Affine Connections and Covariant Derivatives

Welcome to the second module of our course, "Curvature and Dynamics." In the first module, we built the entire static framework of differential geometry: manifolds, tangent spaces, and tensor fields. However, this framework is missing a crucial element. If you have a vector field, how do you determine its rate of change? On a curved manifold, the tangent spaces at different points, $T_{p} M$ and $T_{q} M$ , are distinct vector spaces. We have no natural way to compare a vector at one point to a vector at another, which is the essence of differentiation.

Today, we will introduce the structure that solves this problem. This lesson addresses the learning outcome: Define an affine connection and the covariant derivative of a tensor field.

An affine connection is a new piece of geometric structure on a manifold that provides a rule for differentiating tensor fields. This differentiation operator is called the covariant derivative, denoted by $\nabla$ . It generalizes the concept of a partial derivative to curved spaces in a coordinate-independent manner.

We will approach this topic from two complementary perspectives, satisfying our goal for both physical application and mathematical rigor:

The Axiomatic/Operational View: We'll first see why the ordinary partial derivative fails and then define the covariant derivative by the properties it must have to be a "good" derivative. This leads to the introduction of connection coefficients, $Γ_{μν}^{λ}$ , which are the practical objects used in calculations.
The Geometric View: We will then step back and understand the connection from a more fundamental, geometric standpoint as a structure on the manifold's frame bundle. This rigorous perspective reveals the deeper origin of the covariant derivative.

By the end of this lesson, you will understand what a connection is, how it defines the covariant derivative of any tensor field, and how these concepts are expressed in the coordinate-based language essential for GR calculations.

1. The Problem with Partial Derivatives

Let's start by understanding why our familiar partial derivative, $\partial_{μ}$ , is inadequate on a manifold. The core issue is that it does not produce a tensor.

Consider a vector field $V$ with components $V^{μ}$ in a coordinate system $x^{μ}$ . In another system $x^{' α}$ , the components are $V^{' α} = \frac{\partial x ^{' α}}{\partial x ^{μ}} V^{μ}$ . Let's try to take the partial derivative of these new components with respect to $x^{' β}$ :

\frac{\partial V ^{' α}}{\partial x ^{' β}} = \frac{\partial}{\partial x ^{' β}} (\frac{\partial x ^{' α}}{\partial x ^{μ}} V^{μ})

Using the product rule and chain rule ( $\partial / \partial x^{' β} = (\partial x^{ν} / \partial x^{' β}) \partial / \partial x^{ν}$ ):

\frac{\partial V ^{' α}}{\partial x ^{' β}} = (\frac{\partial x ^{ν}}{\partial x ^{' β}} \frac{\partial}{\partial x ^{ν}} \frac{\partial x ^{' α}}{\partial x ^{μ}}) V^{μ} + \frac{\partial x ^{' α}}{\partial x ^{μ}} (\frac{\partial x ^{ν}}{\partial x ^{' β}} \frac{\partial V ^{μ}}{\partial x ^{ν}})

Rearranging terms, we get:

\frac{\partial V ^{' α}}{\partial x ^{' β}} = \frac{\partial x ^{' α}}{\partial x ^{μ}} \frac{\partial x ^{ν}}{\partial x ^{' β}} \frac{\partial V ^{μ}}{\partial x ^{ν}} + \frac{\partial ^{2} x ^{' α}}{\partial x ^{ν} \partial x ^{μ}} \frac{\partial x ^{ν}}{\partial x ^{' β}} V^{μ}

The first term is exactly how a (1,1) tensor should transform. However, the second term, containing the second derivative of the coordinate transformation, is an unwelcome addition. It is not linear in the vector components and depends on the choice of coordinates. Because of this term, the set of partial derivatives $\partial_{ν} V^{μ}$ does not form the components of a tensor. Our theory must be built from tensors to be independent of coordinates, so the partial derivative is not a valid building block.

2. The Covariant Derivative and Connection Coefficients

The solution is to define a new kind of derivative, the covariant derivative $\nabla$ , which is designed to yield a tensor. The idea is to add a "correction term" to the partial derivative that precisely cancels the unwanted term in the transformation law. This correction is the affine connection.

Axiomatic Definition

We can define the covariant derivative operator, $\nabla$ , by the properties we expect any reasonable derivative to satisfy. An affine connection is a map $\nabla$ that takes two vector fields, $X$ and $Y$ , and produces a third vector field, $\nabla_{X} Y$ (the covariant derivative of $Y$ along $X$ ), satisfying:

Linearity in the first argument: $\nabla_{f X + g Y} Z = f \nabla_{X} Z + g \nabla_{Y} Z$ for functions $f, g$ .
Additivity in the second argument: $\nabla_{X} (Y + Z) = \nabla_{X} Y + \nabla_{X} Z$ .
Leibniz rule for the second argument: $\nabla_{X} (f Y) = (X f) Y + f \nabla_{X} Y$ , where $X f$ is the directional derivative of $f$ along $X$ .
Action on scalars: On a scalar function $f$ , $\nabla_{X} f = X f$ .

Connection and Covariant Derivative Definition and Properties — A summary of the axiomatic definition of a connection, which defines the covariant derivative operator $\nabla$. The image shows the operator taking two vector fields to produce a third, and lists the defining properties.

Connection Coefficients

These abstract properties uniquely determine the form of the covariant derivative in a given coordinate basis ${e_{μ} = \partial_{μ}}$ . The derivative of a basis vector $e_{ν}$ along another basis vector $e_{μ}$ must itself be a vector, so it can be expressed as a linear combination of the basis vectors. We define the components of this expansion as the connection coefficients, $Γ_{μν}^{λ}$ :

\nabla_{e_{μ}} e_{ν} = Γ_{ν μ}^{λ} e_{λ}

Note the index placement: the derivative is along $μ$ , of vector $ν$ , and the components are $Γ_{ν μ}^{λ}$ . This convention is standard in many physics texts, including Carroll's.

Using this definition and the axioms, we can find the expression for the covariant derivative of an arbitrary vector field $V = V^{ν} e_{ν}$ :

\nabla_{μ} V^{λ} \equiv (\nabla_{e_{μ}} V)^{λ} = \partial_{μ} V^{λ} + Γ_{ν μ}^{λ} V^{ν}

The connection coefficients $Γ_{ν μ}^{λ}$ are precisely the correction terms we need. They are defined such that the entire object $\nabla_{μ} V^{λ}$ transforms as a (1,1) tensor, even though $\partial_{μ} V^{λ}$ and $Γ_{ν μ}^{λ}$ do not transform as tensors individually.

The following reading provides the detailed derivation of this result and the non-tensorial transformation law for the connection coefficients themselves.

Lecture Notes on General Relativity - S. Carroll

These notes from Sean Carroll's GR course offer a clear, physicist-friendly introduction to the covariant derivative. It starts by motivating the need for it and then defines it through the connection coefficients.

Please read the first two parts of Section 3, 'Curvature'. The first part (up to equation 3.1) introduces the covariant derivative of a vector. The second part (equations 3.2 to 3.6) derives the transformation law for the connection coefficients, proving they are not tensor components.

Covariant Derivative of General Tensors

The axioms also allow us to extend the definition to one-forms and tensors of arbitrary rank. By requiring that the covariant derivative commutes with contraction (e.g., $\nabla_{μ} (ω_{ν} V^{ν}) = (\nabla_{μ} ω_{ν}) V^{ν} + ω_{ν} (\nabla_{μ} V^{ν})$ ), we find the rule for one-forms:

\nabla_{μ} ω_{ν} = \partial_{μ} ω_{ν} - Γ_{ν μ}^{λ} ω_{λ}

Notice the minus sign and the index placement on $Γ$ . The general rule is wonderfully simple: for any tensor, you add one $+ Γ$ term for each upper index and one $- Γ$ term for each lower index. For a general $(r, s)$ tensor, the formula is:

\nabla_{σ} T_{ν_{1} \dots ν_{s}}^{μ_{1} \dots μ_{r}} = \partial_{σ} T_{ν_{1} \dots ν_{s}}^{μ_{1} \dots μ_{r}} + i = 1 \sum r Γ_{λσ}^{μ_{i}} T_{ν_{1} \dots ν_{s}}^{μ_{1} \dots λ \dots μ_{r}} - j = 1 \sum s Γ_{ν_{j} σ}^{λ} T_{ν_{1} \dots λ \dots ν_{s}}^{μ_{1} \dots μ_{r}}

where in the sums, the $λ$ replaces the appropriate index ( $μ_{i}$ or $ν_{j}$ ). This powerful formula allows you to differentiate any tensor field you encounter.

Lecture Notes on General Relativity - S. Carroll

This next part of Carroll's notes and a similar section from another set of notes by J.D. Finley demonstrate how to extend the covariant derivative to all tensors.

Read the third part of Section 3 in Carroll's notes, which derives the covariant derivative for one-forms and states the general rule for all tensors (equations 3.7 to 3.14).

Covariant Derivatives and Curvature

Finley's notes provide a complementary perspective, defining the connection through its action on basis vectors and then explicitly extending it to 1-forms and general tensors via the product rule.

Read Section I (up to eq 1.7) to see the definition via basis vectors, then read Section II to see the extension to 1-forms and general tensors. This reinforces the rules we just learned.

Test your understanding!

Suppose you are in a 2D space with coordinates $(x,y)$ and a non-zero connection coefficient $\Gamma^x_{yy} = 2x$. All other components of $\Gamma$ are zero. Let $V$ be a vector field with components $V^x = y^2$ and $V^y = 0$. Calculate the components of the covariant derivative $\nabla_y V^x$ and $\nabla_y V^y$.

Show answer

We use the formula $\nabla_{μ} V^{λ} = \partial_{μ} V^{λ} + Γ_{ν μ}^{λ} V^{ν}$ .

For $\nabla_{y} V^{x}$ :
$μ = y$ , $λ = x$ .

\nabla_{y} V^{x} = \partial_{y} V^{x} + Γ_{ν y}^{x} V^{ν} = \partial_{y} V^{x} + Γ_{x y}^{x} V^{x} + Γ_{y y}^{x} V^{y}

We have $\partial_{y} V^{x} = \partial_{y} (y^{2}) = 2 y$ .
The connection components are $Γ_{x y}^{x} = 0$ and $Γ_{y y}^{x} = 2 x$ . The vector components are $V^{x} = y^{2}$ and $V^{y} = 0$ .
Plugging these in:

\nabla_{y} V^{x} = 2 y + (0) (y^{2}) + (2 x) (0) = 2 y

For $\nabla_{y} V^{y}$ :
$μ = y$ , $λ = y$ .

\nabla_{y} V^{y} = \partial_{y} V^{y} + Γ_{ν y}^{y} V^{ν} = \partial_{y} V^{y} + Γ_{x y}^{y} V^{x} + Γ_{y y}^{y} V^{y}

We have $\partial_{y} V^{y} = \partial_{y} (0) = 0$ .
All given connection components with an upper index of $y$ are zero.
So, $\nabla_{y} V^{y} = 0 + (0) (y^{2}) + (0) (0) = 0$ .

The covariant derivative $\nabla_{y} V$ has components $(2 y, 0)$ .

3. The Geometry of Connections

We now have an operational definition of a connection through the coefficients $Γ_{μν}^{λ}$ . But what is a connection, geometrically? Your preference for mathematical rigor calls for a deeper look.

The geometric picture is that a connection provides a way to relate the tangent spaces at nearby points. It does this by defining a notion of "horizontal" directions on a more abstract space called a bundle. For General Relativity, the relevant bundle is the frame bundle, whose "fibers" are the set of all possible basis frames for the tangent space at each point.

A connection is a structure on this bundle that, at every point, splits the tangent space of the bundle itself into "vertical" (directions pointing along a fiber) and "horizontal" parts. This choice of horizontal directions is what allows us to define parallel transport, and from that, the covariant derivative.

The following lectures by Frederic Schuller provide a rigorous, bottom-up construction. This is a more advanced viewpoint but is the correct, modern mathematical formulation.

Conncections and connection 1-forms - Lec 21 - Frederic Schuller

This first video introduces the fundamental idea of a connection on a principal bundle, which is the most general setting. The lecturer, Frederic Schuller, clarifies the hierarchy: a connection is the primary object, which then gives rise to parallel transport and, in the special case of vector bundles, the covariant derivative.

Please watch from the beginning to 35:15. Hierarchy (00:08 - 03:40): Pay attention to the distinction between a connection, parallel transport, and a covariant derivative. This framing is crucial. Vertical Subspace (09:22 - 16:33): Understand the definition of the vertical subspace of the principal bundle's tangent space. Geometric Definition (16:33 - 35:15): This is the core geometric concept. A connection is defined as a choice of a horizontal subspace at each point of the principal bundle, which is complementary to the vertical subspace. Absorb this powerful geometric idea.

This geometric definition, while intuitive, is hard to work with computationally. Fortunately, it is equivalent to defining the connection as a specific type of differential form on the principal bundle.

Conncections and connection 1-forms - Lec 21 - Frederic Schuller

This next segment from the same lecture shows how the geometric choice of horizontal subspaces can be encoded in a more computationally friendly object: a Lie algebra-valued 1-form, called the connection 1-form.

Watch from 35:15 to 43:10. The key takeaway is that the horizontal subspace is precisely the kernel of this connection 1-form. This provides an elegant algebraic way to specify a connection.

Finally, the connection on the principal bundle (the frame bundle) induces a covariant derivative on associated vector bundles, such as the tangent bundle that vector fields live in. The coordinate expression for this induced derivative involves components of the connection 1-form, which play the role of the $Γ$ symbols we saw earlier. The following video makes this link explicit.

Covariant derivatives - Lec 25 - Frederic Schuller

This final video makes the crucial link between the abstract connection on the principal bundle and the covariant derivative operator ( abla) acting on sections of an associated vector bundle (like a vector field).

Watch from 50:58 to 01:13:01. This part is quite technical but it completes the picture. It starts by defining a covariant derivative for functions on the principal bundle (called the 'exterior covariant derivative'). It then shows how this operator, when 'pulled back' to the base manifold, gives rise to the familiar formula for the covariant derivative of a vector field, explicitly showing how the connection coefficients (\Gamma) emerge from the connection 1-form.

Watching these clips should give you a strong appreciation for the deep geometric origins of the covariant derivative, connecting the abstract bundle picture to the practical formulas we use in physics.

Conclusion

In this lesson, we have introduced one of the most fundamental tools in differential geometry and General Relativity: the affine connection and its associated covariant derivative, $\nabla$ .

Key Takeaways:

The ordinary partial derivative $\partial_{μ}$ is not a valid operator in curved space because its action on a tensor field does not produce another tensor field.
An affine connection provides a rule for differentiation that is covariant. It can be defined axiomatically as an operator $\nabla$ that satisfies linearity and the Leibniz rule.
In a coordinate system, the connection is specified by a set of $n^{3}$ functions called connection coefficients, $Γ_{μν}^{λ}$ , which are not tensor components themselves.
The covariant derivative of a tensor is given by the partial derivative plus correction terms involving the connection coefficients: one $+ Γ$ term for each upper index and one $- Γ$ term for each lower index.
Geometrically, a connection is a choice of "horizontal" directions on the manifold's frame bundle, which can be encoded in a connection 1-form. This structure induces the covariant derivative on all tensor bundles.

We now have a way to differentiate vector and tensor fields. This opens the door to dynamics. In our next lesson, we will use the connection to define the concept of parallel transport, which describes how to move a vector along a curve while keeping it "pointing in the same direction." This will lead us directly to the equation for geodesics—the paths that free particles follow through curved spacetime.

Recommended resources