Active -atic Fluids on Riemannian 3-Manifolds
Introduction
Much of the matter around us is built from units that point. A liquid crystal is a fluid of rod-shaped molecules that line up along a common direction without freezing into a solid; a film of swimming bacteria, a layer of crawling cells, or a suspension of microtubules driven by molecular motors all share the same feature, that each unit carries an orientation and the units agree, on average, about which way to face. Matter of this kind has orientational order. It flows like a liquid, yet it has a built-in arrow, or a built-in axis, the way a crystal has a built-in lattice. The study of such materials is soft-matter physics, and the materials that burn their own fuel to move, the bacteria and the motor-driven filaments, are called active matter.
The order is rarely perfect. Almost every sample of an orientationally ordered material is threaded by places where the orientation cannot be defined, points or lines around which the arrows swirl and refuse to settle. These are the topological defects, and they are not blemishes to be polished away. They are forced by topology, they carry conserved charges, and in active matter they move, collide, and reorganise the flow around them. A large part of the physics of these materials is the physics of their defects, so any description of orientational order has to describe its defects in the same breath.
This article is about how to write that description down, first in flat space and then on a curved space. The flat-space question is already rich: what is the right mathematical object to represent a headless direction, or a unit with a threefold symmetry, and how do its defects appear in that object. The curved-space question is the one that motivates the whole programme: a film of active matter does not always live on a flat plane. It can coat a sphere, a torus, or a more intricate surface, and one can ask the same question in three dimensions, where the medium fills a curved region of space instead of a flat box. On a curved background the orientation has to be compared between nearby points using the geometry of the space itself, and the defects feel that geometry. The curvature of the space enters the equations of motion.
The setting where these questions stop being academic is living tissue. A tumour, a developing organ, or a sheet of crawling cells is active matter with orientational order, and its topological defects organise both the flow and the growth. In glioma the cells self-organise into collectively motile streams whose density tracks the malignancy of the tumour,[15][16] and three-dimensional imaging of such tumours finds genuine disclination lines and loops, their planar sections the familiar half-integer defects, interconverting between and as the line twists, with the cell alignment holding quasi-long-range order across the whole tumour.[14] This is active -atic matter on a three-dimensional, curved, and deforming medium, and it is the physical reason for posing the theory covariantly.
To answer these questions the article develops two languages side by side and keeps them in step throughout. The first is traditional tensor calculus, the index notation , , in which most of physics is written, where a geometric object is named by its components in a basis and the rules of the game are bookkeeping rules for those indices. The second is geometric calculus, the Clifford-algebra language in which the product of two vectors is a single object combining their dot product and their oriented area, and in which rotations, spinors, and the algebra of defects all become elementary. The two languages describe the same physics. Holding them side by side is what makes the construction legible: the index notation keeps every formula computable, and the geometric notation keeps every formula meaningful.
The plan is to build from the ground up. The first sections fix the algebra: vectors and tensors and the index machinery, then the geometric product and the multivectors it generates. With that in hand the article says precisely what a -atic is, the orientationally ordered phase of units with a -fold symmetry, and constructs its order parameter first for planar order, then for spatial order, where the jump from the complex numbers to the quaternions is forced by the geometry. From there the article reaches spinors, the square-root objects on which rotations act most naturally, and then lifts everything to a curved manifold: the covariant calculus that compares orientations across a curved space, the spin bundle that houses the order parameter there, and the active dynamics, the Beris-Edwards equations, that set it in motion. The curvature returns through a Weitzenböck identity, and a non-equilibrium variational principle picks out which braided or knotted defect structures the active flow selects. The final sections record the two-dimensional shadow of that selection and the outlook.
Two threads run the length of the article. The first is a single algebraic construction, specialised four times in sequence. We build the Clifford algebra of an arbitrary quadratic-form signature once and in full generality, with directions of positive square, of negative square, and degenerate directions of zero square, and then read off in turn the four cases the physics needs: the flat plane for planar -atics, flat space for spatial -atics, the fibrewise over the tangent bundle of a curved 3-manifold, and the degenerate signatures with that surface when a deforming 3-manifold collapses a direction to zero length. Each case is a restriction of the same algebra, and none is rebuilt; Section 2.2 sets up the general signature and tabulates the four specialisations.
The second thread is the advecting -volume. Well before any dynamics, Section 2.6 sets up how an oriented -dimensional volume element rides along a flow, in both languages at once: as a totally antisymmetric -tensor transported by the minors of the flow's derivative, and as a grade- blade transported by a single outer product. The exponential rate at which these volumes stretch under the active flow is the quantity the late sections identify with the topological complexity of the disclination tangle the flow selects.
The reader is assumed to have met vectors, partial derivatives, and a little tensor notation, but not soft-matter physics. Every symbol is defined where it first appears, and every step is shown. We begin with the most basic object of all, a vector and its components.
1. Vectors, tensors, and index notation
Everything in this article is built from vectors and the objects assembled out of them. Before any orientational order, any defect, or any curvature can be discussed, we need a precise and computable language for vectors, for the machines that eat vectors and return numbers, and for the way a metric lets us measure them. This section fixes that language. It is the index notation of tensor calculus, and it is the first of the two languages the article carries throughout.
1.1. Vectors and their components
We work in a real vector space of dimension , which for almost everything below is the tangent space at a point of flat space, so with or . A basis of is a set of vectors with the property that every vector can be written as a sum of scalar multiples of them in exactly one way.
Given a basis of , every has a unique expansion
The real numbers are the components of in the basis . The index is written as a superscript, an upper index, which marks a component of a vector.
The vector is one object, while its components depend on the basis chosen. Pick a different basis and the same arrow acquires a different list of numbers. Much of tensor calculus is the discipline of writing only those statements about components that hold in every basis.
1.2. The dual space and covectors
Alongside the vectors live the linear machines that measure them. A linear functional on is a map that respects addition and scaling, for all scalars and vectors . These functionals themselves form a vector space.
The dual space of is the set of all linear functionals , with addition and scalar multiplication defined pointwise. Its elements are called covectors (or one-forms). Given a basis of , the dual basis of is defined by
where is the Kronecker delta. A covector expands as , and its components carry a lower index, the mark of a covector.
The reason for the superscript-versus-subscript discipline is now visible. Vector components carry upper indices and covector components carry lower indices, and the dual-basis relation of [eq:dual-basis] makes a covector eat a vector by pairing the two,
A pairing of a covector with a vector is a sum of products in which an upper index meets a lower index of the same name.
1.3. The Einstein summation convention
The sum in [eq:pairing] illustrates a pattern that recurs in every formula below: a repeated index, once up and once down, is summed over its range . Writing the summation sign each time is noise.
Whenever an index appears exactly twice in a single term, once as an upper index and once as a lower index, it is summed over its full range , and the summation sign is omitted. Such a repeated index is a dummy index; its name carries no meaning and may be relabelled. An index that appears once is a free index and must match on both sides of an equation.
Under this convention the pairing of [eq:pairing] is simply , and the expansion of [eq:vector-expansion] is . Every formula in the rest of the article uses this convention.
1.4. The metric and raising and lowering indices
A bare vector space measures no lengths and no angles. The object that supplies them is the metric.
A metric on is a symmetric, positive-definite bilinear form . In a basis its components are
so that the inner product of two vectors is . The length of is . The matrix is symmetric and positive definite, so it is invertible; its inverse has components written with upper indices, , and is fixed by
In the orthonormal basis of flat space the metric components are simply , the identity matrix, and likewise . We keep the general symbol throughout, because on a curved manifold the metric components are functions of position and are no longer the identity.
The metric does one more thing that is used on nearly every page: it converts vectors into covectors and back. Given a vector with components , the map is a covector, whose components are obtained by contracting with the metric.
The metric and its inverse convert between upper and lower indices:
The same rule applies index by index to any tensor. The two operations are mutual inverses, since by [eq:inverse-metric].
Take with the metric
which is symmetric and positive definite, so it is a legitimate metric. Let have components . We lower the index to get the covector components , computing each free value of in turn.
For : .
For : .
So . To check that raising undoes the lowering we need the inverse metric. For a matrix the inverse divides the adjugate by the determinant, here , so
Raising again gives . For : . For : . We recover , the components we started with, as [eq:raise-lower] promises.
1.5. Tensors and their components
A vector eats nothing and a covector eats one vector. A tensor is the general machine that eats several vectors and covectors at once, linearly in each slot.
A tensor of type on is a map
that is linear in each of its arguments separately. Feeding it the basis covectors and basis vectors gives its components
carrying upper indices and lower indices. A vector is a tensor of type , a covector is of type , and the metric is of type .
Tensors are built from vectors and covectors by one operation, the tensor product, which the components of [eq:tensor-components] quietly use.
The tensor product of two vectors is the type- tensor whose components are the products of the components,
It is linear in each factor and associative, and the -fold product is a type- tensor with components . Order matters: and differ unless the factors are equal. Every type- tensor is a sum of products of vectors and covectors, so the tensor product generates all tensors from the vectors of Section 1.1.
Take a unit direction with components . Its self-product is the rank-one symmetric tensor . In two dimensions with ,
Removing the trace by subtracting leaves the symmetric-traceless tensor , the nematic order parameter of Section 3 up to its magnitude. The same construction at higher takes the -fold self-product and removes all traces, giving the rank- order parameter of a -atic; this is the route Zhu, Saintillan, and Chern follow to build -atic order from a single director.[2]
The metric raises and lowers any index of any tensor by the rule of [eq:raise-lower] applied slot by slot, so a tensor of type can be presented with its indices in any up-or-down arrangement that totals . The objects of soft-matter physics are almost all type- tensors, written , and the structure of such a tensor is what the next two subsections take apart.
1.6. Symmetric, antisymmetric, and trace decompositions
A type- tensor in dimension is a square array of numbers. It splits canonically into pieces that transform among themselves under a change of basis, and the splitting is the algebraic backbone of the order parameters that appear later.
A tensor is symmetric if and antisymmetric if . Every is the sum of a symmetric and an antisymmetric part,
where round brackets denote symmetrisation and square brackets antisymmetrisation of the enclosed indices.
The symmetric part carries more structure still. Its trace is the scalar , obtained by raising one index and contracting it against the other. Removing a suitable multiple of the metric leaves a piece whose trace vanishes.
Let be symmetric in dimension , with trace . Its symmetric-traceless part is
A general therefore decomposes into three pieces that do not mix under a change of basis: the antisymmetric part , the trace part , and the symmetric-traceless part .
The symmetric-traceless part is the piece that matters for orientational order. A headless axis in space, a nematic director with no head-to-tail distinction, cannot be a vector, since a vector points one way; it is encoded instead by a symmetric-traceless tensor, the -tensor that reappears in Section 3. The decomposition of [eq:sym-traceless] is the algebraic reason such an object exists, so we work an instance in full.
Work in with the orthonormal metric , so that raising and lowering change nothing and the trace is the ordinary matrix trace. Take
We peel off the three pieces of [eq:sym-antisym] and [eq:sym-traceless] in order.
First the antisymmetric part . Subtracting the transpose and halving gives the entries , , and , with the diagonal zero and the lower triangle the negatives of the upper, so
Next the symmetric part , found by averaging with the transpose,
Its trace is , so in dimension the trace part is , a diagonal matrix with on each diagonal entry. Subtracting it from the symmetric part leaves the symmetric-traceless part of [eq:sym-traceless],
whose trace vanishes, as required. Adding the three pieces, the antisymmetric matrix of [eq:T-antisym], the trace part , and the symmetric-traceless matrix of [eq:T-st], returns the original of [eq:T-example]. The symmetric-traceless part is the headless-axis information, the trace is the scalar magnitude, and the antisymmetric part is an infinitesimal rotation, three pieces that will be kept apart for the rest of the article.
This decomposition is the first appearance of the object that will carry orientational order. With the index language and the symmetric-traceless part in hand, the next section introduces the second language, the geometric product, which packages the same vectors and tensors in a form where rotations and defects become elementary.
2. The geometric product and multivectors
Section 1 gave us two ways to combine vectors. The metric of [eq:metric-components] multiplies two vectors and returns a single number, their inner product , which throws away all directional information and keeps only a scalar. The tensor construction of [eq:tensor-map] goes the other way, gluing vectors into ever larger arrays of components without ever closing the operation: the product of two vectors is a type- object living in a different space from the vectors themselves. Neither operation lets us multiply vectors and stay among vector-like objects, the way the real numbers let us multiply numbers and get a number back.
This section introduces a single associative product that does exactly that. It multiplies any two elements of one algebra and returns an element of the same algebra, it contains the inner product as its symmetric part, and it generates a graded family of new objects, the bivectors, trivectors, and so on, that turn out to be the natural home for rotations and for the order parameters of later sections. It is the second of the two languages the article carries throughout, the geometric calculus, and it is built on the geometric product.
2.1. The geometric product
The defining demand is modest. We keep the vector space with its metric from Section 1, and we ask for a single associative product on the algebra those vectors generate, the geometric product, written by juxtaposition. It is the primary object, and everything else in this section is read off from it. We impose one axiom: the square of a vector returns its squared length.
Let carry the metric of Section 1 with associated quadratic form . The geometric product is an associative, bilinear, unital product on the algebra generated by , written by juxtaposition , subject to the single axiom that every vector squares to its quadratic form,[11][5]
for every , a scalar. The algebra so defined is the Clifford algebra . Rigorously it is the quotient of the tensor algebra by the two-sided ideal generated by the elements , and it is the universal associative unital algebra into which maps with : any linear map sending with factors uniquely through it.
The geometric product is not the inner product. The inner product of Section 1 multiplies two vectors and returns a scalar, discarding the rest; the geometric product keeps everything and stays inside one algebra. Both the inner product and the outer product are derived from it, as the symmetric and antisymmetric parts of the product of two vectors. We extract them now.
Polarise the axiom of [eq:square-axiom]. Apply it to a sum and expand, using associativity and bilinearity but never commutativity, since we have not assumed the product commutes:
The left-hand side is by the axiom and the bilinearity of the metric, while the first and last terms on the right are and . Cancelling those leaves the polarisation identity
The symmetric combination of the two orderings recovers twice the metric. The product therefore carries the metric inside it as a derived quantity, but it also carries something the metric discards: the symmetric part of is a scalar by [eq:symmetric-anticommutator], yet nothing forces the antisymmetric part to vanish. We name the two pieces.
For two vectors the inner product and the outer (wedge) product are defined as the symmetric and antisymmetric parts of their geometric product,
The inner part is the scalar of [eq:symmetric-anticommutator]; the outer part changes sign under exchange, , and is a new object called a bivector. Adding the two halves recovers the geometric product as a derived grade decomposition,
What makes the geometric product powerful, beyond either half on its own, is that it is associative and it makes every nonzero vector invertible. Neither the inner product nor the wedge product has an inverse; a vector has no reciprocal under contraction or under the wedge. Under the geometric product it does. Since by [eq:square-axiom], set
Vectors can now be divided, and this is what lets later sections write rotations as products and quotients of vectors instead of matrices acting on them.
The grade split of [eq:geom-product] is the whole content of the geometric product for two vectors. When and are parallel the outer part vanishes and the product is the scalar ; when they are orthogonal the inner part vanishes and the product is the pure bivector , which then anticommutes, . The bivector is the oriented plane element spanned by the two vectors, the geometric-product version of the oriented area, and its sign records the orientation, the sense of circulation from to .
This is where the two languages first touch. The antisymmetric part of a 2-tensor, defined in [eq:sym-antisym], was an array of components with no obvious geometric reading. The wedge product gives it one. Writing in components against the basis bivectors produces exactly the antisymmetric array , so a bivector is an antisymmetric type- tensor, read as an oriented plane instead of a square array. The dictionary that the rest of the article leans on begins here: antisymmetric tensors are bivectors, and the symmetric scalar trace is the inner product.
2.2. The Clifford algebra and its grades
Multiplying two vectors produced a scalar and a bivector. Multiplying three produces, in general, a vector and a trivector, and so on up the dimension. The set of all such products, with all sums and scalar multiples, is closed under the geometric product and forms an algebra.
The Clifford algebra of the vector space with metric and quadratic form is the associative algebra generated by the vectors of under the geometric product of [geometric-product]. Its elements are multivectors. Each multivector is a sum of homogeneous pieces of definite grade: grade the scalars, grade the vectors, grade the bivectors, and in general grade the -vectors, spanned by wedge products of distinct basis vectors. In dimension the top grade is , its single basis element is the pseudoscalar, and the algebra has dimension
A choice of distinct basis vectors gives a basis blade with , and the basis blades span the algebra.
Most multivectors are not single blades but graded sums of them, and the algebra provides operations that read off and reorder those grades. We set these up for an arbitrary multivector before specialising to any dimension.
The grade- projection returns the grade- part of a multivector , and every multivector is the sum of its graded parts,
with the scalar part, the vector part, and so on to the top grade . The reverse reverses the order of the vector factors in every blade; on a grade- part it contributes the sign
so it fixes scalars and vectors and flips bivectors and trivectors. The reverse furnishes a magnitude , the scalar part of , generalising the squared length of a vector. On the even subalgebra of the plane, met in 2.4, the reverse is ordinary complex conjugation.
The inner and outer products of 2.1 were defined for two vectors. They extend to any two blades by grade. The geometric product of a grade- blade and a grade- blade spreads over the grades ; its lowest grade is the inner product and its highest grade is the outer product,
For two vectors these are the grade- scalar and the grade- bivector of [eq:geom-product], recovering the earlier definitions. The wedge raises grade and the inner product lowers it, and the full geometric product carries both at once. This grade-mixing, which the tensor product does not have, is what closes the algebra and makes it invertible.
A blade of grade carries a geometric meaning. The wedge of vectors is the oriented -volume of the parallelepiped they span: a bivector is an oriented area, a trivector an oriented volume, and a grade- blade an oriented -dimensional hypervolume, its magnitude the size of the parallelepiped and its sign the orientation. It vanishes precisely when the vectors are linearly dependent, since a flattened parallelepiped has no -volume. In dimensions there are independent oriented -volumes, the basis -vectors, and the top grade is one-dimensional: the pseudoscalar is the oriented unit -volume of the whole space. Under a linear map every vector is transported, so the pseudoscalar is scaled by the determinant of the map, and the algebra carries the transformation of volumes in its top grade. This is the fact Section 11.4 turns into the central object of the article: the active flow advects these oriented volumes, and the exponential rate at which their blades stretch is the topological complexity of the disclination tangle the flow selects.
So far the metric has been positive-definite, every basis vector squaring to . The construction never used that. The square axiom of [eq:square-axiom] asks only that a vector square to its quadratic form, and a quadratic form can be diagonalised so that each basis vector squares to one of three values.
Diagonalise the quadratic form on an -dimensional space . By Sylvester's law of inertia its basis splits into three blocks, and on such a basis the square axiom [eq:square-axiom] reads
with . The algebra so generated is the Clifford algebra of signature , written : directions of positive square, of negative square, and degenerate directions of zero square. The null directions span the radical, the subspace on which the bilinear form [eq:symmetric-anticommutator] vanishes identically, for every . When the form is non-degenerate and the notation contracts to ; the positive-definite metric of Section 1 has , written .
Only the squares of the generators changed. The grade structure of [clifford-algebra], the dimension count [eq:clifford-dimension], the grade projection and reverse of [grade-projection], and the inner-outer split of [eq:general-inner-outer] were all assembled from associativity and antisymmetry, never from the sign of a square, so they carry over word for word. The algebra has dimension and the same independent oriented -volumes at each grade. The signature changes the multiplication table while the grades stay fixed.
The degenerate directions are exactly the ones the vector inverse cannot reach, and that failure is the algebraic shadow of a collapsing direction in space.
A vector in the radical has , so the inverse formula [eq:vector-inverse] divides by zero and no reciprocal exists. The invertibility that lifted the geometric product above the bare wedge in Section 2.1 fails precisely along the radical. This is the algebra of a collapsing direction: when a deforming medium flattens a direction of space to zero length, the metric there degenerates, the tangent-space quadratic form acquires a null direction, and the local algebra slips from into a signature with . The zero squares of [eq:signature-squares] record which directions have collapsed, the geometric content of the deforming theory of Section 12.
The four settings the article visits are all restrictions of this one construction.
| Setting | Signature | Built in |
|---|---|---|
| planar -atics, flat | , even part | Section 4 |
| spatial -atics, flat | , even part | Section 5 |
| curved 3-manifold | fibrewise over | Section 8 |
| deforming 3-manifold | , radical the collapsed directions | Section 12 |
The first two are flat and fixed, computed in full below. The third attaches a copy of to every point of a curved space and lets the attachment turn from point to point, the spin bundle of Section 8. The fourth lets the quadratic form itself move, and degenerate where the medium collapses a direction, the deforming theory of Section 12.
The smallest case that shows every feature is the plane, which we now work out in full.
Take with the orthonormal basis , so and the square axiom of [eq:square-axiom] reads . By [eq:clifford-dimension] the algebra has dimension , with basis blades
one scalar, two vectors, and one bivector. We compute the products of the generators step by step. The two diagonal products are scalars by the square axiom. For the mixed product, the basis vectors are orthogonal, so , and [eq:dot-wedge-def] gives , a pure bivector. Reversing the order and using [eq:symmetric-anticommutator] with ,
The two generators anticommute. The full multiplication table follows from associativity, the square axiom, and this one anticommutation relation. For instance , while . A general multivector is the graded sum with four real coefficients, one scalar , two vector components , and one bivector coefficient .
2.3. Multivectors versus tensors
Both languages now carry graded objects. A multivector and a tensor are different objects, and the article uses both at once: the order parameter lives in one language, the geometry of space in the other.
A tensor of Section 1 is a multilinear map, recorded by a component array of a fixed type. Nothing constrains its symmetry: it may be symmetric, antisymmetric, or of mixed symmetry, and the metric , the -tensor, and a generic -array are all tensors. A multivector is an element of the Clifford algebra, a graded sum of oriented magnitudes, a scalar together with a vector, an oriented area, an oriented volume, and so on. The two meet on the antisymmetric content and part on everything else.
A grade- multivector is the same object as a totally antisymmetric rank- tensor: the bivector is the antisymmetric array of [eq:sym-antisym], read as an oriented plane in place of a square array, and the same correspondence holds at every grade. The alternating tensors are exactly the multivectors. They differ in two ways. First, a general tensor also has symmetric and mixed-symmetry parts that no multivector represents. There is no symmetric multivector, so the symmetric-traceless -tensor and the metric are tensors with no single-grade multivector form. Second, multivectors carry the geometric product, which contracts with the metric and mixes grades through [eq:general-inner-outer], whereas the tensor product never contracts and only raises rank. A multivector is therefore an antisymmetric tensor equipped with an invertible, metric-aware product.
Orientation, area, volume, and the determinant live naturally as multivectors, in any dimension; the headless-axis order parameter does not, since it is symmetric. Section 3 resolves the apparent tension: the symmetric -tensor has no grade- multivector form, but it does have an even-multivector form, built from the spin representation instead of from a single grade, and that even-multivector face is what carries through to the curved theory. The reading of multivectors as oriented hypervolumes is also the tool the later sections use for deformation: a diffeomorphism between curved spaces acts on volumes through the determinant of its derivative, the factor by which it scales the pseudoscalar, so the bookkeeping of how a deforming space stretches and shears is bookkeeping of the top-grade multivector.
2.4. The even subalgebra and the complex numbers
The even-grade part of the plane's algebra is a familiar structure. Collect the blades of even grade, the scalar and the bivector, and set the vectors aside. Their products stay within the even part, since multiplying two even-grade blades gives an even-grade result, so the even blades form a subalgebra. To identify it we first record the square of the single bivector. By we mean the field of complex numbers, the numbers with and .
The even subalgebra is the span of the even-grade blades,
where is the pseudoscalar of the plane, the unique top-grade blade of [eq:cl2-basis-list]. It is closed under the geometric product and contains the scalars.
The plane pseudoscalar satisfies .
Proof
Square by writing it out and moving the inner factors past each other with the anticommutation relation of [eq:e1e2-anticommute], then collapse the diagonal squares with the square axiom :
The single sign comes from the one swap of into , and the two unit squares finish it.
The pseudoscalar of [eq:even-subalgebra-2] squares to by [pseudoscalar-square], so it behaves exactly like the imaginary unit. The map sending to the complex number respects addition trivially and respects multiplication because both products are governed by the same rule . This is the isomorphism
The even part of the plane's geometric algebra is the complex numbers, with the pseudoscalar in the role of the unit imaginary. We check that the algebra really multiplies like , and then see what right-multiplication by does to a vector.
Take two even elements and . Multiply them out, using associativity and from [pseudoscalar-square]:
The scalar part and the bivector part are exactly the real and imaginary parts of the complex product , confirming the isomorphism of [eq:cl2-c]. The even subalgebra is commutative for the same reason is, a feature special to the plane that fails in three dimensions, where the even part becomes the non-commutative quaternions of Section 5.
Right-multiplication by rotates a vector by a right angle. Take the vector and multiply on the right by , using , :
The result is the vector turned through a quarter turn, the same map that multiplication by performs on the complex plane . The pseudoscalar generates rotations, which is why later sections build the order parameter from exponentials of .
2.5. Two languages for one object
The pieces of the geometric language line up with the index notation of Section 1 as follows, and the article switches between the two whenever one reads more cleanly than the other.
| Geometric calculus | Index notation of Section 1 |
|---|---|
| inner product | contraction , a scalar |
| bivector | antisymmetric part , a 2-tensor |
| pseudoscalar | oriented unit area (or volume in 3D) |
| even subalgebra | the complex numbers |
A bivector and an antisymmetric tensor are the same object seen two ways, and the symmetric-traceless tensor of [eq:sym-traceless], the one flagged in Section 1 as the carrier of orientational order, will turn out to have its own multivector face as well. When the order parameter arrives in Section 3 we shall write it both ways, as a symmetric-traceless tensor in the index language and as an even multivector in the geometric language, and the equivalence of the two descriptions is what makes the curved-space construction of the later sections legible.
2.6. The outermorphism and advecting -volumes
The second thread named in the introduction is one object tracked through both languages: an oriented -volume carried along a flow. The blade of [eq:general-inner-outer] already supplies the oriented -volume; what remains is to say how a map of vectors moves it. A linear map of the underlying space sends each edge of a parallelepiped to a new edge, and the volume they span follows. The map this induces on volumes has a name.
Let be linear. A grade-preserving linear map of the whole Clifford algebra is an outermorphism of if it restricts to the identity on scalars and to on vectors and distributes over the outer product,
for all vectors . We write it , the exterior-power functor applied to the map ; its grade- part is the -th exterior power of , the induced map on oriented -volumes.
The definition demands a map with these properties; that exactly one exists, and that it composes the way the maps do, is the content of the next proposition.
Every linear map has a unique outermorphism . It is functorial,
so is invertible exactly when is, with . On the one-dimensional top grade it acts by a single scalar, taken as the determinant of ,
the factor by which scales the oriented -volume of the whole space.
Proof
Take the orthonormal basis of Section 1. The basis blades with span the algebra by [clifford-algebra], so any grade-preserving linear map obeying [eq:outermorphism] is determined on a basis and is therefore unique. For existence, set on each basis blade by [eq:outermorphism] and extend linearly. This is well defined because the assignment is linear in each argument and vanishes whenever two arguments coincide, since the wedge does; it is alternating and multilinear, so it factors through the -th exterior power that houses the grade- blades, and the induced map is .
Functoriality is read off the generators. Both sides of [eq:outermorphism-functor] are linear, so it suffices to check them on a blade:
and is immediate from [eq:outermorphism]. Taking gives , so inverts when does. Finally the top grade is one-dimensional, spanned by the pseudoscalar , so is a scalar multiple of ; that scalar is the definition of .
The proposition fixes what the outermorphism is. The next theorem says how much it stretches a -volume, and the answer is read entirely off the singular values of .
Let have singular values . The stretch factors of , the singular values of the outermorphism on grade , are the products
In particular the largest factor by which any -volume stretches is the product of the largest singular values,
attained on the blade spanned by the top singular directions.
Proof
Write the singular-value decomposition with orthogonal and in the orthonormal basis. Functoriality [eq:outermorphism-functor] gives . An orthogonal map preserves the inner product on vectors, hence the inner product it induces on blades, in the magnitude of [grade-projection], so and are isometries of every grade and leave the stretch factors of equal to those of . On the orthonormal basis blade the diagonal map acts by [eq:outermorphism] as
so is diagonal in the blade basis with the eigenvalues [eq:outermorphism-svals]. Their largest is , on , which carries back to the top singular directions of .
Two consequences are immediate; a third, the volume-growth exponent of a flow, follows in Section 2.7 once the Lyapunov exponents are in hand.
For linear maps one has , and is the product of all the singular values.
Proof
Apply functoriality [eq:outermorphism-functor] to the pseudoscalar and use [eq:outermorphism-det] twice,
The magnitude statement is the top-grade case of [eq:outermorphism-svals], where the single product runs over all indices.
A map preserves oriented -volume, , if and only if its outermorphism fixes the pseudoscalar, , immediate from [eq:outermorphism-det]. The top grade is then frozen, and every stretch [eq:outermorphism-svals] respects , so growth in one intermediate grade is paid for by contraction in another. This is the algebra of the incompressible flow of Section 11.4: the conserved oriented volume is the top-grade pseudoscalar, and all stretching lives in the grades .
The two languages read the intermediate grades differently and agree on the answer.
Tensor language carries an oriented -volume as a totally antisymmetric -tensor of [eq:sym-antisym], and acts on it as the array of minors of the matrix of , the Cauchy-Binet formula, one entry for each of the ways to pick rows and columns. Geometric language carries the same object as the grade- blade and transports it by the one outer product [eq:outermorphism]. The bookkeeping minors of the first language are the components of the single blade of the second, and both return the same number, the transported -volume .
A flow applies this transport at every instant. When the linear map is the derivative of a flow map , that is, when at a point of the space, the outermorphism carries a material -volume planted at that point along the flow,
a single outer product of the transported edges in the geometric language, and the Jacobian minors of in the index language. Two features run from here to the end of the article. The top grade is scaled by the one number of [eq:outermorphism-det], so a volume-preserving flow with fixes the oriented -volume and confines all stretching to the intermediate grades by [pseudoscalar-pinned], the advected curves and surfaces. And the exponential rate at which those intermediate blades stretch is the sum of the positive Lyapunov exponents of Section 2.7, the quantity Section 11.4 identifies, through Gromov and Yomdin, with the topological entropy of the flow and the braiding rate of the disclination lines. The same outermorphism advects a volume on the space of shapes when the manifold itself deforms, the construction of Section 12.4.
2.7. Lyapunov exponents
The advection of Section 2.6 needs one notion from dynamics to become a rate. Run a flow for a time and linearise it: the differential is a linear map, and its singular values of [outermorphism-singular-values] record how a small material line planted at a point has stretched after time . As grows these factors grow or shrink exponentially, at rates that name the flow.
The Lyapunov exponents of a flow on are the long-time exponential rates of the singular values of its differential,
the rates at which infinitesimal material lines stretch where and contract where . They are the dynamical successors of the static singular values of [outermorphism-singular-values]: a positive exponent pulls nearby trajectories apart, the signature of chaotic mixing.
That the defining limit exists is not automatic. For the leading exponent it follows from the way the flow composes with itself.
On a closed manifold the leading Lyapunov exponent of [eq:lyapunov-def],
exists and equals .
Proof
The flow composes, , so the chain rule gives the cocycle . Taking operator norms, using submultiplicativity, and then the supremum over ,
so is subadditive, , and finite for each because is compact and the flow smooth. Fekete's subadditive lemma then gives , which is the limit [eq:leading-lyapunov].
The remaining exponents exist almost everywhere by Oseledets' multiplicative ergodic theorem. On a periodic orbit, where the flow returns to itself after one period and its linearisation is a single return map , the exponents per period are over the eigenvalues of , and the largest, the dilatation logarithm, is the pseudo-Anosov stretch of the two-dimensional shadow of Section 11.7.
For the incompressible flow of [pseudoscalar-pinned] the determinant [eq:outermorphism-det] holds the product of the singular values at one, so the exponents sum to zero, : stretching in one direction is balanced by contraction in another, and the volume-growth exponent below collects only the stretching half.
With the Lyapunov exponents [eq:lyapunov-def] of a flow , the exponential growth rate of an advected -volume is the sum of the largest exponents,
so the volume-growth exponent of Section 11.4 equals the sum of the positive Lyapunov exponents, .
Proof
By [eq:outermorphism-topstretch] the operator norm of is the product of the largest singular values of , so
The partial sum grows with while and falls once , so its supremum over includes every positive exponent and no negative one. This is the Pesin-Ruelle quantity named in Section 11.4.
3. What is a -atic?
With two languages in hand, the index notation of Section 1 and the geometric-product language of Section 2, the article can now state precisely what a -atic is and write its order parameter in both. The answer to the title question is a single integer and two equivalent descriptions of what it implies.
A -atogen is a constituent of the medium, a molecule, a filament, or a particle, whose shape is invariant under rotation by about its symmetry axis; it carries the cyclic symmetry group of order . The integer is that order of rotational symmetry. A -atic is the orientationally ordered phase of -atogens, whose macroscopic order parameter is invariant under a rotation of the local frame by .
The ladder of standard cases runs as follows. For the units are polar: they carry a single distinguished direction, with no symmetry under any non-trivial rotation, so the order parameter is a genuine vector. For the units are nematogens: apolar rods with head-tail symmetry, invariant under a half-turn, and the familiar liquid-crystal case. For the units are tetratics, with fourfold rotational symmetry about the axis. For the units are hexatogens, carrying the hexagonal symmetry of bond-orientational order in a triangular packing.
3.1. The order parameter in two languages
We work in the plane to fix ideas; the jump to spatial order is the subject of later sections. In the plane the state of a -atogen at a point is a direction modulo the identification , where is the angle the unit's axis makes with a chosen reference direction. The order parameter must be a quantity that is invariant under that identification, so it cannot be the angle itself; it must be a function of that returns the same value when is replaced by .
Tensor language. Section 1 introduced the symmetric-traceless part of a -tensor as the carrier of headless-axis information. For the nematic , the standard construction builds this object from the unit direction vector with components ,
where is the scalar order parameter measuring how strongly the units are aligned, is the metric of Section 1.4, and the trace is removed by the subtraction of so that . In flat with the orthonormal metric the factor follows from the dimension in the general formula [eq:sym-traceless]. For a general -atic in the plane the order parameter is the symmetric-traceless tensor of rank assembled from copies of , with all traces removed; the rank-2 case of [eq:Q-nematic-2d] is the instance of that family.
Geometric-calculus language. Section 2 showed that the even subalgebra , with the pseudoscalar playing the role of the imaginary unit. The order parameter of a -atic in the plane is the spin- field
where the exponential is taken in the even subalgebra via the usual series , is the same scalar amplitude as in [eq:Q-nematic-2d], and is the director angle. The phase advances by for each physical rotation of the frame: a rotation sends , so the order parameter is a weight- object in the even subalgebra.
3.2. Dictionary between the two descriptions
The two descriptions encode the same physical content. The following table sets out the correspondence for the planar case.
| Director angle | Tensor language | Geometric-calculus language |
|---|---|---|
| , | , | |
| frame rotation | transforms as a rank- tensor under | , phase advances by |
| symmetry identification | ||
| defect: phase winds by around a loop | one net rotation of the director by | phase of winds by , charge |
3.3. The through-line: one integer, three roles
The single integer threads through the entire formalism in three roles that are really one.
First, is the spin of the order parameter. The phase of in [eq:spin-k-order] advances by for each unit physical rotation of the frame, so is a spin- field, the same sense in which a vector field is spin-1 and a scalar field is spin-0. The symmetric-traceless tensor of rank in the index language is the same object expressed in components.
Second, is the weight in the algebra. The order parameter lives in the weight- sector of the even subalgebra : it is an element whose argument under the natural grading by angle is . The even subalgebra itself is graded by the integer weight, and weight is where the nematic order parameter sits.
Third, fixes the elementary disclination charge to . The order parameter need only return to its original value modulo a rotation of the frame by , so a loop that encircles a defect need only carry a total phase winding of to close consistently. The minimum non-trivial winding is therefore , corresponding to a topological charge of . For the nematic this is the familiar disclination; for the hexatic it is a defect.
The rotational symmetry of the constituent propagates unchanged: the -atogen's symmetry becomes spin- for the order parameter, weight- in the algebra, and charge for the defects.
3.4. Worked example: the nematic
Take the nematic in the plane with an orthonormal basis and metric . The director is the unit vector at angle ,
Tensor language. Form the symmetric-traceless order parameter of [eq:Q-nematic-2d] component by component. The outer product with gives
Subtracting the trace part yields
using the double-angle identities and . The trace vanishes, confirming the symmetric-traceless construction of Section 1.6.
Now replace by . The double-angle entries become and , so
The tensor order parameter is invariant under a half-turn of the director. Heads and tails are identified, as a nematogen requires.
Geometric-calculus language. The spin-2 order parameter of [eq:spin-k-order] with is
Replace by :
since in the even subalgebra. The spin-2 field is invariant under , the same half-turn symmetry as [eq:Q-invariance].
Reading off the defect charge. Around a disclination the director angle increases by on a complete circuit of the core. The tensor entries return to their original values after that circuit, since [eq:Q-invariance] shows a shift of is the period of the order parameter. In the geometric language the phase of winds by , a complete loop in . The charge is as expected.
The two descriptions agree at every step: the same matrix entries, the same invariance, the same defect charge. The symmetric-traceless tensor language keeps the physics computable; the even-subalgebra language keeps it meaningful. The rest of the article carries both.
4. Planar -atics in flat space
Section 3 built the order parameter in two languages. We now put it to work in the plane and ask two physical questions: what energy governs the field, and what defects it permits. Each is answered twice, once with indices and once with the geometric product, and the two answers coincide.
4.1. The Landau-de Gennes free energy
A free energy assigns a cost to every configuration of the order parameter. It has two parts: a bulk term that fixes how strongly the units align, and an elastic term that penalises spatial variation of the alignment.
For the planar order parameter of [eq:Q-nematic-2d], with elastic constant and Landau coefficients and , the Landau-de Gennes free energy is[9]
The first term is elastic, the last two are the bulk potential. Repeated indices are summed, and in the flat plane , so index placement is immaterial here.
The single invariant carries the whole bulk potential, so we compute it once.
The matrix form of [eq:Q-nematic-2d-matrix] has entries , , and . Summing the squares of the four entries,
independent of the angle , as a rotational invariant must be. The bulk energy density is then , minimised over , for , at the spontaneously ordered magnitude .
The same energy reads cleanly in the geometric language. The spin- field is from Section 3, with reverse and squared modulus , so that . The free energy becomes the Ginzburg-Landau functional
Differentiating the phase, , so the elastic density unpacks as : a cost for varying the magnitude plus times the cost of bending the axis. The index energy and the geometric energy are one functional in two alphabets.
4.2. Defects and topological charge
A defect is a point where the orientation cannot be defined, the amplitude falling to zero at its core while the axis winds around it. The winding is quantised.
The topological charge of a planar disclination enclosed by a loop is the net rotation of the director, in turns,
The field must be single-valued around , so its phase changes by an integer multiple of , giving . The elementary disclination therefore carries charge .
Put the core at the origin and let the director angle follow the polar angle as . Around any loop encircling the origin increases by , so by [eq:defect-charge]
In the geometric language , the pseudoscalar phase turning through exactly one revolution: returns to itself after one circuit and can vanish only at the core, where . For the nematic this is the disclination, the director rotating by around the loop, the lowest-energy defect a planar nematic admits.
Substituting into the elastic term of [eq:gl-planar] gives a gradient density falling as , whose integral from a core radius to the system size diverges logarithmically,
with the Frank elastic constant set by and . Two disclinations of opposite charge attract with a force falling as , the two-dimensional Coulomb law, and can annihilate; like charges repel. This equilibrium picture is the backdrop against which the activity of Section 9 drives the defects into perpetual motion.
5. Spatial -atics and the jump to quaternions
Three dimensions change two things at once. The order-parameter tensor grows, and the even subalgebra that was the commuting complex numbers of [eq:cl2-c] becomes the non-commuting quaternions. We take the tensor side first, then the algebra.
5.1. The spatial order parameter and its defects
In three dimensions the nematic order parameter is the symmetric-traceless tensor built from the director with components ,
the trace removed with the factor fixed by the dimension of Section 1.6. It has five independent components, the symmetric-traceless part of a array. It is the symmetric-traceless part of the self-product of [eq:tensor-product], and the rank- spatial -atic order parameter is the symmetric-traceless part of the -fold product .
The Landau-de Gennes free energy gains a cubic term that the plane did not have, because in three dimensions the invariant need not vanish,[9]
The cubic invariant is odd under , and its presence makes the isotropic-to-nematic transition first order, a feature of the spatial theory with no planar analogue.
The defects change dimension. In the plane a disclination was a point; in space the locus where loses its orientation is generically a curve, one of codimension two. The order-parameter space of the uniaxial nematic is the real projective plane , the sphere of directions with antipodes identified, since and are the same headless axis. Its low homotopy groups give two families of defect:
| defect | codimension | homotopy class |
|---|---|---|
| disclination line | two | |
| point defect (hedgehog) | three |
The single class of is the half-integer line, its own inverse, and the integer hedgehog charges live in . A disclination line, being a curve in space, can knot and link in ways a planar point cannot, the three-dimensional successor to the braid taken up in Section 11.
5.2. The even subalgebra is the quaternions
The geometric algebra of space is , of dimension : one scalar, the three vectors , three bivectors, and the pseudoscalar . The even subalgebra of [grade-projection] collects grades zero and two, the scalar and the three bivectors, four dimensions in all. By we mean Hamilton's quaternions, the numbers with and , a non-commutative division ring.
Write the three unit bivectors as , , . They satisfy Hamilton's relations , so the even subalgebra is isomorphic to the quaternions.[11]
Proof
Each bivector squares to by the same swap-and-collapse as in the plane, for instance , and likewise for and . For the product, move the inner factors past each other with the anticommutation of distinct basis vectors,
so , and the same computation gives and . Then . These are Hamilton's relations, and the linear map sending to the quaternion units is an algebra isomorphism.
A unit-magnitude even element is a rotor with a unit bivector, acting on a vector by conjugation , a rotation through in the plane of , with the reverse of [eq:reverse]. The unit quaternions form the group , the double cover of the rotation group : and carry out the same rotation, the origin of the half-angle .
5.3. Non-commuting disclinations
In the plane the even subalgebra was , commutative, and two defects encircled in either order produced the same total winding. In space the even subalgebra is , and its units do not commute,
The uniaxial class is still abelian, but biaxial order, whose order-parameter space is , has for its disclination group the non-abelian quaternion group , the eight unit quaternions . Two such lines passed through one another can leave an entanglement that no continuous motion removes. The planar braid, an abelian record of crossings, has nowhere to live, and its three-dimensional successor is the knotting and linking of disclination lines.
These lines are not only a theoretical possibility. Three-dimensional imaging of glioma tumours finds nematic order threaded by disclination lines and loops, whose two-dimensional sections are the defects of Section 4.2 and which interconvert between and along the line as it twists through the tissue, the cell alignment holding quasi-long-range order across the tumour.[14] The knotting and linking set out here is the structure those lines can carry, and the non-abelian charges of are the entanglements only biaxial order can hold.
6. Spinors
The rotor of Section 5.2 acted on a vector by a two-sided product . A spinor is the object a rotor acts on by a one-sided product, and that one-sidedness is the content of the spin representation: a spinor sees a rotation at half its angle, and is in this sense a square root of a rotation.
6.1. The rotor and the half-angle
A rotation is carried by a rotor , the exponential of a bivector. For a unit bivector with the exponential series splits into even and odd powers exactly as it does for the pseudoscalar of the plane, giving an Euler formula,
with the reverse of [eq:reverse] serving as the inverse. Conjugation rotates a vector by the full angle , the two half-angles in and adding. The half-angle has a consequence the full angle hides.
Set in [eq:rotor-exp]:
A full rotation returns the rotor to , and only a second full turn, , brings it back to . The vector it acts on is unaffected, since , so ordinary geometry never sees the sign. An object that transforms by a single factor of does.
6.2. Spinors and the double cover
A spinor is an even multivector that transforms under a rotation by the rotor through the one-sided product
instead of the two-sided conjugation of a vector.
By [eq:rotor-2pi] a spinor changes sign under a rotation, , and is restored only after . This is the signature of the spin representation. The map that sends a rotor to the rotation it performs,
is two-to-one, since and give the same rotation. The rotors of unit magnitude therefore form a group that covers the rotation group twice, the spin group . Vectors and tensors carry the representation of ; spinors carry the representation of its double cover , the finer object from which a vector is recovered as a product of spinorial pieces.
6.3. The ladder of spin groups and the order parameter
The spin group is read off the even subalgebra dimension by dimension.
| dimension | even subalgebra | spin group |
|---|---|---|
| plane | (Section 2.4) | |
| space | (Section 5.2) |
In the plane the spinors are the complex numbers and the spin group is the circle ; in space they are the unit quaternions and the spin group is . The -atic order parameter of Section 3 is a spinor of weight : a rotation of the frame by sends
so it picks up copies of the phase. The half-integer defect charges of Section 4.2 are the spinorial half-windings this weight permits. The spatial order parameter inherits the same structure with in place of : it is valued in a quotient by the symmetry group of the -atogen, the object that the spin bundle of Section 8 carries over a curved manifold.[11]
7. Manifolds and covariant calculus
Everything so far lived in flat space, where a basis can be carried unchanged from point to point. A curved space has no such fixed basis, and differentiating a field requires a rule for comparing vectors at neighbouring points. This section assembles that rule, the covariant derivative, and the curvature it reveals.
7.1. Manifolds and tangent spaces
A manifold of dimension is a space covered by coordinate charts, each a smooth identification of a patch of with an open set of . At each point the tangent space is the vector space of velocities of curves through , with the coordinate vectors as a basis. A Riemannian metric assigns to each tangent space an inner product , varying smoothly with the point.
The vectors and tensors of Section 1 now live in the tangent space at each point, and the metric that raises and lowers their indices is itself a field. The new feature is that the basis turns from point to point, so the naive derivative of a component is not a tensor.
7.2. The covariant derivative and the connection
To differentiate a vector field and keep a tensor, the change in its components must be corrected for the turning of the basis. The correction is a connection.
The covariant derivative of a vector field is
with connection coefficients that account for the change of basis. On a Riemannian manifold the unique connection that is compatible with the metric, , and symmetric in its lower indices is the Levi-Civita connection, whose coefficients are the Christoffel symbols
On the sphere of radius with coordinates the metric is , , the off-diagonal entries zero. The only non-zero derivative of the metric is , and [eq:christoffel] gives
with the rest vanishing. These are the corrections that keep a vector parallel as it is carried over the curved surface.
7.3. Parallel transport, curvature, and the Ricci tensor
A vector is parallel-transported along a curve when its covariant derivative along the curve vanishes. On a curved space the transported vector depends on the path taken, and the failure of two transports to agree is the curvature, measured by the non-commuting of covariant derivatives.
The Riemann curvature tensor is defined by the commutator of covariant derivatives,
and its contraction is the Ricci tensor,
The Ricci tensor measures how a small bundle of geodesics converges or spreads, equivalently how a volume element is distorted by the curvature; positive Ricci curvature focuses geodesics and shrinks volumes.
Feeding the Christoffel symbols of [eq:sphere-christoffel] into [eq:riemann] and contracting gives, for the sphere of radius ,
so the Ricci tensor is a positive multiple of the metric and the scalar curvature is . The great circles, the geodesics of the sphere, converge: two that start parallel at the equator meet at the pole. This is the curvature that will enter the viscous operator of the active flow through the Weitzenböck identity of Section 10.
8. The spin bundle and the order parameter on a manifold
The flat geometric calculus of Sections 2 and 6 was built on a single vector space. A manifold carries a different tangent space at every point, so the algebra must be assembled point by point into a bundle, and the order parameter becomes a section of it.
8.1. The Clifford bundle
At each point of a Riemannian manifold the tangent space with its metric generates a Clifford algebra as in Section 2. These assemble into the Clifford bundle over . A multivector field is a section, an assignment of a multivector in to each point, and the geometric product acts fibre by fibre.
The grade decomposition, the reverse, and the inner and outer products of Section 2.2 all act within each fibre. What is new is differentiation across fibres, since the algebra at one point is a different copy from the algebra at the next.
8.2. The spin bundle and the order parameter
The even subalgebra at each point gives a fibre of rotors, and these assemble into the bundle of spin groups. The order parameter is built from it.
The unit rotors in the even part of each fibre form the spin bundle , with structure group . The -atic order parameter is a section of the bundle associated to with fibre the quotient of Section 6, where is the symmetry group of the -atogen.
The spin bundle exists only when the manifold is spin, that is when the second Stiefel-Whitney class vanishes, the obstruction to lifting the frame bundle from to its double cover . Every orientable 3-manifold is parallelizable and hence spin, so on the spaces of interest here is always available, a fact returned to in Section 11.
8.3. The spin connection
The Levi-Civita connection of Section 7.2 records the turning of the frame in the Christoffel symbols. The same turning, written in the geometric algebra, is a bivector, since a bivector is the generator of a rotation.
The Levi-Civita connection lifts to a bivector-valued spin connection , the geometric-algebra form of the Christoffel symbols of [eq:christoffel]. The covariant derivative of the order-parameter spinor is
the partial derivative corrected by the spin connection acting on the spinor by the one-sided product of [eq:spinor-transform]. In flat space and the covariant derivative reduces to the partial derivative of the earlier sections.
The covariant derivative [eq:spinor-cov-deriv] is the object that replaces every flat gradient of Sections 4 and 5 when the order parameter lives on a curved manifold. Its square is not the flat Laplacian: carrying a spinor around a small loop and comparing it with itself produces the curvature of [eq:riemann], and the resulting curvature term is the Ricci tensor that Section 10 isolates in the Weitzenböck identity. The covariant free energy and the active dynamics of the next section are written with in place of throughout.
9. Active dynamics: the Beris-Edwards equations
The free energy of Sections 4 and 5 governs equilibrium. An active material is held away from equilibrium: each unit consumes energy and exerts a stress, driving a flow that in turn carries the order parameter. The equations of motion that couple the two are the Beris-Edwards nematodynamics.
9.1. The molecular field
The force that drives the order parameter back towards equilibrium is the functional derivative of the free energy.
The molecular field is the symmetric-traceless functional derivative of the Landau-de Gennes free energy,[9]
the trace subtracted so that is symmetric-traceless like itself.
Take the planar free energy of [eq:ldg-planar]. Varying the elastic term gives , and varying the bulk terms gives , so
The elastic term smooths gradients, the bulk terms pull the magnitude towards the ordered value of Section 4.1. On a curved manifold the flat Laplacian here becomes a covariant one, the subject of Section 10.
9.2. The Beris-Edwards equation and the active stress
In a flow the order parameter is advected and rotated, and relaxes towards the molecular field.
With velocity field , strain rate , and vorticity , the order parameter evolves by[10]
with rotational mobility and the generalised-advection term, carrying the tumbling parameter ,
The vorticity rotates the axis with the flow, the strain aligns it, and sets the balance between flow-alignment and tumbling.
The flow is driven in turn by the active stress. Each unit exerts a force dipole, and the coarse-grained stress is proportional to the order parameter,
with activity coefficient , extensile for . In the overdamped, low-Reynolds regime the velocity is set instantaneously by balancing viscous, pressure, and stress forces,
with viscosity , pressure , and the elastic stress . The active force is the gradient of the order parameter: a distortion of the axis pushes fluid, and the motile defects of Section 4.2 are exactly the places where this gradient is largest, which is why activity sets them in motion.
9.3. The covariant active nematic
On a Riemannian manifold the partial derivatives of [eq:beris-edwards] become the covariant derivatives of Section 8.3. The material derivative is covariant transport along the flow, and the strain and vorticity are built from the covariant velocity gradient,
The order parameter then evolves by the covariant Beris-Edwards equation
with the generalised advection of [eq:corotation] read with in place of , and the molecular field of [eq:molecular-field] with its elastic Laplacian promoted to the connection Laplacian of [eq:rough-laplacian] acting on the order-parameter section. In the geometric language the tensor is the section of [spin-bundle]: the vorticity acts as a bivector co-rotating through the spin-connection derivative [eq:spinor-cov-deriv], and [eq:covariant-be] is the relaxation of towards the geometric molecular field. This is the three-dimensional successor of the surface theory of Zhu, Saintillan, and Chern.[2]
The one operator that does not carry over by this substitution is the viscous operator of the Stokes flow [eq:stokes-active]: replacing it covariantly brings in the Ricci tensor of Section 7.3, through the Weitzenböck identity that Section 10 now establishes.
9.4. The rotor evolution of
The order-parameter equation [eq:covariant-be] has a geometric-calculus twin in which the rotor evolves directly. Write the order parameter as , with a unit rotor of Section 5.2 carrying the orientation and the scalar magnitude. The vorticity of [eq:covariant-kinematics] is a bivector , and it turns the rotor by the one-sided product , the operation by which the spin connection acts in [eq:spinor-cov-deriv]. The order parameter evolves by
the covariant material derivative on the left, its the spin-connection derivative [eq:spinor-cov-deriv]; on the right the vorticity co-rotation , the strain-driven flow-alignment weighted by the tumbling parameter of [eq:corotation], and the relaxation towards the geometric molecular field , the grade-two image of [eq:molecular-field]. This is [eq:covariant-be] in the rotor language: the same equation with carried by and the corotation packaged as a bivector acting on the spinor.
Two rotations act on by the identical one-sided bivector product: the vorticity of the flow and the spin connection of the manifold. Transport along the flow and parallel transport across the curved space are one operation in the geometric algebra, a bivector turning the spinor, which is why the rotor form carries the curved dynamics with no extra machinery.
10. Weitzenböck and geometric Ricci
Section 9 left one operator to translate to curved space: the viscous operator of the Stokes flow. Replacing the partial derivatives covariantly is not the end of it, because on a curved manifold there are two natural second-order operators that disagree, and the gap between them is the Ricci curvature. The identity that names the gap is the Weitzenböck identity.
10.1. Two Laplacians
A velocity field is a vector, equivalently a one-form once an index is lowered, and a one-form on a Riemannian manifold has two Laplacians. The connection Laplacian is the covariant divergence of the covariant gradient,
built directly from the covariant derivative of Section 7.2. The Hodge-de Rham Laplacian is built from the exterior derivative and its adjoint, the operator whose harmonic solutions are the physically meaningful ones. In flat space the two coincide. On a curved manifold they do not.
10.2. The Weitzenböck identity
On a Riemannian manifold the Hodge-de Rham Laplacian and the connection Laplacian on one-forms differ by the Ricci tensor,
with the positive connection Laplacian of [eq:rough-laplacian] and the Ricci tensor of [eq:ricci] entering with a plus sign.
Proof
Both Laplacians are built from two covariant derivatives, and they differ only in the order in which the derivatives are contracted. Reordering a pair of covariant derivatives is not free on a curved manifold: by the definition of curvature in [eq:riemann], the commutator acting on a one-form is
Expanding in covariant derivatives and reducing it to the connection Laplacian [eq:rough-laplacian] leaves exactly one such commutator, whose Riemann tensor is contracted on the index pair that distinguishes the two operators, by the contraction of [eq:ricci]. The leftover is the Ricci term of [eq:weitzenbock].
The Ricci tensor enters as the residue of commuting two covariant derivatives on a curved space. This is the sense in which the curvature is geometric: the manifold forces it through the connection, and the dynamics inherit it without any curvature term inserted by hand.
10.3. The curved active Stokes flow
For an incompressible flow the viscous operator of [eq:stokes-active] is the Hodge Laplacian, so the covariant active Stokes equation reads
with the active stress of [eq:active-stress]. The curvature of the manifold acts on the velocity directly through the term . On a surface the curvature here would be the scalar Gaussian curvature; in three dimensions it is the full Ricci tensor, an object with no two-dimensional analogue.
On the sphere of radius the Ricci tensor of [eq:sphere-ricci] is , so the curvature contribution to the viscous operator is
an isotropic drag proportional to the curvature. A flow on a tightly curved surface feels a stiffness the flat theory has no term for, and this stiffness couples the hydrodynamics to the geometry that the selection principle of the next section exploits.
Both halves of the curved construction have been checked on the two-sphere. The Weitzenböck gap of [eq:weitzenbock] was assembled on a discrete icosphere with the cartan library: the difference between the Lichnerowicz and rough Laplacians applied to a vector field recovers the Ricci term of [eq:sphere-weitzenbock] to machine precision, and on a symmetric-traceless field it recovers the space-form value . The covariant active nematic was then run on the sphere with the volterra solver: the total disclination charge stays pinned at , the Euler characteristic of that Poincaré-Hopf demands, every defect quantised at , and the field coarsens towards the four defects in the tetrahedral arrangement of a nematic on a sphere.
10.4. The covariant active -atic system
The pieces assemble into one statement, the complete covariant active -atic hydrodynamics on a Riemannian 3-manifold, posed in both languages.
Let be a closed oriented Riemannian 3-manifold ([manifold]), which is automatically spin ([spin-bundle]). The state is an order parameter together with a flow: the order parameter is a section of the bundle of [spin-bundle], equivalently the symmetric-traceless field of [eq:Q-3d] with ; the flow is a divergence-free velocity . They satisfy the coupled system
with the molecular field [eq:molecular-field] carrying the connection Laplacian [eq:rough-laplacian], the generalised advection [eq:corotation], and the active stress [eq:active-stress]. The Ricci coupling in the viscous term is forced by the Weitzenböck identity [eq:weitzenbock], with the curvature of [eq:ricci]; the covariant derivatives are the Levi-Civita derivative [eq:cov-deriv] on tensors and the spin-connection derivative [eq:spinor-cov-deriv] on .
The two languages state the same system. In the index calculus it is the three lines of [eq:covariant-system]. In the geometric calculus the order parameter is the single rotor field of [spin-bundle]: the first line transports and co-rotates through the spin-connection derivative [eq:spinor-cov-deriv] and relaxes it to the geometric molecular field, the active stress is the symmetric-traceless content of that drives the second line, and the curvature enters as the Ricci bivector of Section 10.2. One order parameter, one flow, one curvature coupling, written in the two languages of Section 2.5.
On a compact manifold with boundary the same system [eq:covariant-system] holds in the interior, closed by an anchoring condition fixing on and a no-slip condition . The boundary then imposes a defect budget through the anchoring, where the closed case of [closed-budget] imposes none.
The velocity of [eq:covariant-system] generates a flow map that advects every material probe in the fluid. The selection principle of the next section is read off : it measures how fast stretches the oriented volumes carried along with it, and identifies that rate with the topological complexity of the disclination tangle.
11. The action that selects tangles
The pieces are now in place: an order parameter on a spin bundle, a covariant free energy, an active dynamics with curvature in its viscous operator. The remaining question is which disclination configuration the active flow actually realises, and the answer is not the one equilibrium would give.
11.1. No free energy to minimise
The active stress of [eq:active-stress] feeds the flow, the flow advects the order parameter through [eq:beris-edwards], and the relaxation feeds back, yet no single functional has the coupled pair as its gradient flow. The force on configuration space carries a curl, so there is no energy to minimise. The realised disclination configuration is whatever the driven dissipative flow runs down to and stays near, an attractor.
11.2. Reduction to the defect coordinates
Away from their cores the field is slaved to its singular set, so the slow dynamics is a flow on the configuration space of disclination tangles. The lines do not float freely there: they ride the viscometric surfaces of the flow, the loci where the vorticity and the strain rate of [eq:beris-edwards] balance. This is the three-dimensional successor to the constraint that pins a planar defect to the nodal set of the order parameter, the self-constraint of Head and collaborators.[4] The reduction leaves a drift on the space of tangles, the slow vector field whose integral curves are the coarse-grained defect motions.
11.3. The Freidlin-Wentzell quasipotential
The reduced dynamics is noisy, , with a small effective noise of covariance . For such a driven flow the object that replaces the free energy is the Freidlin-Wentzell quasipotential, the least cost of a noise-driven path that reaches a configuration from the attractor,[6]
Because is non-gradient, is not a free energy. It solves a stationary Hamilton-Jacobi equation,
and its gradient differs from by a divergence-free circulation. The stable tangle is the global minimum of ; the metastable configurations, the almost-braids that hold for a while and then rearrange, are its local minima; the barriers between them are the dwell times.
11.4. What the flow maximises
The integrand of the cheapest escape path measures a stretching rate, the rate at which the advection grows the size of a material probe. Make it the dominant volume-growth exponent of the advection, in the sense of Gromov and Yomdin,[7]
the largest exponential rate at which the -volume of an advected probe grows under the flow map of [eq:covariant-system].
The probe is a blade. A -dimensional material element spans an oriented -volume, the grade- blade of Section 2.2, with magnitude . This is the advection of Section 2.6 with the linear map taken to be the differential of the flow map: the outermorphism [eq:outermorphism] carries the blade along as the wedge of its transported edges,
The two languages reach this differently, as in [d-volume-transport]. In components the transport is the induced map on , a sum over the antisymmetrised minors of the Jacobian , and is the square root of the Gram determinant of the transported edges. In the geometric calculus it is the single outer product [eq:blade-pushforward]: a blade is transported as a blade, its grade fixed once. Both give the same , and the geometric form needs one operation, which is why the bridge below is plainest there. The top grade is the pseudoscalar, scaled by the one coefficient ; incompressibility of [eq:covariant-system] pins that determinant at one, so the oriented three-volume is conserved and all the growth lives in the intermediate grades, the advected curves and surfaces .
The growth rate of these blades is, by Gromov and Yomdin, the topological entropy of the flow map, and topological entropy is the rate at which the flow manufactures topological complexity. The volume-growth exponent [eq:volume-growth], the topological entropy of , and the braiding-and-linking rate of the defects are the same number,
with the topological complexity the disclination worldlines of Section 5.3 accumulate in time , the braid-word length in the plane and the total linking and knotting in space. A blade cannot grow except by wrapping the moving defects, so its stretching rate is the rate at which the worldlines braid and link: the geometric growth of an advected oriented volume equals the topological complexity of the tangle. For a smooth flow this exponent is the sum of the positive Lyapunov exponents [eq:lyapunov-def] by [volume-growth-lyapunov], which the Pesin-Ruelle relation ties to the metric entropy under SRB conditions and bounds in general. The two-dimensional case of Section 11.7 is the face, where is the braid word and is the logarithm of the pseudo-Anosov dilatation.
Among the attractors of the reduced disclination dynamics, the configuration realised by the active flow maximises the volume-growth exponent [eq:volume-growth], equivalently the topological entropy of the disclination tangle through the bridge [eq:the-bridge]. The stable tangle is the global minimum of the quasipotential [eq:quasipotential], and the metastable configurations are its local minima. In two dimensions on a periodic orbit this reduces to the maximum-topological-entropy braid.
On a closed orientable 3-manifold the Euler characteristic vanishes and the tangent bundle is trivialisable, so the Poincaré-Hopf and Gauss-Bonnet constraints impose no net defect count. The disclination content is fixed by [selection] instead of by topology, with the fundamental group entering through the linking and framing of the disclination loops and the curvature of Section 7.3 through the biasing of their positions.
Proof
Every closed orientable 3-manifold is parallelisable by the theorem of Stiefel, so a global frame exists and a non-vanishing director field has no topologically forced zeros. The Euler characteristic of any closed odd-dimensional manifold vanishes, , so the Poincaré-Hopf index sum is zero and topology forces no net defect.
A fixed tangle has a topological type, its knot and link invariants, but as a frozen object it has zero entropy. The entropy of [eq:the-bridge] is a property of the flow, and a frozen configuration carries none of it: it is the rate at which the moving defects regenerate complexity, the pseudo-Anosov stretch per period of Section 11.7. A configuration the flow leaves untangled contributes nothing; one whose worldlines wind without bound contributes the logarithm of the dominant stretch.
The untangled configuration, its disclination loops unknotted and unlinked, is the minimum of the bridge [eq:the-bridge], since an advected blade that wraps nothing does not grow. By [selection] the active flow maximises , so it runs away from the untangled state towards linked and knotted tangles. The active drive is itself a source of topological complexity.
11.5. Detecting the disclination lines
The selected tangle is a set of disclination lines, and the geometric calculus reads them off the rotor field directly. A disclination is a place where the orientation cannot be combed continuously, found two equivalent ways.
The magnitude vanishes at the core. Where the order melts, and the eigenframe of the -tensor of [eq:Q-3d] degenerates; the locus is codimension two, the disclination curve of Section 5.1.
The holonomy is nontrivial around it. Take a small loop and carry the rotor around it by the spin-connection transport of [eq:spinor-cov-deriv]. The net rotor is the path-ordered Wilson loop
an element of the symmetry group . When a disclination threads , and the element is its charge: for the uniaxial nematic reduces to and is the half-integer line, the rotor that is the spinor sign-flip of Section 6.2; for biaxial order is a quaternion unit of the group of Section 5.3, the non-abelian charge. The index calculus finds the same line by the degeneracy of the -eigenvalues and the winding of the director frame, but the director is defined only up to sign and the winding is awkward to track through that ambiguity; the rotor carries the sign in cleanly, the sharper instrument once more. One scans loops over the manifold, evaluates the holonomy [eq:holonomy], and flags the nontrivial -elements: their loci are the disclination lines and their elements are the charges.
A few tangles show what the three-dimensional theory can hold that the plane cannot.
The simplest tangle with no planar analogue is two disclination loops of linking number one, a Hopf link. A small loop around one strand carries the rotor holonomy [eq:holonomy] to , the charge, while a loop around both returns the identity. The two cannot be drawn apart without one strand crossing the other, a reconnection the incompressible flow forbids, so the link is a protected attractor of the reduced dynamics.
Three biaxial loops can sit in a Borromean arrangement, pairwise unlinked yet collectively inseparable. Every pairwise holonomy [eq:holonomy] is trivial, so no abelian charge records the entanglement; the obstruction is the non-commutativity of [eq:quaternion-noncommute], the failure of the loop holonomies in to commute, a triple-linking the pairwise charges cannot see. This is a tangle the uniaxial nematic cannot hold and the biaxial order of Section 5.3 can, purely non-abelian and purely three-dimensional.
A single uniaxial loop can knot, the lowest being the trefoil, since a curve in space has room a planar point never had. Its complexity is the self-linking of the loop against a framing pushed off along the order-parameter axis, the degree of knottedness of a tangled line in the sense of Moffatt.[12] A knot feeds of [eq:the-bridge] through the rate at which the flow must thread the order parameter around it.
11.6. The curved tangle and how to compute it
The pieces run as one program. Evolve the rotor field by [eq:rotor-evolution] on a discrete 3-manifold, the covariant derivatives furnished by the spin connection [eq:spinor-cov-deriv] and the velocity by the curved active Stokes flow [eq:covariant-stokes]. At each step detect the disclination lines by the holonomy [eq:holonomy] and follow their worldlines. Advect a material -blade by the differential of the flow map [eq:blade-pushforward] and read its growth: by the bridge [eq:the-bridge] the stretching rate is the topological entropy, equal to the braiding and linking rate of the detected lines. The quasipotential [eq:quasipotential] is then estimated from the fluctuation paths, and the tangle that minimises it is the one [selection] predicts.
The curvature enters wherever a derivative does, and its signature is geometric. The Ricci stiffness [eq:sphere-weitzenbock] in the viscous operator makes the flow harder to drive where the manifold curves more, so the active force concentrates the disclination lines towards the geometry that lets them move, biasing their positions by the curvature of Section 7.3. On a closed 3-manifold the topological budget is empty by [closed-budget], so the curvature and the selection principle set the entire disclination content between them: the lines arrange to maximise the curved-volume stretching of [eq:the-bridge] while the Ricci term in the drift pins where on the manifold they sit. This is the behaviour the planar theory cannot show, the advected volumes of Section 2.6 growing in a curved space whose Ricci tensor both stiffens the flow and shapes the tangle. The author's covariant solver carries this out in the plane, the computation of Section 11.7, and the full solver on a 3-manifold mesh is the work of the companion paper.
The established results, the conjecture, and the open problem separate cleanly. The order-parameter bundle of Section 8, the covariant active -atic system [eq:covariant-system] of Sections 9 and 10, its rotor evolution [eq:rotor-evolution], the holonomy detection [eq:holonomy], and the blade-to-entropy bridge [eq:the-bridge] are established. The maximal-stretching selection of [selection] is a conjecture of maximum-entropy-production type, supported in two dimensions by the braiding data and reformulated here as the maximisation of a volume-growth exponent. What remains to be built is the reduced drift of [eq:quasipotential] in closed form and the solver on a 3-manifold mesh that would test the selection directly, the map from the active stress and the Ricci curvature to the slow vector field on the space of tangles.
11.7. The two-dimensional shadow
[selection] has a verified instance in the plane. Restricting to a surface and a few strongly confined defects collapses [eq:volume-growth] to its one-dimensional case: the only probe that grows exponentially is a material curve, so the volume-growth exponent reduces to the topological entropy of the braid the defect worldlines trace,
with the pseudo-Anosov dilatation of the braid , the factor by which the optimal material curve stretches per period.[8]
Three defects in strong cardioid confinement settle on the golden braid, with per-period braid word and topological entropy , where is the golden ratio, numerically . Four defects settle on the silver braid, with , where is the silver ratio, numerically . In a circular boundary five or more defects no longer admit a periodic braid and the motion is aperiodic, but a cusped boundary can restore periodicity through a defect-gyre balance: on the trefoiloid a centrally pinned sixth defect brings the count level with the six flow gyres, and the defects resume a periodic Ceilidh dance, a six-defect orbit carrying the silver dilatation once more, the metal unchanged.[1]
The two entropies are the exact metallic-mean values of the pseudo-Anosov dilatations, the dilatation being the spectral radius of the reduced Burau matrix of the period word at .
| braid | period word | dilatation | |
|---|---|---|---|
| golden, 3 defects | |||
| silver, 4 defects |
These two are the first members of an infinite family. The same Burau computation, carried out at an arbitrary winding, shows that every metallic mean is a pseudo-Anosov dilatation.
For each integer the three-strand braid is pseudo-Anosov with dilatation
where is the -th metallic mean, the positive root of . Its topological entropy is therefore . The first three are the golden , the silver , and the bronze , with dilatations , , and .
Proof
Use the reduced Burau representation of the three-strand group at , the representation the dilatations above are read from,
Both generators are unipotent shears, so and . The word is their product,
with trace and determinant , hence characteristic polynomial . Its roots are a reciprocal pair, and since the matrix is hyperbolic, so the braid is pseudo-Anosov with dilatation the larger root, the spectral radius, and topological entropy its logarithm by Thurston-Nielsen.[8] Finally the metallic mean obeys , so , which identifies the dilatation as the square of the metallic mean.
The golden braid of the table is exactly, and the four-defect silver braid carries the dilatation of . The metal indexes the stretch per period, the Burau spectral radius of [eq:metallic-dilatation-eq], and not the number of defects: the six-defect Ceilidh orbit above returns at the silver dilatation, so adding defects does not climb the metallic ladder. The bronze and the higher means are reached instead by winding, the trace of [metallic-dilatation] climbing past silver's six to eleven and beyond as grows. The golden and silver are also the two endpoints of the Thiffeault-Finn optimisation, the most efficient braids per generator and per operation, so they close a pair under that measure even as the metallic family runs on without bound. The optimum depends on the braid presentation: enlarging the three-strand group to graph-generated surface braids on the torus lifts the per-operation maximum above either metal, to the lattice braid Smith and Dunn identify with entropy per operation , its dilatation a Salem number.[18]
The detection, tracking, and braid-word extraction pipeline recovers these dilatations and entropies from the defect worldlines to machine precision, and the covariant Beris-Edwards run on a confined disk settles into the three defects of the golden configuration, stable across the entire settled window. These cases are reproduced from the author's covariant active-nematic solver, the cartan Riemannian-geometry library and the volterra Beris-Edwards solver built on it: discrete exterior calculus on well-centred meshes, the covariant Laplacian of [eq:molecular-planar] for the relaxation, and defect detection with braid-word extraction. The topological entropy of the extracted braid equals the logarithm of its pseudo-Anosov dilatation, which for the golden and silver braids takes the metallic-ratio values and , and these are corroborated at the field level by line-stretching and ensemble-tracer estimates of the stretching rate, all agreeing with the planar braiding observed by Klein, Soto Franco, and collaborators.[1]
The advected-blade picture of Section 2.6 was checked on the same run. The determinant of the one-period flow-map outermorphism [eq:outermorphism-det] holds at unity across the advected tracer field, a median of , so the incompressibility of [eq:covariant-system] pins the top-grade pseudoscalar and confines the stretching to the intermediate grade , where an advected material line grows exponentially at a rate of the order of the braid entropy. The equality of that rate with the pseudo-Anosov value [eq:shadow-entropy] is the content of the bridge [eq:the-bridge], a theorem of Thurston-Nielsen and Gromov-Yomdin type; the field-level line-stretching estimates above are its numerical corroboration.
The boundary between what is shown and what remains is sharp. These cases are a slice of [eq:volume-growth], a one-dimensional curve advected on a two-dimensional surface. The three-dimensional version asks for the growth rate of an advected surface, , and records its invariants as the linking and knotting of disclination worldsheets instead of a braid word. Computing it requires the solver on a three-dimensional mesh and the reduced drift of [eq:quasipotential]; no three-dimensional disclination-tangle computation is claimed here, and it is the subject of the companion paper.
12. The deforming manifold and active remodelling
Every section so far fixed the manifold and let the order parameter live on it. Living matter does not hold still. An epithelial sheet, a developing tissue, a cell cortex: the active units that carry the order parameter also push their own substrate, so the geometry is a degree of freedom that the order parameter remodels, the active-matter setting of living tissue.[13] A glioma is the three-dimensional example: its active-nematic cells stream collectively and invade the surrounding brain, remodelling the tissue they move through,[15][16] a process modelled from the field level of the order parameter down to the stochastic kinetics of cell numbers,[17] and carrying the three-dimensional disclination lines of Section 5.3. The framework already holds what this needs.
12.1. The space of shapes
A fixed manifold sits at a single point of a larger space. Let be the space of Riemannian metrics on , the cone of positive-definite symmetric two-tensor fields; a deforming manifold is a curve in it. The space carries a natural metric of its own, the DeWitt inner product on symmetric two-tensors ,
so distances between shapes are measured with the same metric that measures lengths within a shape. The state of the deforming theory is a point of the extended configuration space
a metric paired with an order-parameter section of [spin-bundle]. Give the product of the DeWitt metric [eq:dewitt] and the metric on sections, the reference ruler that will measure complexity once the dynamics moves a probe through it. Where Sections 7 to 11 built one tangent space at each point of a fixed , the deforming theory builds one copy of the order-parameter bundle over each point of .
12.2. The metric evolution: elastic and plastic
The active stress [eq:active-stress] drives the flow [eq:covariant-stokes], and the flow carries the material points. The metric evolves by the strain rate of [eq:covariant-kinematics], split into the part a material motion contributes and the part the active rearrangements add,
with the elastic rate, the Lie drag along the velocity, and the plastic remodelling rate. The distinction is geometric. The elastic part alone is a relabelling of points: it carries to the pullback along the flow map of Section 11.4, an isometry, leaving the intrinsic curvature unchanged. The plastic part is what genuinely reshapes the geometry, the cell divisions, neighbour exchanges, and symmetry changes that no smooth material motion accounts for, and it alone moves the Ricci curvature. Where the plastic rate drives a direction of the metric to zero length, a sheet of tissue thinning to a film, the tangent-space quadratic form degenerates and the local algebra passes from into the degenerate signature with of [general-signature], the collapsed direction entering the radical.
The state is the point of the extended configuration space [eq:extended-config], with the velocity set instantaneously by the flow. It evolves by the covariant active -atic system [eq:covariant-system] for coupled to the metric evolution [eq:metric-evolution], the curvature recomputed from the moving metric at each instant. Write the flow it generates on as .
The coupling closes on itself: the order parameter sets the active stress, the stress drives the flow, the plastic part of [eq:metric-evolution] remodels the metric, the metric moves the Ricci curvature, and the Ricci term of [eq:covariant-stokes] changes the flow that carries the order parameter. The metric evolution is the active counterpart of a geometric flow: where Ricci flow evolves a metric by its own curvature, here the metric evolves by the active stress of the order parameter living on it.
12.3. The hexatic-to-nematic remodelling
Much of living active matter is driven by a change of . A confluent tissue at small scales is hexatic, its cells six-fold coordinated, ; coarsened or elongated, the same tissue reads as nematic, . In the language of Section 3 this is a change of the -atogen symmetry from to , hence a change of the fibre of the order-parameter bundle of [spin-bundle].
As drops from to the elementary defects change charge, the disclinations of the hexatic coalescing into the disclinations of the nematic. Each coalescence is a burst of plastic remodelling in [eq:metric-evolution], a local reshaping of the tissue. The holonomy [eq:holonomy] records it, its value passing from the class to the class as the symmetry breaks. The order-parameter transition and the shape change are one event.
12.4. Blades on the space of shapes
The complexity argument of Section 11.4 now runs on . A probe is a -dimensional family of states, a blade built from tangent vectors to , each one a metric variation and an order-parameter variation carried together. The coupled flow advects it by the outermorphism [eq:outermorphism] of Section 2.6, now applied to the differential of the flow on the space of shapes, exactly as in the fixed case,
its volume measured by the reference metric of [eq:dewitt] and [eq:extended-config]. The growth of [eq:shape-blade] has two sources wedged into one oriented volume: the stretching of the order-parameter tangle of Section 11.4 and the stretching of the geometry itself along the curve in . A blade on the space of shapes records both at once.
12.5. The morphogenetic entropy functional
The derivation of Section 11 repeats on . The reduced coupled dynamics is a drift on with a small effective noise, and the object that replaces the free energy is the morphogenetic quasipotential,
the least cost of a fluctuation path through the space of shapes, the quasipotential [eq:quasipotential] lifted from the space of tangles to . Its integrand is again a stretching rate, and the supremum over probe dimension is the morphogenetic entropy,
the topological entropy of the coupled order-parameter-and-geometry flow. The bridge of [eq:the-bridge] rises with it,
with the disclination-tangle complexity of [eq:the-bridge] and the shape complexity the plastic remodelling of [eq:metric-evolution] accumulates in time , their product the count of distinguishable tangle-and-shape histories.
For the self-remodelling system [deforming-system] the configuration the living matter realises is the global minimum of the morphogenetic quasipotential [eq:morpho-quasipotential], equivalently the maximiser of the morphogenetic entropy [eq:morpho-entropy]. The matter settles into the shape and the tangle that together generate topological complexity fastest.
12.6. The topological-entropy lineage
One quantity has run through the article, the exponential rate at which an advected oriented volume grows, and it has been a topological entropy at every level. Topological entropy is an invariant of the flow, independent of the metric used to measure the volumes of [eq:volume-growth], [eq:the-bridge], and [eq:morpho-entropy], so the three are comparable though each measures with a different ruler. They form a tower.
The braid entropy of Section 11.7, the fixed-manifold blade entropy of [eq:the-bridge], and the morphogenetic entropy of [eq:morpho-entropy] are nested,
The first inequality is exact: the braid is the term of the supremum [eq:volume-growth], no larger than the supremum itself. The second is the expected ordering: the fixed-manifold flow is the rigid-geometry case of [eq:metric-evolution] with the metric clamped, and the coupled flow extends it by the geometric degrees of freedom, which a flow does not lose entropy by carrying. Making the second inequality precise, by exhibiting as a factor of , belongs to the open program.
Each level is the shadow of the one above. The planar braid of Section 11.7 is the lowest, a curve advected on a surface; the fixed-manifold tangle is next, blades advected on a rigid 3-manifold; the morphogenetic entropy is highest, blades advected on a 3-manifold the order parameter remodels as it goes. The crowning quantity of the modelling is the topmost, the rate at which living matter builds shape and tangle at once. Solving [eq:metric-evolution] alongside the order parameter on a deforming 3-manifold and computing directly is the most demanding and most physical version of the program, and is left to future work.
13. Outlook
The sections assemble one object. A single order parameter, built from the quaternionic even subalgebra of Section 5.2 and carried over a manifold by the spin bundle of Section 8, holds polar, nematic, biaxial, and cholesteric order on any Riemannian 3-manifold through the single dial . The covariant Onsager flow of Sections 9 and 10 drives it, with curvature entering the viscous operator as the Ricci tensor through the Weitzenböck identity. The active coupling is non-variational, so no free energy governs the realised disclination structure; it is fixed instead by the quasipotential of Section 11, whose two-dimensional shadow is the braid hierarchy.
The algebraic description of liquid-crystal defects in three dimensions through geometric algebra is due to Johnson, Head, Lavrentovich, Morozov, Negro, Orlandini, Smith, Vasil, and Marenduzzo, who distinguished uniaxial, biaxial, and cholesteric defects through distinct Clifford algebras and introduced the defect bivector.[3] The present article places that classification in one bundle, the single dial , and sets it in motion: covariant active hydrodynamics with curvature and topology as variables, and a non-equilibrium principle for which configuration the flow realises. The surface theory of Zhu, Saintillan, and Chern is the two-dimensional covariant predecessor whose Onsager structure is lifted here to three dimensions,[2] and the self-constraint of Head and collaborators is the kinematic reduction underlying the passage to tangle coordinates in Section 11.[4]
The open program has four parts. The first is to write the reduced drift of [eq:quasipotential] in closed form, mapping the active stress and the Ricci curvature of [eq:covariant-stokes] to the slow vector field on the space of disclination tangles. The second is to build the solver on three-dimensional meshes, the successor to the planar solver of Section 11.7. The third is to compute disclination-tangle attractors on closed 3-manifolds, where [closed-budget] leaves the configuration to selection alone, and on confined domains with anchoring boundaries, where the boundary imposes a defect budget. The fourth is to test the maximal-stretching selection of [selection] against those computations. The drift, the solver, and the three-dimensional computations are developed in the companion paper. Beyond the fixed manifold lies the self-remodelling coupling of Section 12, where the order parameter deforms the metric it lives on through [eq:metric-evolution]; the morphogenetic entropy of [morphogenetic-entropy] is its organising conjecture and the most physical reach of the program.
References
- B. Klein, A. J. Soto Franco, M. M. H. Sabbir, M. J. Deutsch, R. Kliegman, R. L. B. Selinger, K. A. Mitchell, D. A. Beller (2026). Chaos-Generating Periodic Orbits of Topological Defects in Confined Active Nematics. arXiv:2503.10880
- C. Zhu, D. Saintillan, A. Chern (2024). Active Nematic Fluids on Riemannian 2-Manifolds. arXiv:2405.06044
- N. Johnson, L. C. Head, O. D. Lavrentovich, A. N. Morozov, G. Negro, E. Orlandini, C. A. Smith, G. M. Vasil, D. Marenduzzo (2025). Clifford Algebras and Liquid Crystalline Fermions. arXiv:2504.08519
- L. C. Head, C. Doré, R. Keogh, L. Bonn, A. Doostmohammadi, K. Thijssen, T. López-León, T. N. Shendruk (2023). Spontaneous Self-Constraint in Active Nematic Flows. arXiv:2306.05328
- C. Doran, A. Lasenby (2003). Geometric Algebra for Physicists. Cambridge University Press. DOI
- M. I. Freidlin, A. D. Wentzell (2012). Random Perturbations of Dynamical Systems, 3rd ed. Springer. DOI
- Y. Yomdin (1987). Volume Growth and Entropy. Israel Journal of Mathematics 57(3), 285–300. DOI
- W. P. Thurston (1988). On the Geometry and Dynamics of Diffeomorphisms of Surfaces. Bulletin of the American Mathematical Society 19(2), 417–431. DOI
- P. G. de Gennes, J. Prost (1993). The Physics of Liquid Crystals, 2nd ed. Oxford University Press.
- A. N. Beris, B. J. Edwards (1994). Thermodynamics of Flowing Systems with Internal Microstructure. Oxford Engineering Science Series 36. Oxford University Press.
- D. Hestenes, G. Sobczyk (1984). Clifford Algebra to Geometric Calculus. Reidel. DOI
- H. K. Moffatt (1969). The degree of knottedness of tangled vortex lines. Journal of Fluid Mechanics 35(1), 117–129. DOI
- M. C. Marchetti, J. F. Joanny, S. Ramaswamy, T. B. Liverpool, J. Prost, M. Rao, R. A. Simha (2013). Hydrodynamics of soft active matter. Reviews of Modern Physics 85, 1143. DOI
- A. E. Argento, M. L. Varela, G. Singh, D. P. Visnuk, B. Jacobovitz, M. E. Rutherford, M. B. Edwards, Q. Chaboche, D. A. Orringer, J. A. Heth, M. G. Castro, D. A. Beller, C. Blanch-Mercader, P. R. Lowenstein (2025). Three-dimensional topological defects and quasi-long-range order in biological liquid crystals. bioRxiv. DOI
- L. Barberis, C. A. Condat, S. M. Faisal, P. R. Lowenstein (2024). The self-organized structure of glioma oncostreams and the disruptive role of passive cells. Scientific Reports 14, 25435. DOI
- A. Comba, S. M. Faisal, P. J. Dunn, A. E. Argento, T. C. Hollon, S. Motsch, M. G. Castro, P. R. Lowenstein, et al. (2020). Spatiotemporal analysis of glioma heterogeneity reveals Col1A1 as an actionable target to disrupt tumor mesenchymal differentiation, invasion and malignancy. bioRxiv. DOI
- K. Poffenbarger, R. Pandey (2024). Analyzing the birth-death model of Oncostreams in Glioma, and the effects of Cytochalasin D treatment. arXiv:2407.10983
- S. A. Smith, S. Dunn (2021). Topological Entropy of Surface Braids and Maximally Efficient Mixing. arXiv:2112.07619