Vector Fields and Integral Curves

The abstract machinery of fields, flows, and brackets is doing concrete work at the frontiers of gravitational physics. In the singularity theorems of Penrose and Hawking, a congruence of timelike or null geodesics is an integral-curve family of a four-velocity field, and its expansion — the divergence $\nabla_\mu U^\mu$ that opens the Raychaudhuri equation — is what the theorems drive to $-\infty$ in finite affine parameter to prove that a smooth spacetime must end: incompleteness of a flow, made physical. In numerical relativity, the codes that produced the waveform templates matched against LIGO–Virgo–KAGRA detections integrate the Einstein equations along the flow of a chosen time-evolution vector field, and the freedom to reshape that field (the lapse and shift) is precisely the freedom to reparametrize a flow. When the Event Horizon Telescope reconstructs the shadow of M87 or Sgr A, the ray-tracing that generates each image integrates the geodesic field backward through Kerr spacetime — millions of integral curves threading a curved manifold. And the two continuous symmetries of a stationary, axisymmetric black hole, the timelike and rotational Killing fields, are what make its energy and angular momentum well-defined conserved charges in the first place. Lie's arrows are still steering the calculation.

Picture a steady wind blowing across a plain. At every point there is exactly one velocity arrow, and the whole pattern of arrows is a vector field. Release a dust grain and it drifts along, always tangent to the local arrow — the path it carves is an integral curve. The instruction "let the wind carry every point downstream for time $t$" is a map of the plain to itself, the flow $\phi_t$, and because the wind does not change, drifting for time $s$ and then time $t$ is the same as drifting for $s+t$: the flows compose by adding their parameters. That single sentence is the group law $\phi_s\circ\phi_t=\phi_{s+t}$, and it is why $\{\phi_t\}$ is called a one-parameter group. The first figure draws a concrete field whose integral curves you can solve in your head.

Acting on functions, a vector field is a directional-derivative machine: $X(f)$ is the rate of change of $f$ as you drift along the arrows, $\mathrm{d}(f\circ\gamma)/\mathrm{d}t$ evaluated on the integral curve. This is why "vector field as differential operator" and "vector field as flow" are two faces of one coin — differentiating $f\circ\phi_t$ in $t$ returns $X(f)$ pushed along the flow. The subtlety that separates flat intuition from curved-space rigour appears the moment you have two fields. Flow along $X$, then along $Y$, then back along $X$, then back along $Y$: you do not return to where you started. The leftover displacement, at second order in the step, is exactly the Lie bracket $[X,Y]$ — and it vanishes precisely when the flows commute, which is why coordinate basis fields $\partial_\mu$ (whose flows are pure coordinate shifts) always close the loop. The second figure makes this non-closure concrete.

Globally, an integral curve is the orbit through its starting point. If every orbit is defined for all $t\in\mathbb{R}$ the field is complete and $\phi_t$ is a genuine action of the real line; otherwise an orbit can reach the edge of the manifold or run to infinity in finite parameter, and the flow is only locally defined. Completeness is guaranteed on a compact manifold and often fails on an open one — a distinction with real teeth in general relativity, where the incompleteness of geodesic congruences is the very definition of a singular spacetime.

A vector field is specified, in any chart, by its components — smooth functions multiplying the coordinate basis fields. Everything computational flows from this single expansion.

where: $X$ — the vector field on $M$; $X^\mu(x)$ — smooth component functions in the chart $\{x^\mu\}$; $\partial/\partial x^\mu$ — the coordinate basis fields, a frame for $TM$ over the chart. Reading: the same object exists without coordinates, but a chart turns it into $n$ ordinary functions you can differentiate and integrate. Because each $\partial_\mu$ is a derivation, so is $X$: it is a linear map on smooth functions obeying the product rule. This is the algebraic definition of a vector field, and it needs no coordinates at all.

where: $f,g$ — smooth functions in $C^\infty(M)$; $a,b$ — real constants; $X(f)$ — the directional derivative of $f$ along $X$, itself a smooth function. Reading: linearity plus Leibniz is the definition; conversely every such derivation of $C^\infty(M)$ is a unique vector field. The components are not invariant — they carry an upper index and transform contravariantly, with the Jacobian of the coordinate change, precisely so that the geometric object $X$ is chart-independent.

where: $X^{\mu'}$ — components in a second chart $\{x^{\mu'}\}$; $\partial x^{\mu'}/\partial x^\nu$ — the Jacobian of the transition map. Reading: the arrow stays fixed while its numerical description rotates with the coordinates — the hallmark of a genuine tensor of type $(1,0)$. The dynamical content is the integral curve: a curve whose velocity equals the field at every point it passes through. This first-order autonomous system is the equation of motion of the flow.

where: $\gamma(t)$ — the integral curve through $p$; $t$ — the curve parameter (flow time); $p$ — the initial point. Reading: this is $n$ coupled ODEs for the $n$ functions $\gamma^\mu(t)$, and its solution threads the arrows into a streamline. Collecting the solutions as the start point varies defines the flow map $\phi_t$. Autonomy — no explicit $t$-dependence in $X$ — makes the flow a one-parameter group.

where: $\phi_t(p)$ — the point reached by flowing $p$ for parameter $t$; $\circ$ — composition of diffeomorphisms. Reading: each $\phi_t$ is invertible with inverse $\phi_{-t}$, so the flow is a local group of diffeomorphisms generated by $X$. Differentiating a function along the flow recovers the derivation: the flow and the operator are the same object seen dynamically versus algebraically.

where: $f$ — a smooth function; $X(f)$ — its directional derivative along $X$; at $t=0$ this is $X_p(f)$. Reading: the field is the infinitesimal generator of the flow, exactly as the Hamiltonian generates time translation. Two fields acting in succession give a second-order operator, but the commutator is again first order — the Lie bracket, a new vector field with a clean component formula.

where: $[X,Y]$ — the commutator vector field; $X,Y$ — two smooth fields acting on $f$ in order. Reading: the second-derivative terms cancel by equality of mixed partials, leaving the first-order operator on the right — the bracket is genuinely a vector field. The bracket is antisymmetric and satisfies the Jacobi identity, making the space of fields a Lie algebra; and it is precisely the obstruction to commuting flows, which is the same statement as the vanishing of the Lie derivative of $Y$ along $X$.

where: $\mathcal{L}_X Y$ — the Lie derivative of $Y$ along $X$, the rate at which the flow of $X$ drags $Y$; $\phi^X_s,\phi^Y_t$ — the two flows. Reading: commuting fields can be straightened into simultaneous coordinate directions, and non-commuting fields cannot — the criterion behind frame fields and Frobenius integrability.

1 · The rotation field and its flow (fully numeric)

Take $X=-y\,\partial_x+x\,\partial_y$ on $\mathbb{R}^2$. Find the integral curve through $(1,0)$, identify the flow $\phi_t$, evaluate it at $t=\pi/2$ and $t=\pi/3$, and verify the group law.

Step 1. Write out the integral-curve ODEs from $\dot\gamma=X$: $\dot x=-y$ and $\dot y=x$, with $(x,y)(0)=(1,0)$. Differentiating once more, $\ddot x=-\dot y=-x$, the simple-harmonic equation. Its solution with the given start is $x(t)=\cos t$, $y(t)=\sin t$. The integral curve is the unit circle traced counterclockwise at unit speed.

Step 2. The same computation from any start $(x_0,y_0)$ gives $\phi_t(x_0,y_0)=(x_0\cos t-y_0\sin t,\ x_0\sin t+y_0\cos t)$ — rigid rotation by angle $t$. So $\phi_t$ is rotation, and the field magnitude $|X|=\sqrt{x^2+y^2}=1$ on the unit circle means $t$ equals arc length there.

Step 3. Evaluate: $\phi_{\pi/2}(1,0)=(\cos 90^\circ,\sin 90^\circ)=(0,1)$, a quarter turn; and $\phi_{\pi/3}(1,0)=(\cos 60^\circ,\sin 60^\circ)=(0.500,\ 0.866)$. A tracer at $(1,0)$ sits at $(0,1)$ after parameter $\pi/2$.

Step 4. Check the group law by composing two rotations: $\phi_s\circ\phi_t$ is rotation by $s$ after rotation by $t$, i.e. rotation by $s+t=\phi_{s+t}$, because rotation angles simply add. The flow is a one-parameter group, and since every orbit is defined for all $t$ the field is complete — as it must be, being the Killing field generating rotational symmetry of the Euclidean metric.

2 · A Lie bracket and a loop that fails to close (fully numeric)

Let $X=\partial_x$ and $Y=x\,\partial_y$ on $\mathbb{R}^2$. Compute $[X,Y]$, then run the commutator loop of parameter $t$ starting at the origin and measure the gap.

Step 1. Use the component formula $[X,Y]^\mu=X^\nu\partial_\nu Y^\mu-Y^\nu\partial_\nu X^\mu$ with $X=(1,0)$ and $Y=(0,x)$. The first term is $\partial_x(0,x)=(0,1)$; the second is $x\,\partial_y(1,0)=(0,0)$. Hence $[X,Y]=(0,1)=\partial_y\neq 0$: these flows do not commute.

Step 2. The flows are exact here. $\phi^X_s(x,y)=(x+s,\,y)$ shifts along $x$; and $Y=x\partial_y$ gives $\dot x=0,\dot y=x$, so $\phi^Y_s(x,y)=(x,\,y+xs)$ shifts $y$ at a rate equal to $x$. These closed forms let us track the loop without any Taylor approximation.

Step 3. Start at $p=(0,0)$ and apply the four legs of parameter $t$ in turn. Flow $X$: $(t,0)$. Flow $Y$: $(t,\ 0+t\cdot t)=(t,t^2)$. Flow $-X$: $(0,t^2)$. Flow $-Y$: $(0,\ t^2+0\cdot(-t))=(0,t^2)$. The loop ends at $(0,t^2)$, not the origin — a gap of $t^2$ in the $y$ direction.

Step 4. Compare with the theorem: the gap should be $t^2[X,Y]=t^2\,\partial_y$, i.e. $(0,t^2)$ — an exact match, and here the $O(t^3)$ corrections vanish identically because the flows are affine. At $t=0.5$ the loop misses closure by $0.5^2=0.25$ in $y$, the concrete number on the figure above.

3 · An incomplete field that escapes in finite parameter (fully numeric)

Consider $X=x^2\,\partial_x$ on $\mathbb{R}$. Solve the integral curve through $x_0=2$, find when it escapes, and contrast with a complete field.

Step 1. The ODE is $\dot x=x^2$. Separating variables, $\int x^{-2}\,\mathrm{d}x=\int\mathrm{d}t$ gives $-1/x=t+C$; imposing $x(0)=x_0$ fixes $C=-1/x_0$, so $x(t)=x_0/(1-x_0 t)$. The unique integral curve is $x(t)=2/(1-2t)$.

Step 2. This is smooth wherever the denominator is nonzero, but $1-x_0 t\to 0$ as $t\to 1/x_0$. For $x_0=2$ the curve diverges at $t=1/2$: at $t=0.4$ it already sits at $x=2/(1-0.8)=10$, and it runs to $+\infty$ as $t\to 0.5^-$.

Step 3. The field and its components are perfectly smooth everywhere on $\mathbb{R}$, so this is not a failure of Picard–Lindelöf — the theorem only ever promised a local solution, on the interval $(-\infty,\ 1/x_0)$ for this starting point. Smoothness gives uniqueness, not global existence.

Step 4. Contrast $\tilde X=\partial_x$, whose curve $x(t)=x_0+t$ exists for all $t\in\mathbb{R}$ — a complete field with a global flow. The difference is entirely in how fast $X$ grows at large $x$: quadratic growth outruns any finite parameter budget. Completeness, not smoothness, is what guarantees $\phi_t$ for all $t$; on a compact manifold every smooth field is automatically complete.

Vector fields, their brackets, and their flows are so woven into modern geometry that it is easy to forget they were once a radical idea — that a symmetry could be studied through its infinitesimal generator, an arrow field, rather than the finite transformation itself. That inversion is the life's work of one mathematician.

Texts. Misner, Thorne & Wheeler, Gravitation, Ch. 2–9 (vectors, congruences, and the Lie derivative) · Wald, General Relativity, Ch. 2 and Appendix C (vector fields, flows, and the fundamental theorem on ODEs) · Carroll, Spacetime and Geometry, §2.4 (vector fields, commutators, and the Lie derivative) · Schutz, A First Course in General Relativity, Ch. 5 · Hartle, Gravity, Ch. 8 (worldlines and four-velocity) · Lee, Introduction to Smooth Manifolds, Ch. 8–9 and 20 (vector fields, flows, and Lie derivatives). The key results in this article are: the component bracket $[X,Y]^\mu=X^\nu\partial_\nu Y^\mu-Y^\nu\partial_\nu X^\mu$ arising from the cancellation of second-derivative terms; the second-order flow-composition expansion yielding a loop gap of $t^2[X,Y]$; the rotation-field solution $\gamma(t)=(\cos t,\sin t)$ with $\phi_{\pi/2}(1,0)=(0,1)$; the exact commutator-loop gap $(0,t^2)$ for $X=\partial_x$, $Y=x\partial_y$, equal to $0.25$ at $t=0.5$; and the incomplete-field solution $x(t)=x_0/(1-x_0 t)$ blowing up at $t=1/x_0$ (here $t=0.5$ for $x_0=2$). No empirical constants are quoted; the only conventions are signature $-+++$ and $c=1$ where four-velocities appear.

Builds on: Smooth Manifolds and Coordinate Charts · Tangent Vectors and the Tangent Space · Atlases, Transition Functions and Differentiable Structure · The Tensor Transformation Law. Leads to: General Tensors and the Tensor Product · Cotangent Space and One-Forms · The Metric Tensor and the Line Element · The Invariant Volume Element.

Ask the tutor: Ask me to flow a point along any vector field you name, or to compute a Lie bracket and show you the little square that refuses to close.