(-)q [xl, * * *,p Xq_,] * (2) isomorphism Hk(X; G) - Hk(Ti; G), k = 1, 2,.. (1) cohomology group of the group 7rj. The latter groups are defined

Transcription

1 MA THEMA TICS: EILENBERG AND MACLANE 443 from unitemporal to multitemporal processes. A significant interpretation of multitemporal processes from the vantage point of such systems of equations would be indicated. I See Lefschetz, S., Algebraic Topology, 1942, p. 31; and Braconnier, Jean, "Spectres d'espaces et de groupes topologiques," Portugaliae Math., 1, (1948). 2 Grundbegriffe der Wahrscheinlichkeitsrechnung, Ergebnisse der Mathematik, 2, (1933). 3 "Stochastic Processes," Ann. Math., 48, (1947). 4 "Diffusion Equation and Stochastic Processes," these PROCEEDINGS, 35, (1949). COHOMOLOG Y THEORY OF ABELIAN GROUPS AND HOMOTOP Y THEORY I By SAMUEL EILENBERG AND SAUNDERS MACLANE* DEPARTMENTS OF MATHEMATICS, COLUMBIA UNIVERSITY AND THE UNIVERSITY OF CHICAGO Communicated May 20, 1950 If a topological space X is aspherical (i.e., if all homotopy groups other than the fundamental group vanish), Hurewicz has shown that the fundamental group 7r of the space determines all the homology and cohomology groups of the space. After further investigations by H. Hopf, the authors and subsequently others' obtained an algebraic formulation for this determination, by exhibiting, for each abelian coefficient group G, a natural isomorphism Hk(X; G) - Hk(Ti; G), k = 1, 2,.. (1) between any cohomology group of the space and the corresponding (algebraic) cohomology group of the group 7rj. The latter groups are defined as the coiiomology groups of a certain cell complex K = K(7r,, 1) depending only on the group 7rj. The q-dimensional cells of this complex are all the q-tuples [xi,..., xj] of elements xi of the group, and for q > 1 the boundary of any such cell is defined as q-1 [Xl,...Xq] = [X2,...,Xq] +. (-1)'[Xi,.X+,...,Xq] + L = 1 (-)q [xl, * * *,p Xq_,] * (2) A "normalization" of this complex is also possible. If KN is the subcomplex spanned by all cells [xi,..., Xq] with some xi = 1, then the cohomology groups Hk(7ri; G) of K are isomorphic2 to the relative groups of K modulo K2N.

2 444 MA THEMATICS: EILENBERG AND MACLANE PROC. N. A. S. In the study of the effect of higher homotopy groups upon cohomology (and homology) groups, it appears efficient to first isolate the effect of a single homotopy group. Hence, in this note we consider any arcwise connected topological space X, in which the mth (abelian) homotopy group II = rm is given, with m > 1, and in which all the other homotopy groups vanish (rw = 1, 7ri = 0 for i 5 1, i 5im). Previously, we obtained,3 in analogy with (1), an expression Hk(X; G) _ Hk[K(H, m); G] k = 1, 2,... (3) for the singular cohomology groups of the space in terms of the cohomology groups of a certain cell complex K(ll, m) defined algebraically in a fashion similar to K(ll, 1), but depending essentially upon the commutativity of II (see the detailed description below). The algebraic cohomology groups appearing in (1) and (3) are also essential for the definition4 of certain "obstruction invariants" of spaces; these in turn, seem to be applicable to the problem of extension and classification of continuous mappings.' In a very few cases, the groups of K(fl, m) have been computed' by applying deeper methods of homotopy theory to a suitably constructed topological space. This note will state some of the results of a systematic study of the groups of K(H, m) by purely algebraic methods. A main result is the fact that these groups obey a curious analog of the Freudenthal suspension theorem.7 We begin with a description of the complex K(II, m). Choose for each positive integer q a standard q-dimensional simplex A. with ordered vertices (, 1,..., q), and let ei,, for i = 0, 1,.., q, denote that mapping of A,-, in A. obtained by mapping the vertices 0, 1,..., q - 1 of A,, in order upon the vertices of A,, omitting the vertex i of 4. The q-cells of the complex K(ll, m) are the m-dimensional cocycles g E Z'(4; II); for each g, the mapping e./ yields a cocycle Fig = geq' e Z"'( A,.; H), and thus a q - 1 cell of K. The boundary of the q-cell g is defined as ag =,(- 1)'Fig, where the addition from i = 0 to q is to be regarded as a sum of cells (and not as addition of cocycles). K(fl, m) is the cell complex with these cells and this boundary formula. The suspension homomorphism, mapping K(II, m) into K(II, m + 1), is obtained by first assigning, to each g, the "suspended" (m + 1) cocycle Tg on 4+,, defined for each m + 1 dimensional ordered simplex (ro,..., rm+i) of A,+, as (Tg)(ro,..., r,+i) = g(ro,..., rm) if rm+i = q + 1, = 0 if r+ < q +1. If go denotes the cocycle which is identically zero, in the appropriate dimension, then the suspension mapping

3 MA THEMA TICS: EILENBERG AND MACLA NE 445 Sg = Tg-go (4) is a chain transformation (raising dimensions by 1) of K(ll, m) into K(H, m + 1), and hence induces homomorphisms S:Hm+k[K(I m + 1); G] -- H`-l+k[K(II, i); G] (5) on the corresponding cohomology groups, for k = 1, 2, THEOREM 1. For k < m, the suspension homomorphism S is an isomorphism onto. For k = m + 1, it is an isomorphism into. The argument depends upon an algebraic reduction of the complex K(II, m) to a simpler "cubical" complex Q(ll) depending on the abelian group H alone. Each element x of II determines a 1-cell [xl of Q, each pair x, y, a 2-cell [x, y] of Q, with boundary [x, y] = [x] + [y] - [x + y] (6) A 3-cell of Q is a 2 X 2 square of elements in H, with boundary xl sl - [x, y] + [r, s]- [x + r, y + s- - [x, r] - [y, s] + [x + y, r + s]. (7) The proof that 66 = 0 uses the hypothesis that H is abelian. In general, an (n + 1)-cell of Q will be a 2 X 2 X... X 2 hypercube of dimension n, with entries in H, and the boundary will consist of 3n terms formed, as in (7), by slicing the hypercube. Explicitly, label the vertices of the hypercube by the n-tuples (el,..., e.) with each ei = 0 or 1. An (n + 1)-cell of Q is any function o- with arguments all n-tuples (el,,en) and with values in H. The faces Ri, Si, and Pi of a are defined, for i = 1,..., n, to be the n-cells (Ria)(ej,, (-l-) = 7(ely.-. pe-1p Oy feix *- n-l)y (Sta) (el *, en-1) = 0-(C1, * * *, C-i-1, 1, ei, *, en-1) I (Pia)(el,e*n-l) = (Rpu)(el,..., e,-l) + (So)(1*,.., where the addition is that of the group I. The boundary of a is a = Z (-1)f(Pi - Ri- Sia) where the addition is that of chains in Q; one has 66 = 0. Within the complex Q consider the subcomplex QN which is spanned by the "slabs" (all those n + 1 cells a with n _ 1 such that for some index i one has either Ri or Sio- identically zero) and by the "diagonals" (all those cells of the form Dio- for some i), where

4 446 MA THEMA TICS: EILENBERG A ND MA CLA NE PRoC. N. A. S. (D1o)((Eb.n..+l, efl) = 0(6..*i 6fi 6f+2, *-- en+l)f 6 = ei+1, = 0 if es $ t+1. For any abelian group G we then define the cubical cohomology groups Qn(f; G) as the relative cohomology groups of Q modulo QN; Qn(TI; G) = Hn(Q, QN; G). (8) In particular, equation (6) shows that a 1-dimensional cubical cocycle is a homomorphism, so that Q1(TI; G) is the group of homomorphisms of II into G. Similarly equation (7) implies that a 2-dimensional cubical cocycle is a (normalized) symmetric factor set f(x, y) of II in G, so that Q2(ll; G) is the group of abelian group extensions8 of G by II. The higher cubical cohomology groups of II appear to be new. THEOREM 2. If II + H' is the direct sum of two abelian groups, there is a natural isomorphism Qn(I(; G) + Qn(H'; G). Qn(ll + HI; G) In a subsequent note we shall show that this property, together with the character of the complex Q in the case when H is an infinite cyclic semigroup, serve to characterize the cubical cohomology groups, in the sense that any other construction for all groups II of suitable cell complexes with these properties yields exactly the cubical cohomology groups here defined. THEOREM 3. There are isomorphisms H`n-l+k[K(nJ, m); G] _ Qk(f1; G), k = 1, 2,..., m. The isomorphism 0 in question is related to the suspension homomorphism S by OS = 0; hence this result at once yields the suspension theorem, for k < m. To extend this result to higher values of k, we introduce a sequence of cohomology groups intermediate to the cubical and the ordinary cohomology groups of HI. They are defined by means of subcomplexes Q, of Q, for t = 0, 1, 2,..., c. Call an index i(i = 1,..., n - 1) critical for an (n + -1)-cell a if R1S1a is not identically zero; that is, if o(ej, 6f-1, 1, Oi+2, 0,.*, 6En) # 0 for some choice of the e's. The level of a cell a is the number of distinct critical indices for a. The complex Q(II) is defined as that subcomplex of Q(HI) which is spanned by all cells of level at most t. In particular, if n. t + 1, any (n + 1)-cell o lies in Q,. We define the cubical cohomology groups of level t for HI as the relative groups of Qt modulo Qt n QN: Qn, t(hj; G) = Hn(Q,, Qt n QN; G). (9) The complex Q,c is the whole complex Q, hence Qn a>(h) = Qn(H), and also Qn. '(H) = Q"(II), if n _ t + 1. On the other hand, the (n + 1)-

5 MA THEMA TICS: EILENBERG AND MACLANE 447 cells of Qo have o(el,..., e.) = 0, unless E_ el.& _.. -< e, hence may be mapped in 1-1 fashion on the (n + 1) tuples [xi,..., xn+1], where xi = a'(o, O,..., 0, 1,..., 1) with i-1 arguments equal to 1. This correspondence can be used to show that Q#, 0(11; G) is the nth cohomology group of K(11, 1), modulo KN; hence Qn, 0(11; G)-_IP(II; G)- (10)- Theorem 3 may now be extended as follows. THEOREM 4. For all k = 1, 2,... there is an isomorphism - Htm 1+k[K(H, m); G] Qk' m-1(n1; G). In view of (3), we thus have THEOREM 5. If X is an arcuise connected topological space with a given homotopy group Tm and with all other homotopy groups trivial, the singular cohomology groups of X with coefficients in any abelian group G are determined by Trm according to the formulae Hl (X; G) = 0, q = l1 2y...,Im -1, Hm-i+7(X; G) = Qk, m-1(tm; G), k = 1, 2, The theorem formally includes the case m = 1. Briefly, we may say that the ordinary cohomology groups for the not necessarily abelian fundamental group yield the effect of that group upon cohomology groups of the space, and that the cubical cohomology groups, defined, at suitable levels, for abelian groups, provide the corresponding effect for higher homotopy groups. In a later note we shall give an alternative description of the groups Q k, =-I(11; G) which is more suitable for algebraic computations and topological applications. * Essential portions of the study here summarized were done during the tenure of a John Simon Guggenheim Memorial Fellowship by one of the authors. 'The literature is summarized in Eilenberg, S., Bull. Am. Math. Soc., 55, 3-37 (1949). 2 Eilenberg, S., and MacLane, S., Ann. Math., 48, (1947). 3 Eilenberg, S., and MacLane, S., Ibid., 46, (1945)j4 4Eilenberg, S., and MacLane, S., "Relations between Homology and Homotopy Groups of Spaces, II," Ibid., to appear. ' Whitney, H., Ibid., 50, (1949). 6 Whitehead, G. W., PROC. NATL. ACAD. SCa., 34, (1948). 7 Freudenthal, H., Compositio Math., 5, (1937). 8 This is the group Ext (G, r) used in Eilenberg and MacLane, Ann. Math., 43, (1942).