The Steenrod Algebra and Its Dual John Milnor The Annals of Mathematics, 2nd Ser., Vol. 67, No. 1. (Jan., 1958), pp. 150-171. Stable URL: http://links.jstor.org/sici?sici=0003-486X%28195801%292%3A67%3A1%3C150%3ATSAAID%3E2.0.CO%3B2-2 The Annals of Mathematics is currently published by Annals of Mathematics. Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/annals.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers, and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take advantage of advances in technology. For more information regarding JSTOR, please contact [email protected]. http://www.jstor.org Sun Jul 1 19:31:29 2007 ANNALS OF MATHEMATICS Vol. 67, No. 1, January, 1958 Printed in Japan T H E STEENROD ALGEBRA AND ITS DUAL' BY JOHN MILNOR (Received May 15, 1957) 1. Summary Let .9* denote the Steenrod algebra corrresponding to an odd prime p. (See $ 2 for definitions.) Our basic results ($3) is that c'!* is a Hopf algebra. That is in addition to the product operation there is a homomorphism satisfying certain conditions. This homomorphism +* relates the cup product structure in any cohomology ring H*(K, 2,) with the action of 9" on H*(K, 2,). For example if .Pne 9'"(P-l) denotes a Steenrod reduced pthpower then The Hopf algebra has a dual Hopf algebra The main tool in the study of this dual algebra is a homomorphism A* : H*(K, 2,) -+ H*(K, 2,) @ 9j which takes the place of the action of .3 " on H*(K, 2,). (See $4.) The dual Hopf algebra turns out to have a comparatively simple structure. In fact as an algebra (ignoring the "diagonal homomorphism" 4,) it has the form where E(ri , 2pL- 1)denotes the Grassmann algebra generated by a certain element r1 e X ? , I - ~and , P(E,, 2p" 2) denotes the polynomial algebra generated by 6,e . % , L - ~ . I T h e author holds an Alfred P. Sloan fellowship. 150 151 THE STEENROD ALGEBRA AND ITS DUAL In 9 6 the above information about cir, is used to give a new description of the Steenrod algebra 9 *. An additive basis is given consisting of elements Q,,EOQIE1 with by E, = 0, 1; Y, 2 ... y' '172'.' 0. Here the elements Q , can be defined inductively Q" = 8,Q L += l '/j21~Q& - QLJ71~L ; while each /"I""A. is a certain polynomial in the Steenrod operations," of dimension r1(2p - 2) r,(2p" 2) . . r,(2pk - 2) . The product operation and the diagonal homomorphism in ,Y * are explicitly computed with respect to this basis. The Steenrod algebra has a canonical anti-automorphism which was first studied by R. Thom. This anti-automorphism is computed in 97. Section 8 is devoted to miscellaneous remarks. The equation 8,-1'' = 0 is studied; and a proof is given that .'/ * is nil-potent. A brief appendix is devoted to the case p = 2. Since the sign conventions used in this paper are not the usual ones (see 3 2), a second appendix is concerned with the changes necessary in order to use standard sign conventions. + + + 2. Prerequisites: sign conventions, Hopf algebras, the Steenrod algebra If a and b are any two objects to which dimensions can be assigned, then whenever a and b are interchanged the sign (- l)dinl""'"'hwill be introduced. For example the formula for the relationship between t h e homology cross product and the cohomology cross product becomes This contradicts the usual usage in which no sign is introduced. In the same spirit we will call a graded algebra commutative if Let A = (. , A _ ,, A,, A,, - ) be a graded vector space over a field F. The dual A' is defined by Al, = Hom ( A _ ,, F). The value of a homomor, ) . I t is understood that phism a' on a e A will be denoted by < a f a <a',a = 0 unless dim a' + dim a = 0. (By an element of A we mean an element of some A , .) Similarly we can define the dual A" of A'. Identify > - T h l s has no relation to the generalized Steenrod operations ,-;/'I defined by Adem. 152 JOHN MILNOR each a e A with the element a" E A" which satisfies a",a' ) = (- 1)dinla" dirn a' < a fa , ) (2) for each a' E A'. Thus every graded vector space A is contained in its double dual A". If A is of finite type (that is if each A, is a finite dimensional vector space) then A is equal to A". Now if f : A -t B is a homomorphism of degree zero then f ': B' A' and f ": A" -t B" are defined in the usual way. If A and B are both of finite type it is clear that f = f ". The tensor product A @ B is defined by ( A@ B), = j = n Ai @ B j , where " " stands for "direct sum ". If A and B are both of finite type and if A, = B, = 0 for all sufficiently small i (or for all sufficiently large i) then the product A @ B is also of finite type. In this case the dual ( A@ B)' can be identified with A' @ B' under the rule < -+ xi+ In practice we will use the notation A, for a graded vector space A satisfying t h e condition Ai = 0 for i < 0 . The dual will then be denoted by A* where An = A'-,= Hom (A,, F). A similar notation will be used for homomorphisms. By a graded algebra (A,, 4,) is meant a graded vector space A, together with a homomorphism +,: It is usually required 1 e A,. The algebra is By a connected Hopf algebra with unit (A,, A,@A,-tA,. that #,: be associative and have a unit element connected if the vector space A, is generated by 1. algebra (A,, +,, 4,) is meant a connected graded +,), together with a homomorphism satisfying the following two conditions. 2.1. 4, is a homomorphism of algebras with unit. Here we refer to the product operation 4, in A, and the product (al@I a,) (a,@ a,) = ( - l)diln '2 '3 (a1a,) @ (a,.a,) in A,@ A,. 2.2. For dim a > 0 , the element &,(a) has the form a @ 1 1 @I a b, @ ci with dim b , , dim c, > 0. A%propriate concepts of associativity and commutativity are defined, not only for the product operation #, , but also for the diagonal homomorphisms 4,. (See Milnor and Moore [3]). To every connected Hopf algebra (A,, #, , 4,) of finite type there is as- + + T H E STEENROD ALGEBRA AND I T S DUAL 153 sociated the dual Hopf algebra (A*,$*, +*), where the homomorphisms are the duals in the sense explained above. For t h e proof that the dual is again a Hopf algebra see [3]. d GxG (As an example, for any connected Lie group G t h e maps G P G give rise to a Hopf algebra (H,(G), p , , d,). The dual algebra (H*(G), v,p*) is essentially the example which was originally studied by Hopf .) For any complex K the Steenrod operation LY'~ is a homomorphism - - The basic properties of these operations are the following. (See Steenrod L41.1 2.3. Naturality. If f maps K into L then f * S = LgJYf". 2.4. For a e H3(K,Z,), if i > j / 2 then dJkx = 0. If i = j / 2 then = an.If i = 0 then ../'la = a . 2.5. ~ ~ " ( a v p ) Piaud3j$. We will also make use of the coboundary operation 6: H9(K,2,) -+ H9+'(K,2,) associated with the coefficient sequence =x,+,=, The most important properties here are 2.6. 66 = 0 and 2.7. 6(a v /3) = (6a) v P (- 1 ) " ' " ' " ~ 6P, as well as the naturality condition. Following Adem [I] the Steenrod algebra .Y* is defined as follows. The free associative graded algebra ./r*generated by the symbols 6, y>0 , /'I, acts on any cohomology ring H*(K, 2,) by the rule (OIOL O,).a = (8,(8, .. (8,a) ... )). (It is understood that 6 has dimension 1 in 3% and that ,lJi has dimension 2i(p - I).) Let 3"denote the ideal consisting of all f e F* such t h a t fa = 0 for all complexes K and all cohomology classes a e H*(K, 2,). Then 9* is defined as the quotient algebra 9-*j,F*. It is clear that 9* is a connected graded as'* is not commutative. sociative algebra of finite type over 2,. However 5 (For an alternative definition of the Steenrod algebra see Cartan [2]. The most important difference is that Cartan adds a sign t o the operation 8.1 The above definition is non-constructive. However it has been shown + - 154 JOHN MILNOR by Adem and Cartan that 9," is generated additively by the "basic monomials '' where each E~ is zero or 1 and Furthermore Cartan has shown that these elements form an additive basis for 9* . 3. The homomorphism 4" LEMMA 1 . F o r each element 8 of 9 * there i s a unique element +*(8)= C O',@ 8;' of 5'""@ 9* such that the identity x 8(a "3) = (- 1)d'"Oi' d l l n a 8I(a) O;'(P> i s satisjied for all complexes K and all elements a , 13 e R " ( K ) . Furtheymore i s a r i n g homomorphism. (By an "element" of a graded module we mean a homogeneous element. The coefficient group 2, is to be understood.) I t will be convenient to let 9* @ 9"act on H * ( X )@ H * ( X ) by the rule (of @ 8")(a @ p) ( - 1)di'n8" d l l n a @ ( a )@ d"(P) . Let c: H * ( X )@ H * ( X )-. H " ( X ) denote the cup product. The required identity can now be written as &(a @ p ) = c+"(O)(a@ 9) . PROOFO F EXISTENCE. Let 5 2 denote the subset of ,5'* consisting of the required identity all 19 such that for some p E Y * @ 9" O P ) = cp(a O P ) is satisfied. We must show that @ , = .9*. The identities 6(a v,?) = 6a wl! and / ' n ( a u p )= + (- l ) d n n aw a 8p xi+,=, clearly show that the operations 6 and to then the identity 7 J L Uu PJi3 lJn belong to .P. If (I,, O2 belong THE STEENROD ALGEBRA AND ITS DUAL 155 show that B,8, belongs to A.Similarly 2 is closed under addition. Thus R is a subalgebra of .i'* which contains the generators 6, .lJn of 9*. This proves that 2 = .Y*. PROOFO F UNIQUENESS. From the definition of the Steenrod algebra we see that given an integer n we can choose a complex Y and an element r e H * ( Y ) so that the correspondence defines an isomorphism of ,Y into H k f i ( Y )f or i 5 n . (For example take Y = K(Z,, k ) with k > n.) I t follows that the correspondence defines an isomorphism j of ( 9* @ 9 *)$ into H2""Y x Y ) for i 5 n . Now suppose that p l y p, e .9* @ 9* both satisfy the identity Bc(a @ P ) = cp,(a @ /3) for the same element 8 of .Y ". Taking X = Y x Y , a = r x 1, p = 1 x r , we have cp,(a @ p) = j(p,). But the equality j(pl) = j(p,) with dim p, = dim p, = n implies that p, = p,. This completes the uniqueness proof. Since the assertion that +* is a ring homomorphism follows easily from the proof used in the existence argument, this completes the proof. As a biproduct of the proof we have the following explicit formulas: THEOREM 1. The homomorphisms give 9* the structure of a Hopf algebra. Furthermore the product 4" is associative and the "diagonal homomorphism" +* i s both associative and commutative. PROOF.I t is known that (9*, 4") is a connected algebra with unit ; and that +* is a ring homomorphism. Hence to show that .'/ * is a Hopf algebra it is only necessary to verify Condition 2.2. But this condition is clearly satisfied for the generators 8, and ,/'" of ,Y *, which implies that it is satisfied for all positive dimensional elements of .9*. I t is also known that the product 4" is associative. The assertions that 4" is associative and commutative are expressed by the identities 156 JOHN MILNOR (2) T+*B = +*B for all 8, where T(8' @ 0") is defined a s (- l)d'"e' 8'' @ 8'. Both identities are clearly satisfied if 8 is one of the generators 8 or :yJn of .9". But since each of the homomorphisms in question is a ring homomorphism, this completes the proof. As an immediate consequence we have: COROLLARY 1. There is a dual Hopf algebra with associative, commutative product operation. 4. The homomorphism A* Let H, , H * denote the homology and cohomology, with coefficients 2, , of a finite complex. The action of 9* on H * gives rise to a n action of 9* on H, which is defined by the rule: for all p E H, , B E 9*, a E H * . This action can be considered as a homomorphism A*: H , @ 9 * -+H*. The dual homomorphism A*: H * -,H * @ .9* will be the subject of this section. @ 9' H, has a Alternatively, the restricted homomorphism Hnti dual which we will denote by -+ In this terminology we have carrying Hn into C,Hn @ .%'. The condition that H * be the cohomology of a finite complex is essential here, since otherwise A* would be a n infinite sum. The identity ~ ( 4 8 2= ) (~8,)8, can easily be derived from the identity (B,8,)a = B,(B,a) which is used to define the product operation in 9 *. In other words the diagram THE STEENROD ALGEBRA AND ITS DUAL -- [A* 0 1 A* H*@ 9" is commutative. Therefore the dual diagram Id* H, Ti* @ 1 Ti* is also commutative. Thus we have proved : LEMMA2. T h e identity (A* 0l)A*(a)= (1 @ $*)A*(a) holds for every a e H*. The cup product in H* and the product in .%induce a product operation in H* @ LEMMA3. T h e homomorphism A* : H* -+ H*@ 9 * i s a ring homomorphism. PROOF.Let K and L be finite complexes, let B be a n element of 9*, and let +*(B)= 8; @ 8;'. Then for any a E H * ( K ) , P E H*(L) we have d . ( a x p) = (- 1)dllne;'dlln a B',a x B;',l. Using the rule +, s. z we easily arive a t the identity In other words the diagram H,(K) @ H,(L) @ 9% @ 9% 4' @ ' @ *' H,(K) @ H,(L) @ 9* = H,(K x L)@ 9* 11@~@1 H*(K)€3 9% @ H*(L) @ 9% 11. '* @ * * H J K ) €3 H J L ) = H*(K x L) is commutative (where T interchanges two factors as in 53). Therefore the dual diagram is also commutative. Setting K = L , and letting d : K -, K x K be the diagonal homomorphism we obtain a larger commutative diagram 158 JOHN MILNOR Now starting with a @ ,3 e H * @ H * and proceeding to the right and up in this diagram, we obtain ]"(a w p). Proceeding to the left and up, and then to the right, we obtain I*(a).I*(P). Therefore which proves Lemma 3. The following lemma shows how the action of .Y*on H * ( K ) can be reconstructed from the homomorphism I*. we have LEMMA4. If A*(a) = C ai @ w, then for any 8 e 9'" ea = C ( - 1)dimaidim w . (8, . PROOF.By definition <P, d a > = <pO,a) = < A * ( P O ~ ) , ~ > = < P Od , ]*a) = C < P , a , > < d ,w , ) * . Since this holds for each p e H, , the above equality holds. @ H * -+ H * REMARK.To complete the picture, the operation 7": 9" has a dual 7,: H, -. ,S/j @ H, . Analogues of Lemmas 2 and 4 are easily obtained for 7,. If a product operation K x K -t K is given, so that H,, @ H,, have product operations; then a straightforward and hence 9; proof shows that 7, is a ring homomorphism. (As an example let K denote the loop space of an (n 1)-sphere, or an equivalent CW-complex. Then H,(K) is known to be a polynomial ring on one generator p E Hn(K). The element + is evidently equal to 1 @ p. Therefore 7,(pk) = 1@ p" for all k. Passing to the dual, this proves that the action of .ir * on H ' Y K ) is trivial.) 5. The structure of the dual algebra 9 ; As an example to illustrate this operation A* consider the Lens space X = SLJtl/Z,where N is a large integer, and where the cyclic group Z, acts freely on the sphere SZS'l.T hus X c a n be considered as the ( 2 N 1)skeleton of the Eilenberg-MacLane space K(Z,, 1). The cohomology ring H * ( X ) is known to have the following form. There is a generator a E H1(X) and H ( X ) is generated by 8 = 8a. For 0 2 i _I N, the group H2i(X)is generated by Pi and HLi+l(X)is generated by a p t . The action of the Steenrod algebra on H*(X) is described a s follows. I t will be convenient to introduce the abbreviations + 159 THE STEENROD ALGEBRA AND ITS DUAL satisfies M,i3 = However i f 8 LEMMA5. T h e element M, E .9'pk-3 i s a n y monomial in the operations 6, ,P1, P " .. which i s not of the VJpY31 t hen 03 = 0. S i m i l a r l y (M,6)a = ppk but Ba = 0 f o r m ,'/'"'-l i f 0 i s a n y monomial in the operations 6, q5I , q5" which does not have _Y16o r 0 = 1. the f o r m 0 = ";7yk-1 PROOF.It is convenient to introduce the formal operation @ , =1 /'I 9 j 2 . It follows from 2.4 that 9 j p = 3 + 3". Since q5is a ring homomorphism according to 2.5, it follows that Bp" ((B P")". In particular if i = p' this gives /lijlir = (3 + / ~ P ) P ' = ,3~' i? p r + l . In other words 3'. if j = 0 qjjpuT = 3 n T + l if j = pr .. .. + + + .. + + 1 otherwise . Since 6p = ip-lap = ipz-'Ma = 0 it follows that the only nontrivial operation 6 or 9 j j which can act on pnr is P p ' . Using induction, this proves the first assertion of Lemma 5. To prove the second it is only necessary to add that ?'a = 0 for all j > 0, according to 2.4. Now consider the operation A*: H * ( X ) -+ H * ( X ) @ % . LEMMA6. T h e element R*a has t h e f o r m a @ 1 3 @ r, 3'' @ r, ... +Ppr @ rTwhere each r, i s a well defined element of 9 & ~ -and ~ , where pJ i s the largest pmoer of p w i t h p' 5 N. Similarly R*P has the f o r m 0 + P @ Eo + P" @ E l + +/3" + + @ Er , where E, = 1, and zc~hereeach E, i s a well defined element of x p k - , . PROOF.For any element 8 of 9; Lemma 5 implies that BP = 0 unless i is the dimension of one of the monomials M o ,M I , : that is unless i has the form 2pk - 2. Therefore, according to Lemma 4, we see that 2'3 = O unless i has the form 2pk - 2. Thus ... A*? = Jo(p) + A2p-Yt3) + ... + Rzpr-2 (8) . Since Rmk-"p) belongs to H*,(X) @ .9&r-2, it must have the form t3*k @ E, for some uniquely defined element E, . This proves the second assertion of Lemma 6. The first assertion is proved by a similar argument. REMARK.T hese elements E, and r, have been defined only for k S r = [log, N ] . However the integer N can be chosen arbitrarily large, so we have actually defined E, and T , for all k 2 0. Our main theorem can now be stated as follows. THEOREM 2. T h e algebra 9*i s the tensor product o f the Grassmann algebra generated by T ~T ,~ , and the polynomial algebra generated by El, E,, .-.. .. 160 JOHN MILNOR The proof will be based on a computation of the inner products of moand 6. The folnomials in ri and t, with monomials in the operations .7" lowing lemma is an immediate consequence of Lemmas 4, 5 and 6. LEMMA7. The inner product equals one, but (8, 6,) = 0 if B is any other monomial. Similarly > but (8, r , = 0 if 8 is any other monomial. Consider the set of all finite sequences I = (E,, r , , El, r,, E~ = 0 , l and rt = 0, 1, 2, . For each such I define .. . - ) where Then we must prove that the collection (w(I)) forms a n additive basis for For each such I define c ~ * . where S, = (El + ri)pi-l, ., '' S, = CT=, (Ei + r i ) ~ ' - ~ I t is not hard to verify that these elements B(I) are exactly the "basic monomials" of Adem or Cartan. Furthermore B(I) has the same dimension as w(I). Order the collection { I ) lexicographically from the right. (For example (1, 2, 0, - - ) < (0, 0, 1, ...).) LEMMA8. The inner product 8(1), w(J) i s equal to zero ij- I < J and *lifI=J. Assuming this lemma for the moment, the proof of Theorem 2 can be completed as follows. If we restrict attention to sequences I such that . < > dim o ( I ) = dim 8(I) = n , < > then Lemma 8 asserts that the resulting matrix B(I), w(J) is a nonsingular triangular matrix. But according to Adem or Cartan the elements B(I) generate .ir ". Therefore the elements o ( J ) must form a basis for .i/,;which proves Theorem 2. (Incidentally this gives a new proof of Cartan's assertion that the @(I)are linearly independent.) PROOF O F LEMMA 8. We will prove the assertion (B(I), w(I) ) = + 1 by induction on the dimension. I t is certainly true in dimension zero. Case 1 . The last non-zero element of the sequence 1 = (E, , r, , , E,-, , r , , 0, . - - ) i s r , . Set I t =(E,,Y,,. - . , ~ , - , , r , - 1, 0, so that o ( I ) = to(I')E,. Then . - 0 . ) THE STEENROD ALGEBRA AND ITS DUAL ... . where the summation extends over all sequences ( E : , , s;) and ( E ; , , s z ) with EL &I' = E , and s: sl' = s,. Substituting this in the previous expression we have + + ( B(I),o ( I )) = C + < 8'6 . 9'' i , o ( l f>) < Oe6' ... .-Y Ek '?';', >. But according to Lemma 7 the right hand factor is zero except for the special case 8';' ... /' = _(3 8 - I . p P . 3 ) I , " .. in which case the inner product is one. Inspection shows that the corresponding expression 6% . " on the left is equal to B(If) and hence < . /' < ; > that B(I),o ( I )) = f B ( I ' ) , o(I') = -+ 1. Case 2. The last non-zero element of I = (E,, Y,, . , r, , E , , 0 , &l, = 1. Define I' = ( E , , r , , , r, , 0 , .) so that .. .. . ...) is Carrying out the same construction as before we find that the only nonvanishing right hand term is .Ppk-' . y l O , T, = 1. The corresponding left hand term is again < 6 ( I f )o, ( I f ) ) ; so that < 6 ( I ) ,w ( I ) >= + ( @ ( I r )w,( I ' ) ) = -+ 1, with completes the induction. The proof that B(I),o ( J ) = 0 for I < J is carried out by a similar induction on the dimension. Case l a . The sequence J ends with the element rk and the sequence I ends a t the corresponding place. Then the argument used above shows that < > .. > < Case l b . The sequence J ends with the elements r, , but I ends earlier. Then in the expansion used above, every right hand factor ,yEb',</~ 8%'-I,Ek < 8;' > ... is zero. Therefore ( @ ( I )o, ( J ) ) = 0. Similarly Case 2 splits up into two subcases which are proved in an analogous way. This completes the proof of Lemma 8 and Theorem 2. To complete the description of .9*as a Hopf algebra it is necessary to compute the homomorphism . But since is a ring homomorphism it +, +, 162 J O H N MILNOR is only necessary to evaluate it on the generators of S , . THEOREM3. T h e following formulas hold. +*(Ed = z:=o Eplt +*(rk)= Cf=oELi 0rt O Et + ~k @1. The proof will be based on Lemmas 2 and 3. Raising both sides of the equation = C B"' O E, to the power p%e obtain A*(#") = pp"' 0E?;I . Now (A* l ) A * ( P ) = (A* @ 1) = Ci,j8""' z8"" 0 EfY( 0Et . Comparing this with (1 0+*)A*(P) = C ppk 63 +*(Ed We obtain the required expression for +,([,). Similarly the identity (A* @ l)A*(a)= (1 0+*),?*(a) can be used to obtain the required formula for +,(re). 6. A basis for 9" Let R = (r,,r,, ..) range over all sequences of non-negative integers which are almost all zero, and define E(R)= E:lEITa . . Let E = (Eo , E l , .) range over all sequences of zeros and ones which are almost all zero, and define T ( E )= r,%~r,'l . Then Theorem 2 asserts that the elements . ... Hence there is a dual basis {p(E,R ) }for form a n additive basis for 9,. 9". That is we define p(E,R) e .9 by * 1 if E = E ' , R = R ' 0 otherwise. Using Lemma 8 it is easily seen that p(0, ( r ,0 , 0 , .)) is equal to the Steenrod power ..Pr. This suggests that we define?-YR as the basis element p(0, R) dual to E(R). (Abbreviations such as .GJolin place of 6 '"'1 will be frequently be used.) .. c 0 3 O j h THE STEENROD ALGEBRA AND ITS DUAL 163 Let Q, denote the basis element dual to r,. For example Qo = p(1, 0 , .), 0 ) is equal to the operation 6. It will turn out that any basis ,P R . element p(E, R) is equal to the product & Q,'oQIE1 THEOREM 4a. The elements . .. form a n additive basis for the Steenrod algebra .i/ * zchich is, u p to sign, The elements Q, e .Y 'P'-' dual to the known basis {r(E):(E)} for generate a Grassmann algebra: that is they satisfy z. They permute with the e l ~ m e n t s3' according to the rule (By the difference (r, , r 2 , - ) - (s, , s,, - ) of two sequences we mean the sequence (r, - s,, r, - s,, .). I t is understood, for example, t h a t . /'R-(~"O,...)is zero in case r, < pk.) As a n example we have the following where [a, b] denote the " commutator " ab - ( - 1)"'" a c"" 0 ba . COROLLARY 2. The elements Q, e 9i/-2pk-' can be defined inductively by the rule Qd= 8 , Qk+l= [ 9 p k , Q t I . .. To complete the description of 9 * a s a n algebra it is necessary to find the product l J K /'&. Let X range over all infinite matrices of non-negative integers, almost all zero, with leading entry ommitted. ), S ( X ) = (s,, s,, -), and For each such X define R ( X ) = (r, , r , , T(X) = ( t , , t,, ...), by (weighted row sum), rl = C, p'x,, (column sum), sj = xLj t n = CL+l=n xlj (diagonal sum). .. EL Define the coefficient b(X) = THEOREM 4b. The product TI t,! / nx,,!. . L ~ ' ~ . is L /equal '~ to 164 J O H N MILNOR where the sum extends over all matrices X satisfying the conditions R ( X ) = R , S ( X ) = S. As a n example consider the case R = ( r ,0 , -), S = (s,0 , -). Then the equations R ( X ) = R, S ( X ) = S become ... ...= x l O f pxI1+ = r , x t j = O for i > l , x ~ ~ + x ~ ~ S+, x i j = O for j > 1 , respectively. Thus, letting x = x,,, the only suitable matrices a r e those of the form with 0 5 x 5 Min (s,[rip]). The corresponding coefficients b ( X ) are the binomial coefficients ( r - px, s - x). Therefore we have 3. The product 9'99 s equal to COROLLARY The simplest case of this product operation is the following 4. If r , < p, r , < p, COROLLARY then 9 R9= ( r , , sl)(r2,s ,) . g j K+S As a final illustration we have : COROLLARY 5. The elements 9( 0 " ' 0 ' 0 "'I can be defined inductively by The proofs are left to the reader. 4b. Given any Hopf algebra A, with basis {a,) PROOFOF THEOREM the diagonal homomorphism can be written as $*(ad = C j , k cika, O a, The product operation in the dual algebra is then given by a ~ a k = $,*(cJJ g a*) = (- l ) d i m a ' d i n l a k cjkai , xi where {ai} is the dual basis. I n carrying out this program for the algebra .'/; we will first use Theorem 3 to compute &(E(T))for any sequence T = ( t , , t,, --.). Let [i,, i,, , i,] denote the generalized binomial coefficient (i,+ i, + . - -+ ik)!,lil!i2! so that the following identity holds - . a i l , ! ; THE STEENROD ALGEBRA AND ITS DUAL + .. + ~ (~1 xil+... k =) ~ +tk=n [il, , i,Iy:l . Y$ Applying this to the expression we obtain + + , x,, satisfying xi , - I 2 0, x,, x,, summed over all integers x,, , = t , . Now multiply the corresponding expressions for k = 1 , 2, 3, . Since the product [x,, , xO1][x,, , xll, x,,] [x,, , , xO3] is equal to b(X), we obtain +*(E(T)) = CTL=,== b(X)E(R(X)) @ E(S(X)) , summed over all matrices X satisfying the condition T ( X ) = X. In order to pass to the dual +* we must look for all basis elements r(E)E(T) such that +,(r(E)E(T)) contains a term of the form (non-zero constant)-E(R) @ E(S) . However inspection shows that the only such basis elements are the ones ((T) which we have just studied. Hence we can write down the dual formula +*(PR @ 9") = CR(x)-E, S ( x ) - S b ( X ) 9 T ( X .) This completes the proof of Theorem 4b. PROOF O F THEOREM 4a. We will first compute the products of the basis . . The dual problem is to study the elements p(E, 0 ) dual to r;or:l : % -+ 9 ; @ S$ ignoring all terms in .Y* @ .9*which homomorphism involve any factor E, . The elements 1 @ El, 1 @ E,, , El @ 1, of genemte a n ideal 3 Furthermore according to Theorem 3: @ . +, + (mod Y ) ( ) = r @ 1 1@ r (mod 3) . $*(Ek) = 0 Therefore &(r(E)E(R) 0 if R f 0 and $,(r(E)) = r(E,) (mod 2). T he dual statement is that - xal+o,-a * r(El) @ where it is understood that the right side is zero if the sequences El and E, both have a " 1" in the same place. Thus the basis elements p(E, 0 ) multiply as a Grassmann algebra. Similar arguments show that the product p(E, O)p(O, R ) is equal to 166 JOHN MILNOR p(E, R). From this the first assertion of 4a follows immediately. Computation of dln&,: We must look for basis elements r(E)[(R') such that +,(r(E)E(R1)) contains a term (non-zero constant).E(R) 0r, . Inspection shows that the only such basis elements are r,E(R), r, +,E(R (pk,0, )), T*+~E(R - (0, pk, 0, )), etc. Furthermore the corre1 . This proves that sponding constants are all + and completes the proof of Theorem 4. To complete the description of V *as a Hopf algebra we must compute the homomorphism +* . LEMMA9. The following formulas hold +*(&,I = Q, 0 1 + 10Qk 0, l J.R ~ 01 + 10 _j'Uil+ 09 +*(PR) =CRI+Ra=R + (For example +*(dJU1l)= gJ J'O' OJ1 /3u01 @ J O1.) REMARK.An operation 0 e ,Y* is called a derivation if it satisfies B(a v ,3) = (Ba) v 3 + ( - l)"n'edim a - 8 3 . a This is clearly equivalent to the assertion that B is primitive. It can be , 3' ', shown that the only derivations in .Y * are the elements Q,, Q, , gJ0,1, 2 ° * 0 z 1 , and their multiples. 7. The canonical anti-automorphism As a n illustration consider the Hopf algebra H,(G) associated with a Lie group G. The map g -+ g-l of G into itself induces a homomorphism c: H,(G) -+ H,(G) which satisfies the following two identities: (1) c(1) = 1 (2) if +,(a) = C a: 0a;', where dim a > 0, then C a;c(al') = 0. More generally, for any connected Hopf algebra A,, there exists a unique homomorphism c: A, -+ A, satisfying (1) and (2). We will call c(a) the conjugate of a. Conjugation is a n anti-automorphism in the sense that al dim '2 c(a2)c(aL) . c(a1a,) = ( - l)dim ; The conjugation operations in a Hopf algebra and its dual are dual homomorphisms. For details we refer the reader to [3]. For the Steenrod algebra .9 this *operation was first used by Thom. (See [5] p. 60). More precisely the operation used by Thom is B (~(6). -+ THE STEENROD ALGEBRA AND ITS DUAL 167 If 8 is a primitive element of 9" then the defining relation becomes 8.1 l.c(B) = 0 so that c(8) = - 8. This shows that c(Q,) = - Q,, ~ ( 9 ~ ' ) - J'~. The elements c(,Y2;'"),n > 0, could be computed from Thom's identity + however it is easier to first compute the operation in the dual algebra and then carry it back. By an ordered partition a of the integer n with length l(a) will be meant a n ordered sequence , of positive integers whose sum is n. The set of all ordered partitions of ?z will be denoted by P a r t (n). (For example P a r t (3) has four elements: (3), ( 2 , l ) (1,2), and ( l , l , l ) . In general P a r t (n) has 2"-l elements.) Given an ordered partition a E P a r t (n), let a(i) denote the partial sum Cj:: a ( j ) . the conjzbgate c(E,,) is equal to LEMMA 10. In, the dual algebra + + EaE? - E ~ E ~ E ? . ) (For example c(E,) = - E3 EIE: PROOF.Since +,(En) = C E Oti-,@ i t , the defining identity becomes This can be written as The required formula now follows by induction. ) an anti-automorphism, we can use LemSince the operation w -+ ~ ( w is ma 10 to determine the conjugate of an arbitrary basis element i(R). Passing to the dual algebra p* we obtain t h e following formula. (The details of the computation are somewhat involved, and will not be given.) consider the equations Given a sequence R = (r, , .. , r, , 0, a ) ; where the symbol 6,,,,, denotes a Kronecker delta; and for i = 1 , 2 , 3 , where the unknowns y, are to be non-negative integers. For each solution Y to this set of equations define S ( Y ) = (s, , s,, - ) by Sn (Thus s1 = YI,8 2 = Ya = C &Part Cn) Ya; . + % , I ,etc.) Define the coefficient b(Y) by T H E STEENROD ALGEBRA A N D ITS DUAL of the ideal L ~ ~ - ' Ywill ' * also be a solution. Now the identity P-1, SISz"' = (p - 1, S,)JJ T ~ + P - F~2 , ... JJ c/j = 1 0 if s, S 0 (mod p) -/slsn-l, sz ... if s, = 0 (mod p) shows that every element ~7"1'%"'Q,'o . with r, -- -- 1(mod p) actually belongs to the ideal. Applying the conjugation operation, this proves the following : PROPOSITION 1. The equation O,/jl = 0 has as solutions the elements o j the ideal p * J'"-'. An additive basis i s given by the elements Next we will study certain subalgebras of the Steenrod algebra. Adem shown that 9 * is generated by the elements Q,, J", d'", . Let .Yd*(n)denote the subalgebra generated by Q,, J", , J, ""-I PROPOSITION 2 . The algebra .i/ *(n) isfi~zitedimensional, having as basis the collection of all elements . r, < pn, r, < pi'-', ., r, < p . Thus .'/." is a union of finite dimensional subalgebras .Y" ( n ) . This clearly implies the following. COROLLARY 7. Every positive dimensional element of '/ * is nil-potent. It would be interesting to discover a complete set of relations between the given generators of .ir *(n). For n = 0 there is the single relation [QolQu] = 0, where [a, b] stands for ab - (- l)d""""'"h ba. For n = 1 there are three new relations [Q, , [./'I1 I, [ Qo]] = 0 , [S Y", Qo]] = 0 and (./jl)" = 0 . For n = 2 there are the relations as well as several new relations involving Q,. (The relations ( 7")" =0 /"In = 0 can be derived from the relations above.) The author and [ has been unable to go further with this. 2. Let -/(n) denote the subspace of .9* PROOFOF PROPOSITION spanned by the elements Qo'o ... QnEnC /Jrl."r, which satisfy the specified restrictions. We will first show that ,^/(n) is a subalgebra. Consider the / / I ) , 170 JOHN M.LNOR product where both factors belong to 2 ( n ) . Suppose that some term b(X)Ptltz"' on the right does not belong to ,W(n). Then t, must be 2 pn+l-zfor some I . If xZ0,x ~ - ~..., , ,x,, , were all < pn+'-5 then the factor t, ! ' " ' X0 1 . X,,! would be congruent to zero modulo p. Therefore x,, 2 pn+'-zfor some i j = I. If i > 0 this implies that + which contradicts the hypothesis that 0, j = I , then 9"l""n E ,?/(n). Similarly if i = which is also a contradiction. Since it is easily verified that ,W(n)Qk c ,W(n) for k 5 n, this proves Since ,-/(n) contains the generators that ,Y(n) is a subalgebra of 9". of P * ( n ) , this implies that ~ / ( n )19 " ( n ) . To complete the proof we must show that every element of &(n) beis the union of the 9 * ( n ) longs to 9 " ( n ) . Adem's assertion that 9" with k < dim ( 9 g " ) automatically beimplies that every element of 9'" longs to 9 * ( n ) . In particular we have: ~ ( n ) belongs to 9 * ( n ) . Case 1. Every element P O . . . O ~ ? S ,r,) lexicographically from the right, t h e Ordering the indices (r,, product formulas can be written as 9r l . . T n psl..'Sn = ( r , ,s,) (r,, s , ) 9 ri+S1," ' 2 r n + s n (higher terms) . .. + Given 9 ' t i " ' t n E -'J(n) assume by induction that (1) every 9 J r 1 " " m e a/(%) of smaller dimension belongs to 9 * ( n ) , and (2) every "higher " . 9 ' l " " n e s / ( n ) in the same dimension belongs to 9"(n). We will prove that 9' ti"'tn E 9"*(n). t,) = (0 Ot,O 0) where t, is not a power of p. Case 2. (t, Choose r , , s, > 0 with r, + s, = t,, ( r , , s,) g 0. Then 9 ' 0 " " i 9 0 " ' ~ = (r, , s , ) ~ O " ' ~ L(higher terms). Case 3. Both t, and t, are positive, i < j. Then 9 3 tl.-tLcLy~o...ot ~ + l " ' ~ n= 9tl"'tn + (higher terms) . . .. . + In either case the inductive hypothesis shows that gJtl"'tn belongs to Q, belong to 9 * ( n ) by Corollary 3, this completes 9 * ( n ) . Since Q,, .-a, T H E S T E E N R O D ALGEBRA A N D I T S D U A L 171 the proof of Proposition 2. Appendix 1. The case p = 2 All the results in this paper apply to the case p = 2 after some minor changes. The cohomology ring of the projective space . J " 'is~ a truncated polynomial ring with one generator a of dimension 1. I t turns out that P(a)e H"(P"', 2%) @ .Y" has the form where C,, = 1 and where each Ci is a well defined element of 9 , 6 - , . The algebra 9% is a polynomial algebra generated by the elements TI, C,, . . Corresponding to the basis {5:1C;2 there is a dual basis .) for tiply to the same for{SqR)for Y " .These elements S ~ ~ 1 ~ ~ . . . m u laccording The other results of this paper generalize in an obvious mula as the yJH. way. Appendix 2. Sign conventions The standard convention seems to be that no signs are inserted in formulas 1 , 2, 3 of $ 2 . If this usage is followed then the definition of (1" becomes more difficult. However Lemmas 2 and 3 still hold as stated, and Lemma 4 holds in the following modified form. LEMMA4'. If i"(a)= a , @ w i then for any 0 E 9": where d = dim 0. It is now necessary to define r, E 92,t-, by the equation Otherwise there are no changes in the results stated. 1. J. A D E MT, h e relations on Steenrod powers of cohomology classes, Algebraic geometry and topology, Princeton University Press, 1957, 191-238. 2. H. CARTAN,S u r l'ithration des opdratwns d e s t e e n r o d , Comment. Math. Helv., 29 (1955), 40-58. 3. J . M I L N O Rand J . MOORE, On the structure of Hopf algebras, to appear. 4. N. S T E N R O D ,Cyclic reduced pozc9ersof cohomolog~classes, Proc. Nat. Acad. Sci. U.S.A., 39 (1953), 217-223. 5. R. T H O M ,Quelques proprihte's globales des varihte's difl&erztinhles, Comment. Math. Helv., 28 (1954), 17-86.
© Copyright 2026 Paperzz