On Cartan’s Examples of Isoparametric Hypersurfaces and Their Focal Submanifolds Thomas E. Cecil, Patrick J. Ryan Abstract This paper is a survey of Cartan’s examples of isoparametric hypersurfaces in spheres and their focal submanifolds that were described in his fundamental work on the subject, which appeared in four papers [ 2 ] – [ 5 ] published during the period 1938–1940. 1 Isoparametric Hypersurfaces We begin with some preliminary notation and definitions, which generally follow those in the book [ 10 ] . By a real space form of dimension n , we mean a complete, connected, simply connected manifold \widetilde{M}^{n}(c) with constant sectional curvature c . If c=0 , then \widetilde{M}^{n}(c) is the n -dimensional Euclidean space {\bf R}^{n} ; if c=1 , then \widetilde{M}^{n}(c) is the unit sphere S^{n}\subset{\bf R}^{n+1} ; and if c=-1 , then \widetilde{M}^{n}(c) is the n -dimensional real hyperbolic space H^{n} (see, for example, [ 32 , Vol. I, pp. 204–209] ). For any value of c>0 , this subject is basically the same as for the case c=1 , and for any value of c<0 , it is very similar to the case c=-1 . So we will restrict our attention to the cases c=0,1,-1 . The original definition of a family of isoparametric hypersurfaces in a real space form \widetilde{M}^{n+1} was formulated in terms of the level sets of an isoparametric function, as we now describe. Let F:\widetilde{M}^{n+1}\rightarrow{\bf R} be a non-constant smooth function. The classical Beltrami differential parameters of F are defined by \Delta_{1}F=|{\rm grad}\ F|^{2},\quad\Delta_{2}F=\Delta F\ ({\rm Laplacian}\ {\rm of}\ F), (1) where grad F denotes the gradient vector field of F . The function F is said to be isoparametric if there exist smooth functions \phi_{1} and \phi_{2} from {\bf R} to {\bf R} such that \Delta_{1}F=\phi_{1}\circ F,\quad\Delta_{2}F=\phi_{2}\circ F. (2) That is, both of the Beltrami differential parameters are constant on each level set of F . This is the origin of the term isoparametric . The collection of level sets of an isoparametric function is called an isoparametric family of sets in \widetilde{M}^{n+1} . An isoparametric family in {\bf R}^{n+1} consists of either parallel planes, concentric spheres, or coaxial spherical cylinders, and their focal sets. This was first shown for n=2 by Somigliana [ 51 ] (see also B. Segre [ 47 ] and Levi-Civita [ 34 ] ), and for arbitrary n by B. Segre [ 48 ] . In the late 1930’s, shortly after the publication of the papers of Levi-Civita and Segre, Cartan [ 2 ] – [ 5 ] began a study of isoparametric families in arbitrary real space forms. In Section 2 of his first paper, Cartan [ 2 ] showed that an isoparametric family in a real space form \widetilde{M} , defined as the collection of level sets of an isoparametric function F on \widetilde{M} , is locally equivalent to a family of parallel hypersurfaces in \widetilde{M} , each of which has constant principal curvatures, together with their focal submanifolds (see [ 10 , pp. 86–91] for the details of this equivalence). Therefore, in the case where the ambient space is a real space form \widetilde{M} , we will say that a connected hypersurface M immersed in \widetilde{M} is an isoparametric hypersurface , if it has constant principal curvatures. Remark 1.1 . Isoparametric hypersurfaces in Riemannian manifolds The definition of an isoparametric hypersurface in the paragraph above is not the appropriate formulation if \widetilde{M} is only assumed to be a Riemannian manifold, as can be seen by examples in complex projective space due to Q.-M. Wang [ 57 ] (see also Thorbergsson [ 55 ] and [ 10 , pp. 526–530] ). Rather the appropriate generalization is the following: a hypersurface M in a Riemannian manifold \widetilde{M} is isoparametric if and only if all nearby parallel hypersurfaces M_{t} have constant mean curvature. For example, when \widetilde{M} is complex n -dimensional projective space {\bf CP}^{n} or complex n -dimensional hyperbolic space {\bf CH}^{n} , a (real) isoparametric hypersurface may have non-constant principal curvatures. Nevertheless, making use of the Hopf map \pi:S^{2n+1}\rightarrow{\bf CP}^{n} (see [ 10 , p. 346] ), one can show that an isoparametric hypersurface in {\bf CP}^{n} lifts to an isoparametric hypersurface in S^{2n+1} . This facilitates the classification of isoparametric hypersurfaces in {\bf CP}^{n} . In fact, similar methods can be used to classify isoparametric hypersurfaces in {\bf CH}^{n} , although in that case one must deal with Lorentzian isoparametric hypersufaces in the anti-de Sitter space. For further details about isoparametric hypersurfaces in complex space forms, see Q.-M. Wang [ 57 ] , G. Thorbergsson [ 56 ] , L. Xiao [ 58 ] , M. Dominguez-Vazquez et al. [ 19 ] – [ 21 ] , and the book [ 10 , 343–531] . 2 Cartan’s Results In Section 3 of his first paper on isoparametric hypersurfaces, Cartan [ 2 ] derived what he called a “fundamental formula,” which is the basis for many of his results on the subject. The formula involves the distinct principal curvatures \lambda_{1},\ldots,\lambda_{g} , and their respective multiplicities m_{1},\ldots,m_{g} , of an isoparametric hypersurface f:M^{n}\rightarrow\widetilde{M}^{n+1}(c) in a space form of constant sectional curvature c . If the number g of distinct principal curvatures is greater than one, Cartan showed that for each i , 1\leq i\leq g , the following equation holds, \sum_{j\neq i}m_{j}\frac{c+\lambda_{i}\lambda_{j}}{\lambda_{i}-\lambda_{j}}=0. (3) Cartan’s original formulation of the formula was given in a more complicated form, but he gave the formulation in equation ( 3 ) in his third paper [ 4 , p. 1484] . Equation ( 3 ) is now known as “Cartan’s formula.” Another proof of this formula is given in [ 10 , pp. 91–96] . Using his formula, Cartan was able to classify isoparametric hypersurfaces in {\bf R}^{n+1} and H^{n+1} . In both cases, one can easily prove using Cartan’s formula that the number g of distinct principal curvatures must be either 1 or 2. Then an elementary argument shows that a connected isoparametric hypersurface is either totally umbilic. i.e., g=1 , or in the case g=2 , it is a an open subset of a tube of constant radius over a totally geodesic submanifold of codimension greater than one (see, for example, [ 10 , pp. 96–98] ). In the sphere S^{n+1} , however, Cartan’s formula does not lead to the conclusion that g\leq 2 , and in fact Cartan produced examples with g=1,2,3 and 4 distinct principal curvatures. Moreover, he classified isoparametric hypersurfaces M^{n}\subset S^{n+1} with g\leq 3 , as we will discuss in Section 3 below. Cartan [ 2 , pp. 186–187] gave a method for determining the values of the principal curvatures \lambda_{i} of an isoparametric hypersurface M^{n} , if one knows the value of g and the multiplicities m_{1},\ldots,m_{g} , as in equation ( 4 ) below, \lambda_{i}=\cot\theta_{i},\ 0<\theta_{i}<\pi,\ 1\leq i\leq g, (4) where the \theta_{i} form an increasing sequence, and \lambda_{i} has multiplicity m_{i} on M . Cartan’s method is based on an analytic calculation involving Cartan’s formula ( 3 ). In the case where all of the principal curvatures have the same multiplicity m=n/g , Cartan’s method [ 4 , pp. 34–35] yields the formula ( 5 ) for the principal curvatures given in Theorem 2.1 below. This formula implies that for any point x\in M , there are 2g focal points of (M,x) along the normal geodesic to M through x , and they are evenly distributed at intervals of length \pi/g . Theorem 2.1 (Cartan) . Let M\subset S^{n+1} be a connected isoparametric hypersurface with g principal curvatures \lambda_{i}=\cot\theta_{i} , 0<\theta_{1}<\cdots<\theta_{g}<\pi , all having the same multiplicity m=n/g . Then \theta_{i}=\theta_{1}+(i-1)\frac{\pi}{g},\quad 1\leq i\leq g. (5) For any point x\in M , there are 2g focal points of (M,x) along the normal geodesic to M through x , and they are evenly distributed at intervals of length \pi/g . Remark 2.1 . Münzner [ 39 ] later showed that formula ( 5 ) is valid for all isoparametric hypersurfaces in S^{n+1} , even those where the multiplicities are not all the same. Münzner’s method is different than the method of Cartan. Münzner’s proof is based on the fact that the set of focal points along a normal geodesic circle to M\subset S^{n+1} is invariant under the dihedral group D_{g} of order 2g that acts on the normal circle and is generated by reflections in the focal points. Münzner’s proof also yields the result that the multiplicities satisfy m_{i}=m_{i+2} (subscripts mod g ) (see, for example, [ 10 , p. 108–109] ). Thus there are at most two distinct multiplicities, m_{1},m_{2} . Using a lengthy calculation involving equation ( 5 ), Cartan [ 3 , pp. 364–367] proved the important result that any isoparametric family with g distinct principal curvatures all having the same multiplicity is algebraic in sense of Theorem 2.2 below. Recall that a function F:{\bf R}^{n+2}\rightarrow{\bf R} is homogeneous of degree g , if g is a positive integer such that F(tx)=t^{g}F(x) , for all t\in{\bf R} and x\in{\bf R}^{n+2} . Theorem 2.2 (Cartan) . Let M\subset S^{n+1}\subset{\bf R}^{n+2} be a connected isoparametric hypersurface with g principal curvatures \lambda_{i}=\cot\theta_{i} , 0<\theta_{1}<\cdots<\theta_{g}<\pi , all having the same multiplicity m . Then M is an open subset of a level set of the restriction to S^{n+1} of a homogeneous polynomial F on {\bf R}^{n+2} of degree g satisfying the differential equations, \Delta_{1}F=|{\mbox{\rm grad }}F|^{2}=g^{2}r^{2g-2}, (6) \Delta_{2}F=\Delta F=0\ (F\ {\mbox{\it is\ harmonic}}), (7) where r=|x| , for x\in{\bf R}^{n+2} . Cartan’s Theorem 2.2 was a forerunner of Münzner’s general result (see Theorem 2.3 below) that every isoparametric hypersurface with g distinct principal curvatures is algebraic (regardless of the multiplicities of the principal curvatures), and its defining homogeneous polynomial F of degree g satisfies certain conditions on \Delta_{1}F and \Delta_{2}F , which generalize the conditions that Cartan found in Theorem 2.2 . (See also the book [ 10 , pp. 111–130] for a discussion of Münzner’s theorem.) Theorem 2.3 (Münzner) . Let M\subset S^{n+1}\subset{\bf R}^{n+2} be a connected isoparametric hypersurface with g principal curvatures \lambda_{i}=\cot\theta_{i} , 0<\theta_{1}<\cdots<\theta_{g}<\pi , with respective multiplicities m_{i} . Then M is an open subset of a level set of the restriction to S^{n+1} of a homogeneous polynomial F on {\bf R}^{n+2} of degree g satisfying the differential equations, \Delta_{1}F=|{\rm grad}\ F|^{2}=g^{2}r^{2g-2}, (8) \Delta_{2}F=\Delta F=cr^{g-2}, (9) where r=|x| , and c=g^{2}(m_{2}-m_{1})/2 . Münzner [ 39 , p. 65] called F the Cartan polynomial of M . Now F is usually referred to as the Cartan-Münzner polynomial of M . Equations ( 8 )–( 9 ) are called the Cartan-Münzner differential equations . The following theorem (see [ 10 , pp. 112–113] ) relates the gradient and Laplacian of the restriction V of F to S^{n+1} to the gradient and Laplacian of F:{\bf R}^{n+2}\rightarrow{\bf R} itself. Theorem 2.4 . Let F:{\bf R}^{n+2}\rightarrow{\bf R} be a homogeneous function of degree g , and let V be the restriction of F to S^{n+1} . Then {\rm(a)} |{\rm grad}^{S}V|^{2}=|{\rm grad}^{E}F|^{2}-g^{2}F^{2} , {\rm(b)} \Delta^{S}V=\Delta^{E}F-g(g-1)F-g(n+1)F , where {\rm grad}^{S}V and \Delta^{S}V denote the gradient and Laplacian of V as a real-valued function on the sphere S^{n+1} , and {\rm grad}^{E}F and \Delta^{E}F denote the gradient and Laplacian of F as a real-valued function on the space {\bf R}^{n+2} . Using this theorem, one can compute that the restriction V of F to S^{n+1} satisfies the differential equations, |{\rm grad}^{S}V|^{2}=g^{2}(1-V^{2}), (10) \Delta^{S}V=c-g(n+g)V, (11) where c=g^{2}(m_{2}-m_{1})/2 . Thus, if F is a Cartan-Münzner polynomial on {\bf R}^{n+2} , then its restriction V to S^{n+1} is an isoparametric function on S^{n+1} in the sense of Cartan (as in equation ( 2 )), since both |{\rm grad}^{S}V|^{2} and \Delta^{S}V are functions of V itself. Note that Cartan [ 3 , pp. 364–365] proved that if the principal curvatures all have the same multiplicity m , then the polynomial F must be harmonic. This agrees with Münzner’s condition ( 9 ) in the case where all of the multiplicities are equal, since then the number c=g^{2}(m_{2}-m_{1})/2 in equation ( 9 ) equals zero, and so F is harmonic. 3 Cartan’s Examples We now begin to describe Cartan’s examples of isoparametric hypersurfaces M\subset S^{n+1}\subset{\bf R}^{n+2} according to the number g of distinct principal curvatures. We will describe each example from various points of view, following the presentations in our books [ 9 , pp. 294–304] and [ 10 , pp. 144–159] very closely. Cartan produced examples with g=1,2,3 and 4 distinct principal curvatures, and he classified isoparametric hypersurfaces with g\leq 3 principal curvatures. In his examples with g=3 or g=4 principal curvatures, the principal curvatures all have the same multiplicity. Cartan noted that all of his examples are homogeneous, each being an orbit of a point under an appropriate closed subgroup of SO(n+2) . Based on his results and the properties of his examples, Cartan [ 3 ] asked the following three questions, all of which were answered in the 1970’s, as we will describe below. Cartan’s questions 1. For each positive integer g , does there exist an isoparametric family with g distinct principal curvatures of the same multiplicity? 2. Does there exist an isoparametric family of hypersurfaces with more than three distinct principal curvatures such that the principal curvatures do not all have the same multiplicity? 3. Does every isoparametric family of hypersurfaces admit a transitive group of isometries? In the early 1970’s, Nomizu [ 41 ] – [ 42 ] wrote two papers describing the highlights of Cartan’s work. He also generalized Cartan’s example with four principal curvatures of multiplicity one to produce examples (described in Subsection 3.4 below) with four principal curvatures having multiplicities m_{1}=m_{3}=m , and m_{2}=m_{4}=1 , for any positive integer m . This answered Cartan’s Question 2 in the affirmative. Nomizu also proved that every focal submanifold of every isoparametric hypersurface is a minimal submanifold of S^{n+1} . This also follows from Münzner’s work [ 39 ] – [ 40 ] , and Münzner’s proof is different than that of Nomizu. Remark 3.1 . In 1972, Takagi and Takahashi [ 54 ] gave a complete classification of all homogeneous isoparametric hypersurfaces in S^{n+1} , based on the work of Hsiang and Lawson [ 28 ] . Takagi and Takahashi showed that each homogeneous isoparametric hypersurface in S^{n+1} is a principal orbit of the isotropy representation of a Riemannian symmetric space of rank 2, and they gave a complete list of examples [ 54 , p.480] . This list contains examples with g=6 principal curvatures as well as those with g=1,2,3,4 principal curvatures. In some cases with g=4 , the principal curvatures do not all have the same multiplicity, so this also provided an affirmative answer to Cartan’s Question 2. At about the same time as the papers of Nomizu and Takagi-Takahashi, Münzner published two preprints that greatly extended Cartan’s work and have served as the basis for much of the research in the field since that time. The preprints were eventually published as papers [ 39 ] – [ 40 ] in 1980–1981. Of course, one of Münzner’s primary results is that the number g of distinct principal curvatures of an isoparametric hypersurface in a sphere equals 1,2,3,4 or 6, and thus the answer to Cartan’s Question 1 is negative. Finally, the answer to Cartan’s Question 3 is also negative, as was first shown by the construction of inhomogeneous isoparametric hypersurfaces with g=4 principal curvatures by Ozeki and Takeuchi [ 43 ] – [ 44 ] in 1975. Their construction was then generalized in 1981 to yield even more inhomogeneous examples by Ferus, Karcher and Münzner [ 24 ] . We now discuss Cartan’s examples in detail according to the value of g . 3.1 g=1 If a connected isoparametric hypersurface M^{n}\subset S^{n+1}\subset{\bf R}^{n+2} has g=1 principal curvature, then M^{n} is a totally umbilic hypersurface in S^{n+1} . As such, M^{n} is an open subset of a great or small hypersphere in S^{n+1} . Each great or small hypersphere is the set where S^{n+1} intersects a hyperplane in {\bf R}^{n+2} . The spheres lying in hyperplanes perpendicular to a given diameter of S^{n+1} make up an isoparametric family whose focal set consists of the two end-points (poles) of the diameter. We now describe these hypersurfaces in terms of the general theory of isoparametric hypersurfaces that we have developed. Let p be a unit vector in {\bf R}^{n+2} which we will use to determine a diameter of S^{n+1} . Let F be the linear height function, F(z)=\langle z,p\rangle, (12) for z\in{\bf R}^{n+2} , where \langle\ ,\ \rangle is the usual inner product on {\bf R}^{n+2} . Then one easily computes that {\rm grad}^{E}F=p,\quad|{\rm grad}^{E}F|^{2}=1,\quad\Delta^{E}F=0, so that F satisfies the Cartan-Münzner differential equations with g=1 and c=0 , and so F is a Cartan-Münzner polynomial. A calculation using Theorem 2.4 shows that the restriction V of F to S^{n+1} satisfies |{\rm grad}^{S}V|^{2}=1-V^{2},\quad\Delta^{S}V=-(n+1)V, (13) so that V is an isoparametric function on S^{n+1} . In fact, it is useful to note that at any z\in S^{n+1} , we have {\rm grad}^{S}V=p-\langle z,p\rangle z. (14) We consider the level sets M_{s}=\{z\in S^{n+1}\mid\langle z,p\rangle=\cos s\},\quad 0\leq s\leq\pi. Except for the two focal submanifolds, M_{0}=\{p\} and M_{\pi}=\{-p\} , each level set is an n -sphere with radius \sin s lying in the hyperplane situated 1-\cos s units below the north pole p . From the point of view of the intrinsic geometry of S^{n+1} , the set M_{s} is the geodesic hypersphere S(p,s) centered at the point p having radius s . Note that M_{s} may also be regarded as the sphere S(-p,\pi-s) . When s=\pi/2 , the great sphere M_{s} is the unique minimal hypersurface in the isoparametric family. Finally, the family M_{s} can also be realized as the set of orbits of a group action as follows. Let G_{p} be the subgroup of SO(n+2) which leaves the pole p fixed. The group G_{p} is a naturally embedded copy of SO(n+1) , so that the orbit G_{p}z has codimension one in S^{n+1} whenever z\neq\pm p . One easily shows that G_{p}z is a geodesic sphere through z with centers \pm p . 3.2 g=2 Cartan [ 2 ] showed that an isoparametric hypersurface in the sphere with two distinct principal curvatures must be a standard product of two spheres, as we now describe. We consider {\bf R}^{n+2}={\bf R}^{p+1}\times{\bf R}^{q+1} , where p+q=n . For z=(x,y) in {\bf R}^{p+1}\times{\bf R}^{q+1} , let F(z)=|x|^{2}-|y|^{2}. (15) Then we compute {\rm grad}^{E}F=2(x,-y),\quad|{\rm grad}^{E}F|^{2}=4r^{2},\quad\Delta^{E}F=2(p-q), (16) so that F satisfies the Cartan-Münzner differential equations ( 8 )–( 9 ) with g=2 and c=2(p-q) , and F is a Cartan-Münzner polynomial. Note that since c=g^{2}(m_{2}-m_{1})/2=2(m_{2}-m_{1}) , we have m_{2}-m_{1}=p-q . Since we also have m_{1}+m_{2}=p+q=n , we conclude that m_{1}=q,\quad m_{2}=p. (17) The restriction V of F to S^{n+1} satisfies \displaystyle{\rm grad}^{S}V \displaystyle= \displaystyle 2((1-V)x,-(1+V)y),\quad|{\rm grad}^{S}V|^{2}=4(1-V^{2}), (18) \displaystyle\Delta^{S}V \displaystyle= \displaystyle 2(p-q)-2(n+2)V, so that V is an isoparametric function on S^{n+1} . 3.2.1 Level Sets of F in S^{n+1} Consider the level set of F in S^{n+1} , M_{s}=\{z\in S^{n+1}\mid F(z)=\cos 2s\},\quad 0\leq s\leq\frac{\pi}{2}. Each M_{s} is the Cartesian product of a p -sphere of Euclidean radius \cos s and a q -sphere of radius \sin s , except for the two focal submanifolds, M_{0}=\{(x,y)\mid y=0\}=S^{p}\times\{0\},\quad M_{\frac{\pi}{2}}=\{(x,y)\mid x=0\}=\{0\}\times S^{q}. The M_{s} form a family of parallel hypersurfaces, all of which are tubes over each of the focal submanifolds. Since the focal submanifolds are totally geodesic, it is easy to compute the shape operator of M_{s} using the formula for the shape operator of a tube (see, for example, [ 10 , pp. 17–18] ). By considering M_{s} as a tube of radius s over the totally geodesic submanifold S^{p}\times\{0\} , we get that M_{s} has two constant principal curvatures, \lambda=\cot\left(\frac{\pi}{2}-s\right)=\tan s,\quad\mu=-\cot s, with respective multiplicities p , q . The family \{M_{s}\} can also be realized as the set of orbits of a group action. Considering the decomposition {\bf R}^{n+2}={\bf R}^{p+1}\times{\bf R}^{q+1} , the group G=SO(p+1)\times SO(q+1) is naturally embedded as a subgroup of SO(n+2) . Take any point z=(x,y) with both x and y non-zero. Then the isotropy subgroup of the G -action at z is isomorphic to SO(p)\times SO(q) , so that the orbit M of z has codimension 2 in {\bf R}^{n+2} , and hence codimension 1 in S^{n+1} . Considering the action of G on each factor, it is clear that the orbit consists of the product of standard spheres of radii |x| and |y| , and hence it coincides with M_{s} , where \cos s=|x| and \sin s=|y| . Furthermore, the orbit of (x,0) is the focal submanifold M_{0} , and the orbit of (0,y) is the focal submanifold M_{\pi/2} . 3.3 g=3 In his paper [ 3 ] , Cartan gives a complete classification of isoparametric hypersurfaces in S^{n+1} with g=3 principal curvatures, and he describes each example in great detail. Let M^{n}\subset S^{n+1}\subset{\bf R}^{n+2} be a connected, oriented hypersurface with field of unit normals \xi having g=3 distinct principal curvatures at every point. The first step in Cartan’s classification theorem is to prove the following theorem [ 3 , pp. 359–360] . Theorem 3.1 . Let M^{n}\subset S^{n+1} be a connected isoparametric hypersurface with g=3 principal curvatures at each point. Then all of the principal curvatures have the same multiplicity m . Cartan proves this using the method of moving frames and conditions on the structural formulas that follow from the isoparametric condition and the assumption that g=3 , and we refer the reader to Cartan’s paper [ 3 , pp. 359–360] for the proof. Next, since all of the multiplicities are equal, Cartan’s Theorem 2.2 mentioned above implies that M^{n} is an open subset of a level set of the restriction to S^{n+1} of a harmonic homogeneous polynomial F on {\bf R}^{n+2} of degree g=3 satisfying the differential equations stated in Theorem 2.2 . Cartan’s approach to classifying isoparametric hypersurfaces with g=3 is to try to directly determine all homogeneous harmonic polynomials F of degree g=3 that satisfy the differential equation from Theorem 2.2 , \Delta_{1}F=|{\mbox{\rm grad }}F|^{2}=g^{2}r^{2g-2}=g^{2}(x_{1}^{2}+x_{2}^{2}+\cdots+x_{n+2}^{2})^{{g-1}}, (19) where x=(x_{1},x_{2},\ldots,x_{n+2})\in{\bf R}^{n+2} . After a long calculation, Cartan shows that the algebraic determination of such a harmonic polynomial requires the possibility of solving the algebraic problem: Problem 1: Represent the product (u_{1}^{2}+u_{2}^{2}+\cdots+u_{m}^{2})(v_{1}^{2}+v_{2}^{2}+\cdots+v_{m}^{2}) of two sums of m squares by the sum of the squares of m bilinear combinations of the u_{i} and the v_{j} . By a theorem of Hurwitz [ 29 ] on normed linear algebras, this is only possible if m=1,2,4, or 8. Cartan [ 3 , pp. 340–341] relates the solution of Problem 1 above to the theory of Riemannian spaces admitting an isogonal absolute parallelism, and he uses results from a joint paper of himself and J. A. Schouten [ 6 ] on that topic to determine the possibilities for F . Ultimately, Cartan [ 3 ] shows that the required harmonic homogeneous polynomial F of degree 3 on {\bf R}^{3m+2} must be of the form F(x,y,X,Y,Z) given by x^{3}-3xy^{2}+\frac{3}{2}x(X\overline{X}+Y\overline{Y}-2Z\overline{Z})+\frac{3\sqrt{3}}{2}y(X\overline{X}-Y\overline{Y})+\frac{3\sqrt{3}}{2}(XYZ+\overline{Z}\overline{Y}\overline{X}). (20) In this formula, x and y are real parameters, while X,Y,Z are coordinates in the division algebra {\bf F}={\bf R},{\bf C},{\bf H} (quaternions), {\bf O} (Cayley numbers), for m=1,2,4,8 , respectively. Note that the sum XYZ+\overline{Z}\overline{Y}\overline{X} is twice the real part of the product XYZ . In the case of the Cayley numbers, multiplication is not associative, but the real part of XYZ is the same whether one interprets the product as (XY)Z or X(YZ) . The isoparametric hypersurfaces in the family are the level sets M_{t} in S^{3m+1} determined by the equation F=\cos 3t , 0