Lie Groups

A Lie group is the exact object that appears whenever a symmetry is continuous rather than discrete: the operations form a group, yet the parameters labelling them — an axis and an angle, a phase, a boost rapidity — vary smoothly, so the operations themselves fill out a curved space, a differentiable manifold. The algebra of composition and the geometry of a manifold, usually kept apart, are here forced to be compatible, and the single technical demand that makes them so is that the map $(g,h)\mapsto gh^{-1}$ be smooth. From that one requirement an enormous amount follows, in a way unreasonably kind to the physicist. The kindness has a name: homogeneity. Multiplying the whole group on the left by a fixed element slides it rigidly onto itself, so the manifold looks identical near every point, and an arbitrarily small neighbourhood of the identity already contains the entire local law. The directions you can move in at the identity form the tangent space $T_eG$ , and the Lie bracket makes that vector space into the Lie algebra $\mathfrak{g}$ , a finite-dimensional object recording the group's non-commutativity to leading order. The exponential map integrates an infinitesimal generator into a finite one-parameter family of symmetries, converting functional equations on $G$ into linear algebra on $\mathfrak{g}$ . What the algebra cannot see is topology: two groups can share a Lie algebra and still differ globally — $SU(2)$ is a simply connected 3-sphere, $SO(3)$ is real projective 3-space and doubly connected — which is exactly why spinors exist and why a $720°$ rotation, not $360°$ , returns a spin- $\tfrac12$ state to itself.

Key takeaways

A Lie group is a smooth manifold $G$ whose multiplication and inversion are smooth; equivalently the single map $(g,h)\mapsto gh^{-1}$ is $C^\infty$ .
Left translation $L_g$ is a diffeomorphism, so $G$ is homogeneous: the whole manifold is a smeared-out copy of a neighbourhood of the identity, and its local law lives in the Lie algebra $\mathfrak{g}=T_eG$ .
The exponential map $\exp:\mathfrak{g}\to G$ turns a generator $X$ into a one-parameter subgroup $\exp(tX)$ ; for matrix groups it is $\sum X^n/n!$ .
Compact groups ( $SO(n)$ , $SU(n)$ , $Sp(n)$ ) carry a finite bi-invariant Haar measure, making every finite-dimensional representation unitarizable and completely reducible.
The classical dimensions are $\dim SU(n)=n^2-1$ , $\dim SO(n)=n(n-1)/2$ , $\dim U(n)=n^2$ : so $SU(3)$ has 8 generators (the 8 gluons) and $SU(2)$ has 3.
Same algebra, different group: $\mathfrak{su}(2)\cong\mathfrak{so}(3)$ , yet $SU(2)$ double-covers $SO(3)$ , which is why a $2\pi$ rotation is $+1$ in $SO(3)$ but $-1$ on a spinor.
Key idea: a Lie group is homogeneous — left translation by any element is a diffeomorphism, so the entire manifold is a smeared-out copy of an arbitrarily small neighbourhood of the identity. Studying the tangent space at the identity (the Lie algebra) together with the exponential map captures the whole local structure, and the global structure is then governed only by discrete topological data — the fundamental group and the centre.

The core insight

A Lie group is homogeneous: left translation by any element is a diffeomorphism, so the entire manifold is a smeared-out copy of an arbitrarily small neighbourhood of the identity. Studying the tangent space at the identity — the Lie algebra — together with the exponential map captures the local structure of the whole group, and the global structure is then governed only by discrete topological data, the fundamental group and the centre.

Where this lives today

Lie groups are the working vocabulary of modern physics. Every fundamental interaction is a gauge theory built on one: the Standard Model on $SU(3)\times SU(2)\times U(1)$ , its would-be unifications on $SU(5)$ , $SO(10)$ , or $E_6$ . Lattice quantum chromodynamics — the only known way to compute the proton mass from first principles — is literally Monte Carlo integration over a product of $SU(3)$ copies, one per lattice link, weighted by exactly the Haar measure of Derivation C; the associated Yang–Mills mass-gap problem remains a Clay Millennium open question. In quantum information the single qubit is a unit vector acted on by $SU(2)$ , so every one-qubit gate is a point of the double cover in this article. Wigner's $SO(3,1)$ -and-translations classification still defines what "particle" means — a mass and a spin labelling an irreducible unitary representation of the Poincaré group. And the newest arena is machine learning: equivariant, geometric deep-learning architectures bake in $SO(3)$ , $SU(2)$ , and $E(3)$ symmetry so a network respecting the Lie group of a problem — molecular structure, protein folding, cosmological fields — generalises from far less data, a direct payoff of homogeneity.

The physical picture

Picture the rotations of a sphere. Each is fixed by an axis and an angle — three real numbers that vary continuously — so the set of rotations is itself a curved three-dimensional space, and composing or inverting rotations moves you smoothly around inside it. That is a Lie group in one image: a continuous knob-turning symmetry whose knobs turn differentiably. The decisive move is to look only at the identity. Because multiplying by a fixed $g$ merely slides the group rigidly onto itself, every point has the same neighbourhood as the identity; the directions you can leave the identity in are the Lie algebra $\mathfrak{g}$ , and the exponential map straightens each into a one-parameter family of finite symmetries. The first figure is this dictionary — a generator in the flat tangent space at $e$ , integrated by $\exp$ into a curve on the manifold, with a concrete rotation instance attached.

Figure

A generator in the tangent space at the identity is integrated by the exponential map into a one-parameter subgroup on the curved group manifold. For rotations the generator J_z gives exp(θ J_z) = R_z(θ), a circle that closes at θ = 2π so that exp(2π J_z) = I, and the algebra dimension equals the manifold dimension, dim g = dim G = 3 for SO(3).

The failure of commutativity is measured infinitesimally by the Lie bracket: step a little along $X$ , then $Y$ , then back along each, and you do not return home — the leftover displacement is $[X,Y]$ to leading order, the entire local fingerprint of the group's non-abelian character. But the algebra is deliberately blind to global topology. $SU(2)$ , the unit quaternions, is a 3-sphere and simply connected; $SO(3)$ is that sphere with antipodes glued, real projective 3-space, with a loop that cannot be contracted. $SU(2)$ wraps twice around $SO(3)$ — a double cover — so a rotation and "the same rotation plus $2\pi$ " are one point downstairs but two distinct points upstairs, differing by a sign. The second figure is that sign made quantitative: the spinor amplitude $\cos(\theta/2)$ that only comes home after $720°$ .

Figure

The SU(2) double cover of SO(3) tracked by the spin-½ amplitude cos(θ/2). A rotation by 360° returns every direction in space to itself, so it is the identity in SO(3), yet it maps a spinor to minus itself; only a 720° rotation restores the spinor to +1. SU(2) is the simply connected 3-sphere covering SO(3) = RP³ two-to-one, and π₁(SO(3)) = ℤ₂.

The core equations

A Lie group is a manifold on which the group operations respect the smooth structure. Remarkably, one condition on a single map delivers both multiplication and inversion at once — hence the economical definition.

\mu:G\times G\to G,\quad \iota:G\to G,\qquad (g,h)\mapsto gh^{-1}\ \text{ is }\ C^\infty \qquad (1)

where: $G$ — the underlying smooth manifold; $\mu$ — multiplication, $\mu(g,h)=gh$ ; $\iota$ — inversion, $\iota(g)=g^{-1}$ ; $C^\infty$ — infinitely differentiable. Reading it: algebra and geometry are welded by smoothness, so calculus is legal everywhere on the group. The engine of the whole theory is that translating the group by a fixed element is a diffeomorphism — this is what makes $G$ homogeneous and exports every construction at the identity to every other point.

L_g:G\to G,\quad L_g(h)=gh,\qquad (L_g)^{-1}=L_{g^{-1}} \qquad (2)

where: $L_g$ — left translation by $g$ ; $h$ — the element being translated. Reading it: $L_g$ has the smooth inverse $L_{g^{-1}}$ , so $G$ looks the same near $g$ as near $e$ . Because of homogeneity you only need the identity's neighbourhood, whose infinitesimal data is the tangent space there. The Lie bracket promotes that vector space to an algebra whose dimension equals the group's.

\mathfrak{g}=T_eG,\qquad \dim\mathfrak{g}=\dim G,\qquad [X,Y]=XY-YX\in\mathfrak{g} \qquad (3)

where: $T_eG$ — tangent space to $G$ at the identity; $\mathfrak{g}$ — the Lie algebra; $[X,Y]$ — the Lie bracket (the matrix commutator for matrix groups). Reading it: the bracket is the leading-order remnant of $ghg^{-1}h^{-1}$ . The bridge from the flat algebra to the curved group is the exponential map, integrating a generator into a one-parameter subgroup, a smooth homomorphism of the real line.

\exp(X)=\sum_{n=0}^{\infty}\frac{X^n}{n!},\qquad \gamma(t)=\exp(tX),\qquad \gamma(s+t)=\gamma(s)\,\gamma(t) \qquad (4)

where: $\exp$ — the exponential map $\mathfrak{g}\to G$ ; $\gamma(t)$ — the one-parameter subgroup generated by $X$ ; $s,t$ — real flow parameters. Reading it: additivity in the exponent holds along one direction only; two different generators need the correction below. Conjugation lets the group act on its own algebra, and differentiating that action recovers the bracket — the adjoint representations tie group and algebra together.

\mathrm{Ad}_g(X)=gXg^{-1},\qquad \mathrm{ad}_X(Y)=[X,Y] \qquad (5)

where: $\mathrm{Ad}_g$ — the adjoint action of $G$ on $\mathfrak{g}$ by conjugation; $\mathrm{ad}_X$ — its differential, the algebra acting on itself. Reading it: $\mathrm{Ad}$ is how generators rotate into one another under a finite transformation. Multiplying two exponentials does not add the exponents unless the generators commute; the exact answer is a universal series in nested brackets, the Baker–Campbell–Hausdorff formula.

\exp(X)\exp(Y)=\exp\!\Big(X+Y+\tfrac12[X,Y]+\tfrac1{12}\big([X,[X,Y]]-[Y,[X,Y]]\big)+\cdots\Big) \qquad (6)

where: $X,Y$ — algebra elements; every term past the first two is an iterated bracket. Reading it: the group law near the identity is completely determined by the bracket, a purely algebraic datum. Compact groups possess a unique (up to scale) invariant volume, the Haar measure, and its finiteness is the hinge of their representation theory.

\int_G f(hg)\,\mathrm{d}\mu(g)=\int_G f(g)\,\mathrm{d}\mu(g)=\int_G f(gh)\,\mathrm{d}\mu(g),\qquad \int_G\mathrm{d}\mu=\mathrm{Vol}(G) \qquad (7)

where: $\mathrm{d}\mu(g)$ — the bi-invariant Haar measure; $f$ — any integrable function; $h$ — a fixed element. Reading it: for compact $G$ the total volume is finite, so averaging over the group is bounded — the source of unitarity. Almost every group a physicist meets is a closed subgroup of a matrix group, and Cartan's theorem guarantees these are automatically smooth manifolds with the standard dimensions.

\dim SU(n)=n^2-1,\quad \dim SO(n)=\tfrac{n(n-1)}{2},\quad \dim U(n)=n^2,\quad \dim Sp(n)=n(2n+1) \qquad (8)

where: $SU(n),SO(n),U(n)$ — classical compact matrix groups inside $GL(n,\mathbb{C})$ ; $Sp(n)$ — the compact symplectic group of rank $n$ , realised by $2n\times2n$ complex matrices (equivalently $n\times n$ quaternionic unitaries), so it sits inside $GL(2n,\mathbb{C})$ , not $GL(n,\mathbb{C})$ . Reading it: counting generators is counting gauge bosons — $\dim SU(3)=8$ gluons, $\dim SU(2)=3$ .

Derivation AFrom left-invariant vector fields to the Lie algebra

The goal is to show that the tangent space at the identity carries a canonical bracket, turning $T_eG$ into a finite-dimensional Lie algebra of dimension exactly $\dim G$ . The mechanism is homogeneity itself: left translation spreads a single tangent vector at $e$ into a smooth vector field over the whole manifold, and the commutator of two such fields, being natural under diffeomorphisms, stays inside the same special family — so it can be read back at the identity as a bracket on $T_eG$ . This is the construction that makes "study the group through its algebra" a theorem rather than a slogan.

Step 1. Given $X\in T_eG$ , define a vector field by pushing $X$ forward with left translation, $X^L_g=(\mathrm{d}L_g)_e\,X$ . Smoothness of $\mu$ makes $X^L$ a smooth field, and by construction it is left-invariant, $(L_h)_*X^L=X^L$ . Why: this converts one vector at a point into a global field determined entirely by its value at $e$ , so the space of such fields is a faithful copy of $T_eG$ .

Step 2. The Lie bracket of vector fields is natural under pushforward by a diffeomorphism, $(L_h)_*[U,V]=[(L_h)_*U,(L_h)_*V]$ . Applying this to two left-invariant fields gives $(L_h)_*[X^L,Y^L]=[X^L,Y^L]$ , so their bracket is again left-invariant. Why: this is the closure property — the bracket does not leak out of the finite-dimensional family, which is what allows a finite-dimensional algebra at all.

Step 3. Since a left-invariant field is fixed by its value at $e$ , define $[X,Y]:=[X^L,Y^L]\big|_e\in T_eG$ . Transporting the field bracket back to the identity yields a bracket on the finite-dimensional space $T_eG$ , of dimension $\dim G$ . Why: this is the identification $\{\text{left-invariant fields}\}\cong T_eG$ that gives the algebra its dimension.

Step 4. Bilinearity, antisymmetry $[X,Y]=-[Y,X]$ , and the Jacobi identity $[X,[Y,Z]]+[Y,[Z,X]]+[Z,[X,Y]]=0$ hold for the vector-field bracket and descend to $T_eG$ . Why: these are precisely the Lie algebra axioms, so $(\mathfrak{g},[\cdot,\cdot])$ is a genuine Lie algebra. *For matrix groups the abstract bracket becomes the commutator $XY-YX$ .*

Go deeper · Derivation B · 1 minThe exponential map and the one-parameter subgroup law

Derivation BThe exponential map and the one-parameter subgroup law

The aim is to prove that $\exp(tX)$ is the unique smooth homomorphism of the real line into $G$ with initial velocity $X$ , and hence that $\exp(sX)\exp(tX)=\exp((s+t)X)$ . The route is to recognise a one-parameter subgroup as the integral curve through the identity of the left-invariant field built from $X$ : homogeneity turns its defining condition into a first-order linear matrix ODE, whose solution is the matrix exponential and whose uniqueness delivers the group law for free — the calculation that makes finite transformations computable from generators.

Step 1. A one-parameter subgroup is the integral curve of $X^L$ through $e$ , so $\dot\gamma(t)=X^L_{\gamma(t)}=(\mathrm{d}L_{\gamma(t)})_e X$ . For a matrix group $(\mathrm{d}L_g)_eX=gX$ , giving the linear ODE $\dot\gamma=\gamma X$ with $\gamma(0)=I$ . Why: left invariance is exactly what makes the velocity at $\gamma(t)$ a fixed left-translate of $X$ , collapsing a manifold flow to a constant-coefficient matrix equation.

Step 2. The matrix exponential $\gamma(t)=\sum_{n\ge0}(tX)^n/n!$ solves $\dot\gamma=\gamma X$ , $\gamma(0)=I$ , and the series converges for all $t$ in operator norm since $\lVert(tX)^n/n!\rVert\le (|t|\,\lVert X\rVert)^n/n!$ . Why: convergence in norm licenses term-by-term differentiation, so the formal solution is a genuine, everywhere-defined curve on $G$ .

Step 3. Fix $s$ and compare $t\mapsto\gamma(s+t)$ with $t\mapsto\gamma(s)\gamma(t)$ . Both satisfy $\dot\alpha=\alpha X$ with the same initial value $\alpha(0)=\gamma(s)$ , so uniqueness forces $\gamma(s+t)=\gamma(s)\gamma(t)$ . Why: the homomorphism property is not assumed — it is squeezed out of ODE uniqueness, so it holds exactly, not approximately.

Step 4. This additivity is confined to a single direction. For independent $X,Y$ the product $\exp(X)\exp(Y)$ is governed by the BCH series, and equals $\exp(X+Y)$ if and only if $[X,Y]=0$ . Why: forgetting this caveat is the single most common error in the subject. *As a consistency check: when $G=\mathbb{R}$ this reduces to the familiar scalar identity $e^{s+t}=e^se^t$ , and the BCH residual at second order has been confirmed to be $X+Y+\tfrac12[X,Y]$ .*

Go deeper · Derivation C · 1 minWeyl's unitarian trick — why compact groups are so tame

Derivation CWeyl's unitarian trick — why compact groups are so tame

The objective is to prove that every finite-dimensional representation of a compact Lie group can be made unitary, and is therefore completely reducible into irreducibles. The idea is disarmingly simple once the Haar measure is in hand: average an arbitrary inner product over the whole group. Compactness makes the average finite and positive-definite, and invariance of the measure makes the group preserve it. This one step is why the orthogonality theorems, character tables, and multiplet decompositions of angular momentum and isospin all work — the sharpest single dividing line between compact and non-compact groups.

Step 1. Start with any Hermitian inner product $\langle\cdot,\cdot\rangle$ on the finite-dimensional carrier space $V$ of a representation $\rho:G\to GL(V)$ . Why: such a form always exists; the point is to repair its lack of invariance, not to invent it.

Step 2. Average it against the Haar measure, $\langle u,v\rangle_G=\int_G\langle\rho(g)u,\rho(g)v\rangle\,\mathrm{d}\mu(g)$ . The integral converges because $G$ is compact ( $\mathrm{Vol}(G)<\infty$ ) with continuous integrand, and stays positive-definite Hermitian. Why: finiteness of the total measure does all the work — on a non-compact group this integral generically diverges.

Step 3. For fixed $h$ , substitute $g\mapsto gh^{-1}$ and use right-invariance of $\mathrm{d}\mu$ : $\langle\rho(h)u,\rho(h)v\rangle_G=\int_G\langle\rho(gh)u,\rho(gh)v\rangle\,\mathrm{d}\mu(g)=\langle u,v\rangle_G$ . Hence $\rho(h)$ preserves the averaged form for every $h$ . Why: invariance of the measure is precisely what turns "true on average" into "true for every element", which is the unitarity we wanted.

Step 4. So $\rho$ is unitary in $\langle\cdot,\cdot\rangle_G$ . The orthogonal complement of any invariant subspace is then also invariant, so $V$ splits into irreducibles — complete reducibility. Why: this payoff fails for non-compact groups, whose nontrivial unitary representations must be infinite-dimensional. *When $G$ is finite and $\mathrm{d}\mu$ is the normalised counting measure, this argument recovers Maschke's theorem.*

Limiting cases

Which regime a symmetry lives in fixes, before any calculation, the available integration measure and the shape of its representation theory.

Case	Result	Meaning
Abelian / flat	$[X,Y]=0$ for all $X,Y\in\mathfrak{g}$	Groups like $U(1)$ and $(\mathbb{R}^n,+)$ commute; every irreducible unitary representation is a one-dimensional character $e^{i\alpha\theta}$ , and $\exp$ is a homomorphism onto the identity component.
Compact	$\mathrm{Vol}(G)=\int_G\mathrm{d}\mu<\infty$	$SO(n),SU(n),Sp(n)$ carry a finite bi-invariant Haar measure; every finite-dimensional representation is unitarizable and completely reducible, so orthogonality theorems hold.
Non-compact	$\mathrm{Vol}(G)=\infty$	The Lorentz group $SO(3,1)$ and translations have no finite invariant measure; their nontrivial unitary representations are infinite-dimensional.
Simply connected cover	$\pi_1(\tilde G)=\{e\}$	Each connected $G$ has a unique simply connected cover $\tilde G$ with the same algebra; every algebra representation integrates to $\tilde G$ , e.g. $SU(2)\to SO(3)$ with kernel $\mathbb{Z}_2$ .
Semisimple	Killing form non-degenerate	No abelian ideals; the Cartan classification ( $A_n,B_n,C_n,D_n$ plus five exceptionals) enumerates every simple compact group — the backbone of gauge and GUT model-building.

Common trapThe Lie algebra does not determine the group, and exp is neither onto nor a homomorphism

Three beliefs feel true and are false. First, the Lie algebra fixes only the local structure: $\mathfrak{su}(2)\cong\mathfrak{so}(3)$ yet $SU(2)$ (a simply connected 3-sphere) and $SO(3)$ ( $\mathbb{RP}^3$ , with $\pi_1=\mathbb{Z}_2$ ) are genuinely different groups — the algebra cannot see the double cover, the centre, or the fundamental group. Second, $\exp(X)\exp(Y)\ne\exp(X+Y)$ in general; the true exponent carries the BCH tail $\tfrac12[X,Y]+\cdots$ , and equality holds only when $X$ and $Y$ commute. Third, the exponential map is generally neither injective nor surjective: on $SL(2,\mathbb{R})$ it misses whole open sets, so "every group element is $\exp$ of a generator" is onto only for connected compact groups. And a trap of vocabulary — it is compactness, not mere finite dimensionality, that guarantees unitary finite-dimensional representations.

Worked examples

1 · The Lie algebra su(2), the double cover, and a finite rotation (fully worked)

From the Pauli matrices $\sigma_x,\sigma_y,\sigma_z$ define generators $T_a=\sigma_a/2$ . Verify the $\mathfrak{su}(2)$ brackets, compute $\exp(-i\theta\,\sigma_z/2)$ , and show that a $2\pi$ rotation acts as $-1$ on a spinor while $4\pi$ acts as $+1$ .

Step 1. From $\sigma_x\sigma_y=i\sigma_z$ (and cyclic), $[\sigma_x,\sigma_y]=2i\sigma_z$ , so $[T_x,T_y]=\tfrac14(2i\sigma_z)=iT_z$ , and cyclically $[T_i,T_j]=i\,\epsilon_{ijk}T_k$ . Restoring $\hbar$ via $J_a=\hbar T_a$ gives the angular-momentum algebra $[J_i,J_j]=i\hbar\,\epsilon_{ijk}J_k$ . The three $T_a$ close into $\mathfrak{su}(2)$ , confirming $\dim SU(2)=2^2-1=3$ .

Step 2. Exponentiate along $\hat z$ . Because $\sigma_z^2=I$ , the series splits into cosine and sine parts:

\exp\!\Big(-i\tfrac{\theta}{2}\sigma_z\Big)=\cos\tfrac{\theta}{2}\,I-i\sin\tfrac{\theta}{2}\,\sigma_z=\begin{pmatrix}e^{-i\theta/2}&0\\[2pt]0&e^{i\theta/2}\end{pmatrix},\qquad \exp\!\Big(-i\tfrac{\theta}{2}\hat n\cdot\boldsymbol{\sigma}\Big)=\cos\tfrac{\theta}{2}\,I-i\sin\tfrac{\theta}{2}\,(\hat n\cdot\boldsymbol{\sigma}).

The finite rotation is trigonometric in the half-angle $\theta/2$ — the tell-tale of the spin-½ representation.

Step 3. Evaluate at full turns. At $\theta=2\pi$ : $\cos\pi\,I-i\sin\pi\,\sigma_z=-I$ , a global sign flip. At $\theta=4\pi$ : $\cos2\pi\,I=+I$ . So $U(2\pi)=-I$ and $U(4\pi)=+I$ : the spinor changes sign under a $360°$ rotation and only returns after $720°$ .

Step 4. Interpret via the cover. In $SO(3)$ a $2\pi$ rotation is the identity, so $U(2\pi)=-I$ cannot be a function of the $SO(3)$ element alone; it is single-valued only on the double cover $SU(2)$ , where $2\pi$ and $0$ are distinct points — the content of the second figure and of the measured neutron-interferometry sign change. Half-integer spins live on $SU(2)$ , not on $SO(3)$ ; the double cover is what makes them consistent.

2 · Counting generators: gauge bosons of the Standard Model

The Standard Model gauge group is $SU(3)_c\times SU(2)_L\times U(1)_Y$ . Use the dimension formulas to count its gauge bosons, then contrast with the GUT candidates $SU(5)$ and $SO(10)$ .

Step 1. One gauge boson lives in each generator, i.e. each dimension of the algebra. From equation (8): $\dim SU(3)=3^2-1=8$ , $\dim SU(2)=2^2-1=3$ , $\dim U(1)=1$ . Total $8+3+1=12$ gauge bosons.

Step 2. Identify them. The 8 are the coloured gluons. The electroweak $3+1=4$ are the $W^1,W^2,W^3$ of $SU(2)_L$ and the $B$ of $U(1)_Y$ ; after spontaneous symmetry breaking these mix into $W^\pm,Z^0$ and the massless photon $\gamma$ . The 12 = 8 gluons + 4 electroweak bosons, exactly the observed force carriers.

Step 3. A grand-unified group must contain all 12 and more. $\dim SU(5)=5^2-1=24$ , so $SU(5)$ adds $24-12=12$ new gauge bosons — the $X,Y$ leptoquarks that would mediate proton decay. $\dim SO(10)=\tfrac{10\cdot9}{2}=45$ , larger still and able to place a full generation in one 16-dimensional spinor. $SU(5)$ has 24 generators, $SO(10)$ has 45; the extras beyond 12 are the prediction of unification.

Step 4. This is no coincidence: gauge bosons always transform in the adjoint representation, whose dimension equals $\dim G$ — for $SU(3)$ the octet ( $\dim 8$ ), for $SU(2)$ the triplet ( $\dim 3$ ). Gauge-boson count $=\dim(\text{adjoint})=\dim G$ .

Physics of the portrait

The Lie group is one of the rare structures named for the person who invented it — and he invented it not for physics but to do for differential equations what Galois had just done for polynomials.

Origins · 1 minSophus Lie and how this was discovered

Sophus Lie

1842–1899, Norwegian mathematician · the origin of continuous transformation groups

Sophus Lie was a late starter who became the founder of an entire branch of mathematics. Trained in Christiania (Oslo), he found his direction only around 1870 in Paris and Berlin, where — in conversation with Felix Klein — he conceived the idea that would occupy him for life: just as Galois had attached a finite group to a polynomial and read off solvability, one could attach a continuous group of transformations to a differential equation and read off its integrability from the symmetries leaving it invariant. Beginning in 1873 Lie built the theory of these continuous transformation groups, showing their infinitesimal generators close under a bracket — the object we now call the Lie algebra — and that finite transformations are recovered by exponentiating the infinitesimal ones. With Friedrich Engel he set it down in the three-volume Theorie der Transformationsgruppen (1888–1893). The classification that crowns the subject was not Lie's own: Wilhelm Killing enumerated the simple complex Lie algebras in 1888, and Élie Cartan's 1894 thesis made it rigorous and complete, with four infinite families and five exceptionals. The physical harvest came a generation later: Hermann Weyl's 1925 proof of complete reducibility and his character formula, Emmy Noether's theorem tying each continuous symmetry to a conservation law, and Eugene Wigner's classification of the Poincaré group's representations turned Lie's tool into the organising language of quantum mechanics, relativity, and the entire Standard Model — every gauge symmetry a Lie group, every conserved charge a generator of one.

History

1873

— Sophus Lie founds the theory of continuous transformation groups

1888

— Killing classifies the simple complex Lie algebras

1894

— Élie Cartan's thesis rigorously completes the classification

1897

— Hurwitz introduces invariant integration over a compact group

1925

— Weyl proves complete reducibility and the character formula for compact groups

1933

— Haar constructs the invariant measure on locally compact groups

Going further

Texts. Stone & Goldbart, Mathematics for Physics, Ch. 16 · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 4 · Hassani, Mathematical Physics, chapters on group theory · Nakahara, Geometry, Topology and Physics, §5 · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 10 · B. C. Hall, Lie Groups, Lie Algebras, and Representations, 2nd ed. · J. F. Cornwell, Group Theory in Physics, Vol. 1. Key anchor results: (i) $[T_i,T_j]=i\epsilon_{ijk}T_k$ from the Pauli matrices; (ii) $\exp(-i\tfrac{\theta}{2}\sigma_z)=\mathrm{diag}(e^{-i\theta/2},e^{i\theta/2})$ , giving $-I$ at $2\pi$ and $+I$ at $4\pi$ ; (iii) the BCH exponent to second order is $X+Y+\tfrac12[X,Y]$ ; (iv) $\dim SU(3)=8$ , $\dim SU(5)=24$ , $\dim SO(10)=45$ , and $8+3+1=12$ .

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

Ask the tutor: Ask me to exponentiate a specific generator — say a rotation axis in $SU(2)$ — and watch the algebra turn into a finite group element step by step.

What is the difference between a Lie group and its Lie algebra?

The Lie group $G$ is the full curved space of finite transformations — all the rotations, phases, and boosts. The Lie algebra $\mathfrak{g}$ is the flat tangent space at the identity, the infinitesimal generators, equipped with the bracket $[X,Y]$ . They are linked by the exponential map $\exp:\mathfrak{g}\to G$ . The algebra is vastly easier to work with — a finite-dimensional vector space, so questions about it are linear algebra — and it captures the group completely near the identity. What it cannot capture is global topology: the algebras of $SU(2)$ and $SO(3)$ are identical, but the groups differ. The working slogan: compute in the algebra, then check the topology of the group.

Why is homogeneity — left translation being a diffeomorphism — such a big deal?

Because it means the group looks the same near every point, so all the local geometry is concentrated at one place, the identity. Without homogeneity you would study the manifold neighbourhood by neighbourhood, as for a generic curved surface. With it, one tangent space and one bracket determine the local law everywhere, and every left-invariant vector field is fixed by a single value at $e$ . This is why the finite-dimensional Lie algebra exists and why differential equations on it replace functional equations on the group — the reason a continuous symmetry with infinitely many elements can be pinned down by a finite table of structure constants.

If SU(2) and SO(3) have the same Lie algebra, why do we need both?

Because half-integer spin lives on $SU(2)$ and cannot live on $SO(3)$ . $SO(3)$ is the group of actual rotations of space, but it is doubly connected: a full $2\pi$ rotation is a loop that cannot be shrunk to a point. $SU(2)$ is the simply connected double cover — $SO(3)$ with that loop "unwound", so $2\pi$ and $0$ become different elements. A spin-½ state picks up a sign under $2\pi$ and only returns after $4\pi$ : single-valued on $SU(2)$ , double-valued on $SO(3)$ . Nature uses the cover — electrons, quarks, and neutrinos are spinors — and the same relationship recurs one level up as $\mathrm{Spin}(n)\to SO(n)$ .

Is the exponential map always onto — can every group element be written as exp of a generator?

No, and this is a common trap. For a connected compact group the exponential map is surjective, so there every element is $\exp(X)$ for some $X$ . But for non-compact groups it can fail badly: on $SL(2,\mathbb{R})$ whole open regions are not the exponential of any single algebra element, and you need a product of two exponentials to reach them. It can also fail to be injective — on $U(1)$ many generators map to the same element because $\exp$ wraps the line around the circle. So " $\exp$ recovers a neighbourhood of the identity" is always true, but " $\exp$ recovers the whole group" is special to connected compact groups.

What is the Haar measure and why does compactness matter so much?

The Haar measure is the unique (up to scale) way to integrate over a Lie group invariant under group multiplication — the analogue of the flat $\mathrm{d}x$ invariant under translation. Its decisive property is total volume. For a compact group like $SU(2)$ the volume is finite, so you can average any quantity over the group and get a finite answer; that single fact powers the unitarian trick, the orthogonality of characters, and complete reducibility. For a non-compact group like the Lorentz group the volume is infinite, averaging diverges, and all of that machinery breaks — which is why its unitary representations are infinite-dimensional. Compactness is what makes representation theory finite and tame.

How do Lie groups connect to conservation laws in physics?

Through Noether's theorem: every continuous symmetry of a system's action corresponds to a conserved quantity, and a continuous symmetry is precisely a one-parameter subgroup $\exp(tX)$ of a Lie group. Time-translation symmetry gives energy conservation; spatial translation gives momentum; rotational symmetry, the group $SO(3)$ , gives angular momentum, with the generators $J_x,J_y,J_z$ becoming the conserved components. In the Hamiltonian picture the generators become phase-space functions whose Poisson brackets reproduce the Lie bracket, so the conserved charges literally realise $\mathfrak{g}$ . The unifying statement is that the conserved charges of physics are the generators of its symmetry Lie groups.