Lie Algebras and the Exponential Map

The algebra-to-group dictionary is the working machinery of modern physics. In quantum mechanics every conserved quantity is a generator: angular momentum spans $\mathfrak{su}(2)$, finite rotations are $\exp(-i\theta\,\hat n\!\cdot\!\vec J/\hbar)$, and time evolution is the one-parameter subgroup $\exp(-iHt/\hbar)$ generated by the Hamiltonian. The Standard Model is built entirely on exponentiated algebras: the gauge group $\mathrm{SU}(3)\times\mathrm{SU}(2)\times U(1)$ acts through $U=\exp(i\alpha^aT_a)$, and the non-Abelian field strength $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu+g[A_\mu,A_\nu]$ carries the bracket explicitly, so that the self-interaction of gluons is just the structure constants of $\mathfrak{su}(3)$ made dynamical. In quantum computing single-qubit gates are elements of $\mathrm{SU}(2)$ produced by exponentiating Pauli generators, with control pulses engineered by choosing which algebra element to exponentiate and for how long. Lattice gauge theory simulates $\mathrm{SU}(3)$ by discretizing these exponentials; robotics interpolates rigid-body motions through $\mathfrak{se}(3)$ (the screw/twist formalism); and the Killing–Cartan classification tells model-builders precisely which grand-unified groups — $\mathrm{SU}(5)$, $\mathrm{SO}(10)$, $E_6$ — are even available to try.

Picture the group as a curved surface and the identity $e$ as a marked point on it. The tangent plane touching the surface at $e$ is flat, and that plane — the velocities of curves through the identity — is the Lie algebra. A tangent vector $X$ is a pure direction and rate of infinitesimal transformation; the exponential map wraps that straight arrow onto an actual path in the group, the one-parameter subgroup $t\mapsto \exp(tX)$. The cleanest case is the circle group: its algebra is a single tangent line, and $\exp$ rolls that line around the circle, turning the rate $\theta$ into the rotation by angle $\theta$.

The bracket enters the moment the group is non-commutative. Walk a tiny closed loop: flow by $X$, then $Y$, then back by $X$, then back by $Y$. On a commutative group you return exactly; on a non-commutative one you miss, and the gap — to leading order — is precisely $[X,Y]$. So the algebra literally stores how much the group fails to commute, one bracket per pair of directions. This is not toothless abstraction: two small rotations about perpendicular axes, done in the two possible orders, differ by a real, measurable third rotation whose size is set by the bracket. The second figure makes that number explicit.

The starting definition is geometric: the Lie algebra is the tangent space at the identity, equipped with a product that lands back in the same space. Everything that follows is a property of this bracket or a computation with the exponential.

where: $\mathfrak{g}$ — the Lie algebra; $T_eG$ — the tangent space to the group $G$ at the identity $e$; $[\,\cdot\,,\,\cdot\,]$ — the Lie bracket, a bilinear map. Reading: $\mathfrak{g}$ is a finite-dimensional vector space, and the bracket is the only extra structure — it closes, so the bracket of two algebra elements is again in the algebra. The bracket is not an arbitrary product: it must be antisymmetric and satisfy the Jacobi identity. These two axioms, together with bilinearity, are the entire definition of a Lie algebra, and they are exactly the properties the matrix commutator turns out to have.

where: $X,Y,Z$ — elements of $\mathfrak{g}$. Reading: antisymmetry forces $[X,X]=0$; the Jacobi identity is the algebra's substitute for associativity, making $\operatorname{ad}_X=[X,\,\cdot\,]$ act as a derivation. For every matrix Lie group the bracket is completely concrete: it is the ordinary commutator of matrices, which is why so much of the physics reduces to matrix algebra rather than differential geometry.

where: $X,Y$ — matrices in a matrix Lie algebra such as $\mathfrak{gl}(n)$, $\mathfrak{so}(n)$, or $\mathfrak{su}(n)$. Reading: closure becomes a linear-algebra check — the commutator of two antisymmetric matrices is antisymmetric, and of two anti-Hermitian traceless matrices is again anti-Hermitian and traceless. Choosing a basis $\{T_a\}$ of generators reduces the whole algebra to a table of numbers: the brackets of basis elements re-expand in the same basis, with coefficients called the structure constants.

where: $T_a$ — a basis of $\mathfrak{g}$ (the generators); $f^{c}{}_{ab}$ — the structure constants, antisymmetric in $a,b$. Reading: the $f^{c}{}_{ab}$ fix the algebra up to isomorphism, satisfy a quadratic constraint from Jacobi, and are reproduced by every representation. The map back from algebra to group is the exponential. For matrices it is the familiar power series, and its defining feature is that its velocity at the identity is exactly the generator it was built from.

where: $\exp(X)$ — the group element generated by $X$; $t$ — a real parameter. Reading: the series converges for every matrix $X$; near $X=0$ the map is a diffeomorphism giving canonical coordinates, but globally $\exp$ need be neither injective nor surjective. The curve $t\mapsto e^{tX}$ is not just any path: it is a homomorphism from the additive line into the group, a one-parameter subgroup — the precise sense in which a single generator generates a continuous family of transformations.

where: $\gamma(t)$ — a smooth homomorphism $\mathbb{R}\to G$; $X=\gamma'(0)$ — its tangent at the identity, the generator. Reading: additivity $\gamma(s+t)=\gamma(s)\gamma(t)$ holds because $sX$ and $tX$ commute; it fails for $e^{X}e^{Y}$ when $X$ and $Y$ do not. That failure is quantified exactly by the Baker–Campbell–Hausdorff formula, which writes the product of two exponentials as a single exponential whose exponent is built entirely from iterated brackets — the deepest statement that the local group is stored in the algebra.

where: $\log$ — the local inverse of $\exp$ near $\mathbb{1}$. Reading: no ordinary products $XY$ appear anywhere — only brackets — so knowing the algebra's structure constants determines the multiplication law of $G$ in a neighbourhood of the identity to all orders.

The exponential and the bracket collapse into recognizable and much simpler behaviour in a handful of regimes that recur throughout physics.

1 · Rotating a spin-½ state by exponentiating the Pauli algebra (fully worked)

The generators of $\mathrm{SU}(2)$ are $-\tfrac{i}{2}\sigma_k$, with $\sigma_k$ the Pauli matrices. A finite rotation by $\theta$ about the unit axis $\hat n$ is $U(\theta,\hat n)=\exp\!\big(-\tfrac{i}{2}\theta\,\hat n\!\cdot\!\vec\sigma\big)$. Evaluate it in closed form, compute the $\theta=\pi/2$ rotation about $\hat z$, and use $\theta=2\pi$ to expose the double cover.

Step 1. Exploit the algebra. Because $(\hat n\!\cdot\!\vec\sigma)^2=\mathbb{1}$, the series resums into a cosine and sine, exactly as $e^{i\alpha}=\cos\alpha+i\sin\alpha$: $U(\theta,\hat n)=\cos\tfrac{\theta}{2}\,\mathbb{1}-i\sin\tfrac{\theta}{2}\,(\hat n\!\cdot\!\vec\sigma)$. The half-angle is the fingerprint of the $\tfrac12$ in the generator.

Step 2. Specialize to $\hat n=\hat z$, $\theta=\pi/2$, $\sigma_z=\operatorname{diag}(1,-1)$: $U=\cos\tfrac{\pi}{4}\,\mathbb{1}-i\sin\tfrac{\pi}{4}\,\sigma_z=\tfrac{1}{\sqrt2}(\mathbb{1}-i\sigma_z)=\operatorname{diag}\!\big(e^{-i\pi/4},\,e^{+i\pi/4}\big)$. Its determinant is $e^{-i\pi/4}e^{+i\pi/4}=1$, confirming $U\in\mathrm{SU}(2)$ as the det–trace identity demands ($\operatorname{tr}(-\tfrac{i}{2}\theta\sigma_z)=0$).

Step 3. Push to a full turn, $\theta=2\pi$: $U=\cos\pi\,\mathbb{1}-i\sin\pi\,(\hat n\!\cdot\!\vec\sigma)=-\mathbb{1}$. A $360^\circ$ rotation returns a spin-½ state to $-1$ times itself, not to itself; only at $\theta=4\pi$ does $U=+\mathbb{1}$. A spin-½ needs $720^\circ$ to come home. This is the algebra-level signature that $\mathrm{SU}(2)$ double-covers $\mathrm{SO}(3)$: the same $\mathfrak{su}(2)$ exponentiates to a group with twice the turns, which is why half-integer spin exists at all.

2 · How much do two small rotations fail to commute? (fully worked)

Take $X=\varepsilon L_1$ and $Y=\varepsilon L_2$ in $\mathfrak{so}(3)$ with $\varepsilon=0.1$ rad — small rotations about the $x$ and $y$ axes. Quantify, as a concrete angle, by how much the order of the two rotations matters.

Step 1. Use the commutator result. The order-of-operations discrepancy is $e^{X}e^{Y}e^{-X}e^{-Y}=\exp\!\big([X,Y]+O(\varepsilon^3)\big)$, and $[X,Y]=\varepsilon^2[L_1,L_2]=\varepsilon^2 L_3$ from the $\mathfrak{so}(3)$ bracket above.

Step 2. Read off the axis and angle. The leftover generator $\varepsilon^2 L_3$ is a rotation about the $z$ axis — the third, unused direction — by angle $\varepsilon^2=(0.1)^2=\mathbf{0.01\ rad}\approx0.57^\circ$. So the two orderings of an $x$- and a $y$-rotation differ by a real $z$-rotation of about half a degree.

Step 3. Cross-check and interpret. Multiplying the four exponentials as matrices gives a rotation whose off-diagonal $z$-block is $\sin(0.00997)$, agreeing with $\varepsilon^2=0.01$ to within the expected $O(\varepsilon^3)$ BCH correction. The non-commutativity of rotations is not a paradox but a measurable $\varepsilon^2$ effect — the microscopic origin of everything from the gyroscopic drift of a tumbling body to the geometric (Berry) phase, all bookkept by $[L_1,L_2]=L_3$.

The whole edifice — algebras, generators, the exponential dictionary — descends from one man's attempt to do for differential equations what Galois had done for polynomials.

Texts. Stone and Goldbart, Mathematics for Physics, ch. 17 (Lie groups and Lie algebras) · Arfken, Weber and Harris, Mathematical Methods for Physicists, 7th ed., ch. 17 · Nakahara, Geometry, Topology and Physics, 2nd ed., ch. 5 · Hall, Lie Groups, Lie Algebras, and Representations, chs. 2–3 · Gilmore, Lie Groups, Physics, and Geometry · Riley, Hobson and Bence, Mathematical Methods for Physics and Engineering, 3rd ed., ch. 28 · Byron and Fuller, Mathematics of Classical and Quantum Physics, ch. 10. Key self-consistency checks: (1) re-expanding $e^{sX}e^{tY}e^{-sX}e^{-tY}$ to second order leaves $st[X,Y]$ as the sole surviving term — the cross term of $e^{-sX}e^{-tY}$ is $+stXY$, not $stYX$; (2) $[L_1,L_2]=L_3$ follows directly from $(L_i)_{jk}=-\epsilon_{ijk}$; (3) $\det e^X=e^{\operatorname{tr}X}$ is established by triangularization; (4) $\exp(-\tfrac{i}{2}\theta\hat n\!\cdot\!\vec\sigma)=\cos\tfrac\theta2\,\mathbb{1}-i\sin\tfrac\theta2\,\hat n\!\cdot\!\vec\sigma$, with $\theta=2\pi$ giving $-\mathbb{1}$; (5) the $\varepsilon=0.1$ commutator loop returns $\exp(\varepsilon^2L_3)$, a $0.01$ rad twist about $z$.

Prerequisites. Lie Groups · Groups, Subgroups, and Cosets · Group Homomorphisms and Quotient Groups. Leads to. The Rotation Group SO(3) · SU(2) and the Spin Double Cover · Group Representations.

Ask the tutor: Ask me to exponentiate any generator you like — a Pauli matrix, an $\mathfrak{so}(3)$ rotation, or a boost — and watch the algebra unfold into a finite group element with the bracket bookkeeping every correction.