nLab affine space




An affine space or affine linear space is a vector space that has forgotten its origin. An affine linear map (a morphism of affine spaces) is a linear map (a morphism of vector spaces) that need not preserve the origin.

Note that the ‘linear functions’ of elementary algebra —the total functions whose graphs are lines— are in fact (precisely) affine \mathbb{R}-linear maps from \mathbb{R} to itself. (Similarly, the ‘linear relations’ —the relations whose graphs are lines— are precisely the projective \mathbb{R}-linear maps.)

Alternatively, in algebraic geometry, the terminology “nn-dimensional affine space” 𝔸 nk\mathbb{A}^n k (affine line, affine plane, etc.) over a field kk refers to, depending on context, the set k nk^n, or the set of maximal ideals of the polynomial algebra k[x 1,,x n]k[x_1, \ldots, x_n] – these definitions coinciding if kk is an algebraically closed field – and typically considered as equipped with relevant extra structure such as a Zariski topology or, going even further, the locally ringed space structure adhering to the affine variety or affine scheme corresponding to the polynomial algebra k[x 1,,x n]k[x_1, \ldots, x_n]. Whatever the precise sense chosen, the idea is that an affine space 𝔸 nk\mathbb{A}^n k is a setting in which the study of loci of polynomial equations, i.e. definable sets in the theory of commutative algebras over kk, is carried out.

Most of this article concerns affine spaces in the sense of vector spaces that have forgotten their origins or identities; the algebraic geometry sense is very briefly touched upon in the section Affine spaces as model spaces.



The definition of affine space can be made precise in various (equivalent) ways. We give a name to some of the definitions for later reference.

  • An affine space is simply a vector space, but with different morphisms; an affine linear map is a function that is the difference between a linear map and a constant function.

  • An affine space is a set equipped with an equivalence class of vector space structures, where two vector space structures are considered equivalent if the identity function is affine linear as a map from one structure to the other; whether a map between affine spaces is affine linear is independent of the representative vector space structures.

  • An affine space is a set AA together with a vector space VV and an action of (the additive group or translation group of) VV on AA that makes AA into a VV-torsor (over the point); an affine linear map is a VV-equivariant map. For this point of view, see also affine space.

  • An affine space is a heap whose automorphism group is equipped with structure making it the additive group of a vector space; an affine linear map is a heap morphism.

  • An affine space is an inhabited set AA together with a vector space VV and a function Λ:A×AV\Lambda\colon A \times A \to V (thought of Λ(x,y)xy\Lambda(x,y) \coloneqq x - y) that satisfies some equations; an affine linear map AAA \to A' is a function equipped with a linear map VVV \to V' relative to which it preserves subtraction (the “vector-valued difference” definition).

  • An affine space over the ground field kk is an inhabited set AA together with functions μ:A×A×AA\mu\colon A \times A \times A \to A (thought of as μ(x,y,z)xy+z\mu(x,y,z) \coloneqq x - y + z) and Λ *:k×A×AA\Lambda_*\colon k \times A \times A \to A (thought of as Λ r(x,y)xrx+ry\Lambda_r(x,y) \coloneqq x - r x + r y) that satisfy some equations; an affine linear map is a function that preserves these operations (the “two ternary operations” definition).

  • An affine space over kk is an inhabited set AA together with a function μ *:k×A×A×AA\mu_*\colon k\times A\times A\times A\to A (thought of as μ r(x,y,z)rxry+z\mu_r(x,y,z) \coloneqq r x - r y + z) that satisfies some equations; an affine linear map is a function that preserves this operation (the “one quaternary operation”) definition.

  • Assuming that 22 is invertible in the field kk (i.e. the characteristic of kk is not 22), an affine space over kk is an inhabited set AA together with a function Λ *:k×A×AA\Lambda_*\colon k \times A \times A \to A that satisfies some equations; an affine linear map is a function that preserves this operation (the “one ternary operation” definition).

  • An affine space over the field kk is an inhabited set AA together with, for every natural number n0n \geq 0 and every (n+1)(n+1)-tuple (r 0,,r n)(r_0,\dots,r_n) of elements of kk such that r 0++r n=1r_0 + \dots + r_n = 1, a function γ r 0,,r n:A n+1A\gamma_{r_0,\ldots,r_n}\colon A^{n+1}\to A (thought of as γ r 0,,r n(x 0,,x n)r 0x 0++r nx n\gamma_{r_0,\ldots,r_n}(x_0,\ldots,x_n) \coloneqq r_0 x_0 + \cdots + r_n x_n), satisfying some equations; an affine linear map is a function that preserves these operations (the “unbiased” definition).

  • An affine space over the field kk is a vector space AA' together with a surjective linear map π:Ak\pi:A'\to k (the “slice of VectVect” definition). The affine space itself (the set being regarded as equipped with affine-space structure) is the fiber π 1(1)\pi^{-1}(1).

Mike Shulman: I think there should also be a definition of the form “an affine space is a projective space” with a distinguished line called “infinity”, which should also be equivalent to a “synthetic” description involving points and lines and incidence axioms. This definition would not fix the field kk at the outset, but rather recover it synthetically using cross-ratios. Accordingly, it ought to define an equivalent groupoid to the groupoid of pairs (k,A)(k,A) where kk is a field and AA is an affine space over AA. I don’t know how one could recover the non-invertible affine transformations from it directly.

There should be another characterisation, which I don't quite see how to phrase, at least when k=k = \mathbb{R}, which is that an affine space is a manifold (perhaps Riemannian) that is sufficiently flat and unbounded in some sense. —Toby

Mike Shulman: It’d have to be at least Riemannian, otherwise you don’t have enough structure. I don’t suppose it’s enough to say that a (finitely generated) affine space is a Riemannian manifold isometric to some n\mathbb{R}^n?

Toby: I intended ‘that is [] in some sense’ to include the possibility of structure that should be preserved by the morphisms. Note that a Riemannian manifold is too much structure, although it allows a definition like the first one above. (A Riemannian manifold isometric to some n\mathbb{R}^n is precisely a Euclidean space.) But really, I'm hoping for some phrasing such that ‹isomorphic to some n\mathbb{R}^n› actually becomes a (not too obvious) theorem. I'll keep thinking about it.

Mike Shulman: Shouldn’t a Riemannian manifold isometric to some n\mathbb{R}^n be a “Euclidean affine space” (a torsor over a Euclidean space)? Seems that a Euclidean space would be a Riemannian manifold equipped with an isometry to some n\mathbb{R}^n. It does seem like there should be a natural way to say this, but I don’t know what it is.

Toby: I guess that this depends on what you think ‘Euclidean space’ means; I've known people to define it to be n\mathbb{R}^n, but that seems quite ahistorical to me; I like that Urs calls such a thing Cartesian space instead. Euclid did not have coordinates; he did not even have an origin, so a Euclidean space should be a heap rather than a group. For my comment above, I would define a Euclidean space to be an affine inner product space; FWIW Wikipedia agrees. (However, Wikipedia doesn't go as far as I do when I claim that the inner product should be valued in an \mathbb{R}-line rather than in \mathbb{R} itself; then again, I ignored that subtlety myself in my previous comment.)

Clearly every vector space has an underlying affine space (and every linear map is affine linear), giving a forgetful functor U:VectAffU:Vect \to Aff. Conversely, any affine space gives rise to a canonical vector space, sometimes called its space of displacements. This is obvious from the definitions that involve a vector space as part of the structure, but a vector space can also be reconstructed from the other definitions as well, analogously to how a group can be reconstructed from a heap. This gives a functor D:AffVectD:Aff\to Vect in the other direction. One can verify that D(U(V))VD(U(V))\cong V and U(D(A))AU(D(A))\cong A; the first isomorphism is natural, but the second is not (otherwise VectVect and AffAff would be equivalent categories, which they are not).

The category of affine spaces is almost a variety of algebras, as can be seen from the last few definitions, except for the requirement that an affine space be inhabited. To rectify this, sometimes one allows the empty set to be an affine space, although it does not have any particular vector space of displacements. (See heap#empty for discussion.)

Note that there are a few different ways to think about the operations involved in the final three definitions (those not explicitly involving a vector space). The operation μ:x,y,zxy+z\mu\colon x,y,z \mapsto x - y + z is the same as the Mal'cev operation (i.e. heap structure) of the additive group of a vector space. It can be viewed as the point completing a parallelogram with given vertices x,y,zx,y,z, or equivalently as the result of adding xx and zz, relative to a choice of yy as the origin. The operation Λ *:r,x,yxrx+ry\Lambda_*\colon r,x,y \mapsto x - r x + r y can be viewed as either a weighted average of xx and yy (i.e. as (1r)x+ry(1-r)x + r y) or as the result of multiplying the “displacement vector” yxy-x by rr, relative to the origin xx (i.e. as x+r(yx)x + r(y-x)).

Details and comparisons

The first few definitions, which explicitly involve a vector space, make no especial use of the fact that the vector space is a vector space rather than merely an abelian group. Thus, they are valid (and equivalent) in the more general context of torsors and heaps. They are also mostly complete as stated, except for the final one.

Vector-valued differences

In this definition, an affine space over a vector space VV is a set AA together with a “subtraction” function Λ:A×AV\Lambda\colon A\times A\to V, written Λ:x,yxy\Lambda\colon x,y \mapsto x-y, such that:

  • Λ(x,x)=0\Lambda(x,x) = 0, or xx=0x-x = 0, for any xx in AA.
  • Λ(x,y)+Λ(y,z)=Λ(x,z)\Lambda(x,y) + \Lambda(y,z) = \Lambda(x,z), or (xy)+(yz)=(xz)(x-y) + (y-z) = (x-z), for any x,y,zx,y,z in AA.
  • For any xx in AA and vv in VV there exists a unique yy in AA such that Λ(y,x)=v\Lambda(y,x) = v, or yx=vy - x = v.

If yx=vy - x = v, then we write y=x+vy = x + v, which we can regard as an operation on xx and vv by the third axiom. Hence we have (x+v)x=v(x + v) - x = v and (by uniqueness) x+(yx)=yx + (y - x) = y, and also x+0=xx + 0 = x and (x+v)+w=x+(v+w)(x + v) + w = x + (v + w) by the first two axioms. Thus, these axioms suffice to make AA into a torsor over the additive group of VV with the action ++, which is one of the previous definitions given.

Note again that this would makes sense if VV is any group, not just the additive group of a vector space.

Two ternary operations

This definition is an affine version of the usual definition of a vector space in terms of addition and scalar multiplication. However, in each case the affine operation needs to take an extra parameter. In reading the following axioms it helps to think of μ(x,y,z)\mu(x,y,z) as “the sum of xx and zz relative to the basepoint yy” and likewise Λ r(x,y)\Lambda_r(x,y) as “the product ryr\cdot y relative to the basepoint xx”.

  • μ(x,x,y)=y\mu(x,x,y) = y (identity for addition)
  • μ(x,y,μ(z,w,v))=μ(μ(x,y,z),w,v)\mu(x,y,\mu(z,w,v)) = \mu(\mu(x,y,z),w,v) (associativity of addition)
  • μ(x,y,z)=μ(z,y,x)\mu(x,y,z) = \mu(z,y,x) (commutativity of addition)
  • Λ rs(x,y)=Λ r(x,Λ s(x,y))\Lambda_{r s}(x,y) = \Lambda_r(x, \Lambda_s(x,y)) (associativity of scalar multiplication)
  • Λ 1(x,y)=y\Lambda_1(x,y) = y (identity for scalar multiplication)
  • Λ 0(x,y)=x\Lambda_0(x,y) = x (left zero for scalar multiplication)
  • Λ r+s(x,y)=μ(Λ r(x,y),x,Λ s(x,y))\Lambda_{r+s}(x,y) = \mu(\Lambda_r(x,y), x, \Lambda_s(x,y)) (left distributivity of scalar multiplication)
  • Λ r(w,μ(x,y,z))=μ(Λ r(w,x),Λ r(w,y),Λ r(w,z))\Lambda_r(w, \mu(x,y,z)) = \mu(\Lambda_r(w,x), \Lambda_r(w,y), \Lambda_r(w,z)) (right distributivity of scalar multiplication)

Joost: Could it be that there is an axiom missing here ? One can go from Vector spaces to the 2 ternary operations definition and back, but I can’t see that by starting with the two ternary operations definition, going to vectorspaces and back, you get the same Λ\Lambda. I guess you need an extra axiom as μ(x,y,Λ r(y,z))=Λ r(x,μ(x,y,z))\mu(x,y,\Lambda_r(y,z))=\Lambda_r(x,\mu(x,y,z)).

Toby: Conceptually, there is something missing, which I've inserted as the left zero property of scalar multiplication. (The right zero property Λ r(x,x)=x\Lambda_r(x,x) = x follows from this using associativity of scalar multiplication with s=0s = 0, same as with vector spaces.) But I have to leave, and I haven't yet derived your axiom, even with this aid.

One quaternary operation

This definition is an affine version of the less standard definition of a vector space in terms of a single operation r,x,yrx+yr,x,y\mapsto r\cdot x + y. Here an affine space over kk is a set AA together with a single operation μ:k×A×A×AA\mu\colon k\times A\times A\times A\to A, written as (r,x,y,z)μ r(x,y,z)(r,x,y,z)\mapsto \mu_r(x,y,z) and thought of as the sum “rx+zr\cdot x + z relative to the basepoint yy,” such that:

  • μ 1(x,y,y)=x\mu_1(x,y,y) = x
  • μ r(x,x,y)=y\mu_r(x,x,y) = y
  • μ 1(x,y,z)=μ 1(z,y,x)\mu_1(x,y,z) = \mu_1(z,y,x)
  • μ r(μ s(x,y,z),w,v)=μ rs(x,y,μ r(z,w,v))\mu_r(\mu_s(x,y,z),w,v) = \mu_{r s}(x,y,\mu_r(z,w,v))
  • μ r+s(x,y,z)=μ r(x,y,μ s(x,y,z))\mu_{r+s}(x,y,z) = \mu_r(x,y,\mu_s(x,y,z))

One ternary operation

In the affine case (in contrast to the vector space case), it turns out that if 22 is invertible the “addition” (x,y,z)xy+z(x,y,z)\mapsto x-y+z can be recovered from the “scalar multiplication” (r,x,y)rx+(1r)y(r,x,y)\mapsto r x + (1-r)y by μ(x,y,z)=Λ 2(y,Λ 1/2(x,z))\mu(x,y,z) = \Lambda_2(y,\Lambda_{1/2}(x,z)). Thus, in this case we can define an affine space over kk to be a set AA together with a single operation Λ:k×A×AA\Lambda\colon k\times A\times A\to A such that the axioms for the two-ternary-operations definition are satisfied with this definition of μ\mu.

However, we can also simplify the requisite axioms in this presentation. The following axioms are easier to state if we write Λ r(x,y)\Lambda_r(x,y) as (1r)x+ry(1-r) x + r y, or equivalently as rx+syr x + s y, where we require r+s=1r+s=1 for the expression to be defined.

  • Idempotence: rx+sx=xr x + s x = x whenever r+s=1r+s=1.
  • Commutativity: rx+sy=sy+rxr x + s y = s y + r x whenever r+s=1r+s=1.
  • Associativity: whenever r+s+t=1r+s+t=1, whichever of the following three expressions are defined are equal:
    rx+(1r)(s1ry+t1rz) (1s)(r1sx+t1sz)+sy (1t)(r1tx+s1ty)+tz \array{ r x + (1-r)\left(\frac{s}{1-r} y + \frac{t}{1-r} z\right)\\ (1-s)\left(\frac{r}{1-s} x + \frac{t}{1-s} z\right) + s y\\ (1-t)\left(\frac{r}{1-t} x + \frac{s}{1-t} y\right) + t z }

    The first is defined whenever r1r\neq 1, the second whenever s1s\neq 1, and the third whenever t1t\neq 1. Since kk has characteristic 2\neq 2, we cannot have r=s=t=1r=s=t=1 and r+s+t=1r+s+t=1 at the same time, so at least one of these expressions is always defined. We write rx+sy+tzr x + s y + t z for the common value of whichever of them are defined.

  • Cancellation: for any rkr\in k and x,yAx,y\in A, we have x+ryry=xx + r y - r y = x.

Unbiased definition

Let Th vectTh_{vect} denote the Lawvere theory of kk-vector spaces. For any nn, its nn-ary operations are nn-tuples (r 1,,r n)k n(r_1,\dots,r_n)\in k^n representing the linear combination operation (x 1,,x n)r 1x 1++r nx n(x_1,\dots,x_n)\mapsto r_1 x_1 +\dots+ r_n x_n. Composition of operations is by substitution in the obvious way, and the identity operation is (1)(1). A model of this theory is simply a vector space. With this ‘unbiased’ definition, a vector space comes equipped with, for every integer n0n\ge 0 and nn-tuple (r 1,,r n)(r_1,\dots,r_n) of elements of kk, a function V nVV^n\to V (thought of as (v 1,,v n)r 1v 1+r nv n(v_1,\dots,v_n)\mapsto r_1 v_1+\dots r_n v_n), satisfying some axioms.

Let Th affTh_{aff} denote the subtheory of Th vectTh_{vect} containing only those operations (r 1,,r n)(r_1,\dots,r_n) such that r 1++r n=1r_1+\dots+r_n=1; an affine space is a nonempty model of Th affTh_{aff}. (We have to observe that these are closed under the theory operations and thus define a subtheory. Note that this excludes all zero-ary operations, so an affine space has no distinguished constants, and it also excludes all nonidentity unary operations.) The basic operations r 0x 0++r nx nr_0x_0+\dots+r_n x_n, when r 0++r n=1r_0+\dots+r_n=1, are called affine (linear) combinations of elements of AA.

The axioms for the unbiased definition are most straightforward to see by writing out the operations of Th affTh_{aff}. In particular, this includes “substitution” axioms of the form

r 0(s 00x 00++s 0m 0x 0m 0)++r n(s n0x n0++s nm nx nm n)=r 0s 00x 00++r ns nm nx nm n. r_0(s_{00} x_{00} + \dots + s_{0m_0} x_{0m_0}) + \dots + r_n (s_{n0} x_{n0} + \dots + s_{n m_n} x_{n m_n}) = r_0 s_{00} x_{00} + \dots + r_n s_{n m_n} x_{n m_n}.

However, it also includes “permutation” axioms of the form

r 0x 0++r nx n=r σ0x σ0++r σnx σn r_0 x_0 + \dots + r_n x_n = r_{\sigma 0} x_{\sigma 0} + \dots + r_{\sigma n} x_{\sigma n}

and also “duplication” and “omission” axioms. This Lawvere theory can be defined concisely as follows. The Lawvere theory of vector spaces is the opposite of the category of finite-dimensional vector spaces; its operations are all linear combinations. The Lawvere theory for affine spaces is the sub-theory of this consisting of only the affine combinations. (The Lawvere theory of vector spaces also has other interesting sub-theories, such as that consisting of convex combinations whose algebras are abstract convex spaces in one sense of the term.) Note that the empty set is a model (algebra) of this Lawvere theory; an affine space is an inhabited model.

Given the unbiased definition in terms of a Lawvere theory, the previous three “biased” vector-space-free definitions can then be recovered by finding particular generating operations for the theory. In particular, this Lawvere theory is generated by 22-ary operations if char(k)2char(k)\neq 2, and by 33-ary ones if char(k)=2char(k)=2. To wit, suppose given (r 0,,r n)k n+1(r_0,\dots,r_n)\in k^{n+1} with n3n\ge 3 such that r 0++r n=1r_0+\dots+r_n=1. Suppose for the moment that the r ir_i are not all 11, and WLOG suppose that r 01r_0\neq 1. (Note that here we use the invariance under permutations.) Then we have

r 0x 0++r nx n=r 0x 0+(1r 0)(11r 0r 1x 1++11r 0r nx n) r_0 x_0 + \dots + r_n x_n = r_0 x_0 + (1-r_0)\left(\frac{1}{1-r_0} r_1 x_1 + \dots + \frac{1}{1-r_0}r_n x_n\right)

so we have expressed the given (n+1)(n+1)-ary operation in terms of a 22-ary one and an nn-ary one. By induction, in this way we can express any (n+1)(n+1)-ary operation in terms of 22-ary ones (note that there is only one 11-ary operation, namely the identity, and no 00-ary ones) — as long as we never hit a tuple where every r i=1r_i=1. But since we always have the requirement r 0++r n=1r_0+\dots+r_n=1, this badness can only happen if the characteristic of kk is nn. Moreover, we still have

x 0++x n=x 0x 1+(2x 1+x 2++x n) x_0 + \dots + x_n = x_0 - x_1 + (2x_1 + x_2 + \dots + x_n)

so we can still write this (n+1)(n+1)-ary operation in terms of a 33-ary one and an nn-ary one. So only if n+1=3n+1=3 (i.e. n=char(k)=2n=char(k)=2) are we prevented from getting down to 22-ary operations only, and in this case we can still get down to 33-ary ones. Finally, we observe that any 33-ary operation can be written in terms of 22-ary ones and the particular 33-ary operation x 0x 1+x 2x_0 - x_1 + x_2:

r 0x 0+r 1x 1+r 2x 2=(r 0x 0+(1r 0)x 2)x 2+(r 1x 1+(1r 1)x 2). r_0 x_0 + r_1 x_1 + r_2 x_2 = \big(r_0 x_0 + (1-r_0) x_2\big) - x_2 + \big(r_1 x_1 + (1-r_1) x_2\big).

Slice of VectVect

Given an affine space AA (with any other definition), the corresponding π:Ak\pi:A'\to k is constructed as follows. Let A=1AA' = 1 \sqcup A, where \sqcup is the coproduct in affine spaces (akin to a simplicial join), 11 is the terminal affine space, and π\pi is the composite of 1!:1A111 \sqcup !: 1 \sqcup A \to 1 \sqcup 1 with a natural identification μ:11k\mu: 1 \sqcup 1 \cong k. Both 1!1 \sqcup ! and μ\mu which are morphisms of AffAff may be regarded as morphisms of 1AffVect1 \downarrow Aff \simeq Vect (pointed affine spaces are vector spaces) if we let the first inclusion i 0:111i_0: 1 \to 1 \sqcup 1 be the pointing of 111 \sqcup 1 and 0:1k0: 1 \to k the pointing of kk and define μ\mu by μi 0=0\mu \circ i_0 = 0, μi 1=1\mu \circ i_1 = 1 (the element 1k1 \in k). (So μ\mu is like two ends of a meter stick used to set up coordinates on the line kk.)

Conversely, given π:Ak\pi:A'\to k, the fiber π 1(1)\pi^{-1}(1) naturally acquires a “vector-valued difference” affine space structure by simple subtraction in the vector space AA', where the vector space of displacements is V=π 1(0)V = \pi^{-1}(0).

Note that this definition embeds the category AffAff of (inhabited) affine spaces fully-faithfully in the slice category Vect/kVect/k. The objects of Vect/kVect/k not in AffAff are those of the form 0:Vk0:V\to k, which form a category equivalent to VectVect itself. Moreover, there are no morphisms from objects of AffAff to objects not in AffAff; while by the above construction, a morphism from 0:Vk0:V\to k to an affine space π:Ak\pi:A'\to k is just a map from VV to the vector space of displacements of AA. Hence, Vect/kVect/k is equivalent to the (dual) cograph of D:AffVectD:Aff\to Vect.

Closed monoidal structure

If we allow affine spaces to be empty, then they are the models of an algebraic theory Th AffTh_{Aff}. Moreover, like Th VectTh_{Vect}, the theory Th AffTh_{Aff} is a commutative theory. It follows that if A,BA, B are affine spaces, then the set hom(A,B)\hom(A, B) is closed under all affine space operations pointwise defined on the set of all functions from AA to BB. This gives AffAff a closed category structure; on general grounds, it is in fact a symmetric monoidal closed category. The unit of this structure is the terminal or one-pointed affine space 11, via the natural isomorphism hom(1,B)B\hom(1, B) \cong B. Thus AffAff is a closed semicartesian monoidal category.

Analogous to the case of VectVect, every affine space is a coproduct of copies of the monoidal unit: an affine space AA of dimension nn admits an affine basis, which amounts to an isomorphism 111A1 \sqcup 1 \sqcup \ldots \sqcup 1 \cong A, represented by n+1n+1 [sic] points of AA. Such basis representations allow one to coordinatize spaces of maps hom(A,B)B n+1\hom(A, B) \cong B^{n+1}, with dimension (n+1)dim(B)(n+1)\dim(B). If one uses the first of the affine basis elements to give a pointing of the affine space (equivalent to a vector space structure), then the remaining affine basis elements provide a vector space basis, and in those coordinates every element fhom(A,B)f \in \hom(A, B) may be written in matrix-vector form f(x)=Mx+bf(x) = M x + b, where again the space of such (M,b)(M, b) has dimension mn+mm n + m if mm is the dimension of BB. (There are also more ‘unbiased’ coordinate descriptions, not biased in favor of the first basis element playing the role of the origin.)

Similarly, we can coordinatize affine tensor products ABA \otimes B: the tensor distributes over coproducts (as it does in any symmetric monoidal closed category) and so

( n+11)( m+11) (m+1)(n+1)11 (m+1)(n+1)1(\bigsqcup_{n+1} 1) \otimes (\bigsqcup_{m+1} 1) \cong \bigsqcup_{(m+1)(n+1)} 1 \otimes 1 \cong \bigsqcup_{(m+1)(n+1)} 1

with dimension mn+m+nm n + m + n. In other words, if a 1,,a n+1a_1, \ldots, a_{n+1} is an affine basis of nn-dimensional AA and b 1,,b m+1b_1, \ldots, b_{m+1} a basis of mm-dimensional BB, then ABA \otimes B has an affine basis consisting of the mn+m+n+1m n + m + n + 1 many elements a ib ja_i \otimes b_j.

The embedding AffVect/kAff \to Vect/k described in the previous subsection, sending AA to 1!:1A111 \sqcup !: 1 \sqcup A \to 1 \sqcup 1, is a strong monoidal functor (preserves the tensor product up to coherent isomorphism) if Vect/kVect/k is endowed with the obvious tensor product acquired from VectVect. Note that Vect/kVect/k is the coreflection of VectVect from monoidal categories to semicartesian monoidal categories; the embedding AffVect/kAff \to Vect/k was in fact discovered by one of us in conjunction with this fact, and is the same as the functor induced by universality from the strong monoidal functor AffVectAff \to Vect given by A1AA \mapsto 1 \sqcup A.

Further Remarks

Every finitely-generated affine space is isomorphic to the nn-fold direct sum k nk^n, where kk is the base field and nn is a natural number (possibly 00). In algebraic geometry, an nn-dimensional affine space is often denoted 𝔸 n\mathbb{A}^n and identified with k nk^n. If one accepts the empty set as an affine space, then this is considered to have dimension 1-1 by convention (so k 1=k^{-1} = \empty).

The notion of affine space may be generalised to affine module by replacing the vector space above by a module and the base field kk by a commutative ring. Then an affine module over the ring \mathbb{Z} of integers is precisely a commutative heap, just like a module over \mathbb{Z} is an abelian group. Note that the definition involving only one “scalar multiplication” operation works if and only if 22 is invertible in kk; it's not enough that 202 \ne 0 in kk.

Mike Shulman: I haven’t thought much about affine modules, but it seems likely to me that the “biased” module-free definitions won’t be right any more, since the Lawvere theory needn’t be generated by 2-ary or 3-ary operations (as far as I can see). More explicitly, I don’t immediately see how to write an operation like

(x 0,x 1,x 2,x 3)4x 06x 12x 2+5x 3(x_0,x_1,x_2,x_3) \mapsto 4x_0 - 6x_1 - 2x_2 + 5x_3

in terms of A 3AA^3\to A and ×A 2A\mathbb{Z}\times A^2\to A, but it seems to me that this operation should still exist in an affine \mathbb{Z}-module.

Mike Shulman: Well that’s rubbish isn’t it. The operation A 3AA^3\to A is enough to give you a heap, hence an additive group, and then ×A 2A\mathbb{Z}\times A^2\to A gives you the scalar multiplication. And so

4x 06x 12x 2+5x 3=((4x 03y)y+(6x 1+7y))y+((2x 2+3y)y+(5x 34y))4x_0 - 6x_1 - 2x_2 + 5x_3 = \big((4x_0 - 3y) - y + (-6x_1+7y)\big) - y + \big((-2x_2+3y) - y + (5x_3 - 4y)\big)

for any yy at all.

Toby: Right. But I find an affine module of a rig to be a trickier concept.

Mike Shulman: Quite so. Perhaps first one should look for a version of a heap corresponding to a monoid?

Toby: Yes, that would be an affine \mathbb{N}-module.

Affine spaces as model spaces

Affine spaces typically serve as local models for more general kinds of spaces.

For instance a manifold is a topological space that is locally isomorphic to an affine space over the real numbers.

Similarly, in algebraic geometry a scheme is locally isomorphic to an affine scheme.

Therefore there are attempts to axiomatize properties of categories of affine spaces for the purpose of using these as model spaces for more complicated geometries. One such axiomatization is the notion of geometry (for structured (∞,1)-toposes). and in particular that of pregeometry.


Textbook accounts:

See also:

  • Aurelio Carboni, Categories of Affine Spaces , JPAA 61 (1989) pp.243-250.

  • Aurelio Carboni, George Janelidze, Modularity and Descent , JPAA 99 (1995) pp.255-265.

  • Michel Thiébaud, Modular Categories , pp.386-400 in Proc. Como conference - Category Theory , LNM 1488 Springer Heidelberg 1991.

Last revised on September 5, 2023 at 15:04:31. See the history of this page for a list of all contributions to it.