Knot Theory: Topology, Physics, & Computation

• Knot theory is a mathematical field examining embeddings of circles in 3-space through continuous deformations and diagrammatic representations.
• Modern research extends knots into physics and computation by employing algebraic invariants, topological solitons, and optimized rendering techniques.
• Applied studies reveal how elasticity, self-contact, and quantum defects introduce practical complexity to classical knot models in varied scientific contexts.

A knot is classically a smooth or piecewise-linear locally flat embedding of a circle $S^1$ in $\mathbb{R}^3$ or, equivalently after one-point compactification, in $S^3$, considered up to ambient isotopy. In current research the term also denotes a wider family of topologically constrained objects: diagrammatic knot types, knotoids obtained from embedded intervals, knotted defects in ordered media, Hopf-charged textures in quantum fluids, eigenenergy-string braids in non-Hermitian band theory, and materially tied elastic rods. Across these settings, the common content of “knot” is invariance under continuous deformation together with a nontrivial global organization of crossings, preimages, or linked loops (Carter, 2012, Hall et al., 2015, Hu et al., 2020).

1. Classical topological object and diagrammatic representation

In classical knot theory, a knot is an embedding

$S^1 \hookrightarrow \mathbb{R}^3$

considered up to ambient isotopy. It is often convenient to work in $S^3$, the one-point compactification of $\mathbb{R}^3$. Two knots are equivalent when there is an orientation-preserving homeomorphism of pairs carrying one embedded circle to the other. This ambient viewpoint turns the subject into a study of embeddings together with the topology of their complements (Carter, 2012).

Knots are ordinarily studied through planar diagrams: generic projections with transverse double points equipped with over/under information. Diagrammatic equivalence is generated by the three Reidemeister moves together with planar isotopy. This makes knot theory simultaneously geometric and combinatorial: a three-dimensional embedding is encoded by a planar crossing pattern, and isotopy becomes a finite calculus of local moves. Standard examples include the trefoil $3_1$, the simplest nontrivial knot, and the figure-eight knot $4_1$, which already exhibits behavior distinct from the trefoil in coloring and group-theoretic invariants (Carter, 2012).

A central quantitative invariant is the crossing number $c(K)$, the minimum number of crossings over all diagrams of a knot. Modern diagrammatic knot theory also studies bridge number, unknotting number, petal number, and other complexity measures. These invariants do not replace one another: they capture different normal forms and different obstructions to simplification. The introduction of straight number and contained straight number makes this explicit by measuring how efficiently a knot can be drawn when every crossing is forced onto a single straight strand (Owad, 2018).

2. Algebraic, group-theoretic, and quandle structures

The complement of a knot carries a fundamental algebraic invariant, the knot group. For a knot $K \subset S^3$ with tubular neighborhood $\mathbb{R}^3$0, the knot exterior is

$\mathbb{R}^3$1

and the knot group is

$\mathbb{R}^3$2

From a diagram one obtains a Wirtinger presentation whose relations have the form $\mathbb{R}^3$3, reflecting conjugation at crossings. This already shows that crossing combinatorics are encoded in the complement group (Carter, 2012).

Quandles refine this perspective. A quandle is a set with a binary operation satisfying idempotence, right-invertibility, and self-distributivity, precisely the axioms needed for coloring rules to survive Reidemeister moves. Dihedral quandles recover classical 3-colorability, and conjugation quandles arise from groups via $\mathbb{R}^3$4. In Carter’s formulation, the fundamental quandle is “somewhat stronger” than the fundamental group, because it retains the directed crossing action rather than only the induced conjugation structure (Carter, 2012).

Knot contact homology pushes the algebraic theory further. For a knot $\mathbb{R}^3$5, the degree-zero knot contact homology $\mathbb{R}^3$6 is identified with the cord algebra, and augmentations of $\mathbb{R}^3$7 encode geometric information related to representation varieties. Cornwell studies KCH representations $\mathbb{R}^3$8 and shows that they induce augmentations of knot contact homology, clarifying the relation between the augmentation polynomial and the $\mathbb{R}^3$9-polynomial. For 2-bridge knots, the augmentation polynomial agrees with the classical $S^3$0-polynomial after the $S^3$1 substitution, while torus knots and a family of pretzel knots exhibit genuinely higher-dimensional representation-theoretic structure (Cornwell, 2013).

These constructions extend to higher-dimensional knotting. For surface-links in $S^3$2, including non-orientable ones, knot groups admit braid-theoretic characterizations via plat presentations, and there is a parallel characterization for knot symmetric quandles. In particular, every dihedral quandle with an arbitrarily good involution can be realized as the knot symmetric quandle of a surface-link, showing that quandle-theoretic realizability is not restricted to orientable or classical settings (Yasuda, 2024).

3. Alternative diagrammatic regimes: straight knots and knotoids

A straight diagram is a knot diagram in which all crossings lie on one straight horizontal strand, while the remainder of the knot is a collection of semicircles based on that strand. If a straight diagram with $S^3$3 crossings has knot word

$S^3$4

then $S^3$5 is its straight word. The straight number $S^3$6 is the minimum crossing number among all straight diagrams of $S^3$7, while the contained straight number $S^3$8 imposes the stronger condition that no arc crosses the extended straight strand. Both are well defined for every knot, and they satisfy

$S^3$9

A knot is perfectly straight when $S^1 \hookrightarrow \mathbb{R}^3$0 (Owad, 2018).

This straight-position formalism generalizes the meander and OGC constructions of Jablan and Radović from special classes to all knots. Owad proves that every knot admits a straight diagram and a contained straight diagram, gives an explicit upper bound $S^1 \hookrightarrow \mathbb{R}^3$1, and derives linear relations with petal number: $S^1 \hookrightarrow \mathbb{R}^3$2 He also shows that every torus knot $S^1 \hookrightarrow \mathbb{R}^3$3, every $S^1 \hookrightarrow \mathbb{R}^3$4-pretzel knot, every 2-bridge knot with continued-fraction expansion of length $S^1 \hookrightarrow \mathbb{R}^3$5, and every knot with 7 or fewer crossings is perfectly straight (Owad, 2018).

Knotoids replace the embedded circle by an embedded interval with two endpoints. A knotoid diagram is an immersed interval in $S^1 \hookrightarrow \mathbb{R}^3$6 or $S^1 \hookrightarrow \mathbb{R}^3$7 with transverse double points and over/under data, subject to Reidemeister moves away from endpoints and excluding the forbidden move in which an endpoint passes over or under a strand. This produces two related but distinct theories: spherical knotoids and planar knotoids. Hyperbolic knotoid theory associates spherical knotoids to knots in a thickened torus and planar knotoids to knots in handlebodies with totally geodesic boundary. In the spherical setting, the product of hyperbolic knotoids is again hyperbolic and the corresponding volumes add. Rational knotoids form a particularly tractable class: all rational knotoids are hyperbolic, and the unique rational spherical knotoid of least volume is $S^1 \hookrightarrow \mathbb{R}^3$8, with volume approximately $S^1 \hookrightarrow \mathbb{R}^3$9 (Adams et al., 2022).

4. Knots in ordered media, quantum matter, and band topology

In nematic liquid crystals, a knot need not be a material string at all. The order parameter is a director field

$S^3$0

so the ground-state manifold is $S^3$1. Line defects are disclinations, and these disclination lines may themselves form knots or links. Machon and Alexander characterize such knotted nematics via the Pontryagin–Thom construction and Milnor fibrations. In this framework the defect set is the boundary of a Seifert-like surface, and “color winding” on that surface records director rotation, hedgehog charge, and linking information. Explicit constructions using polynomials such as $S^3$2 generate torus knots and links, while the Hopf link arises from $S^3$3 (Machon et al., 2013).

In a spin-1 Bose–Einstein condensate, the knot becomes a nonsingular three-dimensional topological soliton in the nematic vector field. The relevant map is

$S^3$4

classified by

$S^3$5

with topological charge given by the Hopf invariant $S^3$6. Hall and collaborators experimentally realized the standard Hopf map with $S^3$7 in a spinor condensate of $S^3$8, using spatially dependent Larmor precession induced by an inhomogeneous magnetic field. The observed density images reveal the circular core and linked preimage rings of the Hopf fibration. This was reported as the first experimental observation of a knot soliton in quantum matter (Hall et al., 2015).

A different but mathematically related use of “knot” appears in non-Hermitian band theory. For a one-dimensional non-Hermitian Bloch Hamiltonian with separable complex bands $S^3$9, the trajectories of the eigenenergies as functions of momentum form closed strings in $\mathbb{R}^3$0, topologically a solid torus. Hu and Zhao show that the resulting closed braids give a complete topological classification of 1D non-Hermitian Hamiltonians with separable bands and no symmetry. The theory includes Hopf links, trefoil knots, figure-eight knots, and Whitehead links realized as eigenenergy-string closures. In this setting the global biorthogonal Berry phase

$\mathbb{R}^3$1

is a $\mathbb{R}^3$2 invariant equal to the permutation parity of the non-Hermitian bands and to the sum of the Wilson-loop eigenphases (Hu et al., 2020).

Context	Knotted object	Topological descriptor
Nematic liquid crystal	Disclination line in a director field $\mathbb{R}^3$3	PT surface, linking data, Seifert-surface structure
Spin-1 Bose–Einstein condensate	Knot soliton in $\mathbb{R}^3$4	Hopf invariant $\mathbb{R}^3$5
1D non-Hermitian Bloch bands	Closed braids of eigenenergy strings in a solid torus	Knot/link type and a $\mathbb{R}^3$6 Berry-phase invariant

Taken together, these cases show that in modern physics a knot may be realized as a defect set, a preimage-linking structure, or a spectral braid rather than as a tangible piece of cord. This suggests that knotting is best understood as a topological organization principle that can be instantiated in fields, order parameters, and spectra as well as in curves (Machon et al., 2013, Hall et al., 2015, Hu et al., 2020).

5. Physical knots as elastic and dynamical objects

For physical knots, finite thickness, elasticity, friction, and self-contact become unavoidable. In the study of twist deformation of closed trefoils, a physical knot is modeled as an elastomeric rod tied into a trefoil, closed by joining its ends, and then twisted by a relative angle

$\mathbb{R}^3$7

X-ray microtomography and discrete-rod simulations show that the untwisted trefoil has a three-fold symmetric curvature profile, while applied twist either tightens or loosens the knot until symmetry-breaking buckling occurs at critical angles. The analysis is organized with a Cosserat frame and the kinematic quantities $\mathbb{R}^3$8, with scalar curvature $\mathbb{R}^3$9. The observed transition is interpreted as a Michell-type instability in which imposed twist energy is converted into bending energy; the global relation between topology and geometry is expressed through the Călugăreanu–White–Fuller identity

$3_1$0

This places knot buckling within the mechanics of slender rods rather than purely within abstract topology (Goto et al., 4 Mar 2025).

A distinct dynamical viewpoint treats knots as charged polygonal curves evolving under self-repulsion. In Kauffman’s knot-dynamics experiments, the discrete Simon energy is

$3_1$1

with repulsive forces between nonadjacent vertices and spring forces along adjacent edges. Rational tangles are used to construct families of hard unknots and complexified knots. These are unknotted or topologically simple configurations that nevertheless fail, under self-repulsive evolution, to relax to simple round or low-energy forms. The paper argues that sufficiently complex hard unknots and sufficiently complex complexified knots may fail to reach global minimal-energy states in self-repulsion environments, emphasizing that topological triviality does not imply dynamical simplicity (Kauffman, 2021).

These physical studies sharpen the distinction between ideal and realized knotting. In ideal knot theory, isotopy removes geometry from the problem; in elastic and dynamical settings, the same knot type can support multiple metastable geometries, bifurcations, and distinct energy pathways. A plausible implication is that any realistic account of applied knotting must combine topology with constitutive mechanics and contact geometry (Goto et al., 4 Mar 2025, Kauffman, 2021).

6. Computational representation, rendering, and recognition

Computational knot theory increasingly treats knot diagrams and embeddings as optimization objects. The R package knotR takes a closed Bezier path drawn in Inkscape, imports it into R, and optimizes an explicit badness functional using nlm(). The ingredients include crossing-angle quality, bending energy, and penalties for close approaches of noncrossing strands. The result is a systematic way to produce vectorized, publication-quality knot diagrams with symmetry constraints, exemplified on knots such as $3_1$2, $3_1$3, $3_1$4, the Perko pair, and the full family of prime knots with 8 or fewer crossings (Hankin, 2016).

Differentiable geometry has pushed this further into inverse design. In knot-based inverse perceptual art, the object is a tubular knot in $3_1$5 whose specified projections resemble target images. The knot embedding is represented as

$3_1$6

where $3_1$7 is a homeomorphism parameterized by an invertible neural network, so the optimization preserves knot type. A differentiable renderer for tubular knots, together with loss terms enforcing absence of tube self-intersection, occupancy inside a prescribed region, bounded bending, and a material budget, enables gradient-based search over embeddings. The framework is demonstrated for single-view and multi-view designs and culminates in a real-world 3D-printed object (Gangopadhyay et al., 2023).

Machine learning has also turned knot recognition into a fine-grained visual classification problem. The Knots-10 benchmark contains 1,440 images, with a deployment-oriented split that trains on loosely tied knots and tests on tightly dressed ones. Under this protocol, Swin-T and TransFG both average $3_1$8 accuracy, while PMG scores $3_1$9, consistent with the hypothesis that jigsaw shuffling disrupts crossing continuity. McNemar tests cannot separate four of the five general-purpose backbones, and a Mantel permutation test shows that topological distance significantly correlates with confusion patterns in three of the five models. The proposed TACA regularization improves embedding-topology alignment from $4_1$0 to $4_1$1 without improving classification accuracy, while a pilot cross-domain test with 100 phone photographs produces a $4_1$2–$4_1$3 percentage-point drop, exposing rope appearance bias as the dominant failure mode (Nie et al., 24 Mar 2026).

Across these computational settings, the knot is not merely an invariant to be computed after the fact. It becomes a representational prior, an optimization constraint, and, in the recognition setting, the latent source of discriminative structure. This suggests that the next stage of computational knot research will likely couple topological guarantees to physically grounded geometry and to more robust visual representations (Hankin, 2016, Gangopadhyay et al., 2023, Nie et al., 24 Mar 2026).

Classical knot theory, field-theoretic knotting, elastic-knot mechanics, and computational knot design all preserve a common core: a knot is a configuration whose essential content survives deformation but whose realization depends on the ambient category. In $4_1$4 it is an embedded circle; in a nematic or condensate it is a linked preimage structure; in non-Hermitian band theory it is a closed braid of eigenenergy strings; in rod mechanics it is a self-contacting elastic body; and in modern computation it is an object to be rendered, optimized, and classified. The persistent utility of the concept lies precisely in that portability between topology, geometry, physics, and algorithmics.