.
We would like to consider the family of such
$Z$
, but for technical reasons, we single out those schemes whose ideals are wellbehaved.
Definition 2.4.
A (zerodimensional) length
$s$
scheme
$Z\subseteq {\mathbb{P}}^{n}$
will be called resolutiongeneral, if the graded Betti numbers in the minimal resolution of
${I}_{Z}$
are the same as those in the resolution of
$s$
general points.
For instance, a length
$7$
subscheme
$Z\subseteq {\mathbb{P}}^{2}$
is resolutiongeneral iff its minimal resolution looks like
$$0\to R(5)\oplus R(4)\to R(3{)}^{3}\to R\to R/{I}_{Z}\to 0.$$
In particular,
$Z$
does not lie on a conic.
Definition 2.5.
A zerodimensional scheme
$Z\subseteq {\mathbb{P}}^{n}$
will be called a polar polyhedron of
$\Lambda $
, if it is apolar to
$\Lambda $
and resolutiongeneral.
Let
$\text{Hilb}(s,{\mathbb{P}}^{n})$
be the Hilbert scheme parametrising length
$s$
subschemes of
${\mathbb{P}}^{n}$
. Let
$\mathcal{A}(s,\Lambda )$
denote the set of polar
$s$
hedra of
$\Lambda $
, it is then a constructible subset of
$\text{Hilb}(s,{\mathbb{P}}^{n})$
. We will write
$\mathcal{A}$
for
$\mathcal{A}(s,\Lambda )$
if no confusion is likely.
Remark 2.6.
In the literature there is no unanimity on the definition of a `polar polyhedron', in particular the approaches in [
7]
and [
23]
are different from ours and from each other. It is understood that if
$Z=\{{L}_{1},...,{L}_{s}\}$
are
$s$
general points, then morally
$Z$
should count as a polar
$s$
hedron of any
$\Lambda \subseteq \text{span}\left\{{L}_{i}^{d}\right\}$
. However, it is not obvious which degenerations of
$Z$
should be allowed, and it seems that (within reason) we should tailor our definition to the specific problem at hand.
Many of our results depend on a free resolution of
${I}_{Z}$
, and hence `resolutiongeneral' seems to be the most suitable notion. This issue never arises in [
22]
, because there it is tacitly assumed that all geometric configurations are nondegenerate.
If
$\mathcal{A}(s,\Lambda )$
is nonempty, so is
$\mathcal{A}(t,\Lambda )$
for any
$t>s$
. It is the case that every
$\Lambda $
in
$G(r,{S}_{d})$
admits a polar
$\left(\genfrac{}{}{0ex}{}{n+d}{d}\right)$
hedron. An elementary parameter count (see [2] ) shows that a general
$\Lambda $
in
$G(r,{S}_{d})$
will admit a polar
$s$
hedron only if
$$\begin{array}{c}s\ge \frac{r\left(\genfrac{}{}{0ex}{}{n+d}{d}\right)}{n+r}.\end{array}$$ 
(3)

Definition 2.7.
A quadruple
$(n,d,r,s)$
which satisfies ( 3 ) is said to be nondegenerate, if a general
$\Lambda $
admits a polar
$s$
hedron.
A quadruple satisfying ( 3 ) is degenerate if the set
$\{\Lambda :\mathcal{A}(s,\Lambda )\ne \varnothing \}$
fails to be dense in
$G(r,{S}_{d})$
. Very few such examples are known (see [2] for the list), but none of them is without its geometric peculiarity.
In general it is not trivial to prove that a particular quadruple is nondegenerate.
For
$r=1$
, we have the following classification theorem by Alexander and Hirschowitz.
Theorem 2.8 (see [19] ).
Assuming
$r=1$
and
$d>2$
, the only degenerate cases are
$(n,d,s)=(2,4,5),(3,4,9),(4,4,14)$
and
$(4,3,7)$
.
For
$r>1$
we have the following results by Dionisi and Fontanari.
Theorem 2.9.
Assume
$r>1$
. Then

(i)
for
$n=2$
, the only degenerate quadruple is
$(2,3,2,5)$
;

(ii)
there are no degenerate quadruples with
$r\ge n+1$
.
The proofs may be found in [4, 12] respectively. Part (i) was claimed by Terracini [27] , but his proof is obscure.
If
$(n,d,r,s)$
is nondegenerate, then with a slight abuse of notation we will write
$\mathcal{A}$
for
$\mathcal{A}(s,\Lambda )$
, where
$\Lambda $
is understood to be a general point of
$G(r,{S}_{d})$
. It has dimension
$s(n+r)r\left(\genfrac{}{}{0ex}{}{n+d}{d}\right)$
.
3 Associated systems of points
Recall ([
10,p. 313]
) that if
$\Gamma $
is a zerodimensional Gorenstein scheme, then any closed subscheme
${\Gamma}^{\prime}\subseteq \Gamma $
has a residual scheme
${\Gamma}^{\prime \prime}\subseteq \Gamma $
, such that
$$deg{\Gamma}^{\prime}+deg{\Gamma}^{\prime \prime}=deg\Gamma .$$
In particular this applies if
$\Gamma $
is a (global) complete intersection in
${\mathbb{P}}^{n}$
, which is the only case we will need.
Now let
$\Lambda $
denote a general pencil of planar quartics. Then
$\mathcal{A}=\mathcal{A}(8,\Lambda )$
is
$2$
dimensional; we will show that it is rational. Every
$Z\in \mathcal{A}$
has a HilbertBurch resolution
$$0\to R(5{)}^{2}\stackrel{\mu}{\to}R(4)\oplus R(3{)}^{2}\to R\to R/{I}_{Z}\to 0.$$
(See [
3]
for the basic theory behind the HilbertBurch theorem.) In particular
$dim({I}_{Z}{)}_{3}=2$
, so
$Z$
has an associated point
$\alpha \left(Z\right)$
, defined to be the residual intersection of cubics passing through
$Z$
. The matrix of the map
$\mu $
has the form
$$\begin{array}{c}M=\left[\begin{array}{ccc}\underline{2}& \underline{2}& \underline{1}\\ \underline{2}& \underline{2}& \underline{1}\end{array}\right],\end{array}$$ 
(4)

with the convention that
$\underline{j}$
stands for a degree
$j$
form.
Theorem 3.1.
Let
$\Lambda $
be a general pencil of planar quartics. Then the morphism
$\alpha :\mathcal{A}\u27f6{\mathbb{P}}^{2}$
admits a rational inverse, hence
$\mathcal{A}$
is a rational surface.
Proof. Fix a general point in the image of
$\alpha $
, by change of coordinates we assume it to be
$P=[0,0,1]$
. We would like to show that there is a unique resolution general length
$8$
scheme
$Z$
with associated point
$P$
.
Now
$P$
is defined by the vanishing of the rightmost column in ( 4 ), hence, after rowoperations,
$M$
can be brought into the form
$$M=\left[\begin{array}{ccc}{q}_{1}& {q}_{2}& {u}_{0}\\ {q}_{3}& {q}_{4}& {u}_{1}\end{array}\right],{q}_{i}\in {R}_{2}.$$
We start with the
$24$
dimensional vector space of
$2\times 2$
matrices
$${V}_{1}=\{N=\left[\begin{array}{cc}{q}_{1}& {q}_{2}\\ {q}_{3}& {q}_{4}\end{array}\right]:{q}_{i}\in {R}_{2}\}.$$
For
$N\in {V}_{1}$
, write
$$\begin{array}{c}{\theta}_{N}={u}_{1}{q}_{1}{u}_{0}{q}_{3},{\theta}_{N}^{\prime}={u}_{1}{q}_{2}{u}_{0}{q}_{4},{\omega}_{N}={q}_{1}{q}_{4}{q}_{2}{q}_{3},\end{array}$$ 
(5)

and let
${J}_{N}$
be the ideal generated by
${\theta}_{N},{\theta}_{N}^{\prime},{\omega}_{N}$
. Thus
${V}_{1}$
is a parameter space for all HilbertBurch matrices as above. For a dense open set of elements
$N$
in
${V}_{1}$
, the ideal
${J}_{N}$
defines a planar length
$8$
scheme.
We let
$G{L}_{2}(\mathbf{C})$
act on
${V}_{1}$
by right multiplication, i.e., for
$g=\left[\begin{array}{cc}\alpha & \beta \\ \gamma & \delta \end{array}\right]\in G{L}_{2}$
, and
$N$
as above,
$$\begin{array}{c}Ng=\left[\begin{array}{cc}{q}_{1}\alpha +{q}_{2}\gamma & {q}_{1}\beta +{q}_{2}\delta \\ {q}_{3}\alpha +{q}_{4}\gamma & {q}_{3}\beta +{q}_{4}\delta \end{array}\right]\end{array}$$ 
(6)

Define
${V}_{2}=\{N\in {V}_{1}:{\theta}_{N},{\theta}_{N}^{\prime}\in {\Lambda}_{3}^{\perp}\}$
, which is a
$12$
dimensional subspace of
${V}_{1}$
. (If
$F\in \Lambda $
, then
${\theta}_{N}\circ F={\theta}_{N}^{\prime}\circ F=0$
is a set of six linear equations. In all,
${V}_{2}$
is defined by
$12$
linear equations which are independent for a general
$\Lambda $
, hence
$dim{V}_{2}=12$
.) Inside
${V}_{2}$
, there is a
$6$
dimensional subspace
$${V}_{3}=\{\left[\begin{array}{cc}a{u}_{0}& b{u}_{0}\\ a{u}_{1}& b{u}_{1}\end{array}\right]:a,b\in {R}_{1}\}.$$
(Since
${\theta}_{N},{\theta}_{N}^{\prime}=0$
for
$N\in {V}_{3}$
, the containment
${V}_{3}\subseteq {V}_{2}$
is clear.) Form the
$6$
dimensional space
$W={V}_{2}/{V}_{3}$
. For
$N\in {V}_{2}$
, write
$\left[N\right]$
for the corresponding point in the projective space
$\mathbb{P}W\simeq {\mathbb{P}}^{5}$
. Since
${V}_{3}\subseteq {V}_{2}\subseteq {V}_{1}$
are inclusions of
$G{L}_{2}$
modules,
$W$
is also a (right)
$G{L}_{2}$
module; in particular
$PG{L}_{2}$
acts on
$\mathbb{P}W$
. The point of this construction lies in the following lemma:
Lemma 3.2.

(i)
If
$N,\stackrel{~}{N}\in {V}_{2}$
are such that
$\left[N\right],\left[\stackrel{~}{N}\right]$
lie in the same
$PG{L}_{2}$
orbit of
$\mathbb{P}W$
, then
${J}_{N}={J}_{\stackrel{~}{N}}$
.

(ii)
Let
$Z\in {\alpha}^{1}\left(P\right)$
. Consider two minimal resolutions of
${I}_{Z}$
with corresponding HilbertBurch matrices
$M,\stackrel{~}{M}$
, and let
$N,\stackrel{~}{N}$
denote their leftmost minors. Then
$\left[N\right],\left[\stackrel{~}{N}\right]$
lie in the same
$PG{L}_{2}$
orbit in
$\mathbb{P}W$
.
Proof. By straightforward calculation,
$$\begin{array}{c}{\theta}_{Ng}=\alpha {\theta}_{N}+\gamma {\theta}_{N}^{\prime},{\theta}_{Ng}^{\prime}=\beta {\theta}_{N}+\delta {\theta}_{N}^{\prime},{\omega}_{Ng}=det\left(g\right){\omega}_{N},\end{array}$$ 
(7)

so
${J}_{N}={J}_{Ng}$
. Let
$Q=\left[\begin{array}{cc}a{u}_{0}& b{u}_{0}\\ a{u}_{1}& b{u}_{1}\end{array}\right]\in {V}_{3}$
. Then
$$\begin{array}{c}{\theta}_{N+Q}={\theta}_{N},{\theta}_{N+Q}^{\prime}={\theta}_{N}^{\prime},{\omega}_{N+Q}={\omega}_{N}a{\theta}_{N}^{\prime}+b{\theta}_{N},\end{array}$$ 
(8)

so
${J}_{N+Q}={J}_{N}$
. This proves (i).
Any two minimal resolutions of
${I}_{Z}$
are isomorphic (see [
9,§20.1]
), which translates into the statement that
$N$
and some
$G{L}_{2}$
translate of
$\stackrel{~}{N}$
must differ by an element of
${V}_{3}$
. This says that
$\left[N\right],\left[\stackrel{~}{N}\right]$
must be in the same orbit, which is (ii). □ Now define a subvariety
$Y=\left\{\right[N]\in \mathbb{P}W:{\omega}_{N}\circ \Lambda =0\}\subseteq \mathbb{P}W$
.
Formulae (
8 ) imply that
${\omega}_{N+Q}\circ \Lambda =0\u27fa{\omega}_{N}\circ \Lambda =0$
(since
${\theta}_{N}\circ \Lambda ={\theta}_{N}^{\prime}\circ \Lambda =0$
), hence this definition is meaningful. The inclusion
$Y\subseteq \mathbb{P}W$
is a
$PG{L}_{2}$
stable by formulae ( 7 ). By the previous lemma, each
$Z\in {\alpha}^{1}\left(P\right)$
defines an orbit
${\Omega}_{Z}\subseteq Y$
. The
$PG{L}_{2}$
stabilizer of a point in
${\Omega}_{Z}$
is trivial, hence
$dim{\Omega}_{Z}=3$
. The union of
$\{{\Omega}_{Z}{\}}_{Z\in {\alpha}^{1}\left(P\right)}$
fills a dense open subset in
$Y$
. Hence it is enough to show that
$Y$
contains only one threedimensional component, this will imply that
${\alpha}^{1}\left(P\right)$
is singleton. Define
$$\begin{array}{cc}{\Gamma}_{1}& =\left\{\right[N]\in \mathbb{P}W:N=\left[\begin{array}{cc}{q}_{1}& 0\\ {q}_{3}& 0\end{array}\right]\text{for some}{q}_{i}\text{and}{\theta}_{N}\circ \Lambda =0\text{}\},\end{array}$$  
$$\begin{array}{cc}{\Gamma}_{2}& =\left\{\right[N]\in \mathbb{P}W:N=\left[\begin{array}{cc}0& {q}_{2}\\ 0& {q}_{4}\end{array}\right]\text{for some}{q}_{i}\text{and}{\theta}_{N}^{\prime}\circ \Lambda =0\text{}\},\end{array}$$  
each of which is a copy of
${\mathbb{P}}^{2}$
in
$Y$
. Define a birational map
$h:{\Gamma}_{1}\u27f6{\Gamma}_{2}$
as follows. Let
$\left[N\right]\in {\Gamma}_{1}$
, then there is a
$4$
dimensional family of solutions
$({q}_{2},{q}_{4})$
to the equations
$${\theta}_{N}^{\prime}\circ \Lambda ={\omega}_{N}\circ \Lambda =0.$$
(This is so because
${q}_{2},{q}_{4}$
together depend upon
$12$
parameters and there are
$8$
equations.) However, if
$({q}_{2},{q}_{4})$
is one such solution, then
$({q}_{2}+a{u}_{0},{q}_{4}+a{u}_{1})$
is also one for any
$a\in {R}_{1}$
, and this accounts for all the solutions. Hence the class in
$\mathbb{P}W$
of the matrix
$\left[\begin{array}{cc}0& {q}_{2}\\ 0& {q}_{4}\end{array}\right]$
is uniquely determined. We define
$h\left(\right[N\left]\right)$
to be this class. (The reader should verify that this definition is independent of the choice of coset representative for
$\left[N\right]$
.) Now a general element in
$Y$
can be written as a sum
$\left[N\right]+\left[h\right(N\left)\right]$
for
$\left[N\right]\in {\Gamma}_{1}$
, i.e., the ruled join of
${\Gamma}_{1},{\Gamma}_{2}$
along
$h$
contains a dense open subset of
$Y$
. Since this join is irreducible (it is the image of the Segre imbedding
${\mathbb{P}}^{2}\times {\mathbb{P}}^{1}\subseteq {\mathbb{P}}^{5}$
), we are done. □ The argument for the following proposition is similar. As before,
$(2,3,4,7)$
is nondegenerate by Theorem 2.9 .
Proposition 3.3.
Let
$\Lambda $
be a general web of planar cubics. Then
$\mathcal{A}(7,\Lambda )$
is a rational surface.
Proof. The HilbertBurch matrix for
$Z\in \mathcal{A}$
is
$\left[\begin{array}{ccc}\underline{1}& \underline{1}& \underline{1}\\ \underline{2}& \underline{2}& \underline{2}\end{array}\right]$
. For a general
$Z$
, the linear forms in the top row are independent, hence after column operations we can assume the matrix to be
$$\left[\begin{array}{ccc}{u}_{0}& {u}_{1}& {u}_{2}\\ {q}_{0}& {q}_{1}& {q}_{2}\end{array}\right],{q}_{i}\in {R}_{2}.$$
Let
${V}_{1}$
denote the
$18$
dimensional vector space
$\left\{\right[{q}_{0},{q}_{1},{q}_{2}]:{q}_{i}\in {R}_{2}\}$
, and
${V}_{2}$
the
$3$
dimensional subspace
$\left\{\right[a{u}_{0},a{u}_{1},a{u}_{2}]:a\in {R}_{1}\}$
. Let
$W={V}_{1}/{V}_{2}$
. Then the
$12$
equations
$\left\{\right({u}_{i}{q}_{j}{u}_{j}{q}_{i})\circ \Lambda =0\}$
cut out a
$2$
plane in
$\mathbb{P}W$
which is birational to
$\mathcal{A}$
. □ Now let
$(n,d,r,s)=(3,2,6,7)$
, we will show that
$\mathcal{A}$
is birational to the projective
$3$
space. The ideal of every
$Z\in \mathcal{A}$
is generated by three quadrics and and a cubic. The associated point
$\alpha \left(Z\right)$
is defined to the residual intersection of the quadrics through
$Z$
.
Proposition 3.4.
Let
$\Lambda $
be a general point of
$G(6,{S}_{2})$
. Then the map
$\alpha :\mathcal{A}\u27f6{\mathbb{P}}^{3}$
is birational.
Proof. Let
$Z$
be a resolutiongeneral scheme of length
$7$
. It is apolar to
$\Lambda $
iff the three generating quadrics lie in
${\Lambda}_{2}^{\perp}$
.
Let
$P$
be a general point of
${\mathbb{P}}^{3}$
, and let
$W\subseteq {\Lambda}_{2}^{\perp}$
be the
$3$
dimensional subspace of forms vanishing at
$P$
. Then
$W$
defines a length
$8$
scheme
$Y$
. Now the residual scheme of
$P$
in
$Y$
is the only point of
$\mathcal{A}$
mapping to
$P$
. □
Remark 3.5.
The case
$(2,3,7,8)$
has a similar geometry, where
$\mathcal{A}$
is birational to
${\mathbb{P}}^{2}$
. For
$(2,3,8,8)$
(resp.
$(3,2,7,7)$
),
$\mathcal{A}$
is a finite set consisting of
$9$
(resp.
$8$
) points.
4 Symmetric Products of Elliptic Curves
For the examples in this section, the determination of
$\mathcal{A}$
reduces to an intersectiontheoretic calculation on the symmetric product of an elliptic curve. If
$E$
is a smooth projective curve, then
${E}^{\left(m\right)}$
will denote its
$m$
th symmetric product. This is a smooth projective variety whose points are naturally seen as effective degree
$m$
divisors on
$E$
.
Let
$\Lambda $
be a general net of planar quartics. Since
$(2,4,3,9)$
is nondegenerate,
$\mathcal{A}$
is a finite set. In the next theorem we calculate its cardinality.
Theorem 4.1.
Let
$\Lambda $
be a general net of planar quartics. Then
$\Lambda $
admits
$4$
polar enneahedra.
Proof. The ideal of
$Z\in \mathcal{A}$
is generated by one cubic and
$3$
quartics.
The space
${\Lambda}_{3}^{\perp}$
is onedimensional, i.e.,
$\Lambda $
is apolar to a unique cubic curve
$E\subseteq {\mathbb{P}}^{2}$
. Since
$\Lambda $
is general, we may (and will) assume that
$E$
is smooth. If
$H$
denotes the hyperplane divisor on
$E$
, then we have an identification
${H}^{0}(E,4H)={R}_{4}/({I}_{E}{)}_{4}$
. This is a
$12$
dimensional space, denoted
$U$
.
Let
$W={\Lambda}_{4}^{\perp}/({I}_{E}{)}_{4}$
, which is a
$9$
dimensional space inside
$U$
. Every scheme
$Z\subseteq {\mathbb{P}}^{2}$
of length
$9$
which is apolar to
$\Lambda $
is contained in
$E$
, and thus defines an effective divisor on
$E$
. Then the
$3$
dimensional space
${H}^{0}(E,4HZ)$
, which is a priori inside
$U$
, is in fact contained in
$W$
.
The argument shows that the following diagram is a fibre square:
$$\mathcal{A}\to G(3,W)\downarrow {i}_{1}\downarrow {E}^{\left(9\right)}\to {i}_{2}G(3,U)$$
Here
${i}_{1}$
is the natural inclusion and
${i}_{2}\left(Z\right)={H}^{0}(E,4HZ)$
. Since the images of both inclusions have complementary codimensions, it is enough to take the intersection of their classes inside
${H}^{*}\left(G\right(3,U),\mathbf{Z})$
in order to calculate the degree of
$\mathcal{A}$
as a zerocycle.
Conventions. The notation for Schubert calculus follows [
13,§14.7]
.
We refer to [
1]
for some basic cohomological calculations on curves. If
${X}_{1},{X}_{2}$
are varieties, then denote projections by
${\pi}_{i}:{X}_{1}\times {X}_{2}\u27f6{X}_{i}$
. All cohomology is with
$\mathbf{Z}$
coefficients. If
$\alpha $
is a class in
${H}^{*}\left({X}_{1}\right)$
(resp.
${H}^{*}\left({X}_{2}\right)$
), then its pullback to
${H}^{*}({X}_{1}\times {X}_{2})$
is denoted
$\alpha \otimes 1$
(resp.
$1\otimes \alpha $
). Cup product is written as juxtaposition.
Firstly, we should find the rank
$3$
subbundle of
$U\otimes {\mathcal{O}}_{{E}^{\left(9\right)}}$
which defines the inclusion
${i}_{2}$
. Let
$\Delta $
denote the universal divisor on
${E}^{\left(9\right)}\times E$
(see [
1,Ch.IV]
), so that
$\Delta {}_{\left\{Z\right\}\times E}=Z\times E$
. Define a line bundle
$\mathcal{\mathcal{M}}={\pi}_{2}^{*}\left({\mathcal{O}}_{E}\right(4H\left)\right)\otimes \mathcal{O}(\Delta )$
on
${E}^{\left(9\right)}\times E$
. Applying
${{\pi}_{1}}_{*}$
to the inclusion
$$\mathcal{\mathcal{M}}\subseteq {\pi}_{2}^{*}\left({\mathcal{O}}_{E}\right(4H\left)\right),$$
we have
$$(\mathcal{G}=){{\pi}_{1}}_{*}(\mathcal{\mathcal{M}})\subseteq U\otimes {\mathcal{O}}_{{E}^{\left(9\right)}}.$$
A moment's reflection will show that
${i}_{2}$
is induced by the last inclusion.
The image of
${i}_{2}$
has class
$\{3,3,3\}$
. Hence by the JacobiTrudi identity, the class of
$\mathcal{A}$
in
${H}^{18}\left({E}^{\left(9\right)}\right)$
is given by
${c}_{3}({\mathcal{G}}^{*}{)}^{3}$
, which we now calculate.
The cohomology rings of
$E$
and
${E}^{\left(9\right)}$
. Let
${\delta}_{1},{\delta}_{2}\in {H}^{1}\left(E\right)$
be a symplectic basis, it will then generate
${H}^{*}\left(E\right)$
. The product
$\eta ={\delta}_{1}{\delta}_{2}\in {H}^{2}\left(E\right)$
is the class of a point.
Let
$\mathcal{\mathcal{L}}$
be a Poincaré line bundle ([
1,Ch.IV]
) on
$E\times {\text{Pic}}^{9}\left(E\right)$
, then
$\mathcal{\mathcal{E}}={{\pi}_{2}}_{*}(\mathcal{\mathcal{L}})$
is a rank
$9$
bundle on
${\text{Pic}}^{9}\left(E\right)$
. Fix an isomorphism
${\text{Pic}}^{9}\left(E\right)=E$
, then by the calculation of [
1,p. 336]
,
${c}_{1}(\mathcal{\mathcal{E}})=\eta $
. Now let
$\xi ={c}_{1}\left({\mathcal{O}}_{\mathbb{P}\mathcal{\mathcal{E}}}\right(1\left)\right)\in {H}^{2}(\mathbb{P}\mathcal{\mathcal{E}})$
. With the identification
$\mathbb{P}\mathcal{\mathcal{E}}={E}^{\left(9\right)}$
, the ring
${H}^{*}\left({E}^{\left(9\right)}\right)$
is generated by
$\xi $
and (the pullbacks of )
${\delta}_{1},{\delta}_{2}$
, subject to the relation
${\xi}^{9}={\xi}^{8}\eta $
.
The Chern class of
$\mathcal{\mathcal{M}}$
and GRR. Let
$$\gamma =({\delta}_{1}\otimes 1)(1\otimes {\delta}_{2})({\delta}_{2}\otimes 1)(1\otimes {\delta}_{1}),$$
a class in
${H}^{1,1}({E}^{\left(9\right)}\times E)$
. By [
1,p. 337338]
,
$${c}_{1}(\mathcal{O}(\Delta \left)\right)=\xi \otimes 1+\gamma +9(1\otimes \eta ),$$
hence
$${c}_{1}(\mathcal{\mathcal{M}})=\xi \otimes 1\gamma +3(1\otimes \eta ).$$
Now we apply Grothendieck–Riemann–Roch to
$\mathcal{\mathcal{M}}$
along the projection
${E}^{\left(9\right)}\times E\stackrel{{\pi}_{1}}{\u27f6}{E}^{\left(9\right)}$
. Thus
$$\text{ch}({{\pi}_{1}}_{!}\mathcal{\mathcal{M}})\text{td}\left({E}^{\left(9\right)}\right)={{\pi}_{1}}_{*}\left(\text{ch}\right(\mathcal{\mathcal{M}}\left)\text{td}\right({E}^{\left(9\right)}\times E\left)\right).$$
Since
${R}^{i}{{\pi}_{1}}_{*}\mathcal{\mathcal{M}}=0$
for
$i>0$
and
$\text{td}\left(E\right)=1$
, this simplifies to
$$\text{ch}(\mathcal{G})={{\pi}_{1}}_{*}\left({e}^{{c}_{1}(\mathcal{\mathcal{M}})}\right).$$
Let
${n}_{i}$
denote the
$i$
th Newton class of
$\mathcal{G}$
(i.e., the sum of
$i$
th powers of the Chern roots of
$\mathcal{G}$
), then
$\text{ch}(\mathcal{G})={\sum}_{i\ge 0}{n}_{i}/i!$
. Now we expand the exponential series, and apply
${{\pi}_{1}}_{*}$
term by term, to get
$$\begin{array}{cc}{n}_{0}=3,& {n}_{1}=\frac{1}{2}(6\xi 2\eta ),\end{array}$$  
$$\begin{array}{cc}{n}_{2}=\frac{1}{3}(9{\xi}^{2}+6\xi \eta ),& {n}_{3}=\frac{1}{4}(12{\xi}^{3}12{\xi}^{2}\eta ).\end{array}$$  
Then
$${c}_{3}(\mathcal{G})=\frac{1}{6}{n}_{1}^{3}\frac{1}{2}{n}_{1}{n}_{2}+\frac{1}{3}{n}_{3}=({\xi}^{3}+{\xi}^{2}\eta ).$$
Hence finally
$${c}_{3}({\mathcal{G}}^{*}{)}^{3}=({\xi}^{3}+{\xi}^{2}\eta {)}^{3}=4{\xi}^{8}\eta .$$
Since
${\xi}^{8}\eta $
is the class of a point on
${E}^{\left(9\right)}$
, we deduce that
$\mathcal{A}$
has degree
$4$
.
In order to show that
$\mathcal{A}$
is reduced and hence consists of
$4$
geometric points, we use Kleiman's transversality result (see [
18,Theorem10.8]
).
We can reformulate the entire construction in the following way: start with a smooth
$E$
and hence
$U$
, then specifying a codimension
$3$
subspace
$W\subseteq U$
is tantamount to specifying
$\Lambda $
. Since
$G(3,U)$
is a homogeneous space for
$GL\left(U\right)$
, the intersection is transversal for a general
$W$
, so
$\mathcal{A}$
is reduced. □ The next example is that of a pencil of cubic surfaces. We need to show that
$(3,3,2,8)$
is nondegenerate, the proof is given in § 6 .
Proposition 4.2.
Let
$\Lambda $
be a general pencil of cubic surfaces. Then
$\Lambda $
admits
$3$
polar octahedra.
Proof. The calculation is very similar to Theorem 4.1 . The ideal of
$8$
general points in
${\mathbb{P}}^{3}$
is generated by
$2$
quadrics and
$4$
cubics. Now
${\Lambda}_{2}^{\perp}$
is
$2$
dimensional, hence generates the ideal of a smooth normal elliptic quartic
$E\subseteq {\mathbb{P}}^{3}$
apolar to
$\Lambda $
, and every
$Z\in \mathcal{A}$
is in fact contained in
$E$
. Let
$$U={R}_{3}/({I}_{E}{)}_{3},W={\Lambda}_{3}^{\perp}/({I}_{E}{)}_{3},$$
which are spaces of dimension
$12,10$
respectively. Define
${i}_{1},{i}_{2}$
as before, then the following diagram is a fibre square
$$\mathcal{A}\to G(4,W)\downarrow {i}_{1}\downarrow {E}^{\left(8\right)}\to {i}_{2}G(4,U)$$
Now
${i}_{2}$
is induced by a rank
$4$
bundle
$\mathcal{G}$
on
${E}^{\left(8\right)}$
. The class of
$\mathcal{A}$
in
${E}^{\left(8\right)}$
equals
$${c}_{4}({\mathcal{G}}^{*}{)}^{2}=({\xi}^{4}+{\xi}^{3}\eta {)}^{2}=3{\xi}^{7}\eta .$$
The argument for transversality is the same as before. □ Using similar calculations, we can give alternate proofs of the following results by Schlesinger [26,p. 212] ). The original argument uses
$\vartheta $
functions.
Proposition 4.3 (Schlesinger).

(1)
Let
$\Lambda $
be a general pencil of planar cubics. Fix a general elliptic curve
$E\subseteq {\mathbb{P}}^{2}$
apolar to
$\Lambda $
. Then there are
$3$
polar hexahedra of
$\Lambda $
which are contained in
$E$
.

(2)
Let
$\Lambda $
be a general pencil of planar quartics. Fix a general elliptic curve
$E\subseteq {\mathbb{P}}^{2}$
apolar to
$\Lambda $
. Then there are
$3$
polar octahedra of
$\Lambda $
which are contained in
$E$
.
Proof. We will only prove (1), the argument for (2) is identical in essence. Recall that the ideal of
$6$
general planar points is generated by
$4$
cubics. Since
$(2,3,2,6)$
is nondegenerate^{1 }
,
$\mathcal{A}(6,\Lambda )$
is
$4$
dimensional.
Consider the incidence correspondence
$$\Phi \subseteq \mathcal{A}\times \mathbb{P}{\Lambda}_{3}^{\perp},\Phi =\left\{\right(Z,E):Z\subseteq E\}.$$
The projection
${\pi}_{1}:\Phi \u27f6\mathcal{A}$
is generically a
${\mathbb{P}}^{3}$
bundle, so
$dim\Phi =7$
. Fix a general elliptic curve
$E$
apolar to
$\Lambda $
, and consider the diagram
$$G(3,{\Lambda}_{3}^{\perp}/({I}_{E}{)}_{3}){i}_{1}\downarrow {E}^{\left(6\right)}\to {i}_{2}G(3,{R}_{3}/\left({I}_{E}{)}_{3}\right)$$
As usual,
${i}_{1}$
is the inclusion and
${i}_{2}\left(Z\right)={H}^{0}(E,3HZ)$
. Then
${i}_{2}\left(Z\right)$
lies in the image of
${i}_{1}$
, iff
$Z$
is apolar to
$\Lambda $
. Calculating as before, the product
$\left[\text{image}{i}_{1}\right].\left[\text{image}{i}_{2}\right]$
equals thrice the class of a point. Hence
${\pi}_{2}^{1}\left(E\right)$
must be nonempty. This implies that
${\pi}_{2}:\Phi \u27f6\mathbb{P}{\Lambda}_{3}^{\perp}(\simeq {\mathbb{P}}^{7})$
is dominant. But then it is generically finite, hence for a general
$E$
, the fibre
${\pi}_{2}^{1}\left(E\right)$
consists of
$3$
points. □ It is shown in [
2]
(using a machine calculation) that
$(5,2,3,9)$
is nondegenerate. Now there is a (unique) elliptic sextic curve passing through
$9$
general points of
${\mathbb{P}}^{5}$
. (The classical reference is [
25]
, also see [
5]
for a proof using Gale duality.) Hence if
$\Lambda $
is a general net of quadrics in
${\mathbb{P}}^{5}$
and
$Z$
a set of
$9$
general points apolar to
$\Lambda $
, then the elliptic sextic passing through
$Z$
is apolar to
$\Lambda $
.
Proposition 4.4.
Let
$\Lambda $
be a general net of quadrics in
${\mathbb{P}}^{5}$
. Fix a general elliptic sextic curve
$E\subseteq {\mathbb{P}}^{5}$
apolar to
$\Lambda $
. Then there are
$4$
polar enneahedra of
$\Lambda $
which are contained in
$E$
.
Proof. Similar to above. Use the fact that the ideal of
$9$
general points (resp. an elliptic sextic curve) is generated by
$12$
(resp.
$9$
) quadrics. □
5 The
$(2,3,3,6)$
case
Now we come to London's beautiful calculation in [
22]
, where he determines the number of polar hexahedra of a general net of cubic curves. I have rewritten the proof so as to make it more transparent, but all the key ideas are already in the original.
Let
$\Lambda $
be such a net. By Theorem 2.9 (i),
$\Lambda $
has a finite number of polar hexahedra. We will count them by setting up a correspondence on a certain elliptic curve.
5.1
We begin by motivating the constructions which are to follow. Say
$\{{F}_{1},{F}_{2},{F}_{3}\}$
is a basis of
$\Lambda $
and
$Z=\{{L}_{1},...,{L}_{6}\}$
one of its polar hexahedra. We have expressions
$${F}_{j}={c}_{1j}{L}_{1}^{3}+\cdot \cdot \cdot +{c}_{6j}{L}_{6}^{3},j=1,2,3.$$
Let
$\psi \in {R}_{2}$
be the form which defines the conic passing through
$\{{L}_{2},...,{L}_{6}\}\subseteq \mathbb{P}{S}_{1}$
. Since
$\psi $
annihilates
${L}_{2}^{3},...,{L}_{6}^{3}$
, we have
$\psi \circ {F}_{j}=\text{constant}\times {L}_{1}$
for every
$j$
, so
$\psi \circ \Lambda $
is only a
$1$
dimensional vector space.
It will be seen below (
§ 5.2 ) that all
$\psi $
with this property lie on a curve.
Similarly if
${l}_{1},{l}_{1}^{\prime}\in {R}_{1}$
annihilate
${L}_{1}$
, then the six derivatives
$\{{l}_{1}\circ {F}_{j},{l}_{1}^{\prime}\circ {F}_{j}:j=1,2,3\}$
span only a
$5$
dimensional space. It will be seen below (§ 5.3 ) that all
$2$
dimensional spaces
$\text{span}\{{l}_{1},{l}_{1}^{\prime}\}\subseteq {R}_{1}$
with this property lie on a curve, isomorphic to the previous one.
5.2
Now we come to the actual constructions. The symbol (
$\looparrowleft $
) will appear frequently, it is explained in Remark 5.2 . Consider the vector bundle morphism on
$\mathbb{P}{R}_{2}(={\mathbb{P}}^{5})$
$${f}_{23}:{\mathcal{O}}_{{\mathbb{P}}^{5}}(1)\otimes \Lambda \u27f6{S}_{1}$$
coming from the internal product map of § 2 . Define the degeneracy locus
$\Psi =\{\text{rank}{f}_{23}\le 1\}$
. For a general
$\Lambda $
, it is a degree
$6$
normal elliptic curve in
${\mathbb{P}}^{5}$
(
$\looparrowleft $
). Note that
${\Lambda}_{2}^{\perp}=0$
by the generality of
$\Lambda $
, so
$\text{rank}{f}_{23}$
is exactly
$1$
at each
$\psi \in \Psi $
.
5.3
Now identify
$\mathbb{P}{S}_{1}$
with the Grassmannian
$G(2,{R}_{1})$
, the latter is equipped with a rank two tautological bundle
$\mathcal{\mathcal{B}}\subseteq {R}_{1}\otimes {\mathcal{O}}_{G}$
. The internal product
${f}_{13}$
gives a morphism
$${f}_{13}:\mathcal{\mathcal{B}}\otimes \Lambda \u27f6{S}_{2}$$
The locus
$E=\{\text{rank}{f}_{13}\le 5\}=\{det{f}_{13}=0\}$
is given by a section of
${\mathcal{O}}_{\mathbb{P}{S}_{1}}\left(3\right)$
, hence it is a smooth (
$\looparrowleft $
) degree
$3$
curve in
$\mathbb{P}{S}_{1}$
. By the generality of
$\Lambda $
, the rank of
${f}_{13}$
is exactly
$5$
at every
$L\in E$
(
$\looparrowleft $
).
5.4
We have an isomorphism
$$\alpha :E\u27f6\Psi $$
defined as follows: let
$L\in E$
, and
$U={L}^{\perp}$
. By hypothesis, the space
${f}_{13}(U\otimes \Lambda )$
is
$5$
dimensional, so it is annihilated by a unique form in
$\mathbb{P}{R}_{2}$
, we declare
$\alpha \left(L\right)$
to be this form. It is clear that
${f}_{23}\left(\alpha \right(L)\otimes \Lambda )$
is only
$1$
dimensional (since
$U$
annihilates it), so
$\alpha \left(L\right)\in \Psi $
.
If
$Z$
is as in § 5.1 above, then
$\alpha \left({L}_{1}\right)$
is the conic envelope containing the lines defined by
${L}_{2},...,{L}_{6}$
.
5.5
Define a correspondence
$\mathbb{T}$
on
$E$
as follows:
$(L,M)\in \mathbb{T}$
iff
$M$
lies on the conic defined by
$\alpha \left(L\right)$
. For a fixed
$L$
, there are
$6$
positions of
$M$
such that
$(L,M)\in \mathbb{T}$
. For a fixed
$M$
, the elements of
$\Psi $
which vanish at
$M$
lie on a hyperplane section of
$\Psi $
. Via
${\alpha}^{1}$
, the points of this hyperplane section correspond to
$6$
positions of
$L$
. This shows that
$\mathbb{T}$
has degree
$(6,6)$
and valence zero.
5.6
By the general theory of correspondences (see [
16,§2.5]
), there are
$12$
elements in
$\mathbb{T}$
of the form
$(L,L)$
, they are called the united points of
$\mathbb{T}$
.
Moreover
$\mathbb{T},{\mathbb{T}}^{1}$
have
$72$
common points, i.e., pairs
$(L,M)$
such that
$(L,M),(M,L)\in \mathbb{T}$
. Hence there are
$7212=60$
such pairs where
$(L,M)$
are distinct.
It is clear that starting from
$Z$
, the pairs
$({L}_{1},{L}_{2})$
etc. are common to
$\mathbb{T},{\mathbb{T}}^{1}$
. The next lemma says that the implication is reversible.
Lemma 5.1.
Assume
$(L,M),(M,L)\in \mathbb{T}$
, and
$L\ne M$
. Let the conics
$\alpha \left(L\right),\alpha \left(M\right)$
intersect in
$\{{P}_{1},{P}_{2},{P}_{3},{P}_{4}\}$
. Then
$Z=\{L,M,{P}_{1},...,{P}_{4}\}$
is a polar hexahedron of
$\Lambda $
.
Proof. Recall that the ideal of
$6$
general points is generated by
$4$
cubics. Let
$l,{l}^{\prime}\in {R}_{1}$
be generators of
${L}^{\perp}$
, and
$m,{m}^{\prime}$
of
${M}^{\perp}$
. Consider the four cubic forms
$$\left\{l\alpha \right(L),{l}^{\prime}\alpha (L),m\alpha (M),{m}^{\prime}\alpha (M\left)\right\}.$$
They are linearly independent and each of them vanishes at all points of
$Z$
. Hence together they generate
$({I}_{Z}{)}_{3}$
. Moreover, the definition of
$\alpha $
implies that each of them annihilates
$\Lambda $
. Hence
${I}_{Z}\subseteq {\Lambda}^{\perp}$
. □ Now a polar hexahedron of
$\Lambda $
gives
$2\left(\genfrac{}{}{0ex}{}{6}{2}\right)=30$
pairs
$({L}_{i},{L}_{j})$
common to
$\mathbb{T},{\mathbb{T}}^{1}$
. Alternately, starting from a common point we can reconstruct a polar hexahedron as shown above. Hence, following London, we conclude that
$\Lambda $
has
$60\xf730=2$
polar hexahedra.