Vector Fields on Spheres

Lars Hesselholt

The unit sphere in Euclidean n-space is the subset S n1 Rn of all vectors x of norm 1. The tangent space to S n1 at x is the hyperplane Tx (S n1 ) Rn of all vectors v Rn that are perpendicular to x. A continuous tangent vector eld on the sphere S n1 is dened to be a continuous function X : S n1 Rn such that X(x) Tx (S n1 ), for all x S n1 . The vector eld problem asks for the maximal number k (n) of continuous vector elds X1 , . . . , Xk on S n1 such that the vectors X1 (x), . . . , Xk (x) are linearly independent, for all x S n1 . We note that it is equivalent to ask that the vectors X1 (x), . . . , Xk (x) form an orthonormal frame, for all x S n1 . To see this, we recall that the Gram-Schmidt process replaces the linearly independent vectors X1 (x), . . . , Xk (x) by orthonormal vectors X1 (x), . . . , Xk (x) that span the same subspace of Rn . Moreover, this process is continuous, and therefore, the maps X1 , . . . , Xk : S n1 Rn dened in this way are again continuous vector elds on S n1 . One possible way to construct a vector eld on S n1 is as follows. Let A be an n n matrix. Then the function X : S n1 Rn dened by X(x) = Ax is a tangent vector eld if and only if the inner product x, Ax = 0, for all x S n1 . This, in turn, is equivalent to the requirement that A be skew symmetric, that is, A + At = 0, where At is the transpose of the matrix A. Indeed, suppose rst that x, Ax = 0, for all x S n1 , or equivalently, for all x Rn . Then x, (A + At )y = x, Ay + Ax, y = x, Ax + x, Ay + Ax, y + Ay, y = x + y, A(x + y ) = 0, for all x, y Rn , and hence, A + At = 0. Conversely, if A + At = 0, then
1 x, Ax = 2 ( x, Ax + Ax, x ) t =1 2 ( x, Ax + x, A x ) = 1 2

x, (A + At )x = 0.

We will say that the vector eld X obtained in this way is a linear vector eld. Let X1 , . . . , Xk : S n1 Rn be linear vectors elds corresponding to the skew symmetric n n matrices A1 , . . . , Ak . Then the vectors X1 (x), . . . , Xk (x) form t an orthonormal frame, for all x S n1 if and only if At i Aj + Aj Ai = 0, for t i k . Since the matrices Ai all 1 i < j k , and Ai Ai = I , for all 1
1

are skew symmetric, these requirements are equivalent to the requirements that Ai Aj + Aj Ai = 0, for all 1 i < j k and A2 i k . Here are i = I , for all 1 some examples: If n = 2, the skew symmetric matrix A= 0 1 1 0 eld. If n = 4, 0 0 1 0 1 0 1 0 0 0 0 0

satises A2 = I which shows that S 1 has one linear unit vector the three skew symmetric matrices 0 0 0 1 0 0 1 0 0 0 0 0 1 0 0 0 0 1 , A3 = A1 = 0 0 0 0 1 , A2 = 1 0 0 0 1 0 1 0 0 0 0 1 0

satises A2 = I which shows that S 5 has one linear unit vector eld. The following result was proved independently by Hurwitz [5] and Radon [7] around 1923; see also Eckmann [4]. Theorem A. Let n be a positive integer and write n = 24+ u, where u is odd, 0, and 0 < 4. Then the maximal number of orthonormal linear vector elds on S n1 is equal to l(n) = 8 + 2 1.

2 2 satisfy A2 1 = A2 = A3 = I and A1 A2 + A2 A1 = A1 A3 + A3 A1 = A2 A3 + A3 A2 = 0 3 which shows that S has three orthonormal linear vector elds. If n = 6, the matrix 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 A= 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0

The theorem of Hurwitz and Radon determines the maximal number l(n) of orthogonal linear vector elds on S n1 . However, the maximal number k (n) of orthogonal continuous vector elds on S n1 could possibly be larger. It was proved by Adams in 1962 that, in fact, k (n) = l(n). To explain how one may prove such a thing, we rst reformulate the problem. Let p n be positive integers. The Stiefel manifold Vn,p is dened to be the set of all p-tuples (x1 , . . . , xp ) of orthonormal vectors in Rn . Let xi,s be the sth coordinate of the vector xi . Then Vn,p (Rn )p = Rnp is equal to the set of solutions to the p(p + 1)/2 equations
n

xi,s xj,s = i,j

s=1

(1

p).

The implicit function theorem shows that, locally, we can express p(p + 1)/2 of the np coordinates xi,s , 1 i p, 1 s n, as smooth functions of the remaining coordinates. This shows that Vn,p is a smooth manifold of dimension np p(p +1)/2. Hence, locally, Vn,p is dieomorphic to Euclidean np p(p + 1)/2 space. However, globally, Vn,p has a rich topology. For example, Vn,1 is the unit sphere S n1 , Vn,n = O(n) is the Lie group of orthogonal n n matrices, and Vn,n1 = SO(n) is
2

the closed subgroup of orthogonal n n matrices whose determinant is equal to 1. Now, there is a continuous projection map : Vn,p Vn,1 = S n1 that takes the p-frame x1 , . . . , xp to the last vector xp . Suppose that X1 , . . . , Xp1 are orthonormal continuous vector elds on S n1 . Then the map : S n1 Vn,p dened by (x) = (X1 (x), . . . , Xp1 (x), x) is continuous and the composite map S n1 Vn,p S n1 is equal to the identity map idS n1 . Conversely, if : S n1 Vn,p is continuous and = idS n1 , then the maps X1 , . . . , Xp1 : S n1 Rn dened by the formula (x) = (X1 (x), . . . , Xp1 (x), x) are continuous orthonormal vector elds on S n1 . Hence, we wish to prove that if p l(n) + 2, then there does not exists a continuous map : S n1 Vn,p such that = idS n1 . The method of algebraic topology is to construct an image in algebra of our problem in topology. Here is one such image. Let M be a smooth manifold such as Vn,p . Then we have the notion of a dierential q -form on M . The dierential d of a dierential q -form on M is a dierential (q + 1)-form on M . We say that is a closed dierential q -form, if d = 0, and we say that is an exact dierential q form, if = d , for some dierential (q 1)-form . The set of all closed dierential q -forms on M forms a real vector space, and the set of all exact dierential q -forms on M forms a real subspace of this vector space. These vector spaces are both innite dimensional. But the quotient vector space
q HdR (M ) =

{closed dierential q -forms on M } {exact dierential q -forms on M }

is often a nite dimensional vector space. This is the case, for instance, if M is a q compact smooth manifold such as Vn,p . The vector space HdR (M ) is called the q th de Rham cohomology group of M . Suppose that f : N M is a smooth map from a smooth manifold N to the smooth manifold M . Then a dierential q -form on M gives rise to a dierential q -form f on N called the pull-back of by f . The pull-back f is closed, if is closed, and exact, if is exact, and therefore, we q q have a well-dened map f : HdR (N ) HdR (M ) that takes the class of to the q class of f . This map is a linear map from the real vector space HdR (N ) to the q real vector space HdR (M ). In fact, one can use the Weierstrauss approximation q q theorem to associate a linear map f : HdR (M ) HdR (N ) to every continuous map f : N M . This association has the following properties:
q (i) (idM ) = idHdR (M ) .

(ii) (f g ) = g f . q We say that HdR () is a functor smooth manifolds continuous maps

dR

H q ()

real vector spaces linear maps

from the category of smooth manifolds and continuous maps to the category of vector spaces and linear maps. We refer to Madsen and Tornehaves book [6] for a detailed introduction to dierential forms and de Rham cohomology.
3

Let n be an odd number. Then l(n) = 0 and we wish to prove that there does not exist a continuous map : S n1 Vn,2 such that the composition S n1 Vn,2 S n1 is the identity map idS n1 . So we assume that such a map exists and proceed to derive a contradiction. The maps and give rise to linear maps
q q q HdR (S n1 ) HdR (Vn,2 ) HdR (S n1 ), q and since HdR () is a functor, the composition of these two maps is the identity q map of the real vector space HdR (S n1 ). Now, one calculates q dimR HdR (S n1 ) =

1 (q = 0 or q = n 1) 0 (otherwise)

and, if n is odd,
q dimR HdR (Vn,2 ) =

1 (q = 0 or q = 2n 3) 0 (otherwise).

Hence, for q = n 1, the composite map

n1 n1 n1 HdR (S n1 ) HdR (Vn,2 ) HdR (S n1 )

is the zero map, because the real vector space in the middle is zero. But then this n1 map is not the identity map of the 1-dimensional real vector space HdR (S n1 ) which is a contradiction. We can therefore conclude that there are no continuous unit vector elds on S n1 if n is odd. Let us also consider the case n = 6. We have l(6) = 1 and wish to show that also k (6) = 1. Again, we assume that there exists a smooth map : S 5 V6,3 such that is the identity map of S 5 . However, in this case, one calculates
q dimR HdR (V6,3 ) =

1 (q = 0, 5, 7, or 12) 0 (otherwise),

so we cannot rule out that the linear maps

q q q HdR (S 5 ) HdR (V6,3 ) HdR (S 5 )

exist. Therefore, we need an invariant that more fully captures the topology of the manifold Vn,p than does de Rham cohomology. The more suttle invariant that turns out to give the solution to the problem is called topological K -theory and was introduced by Atiyah and Hirzebruch [3] based on ideas of Grothendieck. It assigns to the topological space X , a -ring KO(X ), and to the continuous map f : X Y , a -ring homomorphism f : KO(Y ) KO(X ) such that (idX ) = idKO(X ) and (f g ) = g f . Hence, KO() is a functor topological spaces continuous maps
KO ()

-rings -ring homomorphisms

from the category of topological spaces and continuous maps to the category of -rings and -ring homomorhisms. We will not give the denition of KO() here but refer to Atiyahs book [2].
4

Now, let p = l(n) + 2 and assume there exists a continuous map : S n1 Vn,p such that the composition S n1 Vn,p S n1 is the identity map of S n1 . Then the composition KO(S n1 ) KO(Vn,p ) KO(S n1 ) is also the identity map, because KO() is a functor. It is now possible as before to derive a contradiction and conclude that the map cannot exist. This was achieved by Adams [1] in 1962 who proved the following result. Theorem B. Let n be a positive integer and write n = 24+ u where u is an odd integer, and and integers with 0 and 0 < 4. Then there are at most k (n) = 8 + 2 1 linearly independent continuous vector elds on S n1 . Together the theorems of Hurwicz-Radon and Adams show that there exists exactly k (n) = l(n) linearly independent continuous vector elds on the unit sphere S n1 in Euclidean n-space. References
[1] J. F. Adams, Vector elds on spheres, Ann. of Math. 75 (1962), 603632. [2] M. F. Atiyah, K -theory. Notes by D. W. Anderson. Second edition, Advanced Book Classics, Addison-Wesley Publishing Company, Advanced Book Program, Redwood City, CA, 1989. [3] M. F. Atiyah and F. Hirzebruch, Riemann-Roch theorems for dierentiable manifolds, Bull. Amer. Math. Soc. 65 (1959), 276281. [4] B. Eckmann, Gruppentheoretischer Beweis des Satzes von Hurwitz-Radon u ber die Komposition quadratischer Formen, Comment. Math. Helv. 15 (1943), 358366. [5] A. Hurwitz, Uber die Komposition der quadratischen Formen, Math. Ann. 88 (1923), 125. [6] I. Madsen and J. Tornehave, From calculus to cohomology. De Rham cohomology and characteristic classes, Cambridge University Press, Cambridge, 1997. [7] J. Radon, Lineare Scharen orthogonaler Matrizen, Abh. Sem. Hamburg I (1923), 114.