Tensor analysis is the mathematics of quantities whose meaning survives a change of coordinates. Vectors, covectors, bilinear forms and higher-order tensors can be represented by arrays of components, but the tensor itself is the coordinate-independent object those components describe.
This distinction becomes essential whenever geometry, mechanics or fields are described in different coordinate systems. A physically meaningful law should not become false merely because one observer uses Cartesian coordinates and another uses curvilinear coordinates.
Series route: Mathematics Learning Hub → How Mathematics Works → Tensor Analysis. Useful foundations include Linear Algebra, Differential Geometry and Mathematical Physics.
1. A tensor is not merely a multidimensional array
An array is a list of numbers arranged by indices. A tensor is a multilinear geometric object whose components transform according to specific rules under change of basis.
The array is one coordinate representation of the tensor.
2. Scalars are rank-zero tensors
A scalar has one value independent of basis.
Temperature at a point is a common example. Its numerical value can depend on the chosen unit system, but not on the orientation of coordinate axes.
3. Vectors are rank-one contravariant tensors
A vector can be expanded in a basis as v=v^i e_i.
When the basis changes, the components v^i change in the opposite way so that the geometric vector remains the same.
4. Covectors are linear measurement rules
A covector is a linear functional that maps vectors to scalars.
In coordinates, covector components transform differently from vector components.
Differentials of scalar fields are naturally covectors.
5. Dual spaces separate directions from measurements
The vector space V contains directions; the dual space V* contains linear measurements on those directions.
Without additional structure, vectors and covectors are related but not identical.
6. Tensors combine vectors and covectors multilinearly
A tensor of type (r,s) can be viewed as a multilinear map taking r covectors and s vectors, or equivalently as an element of a tensor product of vector and dual spaces.
The pair (r,s) records contravariant and covariant slots.
7. Tensor products build higher-order structure
The tensor product combines vectors, covectors or other tensors into a larger multilinear object.
If u and v are vectors, u⊗v is a rank-two contravariant tensor.
The product is bilinear and preserves separate linear dependence in each factor.
8. Basis changes control component transformations
A tensor’s components follow transformation rules built from the change-of-basis matrix and its inverse.
These rules are not arbitrary bookkeeping. They guarantee that the underlying multilinear map is unchanged.
9. Einstein summation compresses repeated-index notation
In tensor notation, a repeated index appearing once up and once down is conventionally summed.
Thus a_i v^i means Σ_i a_i v^i.
The notation reveals contraction structure while reducing visual clutter.
10. Free indices must match across an equation
Indices that are not summed are free. Every term in a valid tensor equation must carry the same free-index structure.
This provides a fast consistency check analogous to dimensional analysis.
11. Contraction reduces tensor rank
Contraction pairs one covariant and one contravariant index and sums them.
The trace of a linear operator is a familiar contraction.
Contraction is coordinate-independent even though it is calculated through components.
12. Symmetric tensors remain unchanged under slot exchange
A rank-two covariant tensor T is symmetric when T(u,v)=T(v,u).
Metrics and many stress or covariance-type tensors have symmetry under appropriate modelling conditions.
13. Antisymmetric tensors change sign
An antisymmetric tensor changes sign when two relevant slots are exchanged.
Differential forms are fully antisymmetric covariant tensors and support coordinate-independent integration.
14. The metric tensor measures inner products
A metric tensor g assigns an inner product to tangent vectors at each point of a manifold.
It determines lengths, angles and local geometric measurement.
This is the central bridge to Differential Geometry.
15. The metric lowers indices
Given vector components v^j, the metric produces covector components v_i=g_{ij}v^j.
This operation converts a vector into its metric-dual covector.
The ability to identify vectors and covectors therefore depends on having a nondegenerate metric or comparable structure.
16. The inverse metric raises indices
The inverse metric g^{ij} satisfies g^{ik}g_{kj}=δ^i_j.
It converts covector components into vector components and raises tensor indices.
17. Kronecker delta represents the identity map
The mixed tensor δ^i_j equals 1 when i=j and 0 otherwise in a basis representation.
It acts as the identity under contraction.
18. Levi-Civita symbols encode orientation
The alternating Levi-Civita symbol changes sign under odd permutations of indices and vanishes when indices repeat.
With appropriate density or tensor treatment, it helps express cross products, determinants and volume forms.
19. Tensors vary from point to point on manifolds
A tensor field assigns a tensor to each point smoothly.
Vector fields, metric fields and stress fields are examples.
Once tensors vary spatially, differentiation becomes subtle because tangent spaces at different points are distinct.
20. Ordinary partial derivatives are not generally tensorial
Differentiating tensor components with ordinary partial derivatives introduces extra terms under nonlinear coordinate changes.
A derivative suitable for geometry must correct for how the basis itself changes.
21. Covariant derivatives repair coordinate dependence
A connection defines a covariant derivative ∇ that differentiates tensor fields while preserving tensorial transformation behaviour.
In coordinates, correction terms involving connection coefficients are added to ordinary derivatives.
22. Christoffel symbols are connection coefficients, not tensor components
Christoffel symbols describe how coordinate basis vectors change and appear in covariant derivatives.
They do not transform as tensor components.
This is a classic warning: index notation alone does not make an object a tensor.
23. The Levi-Civita connection is determined by the metric
On a Riemannian or pseudo-Riemannian manifold, there is a unique torsion-free connection compatible with the metric.
This Levi-Civita connection defines parallel transport, geodesics and curvature.
24. Curvature is encoded by a rank-four tensor
The Riemann curvature tensor measures the failure of covariant derivatives to commute and the failure of parallel transport around a small loop to return a vector unchanged.
Curvature is therefore not merely visual bending; it is a tensorial obstruction.
25. Ricci curvature contracts the Riemann tensor
Contracting suitable indices of the Riemann tensor produces the Ricci tensor.
A further contraction with the inverse metric gives scalar curvature.
Each contraction compresses geometric information while preserving particular aggregate features.
26. Tensor analysis enters general relativity
General relativity represents spacetime through a Lorentzian metric and relates curvature to matter-energy structure.
Tensor equations ensure that the physical law is coordinate-independent.
The companion Mathematical Physics route places this tensor machinery inside the wider physical framework.
27. Stress is naturally a second-order tensor
In continuum mechanics, stress relates a surface normal direction to the traction force acting across that surface.
This dependence on two directions is naturally represented by a rank-two tensor rather than one vector.
28. Strain is also tensorial
Small-deformation strain tensors measure directional changes in length and angle.
Constitutive laws connect stress and strain through fourth-order elasticity tensors in anisotropic materials.
29. Inertia is tensorial when rotation depends on direction
A rigid body’s rotational response depends on how mass is distributed relative to the rotation axis.
The inertia tensor relates angular velocity and angular momentum in a coordinate-independent way.
30. Coordinate systems change components, not the physical object
A vector can have simple Cartesian components and complicated spherical-coordinate components while representing the same geometric direction.
Tensor analysis keeps this distinction explicit so equations can be compared across coordinate systems without confusing representation with reality.
31. Principal axes simplify symmetric tensors
A real symmetric rank-two tensor can be diagonalised in an orthonormal basis.
Its eigenvectors define principal directions and eigenvalues define principal magnitudes.
Stress and inertia tensors become especially interpretable in their principal-axis bases.
32. Tensor invariants survive basis change
Trace, determinant and eigenvalue combinations of suitable linear operators or rank-two tensors can produce basis-independent quantities.
Invariants provide compressed descriptions that do not depend on coordinate orientation.
33. Differential forms are antisymmetric tensors built for integration
A k-form is a fully antisymmetric covariant rank-k tensor field.
Forms integrate naturally over k-dimensional oriented surfaces and interact with the exterior derivative.
This connects tensor analysis to topology through Stokes’ theorem.
34. Tensor densities require extra transformation weight
Some coordinate expressions transform like tensors multiplied by powers of a Jacobian determinant.
These tensor densities appear in integration and orientation-sensitive constructions.
Recognising the distinction prevents incorrect use of symbols such as the Levi-Civita object.
35. A worked mechanism: coordinate-independent dot product
Let v have components v^i and w have components w^j. With metric g_{ij}, their inner product is g_{ij}v^i w^j.
- Change coordinates.
- The vector components transform contravariantly.
- The metric components transform covariantly in both indices.
- All transformation factors cancel under contraction.
- The resulting scalar is unchanged.
Tensor transformation rules are precisely what make the final geometric measurement independent of the chosen coordinates.
36. Common tensor-analysis failure modes
- Array=tensor confusion: assuming any indexed array is automatically tensorial.
- Index-position blindness: treating covariant and contravariant indices as interchangeable without a metric.
- Free-index mismatch: writing equations whose unsummed indices differ across terms.
- Christoffel confusion: treating connection coefficients as tensor components.
- Coordinate fixation: mistaking a convenient component representation for the geometric object itself.
- Partial-derivative misuse: differentiating tensor fields in curvilinear coordinates without connection corrections.
- Invariant overreach: assuming one contracted scalar retains all tensor information.
37. Tensor analysis as a mathematical machine
Coordinate-Independent Quantity → Basis Components → Transformation Law → Tensor Operations → Metric/Connection → Covariant Derivative → Invariants/Curvature → Geometric or Physical Interpretation.
38. What mastery looks like
- distinguish tensors from component arrays;
- understand vectors and covectors as dual structures;
- use tensor products, symmetry and contraction correctly;
- track free and dummy indices consistently;
- raise and lower indices using a metric;
- differentiate tensor fields using a connection;
- distinguish connection coefficients from tensors;
- interpret curvature through tensorial invariants;
- return tensor equations to geometry, mechanics or physics without confusing coordinates with the underlying object.
39. Conclusion
Tensor analysis works by separating geometric meaning from coordinate representation. Components change when the basis changes; the tensor does not. Metrics connect vectors and covectors. Contractions create invariants. Connections permit differentiation across changing tangent spaces. Curvature emerges from the failure of covariant derivatives to commute.
Linear algebra studies vectors and matrices in a chosen space. Tensor analysis supplies the transformation discipline needed when the coordinates themselves move, curve or change.
How Mathematics Works | Batch 10
- How Mathematics Works | Lie Groups & Lie Algebras
- How Mathematics Works | Harmonic Analysis
- How Mathematics Works | Calculus of Variations
- Tensor Analysis — this article
Return to the Mathematics Learning Hub or the How Mathematics Works root.