Loading field map
Differential topology is the study of smooth manifolds and the smooth maps between them. A smooth manifold is a space that locally resembles Euclidean space, but with enough structure to define what it means for a function or map to be differentiable. The subject asks which properties of such spaces are invariant under smooth deformation, and it seeks to classify manifolds up to diffeomorphism—the smooth analogue of homeomorphism, where two manifolds are considered the same if there is a bijective smooth map between them whose inverse is also smooth.
The central objects are therefore not rigid geometric shapes with distances and angles, but flexible surfaces and higher-dimensional spaces that can be bent, stretched, and smoothly reshaped. What matters is not how a manifold is embedded in some ambient space, but its intrinsic structure: the way its points can be smoothly coordinated, the tangent spaces attached to each point, and the global ways these local data fit together.
Three enduring questions organize much of the field. First, classification: when are two smooth manifolds diffeomorphic? This is generally impossible to answer completely for all manifolds, but the field has developed powerful invariants—algebraic, topological, and analytic—that can distinguish manifolds or prove them equivalent. Second, existence: which abstractly defined spaces admit smooth structures, and which smooth structures are possible on a given topological manifold? This question became unexpectedly subtle in the late twentieth century. Third, structure: what can be said about the geometric or topological features that every smooth manifold of a given type must possess, such as the presence of critical points of functions, the existence of special submanifolds, or the behavior of flows?
A fourth, more methodological question runs through the subject: how much of the topology of a manifold is captured by its smooth structure, and how much is independent of it? This distinction between topological and smooth information is one of the field's deepest recurring themes.
The roots of differential topology lie in nineteenth-century analysis and geometry, particularly in the study of curves and surfaces, and in the work of Bernhard Riemann, who conceived of spaces that are locally Euclidean but globally curved. However, the modern subject emerged only in the mid-twentieth century, when mathematicians began to treat smooth manifolds as abstract objects in their own right rather than as subsets of Euclidean space. The work of Hassler Whitney in the 1930s and 1940s established foundational results: every smooth manifold can be embedded in Euclidean space, and every topological manifold of dimension at least five can be given a smooth structure. These results made it possible to study manifolds intrinsically while still using the tools of Euclidean analysis.
A decisive turning point came with the development of Morse theory, initiated by Marston Morse in the 1920s and 1930s and later refined by many others. Morse theory studies a smooth real-valued function on a compact manifold and shows how the topology of the manifold is encoded in the critical points of the function—points where the derivative vanishes. The manifold can be built up by attaching handles corresponding to these critical points, and the sequence of critical values describes a filtration of the manifold by simpler pieces. This provided a powerful bridge between analysis and topology, and it remains a central tool.
The 1950s and 1960s saw the field mature into its modern form. The work of John Milnor, René Thom, Stephen Smale, and others produced striking results: the classification of exotic smooth structures on spheres, the proof of the higher-dimensional Poincaré conjecture via the h-cobordism theorem, and the development of cobordism theory, which studies when a manifold is the boundary of another manifold. These achievements established differential topology as a distinct discipline with its own methods and questions, separate from both algebraic topology and differential geometry.
Differential topology is not divided into rival schools in the way that, say, foundations of mathematics or philosophy of science are. Rather, it is organized around a set of complementary methods that address different aspects of the same questions. These methods are best understood as tools in a shared toolkit, each with its own strengths and limitations.
Morse theory provides a way to decompose a smooth manifold into simple building blocks. A smooth function on a compact manifold, if chosen generically, has only nondegenerate critical points—points where the Hessian matrix is nonsingular. The index of a critical point is the number of negative eigenvalues of the Hessian, and it determines the dimension of the handle that must be attached to build the manifold. The sequence of critical points, read in order of increasing function value, gives a handle decomposition of the manifold.
This approach is particularly powerful for proving existence results: if one can construct a manifold with a given handle decomposition, one can often show that a smooth manifold with prescribed properties exists. It also underlies the proof of the h-cobordism theorem, which states that if a compact manifold has two boundary components and is simply connected, and if certain homological conditions hold, then the manifold is a product of one boundary with an interval. This theorem, proved by Smale in the early 1960s, was the key to the higher-dimensional Poincaré conjecture and to the classification of simply connected manifolds of dimension at least five.
The limitation of Morse theory is that it requires a choice of function, and different functions can give very different handle decompositions. Understanding when two decompositions describe the same manifold is a difficult problem, addressed by the theory of handle moves and by Cerf theory, which studies how handle decompositions change as the Morse function is deformed.
A second fundamental method is the systematic use of transversality. Two submanifolds of a manifold are transverse if at every point of intersection, their tangent spaces together span the tangent space of the ambient manifold. Transverse intersections are stable: small perturbations of the submanifolds do not change the intersection pattern. The key principle, due largely to Thom, is that transversality is a generic property—any smooth map can be slightly perturbed to make it transverse to a given submanifold, and the space of transverse maps is dense in the space of all smooth maps.
This principle underlies many constructions in the field. It allows one to define intersection numbers, to count the number of solutions to equations in a way that is independent of perturbations, and to prove that certain desirable properties hold for almost all choices of a geometric object. For example, the Whitney embedding theorem uses transversality to show that a smooth map can be perturbed to an embedding, provided the dimension of the ambient space is large enough.
The limitation of transversality is that it is a qualitative tool: it tells one that generic objects behave well, but it does not by itself produce explicit constructions or compute invariants. It must be combined with other methods to yield quantitative results.
Cobordism theory studies manifolds up to the relation of being the boundary of a higher-dimensional manifold. Two closed manifolds are cobordant if their disjoint union is the boundary of some compact manifold. The set of cobordism classes forms a ring under disjoint union and Cartesian product, and this ring was completely computed by Thom in the 1950s using methods from algebraic topology. The result is that cobordism classes are determined by certain characteristic numbers—integrals of characteristic classes over the manifold.
Characteristic classes are cohomology classes associated to vector bundles, and they provide invariants of manifolds via their tangent bundles. The most important are the Stiefel–Whitney classes, which take values in mod-2 cohomology, and the Pontryagin classes, which take values in integer cohomology. These classes are natural under pullback and satisfy certain axioms, and they can be used to distinguish manifolds that are not diffeomorphic.
The relationship between cobordism and characteristic classes is one of the field's most beautiful results: the cobordism ring is a polynomial ring over the mod-2 Steenrod algebra, and the characteristic numbers provide a complete set of invariants for cobordism. This shows that the global structure of manifolds, up to the flexible relation of cobordism, is governed by algebraic invariants that can be computed locally.
Surgery theory, developed in the 1960s by Milnor, Kervaire, Browder, Novikov, Sullivan, and Wall, addresses the classification of manifolds within a fixed homotopy type. The basic operation is surgery: removing a sphere of one dimension and replacing it with a sphere of another dimension, in a way that changes the manifold's topology in a controlled manner. The goal is to determine, for a given space, which manifolds are homotopy equivalent to it, and to classify these manifolds up to diffeomorphism.
The theory proceeds in two steps. First, one asks whether a given homotopy equivalence can be improved to a diffeomorphism by performing surgeries; this is the surgery obstruction problem, and the answer is given by an invariant in the Wall group, an algebraic object built from the fundamental group. Second, one classifies the different ways of performing surgery, which is governed by the structure set of the manifold. This theory is most complete in dimensions at least five, where the fundamental group is finite and the algebraic obstructions are computable.
Surgery theory represents the most ambitious attempt to classify manifolds, and it succeeded spectacularly for simply connected manifolds of dimension at least five. Its limitation is that the algebraic invariants become intractable for manifolds with complicated fundamental groups, and the theory says little about dimensions three and four, where different phenomena occur.
A recurring theme is the relationship between smooth manifolds and topological manifolds—spaces that are locally Euclidean but not necessarily equipped with a smooth structure. Whitney's results showed that in dimensions at least five, every topological manifold admits a smooth structure, and that this structure is unique up to diffeomorphism. However, in 1956 Milnor discovered exotic spheres: manifolds that are homeomorphic to the standard sphere but not diffeomorphic to it. This showed that the smooth structure is not determined by the topology alone.
The situation became even more striking in dimension four. In 1982, Michael Freedman classified simply connected topological four-manifolds, showing that they are determined by their intersection form. Shortly thereafter, Simon Donaldson showed that many of these topological manifolds admit no smooth structure at all, and that those that do admit smooth structures can support infinitely many inequivalent ones. This dramatic divergence between the topological and smooth categories in dimension four remains one of the field's central mysteries.
The tools used to study this distinction are often analytic: Donaldson's invariants come from gauge theory, specifically from the study of instantons—solutions to certain partial differential equations on four-manifolds. Later, the Seiberg–Witten invariants provided a simpler but equally powerful set of tools. These analytic methods show that the smooth structure of a four-manifold is a subtle and rich invariant, far more complex than the underlying topology.
Contemporary differential topology is characterized by a productive interaction between its classical methods and newer analytic and algebraic tools. The field is not a single unified theory but a network of techniques, each adapted to particular dimensions and questions.
In dimensions at least five, surgery theory and the h-cobordism theorem provide a nearly complete classification of simply connected manifolds, and the main open problems concern manifolds with nontrivial fundamental groups. In dimension three, the situation was transformed by the proof of the Poincaré conjecture and the geometrization conjecture by Grigori Perelman in the early 2000s, using the Ricci flow—an analytic method from geometric analysis. This showed that every three-manifold can be decomposed into pieces with geometric structures, providing a classification that had long been sought.
Dimension four remains the most challenging and the most active. The gauge-theoretic invariants of Donaldson and the Seiberg–Witten invariants have been supplemented by more recent developments, including Floer homology theories that assign algebraic invariants to three- and four-manifolds. These invariants are powerful enough to distinguish many smooth structures, but the full classification of smooth four-manifolds remains far out of reach.
The field also continues to develop new tools. Pseudoholomorphic curves, introduced by Gromov in the 1980s, have become a central technique, leading to the development of symplectic topology as a closely related but distinct subject. The interaction between differential topology and algebraic geometry, via the study of complex manifolds and their smooth structures, remains fruitful. And the use of homotopy theory to compute the algebraic invariants that arise in surgery and cobordism continues to be an active area.
Throughout these developments, the core questions remain the ones that defined the field: what are the possible smooth structures on a given topological space, how can they be distinguished, and what geometric and topological features are forced by the existence of a smooth structure? The answers have turned out to be both more subtle and more beautiful than the founders of the subject could have anticipated, and the field continues to be defined as much by its open problems as by its established results.