proper homotopy theory


Homotopy theory

homotopy theory, (∞,1)-category theory, homotopy type theory

flavors: stable, equivariant, rational, p-adic, proper, geometric, cohesive, directed

models: topological, simplicial, localic, …

see also algebraic topology



Paths and cylinders

Homotopy groups

Basic facts



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


Idea of some problems

Proper homotopy theory is both an old and a fairly new area of algebraic topology. It deals with properties of non-compact topological spaces, that cannot be detected using maps from simplices in these spaces, such as used in the singular complex.

Historically the subject is traced back to Kerekjarto’s classification of non-compact surfaces in 1923, but its emergence as an important tool in geometric topology came with Larry Siebenmann’s work in 1965.

Suppose MM is a smooth manifold with boundary, M\partial M, then MMM\setminus \partial M is an open manifold. Now suppose someone gives us an open manifold NN, is it possible to detect if there is a compact manifold MM, with MMNM\setminus \partial M \cong N. Siebenmann showed that certain conditions on the ends of NN were necessary and that there were obstructions if the dimension of MM was greater than 5.

Its potential importance of proper homotopy, for example for physical applications, comes from the fact that the phenomena it studies include the limiting behaviour of the system.


The basic hypothesis will be that XX will be a connected and locally connected compact Hausdorff space. It will usually be sigma-compact, i.e., there will be an increasing sequence, {K n}\{K_n\}, of compact subspaces with each K nK_n in the interior of K n+1K_{n+1} and such that

X= n=0 K n,X = \bigcup^\infty_{n=0}K_n,

These spaces will most often than not be locally finite simplicial complexes.

We will be interested in the homotopy of such spaces ‘out towards its ends’


To illustrate the idea of the ends of a space XX, we note that naively \mathbb{R} has two ends, \infty and -\infty, whilst 2\mathbb{R}^2 has only one as it is S 2{}S^2\setminus \{\infty\}, (but that is vague!).

More exactly, consider the system of spaces

ε(X)={closure(XK):KcompactX},\varepsilon(X) = \{closure(X\setminus K) : K compact \subset X\},

This is an inverse system or pro-object in the category of spaces. Applying the connected component functor, π 0\pi_0, to this system of spaces gives π 0(X)\pi_0(X), and, classically, one takes the limit of this to get

e(X)=limπ 0ε(X),e(X) = lim \pi_0\varepsilon(X),

the set of ends of XX. In general, e(X)e(X) would be given the inverse limit topology?, to preserve more of the information coming from its construction. This space is the space of (Freudenthal) ends of XX. It is a profinite space.


  • Let X 8X_8 be the figure eight space, the one-point union of two circles, and let XX be its universal cover. This is an infinite ‘thorn bush’. It has infinitely many ends and

    e(X)2 0.e(X) \cong 2^{\aleph_0}.
  • Let MM be a compact manifold and X=MMX = M\setminus \partial M, then e(X)π 0(M)e(X) \cong \pi_0(\partial M).

Proper maps

The assignment sending XX to e(X)e(X) cannot be functorial on the category of spaces and continuous maps, since the contracting map {0}\mathbb{R}\to \{0\} is continuous, e()e(\mathbb{R}) is {,}\{-\infty, \infty\}, whilst e({0})e(\{0\}) is empty since {0}\{0\} is compact. The problem is that continuity is really about inverse images (inverse image of open is open), but the inverse image mapping does not preserve compactness (as in the example!).

Definition (recall)

A map f:XYf:X\to Y is a proper map if for each subset KK compact in YY then f 1(K)f^{-1}(K) is compact in XX.

End spaces continued

If f:XYf:X\to Y is proper, then it induces a pro-morphism

ε(f):ε(X)ε(Y)\varepsilon(f) : \varepsilon(X)\to \varepsilon(Y)

and hence a continuous map of the end spaces

e(f):e(X)e(Y),e(f):e(X)\to e(Y),

and ee becomes a functor from some category ProperProper of spaces and proper maps to StoneStone, the category of profinite spaces / Stone spaces.

Proper homotopy


For any space XX, the natural inclusions of XX into X×IX \times I, e i(x)=(x,i)e_i(x) = (x,i), i=0,1i = 0,1 together with the projection maps from X×IX\times I to XX, are proper maps.


The category ProperProper has a cylinder functor.

We call the corresponding notion of homotopy, ‘proper homotopy’. We get a ‘Proper category’ and an associated ‘proper homotopy category’, which we will denote Ho(Proper)Ho(Proper). If we are restricting to σ\sigma-compact spaces we may write Proper σProper_\sigma, and so on.

Germs at \infty

Although a proper map f:XYf: X\to Y will induce a continuous ε(f)\varepsilon(f), on the end spaces, it is clear that ff does not need to be defined on the whole of XX for this to work, as ε\varepsilon encodes behaviour ‘out towards \infty’. This leads to the notion of a ‘germ at \infty’.

Suppose XX is locally compact Hausdorff and AXA\subset X. The inclusion j:AXj: A\to X is ‘cofinal’ if the closure of XAX\setminus A is compact. Note that a cofinal inclusion is proper and induces an isomorphism ε(A)ε(X)\varepsilon(A)\cong \varepsilon(X). Let Σ\Sigma be the class of all cofinal inclusions in ProperProper and let Proper =Proper[Σ 1]Proper_\infty = Proper[\Sigma^{-1}], the category obtained by formally inverting the cofinal inclusions.


This category is called the proper category at \infty.

Note that (Proper,Σ)(Proper,\Sigma) admits a calculus of right fractions, so any morphism from XX to YY in Proper Proper_\infty can be represented by a diagram

XjAfY,X\stackrel{j}{\leftarrow} A \stackrel{f}{\rightarrow}Y,

with jj a cofinal inclusion, i.e., ff is defined on some ‘neighbourhood of the end of XX’.

Two such diagrams

Xj A f YandXj A f YX\stackrel{j^{\prime}}{\leftarrow} A^{\prime} \stackrel{f^\prime}{\rightarrow}Y and X\stackrel{j^{\prime\prime}}{\leftarrow} A^{\prime\prime} \stackrel{f^{\prime\prime}}{\rightarrow}Y

represent the same germ if f |A=f |Af^\prime | A = f^{\prime\prime}|A for some cofinal subspace AA with A A AA^\prime \cup A^{\prime\prime}\subset A.

There is also a homotopy category Ho(Proper )Ho(Proper_\infty)

The end space e(X)e(X) is a Stone space so is Max(R)Max(R) for some Boolean algebra RR (Stone duality) In the 1960s someone (Goldman?) looked at a ring, RR, of ‘almost continuous functions’ from XX to /2\mathbb{Z}/2\mathbb{Z}, that gave the right e(X)e(X). Can this idea help integrate better the ideas of proper homotopy etc. with modern methods of algebraic geometry?

Proper analogues of the fundamental group

The end space behaves a bit like a π 0\pi_0 and usually spaces will have many ends, so are not ‘connected at infinity’. If we try to do a fundamental group or groupoid analogue, this means life will get more complicated. We will try with the assumption of a space having a single end for simplicity! We will also assume XX is σ\sigma-compact.

First attempt

We could try defining π 1(ε(X))\pi_1(\varepsilon(X)) as a progroup, then taking its limit. For this we would take {K n}\{K_n\} an exhausting increasing sequence of compact subsets and setting U i=XK iU_i = X\setminus K_i, pick a base point x ix_i in each U iU_i, and we will get groups π 1(U i,x i)\pi_1(U_i,x_i). We however need induced homomorphisms π 1(U i+1,x i+1)π 1(U i,x i)\pi_1(U_{i+1},x_{i+1})\to \pi_1(U_i,x_i), and for this we have to choose an arc in U iU_i from x i+1x_{i+1} to x ix_i. We can combine these to get a base ray, rather than a base point, that is, we need a proper map, α:[0,)X\alpha : [0,\infty)\to X. With that we do get an inverse sequence of groups, but there are problems. What is the dependence of the inverse system on the choice of α\alpha?

<inkscape:perspective id='perspective2414' inkscape:persp3d-origin='372.04724 : 350.78739 : 1' inkscape:vp_x='0 : 526.18109 : 1' inkscape:vp_y='0 : 1000 : 0' inkscape:vp_z='744.09448 : 526.18109 : 1' sodipodi:type='inkscape:persp3d'></inkscape:perspective> <sodipodi:namedview bordercolor='#666666' borderopacity='1.0' gridtolerance='10000' guidetolerance='10' id='base' inkscape:current-layer='layer1' inkscape:cx='762.19162' inkscape:cy='154.01741' inkscape:document-units='px' inkscape:pageopacity='0.0' inkscape:pageshadow='2' inkscape:window-height='701' inkscape:window-width='1152' inkscape:window-x='0' inkscape:window-y='22' inkscape:zoom='1.55625' objecttolerance='10' pagecolor='#ffffff' showgrid='false'></sodipodi:namedview> <rdf:RDF> <cc:Work rdf:about=''> <dc:format>image/svg+xml</dc:format> <dc:type rdf:resource=''></dc:type> </cc:Work> </rdf:RDF> base ray

Let XX be an infinite cylinder with an infinite string of circles attached via a proper ray α:[0,)X\alpha: [0,\infty) \to X. The space has just one ‘end’ but you can choose different ways of going from π 1(U i+1,x i+1)\pi_1(U_{i+1},x_{i+1}) to π 1(U i,x i)\pi_1(U_i,x_i) for fairly obvious choices of base points such that the limit groups of the resulting two inverse systems are non-isomorphic! (In the survey listed below, this example is examined in detail, and one of the limits is a free group on one element, the other is trivial! Definitely non-isomorphic!)

This means that limπ 1(ε(X))lim \pi_1(\varepsilon(X)) is not an invariant of the end. This phenomenon is linked to the fact that π 1(ε(X))\pi_1(\varepsilon(X)) does not satisfy the Mittag-Leffler condition for either choice of the base rays.

Waldhausen boundary

If XX and YY are locally compact Hausdorff spaces, there is no obvious candidate for a space of proper maps from XX to YY, but one can form a simplicial set (X,Y)\mathbb{P}(X,Y) with (X,Y) n=Proper(X×Δ n,Y)\mathbb{P}(X,Y)_n = Proper(X\times \Delta^n,Y), which acts as if it was the singular complex of the mythical space of proper maps from XX to YY.


The Waldhausen boundary of XX is the simplicial set ([0,),X)\mathbb{P}([0,\infty),X).

There is an epimorphism from π 0(([0,),X))\pi_0(\mathbb{P}([0,\infty),X)) to e(X)e(X).

  • In the example above of the cylinder with the string of circles attached, π 0(([0,),X))\pi_0(\mathbb{P}([0,\infty),X)), is uncountable, and π 1(([0,),X))\pi_1(\mathbb{P}([0,\infty),X)) maps onto limSlim S.

  • When XX has a single end and π 0(ε(X))\pi_0(\varepsilon(X)) is Mittag-Leffler, then π 0(([0,),X))\pi_0(\mathbb{P}([0,\infty),X)), is a single point, i.e. all possible base rays are properly homotopic.

Second attempt

Even if we did not have the above difficulty with the limit groups, we would still have the problem that, as the limit functor is not exact, the resulting limiting homotopy groups would not be that well behaved. There would not be any general long exact sequence results (just as with Čech homology). There is at least one possible replacement for those limiting homotopy groups, but first we note that it is not appropriate to base any such things at a point, rather we should be using a base ray as was discussed above.

One fairly obvious attempt to define a ‘fundamental group’ for XX, based at a proper ray α:[0,)X\alpha: [0,\infty) \to X, would be to note that α\alpha gives ([0,),X)\mathbb{P}([0,\infty),X) a base point so we could look at π 1(([0,),X),α)\pi_1(\mathbb{P}([0,\infty),X),\alpha) and more generally at π n(([0,),X),α)\pi_n(\mathbb{P}([0,\infty),X),\alpha), and we will denote these groups by π̲̲ n(X,α)\underline{\underline{\pi}}_n(X,\alpha).

There are variants ‘at infinity’ of both the Waldhausen boundary and these groups, otained using germs instead of proper maps. These will be denoted with a \infty as a super- or suffix on the above notation.

Brown–Grossman homotopy groups

(Once over lightly here, more details at Brown-Grossman homotopy group.) In fact there is another different way of looking at these groups, which has a more geometric feel to it. Historically these groups were not the first successful attempt. This was due to Ed Brown and uses strings of spheres. These are examples of spherical objects and the resulting ‘groups’ (better thought of as ‘Π 𝒜\Pi_\mathcal{A}-algebras’) have a rich structure. They are discussed at Brown-Grossman homotopy groups. They do link with the above homotopy groups of the Waldhausen boundary, which are called the Steenrod homotopy group?s, (for reasons that will be explained there).


Survey article:


  • H. J. Baues and A. Quintero, Infinite homotopy theory, Volume 6 of K-monographs in mathematics, Springer, 2001

  • Bruce Hughes and Andrew Ranicki, Ends of Complexes, Cambridge Tracts in Mathematics (No. 123), C.U.P.

Lecture Notes:

Revised on July 11, 2017 17:23:02 by Bartek (