Topology Applications and Counterexamples | Data, Robotics, Analysis and Theorem Boundaries

Topology is useful twice: when it reveals structure that survives deformation, and when a counterexample shows exactly where a theorem stops working.

The first seven cells of the BTT Topology route built the language of open sets, metrics, compactness, connectedness, products, quotients, fundamental groups, homology, cohomology, manifolds and knots. This final cell turns those ideas outward. It asks where topology appears in other mathematical systems, how topological hypotheses control existence and stability, and why counterexamples are not curiosities but essential instruments for mathematical judgement.

This guide serves R22.08 · Applications and counterexamples in the BTT Mathematics Atlas. It is an advanced educational synthesis, not a claim that every modern application of topology is covered. Its job is to connect the preceding theory to recurring patterns in data, robotics, optimisation, analysis and geometry while teaching a disciplined habit: every theorem travels with hypotheses.

Topology route: Topological Spaces and Continuity · Metric Spaces and Examples · Compactness and Connectedness · Products and Quotients · Fundamental Groups and Covering Spaces · Homology and Cohomology · Manifolds, Knots and Geometric Topology.

Return route: BTT Mathematics Hub.

The simple answer

Topology studies properties preserved by continuous deformation and continuous maps. That perspective becomes useful whenever a problem depends more on connectivity, holes, neighbourhoods, continuity or global structure than on exact coordinates and distances.

A counterexample does the complementary job. A theorem often has the form “if assumptions A, B and C hold, then conclusion D follows.” A counterexample shows what can happen when one assumption is removed. Good counterexamples therefore map the boundary of valid reasoning.

Application asks: what structure survives enough to be useful? Counterexample asks: which hypothesis was carrying the theorem?

1. Topology in data: shape before coordinates

A data set can be represented as a cloud of points in a metric space. At one scale, points may appear isolated. At a larger scale, clusters connect. At still larger scales, loops may fill in. Topological data analysis studies how features such as connected components and holes appear and disappear as the scale changes.

Persistent homology provides one major framework. Build a filtration—a nested family of spaces indexed by scale—and compute homology at each stage. A connected component or cycle that exists only over a tiny scale interval may be unstable; a feature that persists over a wider interval may deserve more attention.

The topology does not remove modelling choices. Results depend on the metric, filtration, sampling density, noise and preprocessing. Persistent homology is therefore a structured measurement of shape across scale, not an automatic truth detector.

Worked example 1: a sampled circle

Imagine points sampled approximately around a circle. For a very small radius, every point may form its own connected component. As radius grows, neighbouring points connect and eventually form one component. Before the centre fills in, a one-dimensional cycle may appear, producing a nontrivial H₁ class. At a still larger radius, higher-dimensional simplices can fill the loop and the H₁ class disappears.

The interval between birth and death of that class is a persistence interval. The calculation translates a visual “ring-like” pattern into an algebraic record that can be compared across data sets.

2. Čech and Vietoris–Rips constructions

Two common ways to turn point clouds into simplicial complexes are Čech and Vietoris–Rips constructions.

• In a Čech complex at scale r, a finite collection of points spans a simplex when the corresponding radius-r balls have a common intersection.
• In a Vietoris–Rips complex at scale r, a finite collection spans a simplex when every pair is sufficiently close under the chosen threshold convention.

The Rips condition is computationally convenient because it is determined by pairwise distances. The Čech construction has a strong geometric relation to unions of balls through the nerve theorem. Their exact scale conventions vary in the literature, so formulas should state whether r or 2r is being used.

3. The nerve theorem: local overlaps recover global shape

Suppose a space is covered by sets whose finite nonempty intersections are contractible—a good cover. The nerve of the cover is a simplicial complex with one vertex for each cover set and a simplex whenever the corresponding sets have nonempty common intersection.

The nerve theorem says, under the suitable hypotheses, that the nerve has the same homotopy type as the covered space. This is a powerful compression principle: a continuous geometric object can be replaced by combinatorial overlap data without losing its homotopy type.

The hypothesis matters. Arbitrary covers do not automatically preserve the homotopy type through their nerves. The quality of intersections is doing mathematical work.

4. Sensor coverage as a topological problem

Suppose sensors each cover a region around their location. A practical question is whether the union of sensing regions has gaps. If full coordinate information is available, geometry can be used directly. If only overlap information is known, a nerve complex can encode which sensor regions intersect.

Under suitable convexity or good-cover conditions, homology of the nerve can detect coverage holes. The application illustrates topology’s strength: global connectivity and hole information can sometimes be recovered from local overlap relations.

But a detected homology class must still be interpreted within the physical model. Boundary effects, communication range, obstacles and imperfect sensing can change what the topological signal means operationally.

5. Robotics: configuration spaces turn motion into topology

A robot has many possible states: position, orientation and joint angles. The collection of all allowed states is its configuration space. Obstacles remove forbidden configurations. A motion-planning problem becomes a path-planning problem in the remaining free configuration space.

For a point robot moving in a room, configuration space may look much like the room itself. For a rigid body or articulated arm, configuration space can have several dimensions and circular coordinates. A planar rigid robot has two translational coordinates and one angular coordinate; the angle naturally lives on S¹.

Topology helps answer questions before metric optimisation begins: are the start and goal in the same path component? Are there distinct homotopy classes of paths around obstacles? Can one choose representatives from each class before minimising length or energy?

Worked example 2: two routes around an obstacle

Remove a closed disc obstacle from the plane. A path from a point on the left to a point on the right can pass above or below. Relative to fixed endpoints, these routes can represent different homotopy classes depending on winding around the obstacle. Optimisation within one class cannot automatically discover a shorter path in another class.

This suggests a two-stage algorithmic idea: first identify relevant topological path classes; then optimise geometrically within each class.

6. Topological complexity and motion planning

There is a formal invariant called topological complexity, introduced in the study of motion planning, that measures how many continuous local motion-planning rules are needed because a globally continuous planner may be impossible.

The deeper idea is accessible without the full theory: topology can obstruct one universal continuous rule. Discontinuity in planning may reflect the space itself rather than poor algorithm design.

That distinction matters in many systems. When a global rule cannot exist continuously, a robust design should use regions, modes or local planners rather than pretend one formula will remain stable everywhere.

7. Optimisation: topology helps guarantee existence

Optimisation asks for points that minimise or maximise an objective. Topology enters before differentiation. If a feasible set K is compact and f:K→R is continuous, the Extreme Value Theorem guarantees f attains both a minimum and a maximum.

This separates two questions:

• Existence: does an optimum actually occur?
• Computation: how can we find it efficiently?

Compactness may solve the first question even when the second is difficult.

Worked example 3: why an open feasible set can fail

Minimise f(x)=x on the open interval (0,1). The infimum is 0, but no point of the feasible set achieves it. The function is continuous and the set is bounded, but the missing boundary point prevents attainment.

On the compact set [0,1], the same function attains its minimum at 0 and maximum at 1. The change is topological, not computational.

8. Fixed-point theorems: topology constrains self-consistent states

A fixed point of f:X→X is a point x with f(x)=x. Fixed-point theorems state conditions under which such a point must exist.

Brouwer’s fixed-point theorem says that every continuous map from a closed n-dimensional ball to itself has a fixed point. In one dimension, every continuous f:[0,1]→[0,1] has a fixed point. Define g(x)=f(x)−x. Then g(0)≥0 and g(1)≤0, so the Intermediate Value Theorem gives g(c)=0.

The higher-dimensional theorem is deeper, but the lesson is similar: continuity plus the topology of a compact convex domain forces a self-consistent state.

Do not overgeneralise. Remove compactness, convexity or continuity and fixed points can disappear.

9. Analysis: compactness controls infinite behaviour

Many analysis proofs need to convert infinitely many local or sequential possibilities into a manageable limit. Compactness provides exactly this kind of control.

• In metric spaces, compactness implies every sequence has a convergent subsequence.
• Continuous functions on compact spaces are uniformly continuous when the codomain is metric.
• Continuous real-valued functions on compact spaces are bounded and attain extremes.
• Compactness helps turn pointwise choices of neighbourhoods into finite collections.

Functional analysis develops weaker topologies precisely because bounded sets that are noncompact in a strong norm topology can become compact or relatively compact in weaker topologies under suitable hypotheses. Topology is therefore part of the design of the analytical setting.

10. Dynamical systems: state spaces and invariant sets

A dynamical system evolves states over time. The state space may be Euclidean, a manifold, a quotient or a function space. Continuity determines whether nearby states have nearby images over one step; topology determines connected components, invariant regions and global orbit structure.

Periodic boundary conditions often create quotient spaces. An angular variable lives on S¹ rather than an interval with two physically distinct endpoints. Two angular variables naturally produce a torus T². Treating them as ordinary unconstrained real coordinates can obscure the true global geometry.

Topology does not replace stability analysis, differential equations or probability. It supplies the correct state space in which those analyses operate.

11. Networks: from graphs to higher-order topology

A graph records pairwise relationships through vertices and edges. A simplicial complex can additionally record higher-order interactions: triangles can represent three-way relationships that are present as a joint object rather than merely three pairwise links.

This distinction matters when studying systems where groups of agents interact collectively. Homology can then detect cycles and cavities in the higher-order relational structure.

However, adding a filled triangle is mathematically different from merely having all three graph edges. A clique complex chooses to fill every clique by a simplex; that is a modelling decision. The topology reflects the chosen complex, not the graph alone.

12. Shape matching and manifold-valued data

Some data naturally live on manifolds rather than in unconstrained Euclidean space. Directions live on spheres, rotations can be represented by rotation groups, and periodic phases live on circles.

Embedding such data in a large Euclidean coordinate system can be convenient, but the intrinsic topology may matter. A circle has no globally continuous one-dimensional coordinate without a cut. Rotation spaces have nontrivial global identifications. Algorithms that ignore these facts may create artificial discontinuities.

The lesson is not “always use advanced topology”. It is “model the state space honestly before choosing computational coordinates”.

13. Counterexample laboratory: arbitrary intersections of open sets

Every finite intersection of open sets is open. Why does the topology axiom not demand arbitrary intersections?

In R, each Uₙ=(−1/n,1/n) is open. But

⋂_n=1^∞Uₙ={0},

which is not open in the usual topology. The example shows that “finite” in the topology axiom is essential.

14. Counterexample: continuous bijection is not enough

A continuous bijection need not be a homeomorphism. Let X be R with the discrete topology and Y be R with the usual topology. The identity set map f:X→Y is continuous because every inverse image of a usual open set is open in discrete X. It is bijective.

But f⁻¹:Y→X is not continuous because a singleton {0} is open in X but not in Y. Therefore f is not a homeomorphism.

The compact-to-Hausdorff theorem explains one important repair: a continuous bijection from compact X to Hausdorff Y is automatically a homeomorphism.

15. Counterexample: bounded does not mean compact

The interval (0,1) is bounded in R but not compact. The infinite discrete metric space X with d(x,y)=1 for x≠y is also bounded, but not compact because for radius ε<1 every ε-ball is a singleton and finitely many cannot cover X.

“Closed and bounded implies compact” is the Heine–Borel theorem for subsets of finite-dimensional Euclidean space, not the definition of compactness in arbitrary metric spaces.

16. Counterexample: bounded and Cauchy do not guarantee a limit in the space

The space (0,1) with the usual metric is bounded but incomplete. The sequence xₙ=1/n lies in (0,1), is Cauchy, and converges in R to 0. But 0 is not in the space, so the sequence has no limit in (0,1).

This isolates completeness from boundedness. A metric space can be bounded and incomplete, bounded and complete, unbounded and complete, or unbounded and incomplete.

17. Counterexample: connected need not mean path connected

The topologist’s sine curve provides a classical example. Take the graph {(x,sin(1/x)):x>0} and its closure, which adds the vertical segment {0}×[−1,1]. The resulting space is connected but not path connected.

The graph oscillates infinitely often as x approaches zero. Its closure cannot be separated into two open pieces, yet points on the limiting vertical segment cannot all be joined to the oscillating graph by paths.

The example protects the implication direction: path connected ⇒ connected, but not conversely.

18. Counterexample: images of open sets under continuous maps

Continuity says inverse images of open sets are open. It does not say images of open sets are open.

The continuous map f:R→R, f(x)=x², sends the open interval (−1,1) to [0,1), which is not open in R.

If a map does send open sets to open sets, it is called an open map. Openness and continuity are separate properties.

19. Counterexample: quotient of Hausdorff can fail to be Hausdorff

Quotients can destroy separation. A standard construction starts with R and identifies all integers to a single point while leaving other points distinct. Depending on the exact quotient, related examples show that when equivalence classes accumulate badly, the quotient may fail Hausdorff separation.

A safer structural theorem is this: if X is compact Hausdorff and the equivalence relation is closed in X×X, then the quotient is Hausdorff. The closedness hypothesis prevents distinct equivalence classes from accumulating into inseparable points.

When working with quotients, do not inherit Hausdorffness by assumption. Check a theorem that actually preserves it.

20. Counterexample: sequence tests are not universal topology tests

In metric spaces, topology is completely controlled by sequences in many familiar ways: closedness can be tested by limits of sequences, and continuity can be tested sequentially.

In arbitrary topological spaces, sequences may not detect all closure behaviour. Nets and filters generalise sequences by allowing more flexible directed indexing and recover the full topological notion of convergence.

The practical lesson is not to abandon sequences. It is to state the setting. Metric and first-countable spaces justify powerful sequential shortcuts that cannot be promoted without conditions.

21. Counterexample: compactness and sequential compactness can separate

In metric spaces, compactness, sequential compactness and several related conditions coincide. In general topological spaces they can diverge.

This is why a proof that extracts a convergent subsequence is a metric-space proof pattern unless a more general theorem has been established. The open-cover definition remains the general definition of compactness.

Counterexamples in general topology are often constructed using products, ordinal spaces or unusual topologies. Their purpose is not artificial difficulty; they reveal which familiar equivalences depend on metrizability or countability.

22. Counterexample: product topology is not box topology

For finitely many factors, product and box topologies agree. For infinitely many factors, product-basic sets restrict only finitely many coordinates, whereas box-basic sets may restrict every coordinate at once.

On R^N, the box topology is strictly finer than the product topology. A set such as ∏_n(−1/n,1/n) is box-open because every coordinate is assigned an open interval. It is not product-open because no product-basic neighbourhood can impose infinitely many simultaneous coordinate restrictions.

The distinction changes convergence, compactness and continuity behaviour. The word “product” therefore names a precise topology, not just a Cartesian point set.

23. Counterexample: same π₁ does not mean same space

Both a point and S² have trivial fundamental group, but they are not homeomorphic and not homotopy equivalent. Their second homology groups differ: H₂(point)=0 while H₂(S²)≅Z.

Likewise, even matching homology groups do not generally force homeomorphism. Different spaces can share the same additive invariants.

An invariant can prove non-equivalence when values differ. Equal values usually mean only that the chosen test has not distinguished the spaces.

24. Counterexample: local sameness does not imply global sameness

Every connected 2-manifold is locally modelled on R², yet S², T² and higher-genus surfaces are not homeomorphic. Local charts see only neighbourhoods. Global topology records how those neighbourhoods are assembled.

This distinction recurs in many applications. A locally linear approximation can be excellent while the global state space still contains holes, identifications or multiple path classes.

Local models explain nearby behaviour. Global topology decides how the neighbourhoods fit together.

25. Counterexample: dense does not mean large in every sense

Q is countable and dense in R. Its complement R\Q is uncountable and also dense. Density means every nonempty open interval meets the set; it says nothing by itself about cardinality, measure or proportion.

Indeed Q has Lebesgue measure zero despite being dense. Topological size, measure-theoretic size and cardinal size are different notions.

26. Counterexample: open and closed are not opposites in ordinary language

A subset can be both open and closed. The empty set and whole space always are. In a disconnected space, nontrivial clopen subsets appear.

For X=(−2,−1)∪(1,2) with the subspace topology from R, each component is both open and closed in X. This is exactly how disconnectedness can be detected.

“Closed” means the complement is open, not “not open”.

27. Counterexample: completeness is not topological

The usual metric on R is complete. But R also admits metrics inducing the same usual topology that are incomplete. For example, transport the ordinary metric from the open interval (−π/2,π/2) through the homeomorphism arctan:R→(−π/2,π/2).

The resulting metric d′(x,y)=|arctan x−arctan y| induces the usual topology on R, but sequences with arctan values tending to π/2 can be Cauchy without a limit in R under d′.

Thus homeomorphisms preserve topology but not metric completeness.

28. Counterexample: compact does not imply finite

The interval [0,1] is infinite and compact. The Cantor set is uncountable and compact. Compactness is not a finiteness statement about the number of points.

Its finite character lies elsewhere: every open cover has a finite subcover. This is a finiteness principle about how neighbourhoods control the space.

29. Counterexample: a continuous map need not preserve distance, area or angle

Topology remembers continuity, not metric geometry. A homeomorphism can stretch lengths, distort angles and change areas drastically while preserving open-set structure.

For example, x↦x³ is a homeomorphism R→R but does not preserve ordinary distances. The map from an open interval to R given by a tangent function is a homeomorphism even though a bounded Euclidean interval is stretched across the entire line.

If length or angle matters, additional geometric structure is required.

30. A theorem-boundary workflow

When using a topological theorem, do not begin by recalling only the conclusion. Use this operating sequence:

1. Write the theorem with every hypothesis.
2. Match each hypothesis to the current space or map.
3. Mark which assumptions are structural: compact, Hausdorff, connected, metric, locally path connected, second countable and so on.
4. If one hypothesis is missing, do not silently continue.
5. Search for a repaired theorem or construct a counterexample.
6. Use the counterexample to identify what failure becomes possible.
7. Return to the original problem and decide whether extra structure can be proved.

This is the topology version of mathematical verification. A theorem is a conditional machine. The hypotheses are its input contract.

31. An application workflow

1. Identify the real objects and question.
2. Choose the state space or data space.
3. Decide which topology is mathematically justified.
4. Separate topological properties from metric or probabilistic properties.
5. Choose an invariant or theorem aligned with the task.
6. Check the hypotheses before applying it.
7. Compute or estimate the invariant.
8. Test sensitivity to modelling choices.
9. Return the result to the original domain with limitations stated.

Topological sophistication does not excuse a weak model. The topology is only as relevant as the space and relation it represents.

32. Worked synthesis: path planning around obstacles

Consider a point robot moving in the plane with several disjoint closed disc obstacles. The free configuration space is the plane with those discs removed.

• Topology: the obstacles create nontrivial loop classes.
• Homotopy: paths from start to goal can be grouped by how they wind around obstacles.
• Fundamental group: the free-space loop structure can distinguish winding behaviours.
• Optimisation: within each relevant path class, minimise length, energy or risk.
• Verification: ensure the final continuous path remains inside free space with required clearance.

Topology does not solve the numerical optimisation by itself. It organises the search so that globally different route families are not accidentally treated as one local optimisation basin.

33. Worked synthesis: compactness in model calibration

Suppose a continuous loss function L(θ) is defined on a compact parameter set K. Compactness guarantees that L attains a global minimum somewhere in K. This does not tell us which numerical algorithm finds it, whether the minimum is unique, or whether the model is scientifically adequate.

But it establishes an existence layer. If K were open or noncompact, an infimum might be approached indefinitely without being attained. The topological property therefore changes the mathematical status of the optimisation problem before computation begins.

34. Worked synthesis: topology in periodic variables

Angles θ and θ+2π describe the same physical orientation. Treating θ as an unconstrained point of R duplicates states. Treating it as a point of S¹ captures the identification directly.

For two independent periodic angles, the state space becomes S¹×S¹=T². A trajectory that crosses θ=2π and reappears at θ=0 is continuous on the circle even though a cut coordinate may appear to jump.

This is a concrete reason to distinguish coordinates from the space they represent.

35. Common misconceptions

• “Topology ignores geometry, so it cannot be applied.” It isolates a different layer of structure and often works alongside geometry.
• “Persistent homology automatically discovers meaningful features.” Results depend on metric, filtration and data quality.
• “A theorem conclusion is the theorem.” The hypotheses are part of the theorem.
• “One counterexample disproves topology.” A counterexample disproves an overgeneralised statement and sharpens the correct theorem.
• “If two spaces share an invariant, they are the same.” Most invariants are incomplete.
• “Compactness means physically small.” It is an open-cover property.
• “Local coordinates reveal global structure.” Charts can hide global identifications and holes.
• “The finest topology is always most informative.” Finer topologies change continuity and compactness; usefulness depends on the model.
• “Topological methods replace domain validation.” They provide mathematical information about a chosen model, not automatic real-world validity.

36. Practice set

1. Give an infinite intersection of open subsets of R that is not open.
2. Give a continuous bijection that is not a homeomorphism.
3. Give a bounded noncompact metric space.
4. Give a bounded incomplete metric space.
5. Name a connected space that is not path connected.
6. Give a continuous map that sends an open set to a nonopen image.
7. Explain why quotient Hausdorffness needs checking.
8. What replaces sequences when arbitrary topological convergence must be captured?
9. Explain the product-versus-box distinction for infinitely many factors.
10. Give two nonhomeomorphic spaces with the same trivial fundamental group.
11. Give a countable dense subset of R.
12. Give an uncountable dense subset of R.
13. Explain why completeness is not topological.
14. State the compact-domain extreme-value principle.
15. Why can configuration-space topology help robot path planning?
16. What does persistent homology track?
17. What theorem can relate the nerve of a good cover to the covered space?
18. What does Brouwer’s fixed-point theorem guarantee on a closed ball?
19. Why can periodic coordinates create artificial discontinuities?
20. What is the first question to ask before applying a theorem?

37. Answers and checks

1. ⋂_n≥1(−1/n,1/n)={0}.
2. The identity from discrete R to usual R is continuous and bijective but its inverse is not continuous.
3. An infinite set with the discrete metric is bounded by 1 but noncompact.
4. (0,1) with the usual metric.
5. The topologist’s sine curve closure.
6. x↦x² sends (−1,1) to [0,1).
7. Quotient maps can collapse accumulating equivalence classes so distinct quotient points may fail to have disjoint neighbourhoods.
8. Nets or filters.
9. Product-basic opens restrict finitely many coordinates; box-basic opens may restrict all coordinates.
10. A point and S².
11. Q.
12. R\Q.
13. Two metrics can induce the same topology while one is complete and another is not.
14. A continuous real-valued function on a compact space attains its maximum and minimum.
15. Obstacles create distinct path components or homotopy classes that can organise the search before geometric optimisation.
16. The birth and death of homology classes across a filtration parameter.
17. The nerve theorem.
18. Every continuous self-map of the closed n-ball has a fixed point.
19. A cut coordinate identifies one physical state with two endpoint values; S¹ represents the periodic state continuously.
20. Check every hypothesis and match it to the current setting.

38. What this topology route has built

The eight Atlas cells in R22 now form a coherent learning route: open sets and continuity; metric spaces; compactness and connectedness; products and quotients; fundamental groups and covering spaces; homology and cohomology; manifolds and knots; applications and counterexamples.

The sequence does not make topology “finished”. It creates an introductory-to-advanced bridge through the region’s agreed Atlas scope. Further directions include higher homotopy groups, fibre bundles, spectral sequences, characteristic classes, homotopical algebra, geometric group theory, low-dimensional topology and many specialised research areas.

The completion test remains practical: use definitions correctly, identify which properties are topological rather than metric, apply theorems with hypotheses attached, and use counterexamples to detect when an argument has crossed its valid boundary.

Return to the BTT Mathematics Hub.