The Lie Bracket and Commutator of Vector Fields

Two vector fields on a manifold are two rival sets of marching orders: at every point, $X$ says "step this way" and $Y$ says "step that way." The Lie bracket answers a deceptively simple question — does the order in which you obey them matter? Walk a tiny bit along $X$ , then along $Y$ , then retrace along $X$ , then along $Y$ , and on a generic field of arrows you do not arrive back where you started. The residual gap, at second order in the step size, is a brand-new vector field $[X,Y]$ , the commutator of $X$ and $Y$ . It is the infinitesimal measure of how badly two flows fail to commute. There is a purely algebraic face of the same object. Each vector field acts on functions as a directional-derivative operator, and the composition $X(Yf)$ is a second-order differential operator — too many derivatives to be a vector field. The magic is that the antisymmetric combination $X(Yf)-Y(Xf)$ has its second-derivative terms cancel identically, because mixed partials commute. What survives is a first-order derivation: a genuine vector field. The bracket therefore turns the infinite-dimensional space of smooth vector fields into a Lie algebra — bilinear, antisymmetric, and obeying the Jacobi identity — the Lie algebra of the diffeomorphism group of the manifold. This single construction is the hidden engine under a large fraction of the differential geometry used in relativity. It is the Lie derivative acting on vectors, $\mathcal{L}_X Y=[X,Y]$ ; it is the obstruction in the Frobenius integrability theorem; it is the correction term that makes the curvature operator tensorial; and it is the source of the structure constants of every symmetry algebra, from the Poincaré algebra of Minkowski space to the anholonomy coefficients of a tetrad. Master the bracket and the rest of the machinery clicks into place.

Key takeaways

The Lie bracket is the commutator of directional-derivative operators, $[X,Y]f=X(Yf)-Y(Xf)$ , and it is again first-order — hence a bona fide vector field — because antisymmetrization annihilates the second-derivative terms.
In coordinates $[X,Y]^\mu=X^\nu\partial_\nu Y^\mu-Y^\nu\partial_\nu X^\mu$ ; the symmetric $X^\nu Y^\mu\partial_\nu\partial_\mu f$ pieces cancel, which is exactly why the result transforms as a vector.
Geometrically $[X,Y]$ is the gap left when you flow along $X$ , then $Y$ , then back along $X$ , then back along $Y$ : the leading non-closure is $\epsilon^2[X,Y]$ , so flows commute if and only if the bracket vanishes.
The bracket is $\mathbb{R}$ -bilinear, antisymmetric, and satisfies the Jacobi identity, making $\mathfrak{X}(M)$ an infinite-dimensional Lie algebra — the Lie algebra of $\mathrm{Diff}(M)$ .
It is not $C^\infty(M)$ -linear in its arguments: $[fX,gY]=fg[X,Y]+f(Xg)Y-g(Yf)X$ , and this failure is precisely what a connection later repairs.
The bracket equals the Lie derivative $\mathcal{L}_X Y=[X,Y]$ and underlies Killing symmetries, Frobenius integrability, torsion, and the curvature operator throughout general relativity.
Key idea: the commutator $[X,Y]f=X(Yf)-Y(Xf)$ is a vector field because antisymmetrization kills the second-derivative terms; geometrically it is the infinitesimal failure of the flows of $X$ and $Y$ to commute.

Where this lives today

The bracket is not a museum piece — it is executed billions of times a second inside modern general-relativity codes. In the 3+1 (ADM/BSSN) formulation used for simulating black-hole and neutron-star mergers, spacetime is sliced into spatial hypersurfaces and every field is evolved by a Lie derivative along a chosen time-flow vector, split into lapse and shift. The popular Gamma-driver and 1-plus-log gauge conditions that keep a moving-puncture simulation stable are literally evolution equations built from Lie derivatives, and $\mathcal{L}_\beta$ along the shift vector $\beta$ appears in essentially every evolved quantity. When LIGO and Virgo compare data against numerical waveform templates, those templates were computed by machinery in which the bracket is a load-bearing beam. Elsewhere in gravitation, the algebra of Killing vectors — closed under the bracket — is what reduces the Einstein equations of a symmetric spacetime to a tractable system, and the structure constants of the spatial homogeneity group classify the Bianchi cosmologies. The same mathematics escapes gravity entirely into geometric control theory, where it has a charmingly tangible avatar: parallel parking. A car cannot slide directly sideways — "drive forward" and "steer" are its two available vector fields — yet by alternating them (forward, turn, back, turn) a driver produces a net sideways motion. That sideways displacement is exactly a nonzero Lie bracket of the two control fields, and the Chow–Rashevskii theorem guarantees that as long as iterated brackets span all directions, any configuration is reachable. Robot motion planners, spacecraft attitude controllers, and the design of quantum gates all exploit this "steering by commutators," making Lie's nineteenth-century bracket a working tool in twenty-first-century engineering.

The physical picture

Picture two vector fields as two independent autopilots. Following $X$ for a moment nudges you one way; following $Y$ nudges you another. If the two autopilots were a coordinate grid — "east" and "north" on a flat map — the order would never matter: go east then north, or north then east, and you reach the same corner. But a general pair of fields twists relative to each other, and then the four-step loop leaves you displaced. That displacement, per unit area of the loop, is the Lie bracket. It is a local quantity built entirely from how the fields and their derivatives sit at one point, yet it controls a global fact: whether the two flows can ever be simultaneously straightened into coordinate lines.

Figure

Flow along X then Y then back along X then back along Y fails to close: the residual gap equals ε² times the Lie bracket of X and Y — here exactly ε²∂_y for the fields X = ∂_x and Y = x∂_y

The algebraic side of this picture is where the "vector field, not operator" miracle happens. Applying $Y$ and then $X$ to a function produces a second-order operator: schematically $X(Yf)$ contains a piece $X^\nu Y^\mu\,\partial_\nu\partial_\mu f$ . A vector field is only allowed one derivative, so $X\!\circ\!Y$ on its own is disqualified. But in the difference $X(Yf)-Y(Xf)$ the two second-order pieces are $X^\nu Y^\mu\,\partial_\nu\partial_\mu f$ and $Y^\nu X^\mu\,\partial_\nu\partial_\mu f$ ; because $\partial_\nu\partial_\mu=\partial_\mu\partial_\nu$ , relabelling dummy indices shows they are identical and subtract to zero. Only first derivatives of the components survive, and those assemble into a legitimate vector field. Antisymmetrization is not a cosmetic choice — it is the surgical operation that removes exactly the illegal derivative.

Figure

The same construction with numbers: starting at the point 1, 1 with ε = 0.6, the four flows of X = ∂_x and Y = x∂_y land at 1, 1.36, a gap of exactly Δy = ε² = 0.36 in the ∂_y direction

A coordinate basis is the degenerate case that anchors the whole story. Because mixed partials commute, $[\partial_\mu,\partial_\nu]=0$ always — the coordinate autopilots close their loop perfectly. Turn this around and it becomes a powerful diagnostic: if you can find two fields whose bracket does not vanish, no coordinate system can have both as basis vectors simultaneously. Non-commutativity is a genuine geometric obstruction, and the size of the bracket measures exactly how far a chosen frame is from being a holonomic coordinate grid.

The core equations

The definition is operator-algebraic: the bracket is the commutator of the two directional-derivative operators. Read the right-hand side as "apply $Y$ , then $X$ , then subtract the reverse order." Everything else — the coordinate formula, the flow interpretation, the Lie-algebra axioms — is a consequence of this one line.

[X,Y]f \;=\; X(Yf)-Y(Xf) \qquad (1)

where: $X,Y$ — smooth vector fields, each acting as a first-order derivation on functions; $f$ — a smooth function $f\in C^\infty(M)$ ; $Xf$ — the directional derivative of $f$ along $X$ ; $[X,Y]$ — the Lie bracket, itself a vector field. The physical reading: the bracket is the leftover derivative that a round-trip of the two operators cannot undo. Expanding the definition in any chart gives the working formula. Notice there is no metric and no connection anywhere — the bracket is defined on a bare smooth manifold, using only partial derivatives of the components.

[X,Y]^\mu \;=\; X^\nu\,\partial_\nu Y^\mu \;-\; Y^\nu\,\partial_\nu X^\mu \qquad (2)

where: $X^\mu,Y^\mu$ — components in a coordinate basis, $X=X^\mu\partial_\mu$ ; $\partial_\nu=\partial/\partial x^\nu$ ; $[X,Y]^\mu$ — the $\mu$ -component of the bracket field. Physical reading: each field differentiates the other's components; the two terms are swapped copies, which is the source of the antisymmetry. The first two Lie-algebra axioms are immediate from the definition: swapping $X$ and $Y$ flips the sign, and the operation is linear over the real numbers in each slot.

[X,Y]=-[Y,X],\qquad [aX+bY,Z]=a[X,Z]+b[Y,Z] \qquad (3)

where: $a,b$ — real constants; $X,Y,Z$ — smooth vector fields. Physical reading: antisymmetry forces $[X,X]=0$ (a field's flow trivially commutes with itself); bilinearity is only over $\mathbb{R}$ , a caveat sharpened in equation (5). The third and deepest axiom is the Jacobi identity: the cyclic sum of double brackets vanishes. It is what upgrades an antisymmetric bilinear product into a Lie algebra.

\big[[X,Y],Z\big]+\big[[Y,Z],X\big]+\big[[Z,X],Y\big]=0 \qquad (4)

where: $X,Y,Z$ — smooth vector fields; the cyclic sum runs over the three fields. Physical reading: the identity is the infinitesimal shadow of the associativity of composing finite diffeomorphisms. The bracket is emphatically not linear over functions in its arguments — pull a function inside and extra derivative terms are spat out. This is the single most-forgotten property.

[fX,gY]=fg\,[X,Y]+f\,(Xg)\,Y-g\,(Yf)\,X \qquad (5)

where: $f,g$ — smooth functions; $Xg$ — derivative of $g$ along $X$ . Physical reading: because the bracket differentiates, rescaling a field by a function leaks new terms; setting $g=1$ gives the compact special case $[fX,Y]=f[X,Y]-(Yf)X$ . Acting on a vector field, the Lie derivative along $X$ is the bracket. This identifies the algebraic commutator with the geometric operation of dragging one field along the flow of the other.

\mathcal{L}_X Y \;=\; [X,Y] \qquad (6)

where: $\mathcal{L}_X$ — Lie derivative along $X$ ; $Y$ — the field being dragged along the flow of $X$ . Physical reading: the Lie derivative generalizes to tensors of any rank, but on vectors it collapses to the humble commutator. The dragging is made explicit by the flow (pushforward) limit, which is the coordinate-free definition of $\mathcal{L}_X Y$ and the bridge to the parallelogram picture.

\mathcal{L}_X Y=\lim_{t\to 0}\frac{1}{t}\Big[\big(\phi^X_{-t}\big)_* Y_{\phi^X_t(p)}-Y_p\Big] \qquad (7)

where: $\phi^X_t$ — flow of $X$ for parameter $t$ ; $(\phi^X_{-t})_*$ — pushforward (differential) of the inverse flow; $p$ — the evaluation point. Physical reading: carry $Y$ back along the flow of $X$ and compare it to the original $Y$ at $p$ ; the leading discrepancy per unit time is the bracket. Once a connection is introduced, the bracket relates to covariant derivatives through the torsion. For the Levi-Civita connection of general relativity the torsion vanishes, giving a formula worth memorizing.

[X,Y]=\nabla_X Y-\nabla_Y X - T(X,Y)\;\xrightarrow{\ T=0\ }\;\nabla_X Y-\nabla_Y X \qquad (8)

where: $\nabla_X Y$ — covariant derivative of $Y$ along $X$ ; $T$ — torsion tensor. Physical reading: for a torsion-free connection you may compute the bracket with $\nabla$ in place of $\partial$ — the extra Christoffel pieces cancel by symmetry — but the partial-derivative formula (2) is always valid, connection or not. In a non-coordinate (tetrad) frame the basis fields no longer commute, and their brackets define the anholonomy coefficients that feed the curvature in the tetrad formalism.

[e_a,e_b]=C^c{}_{ab}\,e_c,\qquad C^c{}_{ab}=-C^c{}_{ba} \qquad (9)

where: $\{e_a\}$ — a frame (tetrad) of vector fields; $C^c{}_{ab}$ — commutation (anholonomy / structure) coefficients. Physical reading: for a coordinate frame all $C^c{}_{ab}=0$ ; a nonzero $C$ signals a frame that cannot be integrated to coordinates, exactly as in equation (2).

Derivation AThe second derivatives cancel, leaving a genuine vector field

The goal of this derivation is to prove the foundational claim on which everything rests: that although $X$ and $Y$ are each first-order operators and their composition $X\!\circ\!Y$ is second-order, the antisymmetric combination $[X,Y]=X\!\circ\!Y-Y\!\circ\!X$ is again first-order and therefore represents a vector field. We do this two ways that reinforce each other — first abstractly by verifying the Leibniz rule (the defining property of a derivation), then concretely by expanding in a chart and watching the offending second-derivative terms annihilate.

Step 1. Recall that a vector field is precisely an $\mathbb{R}$ -linear map $C^\infty(M)\to C^\infty(M)$ satisfying the Leibniz product rule $V(fg)=f\,Vg+g\,Vf$ . Linearity of $[X,Y]$ is inherited from the linearity of $X$ and $Y$ , so the only nontrivial thing to check is Leibniz. Why: "being a vector field" is not about how an object is written but about how it acts on functions — Leibniz plus linearity is the complete test.

Step 2. Expand $X(Y(fg))$ using Leibniz for $Y$ first and then $X$ : one obtains $Xf\,Yg+f\,X(Yg)+Xg\,Yf+g\,X(Yf)$ . Writing the same for $Y(X(fg))$ simply swaps the roles of $X$ and $Y$ throughout. Why: pushing the product rule through the composition exposes the cross terms $Xf\,Yg$ and $Xg\,Yf$ , which are symmetric under $X\leftrightarrow Y$ and will therefore cancel in the difference.

Step 3. Subtract: $[X,Y](fg)=X(Y(fg))-Y(X(fg))=f\,[X,Y]g+g\,[X,Y]f$ . The four cross terms cancel in pairs, leaving exactly the Leibniz rule for $[X,Y]$ . Hence $[X,Y]$ is a derivation, i.e. a vector field. Why: this is the abstract proof — no coordinates, valid on any smooth manifold — that the commutator lives in $\mathfrak{X}(M)$ .

Step 4. Now the concrete version. In a chart, $X(Yf)=X^\nu\partial_\nu(Y^\mu\partial_\mu f)=X^\nu(\partial_\nu Y^\mu)\,\partial_\mu f+X^\nu Y^\mu\,\partial_\nu\partial_\mu f$ , and $Y(Xf)=Y^\nu(\partial_\nu X^\mu)\,\partial_\mu f+Y^\nu X^\mu\,\partial_\nu\partial_\mu f$ . Why: the product rule on the inner derivative manufactures one first-order piece and one second-order piece in each composition; we need to see the second-order pieces line up.

Step 5. Subtract. The second-order terms $X^\nu Y^\mu\,\partial_\nu\partial_\mu f$ and $Y^\nu X^\mu\,\partial_\nu\partial_\mu f$ are equal — relabel $\mu\leftrightarrow\nu$ in the second and use $\partial_\nu\partial_\mu=\partial_\mu\partial_\nu$ — and cancel exactly, leaving $[X,Y]f=(X^\nu\partial_\nu Y^\mu-Y^\nu\partial_\nu X^\mu)\,\partial_\mu f$ . This is equation (2). Why: the cancellation is not approximate; it is an identity flowing from the equality of mixed partials, the coordinate incarnation of the abstract Leibniz argument.

Go deeper · Derivation B · 2 minThe bracket as the gap of a flow parallelogram

Derivation BThe bracket as the gap of a flow parallelogram

The goal here is to make the geometric slogan precise: that the Lie bracket is the leading non-closure of the loop built from the four flows $\phi^X_\epsilon$ , $\phi^Y_\epsilon$ , $\phi^X_{-\epsilon}$ , $\phi^Y_{-\epsilon}$ . We evaluate a test function along the commutator path, Taylor-expand each leg to second order in the step $\epsilon$ , and watch every zeroth-, first-, and symmetric second-order term cancel, leaving a single antisymmetric remainder equal to $\epsilon^2[X,Y]f$ . This is the calculation behind both figures above.

Step 1. Expand $f$ along a single $X$ -flow: $f(\phi^X_\epsilon(p))=f+\epsilon\,Xf+\tfrac12\epsilon^2\,X(Xf)+O(\epsilon^3)$ , all evaluated at $p$ . The operator $X$ appears because differentiating along the integral curve of $X$ is exactly applying $X$ . Why: this is the engine of the whole expansion — flowing a distance $\epsilon$ is, to each order, repeated application of the generating field.

Step 2. Compose the forward legs. Applying the $Y$ -expansion to the result of the $X$ -flow gives $f(\phi^Y_\epsilon\phi^X_\epsilon(p))=f+\epsilon(Xf+Yf)+\tfrac12\epsilon^2\big(X(Xf)+2X(Yf)+Y(Yf)\big)+O(\epsilon^3)$ . The cross term carries $X$ acting first, so it correctly reads $X(Yf)$ . Why: operator ordering is the entire subtlety — the cross term is where the non-commutativity will eventually surface, so its ordering must be tracked faithfully.

Step 3. Append the return legs by repeating with $\epsilon\to-\epsilon$ for $\phi^X_{-\epsilon}$ and $\phi^Y_{-\epsilon}$ , keeping the operators in composition order. The first-order terms $\epsilon(Xf+Yf)$ from the forward legs are exactly undone by the $-\epsilon$ legs. Why: a commutator loop is designed so that all the "there and back" pieces cancel; only the ordering-sensitive residue can survive.

Step 4. Collect order $\epsilon^2$ . The symmetric self-terms $X(Xf)$ and $Y(Yf)$ cancel between forward and return legs, and what remains is $f(c(\epsilon))-f(p)=\epsilon^2\big(X(Yf)-Y(Xf)\big)+O(\epsilon^3)=\epsilon^2\,[X,Y]f\big|_p+O(\epsilon^3)$ . Why: this is the payoff — the only uncancelled piece is precisely the commutator, so the gap of the loop is $\epsilon^2[X,Y]$ .

Step 5. Read off the geometry. The displacement of the open parallelogram is $[X,Y]$ to leading order, and it vanishes to this order if and only if $[X,Y]=0$ ; a theorem (commuting flows) promotes this to the exact statement $\phi^X_t\circ\phi^Y_s=\phi^Y_s\circ\phi^X_t$ . Why: the worked example $X=\partial_x,\ Y=x\partial_y$ shows the coefficient is not merely leading-order but can be exact — there the gap is precisely $\epsilon^2\partial_y$ with no higher corrections.

Go deeper · Derivation C · 1 minThe Jacobi identity from operator associativity

Derivation CThe Jacobi identity from operator associativity

The goal is to prove the Jacobi identity (4) with essentially no computation beyond bookkeeping, by exploiting the fact that on functions the bracket is the plain operator commutator $[X,Y]=XY-YX$ , where juxtaposition means composition. Composition of operators is associative, and associativity is the only ingredient needed — which is why the Jacobi identity holds for every associative algebra's commutator and is not special to vector fields. We expand each double bracket into a signed sum of triple products and show they cancel in pairs.

Step 1. Expand the first term: $[[X,Y],Z]=(XY-YX)Z-Z(XY-YX)=XYZ-YXZ-ZXY+ZYX$ . Every term is a composition of the three operators in some order. Why: turning brackets into ordered triple products converts an abstract identity into a finite counting problem over the six permutations of $X,Y,Z$ .

Step 2. Cycle $X\to Y\to Z\to X$ to get the second term: $[[Y,Z],X]=YZX-ZYX-XYZ+XZY$ . Why: the cyclic structure of the identity means the remaining terms are obtained by relabelling, not by new work.

Step 3. Cycle once more for the third term: $[[Z,X],Y]=ZXY-XZY-YZX+YXZ$ . Why: all three double brackets are now written in the common basis of the six orderings $XYZ,XZY,YXZ,YZX,ZXY,ZYX$ .

Step 4. Add the three lines. Each of the six orderings appears exactly twice — once with a plus sign and once with a minus — so the entire cyclic sum collapses to zero. Why: this is the whole proof; the identity is a consequence of associativity alone, confirming that $\mathfrak{X}(M)$ with the bracket is a Lie algebra.

Limiting cases

The special and limiting situations sharpen what the bracket actually detects.

Case	Result	Meaning
Coordinate basis	$[\partial_\mu,\partial_\nu]=0$	mixed partials commute, so coordinate autopilots always close their loop; a holonomic frame has zero anholonomy
Self-bracket	$[X,X]=0$	antisymmetry forces it; a field's own flow commutes with itself
Commuting flows	$[X,Y]=0\iff\phi^X_t\circ\phi^Y_s=\phi^Y_s\circ\phi^X_t$	vanishing bracket is exactly equivalent to commuting one-parameter flows, locally where the flows are defined
Common coordinates	$[X,Y]=0$ , $X,Y$ independent $\Rightarrow$ $X=\partial_s,\ Y=\partial_t$	independent commuting fields can be simultaneously straightened into a coordinate grid (canonical-form theorem)
Function rescaling	$[fX,gY]=fg[X,Y]+f(Xg)Y-g(Yf)X$	the bracket is $\mathbb{R}$ -bilinear but not $C^\infty(M)$ -bilinear; the extra terms are its differentiating nature showing through
Levi-Civita connection	$[X,Y]=\nabla_XY-\nabla_YX$	for zero torsion, partials may be replaced by covariant derivatives; the Christoffel terms cancel by symmetry

Common trapThe bracket is X^ν∂_νY^μ, and ∂ can always be replaced by ∇

Two traps in one. First, the components are not $[X,Y]^\mu=X^\nu\partial_\nu Y^\mu$ alone — you must subtract the swapped partner $Y^\nu\partial_\nu X^\mu$ . Dropping it is the single most common error, and it produces an object that is not even a vector (it fails the transformation law precisely because the second-derivative anomaly no longer cancels). Second, you cannot blindly swap $\partial_\nu$ for a covariant derivative $\nabla_\nu$ : the identity $[X,Y]=\nabla_XY-\nabla_YX$ holds only when the connection is torsion-free. For the Levi-Civita connection of general relativity that is fine and the Christoffel symbols cancel by their lower-index symmetry — but for a connection with torsion the two differ by exactly $-T(X,Y)$ . The partial-derivative formula (2), by contrast, needs no connection at all and is always correct.

Worked examples

1 · The flow loop that does not close (exact, not just leading order)

Take $X=\partial_x$ and $Y=x\,\partial_y$ on the plane. Compute the bracket algebraically, then verify it geometrically by composing the four flows explicitly for a finite step $\epsilon$ .

Step 1. Algebra first. With $X^x=1$ (others zero) and $Y^y=x$ (others zero), equation (2) gives $[X,Y]^\mu=X^\nu\partial_\nu Y^\mu-Y^\nu\partial_\nu X^\mu$ . The only surviving piece is $X^x\partial_x Y^y=\partial_x(x)=1$ in the $y$ -slot, since every derivative of $X$ 's constant components vanishes. Thus $[X,Y]=\partial_y$ — the fields do not commute.

Step 2. Now the flows. Solving $\dot{\mathbf r}=X$ gives $\phi^X_t(x,y)=(x+t,\,y)$ ; solving $\dot{\mathbf r}=Y$ gives $\phi^Y_s(x,y)=(x,\,y+sx)$ , since $\dot x=0$ and $\dot y=x$ stays constant along the $Y$ -flow. These are the two autopilots in closed form.

Step 3. Compose the loop starting at $p=(1,1)$ with $\epsilon=0.6$ . Flow $\phi^X_\epsilon$ : $(1,1)\to(1.6,1)$ . Flow $\phi^Y_\epsilon$ : $(1.6,1)\to(1.6,\,1+0.6\times1.6)=(1.6,1.96)$ . Flow $\phi^X_{-\epsilon}$ : $(1.6,1.96)\to(1.0,1.96)$ . Flow $\phi^Y_{-\epsilon}$ : $(1.0,1.96)\to(1.0,\,1.96-0.6\times1.0)=(1.0,1.36)$ .

Step 4. Read the gap. The loop starts at $(1,1)$ and ends at $(1,1.36)$ : a pure $y$ -displacement of $\Delta y=0.36$ . Compare to $\epsilon^2=0.6^2=0.36$ . The gap is exactly $\epsilon^2\partial_y=\epsilon^2[X,Y]$ , with no higher-order terms — a rare case where the leading-order formula is the whole answer because both flows are affine in the relevant coordinate. The bracket $[X,Y]=\partial_y$ is literally the residual displacement, $0.36$ in the $y$ -direction.

2 · Rotation generators close into the algebra so(3)

The vector fields generating rotations of $\mathbb{R}^3$ about the three axes are $L_x=y\partial_z-z\partial_y$ , $L_y=z\partial_x-x\partial_z$ , $L_z=x\partial_y-y\partial_x$ . Show they form a closed Lie algebra and find the structure constants.

Step 1. Compute $[L_x,L_y]f=L_x(L_yf)-L_y(L_xf)$ directly. Applying $L_x=y\partial_z-z\partial_y$ to $L_yf=zf_x-xf_z$ , and $L_y=z\partial_x-x\partial_z$ to $L_xf=yf_z-zf_y$ , generates several first- and second-order terms in each.

Step 2. Subtract and watch the second derivatives cancel. Every mixed-partial term — $yz\,f_{xz}$ , $xy\,f_{zz}$ , $z^2f_{xy}$ , $xz\,f_{zy}$ and their partners — appears once in each composition and annihilates, exactly as the general theorem guarantees. The first-order remainder is $y f_x-x f_y$ .

Step 3. Identify the result: $y\partial_x-x\partial_y=-(x\partial_y-y\partial_x)=-L_z$ . Hence $[L_x,L_y]=-L_z$ , and by cyclic symmetry $[L_y,L_z]=-L_x$ , $[L_z,L_x]=-L_y$ . Compactly, $[L_i,L_j]=-\epsilon_{ijk}L_k$ : the three fields close among themselves, so they span a 3-dimensional Lie algebra, namely $\mathfrak{so}(3)$ .

Step 4. Reconcile the sign with quantum mechanics. Physicists use the Hermitian operators $\hat L_i=-i\,L_i$ ; then $[\hat L_x,\hat L_y]=(-i)^2[L_x,L_y]=-(-L_z)=+i\hat L_z$ , recovering the familiar $[\hat L_i,\hat L_j]=i\epsilon_{ijk}\hat L_k$ . The geometric vector fields obey $[L_i,L_j]=-\epsilon_{ijk}L_k$ ; the factor of $-i$ that makes them Hermitian operators flips the sign to the textbook $+i\epsilon_{ijk}$ . The structure constants of $\mathfrak{so}(3)$ are read straight off the brackets.

Physics of the portrait

The bracket carries the name of the Norwegian who saw that continuous symmetry, differential equations, and this one antisymmetric product are the same subject viewed from three sides.

Origins · 2 minSophus Lie and how this was discovered

Sophus Lie

1842–1899 · founder of the theory of continuous transformation groups

Marius Sophus Lie was a late bloomer who became one of the nineteenth century's most original geometers. Born in 1842 in Nordfjordeid, Norway, he drifted after university until, around 1868, he discovered geometry and never looked back. A pivotal year in Paris in 1870, sharing ideas with the young Felix Klein, set the course of his life's work: the realization that the bewildering variety of methods for solving differential equations could be unified by studying the continuous groups of transformations that leave the equations invariant. Where Évariste Galois had tamed polynomial equations with finite groups, Lie set out to tame differential equations with continuous ones. His decisive insight was that a continuous group is controlled by its infinitesimal generators — vector fields — and that the commutator of two generators is again a generator. That bracket, closing the generators into what we now call a Lie algebra, is the linearized skeleton of the whole group; the Jacobi identity it satisfies is the infinitesimal echo of the group's associativity. With his collaborator Friedrich Engel he compiled the monumental three-volume Theorie der Transformationsgruppen (1888–1893), and though his relationship with Klein soured and a nervous breakdown shadowed his later years, the structures he built now pervade physics. Every conservation law read off a symmetry, every Killing vector of a spacetime, every gauge group of a field theory, and the anholonomy coefficients of a tetrad are all speaking Lie's language. He died in 1899, shortly after returning to a chair created for him in Kristiania (Oslo); the term "Lie algebra" for his bracket structure was coined by Hermann Weyl a generation later.

History

1866

— Jacobi's Vorlesungen über Dynamik, containing the cyclic bracket identity now named for him, is published posthumously

1870

— Sophus Lie, in Paris with Felix Klein, begins the theory of continuous transformation groups

1888

— Lie and Friedrich Engel publish the first volume of Theorie der Transformationsgruppen

1925

— Élie Cartan and Hermann Weyl complete the structure and representation theory of Lie algebras

1934

— Weyl coins the term Lie algebra for the bracket structure of infinitesimal generators

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), §9.6 and Box 9.2 (flows and the commutator) · Wald, General Relativity (1984), §B.1 and Appendix C · Carroll, Spacetime and Geometry (2004), §2.3 · Schutz, Geometrical Methods of Mathematical Physics (1980), Ch. 3 · Lee, Introduction to Smooth Manifolds (2nd ed., 2013), Ch. 8–9 (the definitive modern treatment of brackets and flows). Key results to keep in mind: the component formula $[X,Y]^\mu=X^\nu\partial_\nu Y^\mu-Y^\nu\partial_\nu X^\mu$ ; the exact flow-loop gap $\epsilon^2\partial_y$ for $X=\partial_x,\ Y=x\partial_y$ ; the $\mathfrak{so}(3)$ closure $[L_i,L_j]=-\epsilon_{ijk}L_k$ and its $-i$ conversion to the quantum $+i\epsilon_{ijk}$ ; and the six-term cancellation proving Jacobi. Structural facts: $[\partial_\mu,\partial_\nu]=0$ , the Poincaré algebra has dimension 10, and the maximal-symmetry dimension in $n$ dimensions is $n(n+1)/2$ .

Prerequisites. Tangent Vectors and the Tangent Space · Vector Fields and Integral Curves · Smooth Manifolds and Coordinate Charts · Tensor Algebra: Contraction, Symmetrization and Antisymmetrization. Leads to. The Tensor Transformation Law · General Tensors and the Tensor Product · The Metric Tensor and the Line Element.

Ask the tutor: Want to see the bracket in action — pick any two vector fields and I will compute $[X,Y]$ , trace the flow loop that fails to close, and tell you whether they can share a coordinate grid.

Why is X of Yf not a vector field, but the bracket applied to f is?

A vector field is allowed exactly one derivative — it must obey the Leibniz product rule. The composition $X(Yf)$ applies two derivatives in succession and contains a genuinely second-order piece $X^\nu Y^\mu\,\partial_\nu\partial_\mu f$ , which violates Leibniz, so it is a second-order operator, not a vector field. In the antisymmetric combination $X(Yf)-Y(Xf)$ those second-order pieces are identical (mixed partials commute, indices relabel) and cancel exactly, leaving only first derivatives of the components. That surviving first-order object satisfies Leibniz and so is a legitimate vector field. Antisymmetrization is the precise operation that deletes the illegal second derivative.

Does the Lie bracket depend on a metric or a connection?

No — and this is one of its most important features. The bracket is defined on a bare smooth manifold using only partial derivatives of components (equation 2); it needs no metric, no connection, no notion of length or parallel transport. This makes it the most primitive way to differentiate one vector field along another. When you do have a connection, the bracket relates to covariant derivatives by $[X,Y]=\nabla_XY-\nabla_YX-T(X,Y)$ , but that is a derived relationship, not the definition. The independence from extra structure is exactly why the bracket appears in the definitions of torsion and curvature: it provides a connection-free baseline against which the connection's own effects are measured.

What is the difference between the Lie bracket and the Lie derivative?

They coincide on vector fields and differ in scope. The Lie derivative $\mathcal{L}_X$ measures the change of any tensor field as it is dragged along the flow of $X$ ; it acts on functions, vectors, one-forms, and tensors of every rank. The Lie bracket is specifically the commutator of two vector fields. The bridge is the identity $\mathcal{L}_X Y=[X,Y]$ : when the Lie derivative acts on a vector field, it reduces to the bracket. On a function it reduces instead to the directional derivative, $\mathcal{L}_X f=Xf$ . So think of the bracket as the vector-on-vector special case of the more general Lie derivative, and note that both are built from flows and require no connection.

If the bracket vanishes, can I always find coordinates with X and Y as basis fields?

Locally yes, provided $X$ and $Y$ are linearly independent where you want the coordinates. Vanishing bracket is equivalent to the flows commuting, $\phi^X_t\circ\phi^Y_s=\phi^Y_s\circ\phi^X_t$ , and commuting flows can be used as coordinate lines: set $(s,t)$ to be the flow parameters and the fields become $\partial_s$ and $\partial_t$ . This is the canonical-form (commuting-flows) theorem, a special case of Frobenius. Two caveats: it is a local statement — global obstructions can prevent extending the chart — and the fields must stay independent, since at points where they become parallel the "grid" degenerates. The result generalizes: $k$ pointwise-independent, pairwise-commuting fields can be simultaneously straightened into $k$ coordinate directions.

What are structure constants and anholonomy coefficients?

When you pick a frame of basis vector fields $\{e_a\}$ that is not a coordinate basis, the basis fields generally fail to commute, and their brackets expand back in the same frame: $[e_a,e_b]=C^c{}_{ab}\,e_c$ . The coefficients $C^c{}_{ab}$ are called commutation or anholonomy coefficients, and they are antisymmetric in the lower pair. For a coordinate (holonomic) frame they all vanish, recovering $[\partial_\mu,\partial_\nu]=0$ ; a nonzero value signals a frame that cannot be integrated into coordinates. When the frame is a set of group-invariant fields — as for the isometry group of a homogeneous space — these coefficients are constant and are precisely the structure constants of the group's Lie algebra. In the tetrad formulation of general relativity they feed directly into the connection coefficients and hence the curvature.

Why does the Jacobi identity matter physically?

The Jacobi identity is what makes $\mathfrak{X}(M)$ a Lie algebra rather than just an antisymmetric product, and Lie algebras are the linearizations of symmetry groups. Physically the identity is the infinitesimal shadow of the associativity of composing finite transformations: because doing diffeomorphisms in sequence is associative, their infinitesimal generators must satisfy Jacobi. Its consequences are everywhere — it guarantees that the bracket of two Killing vectors is again a Killing vector (so symmetries close into an algebra), it underlies the consistency of gauge transformations in field theory, and in Hamiltonian mechanics the same identity for Poisson brackets guarantees that conserved quantities close into an algebra. Remarkably, it costs nothing to prove: it follows from the associativity of operator composition alone.

Can the bracket of two vectors point outside the plane they span?

Yes, and that is precisely the interesting case. At a single point $X$ and $Y$ are just two tangent vectors spanning (at most) a 2-plane, but the bracket depends on their derivatives in a neighbourhood, so $[X,Y]$ can point in a direction transverse to that plane. Worked example 1 shows it vividly: $\partial_x$ and $x\partial_y$ at the point $(1,1)$ span the whole plane already, but the bracket $\partial_y$ is a definite new direction generated by how the fields twist. This transversality is the whole content of the Frobenius theorem: a distribution of planes is tangent to a family of surfaces if and only if the bracket of any two fields lying in the distribution stays inside it. When brackets escape the distribution, the planes cannot be knitted into surfaces.