Then he showed that the set of clans, which are LSA's with the positive definite Hessian type inner product such that the left multiplication of any element has only real eigenvalues, is in one-to-one correspondence with the set of the homogeneous convex domains. The terminology of Hessian type was introduced by Hirohiko Shima. He introduced the Hessian type inner product for the study of homogeneous Hessian manifolds[15] . Using this inner product, he developed the Hessian algebra which is the LSA with the Hessian type inner product.

We have just shown that the homogeneous affine hypersurface which is a graph of a function $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  and the automorphism group contains a solvable unimodular simply transitive subgroup gives a complete LSA with the Hessian type inner product. We will consider the converse of this in the following. We start from a complete LSA $\mathcal{A}$  with a Hessian type inner product $H$  with $|detH|=1$  . (Abusing the notation, we denote by $H$  in this paper both the inner product and its associated symmetric matrix with respect to the standard basis on ${\mathbb{R}}^n$  .) Let $\overline{\mathfrak{a}}$  be the associated Lie algebra of $\mathcal{A}$  . Define a map $\phi :\overline{\mathfrak{a}}\to aff(n+1)$  by $\phi \left(a\right)=\left(\left(\begin{array}{cc}{\lambda }_a& 0\\ a^{\prime }H& 0\end{array}\right)\left(\begin{array}{c}a\\ 0\end{array}\right)\right)$  , where $a^{\prime }$  is the transpose of $a\in \overline{\mathfrak{a}}$  . Then $\phi$  is the Lie algebra homomorphism since, for any $a,b\in \overline{\mathfrak{a}}$  ,

$$\begin{array}{ccc}\left[\phi \right(a),\phi (b\left)\right]& =& \phi \left(a\right)\phi \left(b\right)-\phi \left(b\right)\phi \left(a\right)\end{array}$$

$$\begin{array}{ccc}& =& \left(\left(\begin{array}{cc}{\lambda }_a{\lambda }_b-{\lambda }_b{\lambda }_a& 0\\ a^{\prime }H{\lambda }_b-b^{\prime }H{\lambda }_a& 0\end{array}\right)\left(\begin{array}{c}{\lambda }_{a}b-{\lambda }_{b}a\\ a^{\prime }Hb-b^{\prime }Ha\end{array}\right)\right)\end{array}$$

$$\begin{array}{ccc}& =& \left(\left(\begin{array}{cc}[{\lambda }_a,{\lambda }_b]& 0\\ (ab{)}^{\prime }H-(ba{)}^{\prime }H& 0\end{array}\right)\left(\begin{array}{c}ab-ba\\ 0\end{array}\right)\right)\end{array}$$

$$\begin{array}{ccc}& =& \left(\left(\begin{array}{cc}{\lambda }_{[a,b]}& 0\\ [a,b{]}^{\prime }H& 0\end{array}\right)\left(\begin{array}{c}[a,b]\\ 0\end{array}\right)\right)\end{array}$$

$$\begin{array}{ccc}& =& \phi \left(\right[a,b\left]\right),\end{array}$$

where the third equality follows from ( 2.7 ) and symmetry of the Hessian inner product $H$  , the forth equality is the LSA condition and the compatibility of LSA product and Lie product. Since $\mathfrak{a}=\phi \left(\overline{\mathfrak{a}}\right)$  is a Lie subalgebra of $aff(n+1)$  , we have a corresponding Lie subgroup $\mathbb{A}=exp\mathfrak{a}\subset Aff(n+1)$  whose elements are given by

$$\begin{array}{c}\left(\left(\begin{array}{cc}{e}^{{\lambda }_a}& 0\\ a^{\prime }H(I+\frac{{\lambda }_a}{2!}+\frac{({\lambda }_a{)}^2}{3!}+\cdots )& 1\end{array}\right)\left(\begin{array}{c}{e}^a-1\\ a^{\prime }H(\frac{a}{2!}+\frac{a^2}{3!}+\cdots )\end{array}\right)\right),\end{array}$$

(2.9)

for $a\in \overline{\mathfrak{a}}$  . Note that the orbit space of the Lie group $\overline{\mathbb{A}}=\left\{\right(e^{{\lambda }_a},e^a-1)\in aff(n\left)\right|a\in \mathcal{A}\}$  at the origin is the whole space ${\mathbb{R}}^n$  because $\mathcal{A}$  is complete.

Put $x=e^a-1\in {\mathbb{R}}^n$  and let $F\left(x\right)=F(e^a-1)=a^{\prime }H(\frac{a}{2!}+\frac{a^2}{3!}+\cdots )$  be the function from ${\mathbb{R}}^n$  to $\mathbb{R}$  . Then the function $F$  defines a homogeneous hypersurface $\Sigma =\left\{\right(x,F\left(x\right))\in {\mathbb{R}}^{n+1}|x\in {\mathbb{R}}^n\}$  containing the origin of ${\mathbb{R}}^{n+1}$  since $F\left(0\right)=F(e^0-1)=0$  and $\mathbb{A}$  acts simply transitively on $\Sigma$  since $\Sigma$  is the $\mathbb{A}$  -orbit of the origin. For $a,b\in \mathcal{A}$  and $u\in \mathbb{R}$  , the group multiplication in $\mathbb{A}$  shows that

$$\begin{array}{ccc}F\left(\right(e^a-1)+e^{{\lambda }_a}(e^{ub}-1\left)\right)=a^{\prime }H(\frac{a}{2!}+\frac{a^2}{3!}+\cdots )& & \end{array}$$

$$\begin{array}{ccc}& & +a^{\prime }H(I+\frac{{\lambda }_a}{2!}+\frac{({\lambda }_a{)}^2}{3!}+\cdots )(e^{ub}-1)\end{array}$$

(2.10)

$$\begin{array}{ccc}& & +\left(ub{)}^{\prime }H\right(\frac{ub}{2!}+\frac{(ub{)}^2}{3!}+\cdots ).\end{array}$$

By differentiating ( 2.10 ) with respect to $u$  at $u=0$  , we obtain

$$\begin{array}{c}dF(e^a-1)e^{{\lambda }_a}=a^{\prime }H(I+\frac{{\lambda }_a}{2!}+\frac{({\lambda }_a{)}^2}{3!}+\cdots ).\end{array}$$

(2.11)

Therefore, if we put $M_x=e^{{\lambda }_a}$  where $x=e^a-1$  , then the elements of $\mathbb{A}$  in ( 2.9 ) is equal to the matrix in Proposition  2.4 . Differentiating ( 2.11 ) at the origin, we have $$Dd{F}_0=H.$$  Now the group multiplication in $\mathbb{A}$  gives us : for $a,b\in \mathcal{A}$  and $u\in \mathbb{R}$  ,

$$\begin{array}{ccc}dF\left(\right(e^a-1)+e^{{\lambda }_a}(e^{ub}-1\left)\right)e^{{\lambda }_a}{e}^{{\lambda }_{ub}}& & \end{array}$$

$$\begin{array}{ccc}& =& dF(e^a-1)e^{{\lambda }_a}{e}^{{\lambda }_{ub}}+dF(e^{ub}-1)e^{{\lambda }_{ub}}.\end{array}$$

(2.12)

Once more, differentiating ( 2.12 ) with respect to $u$  at $u=0$  gives $$Dd{F}_x(M_x,M_x)=DdF(e^a-1)(e^{{\lambda }_a},e^{{\lambda }_a})=DdF\left(0\right)(\cdot ,\cdot )=Dd{F}_0(\cdot ,\cdot ).$$  Since $det{M}_x=1$  , we have $detDd{F}_x=detDd{F}_0$  and hence $|detDd{F}_x|=|detH|=1$  for all $x\in {\mathbb{R}}^n$  . Therefore the affine normals of the hypersurface $\Sigma$  are equal to $\xi =(0,\cdots ,0,1)$  , so they are parallel. Now we can summarize as follows.

Theorem 2.7There is a one-to-one correspondence between the set of the graph of $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  on which a unimodular Lie subgroup $\mathbb{A}$  of affine transformations acts simply transitively and the set of the complete LSA $\mathcal{A}$  with a nondegenerate Hessian type inner product $H$  with $|detH|=1$  .

3 The transitive action of abelian subgroup

With the first assumption (E1) of Eastwood and Ezhov conjecture, we will assume that the Lie group $\mathbb{A}$  is a transitive abelian subgroup of $Aut(\Sigma {)}_0$  in this section. Let's denote by $\mathbb{J}$  the isotropy subgroup of $Aut(\Sigma {)}_0$  at the origin $0\in \Sigma$  .

Proposition 3.1Let $\Sigma$  be the graph of a function $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  and let $\mathbb{A}$  be an abelian subgroup of $Aut(\Sigma {)}_0$  which acts on $\Sigma$  transitively. Then the action of $\mathbb{A}$  on $\Sigma$  is simply transitive.

Proof. Suppose that the action of $\mathbb{A}$  is not simply transitive. Then $\mathbb{A}$  contains a nonzero element $j$  of $\mathbb{J}$  and by the commuting property, we have that $jg\cdot 0=gj\cdot 0=g\cdot 0$  for all $g\in \mathbb{A}$  . It follows that $g\cdot 0$  is fixed under $j$  for any $g\in \mathbb{A}$  .

Since $\mathbb{A}$  acts transitively on $\Sigma$  , each $j\in \mathbb{A}\cap \mathbb{J}$  fixes every point of $\Sigma$  . Hence we see that $\Sigma$  is a hyperplane through $0$  in ${\mathbb{R}}^{n+1}$  because the fixed point set of a linear map $j$  is a subspace. But this contradicts the nondegeneracy of the hypersurface $\Sigma$  . $\square$ 

For the abelian LSA $\mathcal{A}$  , the condition ( 2.7 ) of Hessian type inner product is reduced to the following

$$\begin{array}{c}H(a,bc)=H(b,ac),a,b,c\in \mathcal{A}.\end{array}$$

(3.1)

Lemma 3.2Let $\Sigma$  be the graph of a function $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  and let $\mathbb{A}$  be an abelian subgroup of $Aut(\Sigma {)}_0$  which acts on $\Sigma$  simply transitively. Then any element of $\mathbb{A}$  can be written as $$g_x=\left(\left(\begin{array}{cc}I+{\lambda }_x& 0\\ x^{\prime }H& 1\end{array}\right)\left(\begin{array}{c}x\\ F\left(x\right)\end{array}\right)\right),$$  for some $x\in {\mathbb{R}}^n$  and $\lambda :{\mathbb{R}}^n\to gl\left({\mathbb{R}}^n\right)$  is a left multiplication map coming from the associated abelian LSA $\mathcal{A}$  .

Proof. For the abelian LSA $\mathcal{A}$  , we have $({\lambda }_a{)}^n={\lambda }_{a^n}$  for all $a\in \mathcal{A}$  and $n=2,3,\cdots$  . Therefore,

$$\begin{array}{ccc}{e}^{{\lambda }_a}& =& I+{\lambda }_a+\frac{1}{2!}({\lambda }_a{)}^2+\frac{1}{3!}({\lambda }_a{)}^3+\cdots \end{array}$$

$$\begin{array}{ccc}& =& I+{\lambda }_a+\frac{1}{2!}{\lambda }_{a^2}+\frac{1}{3!}{\lambda }_{a^3}+\cdots \end{array}$$

$$\begin{array}{ccc}& =& I+{\lambda }_{a+\frac{1}{2!}{a}^2+\frac{1}{3!}{a}^3+\cdots }\end{array}$$

$$\begin{array}{ccc}& =& I+{\lambda }_x.\end{array}$$

Then from the equation ( 2.11 ) and ( 3.1 ), we have

$$\begin{array}{ccc}d{F}_x(I+{\lambda }_x)& =& a^{\prime }H(I+\frac{{\lambda }_a}{2!}+\frac{({\lambda }_a{)}^2}{3!}+\cdots )\end{array}$$

$$\begin{array}{ccc}& =& (a+\frac{1}{2!}{a}^2+\frac{1}{3!}{a}^3+\cdots {)}^{\prime }H\end{array}$$

$$\begin{array}{ccc}& =& (e^a-1{)}^{\prime }H=x^{\prime }H.\end{array}$$

Then from Proposition  2.4 , the elements of $\mathbb{A}$  can be written in the form given above. $\square$ 

Lemma  3.2 shows that from the simply transitive abelian Lie subgroup we can obtain the structure of the associated abelian LSA and the Hessian type inner product directly, and vice versa.

Theorem 3.3Let $\Sigma$  be the graph of a function $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  . Suppose that $Aut(\Sigma {)}_0$  contains an abelian subgroup which acts on $\Sigma$  simply transitively. Then we have the followings :

• (a) The function $F$  is a polynomial which is given by

  $$\begin{array}{c}F\left(x\right)=x^{\prime }H(\frac{1}{2}I-\frac{1}{3}{\lambda }_x+\frac{1}{4}({\lambda }_x{)}^2-\cdots )x,\end{array}$$

  (3.2)

  where $H=Dd{F}_0$  and $\lambda$  is the left multiplication map obtained from the associated complete abelian LSA.
• (b) An abelian subgroup of $Aut(\Sigma {)}_0$  which acts simply transitively on $\Sigma$  is unique. Therefore $\mathbb{A}$  is a normal subgroup of $Aut(\Sigma {)}_0$  .

Proof. (a) Let $\mathbb{A}$  be the simply transitive abelian subgroup of $Aut(\Sigma {)}_0$  and let $\mathcal{A}$  be the associated complete abelian LSA. Let $\lambda$  be the left multiplication map obtained from the LSA $\mathcal{A}$  and let $H=Dd{F}_0$  be the Hessian type inner product on $\mathcal{A}$  . Then $H{\lambda }_x=({\lambda }_x{)}^{\prime }H$  for all $x\in {\mathbb{R}}^n$  from ( 3.1 ). Define the polynomial $G:{\mathbb{R}}^n\to \mathbb{R}$  as the right hand side of ( 3.2 ). If we differentiate $G\left(x\right)$  by using $d\left(\right({\lambda }_x{)}^n\left)\right(v)=n({\lambda }_x{)}^{n-1}{\lambda }_v$  , then

$$\begin{array}{ccc}d{G}_x& =& H(\frac{1}{2}I-\frac{1}{3}{\lambda }_x+\frac{1}{4}({\lambda }_x{)}^2-\cdots )x\end{array}$$

$$\begin{array}{ccc}& & +x^{\prime }H(\frac{1}{2}I-\frac{2}{3}{\lambda }_x+\frac{3}{4}({\lambda }_x{)}^2-\cdots )\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }H(I-{\lambda }_x+({\lambda }_x{)}^2-({\lambda }_x{)}^3+\cdots )\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }H(I+{\lambda }_x{)}^{-1}.\end{array}$$

where the second equality follows from the associativity of $H$  . On the other hand, from the proof of Lemma  3.2 , $d{F}_x=x^{\prime }H(I+{\lambda }_x{)}^{-1}$  . Hence we have $F\left(x\right)=G\left(x\right)+c$  for some constant $c$  . But $G\left(0\right)=0=F\left(0\right)$  , so we conclude that they are equal.

(b) Using the associativity of $H$  , $d{F}_x=x^{\prime }\left(\right(I+{\lambda }_x{)}^{-1}{)}^{\prime }H$  . Then, we have

$$\begin{array}{c}\left(\right(I+{\lambda }_x{)}^{-1}x{)}^{\prime }=d{F}_x{H}^{-1}.\end{array}$$

(3.3)

Because the right hand side of ( 3.3 ) depends only on the given function, we can show that the LSA structure is uniquely determined as follows : If we assume that there exists a Lie algebra homomorphism $\tilde{\lambda }:{\mathbb{R}}^n\to gl\left(n\right)$  which gives another associated abelian LSA structure, so that $(I+{\tilde{\lambda }}_x{)}^{-1}x=(I+{\lambda }_x{)}^{-1}x$  for all $x\in {\mathbb{R}}^n$  by ( 3.3 ), then we have

$$\begin{array}{c}\left[\right({\lambda }_x-{\tilde{\lambda }}_x)-(({\lambda }_x{)}^2-({\tilde{\lambda }}_x{)}^2)+(({\lambda }_x{)}^3-({\tilde{\lambda }}_x{)}^3)-\cdots ]x=0,\end{array}$$

(3.4)

for all $x\in {\mathbb{R}}^n$  . If we differentiate ( 3.4 ), then we can easily derive that

$$\begin{array}{c}2({\lambda }_x-{\tilde{\lambda }}_x)-3\left(\right({\lambda }_x{)}^2-\left({\tilde{\lambda }}_x{)}^2\right)+4\left(\right({\lambda }_x{)}^3-\left({\tilde{\lambda }}_x{)}^3\right)-\cdots =0,\end{array}$$

(3.5)

for all $x\in {\mathbb{R}}^n$  . By differentiating ( 3.5 ) at $x=0$  , we obtain $2(\lambda -\tilde{\lambda })=0$  , that is, $\lambda =\tilde{\lambda }$  .

Now (a) and Lemma  3.2 shows that the abelian subgroup $\mathbb{A}$  is also uniquely determined from the LSA structure and the Hessian type inner product. $\square$ 

Any element $j\in \tilde{\mathbb{J}}$  , the isotropy subgroup at the origin $0\in \Sigma$  , must be linear. So it is given by $j=\left(\begin{array}{cc}A& 0\\ c^{\prime }& t\end{array}\right)\in GL(n+1).$  Since $d{F}_0=0$  , the tangent space of $\Sigma$  at $0$  is ${\mathbb{R}}^n$  and it is invariant by the action of $j_{*}=j$  . Hence we have the following :

Lemma 3.4Any element $j\in \tilde{\mathbb{J}}$  is represented as the following form $$j=\left(\begin{array}{cc}A& 0\\ 0& t\end{array}\right),$$  where $A\in GL\left(n\right)$  and $t=(detA{)}^{\frac{2}{n}}$  .

Proposition 3.5Let $\tilde{\mathbb{J}}$  be the isotropy subgroup of $Aut(\Sigma )$  at $0$  .

Then the followings are equivalent :

• (a) $g\in \tilde{\mathbb{J}}$  .
• (b) $g\in GL(n+1)$  normalizes $\mathbb{A}$  .
• (c) $g=\left(\begin{array}{cc}A& 0\\ 0& t\end{array}\right)\in GL(n+1)$  and $F\left(Ax\right)=tF\left(x\right)$  .
• (d) $g=\left(\begin{array}{cc}A& 0\\ 0& t\end{array}\right)\in GL(n+1)$  and $A$  is an automorphism of the associated complete abelian LSA $\mathcal{A}$  and $A^{\prime }HA=tH$  .

Proof. (a) $\Rightarrow$ (b) : Since $\tilde{\mathbb{J}}$  is the isotropy subgroup at 0, $g$  is contained in $GL(n+1)$  . From Theorem  3.3 , $\mathbb{A}$  is the unique abelian normal subgroup of $Aut(\Sigma {)}_0$  . Then $g\mathbb{A}{g}^{-1}\subset Aut(\Sigma {)}_0$  must be equal to $\mathbb{A}$  . Hence $g$  normalizes $\mathbb{A}$  .

(b) $\Rightarrow$ (a),(c) and (d): Let $g\in GL(n+1)$  be a normalizer of the abelian subgroup $\mathbb{A}$  , then for any $h\in \mathbb{A}$  there exists $\tilde{h}\in \mathbb{A}$  such that $g\cdot (h\cdot 0)=\tilde{h}\cdot (g\cdot 0)=\tilde{h}\cdot 0$  . Since $\mathbb{A}$  acts on $\Sigma$  transitively, this shows that $g$  belongs to $Aut(\Sigma )$  , that is, $g\in \tilde{\mathbb{J}}$  . Therefore from Lemma  3.4 , $g=\left(\begin{array}{cc}A& 0\\ 0& t\end{array}\right)$  where $A\in GL\left(n\right)$  and $t=(detA{)}^{\frac{2}{n}}$  . Then for any element $g_x=\left(\left(\begin{array}{cc}{M}_x& 0\\ x^{\prime }H& 1\end{array}\right)\left(\begin{array}{c}x\\ F\left(x\right)\end{array}\right)\right)\in \mathbb{A}$  , we have

$$\begin{array}{c}g{g}_x{g}^{-1}=\left(\left(\begin{array}{cc}A{M}_x{A}^{-1}& 0\\ t{x}^{\prime }H{A}^{-1}& 1\end{array}\right)\left(\begin{array}{c}Ax\\ tF\left(x\right)\end{array}\right)\right)\in \mathbb{A}.\end{array}$$

(3.6)

Hence we have $A{\lambda }_x{A}^{-1}={\lambda }_{Ax}$  , that is, $A\in Aut(\mathcal{A})$  and $F\left(Ax\right)=tF\left(x\right)$  .

Moreover we have $t{x}^{\prime }H{A}^{-1}=(Ax{)}^{\prime }H=x^{\prime }{A}^{\prime }H$  , and hence $tH=A^{\prime }HA$  .

(c) $\Rightarrow$ (a): Since $g\cdot \left(\begin{array}{c}x\\ F\left(x\right)\end{array}\right)=\left(\begin{array}{c}Ax\\ tF\left(x\right)\end{array}\right)\in \Sigma$  , $g\in \tilde{\mathbb{J}}\subset Aut(\Sigma )$  .

(d) $\Rightarrow$ (b) and (c): From the conditions of (d), we have

$$\begin{array}{ccc}& A{M}_x{A}^{-1}=I+A{\lambda }_x{A}^{-1}=I+{\lambda }_{Ax}=M_{Ax}& \end{array}$$

$$\begin{array}{ccc}& t{x}^{\prime }H{A}^{-1}=x^{\prime }{A}^{\prime }H=(Ax{)}^{\prime }H.& \end{array}$$

Since $A$  is an automorphism of the LSA $\mathcal{A}$  ,

$$\begin{array}{ccc}F\left(Ax\right)& =& x^{\prime }{A}^{\prime }H(\frac{1}{2}I-\frac{1}{3}A{\lambda }_x{A}^{-1}+\frac{1}{4}A({\lambda }_x{)}^2{A}^{-1}-\cdots )Ax\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }{A}^{\prime }HA(\frac{1}{2}I-\frac{1}{3}{\lambda }_x+\frac{1}{4}({\lambda }_x{)}^2-\cdots )x\end{array}$$

$$\begin{array}{ccc}& =& tF\left(x\right).\end{array}$$

Using the above, ( 3.6 ) shows that $g$  normalizes the abelian subgroup $\mathbb{A}$  . $\square$ 

From Proposition  3.5 , we have the following theorem :

Theorem 3.6Let $\Sigma$  be a graph of a function $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  . If the $Aut(\Sigma {)}_0$  contains a transitive abelian subgroup, then the followings hold.

• (a) $Aut(\Sigma {)}_0=\mathbb{A}\mathbb{\setminus }\mathbb{J}$  , where $\mathbb{A}$  is the simply transitive abelian subgroup and $\mathbb{J}$  is the isotropy subgroup of $Aut(Sigma{)}_0$  .
• (b) $\mathbb{J}$  acts on the associated complete abelian LSA $\mathcal{A}$  as an automorphism and similarity with respect to the Hessian type metric.

A derivation $B$  on an LSA $\mathcal{A}$  is the linear map satisfying $$B\left(ab\right)=\left(Ba\right)b+a\left(Bb\right),$$  for all $a,b\in \mathcal{A}$  , that is, $B{\lambda }_a={\lambda }_{Ba}+{\lambda }_{a}B$  for all $a\in \mathcal{A}$  . Let $Der(\mathcal{A})$  be the set of all the derivation on $\mathcal{A}$  , which is the Lie algebra of the automorphism group of $\mathcal{A}$  , that is, $Der(\mathcal{A})=LieAut(\mathcal{A})$  .

Definition 3.7For an LSA $\mathcal{A}$  with an inner product $H$  , a map $B:\mathcal{A}\to \mathcal{A}$  is called an infinitesimal similarity if it satisfies the following :

$$\begin{array}{ccc}& (Ba{)}^{\prime }Hb+a^{\prime }HBb=s{a}^{\prime }Hb,& \end{array}$$

(3.7)

$$\begin{array}{ccc}& (\text{or}\langle Ba,b\rangle +\langle a,Bb\rangle =s\langle a,b\rangle )& \end{array}$$

where $s$  must be equal to $\frac{2}{n}trB$  for $n=dim\mathcal{A}$  .

Let's denote the subalgebra of $Der(\mathcal{A})$  whose elements are infinitesimal similarities by $sDer(\mathcal{A})$  . For an element $B\in sDer(\mathcal{A})$  , if $trB=0$  , we call it infinitesimal isometry derivation. We will denote by $iDer(\mathcal{A})$  the subalgebra of all infinitesimal isometry derivations. It readily follows that the codimension of $iDer(\mathcal{A})$  in $sDer(\mathcal{A})$  must be 0 or 1.

Proposition 3.8Let $\Sigma$  be the graph of a function $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  , on which an abelian subgroup acts simply transitively.

Let $\mathfrak{j}$  be the Lie algebra of the isotropy subgroup $\mathbb{J}$  of $Aut(\Sigma {)}_0$  . Then $$\mathfrak{j}=\left\{\left(\begin{array}{cc}B& 0\\ 0& s\end{array}\right)|B\in sDer(\mathcal{A})\text{and}s=\frac{2}{n}trB\right\},$$  where $\mathcal{A}$  is the associated complete abelian LSA. Therefore the dimension of $\mathbb{J}$  is equal to the dimension of the Lie algebra $sDer(\mathcal{A})$  .

Proof. From the Lemma  3.4 , the elements of $\mathfrak{j}$  are given by the matrix $\left(\begin{array}{cc}B& 0\\ 0& s\end{array}\right)\in gl(n+1)$  where $s=\frac{2}{n}trB$  . Then, from the Proposition  3.5 (d), $B$  must be contained in $sDer(\mathcal{A})$  . $\square$ 

If $sDer(\mathcal{A})⫌︀iDer(\mathcal{A})$  , then the complete abelian LSA $\mathcal{A}$  has a derivation $B$  , which is an infinitesimal similarity as well with $trB=\frac{n}{2}$  , that is, $a^{\prime }Hb=a^{\prime }{B}^{\prime }Hb+a^{\prime }HBb$  . Hence the associated polynomial $F$  given in Theorem  3.3 can be represented as the following form,

$$\begin{array}{ccc}F\left(x\right)& =& x^{\prime }H(\frac{1}{2}I-\frac{1}{3}{\lambda }_x+\frac{1}{4}({\lambda }_x{)}^2-\cdots )x\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }H(\frac{1}{2}x-\frac{1}{3}{x}^2+\frac{1}{4}{x}^3-\cdots )\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }{B}^{\prime }H(\frac{1}{2}x-\frac{1}{3}{x}^2+\frac{1}{4}{x}^3-\cdots )+x^{\prime }H(\frac{1}{2}Bx-\frac{2}{3}xBx+\frac{3}{4}{x}^{2}Bx-\cdots )\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }H(Bx-xBx+x^{2}Bx-\cdots )\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }H(I-{\lambda }_x+({\lambda }_x{)}^2-\cdots )Bx\end{array}$$

$$\begin{array}{ccc}& =& x^{\prime }H(I+{\lambda }_x{)}^{-1}Bx,\end{array}$$

(3.8)

by using the symmetry of $H$  , the associativity $(xy{)}^{\prime }Hz=x^{\prime }H(yz)$  and the derivation property, $B{x}^k=k{x}^{k-1}Bx$  for all $x,y,z\in {\mathbb{R}}^n$  .

4 Complete abelian LSA and Filiform LSA

Let $\mathcal{A}$  be a complete abelian LSA and let ${\mathcal{A}}^i$  's be ideals given by the following : $${\mathcal{A}}^1=\mathcal{A},{\mathcal{A}}^{i+1}=\mathcal{A}\cdot {\mathcal{A}}^i\text{for}i=1,2,3,\cdots .$$  Then there exists $m$  such that ${\mathcal{A}}^{m+1}=\left\{0\right\}$  since ${\lambda }_x$  's are nilpotent for all $x\in \mathcal{A}$  . In fact they make a descending central series, that is, $$\mathcal{A}={\mathcal{A}}^1\supset {\mathcal{A}}^2\supset \cdots \supset {\mathcal{A}}^m\supset {\mathcal{A}}^{m+1}=\left\{0\right\}.$$  If $a\in {\mathcal{A}}^m$  , then $b\cdot a\in {\mathcal{A}}^{m+1}=\left\{0\right\}$  for any $b\in \mathcal{A}$  . This says that ${\mathcal{A}}^m$  is contained in the annihilator of $\mathcal{A}$  , $Ann(\mathcal{A})=\{x\in \mathcal{A}|xy=0=yx,\text{for all}y\in \mathcal{A}\}$  . Notice that $Ann(\mathcal{A})$  is a nontrivial ideal of a complete abelian $\mathcal{A}$  .

Definition 4.1 ([3] )Let $\mathcal{A}$  be a complete abelian LSA. A (vector space) basis of $\mathcal{A}$  $$\{e_{11},\cdots ,e_{1k_1},e_{21},\cdots ,e_{2k_2},\cdots ,e_{m1},\cdots ,e_{m{k}_m}\}$$  is called adequate if for all $i$  , $e_{il}$  belongs to ${\mathcal{A}}^i$  and $k_i+k_{i+1}+\cdots +k_m=dim{\mathcal{A}}^i$  .

Let ${\mathcal{C}}_i$  be ideals defined by $${\mathcal{C}}_1=Ann(\mathcal{A}),{\mathcal{C}}_{i+1}=\{x\in \mathcal{A}|yx\in {\mathcal{C}}_i\text{for any}y\in \mathcal{A}\}.$$  Then they make an ascending central series, i.e., $${\mathcal{C}}_0=\left\{0\right\}\subset {\mathcal{C}}_1\subset {\mathcal{C}}_2\subset \cdots .$$ 

Lemma 4.2

• (a) $\mathcal{A}/{\mathcal{C}}_i$  is a complete abelian LSA for all $i=1,2,\cdots$  .
• (b) $Ann(\mathcal{A}/{\mathcal{C}}_i)={\mathcal{C}}_{i+1}/{\mathcal{C}}_i$  for all $i=1,2,\cdots$  .
• (c) There exists an integer $m^{\prime }$  such that ${\mathcal{C}}_{m^{\prime }}=\mathcal{A}$  , so the ascending central series is terminated finitely.
• (d) ${\mathcal{C}}_i=\{x\in \mathcal{A}|x\cdot {\mathcal{A}}^i=\left\{0\right\}\}$  for $i\in \{1,2,\cdots ,m^{\prime }\}$  .

Proof. (a) $\mathcal{A}/{\mathcal{C}}_i$  is also an abelian LSA since ${\mathcal{C}}_i$  is ideal. Since ${\lambda }_x={\rho }_x$  are nilpotent for all $x\in \mathcal{A}$  , ${\lambda }_{\overline{x}}={\rho }_{\overline{x}}$  are also nilpotent for $\overline{x}=x+{\mathcal{C}}_i\in \mathcal{A}/{\mathcal{C}}_i$  . So we have $tr{\rho }_{\overline{x}}=0$  for all $x\in \mathcal{A}$  and $\mathcal{A}/{\mathcal{C}}_i$  is complete.

(b) is clear from the definition.

(c) From (a) and (b), ${\mathcal{C}}_{i+1}/{\mathcal{C}}_i$  are all nontrivial if ${\mathcal{C}}_i\ne \mathcal{A}$  . Since $\mathcal{A}$  is of finite dimensional, there exist $m^{\prime }$  such that ${\mathcal{C}}_m^{\prime }=\mathcal{A}$  .

(d) Since $x\in {\mathcal{C}}_i$  is equivalent to that $xy\in {\mathcal{C}}_{i-1}$  for all $y\in \mathcal{A}$  , it follows immediately by induction. $\square$ 

Proposition 4.3

• (a) ${\mathcal{A}}^{m-i}\subset {\mathcal{C}}_{i+1}$  for $i\in \{0,\cdots ,m-1\}$  .
• (b) $m^{\prime }=m$  .

Proof. (a) Since ${\mathcal{A}}^{m-i}{\mathcal{A}}^{i+1}\subset {\mathcal{A}}^{m+1}=\left\{0\right\}$  , ${\mathcal{A}}^{m-i}$  is a subspace of ${\mathcal{C}}_{i+1}$  for all $i\in \{0,1,2,\cdots ,m-1\}$  from Lemma  4.2 (d).

(b) From (a), $\mathcal{A}={\mathcal{A}}^1\subset {\mathcal{C}}_m$  , hence we have $m^{\prime }\le m$  . If $m^{\prime }<m$  , then ${\mathcal{A}}^m={\mathcal{A}}^{m-1}\mathcal{A}={\mathcal{A}}^{m-1}{\mathcal{C}}_{m^{\prime }}=\left\{0\right\}$  . This is a contradiction. $\square$ 

Definition 4.4Let $V$  be a vector space and let $\langle ,\rangle$  be an inner product on $V$  . For a subset $W$  , the perpendicular subspace $W^{\perp }$  is defined as the following : $$W^{\perp }=\{x\in V|\langle x,y\rangle =0,\text{for all}y\in W\}.$$ 

Proposition 4.5Let $\mathcal{A}$  be a complete abelian LSA with a nondegenerate Hessian type inner product $\langle ,\rangle$  . Then ${\mathcal{C}}_i=({\mathcal{A}}^{i+1}{)}^{\perp }$  and the dimension of ${\mathcal{C}}_i/{\mathcal{C}}_{i-1}$  is equal to the dimension of ${\mathcal{A}}^i/{\mathcal{A}}^{i+1}$  for $i=1,2,\cdots ,m$  .

Proof. By the associativity ( 3.1 ) of the Hessian type inner product, ${\mathcal{C}}_i\subset ({\mathcal{A}}^{i+1}{)}^{\perp }$  . Let $x\in ({\mathcal{A}}^{i+1}{)}^{\perp }$  , then for any $a\in \mathcal{A}$  and $b\in {\mathcal{A}}^i$  , we have $\langle a,bx\rangle =\langle ab,x\rangle =0.$  Since $\langle ,\rangle$  is nondegenerate, $bx=0$  , that is, $x\in {\mathcal{C}}_i$  from Lemma  4.2 (d). Therefore ${\mathcal{C}}_i=({\mathcal{A}}^{i+1}{)}^{\perp }$  . This says that $dim{\mathcal{C}}_i=dim\mathcal{A}/{\mathcal{A}}^{i+1}$  for each $i=1,2,\cdots ,m$  . So we have $dim{\mathcal{C}}_i/{\mathcal{C}}_{i-1}=dim(\mathcal{A}/{\mathcal{A}}^{i+1})/(\mathcal{A}/{\mathcal{A}}^i)=dim{\mathcal{A}}^i/{\mathcal{A}}^{i+1}$  . $\square$ 

Let $\mathcal{\mathscr{H}}$  be an $n$  -dimensional LSA. $\mathcal{\mathscr{H}}$  is called a filiform LSA if it satisfies : $$dim{\mathcal{\mathscr{H}}}^i=dim{\mathcal{\mathscr{H}}}_i=n+1-i\text{for all}i\in \{1,\cdots ,n\},$$  where ${\mathcal{\mathscr{H}}}^i=\mathcal{\mathscr{H}}{\mathcal{\mathscr{H}}}^{i-1}$  and ${\mathcal{\mathscr{H}}}_i={\mathcal{\mathscr{H}}}_{i-1}\mathcal{\mathscr{H}}$  with ${\mathcal{\mathscr{H}}}^1={\mathcal{\mathscr{H}}}_1=\mathcal{\mathscr{H}}$  .(cf. [3] ) Note that $\mathcal{\mathscr{H}}$  is a $n$  -dimensional filiform LSA if and only if the nilpotency of $\mathcal{\mathscr{H}}$  is $n+1$  , that is, ${\mathcal{\mathscr{H}}}^{n+1}={\mathcal{\mathscr{H}}}_{n+1}=\left\{0\right\}$  and ${\mathcal{\mathscr{H}}}^n\ne \left\{0\right\}$  , ${\mathcal{\mathscr{H}}}_n\ne \left\{0\right\}$  . It has the following properties :

Proposition 4.6 ([3] )Let $\mathcal{\mathscr{H}}$  be an $n$  -dimensional filiform LSA.

• (a) The associated Lie algebra of $\mathcal{\mathscr{H}}$  is nilpotent.
• (b) $\mathcal{\mathscr{H}}$  is a complete LSA.
• (c) ${\mathcal{\mathscr{H}}}^i$  is equal to ${\mathcal{\mathscr{H}}}_i$  for all $i\in \{1,\cdots ,n\}$  .
• (d) $\mathcal{\mathscr{H}}$  has a one-dimensional annihilator $Ann(\mathcal{\mathscr{H}})={\mathcal{\mathscr{H}}}^n={\mathcal{\mathscr{H}}}_n$ 
• (e) For all $i\in \{1,\cdots ,n\}$  , $\mathcal{\mathscr{H}}/{\mathcal{\mathscr{H}}}^i$  is a filiform LSA.
• (f ) There is an adequate basis $\{e_1,\cdots ,e_n\}$  satisfying $$e_1{e}_j=e_{j+1}\text{for all}j\in \{1,2,\cdots ,n-1\}.$$ 

The adequate basis in the Proposition  4.6 (f ) is called strongly adequate.

Note that if $\mathcal{A}$  is an abelian filiform LSA and $\{e_1,\cdots ,e_n\}$  is a strongly adequate basis of $\mathcal{A}$  , then $e_i=e_1^i$  for all $i\in \{1,\cdots ,n\}$  and we have $e_i{e}_j=e_{i+j}$  .(cf. [3] )

Lemma 4.7Let $\mathcal{A}$  be an abelian filiform LSA with a strongly adequate basis $\{e_1,\cdots ,e_n\}$  , then we have the following :

• (a) There is a nondegenerate Hessian type inner product $\langle ,\rangle$  on $\mathcal{A}$  which is given by

  $$\begin{array}{c}\langle e_i,e_j\rangle =\{\begin{array}{cc}1,& i+j=n+1,\\ 0,& i+j\ne n+1,\end{array}\end{array}$$

  (4.1)

  which is unique up to LSA automorphism, i.e., if $\langle ,\rangle$  is a nondegenerate Hessian type inner product on $\mathcal{A}$  , then there exists a strongly adequate basis such that ( 4.1 ) holds up to sign.
• (b) The dimension of $sDer(\mathcal{A})$  is $1$  and $iDer(\mathcal{A})=\left\{0\right\}$  .

Proof. (a) Let $\langle ,\rangle$  be the inner product given in ( 4.1 ). Then $\langle ,\rangle$  is obviously nondegenerate. For any $i,j$  and $k\in \{1,\cdots ,n\}$  ,

$$\begin{array}{ccc}\langle e_i\cdot e_j,e_k\rangle =\langle e_{i+j},e_k\rangle =\{\begin{array}{cc}1,& i+j+k=n+1,\\ 0,& i+j+k\ne n+1,\end{array}& & \end{array}$$

$$\begin{array}{ccc}\langle e_i,e_j\cdot e_k\rangle =\langle e_i,e_{j+k}\rangle =\{\begin{array}{cc}1,& i+j+k=n+1,\\ 0,& i+j+k\ne n+1,\end{array}& & \end{array}$$

hence we conclude that $\langle ,\rangle$  is of Hessian type. Now, let $\langle ,\rangle$  be a nondegenerate Hessian type inner product on $\mathcal{A}$  . Then $\langle e_i,e_j\rangle =0$  for $i+j>n+1$  since ${\mathcal{A}}^{n+1}=\left\{0\right\}$  . Let $\langle e_i,e_j\rangle =s_k$  for $k=i+j\le n+1$  . Since $\langle ,\rangle$  is nondegenerate, $s_{n+1}$  must not be zero. By changing a basis inductively, we can obtain the basis $\{{\overline{e}}_1,{\overline{e}}_2,\cdots ,{\overline{e}}_n\}$  satisfying ( 4.1 ) as follows : First, by scaling $e_1$  , we may assume that $s_{n+1}=\pm 1$  . If $s_n\ne 0$  , by putting ${\overline{e}}_1=e_1\mp \frac{s_n}{n}{e}_2$  and ${\overline{e}}_{j+1}={\overline{e}}_1{\overline{e}}_j$  for $j\in \{1,2,\cdots ,n-1\}$  , we have that $\langle {\overline{e}}_i,{\overline{e}}_j\rangle =0$  for $i+j=n$  . If $s_n=\cdots =s_{k+1}=0$  and $s_k\ne 0$  , put ${\overline{e}}_1=e_1\mp \frac{s_k}{k}{e}_{m-k+2}$  and ${\overline{e}}_{j+1}={\overline{e}}_1{\overline{e}}_j$  for $j\in \{1,2,\cdots ,n-1\}$  .

Then $\langle {\overline{e}}_i,{\overline{e}}_j\rangle =0$  for $i+j=k,k+1,\cdots ,n$  .

(b) Define a derivation $B_1:\mathcal{A}\to \mathcal{A}$  by $B_1\left(e_1\right)=e_1$  , so that $B_1\left(e_j\right)=j{e}_j$  for all $j=2,3,\cdots ,n$  . Then $tr{B}_1=\frac{n(n+1)}{2}$  and $$\langle B_1\left(e_i\right),e_j\rangle +\langle e_i,B_1\left(e_j\right)\rangle =\{\begin{array}{cc}n+1,& i+j=n+1\\ 0,& i+j\ne n+1\end{array}\}=\frac{2}{n}tr{B}_1\langle e_i,e_j\rangle .$$  So $B_1$  is contained in $sDer(\mathcal{A})$  , and hence $dimsDer(\mathcal{A})\ge 1$  .

For any $B\in sDer(\mathcal{A})$  and the strong adequate basis $\{e_1,\cdots ,e_n\}$  , $B\left(e_k\right)=B\left(e_1^k\right)=k{e}_{k-1}B\left(e_1\right)$  , $(k=2,3,\cdots ,n)$  by induction. Therefore if $B\left(e_1\right)=u_1{e}_1+u_2{e}_2+\cdots +u_n{e}_n$  , then $B\left(e_k\right)=k{e}_{k-1}B\left(e_1\right)=k{e}_{k-1}(u_1{e}_1+u_2{e}_2+\cdots +u_n{e}_n)$  for all $k=2,3,\cdots ,n$  and hence $trB=\frac{n(n+1)}{2}{u}_1$  . Now assume that $0\ne B\in iDer(\mathcal{A})$  , then the $e_1$  -coefficient of $B\left(e_1\right)$  must be zero so that $B\left(e_1\right)$  is contained in ${\mathcal{A}}^2$  . In this case, there is a number $k$  such that $\langle B\left(e_1\right),e_k\rangle \ne 0$  and $\langle B\left(e_1\right),e_{k+1}\rangle =0$  since $\langle ,\rangle$  is nondegenerate and the basis is adequate. Note that $k$  is less than $n$  because $B\left(e_1\right)\in {\mathcal{A}}^2$  . Since $B\left(e_k\right)=k{e}_{k-1}B\left(e_1\right)$  , we have $$0=\langle B\left(e_1\right),e_k\rangle +\langle e_1,B\left(e_k\right)\rangle =(k+1)\langle B\left(e_1\right),e_k\rangle \ne 0.$$  This is a contradiction. Hence, $iDer(\mathcal{A})=\left\{0\right\}$  and the dimension of $sDer(\mathcal{A})$  has to be 1. $\square$ 

The following key observation of this paper is made by Y. Choi[2] .

Theorem 4.8Let $\mathcal{A}$  be a complete abelian LSA. Then $\mathcal{A}$  is filiform if and only if the associated polynomial $F$  of $\mathcal{A}$  with a nondegenerate Hessian metric is the Cayley polynomial.

Proof. Let $\mathcal{A}$  be an abelian filiform LSA and $\{e_1,\cdots ,e_n\}$  be a strongly adequate basis of $\mathcal{A}$  . Then from Lemma  4.7 , we have a Hessian type inner product $H$  and an infinitesimal similarity derivation $B$  , which are given by the following matrices : $$H=\left(\begin{array}{ccccc}& & & & 1\\ & 0& & 1& \\ & & ...& & \\ & 1& & 0& \\ 1& & & & \\ \end{array}\right),B=\left(\begin{array}{ccccc}\frac{1}{n+1}& & & & \\ & \frac{2}{n+1}& & 0& \\ & & ...& & \\ & 0& & \frac{n-1}{n+1}& \\ & & & & \frac{n}{n+1}\\ \end{array}\right).$$  The left multiplication ${\lambda }_x$  of $x=x_1{e}_1+x_2{e}_2+\cdots +x_n{e}_n$  and $(I+{\lambda }_x{)}^{-1}$  can be written as : $${\lambda }_x=\left(\begin{array}{ccccc}0& & & & \\ x_1& 0& & 0& \\ x_2& x_1& 0& & \\ ...& \ddots & \ddots & \ddots & \\ x_{n-1}& \cdot \cdot \cdot & x_2& x_1& 0\\ \end{array}\right),(I+{\lambda }_x{)}^{-1}=\left(\begin{array}{ccccc}1& & & & \\ p_1& 1& & 0& \\ p_2& p_1& 1& & \\ ...& \ddots & \ddots & \ddots & \\ p_{n-1}& \cdot \cdot \cdot & p_2& p_1& 1\\ \end{array}\right)$$  where $p_r={\sum }_{i+j+\cdots +m=r}(-1{)}^d{\overbrace{x_i{x}_j\cdots x_m}}^d$  for $r=1,2,\cdots ,n-1$  . Then from ( 3.8 ), the associated polynomial is given by $$F\left(x\right)=x^{\prime }H(I+{\lambda }_x{)}^{-1}Bx={\sum }_{d=2}^{n+1}(-1{)}^d\frac{1}{d}{\sum }_{i+j+\cdots +m=n+1}{\overbrace{x_i{x}_j\cdots x_m}}^d,$$  that is, $F$  is a Cayley polynomial. Conversely, suppose that $\mathcal{A}$  is an $n$  -dimensional complete abelian LSA which is not filiform, then the nilpotency class of ${\lambda }_x$  is less than $n$  , that is, ${\lambda }_x^{n-1}=0$  for all $x\in \mathcal{A}$  . This says that the degree of the associated polynomial $F\left(x\right)$  given Theorem  3.3 is less than $n+1$  . Therefore the polynomial $F$  can not be the Cayley polynomial. $\square$ 

Now we can restate the conjecture of Eastwood and Ezhov in terms of LSA

Conjecture 4.9 (LSA version)Let $\mathcal{A}$  be a complete abelian LSA with a nondegenerate Hessian type inner product. If the dimension of $sDer(\mathcal{A})$  is $1$  , then $\mathcal{A}$  is the abelian filiform LSA.

5 Proof of the Main Theorem

The affine $(n+1)$  -space is divided into two connected components by the hypersurface $\Sigma$  given by a function $F:{\mathbb{R}}^n\to \mathbb{R}$  . Choose the one which is lying above $\Sigma$  and denote it by $\Omega$  , then $\Sigma$  becomes a boundary of $\Omega$  . Let $Aut(\Omega )$  be the subgroup of $Aff(n+1)$  whose elements leave the domain $\Omega$  invariant.

Proposition 5.1 $Aut(\Omega )\subset Aut(\Sigma )$  and $Aut(\Sigma {)}_0=Aut(\Omega {)}_0$  .

Proof. For any $g\in Aut(\Sigma {)}_0\subset Aff(n+1)$  and $x\in \Omega$  , if $gx\in \Sigma$  then put $y=gx$  and choose $g^{-1}y=z\in \Sigma$  . Then we have $gx=gz$  , i.e., $g$  is not one-to-one. This contradicts that $g\in Aff(n+1)$  . If $gx\in {\Omega }^{\prime }$  , the other connected component, choose a path $g_t$  from $g_0=e$  to $g_1=g$  in $Aut(\Sigma {)}_0$  . Then $g_{t}x$  is a path from $x$  to $gx$  and we can choose $t_0$  such that $g_{t_0}x\in \Sigma$  . This contradicts as above.

Hence we have $gx\in \Omega$  , i.e., $g\in Aut(\Omega {)}_0$  .

Conversely, for any $g\in Aut(\Omega )\subset Aff(n+1)$  and $x\in \Sigma$  , if $gx\in \Omega$  then put $y=gx$  and $g^{-1}y=x\in \Sigma$  . This contradicts that $g\in Aut(\Omega )$  . If $gx\in {\Omega }^{\prime }$  , choose a sequence $\left\{x_n\right\}\subset \Omega$  which converges to $x$  . Then the sequence $\{g\cdot x_n\}\subset \Omega$  converges to $g\cdot x$  , hence there exists $x_N$  such that $g{x}_N\in {\Omega }^{\prime }$  .

This is a contradiction. Hence we have $gx\in \Sigma$  , i.e., $g\in Aut(\Sigma )$  . Similarly if $g\in Aut(\Omega {)}_0$  , then $g\in Aut(\Sigma {)}_0$  by considering a path $g_t$  with the same argument. $\square$ 

The domain above the Cayley hypersurface is homogeneous, in fact, the identity component of its automorphism group acts on the domain simply transitively(cf.

Lemma  4.7 (b)). In the following, we will assume that the domain above the hypersurface is also homogeneous. We call these hypersurfaces extensible homogeneous.

Lemma 5.2Let $\Sigma$  be a graph of a function $F:{\mathbb{R}}^n\to \mathbb{R}$  with $|detDdF|=1$  , on which the abelian subgroup $\mathbb{A}$  acts simply transitively.

Let $\mathbb{J}$  be the isotropy subgroup of $Aut(\Sigma )$  and let $\mathcal{A}$  be the associated complete abelian LSA with Hessian type inner product $H$  . Then $\Sigma$  is extensible homogeneous if and only if the action of $\mathbb{J}$  on $(\mathcal{A},H)$  contains a similarity which is not an isometry.

Proof. If the action of $\mathbb{J}$  contains only isometries, then from Proposition  3.5 any element $j\in \mathbb{J}$  is represented by $j=\left(\begin{array}{cc}A& 0\\ 0& 1\end{array}\right)$  , hence $j\cdot e_{n+1}=e_{n+1}$  where $e_{n+1}=(0,\cdots ,0,1)$  . This says that $Aut(\Sigma {)}_0\cdot e_{n+1}\subset {\Sigma }_1=\{(x,F(x)+1)\left|\right(x,F\left(x\right))\in \Sigma \}$  , which contradicts that $Aut(\Sigma {)}_0$  acts transitively on $\Omega$  .

Conversely, if the action of $\mathbb{J}$  on $(\mathcal{A},H)$  contains a similarity which is not an isometry, then $\mathbb{J}$  contains 1-parameter subgroup such that $\mathbb{J}\cdot e_{n+1}=\left\{\right(0,\cdots ,0,t\left)\right|t\in {\mathbb{R}}_{+}\}$  from Proposition  3.8 . Moreover $\mathbb{A}$  acts on ${\Sigma }_t=\left\{\right(x,F\left(x\right)+t\left)\right|(x,F(x\left)\right)\in \Sigma ,t\in {\mathbb{R}}_{+}\}$  simply transitively by using the matrix form in Lemma  3.2 . Therefore $Aut(\Sigma {)}_0=\mathbb{A}\mathbb{\setminus }\mathbb{J}$  acts on $\Omega$  transitively. $\square$ 

From Lemma  5.2 , a complete abelian LSA $\mathcal{A}$  , which is associated to the extensible homogeneous hypersurface, has a derivation $B$  such that $B\in sDer(\mathcal{A})$  and $B\notin iDer(\mathcal{A})$  , i.e., $trB\ne 0$  . In this case, for a proof of the Conjecture, it is enough to show that if $\mathcal{A}$  is not filiform, $iDer(\mathcal{A})$  is not trivial. Note that, if $\mathcal{A}$  is an abelian filiform LSA, then the dimension of $sDer(\mathcal{A})$  is 1 and $iDer(\mathcal{A})=\left\{0\right\}$  from Lemma  4.7 .

Proposition 5.3Let $\mathcal{A}$  be a complete abelian LSA with a nondegenerate Hessian type inner product $\langle ,\rangle$  , then $\mathcal{A}$  is filiform if and only if $dimAnn(\mathcal{A})=1$  .

Proof. If $\mathcal{A}$  is filiform, then $dimAnn(\mathcal{A})=1$  from the Proposition  4.6 . Now suppose that $Ann(\mathcal{A})=span\left\{x\right\}={\mathcal{A}}^m$  . Choose $e_1\in \mathcal{A}$  such that $\langle e_1,x\rangle =1$  . Because ${\mathcal{A}}^2$  is perpendicular to ${\mathcal{C}}_1=Ann(\mathcal{A})$  , we have $e_1\notin {\mathcal{A}}^2$  . Since $dim\mathcal{A}/{\mathcal{A}}^2=dim{\mathcal{C}}_1=1$  from Lemma  4.5 , $\mathcal{A}$  can be represented as the following :

$$\mathcal{A}=V+{\mathcal{A}}^2,$$  where $V$  is the 1-dimensional vector space spanned by $\left\{e_1\right\}$  . Then, inductively, we have ${\mathcal{A}}^i=V^i+{\mathcal{A}}^{i+1}$  for all $i$  , and hence $$\mathcal{A}=V+V^2+\cdots +V^m,$$  where $V^i=span\left\{e_1^i\right\}$  for $1\le i\le m$  . Observe that $\{e_1,e_1^2,\cdots ,e_1^m\}$  is linearly independent and hence becomes a basis of $\mathcal{A}$  , that is, $\mathcal{A}$  is a filiform LSA. $\square$ 

Lemma 5.4Let $\mathcal{A}$  be a complete abelian LSA with a nondegenerate Hessian type inner product $\langle ,\rangle$  , and let $d=dimAnn(\mathcal{A})$  be greater than or equal to $2$  . Then the Lie algebra $iDer(\mathcal{A})$  is nontrivial.

Proof. Let $\mathcal{Z}=span\{e_{n-1},e_n\}$  be a subspace of $Ann(\mathcal{A})$  .

(Case 1) $\mathcal{Z}\subset {\mathcal{Z}}^{\perp }$  . Consider a linear map $\psi :\mathcal{A}\to {\mathbb{R}}^2,x\mapsto (\langle x,e_{n-1}\rangle ,\langle x,e_n\rangle )$  .

Then $ker\psi ={\mathcal{Z}}^{\perp }$  . Since ${\mathcal{Z}}^{\perp }$  has codimension 2, $\phi$  is onto and hence there exist elements $e_1,e_2\in \mathcal{A}$  such that

$$\begin{array}{ccc}& \langle e_1,e_{n-1}\rangle =0,\langle e_1,e_n\rangle =1& \end{array}$$

$$\begin{array}{ccc}& \langle e_2,e_{n-1}\rangle =1,\langle e_2,e_n\rangle =0& \end{array}$$

Denote $V=span\{e_1,e_2\}$  , then $\mathcal{A}=V+{\mathcal{Z}}^{\perp }$  . Note that $\mathcal{Z}\subset Ann(\mathcal{A})$  implies ${\mathcal{Z}}^{\perp }\supset Ann(\mathcal{A}{)}^{\perp }={\mathcal{A}}^2$  . Now define a map $B:\mathcal{A}\to \mathcal{A}$  by $$B\left(e_1\right)=-e_{n-1},B\left(e_2\right)=e_n,B\left(x\right)=0\text{for}x\in {\mathcal{Z}}^{\perp }.$$  Then $B$  is a derivation since $B(\mathcal{A})=\mathcal{Z}\subset Ann(\mathcal{A})$  and $B\left({\mathcal{A}}^2\right)=\left\{0\right\}$  . Let $\{e_1,e_2,e_3,\cdots ,e_{n-2},e_{n-1},e_n\}$  be a basis of $\mathcal{A}$  where $e_3,\cdots ,e_{n-2}\in {\mathcal{Z}}^{\perp }$  . Then the matrix form $H$  of the Hessian type inner product $\langle ,\rangle$  and the derivation $B$  are given as follows, $$H=\left(\begin{array}{cccccc}& & & & 0& 1\\ & *& & *& 1& 0\\ & & & & 0& 0\\ & *& & *& ...& ...\\ 0& 1& 0& \cdot \cdot \cdot & 0& 0\\ 1& 0& 0& \cdot \cdot \cdot & 0& 0\\ \end{array}\right),B=\left(\begin{array}{cccccc}0& 0& 0& \cdot \cdot \cdot & 0& 0\\ 0& 0& 0& \cdot \cdot \cdot & 0& 0\\ ...& ...& ...& ...& ...& ...\\ 0& 0& 0& \cdot \cdot \cdot & 0& 0\\ -1& 0& 0& \cdot \cdot \cdot & 0& 0\\ 0& 1& 0& \cdot \cdot \cdot & 0& 0\\ \end{array}\right).$$  By matrix calculation, we have $B^{t}H+HB=0$  . So $B$  is contained in $iDer(\mathcal{A})$  , that is, $iDer(\mathcal{A})\ne \left\{0\right\}$  .

(Case 2) $\mathcal{Z}\not\subset {\mathcal{Z}}^{\perp }$  and $\mathcal{Z}\cap {\mathcal{Z}}^{\perp }\ne \left\{0\right\}$  . By coordinate change, we may assume that $e_{n-1}\in \mathcal{Z}\cap {\mathcal{Z}}^{\perp }$  and $\langle e_n,e_n\rangle =1$  using $\mathcal{Z}\not\subset {\mathcal{Z}}^{\perp }$  . Put ${\mathcal{Z}}_1=span\left\{e_n\right\}$  . Then since ${\mathcal{Z}}_1$  is an ideal, by the associativity of Hessian type inner product, $\overline{\mathcal{A}}={\mathcal{Z}}_1^{\perp }$  becomes an ideal of $\mathcal{A}$  . Since ${\mathcal{Z}}_1$  is a nondegenerate subspace, $\mathcal{A}=\overline{\mathcal{A}}\oplus {\mathcal{Z}}_1$  . Note that $\mathcal{Z}\cap {\mathcal{Z}}^{\perp }\subset \overline{\mathcal{A}}$  . Choose $e_1\in \overline{\mathcal{A}}$  such that $\langle e_1,e_{n-1}\rangle =1$  . In this case define a derivation $B$  by $$B\left(e_1\right)=-e_n,B\left(e_n\right)=e_{n-1},B\left(x\right)=0\text{for}x\in {\mathcal{Z}}^{\perp }.$$  Let $\{e_1,e_2,e_3,\cdots ,e_{n-2},e_{n-1},e_n\}$  be a basis of $\mathcal{A}$  where $e_2,e_3\cdots ,e_{n-2}\in {\mathcal{Z}}^{\perp }$  . Then the matrix forms $H$  of the Hessian type inner product $\langle ,\rangle$  and the derivation $B$  are given as follows $$H=\left(\begin{array}{cccccc}& & & & 1& 0\\ & *& & *& 0& 0\\ & & & & 0& 0\\ & *& & *& ...& ...\\ 1& 0& 0& \cdot \cdot \cdot & 0& 0\\ 0& 0& 0& \cdot \cdot \cdot & 0& 1\\ \end{array}\right),B=\left(\begin{array}{cccccc}0& 0& 0& \cdot \cdot \cdot & 0& 0\\ 0& 0& 0& \cdot \cdot \cdot & 0& 0\\ ...& ...& ...& ...& ...& ...\\ 0& 0& 0& \cdot \cdot \cdot & 0& 0\\ 0& 0& 0& \cdot \cdot \cdot & 0& 1\\ -1& 0& 0& \cdot \cdot \cdot & 0& 0\\ \end{array}\right).$$  By matrix calculation, we have $B^{t}H+HB=0$  . So $B$  is contained in $iDer(\mathcal{A})$  , that is, $iDer(\mathcal{A})\ne \left\{0\right\}$  .

(Case 3) $\mathcal{Z}\cap {\mathcal{Z}}^{\perp }=\left\{0\right\}$  . In this case, the restriction of the inner product on $\mathcal{Z}$  is nondegenerate. So ${\mathcal{Z}}^{\perp }$  becomes a nondegenerate ideal again by the associativity of $\langle ,\rangle$  . Therefore we have an algebra direct sum $\mathcal{A}={\mathcal{Z}}^{\perp }\oplus \mathcal{Z}$  .

Since $\mathcal{Z}$  is a trivial LSA, $iDer(\mathcal{Z})=\mathfrak{o}\left(2\right)$  or $\mathfrak{o}(1,1)$  , which are clearly nontrivial.

For any $\overline{B}\in iDer\mathcal{Z}$  the map $B:\mathcal{A}\to \mathcal{A}$  given by $$B\left(x\right)=\{\begin{array}{cc}\overline{B}\left(x\right)& x\in \mathcal{Z},\\ 0& x\in {\mathcal{Z}}^{\perp }\end{array}$$  is contained in $iDer(\mathcal{A})$  . This says that $iDer(\mathcal{A})\ne \left\{0\right\}$  . $\square$ 

Corollary 5.5Let $\mathcal{A}$  be a complete abelian LSA which is associated to the nondegenerate extensible homogeneous hypersurface. Then the followings are equivalent $:$ 

• (a) $\mathcal{A}$  is filiform.
• (b) $dimAnn(\mathcal{A})=1$  .
• (c) $dimsDer(\mathcal{A})=1$  .

Proof. Since this complete abelian LSA $\mathcal{A}$  has a nondegenerate Hessian type inner product, Proposition  5.3 shows that $\mathcal{A}$  is filiform if and only if $dimAnn(\mathcal{A})=1$  , and the fact that $\mathcal{A}$  is filiform implies that $dimsDer(\mathcal{A})=1$  from Lemma  4.7 . Because the LSA $\mathcal{A}$  is associated to the extensible hypersurface, Lemma  5.2 says that $dimsDer(\mathcal{A})/iDer(\mathcal{A})=1$  . Finally, Lemma  5.4 shows that if $dimsDer(\mathcal{A})=1$  , then $dimAnn(\mathcal{A})=1$  . $\square$ 

Main Theorem Let $\Sigma$  be a nondegenerate hypersurface given by a function on ${\mathbb{R}}^n$  . Then $\Sigma$  is a Cayley hypersurface if and only if the followings are satisfied $:$ 

• (E1) $\Sigma$  admits a transitive abelian group $\mathbb{A}$  of affine motions.
• (E2) $Aut(\Sigma )$  has a 1-dimensional isotropy group.
• (E3) Affine normals to $\Sigma$  are everywhere parallel.
• (E4) The domain above the hypersurface is also homogeneous.

Proof. For an $n$  -dimensional abelian filiform LSA $\mathcal{A}$  with Hessian type inner product $H=\langle ,\rangle$  given by ( 4.1 ), let $\mathbb{A}$  be the associated abelian Lie subgroup of $Aff(n+1)$  obtained as in Theorem  2.7 (see the discussion following Remark  2.6 ), that is, $Lie\mathbb{A}$  is the associated Lie algebra of $\mathcal{A}$  . Then Theorem  4.8 shows that $\mathbb{A}$  acts on the Cayley hypersurface $\Sigma$  simply transitively. The conditions, (E2) and (E4) follow from Lemma  4.7 (b) and Lemma  5.2 . From Theorem  2.7 , the affine normals of $\Sigma$  are parallel.

Conversely, the transitive abelian group $\mathbb{A}$  acts in fact simply transitively on $\Sigma$  from Proposition  3.1 . Since $\Sigma$  is a global graph whose affine normals are parallel, $\mathbb{A}$  gives us a complete abelian LSA $\mathcal{A}$  with nondegenerate Hessian type inner product by Theorem  2.7 . From (E2) with Proposition  3.8 , $dimsDer(\mathcal{A})=1$  and hence by Corollary  5.5 , $\mathcal{A}$  is filiform. Then by Theorem  4.8 , the polynomial $F$  obtained from $\mathcal{A}$  is the Cayley polynomial, that is, $\Sigma$  is a Cayley hypersurface. $\square$ 

It is unclear for us at this moment that (E1) and (E2) implies (E4) so that the condition (E4) can be eliminated in the Main Theorem. In this paper we consider a hypersurface given by a graph of a global function defined on all of ${\mathbb{R}}^n$  . It would be very interesting to know whether only the conditions (E1), (E2) and (E3) would imply the globalness of the function. The study of these two problems will give a complete answer to the conjecture of Eastwood and Ezhov. Furthermore the classification problem of affine homogeneous hypersurfaces could be further studied through the LSA with Hessian type inner product as suggested in this paper.

References

1. Neda Bokan, Katsumi Nomizu and Udo Simon, Affine hypersurfaces with parallel cubic forms, Tohoku Math. J. (2) 42 (1990), 101–108
2. Yuncherl Choi, Cayley hypersurface and left symmetric algebra, PhD thesis, Seoul National Univ., (2002)
3. Karel Dekimpe and Veerle Ongenae, Filiform left-symmetric algebras, Geom. Dedicata 74 (1999), 165–199
4. Franki Dillen and Luc Vrancken, Generalized Cayley surfaces, Lecture Notes in Math. 1481 (1990), 36–47
5. Franki Dillen and Luc Vrancken, Hypersurfaces with parallel difference tensor, Japan. J. Math. (N.S.) 24 (1998), 43–60
6. Franki Dillen, Luc Vrancken and Sahnur Yaprak, Affine hypersurfaces with parallel cubic form, Nagoya Math. J. 135 (1994), 153–164
7. Michael Eastwood and Vladimir Ezhov, Cayley Hypersurfaces, arXiv:math.DG/0001134 (2000)
8. William M. Goldman and Morris W. Hirsch, Affine manifolds and orbits of algebraic groups, Trans. Amer. Math. Soc. 295 (1986) no. 1, 175–198
9. Hyuk Kim, Developing maps of affinely flat Lie groups, Bull. Korean Math. Soc. 34 (1997) no. 4, 373–394
10. Katsumi Nomizu and Ulrich Pinkall, Cayley surfaces in affine differential geometry, Tohoku Math. J. (2) 41 (1989), 589–596
11. Katsumi Nomizu and Takeshi Sasaki, A new model of unimodular-affinely homogeneous surfaces, Manuscripta Math. 73 (1991), 39–44
12. Katsumi Nomizu and Takeshi Sasaki, Affine differential geometry, Cambridge University Press (1994)
13. Dan Segal, The structure of complete left-symmetric algebras, Math. Ann. 293 (1992) no. 3, 569–578
14. Hirohiko Shima, On certain locally flat homogeneous manifolds of solvable Lie groups, Osaka J. Math. 13 (1976), 213–229
15. Hirohiko Shima, Homogeneous Hessian manifolds, Ann. Inst. Fourier (Grenoble) 30 (1976) no. 3, 91–128
16. Koichi Tanabe, Affine surfaces with parallel cubic form relative to the dual connection, Kobe J. Math. 15 (1998) no. 1, 85–89
17. È. B. Vinberg The theory of homogeneous convex cones, Trans. of the Moscow Math. Soc. 12 (1963), 340–403