Lie Theory

The Exponential Map: From Lie Algebra to Lie Group

Take the tangent space at the identity of a Lie group — a flat vector space you can add and scale — and one map bends it back onto the curved group, turning linear algebra into group theory. For a Lie group G with Lie algebra 𝔤 = T_eG, the exponential map exp: 𝔤 → G sends X to γ(1), where γ is the unique one-parameter subgroup with γ′(0) = X. It is a local diffeomorphism near 0, so a whole neighborhood of the identity is coordinatized by the algebra.

For matrix groups this is literally the power series exp(X) = ∑ₙ₌₀^∞ Xⁿ/n!, and the abstract statement specializes to the classical matrix exponential. The map linearizes the group: it converts the bracket [X,Y] on 𝔤 into the failure of G to commute, via exp(X)exp(Y) = exp(X + Y + ½[X,Y] + …), the Baker–Campbell–Hausdorff formula.

  • FieldDifferential geometry / Lie theory
  • Definitionexp(X) = γ_X(1), γ_X the one-parameter subgroup with γ_X′(0) = X
  • Key propertyLocal diffeomorphism near 0 ∈ 𝔤 (d(exp)₀ = id)
  • Matrix caseexp(X) = ∑ Xⁿ/n! (converges absolutely for all X)
  • Surjective whenG compact connected, or G connected nilpotent — not in general
  • Named forSophus Lie (1880s); matrix series back to Cauchy/Peano

Interactive visualization

Press play, or step through manually. The visualization is yours to drive — try it before reading on.

Open visualization fullscreen ↗

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Precise statement: what the exponential map is

Let G be a Lie group with identity e and Lie algebra 𝔤 = T_eG, the tangent space at e equipped with the bracket [·,·] induced by the commutator of left-invariant vector fields. Each X ∈ 𝔤 extends to a unique left-invariant vector field X̃ (X̃(g) = (dL_g)_e X). Because X̃ is complete on a Lie group, its flow through e exists for all time, giving a smooth homomorphism γ_X: ℝ → G with γ_X(0) = e and γ_X′(0) = X — the one-parameter subgroup generated by X.

Define exp: 𝔤 → G by exp(X) = γ_X(1). Then:

  • exp is smooth, and γ_X(t) = exp(tX) for all t ∈ ℝ;
  • the differential d(exp)₀: T₀𝔤 ≅ 𝔤 → T_eG = 𝔤 is the identity;
  • hence, by the inverse function theorem, exp restricts to a diffeomorphism from a neighborhood U of 0 ∈ 𝔤 onto a neighborhood of e ∈ G.

For G ⊆ GL(n,ℝ) a matrix group, 𝔤 ⊆ 𝔤𝔩(n) and exp coincides with the matrix series ∑ₙ Xⁿ/n!.

The picture: flattening the group near the identity

A Lie group is curved, but its algebra 𝔤 is a flat vector space. The exponential map is the standard chart that reconciles the two: it takes a straight line t ↦ tX through the origin of 𝔤 and lays it down on G as the one-parameter subgroup exp(tX), a smooth curve of group elements. Think of geodesics on a sphere emanating from a point: rays in the tangent plane become great circles on the surface. Here the 'geodesics' are one-parameter subgroups, and exp is the map that shoots them out.

Because d(exp)₀ = id, near the origin this laying-down is undistorted to first order — small tangent vectors map to nearby group elements almost isometrically. The nonlinearity of G shows up at second order: exp(X)exp(Y) is not exp(X+Y) unless X and Y commute. The correction is measured by the bracket, which is precisely the infinitesimal residue of noncommutativity. So exp is the dictionary translating additive, linear structure on 𝔤 into multiplicative, curved structure on G.

Key idea of the proof: flows of left-invariant fields

The engine is the theory of flows of vector fields plus left-invariance. For X ∈ 𝔤, form X̃ and let Φ_t be its (a priori local) flow. Left-invariance means dL_g ∘ Φ_t = Φ_t ∘ L_g wherever defined, so the flow through any point is a translate of the flow through e; this lets you extend the flow to all of ℝ (completeness) by concatenating short-time pieces via group multiplication. The resulting curve γ_X(t) = Φ_t(e) satisfies γ_X(s+t) = γ_X(s)γ_X(t) — it is a homomorphism.

Smoothness of exp in X follows from smooth dependence of flows on parameters (X enters the ODE smoothly). To compute d(exp)₀, differentiate t ↦ exp(tX) at t = 0: (d/dt)|₀ exp(tX) = γ_X′(0) = X, so d(exp)₀ = id. The inverse function theorem then delivers the local diffeomorphism. In the matrix case this all reduces to solving the linear ODE γ′(t) = γ(t)X, γ(0) = I, whose solution is the absolutely convergent series e^{tX}.

Canonical example: SU(2), SO(3), and the 2π surprise

Take G = SU(2), with 𝔤 = 𝔰𝔲(2) the traceless anti-Hermitian 2×2 matrices, spanned by (i/2)σ₁, (i/2)σ₂, (i/2)σ₃ (Pauli matrices). For a unit vector n̂ and angle θ,

exp(θ (i/2) n̂·σ) = cos(θ/2) I + i sin(θ/2) (n̂·σ),

a compact closed form from summing the series (since (n̂·σ)² = I). This is surjective onto SU(2) ≅ S³: every group element is a genuine exponential — as guaranteed for compact connected groups. The map is not injective: θ and θ + 4π give the same element, and antipodal directions coincide.

The double cover SU(2) → SO(3) explains the famous factor: a rotation by θ about n̂ in SO(3) lifts to exp(θ(i/2)n̂·σ), and it takes θ = 4π (not 2π) to return to the identity in SU(2) — the topological source of spin-½ behavior. For SO(3) itself, exp of a skew matrix is Rodrigues' rotation formula.

Why the hypotheses matter: where surjectivity and injectivity fail

The local diffeomorphism statement is unconditional for any Lie group, but global claims are delicate. Injectivity fails whenever G has nontrivial topology: on the circle 𝕋 = ℝ/ℤ, exp(t) = e^{2πit} is periodic, so exp is a local diffeo but a many-to-one covering. Surjectivity fails without compactness or nilpotency: in SL(2,ℝ) the element diag(−λ, −1/λ) with λ > 1, λ ≠ 1, is not in the image of exp — real eigenvalues of an exponential of a traceless real matrix must be positive or come in complex-conjugate pairs, and this matrix has two distinct negative eigenvalues. So exp: 𝔰𝔩(2,ℝ) → SL(2,ℝ) is not onto, even though the group is connected.

Compactness restores surjectivity: a bi-invariant metric exists, its geodesics through e are exactly one-parameter subgroups, and Hopf–Rinow guarantees every point is joined to e by a geodesic — so exp is onto. Connectedness is also essential: exp always lands in the identity component G⁰.

Applications: BCH, representations, and physics

The exponential map is the bridge on which most of Lie theory is built. It gives canonical coordinates of the first kind (X ↦ exp X) and second kind (products exp(x₁X₁)…exp(xₖXₖ)), the charts used to prove that a Lie group is determined near e by its algebra. The Baker–Campbell–Hausdorff formula log(exp X exp Y) = X + Y + ½[X,Y] + (1/12)([X,[X,Y]] − [Y,[X,Y]]) + … expresses group multiplication purely in bracket terms, giving the local equivalence of the categories of Lie groups (up to covering) and Lie algebras — the content of Lie's third theorem.

Representation theory uses it constantly: a representation of 𝔤 integrates to one of G (when G is simply connected) via ρ(exp X) = exp(dρ(X)). In physics, exp turns a Hamiltonian or angular-momentum operator into a time-evolution or rotation operator, e^{−itH/ℏ}, making 𝔤 the space of generators and G the group of symmetries. Riemannian geometry, control theory, and numerical integrators on manifolds all exploit the same construction.

Behavior of the exponential map under different hypotheses on the connected Lie group G
Hypothesis on Gexp surjective?exp a diffeomorphism?Representative example
Abelian, e.g. ℝⁿ or 𝕋ⁿYesDiffeo for ℝⁿ; only local for 𝕋ⁿexp = id on ℝⁿ; wrapping on the torus
Compact connectedYes (Hopf–Rinow argument)No (not injective; conjugate points)SO(3), SU(2), U(n)
Nilpotent simply connectedYes, and a global diffeomorphismYes (global diffeo)Heisenberg group
Connected semisimple noncompactNot alwaysNoSL(2,ℝ): exp misses some elements
General connectedNoNo (only local, near 0)Neither onto: SL(2,ℝ) and SL(2,ℂ) both miss elements (e.g. [[-1,1],[0,-1]] in SL(2,ℂ))

Frequently asked questions

Is the exponential map always surjective?

No. It is surjective for compact connected Lie groups (via a Hopf–Rinow geodesic argument) and for connected nilpotent Lie groups (where it is even a diffeomorphism for the simply connected ones). But it fails for general connected groups: in SL(2,ℝ), the matrix diag(−λ, −1/λ) with λ > 1 is not any exp(X). The image of exp always lies in the identity component G⁰.

Why is d(exp)₀ the identity, and why does that matter?

Differentiating the curve t ↦ exp(tX) at t = 0 gives γ_X′(0) = X, so exp does not distort tangent vectors at the origin to first order. Because the differential at 0 is invertible (it is the identity map 𝔤 → 𝔤), the inverse function theorem makes exp a local diffeomorphism near 0, which is what lets 𝔤 serve as a local coordinate chart for G near e.

For matrix groups, does the abstract definition agree with the power series?

Yes. For G ⊆ GL(n), the one-parameter subgroup γ_X solves the linear ODE γ′(t) = γ(t)X with γ(0) = I, whose unique solution is the absolutely convergent series γ(t) = ∑ₙ (tX)ⁿ/n! = e^{tX}. Setting t = 1 recovers exp(X) = ∑ₙ Xⁿ/n!. The series converges for every X since ‖Xⁿ/n!‖ ≤ ‖X‖ⁿ/n!.

Does exp(X)exp(Y) = exp(X+Y)?

Only when X and Y commute, i.e. [X,Y] = 0. In general, the Baker–Campbell–Hausdorff formula gives exp(X)exp(Y) = exp(X + Y + ½[X,Y] + (1/12)([X,[X,Y]] − [Y,[X,Y]]) + …). The bracket terms measure exactly how far the group is from being abelian; the series converges for ‖X‖, ‖Y‖ small.

Is the exponential map injective?

Only locally. It is a diffeomorphism from a neighborhood of 0 ∈ 𝔤 to a neighborhood of e ∈ G, so injective there. Globally it usually is not: on the torus 𝕋ⁿ it is periodic, and on SU(2) both antipodal directions and shifts by 4π give the same element. For simply connected nilpotent groups, however, exp is a global diffeomorphism, hence injective everywhere.

How does exp relate the Lie algebra representation to the group representation?

If ρ: G → GL(V) is a representation, its differential dρ: 𝔤 → 𝔤𝔩(V) is a Lie algebra representation, and they intertwine via exp: ρ(exp X) = exp(dρ(X)). Conversely, any Lie algebra representation integrates to a group representation when G is simply connected — this is why representation theory of a group is largely the linear-algebra study of its algebra.