Cotangent Space and One-Forms

The cotangent space is not a museum piece — its structure runs quietly through modern physics wherever the metric is absent or secondary. Hamiltonian mechanics lives on the cotangent bundle $T^*Q$ of configuration space: the canonical momentum $p_\mu=\partial L/\partial\dot q^\mu$ is a one-form, and the symplectic form $\omega_{\rm sympl}=\mathrm{d}(p_\mu\,\mathrm{d}q^\mu)$ that governs all of classical dynamics is built without any reference to a metric — which is why symplectic geometry, not Riemannian geometry, underlies phase-space methods, the geometry of optics, and much of quantum foundations. In observational relativity, the cleanest way to compute what a detector actually records — the energy of a photon, a gravitational redshift near a compact object, the frequency shift of a pulsar's ticks — is to carry four-momentum as the one-form $p_\mu$ and contract it with an observer's four-velocity, $E=-p_\mu u^\mu$, an expression whose invariance is guaranteed by exactly the Jacobian cancellation derived above. And in gauge theory and numerical relativity, the split of tensors into upper and lower index slots — vectors versus one-forms — is the bookkeeping that keeps a simulation covariant across the coordinate patches stitched together to cover a black-hole spacetime. The humble covector is the load-bearing beam beneath all of it.

Draw a vector as an arrow. Now refuse to draw a one-form as an arrow — that is the whole conceptual battle. The right picture, due to Misner, Thorne and Wheeler, is a stack of slats: a one-form $\omega$ is a family of parallel level surfaces, the contour lines of the linear function it represents, and its action on a vector is simply the number of surfaces the arrow pierces. Denser slats mean a larger one-form. This "bong of a bell" image is powerful because it is manifestly metric-free: counting how many contour lines an arrow crosses needs no notion of length or angle, so the pairing $\omega(v)$ is invariant under any coordinate change, even one that scrambles distances entirely.

The gradient is the cleanest reason to care. In vector calculus we casually write $\nabla f$ as an arrow pointing uphill, but look at where its components live: $\partial f/\partial x^\mu$ carries the coordinate in the denominator, and under a coordinate change it transforms with $\partial x^\mu/\partial x^{\mu'}$ — the inverse Jacobian, the covariant rule, not the vector rule. The honest statement is that $\mathrm{d}f$ is a covector, and turning it into the steepest-ascent arrow requires a metric to raise the index, $\nabla^\mu f = g^{\mu\nu}\partial_\nu f$. Duality also runs both ways: the double dual $T_p^{**}M$ is canonically $T_pM$ again, so a vector can equally be read as a functional on one-forms, $v(\omega):=\omega(v)$. What is missing, until a metric appears, is any canonical bridge between $T_pM$ and $T_p^*M$ themselves. The second figure shows the relation that pins the dual basis down.

The cotangent space is defined by a single act of dualization: take every linear map from the tangent space to the reals, and collect them into a new vector space. Nothing here needs a metric, a connection, or even coordinates — only the linear structure of $T_pM$.

where: $T_pM$ — tangent space at $p$, the vector space of tangent vectors; $T_p^*M$ — cotangent space at $p$, the dual vector space; $\omega$ — a one-form (covector), a linear functional on tangent vectors. Reading: $T_p^*M$ inherits addition and scalar multiplication pointwise, so it is itself an $n$-dimensional vector space. The single most important one-form is the differential of a function. Given a smooth $f:M\to\mathbb{R}$ and a tangent vector $v$ acting as a derivation, the differential $\mathrm{d}f$ returns the rate of change of $f$ along $v$. This is the intrinsic, coordinate-free content of "gradient".

where: $f$ — smooth real function on $M$; $\mathrm{d}f$ — differential of $f$, a one-form at $p$; $v$ — tangent vector acting as a derivation; $v(f)$ — directional derivative of $f$ along $v$. Reading: linearity of $\mathrm{d}f$ in $v$ is inherited from the derivation property, so $\mathrm{d}f\in T_p^*M$ automatically. Choose a chart $x^\mu$. Its coordinate functions have differentials $\mathrm{d}x^\mu$, and these turn out to be exactly the basis of $T_p^*M$ dual to the coordinate basis $\{\partial_\mu\}$. The defining relation is biorthogonality: each basis one-form reads off one basis vector and ignores the rest.

where: $\partial_\nu$ — coordinate basis vector $\partial/\partial x^\nu$ of $T_pM$; $\mathrm{d}x^\mu$ — coordinate basis one-form, the differential of $x^\mu$; $\delta^\mu{}_\nu$ — Kronecker delta, $1$ if $\mu=\nu$ else $0$. Reading: this follows because $\mathrm{d}x^\mu(\partial_\nu)=\partial_\nu(x^\mu)=\partial x^\mu/\partial x^\nu=\delta^\mu{}_\nu$. Because $\{\mathrm{d}x^\mu\}$ is a basis, any one-form expands in it, and biorthogonality lets you extract each component by feeding the matching basis vector into $\omega$.

where: $\omega_\mu$ — components of $\omega$ in the chart (lower index means covariant); $\mathrm{d}x^\mu$ — dual basis one-forms; the repeated index is summed (Einstein convention). Reading: a one-form is fully specified by its $n$ numbers $\omega_\mu$, just as a vector is by its $n$ numbers $v^\mu$. Feeding a vector into a one-form and using biorthogonality collapses the would-be double sum to a single contraction — the pairing that produces an invariant number from a covector and a vector.

where: $v=v^\mu\partial_\mu$ — a tangent vector with components $v^\mu$; $\omega_\mu v^\mu$ — the invariant scalar obtained by contracting one lower with one upper index. Reading: this is the simplest instance of an index contraction, and the whole point of the upper/lower distinction is to make it invariant. That invariance forces the transformation law. Under a change of chart the covector components pick up the inverse Jacobian — the opposite of the vector rule — so that the two Jacobians cancel inside the pairing.

where: $x^{\mu'}$ — new coordinates; $\partial x^\mu/\partial x^{\mu'}$ — inverse Jacobian (covariant, for one-forms); $\partial x^{\mu'}/\partial x^\mu$ — Jacobian (contravariant, for vectors). Reading: "covariant" means components vary with the basis change; "contravariant" means they vary against it. Finally, in components the differential is just the row of partial derivatives — the gradient wearing its correct, lower-index clothes.

where: $\partial f/\partial x^\mu$ — partial derivatives of $f$, the covariant components of the gradient. Reading: the components $\partial_\mu f$ transform with the inverse Jacobian by the chain rule, confirming that $\mathrm{d}f$ is a one-form and not a vector.

1 · Invariance of the contraction under a coordinate change (fully numeric)

In a 2D chart $(x^1,x^2)$ a one-form has components $\omega_\mu=(2,\,3)$ and a vector has components $v^\mu=(4,\,1)$. Change to new coordinates $x'^1=2x^1$, $x'^2=x^1+x^2$. Compute the pairing in both charts and confirm it is invariant.

Step 1. Old-chart pairing: $\omega_\mu v^\mu = (2)(4)+(3)(1) = 8+3 = \mathbf{11}$. This is the candidate invariant.

Step 2. Build the Jacobian $\partial x'^{\mu}/\partial x^\nu$ from the transformation. Row by row it is $\begin{pmatrix}2&0\\1&1\end{pmatrix}$, and its inverse $\partial x^\nu/\partial x'^{\mu}$ is $\begin{pmatrix}1/2&0\\-1/2&1\end{pmatrix}$. These are the two matrices the contravariant and covariant rules call for.

Step 3. Transform the vector with the Jacobian (contravariant): $v'^1=2\cdot4=8$, $v'^2=1\cdot4+1\cdot1=5$, so $v'^\mu=(8,\,5)$.

Step 4. Transform the one-form with the inverse Jacobian (covariant), $\omega'_\mu=\dfrac{\partial x^\nu}{\partial x'^\mu}\omega_\nu$: $\omega'_1=\tfrac12(2)+(-\tfrac12)(3)=1-1.5=-0.5$, and $\omega'_2=0\cdot2+1\cdot3=3$, so $\omega'_\mu=(-0.5,\,3)$. Notice the vector components grew while the covector components shrank — opposite behaviour, as required.

Step 5. New-chart pairing: $\omega'_\mu v'^\mu = (-0.5)(8)+(3)(5) = -4+15 = \mathbf{11}$. The contraction is $11$ in both charts — the Jacobian and inverse Jacobian cancelled exactly, confirming $\omega_\mu v^\mu$ is a scalar.

2 · The energy a moving observer measures (fully numeric)

In flat spacetime with signature $(-+++)$ and $c=1$, a photon travels in the $+x$ direction carrying the four-momentum one-form $p_\mu=(-2,\,2,\,0,\,0)$ (so $E=2$, $p_x=2$, and $p_\mu p^\mu=0$). Find the energy measured by (a) a static observer $u^\mu=(1,0,0,0)$ and (b) an observer moving at $\beta=0.6$ in the $+x$ direction. The measured energy is the invariant pairing $E_{\rm obs}=-p_\mu u^\mu$.

Step 1. Static observer: $E_{\rm obs}=-p_\mu u^\mu = -\big[(-2)(1)+ (2)(0)\big] = \mathbf{2}$. The static observer reads the photon's coordinate energy, as expected.

Step 2. Moving observer's four-velocity: with $\beta=0.6$, $\gamma=1/\sqrt{1-0.36}=1/\sqrt{0.64}=1.25$, so $u^\mu=\gamma(1,\beta,0,0)=(1.25,\,0.75,\,0,\,0)$. A boost in $+x$, the same direction the photon moves.

Step 3. Pair the one-form with this vector: $E_{\rm obs}=-p_\mu u^\mu = -\big[(-2)(1.25)+(2)(0.75)\big] = -\big[-2.5+1.5\big] = -(-1.0) = \mathbf{1.0}$.

Step 4. Cross-check against the relativistic Doppler formula for an observer chasing the source, $E_{\rm obs}/E=\gamma(1-\beta)=1.25(0.4)=0.5$, giving $E_{\rm obs}=0.5\times2=1.0$. The observer chasing the photon measures $E=1.0$, a redshift — and writing energy as the contraction $-p_\mu u^\mu$ made the answer manifestly coordinate-independent, the cleanest route to every redshift calculation.

One-forms are where linear algebra, exterior calculus, and general relativity meet. The modern picture — a covector as a metric-free measuring device, and the differential as the calculus that satisfies $\mathrm{d}^2=0$ — is largely the creation of one geometer who spent decades insisting that differential forms, not coordinates, are the right primitive objects.

Texts. Misner, Thorne & Wheeler, Gravitation, §2.5–2.6 (one-forms and the gradient as a one-form, the piercing picture) · Wald, General Relativity, §2.3 (dual vectors and tensors) · Carroll, Spacetime and Geometry, §2.3 (vectors and dual vectors, transformation laws) · Schutz, A First Course in General Relativity, §3.3 (one-forms) · Hartle, Gravity, §20.2 (vectors, one-forms, tensors) · Lee, Introduction to Smooth Manifolds, Ch. 11 (the cotangent bundle). The key results to keep in mind: $\mathrm{d}x^\mu(\partial_\nu)=\delta^\mu{}_\nu$ follows directly from $\partial_\nu x^\mu$; the covector rule $\omega_{\mu'}=(\partial x^\mu/\partial x^{\mu'})\omega_\mu$ and the Jacobian cancellation in $\omega_\mu v^\mu$ are derived from the chain rule; $\mathrm{d}(\mathrm{d}f)=0$ follows from the symmetry of mixed partials against the antisymmetric wedge; $\dim T_p^*M=\dim T_pM=n$ comes from the dual-basis construction; worked example 1 gives $11=11$ across charts, and worked example 2 gives $E=2$ for the static observer and $E=1.0$ at $\beta=0.6$, consistent with $\gamma(1-\beta)$. The only physical convention used is $c=1$ with signature $-+++$ in the energy-pairing example; no empirical constants are quoted.

Builds on: Tangent Vectors and the Tangent Space · Smooth Manifolds and Coordinate Charts · Atlases, Transition Functions and Differentiable Structure. Leads to: General Tensors and the Tensor Product · The Tensor Transformation Law · The Metric Tensor and the Line Element · Raising and Lowering Indices · Tensor Algebra: Contraction, Symmetrization and Antisymmetrization.

Ask the tutor: Ask me to walk you through why the gradient is secretly a covector, or to work a raising-and-lowering example the moment a metric enters.