in using rows instead of columns?
Yes, and this gives a useful sanity check. The condition is equivalent to (since in a group, a two-sided inverse is unique, and combined with invertible forces as shown in the proof). Expanding entrywise says the columns of form an orthonormal set with respect to the standard dot product; expanding says the same about the rows. So you may check either the columns or the rows are pairwise orthogonal unit vectors — both give equivalent, correct verifications of orthogonality.
What is the smallest example showing is a proper subgroup of ?
Take over : , all nonzero reals under multiplication, while , since the "determinant" of a matrix is just itself, and we need . So already at , is the trivial group while is infinite — an extreme but perfectly valid illustration that the determinant condition is a genuine, non-vacuous restriction, before even considering the more elaborate pictures used elsewhere in this article.