Topological Invariants

A topological invariant is a quantity that stays unchanged under a specified class of continuous deformations. It records global structure that cannot be removed by stretching, bending, changing coordinates, or changing gauge.

The words “specified class” matter. The same number may be invariant under homotopy of loops, under homeomorphism of spaces, under smooth gauge changes of a bundle, or under gap-preserving deformations of a Hamiltonian. A useful topological statement always says what objects are being compared and what deformations are allowed.

This page gives the organizing language behind winding numbers, Chern numbers, flat holonomy on spaces with holes, and the topological quantum numbers that appear later in quantum matter. Topological Order Preview shows why a many-body phase generally needs a package of global data rather than one invariant in isolation.

Basic Definition

Let $\mathcal C$ be a class of objects and let $\sim$ be an equivalence relation on those objects. An invariant is a rule

$$I:\mathcal C\to S$$

such that

$$X\sim Y \quad\Longrightarrow\quad I(X)=I(Y).$$

For topology, $X\sim Y$ usually means that $X$ and $Y$ are related by a deformation that preserves the relevant global structure. Examples include:

• two spaces related by a homeomorphism;
• two loops related by a homotopy that stays inside the allowed space;
• two bundles related by a bundle isomorphism;
• two gapped Hamiltonian families related by a continuous path that never closes the relevant gap;
• two symmetry-protected systems related by a continuous deformation that preserves the protecting symmetry.

The target set $S$ is often discrete, such as $\mathbb Z$, $\mathbb Z_2$, or a finite group. It may also be a group, ring, vector space, or more structured object. The invariant does not have to be a single number, but numbers are the easiest entry point.

What Must Be Specified

Before calling something a topological invariant, specify four pieces of data.

First, specify the objects. Are they loops, manifolds, vector bundles, projectors, Hamiltonians, or phases?

Second, specify the allowed equivalence. Are coordinate changes allowed? Are gauge transformations allowed? Must a spectral gap stay open? Must time-reversal symmetry, particle-number conservation, or lattice translation symmetry be preserved?

Third, specify the normalization. A winding number needs an orientation convention. A Chern number needs the factor $1/(2\pi)$ in the $U(1)$ convention used here. A Hall-conductance formula also needs charge and orientation conventions.

Fourth, specify the failure modes. A topological invariant can change when the assumptions defining it fail: a loop crosses an excluded point, a band touches another band, a gauge choice becomes singular, a boundary condition changes, or a protecting symmetry is broken.

These details are not pedantry. They are what separates a robust topological statement from a slogan.

Robustness from Discreteness

The common robustness argument is simple. Suppose a family of objects $X_\lambda$ depends continuously on a parameter

$$0\le \lambda\le 1,$$

and an invariant $I(X_\lambda)$ is integer-valued and changes continuously as long as all defining assumptions remain true. Since $\mathbb Z$ is discrete, the only continuous path in $\mathbb Z$ is a constant path. Therefore

$$I(X_\lambda)=I(X_0)$$

throughout the deformation.

This is the mathematical skeleton behind many physical statements:

• a winding number cannot change unless a loop crosses the excluded point or the map becomes ill-defined;
• a band Chern number cannot change during a smooth deformation of a band projector unless the band gap closes;
• a quantized response coefficient cannot drift continuously if it is tied to an integer invariant and the hypotheses remain valid.

The argument does not say that every observable is unchanged. Energies, local curvature profiles, wavefunction shapes, and matrix elements may change continuously. The topological invariant stays fixed because it belongs to a discrete classification.

Coordinate, Gauge, and Topological Invariance

Three kinds of invariance are often confused.

A coordinate-invariant quantity is independent of the coordinates used to describe the same object. For example, the integral of a differential form over an oriented manifold is not supposed to depend on a chart after the orientation and form are specified.

A gauge-invariant quantity is independent of a local choice of representative. Berry curvature is gauge invariant in the abelian case:

$$A\mapsto A-d\chi, \qquad F=dA\mapsto F.$$

But local Berry curvature $F(R)$ is not by itself a topological invariant. It can move around the parameter space under smooth deformations.

A topological invariant is stable under a larger deformation class. The Chern number

$$C = \frac{1}{2\pi} \int_M F$$

is topological when $M$ is a closed oriented surface and the line bundle or isolated band remains well-defined. The local curvature distribution may change, but the normalized integral cannot change without violating the assumptions.

Thus:

• gauge invariance means independence from local representative choices;
• coordinate invariance means independence from chart choices;
• topological invariance means stability under allowed global deformations.

The same expression may have more than one kind of invariance, but the reasons are different.

Standard Examples

Connected components are among the simplest invariants. The number of connected components cannot change under a homeomorphism. A continuous deformation can merge two components only if the space itself is changed in a way that violates the comparison.

The fundamental group $\pi_1(X,p)$ records based-loop homotopy classes. For the circle and the punctured plane, the relevant information is an integer winding number. In the punctured plane,

$$w(\gamma) = \frac{1}{2\pi} \oint_\gamma d\theta$$

counts how many times the loop winds around the missing origin.

The Euler characteristic is a topological invariant of many familiar spaces. For a triangulated closed surface it can be computed as

$$\chi = V-E+F_{\mathrm{faces}},$$

where $V$, $E$, and $F_{\mathrm{faces}}$ count vertices, edges, and faces. Different triangulations give the same $\chi$ for the same topological surface.

Chern numbers are characteristic numbers of complex vector bundles. In the basic $U(1)$ line-bundle case over a closed oriented surface,

$$C_1(L) = \frac{1}{2\pi} \int_M F \in\mathbb Z.$$

They detect a global obstruction to choosing one smooth gauge over all of $M$.

More advanced quantum-matter examples include $\mathbb Z_2$ invariants in time-reversal-invariant systems, winding numbers of chiral Hamiltonians, and higher Chern classes. Those require extra structure and are not developed on this first toolkit page.

Quantum-Mechanical Examples

Topological invariants enter quantum mechanics when state spaces, parameter spaces, boundary conditions, or eigenbundles have global structure.

For a particle on a ring, the configuration coordinate is an angle. Periodic boundary conditions and phase winding lead to integer mode labels. The integer is not a local force; it is a global compatibility condition around $S^1$.

For the Aharonov–Bohm geometry, the accessible region can be a plane with a flux region removed. A loop that winds around the removed region is not homotopic to a loop that misses it. The electromagnetic field strength may vanish on the accessible paths, while the holonomy around a noncontractible loop still has physical phase content.

For Berry phase, a closed path in parameter space can have a holonomy determined by the Berry connection. If the connection is flat on a region with noncontractible loops, the holonomy may still depend on the loop’s homotopy class. If curvature is present, the phase may also depend on geometric area and local curvature flux.

For a two-dimensional Bloch band, the Brillouin zone is a torus. An isolated band defines a line bundle over that torus, and the band Chern number is a global invariant of the band projector. This invariant is the mathematical integer behind the Thouless–Kohmoto–Nightingale–den Nijs description of quantized Hall conductance.

For a degeneracy in a parameter space, a small sphere enclosing the degeneracy can carry nonzero Berry flux. The resulting Chern number detects the enclosed singularity in the eigenstate bundle, much as winding around a puncture detects a missing point.

What Can Make an Invariant Change

A topological invariant can change when the object stops satisfying the assumptions under which the invariant was defined.

Common quantum-mechanical failure modes include:

• a loop crosses an excluded degeneracy, flux tube, or singular point;
• a map into $U(1)$ passes through zero, so its phase is no longer defined;
• an eigenvalue crosses another eigenvalue, so an eigenline or isolated band ceases to exist;
• a spectral gap closes somewhere in the Brillouin zone or parameter space;
• a protecting symmetry is broken during the deformation;
• the base space or boundary conditions are changed;
• the parameter space is not compact or the relevant integral is not controlled at infinity.

These events are not small corrections to the invariant. They are changes in the problem statement. The invariant was never required to stay fixed across them.

Invariants Are Not Always Complete

An invariant can distinguish some objects without classifying all objects.

If $I(X)\ne I(Y)$, then $X$ and $Y$ cannot be equivalent under the chosen relation. But if $I(X)=I(Y)$, the objects may or may not be equivalent. The invariant might be too coarse to detect the difference.

For example, winding number completely classifies loops in the punctured plane up to the usual homotopy relation, but a single integer cannot classify all possible manifolds, all vector bundles in all dimensions, or all topological phases of matter. Chern numbers are powerful, but they are not a universal answer to every topological classification problem.

This is why a serious topological statement identifies both the invariant and the classification theorem, if such a theorem is being claimed.

Local Data versus Global Obstruction

Many topological invariants are computed using local data but measure a global obstruction.

For Chern numbers, the local data are connection one-forms $A_\alpha$ and curvature two-forms $F$. On overlaps,

$$A_\beta = A_\alpha-d\chi_{\beta\alpha}.$$

The local curvature forms patch into a global two-form. The integral of that global curvature can be nonzero even though each patch has $F=dA_\alpha$ locally. The nonzero integer measures the failure to choose one smooth global frame.

For winding, the local data are angle functions. Locally one may write $d\theta$, but the angle $\theta$ is not a globally single-valued function on the punctured plane. The winding number measures that global failure.

The pattern is the same:

• local descriptions exist on patches;
• transition data or singular exclusions appear on overlaps or holes;
• a global invariant records what cannot be removed by changing local descriptions.

Reading Physical Claims Carefully

When a paper or textbook says that a feature is “topological”, ask:

• What is the mathematical object: loop, map, bundle, projector, Hamiltonian family, or phase?
• What is the equivalence relation: homotopy, gauge equivalence, homeomorphism, unitary equivalence, or gap-preserving deformation?
• What invariant is being used: winding number, Chern number, Berry phase modulo $2\pi$, a symmetry-protected index, or something else?
• What assumptions keep it fixed: compactness, smoothness, isolation of a band, locality, symmetry, or boundary conditions?
• What physical event can change it: gap closing, singularity, defect crossing, symmetry breaking, or a change of domain?

This checklist prevents two opposite mistakes: treating every robust-looking feature as topological, and missing genuine global structure because the local equations look ordinary.

Common Mistakes

• Calling any integer a topological invariant without specifying the allowed deformations.
• Confusing gauge invariance with topological invariance.
• Treating local Berry curvature as topological rather than the correctly normalized global integral.
• Forgetting that a topological invariant can change when a spectral gap closes.
• Assuming equal invariant values prove two objects are equivalent.
• Ignoring orientation conventions in winding numbers and Chern numbers.
• Applying compact-space formulas to noncompact parameter spaces without boundary or decay conditions.
• Overstating topological protection: an invariant protects only the features tied to the hypotheses of the invariant.

Cross-Links

• Manifolds, First Look
• Homotopy and Winding
• Chern Numbers
• Holonomy
• Fiber Bundles, First Look
• U(1) Bundles and Quantum Phase
• Berry Connection as a Mathematical Object
• Topology in Quantum Matter — gapped deformations, physical protection assumptions, boundaries, and response.
• Berry Phase
• Boundary Conditions
• Time Reversal

References

• J. R. Munkres, Topology, 2nd ed., Prentice Hall, 2000.
• A. Hatcher, Algebraic Topology, Cambridge University Press, 2002.
• M. Nakahara, Geometry, Topology and Physics, 2nd ed., CRC Press, 2003.
• T. Frankel, The Geometry of Physics, 3rd ed., Cambridge University Press, 2011.
• B. Simon, “Holonomy, the quantum adiabatic theorem, and Berry’s phase,” Physical Review Letters 51, 2167-2170, 1983.
• D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, “Quantized Hall conductance in a two-dimensional periodic potential,” Physical Review Letters 49, 405-408, 1982.
• M. Z. Hasan and C. L. Kane, “Colloquium: Topological insulators,” Reviews of Modern Physics 82, 3045-3067, 2010.

Exercises

1. Let $I(X)$ be an integer-valued quantity that depends continuously on a deformation parameter $\lambda\in[0,1]$ as long as a spectral gap remains open. Explain why $I$ cannot change during a gap-preserving deformation.

Solution

The interval $[0,1]$ is connected, while $\mathbb Z$ is discrete. A continuous map from a connected interval to a discrete set must be constant. Therefore $I(X_\lambda)$ cannot change unless the assumptions allowing continuity fail. In the stated setting, the relevant failure is a gap closing.

2. Give an example of a quantity that is gauge invariant but not topological.

Solution

In an abelian Berry problem, the curvature two-form $F=dA$ is gauge invariant under $A\mapsto A-d\chi$. However, the local value of the curvature can change under smooth deformations of the Hamiltonian or connection. The Chern number $(1/2\pi)\int_MF$ over a closed surface is the topological invariant, not the local curvature profile.

3. A loop in $\mathbb R^2\setminus\{0\}$ has winding number $2$. Can it be deformed to a loop of winding number $0$ while staying in $\mathbb R^2\setminus\{0\}$?

Solution

No. The winding number is invariant under homotopies that remain in the punctured plane. To change from winding number $2$ to winding number $0$, the loop would have to cross the removed origin or otherwise leave the allowed class of loops.

4. A two-dimensional isolated band over the Brillouin-zone torus has Chern number $1$. The Hamiltonian is smoothly deformed, and the band remains separated from all other bands everywhere on the torus. What happens to the Chern number?

Solution

It remains $1$. The isolated band continues to define a smooth line bundle or projector over the torus. Since the Chern number is integer-valued and stable under gap-preserving deformations, it cannot change without a band touching or another failure of the hypotheses.

5. Why is it incomplete to say only “this phase is topological”?

Solution

The statement does not identify the object, the equivalence relation, the invariant, or the assumptions. A clearer statement would say, for example, that a Chern number of an isolated Bloch band is unchanged under smooth deformations that preserve the band gap, or that a winding number of a loop is unchanged under homotopy in a punctured space.