1 Covariant tensors and index notation

1.1 Covariant indices (lower indices)

In index notation, a covariant tensor is indicated by indices written with a particular placement convention. Components of a covariant tensor carry lower indices, and each lower index is associated with how the object responds to a change of coordinates. This placement is not merely cosmetic: it encodes the transformation law that distinguishes covariant components from contravariant ones.

Lower indices are commonly interpreted as encoding how the tensor interacts with vectors in a way compatible with the geometry of the underlying space. For example, a covariant vector is often identified with a 1-form (a linear functional acting on vectors), and its components multiply the components of vectors in an invariant pairing.

1.2 Multilinear maps viewpoint

A standard way to define covariant tensors is through multilinear maps. Roughly, a covariant tensor of rank \(k\) is a multilinear rule that takes \(k\) vectors as inputs and returns an output that transforms appropriately so that the overall definition does not depend on the chosen coordinate system. In the covariant setting, the tensor naturally acts by feeding in vectors (or, in dual language, by pairing covariant slots with contravariant arguments) and producing scalars or lower-rank objects.

This viewpoint is central because it separates the abstract tensor from its coordinates. The tensor is first understood as a coordinate-independent multilinear map, and only later are its components computed in a coordinate chart.

1.3 Component form under coordinate changes

Although the tensor itself is coordinate-free, its components depend on the chart. Under a coordinate transformation, covariant tensor components transform according to the rules of tensor calculus: each lower index introduces factors derived from the Jacobian of the coordinate change in a specific way. Concretely, covariant components use the inverse Jacobian relative to contravariant components.

This component transformation law ensures that any scalar obtained by contracting covariant and contravariant indices (and summing over matched index positions) remains invariant under coordinate changes.

2 Transformation rules

2.1 Basic coordinate transformation law

Let a coordinate change be given by \(x^\mu \mapsto x^{\mu'}(x)\). The Jacobian matrix is \( \frac{\partial x^{\mu'}}{\partial x^\nu}\). For a covariant tensor component with one lower index, the transformation has the schematic form \[ T_{\mu'} = \frac{\partial x^\nu}{\partial x^{\mu'}}\, T_\nu. \] Each additional lower index contributes another factor of \(\frac{\partial x^\nu}{\partial x^{\mu'}}\) with the corresponding index placement. For a covariant tensor of rank \(k\) with components \(T_{\mu_1\cdots \mu_k}\), the transformed components are obtained by multiplying by \(k\) such Jacobian inverse factors, one for each lower index.

2.2 Covariant vs. contravariant transformation comparison

The essential contrast with contravariant tensors is the direction of the Jacobian factors. Contravariant components (upper indices) transform with factors \( \frac{\partial x^{\mu'}}{\partial x^\nu}\), while covariant components (lower indices) transform with the inverse factors \( \frac{\partial x^\nu}{\partial x^{\mu'}}\).

This distinction is what makes pairings between covariant and contravariant objects coordinate-invariant. When an upper and a lower index are matched and summed over, the Jacobian factors cancel out, producing an invariant result.

2.3 Index bookkeeping and Einstein summation

Einstein summation convention streamlines calculations by implicitly summing over repeated indices. In this framework:

  • Repeated indices of one covariant (lower) and one contravariant (upper) type are summed.
  • Free indices (those not summed) determine the rank and transform according to the placement pattern.

Index bookkeeping is crucial to avoid mixing transformation behaviors. Keeping track of whether an index is raised or lowered determines which Jacobian factors appear in the transformed components.

2.4 Examples in low ranks

A few elementary cases illustrate the general rule.

  • Rank-0 (scalar): A scalar \(f\) has no indices and remains unchanged in form under coordinate changes.
  • Rank-1 covariant tensor (covector): A covariant vector (often called a covector) \(\omega_\mu\) transforms with an inverse Jacobian factor.
  • Rank-2 covariant tensor: Components \(S_{\mu\nu}\) transform with one inverse Jacobian factor for \(\mu\) and another for \(\nu\). Bilinear expressions built from these components and two vectors can be shown to be invariant when indices are contracted appropriately.

These examples provide the template for higher-rank constructions.

3 Construction and examples

3.1 Gradients of scalar fields

Given a smooth scalar field \(f\), its differential \(df\) is a covariant 1-tensor (a covector field). In local coordinates, \[ df = \frac{\partial f}{\partial x^\mu}\, dx^\mu, \] so the components \(\partial_\mu f\) naturally behave as covariant components. This is not an artifact of notation: the differential of a scalar is defined in a coordinate-independent way, and the coordinate representation must transform as a covector.

Accordingly, the gradient in differential-geometric terms is typically associated with the covariant derivative of a scalar (equivalently the differential), while the “vector gradient” often requires a metric to convert covariant data into contravariant form.

3.2 Differential 1-forms

A differential 1-form is a covariant tensor of rank 1. Locally it can be expressed as \[ \omega = \omega_\mu\, dx^\mu, \] where the basis \(dx^\mu\) transforms so that \(\omega_\mu\) follows the covariant transformation law. The key feature is that a 1-form assigns a scalar to each tangent vector by pairing \(\omega(v)\).

This pairing is invariantly defined and is a prototypical example of how covariant objects interact with vectors.

3.3 Covariant rank-2 tensors (bilinear forms)

A covariant rank-2 tensor can be treated as a bilinear form: it takes two vectors and outputs a scalar, depending linearly on each argument. In a coordinate chart, one writes \[ B(v,w) = B_{\mu\nu} v^\mu w^\nu, \] with \(B_{\mu\nu}\) covariant components and \(v^\mu, w^\nu\) contravariant components.

The bilinear map viewpoint is often more conceptual than component formulas, especially when discussing symmetry properties or invariant meaning.

3.4 Metrics as covariant tensors

In Riemannian and pseudo-Riemannian geometry, a metric is commonly presented as a covariant rank-2 tensor \(g_{\mu\nu}\). It assigns lengths and angles through expressions like \[ g(v,w) = g_{\mu\nu} v^\mu w^\nu. \] Because the metric is defined geometrically, the components \(g_{\mu\nu}\) must transform as a covariant tensor under coordinate changes.

Many further constructions—such as raising and lowering indices—depend on the metric, but the metric itself is naturally a covariant object in its basic definition.

3.5 Inner products and natural contraction patterns

The inner product induced by a metric is an example of a natural contraction pattern. Given a covector \(\omega_\mu\) and a vector \(v^\mu\), the contraction \(\omega_\mu v^\mu\) is coordinate invariant: covariant and contravariant transformation factors cancel.

Similarly, for a covariant rank-2 tensor \(S_{\mu\nu}\), one obtains invariant scalars by contracting with two vectors, as in \(S_{\mu\nu} v^\mu w^\nu\). These patterns motivate why index placement matters so strongly: it determines which contractions yield invariant outcomes.

4 Covariant tensor fields on manifolds

4.1 Tensor fields and smoothness requirements

On a manifold, a covariant tensor field assigns to each point a covariant tensor in a way compatible with the manifold’s smooth structure. Practically, one requires that its component functions in any coordinate chart are smooth (or of a specified regularity class). This ensures that operations like differentiation and integration can be carried out consistently.

The field perspective extends the notion of a single tensor at one vector space to a smoothly varying family of tensors across the manifold.

4.2 Local charts and component expressions

To compute with a tensor field, one chooses a local chart \((U, x^\mu)\). In that chart, the field is represented by component functions \(T_{\mu_1\cdots \mu_k}(x)\). On overlapping charts, the components are related by the covariant transformation law. This compatibility is what makes the field globally meaningful even though its numerical components depend on the chart.

Thus, the tensor field is defined by its transformation behavior across chart overlaps, not by any single local coordinate expression.

4.3 Pullback behavior under maps

If \(\phi : M \to N\) is a smooth map and a covariant tensor field is given on \(N\), the tensor can often be transported to \(M\) using the pullback construction. Intuitively, pullback describes how covariant data on the target manifold produces covariant data on the source manifold.

For a covariant 1-form \(\omega\) on \(N\), the pullback \(\phi^*\omega\) is defined so that \((\phi^*\omega)_p(v) = \omega_{\phi(p)}(d\phi_p(v))\). This definition generalizes to higher-rank covariant tensors by applying the differential \(d\phi\) to each argument slot.

4.4 Tensor product and direct sums

Covariant tensor fields can be combined using tensor products and direct sums. The tensor product of covariant tensors produces a higher-rank covariant tensor whose components are formed by multiplying component functions and appropriately organizing indices. Direct sums correspond to forming a block-like combination of tensors of the same rank and type.

These algebraic operations are coordinate-independent at the abstract level, but their component expressions follow the same consistent index placement rules.

5 Operations involving covariant tensors

5.1 Tensor contraction (index contraction)

Contraction is the operation that pairs a covariant index with a contravariant index and sums over matching labels. For instance, contracting a covector \(\omega_\mu\) with a vector \(v^\mu\) yields a scalar \(\omega_\mu v^\mu\).

Contraction reduces rank: contracting one pair of indices turns a tensor of rank \((k,\ell)\) into one of rank \((k-1,\ell-1)\) (using a common bookkeeping convention). The result is invariant under coordinate changes when indices are contracted correctly.

5.2 Symmetric and antisymmetric covariant tensors

Covariant tensors can be decomposed into symmetric and antisymmetric parts. For rank-2 covariant tensors \(S_{\mu\nu}\), the symmetric part is \(S_{(\mu\nu)}=\frac12(S_{\mu\nu}+S_{\nu\mu})\), while the antisymmetric part is \(S_{[\mu\nu]}=\frac12(S_{\mu\nu}-S_{\nu\mu})\).

This decomposition is valuable because symmetry determines how many independent components exist and how the tensor behaves under index permutations. Antisymmetric covariant tensors of higher rank are related to differential forms.

5.3 Raising and lowering indices (with a metric)

Index raising and lowering convert covariant indices to contravariant ones and vice versa, typically using a metric tensor \(g_{\mu\nu}\) and its inverse \(g^{\mu\nu}\). For example, given a covector \(\omega_\mu\), one defines the corresponding vector components by \[ \omega^\mu = g^{\mu\nu}\omega_\nu. \] Lowering a vector uses the metric in the opposite direction: \[ v_\mu = g_{\mu\nu}v^\nu. \]

These operations depend on the additional structure provided by the metric. Without a metric (or another isomorphism between tangent and cotangent spaces), there is no canonical notion of raising or lowering.

5.4 Hodge dual and orientation-dependent constructions

The Hodge dual maps \(p\)-forms to \((n-p)\)-forms in an \(n\)-dimensional oriented manifold equipped with a metric. In components, it involves the Levi-Civita tensor density, and the result depends on both orientation and metric signature. Because it is built from contraction with the volume element, it uses covariant index structures in a fundamental way.

The dual operator is widely used to express Maxwell-type equations and to relate integrals over complementary-dimensional submanifolds.

6.1 Why partial derivatives are not tensorial

Partial derivatives of tensor components generally do not transform tensorially, because differentiation interacts with the coordinate dependence of the basis vectors (or co-basis forms). As a consequence, the naive derivative \(\partial_\nu T_{\mu\cdots}\) acquires extra terms under coordinate transformations, preventing it from being a tensor in general.

This motivates the introduction of a connection, which adjusts for the change of basis during differentiation.

6.2 Levi-Civita connection for metric-compatible geometry

For a (pseudo-)Riemannian metric, the Levi-Civita connection is the unique torsion-free connection compatible with the metric. Compatibility means the metric is covariantly constant, often written as \(\nabla_\lambda g_{\mu\nu}=0\). Torsion-free means the connection is symmetric in its lower two indices in an appropriate convention.

The Levi-Civita connection provides the canonical notion of covariant differentiation in standard geometric settings of general relativity and Riemannian geometry.

6.3 Covariant derivative of tensor fields

The covariant derivative \(\nabla\) extends the notion of differentiation to tensor fields while preserving tensorial transformation properties. For a covector field \(\omega_\mu\), one writes \[ \nabla_\nu \omega_\mu = \partial_\nu \omega_\mu - \Gamma^\lambda_{\nu\mu}\,\omega_\lambda, \] where \(\Gamma^\lambda_{\nu\mu}\) are the connection coefficients (Christoffel symbols for Levi-Civita). The minus sign reflects that \(\omega_\mu\) is covariant: the correction term has a structure corresponding to its index placement.

For higher-rank covariant tensors, additional connection terms appear for each covariant index, ensuring the output transforms as a tensor of one higher rank in the derivative direction.

6.4 Commutation and curvature connections (overview)

Covariant derivatives do not generally commute. The commutator \([\nabla_\mu,\nabla_\nu]\) acting on a tensor field produces terms involving curvature. This curvature measures the failure of local flatness and depends only on the connection, not on the specific tensor being differentiated.

As an overview, curvature encodes how parallel transport around small loops changes tensor components, and it underlies the geometric interpretation of gravitational tidal effects and the structure of manifolds.

7 Practical computation patterns

7.1 Component calculation workflow in coordinates

A typical workflow in computations is:

  1. Choose a coordinate chart and write tensor components.
  2. Apply the covariant transformation law when changing charts (or verify it for a given expression).
  3. Perform algebraic operations like contraction, symmetrization, and tensor products using index placement rules.
  4. When differentiation is needed, use the covariant derivative formula with the appropriate connection coefficients.

This approach keeps the distinction between geometric objects and their coordinate representations clear.

7.2 Common contractions in physics-style notation

In physics and engineering contexts, expressions often use mixed-index forms and implicit summation. Common contractions include:

  • \(\omega_\mu v^\mu\) for pairing a covector and a vector.
  • \(T_{\mu\nu}v^\mu w^\nu\) for bilinear forms.
  • \(F_{\mu\nu}F^{\mu\nu}\) when indices are raised using a metric.

Although these formulas look compact, correctness depends on using the correct index placement and consistent metric operations when raising indices.

7.3 Verifying tensor character by transformation checks

When presented with a proposed formula, one can test whether it defines a tensor by checking its transformation behavior under coordinate changes. For covariant tensors, each lower index should introduce the correct inverse Jacobian factors. If extra terms appear that do not match the required transformation law, the expression typically represents a non-tensorial quantity (often something involving partial derivatives without a compensating connection).

This method is a practical substitute for deriving tensor properties from first principles.

7.4 Coordinate-invariant statements derived from components

Even though components depend on coordinates, certain derived statements are invariant. Examples include:

  • Scalars formed by full contraction.
  • Integrated quantities that rely on invariant volume forms and appropriate tensor densities.
  • Symmetry and trace properties preserved under coordinate changes.

A common computational pattern is to express the invariant statement in components, compute in a convenient chart, and then rely on the tensorial transformation laws to ensure the result is independent of that chart choice.

8.1 Contravariant tensors and dual spaces

Covariant and contravariant tensors are linked through duality. A covariant tensor slot can be viewed as acting on a vector space, which is related to the cotangent bundle in manifold language. Contravariant tensors instead act on covector arguments (or, equivalently, represent multilinear maps with the opposite index placement).

The dual relationship explains why raising/lowering operations and contractions create coordinate-invariant pairings.

8.2 Dual vector spaces and covectors

A covector is an element of the dual vector space: it is a linear functional on vectors. In coordinates, covectors are represented with lower indices and transform covariantly. This dual viewpoint is foundational for interpreting covariant tensors as arrays of coefficients that encode how linear functionals act on tangent vectors.

It also clarifies why covariant tensors naturally appear when working with differentials, gradients, and linearizations.

8.3 Tensor bundles and sections (conceptual overview)

On a manifold, tensors are organized into vector bundles: for each point, the fiber is a space of tensors of specified type (e.g., covariant of rank \(k\)). A tensor field is then a section of the corresponding bundle. This bundle viewpoint provides a rigorous geometric framework for discussing global behavior, smoothness, and coordinate transformations.

It also makes operations like tensor products and covariant differentiation naturally geometric, rather than merely algebraic.

8.4 Natural pairings and adjoint relationships

Natural pairings are bilinear operations between spaces of tensor types that are invariant under coordinate changes. The pairing between a covector and a vector is the simplest example, yielding a scalar. More generally, covariant tensors pair with contravariant tensors of complementary type through contractions.

Adjoint relationships—used in contexts like functional analysis and differential geometry—often rely on these pairings. In metric settings, an adjoint can be defined by requiring compatibility with the inner product, linking covariant structures to operator theory.