Knot theory is the branch of topology concerned with the ways a closed curve can be embedded in three-dimensional space. A mathematical knot resembles a loop of string whose ends have been joined: it may be stretched, bent, or moved, but it may not be cut or passed through itself. The central problem is to determine when two apparently different arrangements represent the same knot. Closely related objects called links consist of several disjoint closed curves, which may be entangled with one another even when each individual curve is unknotted.
The subject turns a familiar physical phenomenon into a study of spatial structure. Every knot is intrinsically a circle, so its distinguishing features lie not in the curve alone but in how it sits in the surrounding space. Methods from group theory, geometry, combinatorics, and mathematical physics reveal features that cannot reliably be inferred from a drawing. Knot theory also provides important tools for studying three- and four-dimensional spaces, while its models appear in molecular biology, polymer physics, and quantum theory.
Knots, links, and equivalence
A classical knot is an embedding of the circle into three-dimensional Euclidean space , or equivalently into the three-sphere . The latter is obtained by adding one point at infinity to . An embedding is an injective continuous map that is a homeomorphism onto its image. In most classical knot theory, embeddings are assumed to be smooth or piecewise linear. These restrictions exclude pathological examples, called wild knots, whose local behavior can involve infinitely nested complications.
The appropriate notion of sameness is ambient isotopy. Two knots and are equivalent if there is a continuous family of homeomorphisms , starting with the identity, such that . Requiring the surrounding space to move continuously prevents the curve from crossing itself during the deformation. This is stronger than observing that the two embedded curves are abstractly homeomorphic: every knot, considered without its surroundings, is homeomorphic to a circle.
The unknot is the knot represented by a round planar circle. A complicated-looking loop can still be an unknot if its crossings can all be removed by allowed deformations. The trefoil knot, whose simplest diagram has three crossings, is the simplest nontrivial knot. Its left-handed and right-handed forms are not equivalent under ambient isotopy. Reflection therefore does not automatically preserve knot type. A knot equivalent to its mirror image is called amphichiral; the figure-eight knot is a standard example.
A link is an embedding of a finite disjoint union of circles. The Hopf link consists of two unknotted circles linked once. The Borromean rings have three components: removing any one leaves an unlink, although the three together are inseparable. These examples show that the topology of a link cannot be reconstructed solely from the knot types of its components or from pairwise linking information.
Additional conventions refine the classification problem. An oriented knot carries a direction along its curve, and an oriented link carries one on every component. Equivalence may be required to preserve those directions and, for labeled links, the component labels. Reversing a knot’s direction is distinct from taking its mirror image. Some knots are equivalent to their reverses and some are not. Consequently, a classification must specify whether orientation, component order, and reflection are being retained or disregarded.
Historical development
Representations of knots and interlaced patterns long predate mathematical knot theory, but decorative or practical knotting is not the same as classification under continuous deformation. During the nineteenth century, questions about curves in space became connected with emerging ideas in topology and mathematical physics. In 1833, Carl Friedrich Gauss recorded an integral measuring the linking of two disjoint oriented closed curves. This provided an early example of a numerical quantity detecting a global spatial relationship.
A major stimulus came from William Thomson, later Lord Kelvin. In 1867 he proposed that atoms might be knotted vortex tubes in an all-pervading fluid. Different knot types would represent different chemical elements, with stability supplied by the persistence of vortex structure. The physical theory was abandoned, but it encouraged Peter Guthrie Tait to undertake systematic knot tabulation. Tait’s work, together with that of Thomas Kirkman and Charles Little, established a substantial catalogue of knots organized by crossing number.
The tabulation program also exposed the difficulty of recognizing equivalence. A drawing with many crossings may represent a knot with a much simpler diagram, while visually different drawings can represent the same object. Tait formulated conjectures about alternating diagrams, in which overcrossings and undercrossings alternate along a component. Several of these conjectures remained unresolved for approximately a century. Polynomial methods developed in the 1980s established the minimality of reduced alternating diagrams and related properties; the flyping conjecture was proved by William Menasco and Morwen Thistlethwaite in the early 1990s.
During the early twentieth century, knot theory acquired a more systematic algebraic and topological foundation. Max Dehn investigated knot groups and algorithmic questions. James Waddell Alexander introduced his knot polynomial in 1928. Kurt Reidemeister established a diagrammatic characterization of knot equivalence through local moves, while Herbert Seifert developed constructions of spanning surfaces and algebraic data associated with them. These advances made it possible to pass between spatial embeddings, planar diagrams, surfaces, and algebraic calculations.
A further transformation followed Vaughan Jones’s discovery of a new link polynomial in 1984, published in 1985. It arose from operator-algebra methods rather than from classical knot constructions. Subsequent work connected it with statistical mechanics, representations of braid groups, and quantum field theory. William Thurston’s work on three-dimensional geometry supplied another organizing framework, while later homological invariants connected knots to four-dimensional topology. Knot theory consequently developed from a classification enterprise into a meeting point for several major areas of mathematics.
Diagrams and local moves
A knot diagram is a generic projection of a knot onto a plane, augmented with information indicating which strand passes over the other at every crossing. “Generic” means that the projection has only transverse double points: there are no triple crossings or tangencies. The underpassing strand is conventionally interrupted by a small gap. Without over-and-under information, the projected curve generally does not determine the knot.
The fundamental diagrammatic result is the theorem on Reidemeister moves. Two diagrams represent equivalent tame links precisely when one can be transformed into the other by planar deformation and a finite sequence of three local moves. Type I introduces or removes a small curl and one crossing. Type II introduces or removes a pair of crossings involving two strands. Type III slides a strand past a crossing between two other strands, changing no crossing count. These moves translate a continuous spatial problem into a discrete combinatorial one.
A knot invariant assigns the same value to equivalent knots. To define an invariant from diagrams, it is therefore enough to show that it is unchanged by all three Reidemeister moves. This approach is powerful but demands care. For example, the writhe of an oriented diagram—the sum of the signs of its crossings—is unchanged by moves II and III but changes under move I. Writhe is thus an invariant of a more restricted diagrammatic equivalence, not an invariant of ordinary knots.
The crossing number is the minimum number of crossings among all diagrams of a knot . The unknot has crossing number zero, the trefoil three, and the figure-eight knot four. These small values conceal a difficult minimization problem: counting crossings in a particular drawing gives only an upper bound. A diagram may require an initial increase in crossings before a simplifying sequence becomes apparent. Greedily removing visible curls or crossing pairs is therefore not a complete recognition method.
Diagrams can be encoded in finite symbolic form. A Gauss code records the order in which crossings are encountered while traversing a component, together with information sufficient to reconstruct crossing behavior. Dowker–Thistlethwaite notation uses a systematic numbering of crossing encounters and is especially useful in knot tables. Such encodings support computer enumeration, comparison, and invariant calculation. Not every abstract crossing sequence is realizable by a classical planar diagram, however, so encoding and realizability are separate questions.
Other presentations serve different purposes. Grid diagrams represent knots by horizontal and vertical segments with a fixed crossing convention and support combinatorial homology calculations. Tangle diagrams isolate portions of a knot inside a ball with endpoints on its boundary. They allow complicated links to be assembled from simpler pieces and make local replacement rules mathematically precise.
Spanning surfaces and structural decomposition
A Seifert surface for an oriented knot is a compact, connected, orientable surface embedded in whose boundary is that knot. Every oriented knot admits such a surface. Seifert’s algorithm constructs one directly from a diagram: smooth every crossing in a way compatible with the orientation, cap the resulting disjoint circles with disks, and attach twisted bands corresponding to the original crossings. The resulting surface may not have the smallest possible genus, but it provides concrete geometric and algebraic information.
The genus of a knot is the minimum genus among all its Seifert surfaces. It measures how complicated an orientable spanning surface must be. The unknot has genus zero because it bounds an embedded disk; conversely, a knot bounding such a disk is an unknot. Both the trefoil and the figure-eight knot have genus one. If a connected surface produced by Seifert’s algorithm has disks and bands, then its Euler characteristic is , and its genus is
Knots can be combined by connected sum. Small unknotted arcs are removed from two knots, and the resulting endpoints are joined compatibly to produce . A nontrivial knot is prime if it cannot be expressed as a connected sum of two nontrivial knots. Horst Schubert established the prime decomposition theorem: every nontrivial tame knot decomposes into prime knots, uniquely up to order and the relevant equivalences. Knot genus is additive under connected sum,
This decomposition is different from the decomposition of ordinary integers, but it supplies a comparable organizing principle for classification.
Another important family consists of torus knots. A -torus knot winds around an unknotted torus in two independent directions, with coprime integers and . If either absolute value is one, the result is an unknot; otherwise it is nontrivial. The trefoil is a -torus knot, up to choices of sign and orientation. The genus formula is
Satellite knots arise by placing a knot pattern inside a solid torus and then embedding that solid torus as a neighborhood of another knot, called the companion. Cables and Whitehead doubles are important examples. The surrounding torus can remain topologically essential in the knot complement, recording structure invisible in a simple connected-sum decomposition. Prime knots can therefore still have substantial internal organization, and decomposing their complements along suitable tori is a central technique in three-dimensional topology.
Classical algebraic invariants
One of the earliest and most powerful algebraic constructions associates a group to a knot. The knot group is the fundamental group of its complement,
Its elements represent loops in the surrounding space, with multiplication given by concatenation, modulo continuous deformation that avoids the knot. A diagram gives a Wirtinger presentation: generators correspond to diagram arcs separated by undercrossings, and each crossing supplies a conjugation relation. Thus a spatial embedding produces a finite algebraic description.
The group of the unknot is infinite cyclic, isomorphic to . A trefoil group has a presentation
and is nonabelian, proving that the trefoil is not an unknot. However, an abstract group is not always sufficient to distinguish knots. The boundary of a tubular neighborhood also supplies a meridian, which circles the knot once, and a preferred longitude, which follows it with zero linking number. This peripheral information records how the boundary torus sits in the complement and substantially strengthens the group invariant.
For a two-component oriented link, the linking number measures signed mutual winding. In a regular diagram it is half the sum of the signs of crossings involving different components:
The same integer can be expressed through Gauss’s double integral or as an algebraic intersection number with a spanning surface. Reversing one component’s orientation changes its sign. Linking number zero does not imply that the components are unlinked: the Whitehead link is a standard counterexample.
The Alexander polynomial, introduced in 1928, extracts commutative algebra from the topology of an infinite cyclic cover of the knot complement. It can also be calculated from a Seifert matrix , whose entries record linking numbers between surface curves and displaced copies. One common formula is
where means equality up to multiplication by a unit . A symmetric normalization removes much of this ambiguity. For the trefoil, one may take .
The polynomial constrains geometry but does not completely classify knots. Its breadth—the largest exponent minus the smallest—satisfies
For a fibered knot, whose complement fibers over the circle with a Seifert surface as fiber, the Alexander polynomial is monic; the converse is false in general. Nontrivial knots can even have Alexander polynomial . The determinant is another useful invariant, related to branched covers and diagram colorings. These limitations illustrate a general principle: equality of a computable invariant is usually weaker than equivalence of knots.
Braids, polynomial invariants, and quantum constructions
A braid consists of strands running monotonically between two parallel levels, with prescribed endpoints. Braids on strands form the braid group , generated by elementary crossings . Their defining relations include
and
Unlike permutations, braid generators have infinite order: one crossing cannot generally be canceled by another crossing of the same sign.
Closing a braid by joining corresponding top and bottom endpoints produces a link. Alexander’s theorem states that every link can be represented as a closed braid. Markov’s theorem characterizes when two braids have equivalent closures, using conjugation and stabilization or destabilization. These results permit link invariants to be constructed from algebraic representations of braid groups, provided the resulting quantities respect the Markov moves.
The Jones polynomial is a Laurent-polynomial invariant normalized by . In one standard convention it satisfies the skein relation
where the three oriented diagrams agree outside a small disk and differ inside it by a positive crossing, a negative crossing, or an oriented smoothing. Repeated use of the relation reduces calculations to simpler links. Replacing a knot by its mirror substitutes for , so the Jones polynomial can often detect chirality.
Louis Kauffman provided a diagrammatic state-sum construction. At each crossing, two smoothings are weighted by and . The bracket is normalized by assigning value to a single circle, while each additional disjoint circle contributes . The bracket is invariant under Reidemeister moves II and III. Multiplying it by , where is the diagram’s writhe, corrects its behavior under move I. Substitution gives the Jones polynomial in a standard convention.
Further polynomial invariants, including the HOMFLY-PT and Kauffman polynomials, organize additional skein relations and representation-theoretic information. Their construction connects knot theory with representation theory, lattice models in statistical mechanics, and quantum groups. In 1989 Edward Witten interpreted Jones-type invariants through Chern–Simons theory, a three-dimensional quantum field theory in which observables are associated with embedded loops. Rigorous algebraic constructions by Nicolai Reshetikhin and Vladimir Turaev subsequently produced link and three-manifold invariants from suitable representation-theoretic data. These connections explain why crossing relations can reflect algebraic identities arising far beyond the original problem of knotted curves.
Knot complements and three-dimensional geometry
Instead of studying a knot directly, one can study the space remaining around it. Removing the interior of a tubular neighborhood gives the knot exterior
a compact three-dimensional manifold with torus boundary. The open complement and the exterior encode closely related information. In 1989 Cameron Gordon and John Luecke proved that knots in are determined by their complements, with the customary allowance for reflection when orientation is not specified. Thus the ambient space left behind can recover the knot type.
Thurston’s geometrization results for knot complements reveal a striking division among prime knots. A nontrivial prime knot is a torus knot, a satellite knot, or a hyperbolic knot. A hyperbolic knot has a complement carrying a complete finite-volume metric of constant sectional curvature . The alternatives are distinguished by embedded surfaces and the geometric structures they permit. Composite knots have an additional decomposition before this prime classification is applied.
The figure-eight knot is the simplest hyperbolic knot. Its complement can be assembled from two regular ideal hyperbolic tetrahedra and has volume approximately when curvature is normalized to . An ideal tetrahedron has vertices at infinity rather than at ordinary points of hyperbolic space. The missing knot corresponds geometrically to a cusp, whose cross-sections are Euclidean tori. This construction turns a familiar four-crossing diagram into an explicit three-dimensional geometric object.
For a complete finite-volume hyperbolic manifold of dimension three, Mostow–Prasad rigidity implies that the geometry is determined by the topology. Consequently, hyperbolic volume is a knot invariant rather than an arbitrary feature of a chosen metric. Other information, such as cusp shape, also becomes available. Volume is powerful but not complete: distinct knots can have the same volume, including certain examples related by mutation.
Dehn surgery constructs new three-manifolds by removing a knot’s tubular neighborhood and gluing back a solid torus in a different way. The gluing is specified by the boundary curve that becomes a meridian of the inserted solid torus, commonly encoded by a rational slope relative to a meridian–longitude basis. Thurston’s hyperbolic Dehn surgery theorem states that all but finitely many fillings of a one-cusped finite-volume hyperbolic three-manifold are hyperbolic. Exceptional fillings can instead yield nonhyperbolic spaces. Surgery therefore makes knots a practical language for constructing and comparing three-manifolds, not merely objects to be classified within one fixed space.
Computation, recognition, and complexity
A finite knot diagram creates an algorithmic problem: determine whether it represents the unknot, whether two diagrams are equivalent, or whether a particular invariant has a specified value. The Reidemeister theorem guarantees the existence of a finite sequence connecting equivalent diagrams, but it does not supply a short sequence. A naive search can be extremely inefficient because simplifying a diagram may require temporarily increasing its size.
Wolfgang Haken developed a major decision method in the early 1960s using normal surface theory. A triangulated three-manifold is divided into tetrahedra, and a normal surface meets each tetrahedron in prescribed types of triangular and quadrilateral disks. Counts of these pieces satisfy integer matching equations and additional compatibility conditions. In the unknot problem, the relevant topology is detected through properly embedded disks in the knot exterior. This converts a geometric existence question into a finite, though potentially very large, search.
The complexity of recognition is subtler than decidability. Joel Hass, Jeffrey Lagarias, and Nicholas Pippenger showed that unknot recognition lies in the complexity class NP: an unknotted diagram admits a certificate whose validity can be checked in polynomial time. Later results placed the problem in co-NP under additional hypotheses, and Marc Lackenby established an unconditional co-NP result. Thus both unknottedness and knottedness have polynomially checkable certificates. These statements do not provide a known polynomial-time algorithm for finding the certificates or deciding every input.
In practice, software combines many methods. Fast invariants can exclude equivalence immediately; hyperbolic structures can identify geometric features; simplification routines can reduce diagrams; and rigorous topological algorithms can settle difficult cases. A numerical approximation to a geometric quantity is not automatically a proof. Certified calculations must control rounding error and establish that the computed structure corresponds to the intended topological object.
Enumeration presents another set of challenges. Generating all diagrams up to a crossing bound produces enormous redundancy, since many diagrams encode the same knot. Algorithms therefore exploit diagram symmetries, alternating structure, invariants, and equivalence moves. Modern prime-knot tables extend through many more crossings than nineteenth-century hand tabulations, but their size grows rapidly. A table’s conventions matter: mirror pairs may be combined or separated, and alternating and nonalternating knots are often listed independently.
Exact invariant computation can itself be hard. State-sum definitions may involve exponentially many states, and evaluating the Jones polynomial at generic fixed values is computationally intractable in precise complexity-theoretic senses. Special evaluation points can behave differently. Computational knot theory therefore studies both practical recognition methods and the inherent resources required by topological information.
Homological invariants and four-dimensional questions
Polynomial invariants can be strengthened by replacing their coefficients with algebraic structures. This process, called categorification, seeks a graded homology theory whose Euler characteristic recovers the polynomial. Khovanov homology, introduced by Mikhail Khovanov in 2000, assigns a bigraded collection of groups to a link. Its graded Euler characteristic is the Jones polynomial, with choices of normalization. The construction begins with the cube of resolutions of a diagram and equips associated algebraic data with a differential.
Homology can distinguish knots that share the same polynomial because it retains information lost when dimensions are combined into an alternating sum. Peter Kronheimer and Tomasz Mrowka proved that Khovanov homology detects the unknot. The analogous question of whether the Jones polynomial detects the unknot remains open. This contrast demonstrates why an invariant’s categorification can be substantially stronger than the numerical or polynomial data it recovers.
Knot Floer homology was introduced independently by Peter Ozsváth and Zoltán Szabó, and by Jacob Rasmussen, in work announced around 2002. Its graded Euler characteristic recovers the Alexander polynomial. It also detects knot genus and fiberedness, properties that the Alexander polynomial alone does not determine. Its original construction uses methods related to symplectic geometry and pseudo-holomorphic curves, although grid diagrams provide combinatorial models for important versions.
Four-dimensional questions arise by viewing as the boundary of the four-ball . A slice knot bounds a properly embedded disk in . One must distinguish smooth sliceness from topological sliceness, the latter conventionally requiring the disk to be locally flat. The distinction is essential because smooth and topological embeddings behave differently in dimension four. More generally, the smooth slice genus is the smallest genus of a smoothly embedded, connected, orientable surface in bounded by .
Michael Freedman’s work implies that knots with Alexander polynomial are topologically slice, yet some are not smoothly slice. Homological invariants expose such differences. Rasmussen’s integer invariant , extracted from a deformation of Khovanov homology, satisfies
The knot Floer invariant similarly satisfies . These inequalities can obstruct smooth sliceness even when classical algebraic tests do not.
Knot concordance relates two knots when they cobound a suitably embedded annulus in . Under connected sum, concordance classes form an abelian group, with slice knots representing the identity. This group connects knot theory to four-dimensional topology and contains intricate algebraic structure. Questions such as whether every smoothly slice knot is ribbon remain major unresolved problems.
Physical and biological settings
Physical curves have thickness, stiffness, friction, and energetic costs, whereas classical knot type retains only embedding information. A mathematical knot can be stretched arbitrarily without changing its type, but a real rope cannot generally pass through a narrow opening or sustain unlimited curvature. Geometric knot theory adds quantitative constraints. One important quantity is ropelength, the ratio of a curve’s length to its thickness, where thickness can be defined as the largest radius of an embedded tubular neighborhood. Minimizing ropelength seeks an economical realization of a fixed knot type.
Such models help separate topological obstruction from geometric packing. Two ropes can have the same knot type but very different tightness or energy. Energy functionals may penalize close approaches, bending, or both. Minimizers and critical configurations are studied with methods from the calculus of variations, differential geometry, and numerical optimization. Their interpretation depends on the model: an ideal frictionless tube is not equivalent to a woven rope, a polymer chain, or a biological filament.
Closed DNA molecules provide a direct biological setting for knots and links. Covalently closed circular DNA can become knotted, and replicated circular molecules can remain linked. Enzymes called topoisomerases alter DNA topology by transiently breaking strands and allowing controlled passage before resealing them. Type II topoisomerases pass one double-stranded DNA segment through another, enabling changes in knotting and catenation. Topological analysis can therefore reveal features of molecular reactions that are difficult to infer from a static image.
For a closed DNA ribbon, linking, twist, and writhe are related by
Here the linking number concerns the ribbon’s two boundary curves, twist measures local rotation of the ribbon, and writhe measures the spatial coiling of its centerline. This relation is not an identification of these quantities: twist and writhe can change continuously while their sum remains fixed under deformations that preserve the closed ribbon. Experiments on DNA supercoiling exploit precisely this interplay between topology and elastic geometry.
Knotted configurations also occur in some proteins, but a protein backbone is an open chain rather than a mathematical knot. Assigning a classical knot type therefore requires a closure procedure, and different closures can sometimes give different answers. Researchers also study slipknots, in which a subchain forms a knotted configuration even though the whole chain is classified as unknotted under a chosen closure. This distinction matters when relating topology to folding pathways.
In polymer physics, random closed-chain models investigate how knotting probability depends on chain length, excluded volume, and confinement. For broad classes of long self-avoiding polygon models, nontrivial knotting becomes overwhelmingly likely as length increases. The probability of a particular knot type, however, depends on the ensemble and physical assumptions. Topology supplies the classification, while mechanics and statistical modeling determine which configurations are accessible and how frequently they occur.