JOVANA
Explore Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

What Is a Lie Group? The Classical Matrix Groups

A Lie group is a single object that is at once a smooth manifold and a group, with the two structures wired together compatibly. We meet the definition, see why continuous symmetry forces it, and tour the workhorses you will use all rung long: GL, SL, O, SO, U, SU and Sp.

Two structures on one space, talking to each other

You already command both halves of the idea. From this rung's earlier guides you know a smooth manifold M and what it means for a map f: M -> N to be smooth; from undergraduate algebra you know a group: a set with an associative product, an identity e, and inverses. A Lie group is the simplest possible marriage of the two — a set G that carries a smooth manifold structure and a group structure, with the one demand that the group operations respect the smoothness. Nothing here is exotic yet; the whole subject is what happens when you insist these two ordinary structures live on the same space and cooperate.

Spelled out, the compatibility is a single condition: the multiplication map m: G x G -> G, m(g, h) = gh, and the inversion map i: G -> G, i(g) = g^(-1), are both smooth. (G x G is a manifold in the obvious product way, so "smooth" there means exactly what you learned.) That is the entire definition. It is worth pausing on how lean it is — we do not separately demand that inversion be smooth and continuous and so on; one clause about m and one about i, and the rich theory of the rest of this rung follows. A common shortcut, by the way: requiring the single map (g, h) -> g h^(-1) to be smooth is equivalent and saves a line.

Why continuous symmetry hands you a Lie group

The reason Lie groups are unavoidable is that they are the natural home of continuous symmetry. A finite symmetry group — say the eight symmetries of a square — is a discrete set: you can rotate by ninety degrees but not by a hair less. The symmetries of a circle are different in kind: you can rotate by any real angle theta, and those rotations form the group SO(2), which is the circle itself wearing a group structure. The moment a system has a symmetry you can deform continuously, the set of symmetries acquires a topology and a smooth structure, and the group law is smooth because composing two nearby rotations gives a nearby rotation. That is a Lie group arising on its own.

This is also why Lie groups never travel alone — they come paired with the things they act on. A Lie group G acting smoothly on a manifold M is a smooth group action, and when the action is transitive (you can reach any point from any other by some g) the space M is a homogeneous space G/H. Keep this in your pocket: a Lie group is not just a curved set of matrices, it is a machine for moving a geometry around rigidly, and most of the spaces you care about — spheres, hyperbolic space, projective space — are homogeneous spaces of classical groups. Guide 5 of this rung makes that the whole story.

The big example: GL(n, R) and how it spawns the others

The mother of all examples is the general linear group GL(n, R): all invertible n-by-n real matrices, with matrix multiplication as the group law. Why is it a manifold? The set of all n-by-n matrices is just R^(n^2), a flat Euclidean space, and "invertible" means "determinant not zero." Since det is a continuous function, the invertible matrices form an open subset of R^(n^2) — and an open subset of a manifold is a manifold of the same dimension. So GL(n, R) is an open chunk of R^(n^2), hence a smooth manifold of dimension n^2 with one global chart, no atlas-gluing required.

Now check the group law is smooth. Each entry of a product AB is a polynomial in the entries of A and B — sums of products — so m: GL x GL -> GL is smooth, polynomials being as smooth as anything. Inversion is smoother than it looks: by Cramer's rule each entry of A^(-1) is a polynomial in the entries of A divided by det A, and det A never vanishes on GL, so i: A -> A^(-1) is a ratio of smooth functions with nonzero denominator, hence smooth. Both clauses hold, so GL(n, R) is a genuine Lie group — and almost every other classical group is a subgroup of it cut out by equations.

GL(n, R)  =  { A in M_n(R)  :  det A != 0 }     (open in R^(n^2),  dim = n^2)

multiplication   m(A, B) = AB           entries are polynomials  -> smooth
inversion        i(A)    = A^(-1)        Cramer: poly / det A     -> smooth

SL(n, R) = { det A = 1 }     O(n) = { A^T A = I }     SO(n) = O(n) cap SL
GL(n, R) is an open set in matrix space; the other classical groups are level sets of det or of A^T A inside it.

The classical groups, and the theorem that legitimizes them

The classical matrix groups are the named subgroups of GL you will use endlessly, each defined by a clean algebraic constraint. SL(n, R), the special linear group, is det A = 1 (volume-preserving maps). O(n), the orthogonal group, is A^T A = I (maps preserving lengths and angles — rigid rotations and reflections); inside it SO(n) adds det A = 1 to keep only the rotations. Over the complex numbers the parallel cast is U(n), the unitary group with A* A = I (the complex-conjugate-transpose A* replacing A^T), and SU(n) with the extra det A = 1. Finally Sp(2n) preserves a symplectic form — the symmetry of Hamiltonian mechanics. Each is a Lie group, and each will reappear with its own personality.

Why is each of these actually a manifold, and not some jagged set? Two complementary answers. The hands-on one: O(n) is the zero set of the map A -> A^T A - I landing in symmetric matrices, and that map has the constant-rank property of the regular value theorem, so its level set is automatically a smooth submanifold — you can compute its dimension as n(n-1)/2 this way. The structural one is cleaner and is the cornerstone of the whole subject: Cartan's closed-subgroup theorem says any subgroup of a Lie group that is merely closed as a topological subset is automatically a smooth embedded Lie subgroup, with no checking of smoothness required.

A tiny worked example: SU(2) is the 3-sphere

Let us make the abstraction concrete by identifying one classical group completely. Take SU(2): two-by-two complex matrices A with A* A = I and det A = 1. Writing the conditions out, every such matrix has the form with first row (a, b) and second row (-b-bar, a-bar), where a-bar and b-bar are complex conjugates, subject to the single equation |a|^2 + |b|^2 = 1. Split a = x_1 + i x_2 and b = x_3 + i x_4 into real and imaginary parts, and that one equation becomes x_1^2 + x_2^2 + x_3^2 + x_4^2 = 1.

But that is exactly the equation of the unit sphere S^3 in R^4. So as a manifold SU(2) IS the three-sphere — a compact, connected, simply connected 3-manifold — now additionally carrying a smooth group law. This is the smallest place where you can hold a whole Lie group in your hand: a familiar round sphere that also knows how to multiply. It is no accident that SU(2) governs the quantum spin of an electron and double-covers the rotation group SO(3); we will return to that cover when one-parameter subgroups and the exponential map arrive in Guide 3.

  1. Write a general SU(2) matrix using A* A = I and det A = 1; it reduces to first row (a, b), second row (-b-bar, a-bar).
  2. The remaining constraint det A = 1 collapses to the single real equation |a|^2 + |b|^2 = 1.
  3. Set a = x_1 + i x_2, b = x_3 + i x_4 to read |a|^2 + |b|^2 = x_1^2 + x_2^2 + x_3^2 + x_4^2.
  4. Recognize x_1^2 + x_2^2 + x_3^2 + x_4^2 = 1 as the unit sphere S^3 in R^4, so SU(2) is diffeomorphic to S^3.

Where this rung is heading

The definition you now hold is deliberately a starting point, not the payoff. A group element that you can vary continuously begs to be differentiated, and differentiating the multiplication near the identity is exactly how the manifold and the group start trading information. That derivative-at-the-identity is the Lie algebra, the linearization of G, and it is astonishingly powerful: a finite-dimensional vector space with a bracket that, by Lie's theorems, captures nearly everything about G near the identity. Guide 2 builds it carefully, via left-invariant vector fields, and that is where the subject truly opens up.

A caution worth carrying. The classical groups feel concrete because they are matrices, and that is a gift — but it can mislead. Not every Lie group is naturally a matrix group (the universal cover of SL(2, R) is the standard counterexample, admitting no faithful finite-dimensional matrix representation), so a theorem you prove for matrices may need an honest extra argument in general. For this rung the matrix groups are more than enough to learn on, and we will lead with them; just keep a mental asterisk that "Lie group" is broader than "matrix group," and we will flag the gap whenever it actually bites.