Topology

Topology studies the properties of spaces that survive continuous deformation. It loosens measurement so shape can be discussed without depending on a particular ruler, coordinate system, or drawing. The central questions are not "how long is this edge?" or "what is this exact angle?" but "which points are near each other?", "which pieces are connected?", "what counts as a hole?", "what happens at the boundary?", and "which features survive a structure-preserving map?"

That makes topology one of the compendium's bridge subjects. It connects mathematics, graph theory, linear algebra, category theory, type theory, data visualization, maps, graphs, and training neural networks. A route network, a manifold, a knowledge graph, a loss landscape, a GIS layer, and an embedding space are not the same object, but topology gives each of them a language for connectedness, neighborhoods, boundaries, and deformation.

The useful editorial discipline is to keep topology precise. "Shape" can become a metaphor too quickly. A topology note should say what the space is, what transformations are allowed, which invariant is being used, and what the invariant proves. Without that contract, topology becomes decorative abstraction.

Topology matters whenever the relation among parts is more important than exact measurement:

• a graph has connected components, bridges, cycles, and embeddings before it has a pretty layout;
• a map depends on projection, boundary, adjacency, and region topology as much as coordinate values;
• a dataset may contain clusters, loops, voids, outliers, and strata that are hard to see in raw tables;
• a neural representation may preserve local neighborhoods while distorting global distance;
• a proof assistant may formalize spaces, paths, continuity, and equivalence in a machine-checkable foundation;
• a knowledge graph needs typed edges that separate similarity, adjacency, equivalence, containment, and derivation.

Topology is therefore both a mathematical subject and an indexing habit. It asks which relationships stay meaningful when the representation changes.

A reusable topology record should preserve:

• the object or system being modeled;
• the topological space, metric space, graph, simplicial complex, manifold, cover, quotient, or embedding used to represent it;
• the allowed transformations: homeomorphism, homotopy, isotopy, graph isomorphism, projection, simplification, or another equivalence relation;
• the invariants being computed or cited;
• assumptions about dimension, compactness, boundary, separability, smoothness, noise, sampling, and scale;
• examples and counterexamples that keep the definition honest;
• software, parameters, and data provenance when computation is involved;
• the interpretation boundary: what the topological claim proves, suggests, or leaves untouched.

This is the same discipline the compendium uses for data sources, data storage, and the semantic web: identify the object, identify the representation, preserve provenance, and avoid silently upgrading a derived interpretation into a fact.

The classroom slogan says topology allows stretching and bending but not tearing or gluing. The precise version is that topology studies spaces and continuous maps between spaces. Two spaces are topologically the same when there is a homeomorphism between them: a continuous map with a continuous inverse.

A coffee mug and a torus are the famous informal example because each has one hole. That example is useful only if the deformation preserves the relevant structure. It does not mean mugs and doughnuts are physically, materially, or semantically identical. It means they share a topological invariant under a chosen equivalence.

The practical habit is to name the equivalence relation first. Are two objects being compared up to exact equality, relabeling, deformation, homotopy, coordinate change, projection, discretization, or approximate numerical similarity? Different answers create different invariants.

Point-set topology begins with a set plus a collection of open sets. Open sets describe neighborhoods: the local regions a point can move inside while staying near itself. From that simple idea come the core definitions:

• continuity means inverse images of open sets are open;
• closed sets contain their limit points;
• bases generate topologies from simpler neighborhood collections;
• subspace topologies inherit openness from a larger space;
• product topologies describe spaces built from multiple coordinates;
• quotient topologies describe spaces built by identifying points;
• compactness generalizes finite containment and good limiting behavior;
• connectedness distinguishes one-piece spaces from spaces that split apart;
• separation axioms describe how cleanly points can be distinguished.

This vocabulary can feel austere because it is designed to travel. It applies to real lines, manifolds, function spaces, spectra, finite spaces, state spaces, and logical models. It is not only geometry with the measurements erased; it is a general language for locality and continuity.

Topology becomes clearer when the main equivalence relations are separated:

• homeomorphism preserves the full topological structure;
• homotopy equivalence preserves shape at a coarser level, often allowing deformation through continuous paths;
• isotopy tracks one embedding deforming into another without cutting through forbidden space;
• diffeomorphism preserves smooth structure on differentiable manifolds;
• isometry preserves metric distance and is therefore stronger than homeomorphism;
• graph isomorphism preserves adjacency in a discrete graph;
• weak equivalence and related categorical notions preserve selected structure in homotopical settings.

The ladder matters because an invariant is only meaningful relative to the transformations it survives. Connected components survive homeomorphism. Homology survives homotopy equivalence. A metric diameter survives isometry but not arbitrary homeomorphism. A force-directed network cluster may not survive a different layout algorithm at all.

An invariant is a feature that remains unchanged under a chosen transformation. Topology is powerful because many useful invariants are insensitive to local distortion:

• number of connected components;
• compactness and connectedness;
• orientability;
• boundary and dimension;
• Euler characteristic;
• fundamental group;
• homology and cohomology groups;
• Betti numbers;
• knot invariants;
• persistent homology across scales.

The invariant is not the object. It is a compressed answer to a specific question. A Betti number can say how many independent loops or voids a model contains, but it does not explain their domain meaning. That interpretation has to come from the source data, modeling choices, and neighboring evidence.

Good topology is built from examples and counterexamples:

• The real line and an open interval are homeomorphic, even though one may look larger in a drawing.
• A circle is compact and connected; two disjoint circles are compact but not connected.
• A disk and a point are homotopy equivalent, but they are not homeomorphic.
• A sphere and a torus are not homeomorphic because their hole structure differs.
• A Mobius strip is non-orientable, while a cylinder is orientable.
• A quotient space can create new topology by identifying points, as when a square with paired edges becomes a torus.
• Finite topological spaces are legitimate spaces even when they do not look geometric.

Counterexamples are especially valuable for theorem proving. Formal topology libraries need strange spaces, small spaces, and pathological cases because they expose missing assumptions in otherwise plausible claims.

A manifold is a space that locally resembles ordinary Euclidean space. A curve looks locally like a line. A surface looks locally like a plane. Higher-dimensional manifolds generalize that local regularity.

Manifolds are a meeting point for topology, geometry, analysis, physics, robotics, and visualization. Topology asks which global features persist under deformation. Geometry adds metric, curvature, and measurement. Differential topology adds smooth structure. Computation adds meshes, charts, coordinates, and numerical methods.

Boundaries matter. A disk has a boundary circle; a sphere has no boundary; a road network has terminal nodes; a mapped polygon has edges where inclusion changes. Confusing boundary with decoration causes real errors in maps, simulations, datasets, and UI diagrams.

Algebraic topology translates spaces into algebraic objects. The goal is not to replace shape with symbols for their own sake. The goal is to compute and compare invariants.

Homotopy studies continuous deformation of maps. The fundamental group records loop structure based at a point. Homology studies holes through chains, cycles, and boundaries. Cohomology adds dual structure and richer products. These constructions link topology to linear algebra because many computations reduce to matrices over a coefficient field or ring.

The computational recipe is often concrete:

1. Represent the space with cells, simplices, cubical complexes, or another combinatorial model.
2. Build boundary maps that say how higher-dimensional pieces attach to lower-dimensional pieces.
3. Compute cycles and boundaries.
4. Interpret what remains as homology.

This is why a page about topology should also link to graph theory. Graphs are one-dimensional complexes, and higher-dimensional complexes add triangles, tetrahedra, and beyond.

Topology enters graph work in several ways. A graph has connectivity, cycles, bridges, cuts, and planarity. A graph embedded in a surface raises questions about crossings, faces, genus, and duality. A knowledge graph has local neighborhoods, bridge nodes, isolated components, overloaded hubs, and paths that support discovery.

For data visualization, topology is the reason network drawings require caution. A layout can make a cluster look stable when the underlying graph only supports a weak relation. A good visualization should preserve the difference between graph topology, visual proximity, edge weight, and semantic meaning.

For maps, topology tracks adjacency, containment, connectivity, and boundary consistency. GIS software has to know whether polygons close, whether roads connect, whether holes sit inside regions, and whether simplified geometries preserve the relationships needed for the task.

Applied topology uses topological ideas on empirical, computational, or engineered systems. Topological data analysis, often abbreviated TDA, is the best-known version. It tries to extract stable shape from noisy data.

Common TDA methods include:

• persistent homology, which computes topological features across a range of scales;
• persistence diagrams and barcodes, which visualize when features appear and disappear;
• Mapper, which builds a graph-like summary from a cover and clustering rule;
• witness complexes, which approximate larger datasets using landmark points;
• cubical complexes, which suit images, volumes, and grid data.

These methods are useful when raw coordinates are noisy but global shape still carries signal. They can help inspect sensor data, embeddings, materials, biological shapes, road networks, images, and model representations. They can also mislead when scale, distance metric, sampling density, or preprocessing choices dominate the result.

A topological computation should store enough context to be rerun and criticized:

• source observations and snapshot date;
• coordinate system, feature space, or graph model;
• distance metric or similarity rule;
• normalization, filtering, denoising, and sampling choices;
• complex type and filtration rule;
• scale range and coefficient field;
• clustering method and cover parameters for Mapper-style analyses;
• software package, version, and random seeds;
• output diagrams, summaries, and interpretation notes.

A persistence diagram without those choices is hard to reuse. A Mapper graph without its cover and clustering parameters is not evidence; it is a picture. This is why topology belongs beside data sources, data storage, Python, and software libraries.

Topology has a deep relationship with type theory. In homotopy type theory, types can be interpreted as spaces, terms as points, and equalities as paths. This gives equality a geometric flavor and makes topology part of the foundation of mathematics rather than only a branch of geometry.

Proof assistants also make ordinary topology more durable. Definitions of compactness, continuity, filters, manifolds, metric spaces, and homology can be checked mechanically. The challenge is not only formal proof. It is choosing definitions that are reusable, compositional, and close enough to mathematical practice that later theorems can build on them.

This connects topology to category theory as well. Many topological constructions are best understood by maps, functors, universal properties, adjunctions, limits, colimits, and equivalences.

Topology is often about what survives a change of scale. A connected component may vanish when sampling changes. A loop may be real structure, sensor noise, or a consequence of the distance metric. A cluster may be stable only under one cover. Good topological writing should therefore name the scale at which the claim is being made.

For compendium use, this is the bridge between abstraction and evidence. If a page says an embedding space has holes, a route network has bottlenecks, or a corpus has clusters, the note should preserve the distance rule, sampling method, filtration, cover, and interpretation boundary. That keeps topology connected to data visualization, graphs, maps, and linear algebra without treating every visual shape as a mathematical invariant.

Applied topology should make parameter sensitivity visible. If a feature appears only under one distance metric, one normalization, one cover size, or one sampling rule, the result may still be useful, but it should be labeled as fragile. If the same component, loop, bottleneck, or boundary survives across reasonable choices, the claim becomes more persuasive.

A reusable note should therefore record comparison runs, not just the favorite output. For persistent homology, preserve the filtration, coefficient field, scale range, and threshold used to interpret bars. For Mapper, preserve the lens function, cover resolution, overlap, clustering rule, and pruning choices. For graph or route-network examples, preserve whether the topology came from physical adjacency, travel time, directed edges, projected coordinates, or editorial relationships. Each choice changes what "near," "connected," or "hole" means.

This is where topology meets data sources, data storage, graphs, maps, and training neural networks. A topological claim about an embedding space should be rebuildable from the model, dataset, preprocessing, projection, and parameter set. A topological claim about a map should preserve projection, boundary, topology rules, and source date. A topological claim about a knowledge graph should preserve the edge types included and excluded.

Good display does not need to show every run. It should show whether the feature is formal, computed, stable across parameters, fragile, or only illustrative. That one label helps readers keep proof, computation, and visual intuition in the right boxes.

Topological claims need status labels because the same word "shape" can point to different kinds of evidence. Useful labels include formal theorem, definition, computed invariant, visualization, heuristic analogy, empirical pattern, and modeling assumption. A formal theorem depends on stated definitions. A computed invariant depends on data, software, and parameters. A visualization helps a reader see structure but is not itself proof. A heuristic analogy can guide interpretation without carrying mathematical authority.

These labels keep topology useful in applied pages. A loop in an embedding, a bottleneck in a route network, a cluster in a knowledge graph, and a boundary on a map may all be visually compelling, but each needs a different support chain. The graph should connect the claim to the represented object, transformation, invariant, computation, and caveat so readers can tell whether they are looking at proof, measurement, or suggestive structure.

Topology gives the compendium graph a vocabulary for relationship structure. Useful predicates include:

• has_space;
• has_topology;
• has_metric;
• has_neighborhood;
• has_boundary;
• has_component;
• has_hole;
• maps_to;
• homeomorphic_to;
• homotopy_equivalent_to;
• embedded_in;
• quotient_of;
• deformation_retracts_to;
• has_invariant;
• computed_by;
• parameterized_by;
• visualized_as;
• derived_from_source.

Those edges let the graph distinguish local adjacency, visual proximity, semantic similarity, formal equivalence, and computed evidence. That distinction is crucial for a feature-complete wiki: a reader should be able to move from a topological claim to the object, representation, invariant, computation, and source that support it.

Topology fails when the abstraction outruns the evidence:

• saying two things have the same "shape" without naming the equivalence;
• confusing a visualization layout with graph structure;
• treating a metric-space result as if it were purely topological;
• ignoring boundary and coordinate-system assumptions in maps;
• computing persistent homology without preserving scale and preprocessing choices;
• treating a cluster or loop in an embedding as domain truth;
• using advanced vocabulary where a simple graph, table, or diagram would answer the question.

The fix is not to avoid abstraction. The fix is to make the modeling boundary visible.

Start with metric spaces, open sets, continuity, compactness, and connectedness. Then study product spaces, quotient spaces, separation axioms, manifolds, homotopy, the fundamental group, homology, and cohomology. For applied work, learn enough linear algebra, graph theory, and data visualization to understand how topological invariants are represented and displayed.

For formal foundations, follow topology into type theory, category theory, and theorem proving. For empirical systems, follow it into data sources, graphs, maps, and training neural networks.

• Encyclopedia of Mathematics on general topology (opens in new tab)
• MathWorld on topology (opens in new tab)
• nLab on topology (opens in new tab)
• pi-Base, a database of topological spaces (opens in new tab)
• Allen Hatcher's Algebraic Topology (opens in new tab)
• The Homotopy Type Theory book (opens in new tab)
• GUDHI (opens in new tab)
• Ripser (opens in new tab)
• Dionysus (opens in new tab)
• JavaPlex (opens in new tab)
• HomCloud (opens in new tab)

• Mathematics for proof, abstraction, and the larger map of mathematical branches.
• Graph Theory for connectivity, cycles, planarity, cuts, and discrete structure.
• Linear Algebra for boundary matrices, spectra, embeddings, and numerical computation.
• Category Theory for maps, functors, universal constructions, and equivalence.
• Type Theory for homotopy type theory and machine-checkable foundations.
• Data Visualization for drawing spaces, networks, and invariants without misleading readers.
• Maps for boundary, projection, adjacency, containment, and geospatial topology.
• Graphs for knowledge-graph neighborhoods, graph databases, and retrieval paths.
• Training Neural Networks for representation spaces, embeddings, and model diagnostics.