nLab directed topological space




topology (point-set topology, point-free topology)

see also differential topology, algebraic topology, functional analysis and topological homotopy theory


Basic concepts

Universal constructions

Extra stuff, structure, properties


Basic statements


Analysis Theorems

topological homotopy theory



A directed topological space is a topological space XX in which there is some ‘sense of direction’. This can happen in various different ways and the level of the ‘directedness’ can be different in different situations, so naturally there are several ‘competing’ ideas, but the beginning of a consensus on what the overarching idea is.

If one bases homotopy theory on the idea of a singular simplex or more generally a singular cell of any shape, then there is no way in which a ‘sense of direction’ can be encode. If we have a path in a space we can go along it (traverse it) in either direction, from 0 to 1 or from 1 to 0. From this perspective a directed space is one in which not every singular cell Δ nX\Delta^n \to X (for Δ n\Delta^n the standard topological simplex) is supposed to be traversable in all directions, in some sense: instead these kk-dimensional paths may have a direction .

As an example one can base the ‘sense of direction’ on a closed preorder or partial order, (that is a pospace),so that the paths from the directed interval [0,1][0,1] with the usual order to the space XX, can only be ‘traversed’ in one direction. Another example which does not fit into this first type would be the directed circle.


In other words, a circle with direction determined by the anticlockwise sense. Again it is easy to see that there are certain paths that respect the direction, ‘directed paths’ whilst others do not.

So far there exists a well-developed theory for a notion of directed spaces XX where 1-dimensional paths given by maps [0,1]X[0,1] \to X from the interval into the space are equipped with a direction. See in particular the book by Marco Grandis on Directed Algebraic Topology listed below. Another suggested notion for modelling directed spaces is that of framed spaces, which is tailored towards certain higher categorical applications.

Note that a directed space is like a generalised space; not every directed space need be a space in the traditional sense, in accordance with the red herring principle. As an instance of this, note that Marco Grandis in his book Directed Algebraic Topology handles the directed homotopy of small categories, and of cubical complexes, since this is useful for comparison an interpretation of directed homotopy ‘invariants’.

Directed spaces are studied in directed homotopy theory, a relatively young topic. In generalization of how a topological space has a fundamental groupoid, a directed space has a fundamental category.

Homotopy-theoretic perspective

From a homotopy theoretic perspective one would wish that notions of directed spaces might serve to generalize the homotopy hypothesis – which identifies ordinary (undirected) topological spaces with ∞-groupoids, i.e., with (∞,0)-categories – to a more general context where (∞,0)-categories are generalized to (∞,r)-categories with r>0r \gt 0:

An (∞,r)-category in this context might correspond to a rr-directed topological space , one that comes equipped with a notion of orientation of its kk-cells for 0kr0 \leq k \leq r, but was impartial on direction above that dimension.

If such a definition exists, it may need to use filtered topological spaces instead of bare topological spaces.

Even in the absence of a homotopy-theoretic definition of rr-directed space in this sense, from the perspective of homotopy theory one might take the standpoint of the homotopy hypothesis and define a (nice) rr-directed space to be an (∞,r)-category, just as it makes good sense and is nowadays common practice in algebraic topology to define a nice topological space to be an ∞-groupoid.

See (n,r)-category for more on that.

Urs Schreiber: I haven’t looked at Marco Grandis’ book yet: does it say anything about the homotopy hypothesis in the context of the definition of directed space used there?

Tim Porter: No.



A directed topological space or d-space is pair (X,dX)(X, d X) consisting of a topological space XX and a subset dXC(I,X)d X \subset C(I,X) of continuous maps from the interval I=[0,1]I = [0,1] into XX – called directed paths or d-paths – satisfying the following conditions:

  1. (constant paths) every constant map IXI\to X is directed,

  2. (reparametrisation) dXdX is closed under composition with increasing maps III\to I,

  3. (concatenation) dXdX is closed under path-concatenation: if the d-paths a,ba, b are consecutive in XX (a(1)=b(0))(a(1) = b(0)), then their ordinary concatenation a+ba+b is also a d-path

(a+b)(t)=a(2t),if0t12,(a+b)(t) = a(2t),\,\text{if}\, 0\le t\le \frac{1}{2},
(a+b)(t)=b(2t1),if12t1.(a+b)(t) = b(2t-1),\,\text{if}\, \frac{1}{2}\le t\le 1.

A morphism of directed topological spaces f:(X,dX)(Y,dY)f : (X, d X)\to (Y , d Y) is a morphism of topological spaces f:XYf: X \to Y which preserves directed paths in that for every γ:IX\gamma: I \to X in dXd X the path f *γ:IγXfYf_* \gamma : I \stackrel{\gamma}{\to} X \stackrel{f}{\to} Y is in dYd Y.


Basic examples of dd-spaces include:

  • The standard directed interval is I d=([0,1],dI)I_d = ([0,1], d I) with dId I the set of all monotonic continuous maps [0,1][0,1][0,1] \to [0,1] is a d-space

  • Any pospace XX gives rise to a d-space by taking the directed paths to be, well, directed paths, i.e. continuous order-preserving maps from I dI_d to XX.

Many other example can be found in the references.


A different definition comes from Sanjeevi Krishnan, A Convenient Category of Locally Preordered Spaces, Applied Categorical Structures, 2009, vol. 17, no 5, p. 445-466 (arxiv):

Definition A stream is a tuple X, X, \leq_{-}, where \leq_{-} assigns to each open subset UXU \subset X a preorder U\leq_U, such that:

iU i= i U i\leq_{\bigcup_i U_i} = \bigvee_i \leq_{U_i}

Here, U i,iIU_i, i \in I is a collection of open sets, and \bigvee is pointwise or of relations.


  • Morally, a point in a stream is less or equal to another point if it is less or equal in any open set. The relation X\leq_X for the whole space does not hold much information.
  • In the typical example of the clockwise oriented circle, each point is less or equal to every other one in the relation S 1\leq_{S^1}; but for a contractible subset US 1U \subset S^1, a point xx is less or equal than yy if yy can be reached from xx in a monotonous clockwise path.
  • The defining equation of a stream can be thought of as a (co)sheaf condition, and there is indeed a cosheafification result.
  • Every d-space gives rise to a stream.
  • The category of streams has good properties. In particular, there is an further notion of compactly flowing streams_ extending the notion of compactly generated Hausdorff spaces, and indeed the forgetful functor creates limits and colimits.
  • The category of compactly flowing streams is Cartesian closed.

Framed spaces


Another way to endow spaces with directions is via framings, i.e. choices of a “basis of the vector space of tangential directions at each point”. Somewhat abstracting this idea, frames may also be thought about in terms of projections, as the following remark motivates.


Let VV be an nn-dimensional vector space with an inner product gg. The following structures on VV are equivalent.

  • An orthonormal frame of VV, i.e. an ordered sequence of vectors v iv_i, 1in1 \leq i \leq n, such that g(v i,v j)=δ ijg(v_i,v_j)=\delta_{ij}.
  • A sequence of surjections V iV i1V_i \to V_{i-1}, 1in1 \leq i \leq n, where V iV_i is an oriented ii-dimensional vector space (and V n=VV_n = V).

The two structures are related by setting v iv_i to be the unit vector spanning ker(V iV i1)\ker(V_i \to V_{i-1}) such that V i1v iV_{i-1} \oplus v_i recovers the orientation of V iV_i (note that all V iV_i canonically embed in VV as the orthogonal complement of the kernel of the composite map VV iV \to V_i).

In the absence of inner products, one cannot speak of orthonormal frames any longer. However, sequences of projections can still be defined, and may be regarded as playing the role of “metric-free orthonormal” frames. (A vaguely analogous line of thinking is that a Morse function MM \to \mathbb{R} provides useful “direction” information on MM, e.g. for the construction of handlebodies, that is ultimately independent from any chosen metric on MM.) Moreover, this approach offers room for generalization by varying the length of the projection sequence and the vector space dimensions.


The standard orthonormal frame of nn-dimensional euclidean space n\mathbb{R}^n consists for the ordered sequence of vectors e 1=(1,0,...,0)e_1 = (1,0,...,0), e 2=(0,1,0,...,0)e_2 = (0,1,0,...,0), …, e n=(0,...,0,1)e_n = (0,...,0,1). By the previous remark, this orthonormal frame is equivalently described by the sequence of projections π i: i= i1× i1\pi_i : \mathbb{R}^i = \mathbb{R}^{i-1} \times \mathbb{R} \to \mathbb{R}^{i-1} (each i\mathbb{R}^i being endowed with standard orientation).

When forgetting basepoints, then the previous remark and example equally apply to affine spaces, now endowing each point in the space with a basis of frames. Using affine standard framed n\mathbb{R}^n (or rather compact contractible patches of it, that interact nicely with the projections, see below) as our “local models” for framed spaces one may define global framed spaces and their maps.



Inductively in nn \in \mathbb{N}, an nn-framed patch U nU \subset \mathbb{R}^n is a non-empty subspace of n\mathbb{R}^n with the property that its projection π n(U)\pi_n(U) is an (n1)(n-1)-framed patch, such that π n:Uπ n(U)\pi_n : U \to \pi_n(U) has fibers π n 1(x)\pi^{-1}_n(x) of the form [γ (u),γ +(u)][\gamma_-(u),\gamma_+(u)] for two continuous sections γ ±:π n(U)π n(U)×\gamma_\pm : \pi_n(U) \to \pi_n(U) \times \mathbb{R}. Given two nn-framed patches UU and VV, a (partial) nn-framed patch map F:UVF : U \to V is a (partial) continuous map which descends along π n\pi_n to a (partial) (n1)(n-1)-framed patch map F n1:π n(U)π n(V)F_{n-1} : \pi_n (U) \to \pi_n(V) such that mappings of fibers F:π n 1(x)π n 1(F n1(x))F : \pi^{-1}_n(x) \to \pi^{-1}_n(F_{n-1}(x)) are monotone.

Note that nn-framed patches are compact and contractible spaces.


The standard example of an nn-framed patch is the closed nn-cube I n=[1,1] n n\mathbf{I}^n = [-1,1]^n \subset \mathbb{R}^n. However, in general nn-framed patches need not be “nn-dimensional”: for instance, the 00th slice {0}×I n1\{0\} \times \mathbf{I}^{n-1} of the nn-cube is itself an nn-framed patch, and so are the “kk-directed intervals” I k:={0} k1×I×{0} nk\mathbf{I}_k := \{0\}^{k-1} \times \mathbf{I} \times \{0\}^{n-k}.


Let XX be a topological space. Fix nn \in \mathbb{N}.

  • An nn-framed chart (U,γ)(U,\gamma) in XX is an embedding γ:U n\gamma : U \hookrightarrow \mathbb{R}^n of a subspace UXU \subset X whose image im(γ)\im(\gamma) is an nn-framed patch.

  • Two nn-framed charts (U,γ)(U,\gamma), (V,ρ)(V,\rho) in XX are compatible if ργ 1\rho \circ \gamma^{-1} is a partial map of nn-framed patches.

An nn-framed space is a space XX endowed with an nn-framing structure 𝒜\mathcal{A}, which is an “atlas” of compatible nn-framed charts {(U i,γ i)}\{(U_i,\gamma_i)\} such that U iU_i are a locally finite cover of XX.


The condition for covers to be locally finite is convenient as it describes the situation of locally finite cell complexes. However, the condition could be replaced, and the definition generalized, in several ways. The above version of the definition should be considered a first approximation to a potentially more general notion of framed spaces.

Maps of framed spaces could be defined along the following lines.


Given spaces with nn-framing structure (X,𝒜)(Y,)(X,\mathcal{A}) \to (Y,\mathcal{B}) then a framed map F:XYF : X \to Y is a map such that for any charts (U,γ)𝒜(U,\gamma) \in \mathcal{A} and (V,ρ)(V,\rho) \in \mathcal{B}, FF yields a partial map ρFγ 1:im(γ)im(ρ)\rho \circ F \circ \gamma^{-1} : \mathrm{im}(\gamma) \to \mathrm{im}(\rho) of nn-framed patches.

Framed spaces and framed maps are, in a sense, “very rigid” variants of directed spaces. Nonetheless, they are interesting to study as they turn out to have a rich combinatorial theory associated to them. This combinatorial counterpart is particular useful when translating between the topology of directed spaces and the combinatorics of higher categories: an example of this is the definition of manifold diagrams in the language of framed spaces.


Realizations of geometric computads (resp. their functors) have the structure of framed spaces (resp. of framed maps), see Dorn23.

Comparison to dd-spaces


One may want to define directed paths in an nn-framed space as maps from “the framed interval” into that space: but, in fact, there are now nn different framed intervals, corresponding to the nn possible directions of nn-framings. These are precisely modelled by the nn-patches I k\mathbf{I}_k mentioned in a previous example. Note, there is a framed bijection I kI j\mathbf{I}_k \to \mathbf{I}_j if and only if kjk \leq j (in particular, directions aren’t interchangeable at all: analogously, note that in an n-category a (nk)(n-k)-morphisms can be (possibly degenerate) (nj)(n-j)-morphisms only if kjk \leq j). Given a framed space (X,𝒜)(X,\mathcal{A}), and defining d kX=Map fr(I k,(X,𝒜))d_k X = \mathrm{Map}_{\mathrm{fr}}(\mathbf{I}_k, (X,\mathcal{A})), we thus obtain a filtration of directed paths:

d nXd n1X...d 1XMap(I,X) d_n X \subset d_{n-1} X \subset ... \subset d_1 X \subset \Map(\mathbf{I},X)

instead of a single path subset dXMap(I,X)dX \subset \Map(\mathbf{I},X), as required in the definition of dd-spaces.


  • If we can equip directed spaces with an internal hom, then a directed space with at least one directed path should be a strictly directed object in the category of directed spaces, with respect to the standard directed interval as the interval object, while an ordinary topological space regarded as a directed space should be an undirected object.

But for that to work we need the structure of a directed topological space on C(I d,(X,dX))C(I_d,(X,d X)). This requires that XX has directed homotopies! Does Grandis discuss higher directed paths, too? —Urs

Toby: I don't think that you need internal homs and all that. But see my edits to directed object.

Urs: I think we need directed homotopies to check if a “constructed” directed space is actually a directed object in the original definition: that original definition asks us to check if the internal hom [I,X][I,X] is weakly equivalent to XX. Well, I made up this definition because I think it is the right abstraction, but there is room of course to debate this. But if we accept it then we should try to define the internal hom of Grandis’ directed spaces. There is an obvious solution which one should check the details of: namely a directed topological space should be one which singles out not only subsets of hom(I,X)hom(I,X) but subsets of hom(I ×n,X)hom(I^{\times n}, X) for all nn, closed under the obvious reparameterization and gluing. This would induce an obvious notion of directed homotopies and should induce in an obvious way an internal hom for directed topological spaces. I’d think. But I don’t feel like investing much time into finalizing this idea right now…

Tim Porter: As I have now looked at Marco’s book, there are results on exponentiable d-spaces.(p.59). I can give details if anyone is still interested.


The above definition of dd-spaces is from

This has now developed into a book

A discussion of reparameterization of directed paths in directed topological spaces is in

  • Ulrich Fahrenberg and Martin Raussen, Reparametrizations of Continuous Paths (arXiv, (blog))

Further references are given in directed homotopy theory.

Some ‘local combinatorial’ aspects of framed spaces are discussed in:

  • Christoph Dorn and Christopher Douglas, Framed combinatorial topology, 2021 (arXiv)

A global (higher-categorical) perspective on directed spaces is taken in:

  • Christoph Dorn, Nine short stories about geometric higher categories, 2023 (pdf)


  • See also the discussion at the n-Forum.

The above defined directed topological spaces. My impression is that Eric was interested in more general concepts. But the above definition has a straightforward generalization away from topological spaces. The general strategy is really: start with a category with interval object and consider then the category whose objects are pairs (X,dX)(X, d X) for XX an object and dXd X a subobject of [I,X][I,X], and whose morphisms are morphisms XYX \to Y that take dXd X to dYd Y.

For instance, let’s define directed sets: make the ordinary category Set a category with interval object by , say, taking the interval object to be the set I:=[n]I := [n] of nn elements. A map from II into any other set can be regarded as an nn-step path in that set. Then pairs consisting of a set and a subset of all such maps model “directed sets”.

Eric says: Yes, exactly :) That sounds like a good plan. By the way, what you say about I:=[n]I := [n] reminds me a lot of simplicial sets.

Eric says: We have directed spaces and we may soon have directed sets. This makes me wonder if we should have a directed category internal to another category? This way

  • a directed space is a directed category in Top
  • a directed set is a directed category in Set
  • etc

Would that make sense?

Urs: Let’s see, before getting into this idea of realizing a directed space as a space internal to something else or the like,

I don’t see what you want to mean by a “directed category”. See, the point is that a category already is supposed to be a combinatorial model for a directed space. Just as a groupoid is a combinatorial model for an undirected space. This is the very motivation for defining directed spaces: to fill in the question marks in

  • groupoid | space || category | ?? .

This is why a directed space is defined such that its “thing of all paths in it” is not, in general, a fundamental groupoid but a fundamental category.

Methinks that for the application which you have in mind you want to be studying posets and these are special cases of categories and in particular naturally interpreted as combinatorial models for directed space, in exactly the way in which you are thinking of them as directed spaces! So it seems to me you don’t actually need to be looking for what you seem to be looking for, since it is already quite easily there. But of course maybe I misunderstand what you are after.

Eric: I doubt that what I am looking for is new. If you could help put a name on it, that would be great. I’m not exactly sure what I mean by directed category either other than a “category with a direction” :|

Urs: but a category is directed! Recall that underlying every category is a directed graph (it is a directed graph equipped with a composition operation). So I am still puzzled by what you are looking for, because a “directed category” would have underlying it a “directed directed graph”. What’s that supposed to be? And why do you want it?

Eric: Sorry for being so dense. We can delete this once I get a clue. But for now, I’m still confused. Maybe what I wanted to say is more along the line (but probably still not correct)

“A directed space has a fundamental category”

“A directed set has a fundamental category”

“A directed object has a fundamental category”

Ack! light bulb! (those hurt sometimes)

I think that is probably precisely why you defined directed object.

Could we say (and be correct!) that

“a directed space is a directed object in Top”?

“a directed set is a directed object in Set”?

If so, I think I am making some progress.

Urs: Yes, a directed space should be a directed object in the category of possibly directed topological spaces! (In Top itself there are no directed spaces. Every ordinary topological space is undirected). I think I listed that as a should-be example. To make it a proper example one will have to say a few more probabaly straightforward things about directed homotopies etc. But the idea is certainly this, yes, a directed space is a directed object in the category of possibly directed spaces.

And as for categories: the generic category is a directed object in the category of categories. Unless it happens to be a groupoid. In which case it is an undirected object there.

(All this with respect to the “canonical” choice of interval object. The notion of directedness depends on which interval object you choose to test with. For instance the point itself satisfies the axioms of an interval object. But using it of course everything will look undirected.)

Eric: Ugh. I didn’t want a directed space to be a directed object in the category of directed spaces. That is boring :) A set is an object in the Set too, but it doesn’t tell you anything. Hmm. It looks like what I wanted isn’t going to work as is, i.e. a directed space is not a directed object in Top because there are no directed objects in Top apparently.

Urs: I think you do want that. Just don’t let the terminology let mix you up. An ordinary space is already called a space. While from your perspective an ordinary space ought to be called an undirected space. Then “space” could be assigned to mean “not-necessarily but possibly directed space” and then a directed space could be called a directed object in spaces.

But convention is different. So a directed space is a directed object in the category of “not necessarily but possibly drected spaces”.

Toby: Even here, I don't think that you're really using the terminology ideally. The proper term for what you're calling a “not-necessarily but possibly directed space” is just directed space! Much like a non-associative algebra might happen to be associative, so a directed space might happen to be undirected. (In terms of Grandis's definition, any space XX defines a directed space where dd consists of only the constant paths.)

Urs: Right, Toby, I think that is my point. I was just trying to convince Eric that there is nothing wrong or cheating or boring about the fact that “a directed space is a directed object in the category of directed spaces”.

But maybe the the true issue is whether we want to speak of “directed objects” over at directed object or rather restrict to speaking about undirected objects. Then every object would be a directed object, possibly with trivial direction information, while those objects which are propertly directed would be the not undirected objects.

I consider you as an authority on such issues of logical rigour. You should say how we should fix the terminology and we’ll implement that.

Toby: I'll discuss this at directed object.

Last revised on March 13, 2023 at 10:23:21. See the history of this page for a list of all contributions to it.