Unobstructedness of affine Gorenstein terminal toric fourfolds

This follows from [AvS00, Proposition 4.1, Theorem 6.6].

For the corresponding statement for the Hilbert-basis presentation, see [ACF22a, Lemmas 5.3 and 5.6]. The same argument applies to any finite set of boundary generators and any fixed choice of boundary representatives.

Let \Sigma be a finite common rational polyhedral refinement of the normal fans of P,Q_1, and Q_2. Equivalently, the three support functions \eta_P,~\eta_{Q_1},~ \eta_{Q_2} are linear on every cone \tau\in\Sigma. For such a cone, set

Since \eta_P is linear on \tau, the projection to M identifies B_\tau with the affine semigroup \tau\cap M. Therefore B_\tau has a finite Hilbert basis. Adjoin the Hilbert bases of B_\tau for all cones \tau\in\Sigma. This adds only finitely many boundary generators.

Let b=(c,\eta_P(c))\in\partial_P(S_P). Choose a cone \tau\in\Sigma containing c, express b as a sum of elements of the Hilbert basis of B_\tau, and use this expression to define \partial(b). All generators occurring in this representative have first coordinates in the same cone \tau. Since each \eta_{Q_i} is linear on \tau,

Finally, every s\in S_P has the form s=(c,\eta_P(c))+nR^*, n\in\ensuremath{\mathbb{N}}, so the enlarged boundary elements together with R^* generate S_P. Lemma 3.1 concludes the proof.

Fix i\in\{1,2\}, and write

Since \partial(\ensuremath{\mathbf{k}}) and \ensuremath{\mathbf{k}}_{\boldsymbol\alpha} are the chosen adapted representatives of boundary elements, Lemma 3.4 gives

Moreover, projecting the boundary decomposition to M yields c_{\ensuremath{\mathbf{k}}_{\boldsymbol\alpha}} = c_{\partial(\ensuremath{\mathbf{k}})} - \pi_M(m_{\boldsymbol\alpha}). By mutual compatibility,

and hence \pi_M(m_{\boldsymbol\alpha}) vanishes identically on Q_i. Therefore \eta_{Q_i}(c_{\ensuremath{\mathbf{k}}_{\boldsymbol\alpha}}) = \eta_{Q_i}(c_{\partial(\ensuremath{\mathbf{k}})}) and

Using the definition of \eta_{Q_i}(\,\cdot\,) and the two vanishing equalities above, we conclude that

as required.

Set

By (?), \eta_{Q_i}(\ensuremath{\mathbf{a}}+\ensuremath{\mathbf{k}})=A_i+B_i. Moreover, Lemma 3.5 implies

Expanding (?) and using (?), all terms of degree \mathbf t^{\boldsymbol\beta} are multiples of the same character. Their common coefficient is

which is zero by the Vandermonde’s convolution applied twice [GKP94, Equation (5.27), pp. 170]. Hence R_{\ensuremath{\mathbf{a}},\ensuremath{\mathbf{k}}}=0.

Modulo (t_1,t_2), the equations F_{\ensuremath{\mathbf{k}}} reduce to f_{\ensuremath{\mathbf{k}}}, and the relations R_{\ensuremath{\mathbf{a}},\ensuremath{\mathbf{k}}} reduce to r_{\ensuremath{\mathbf{a}},\ensuremath{\mathbf{k}}}. By Lemma 3.1, the latter generate all relations among the equations of the central fibre, while Proposition 3.6 provides their lifts. By the equational flatness criterion [Fil26, Lemma 3.5], the equations therefore define a formal flat deformation over \ensuremath{\mathbb{C}}[[t_1,t_2]].

The relation s_{F_1}+s_{F_2}=R^* gives R^*-s_{F_1}=s_{F_2}, R^*-s_{F_2}=s_{F_1}, and

Since s_{F_1},s_{F_2}\in\sigma^\vee, for every vertex v of P, the values \varphi_{R^*-s_{F_1}}(v) and \varphi_{R^*-s_{F_2}}(v) are nonnegative integers whose sum is \varphi_{R^*}(v)=1. Hence \varphi_{R^*-s_{F_1}}(v)\varphi_{R^*-s_{F_2}}(v)=0.

Moreover, since \pi_M(R^*)=0, we have \pi_M(R^*-s_{F_1})=-\pi_M(R^*-s_{F_2}), and therefore

Together with Q_i\subset\ker\pi_M(R^*-s_{F_i}) for i=1,2, this shows that the two deformation pairs satisfy the compatibility conditions. The result now follows from Corollary 3.7.

After replacing the ambient lattice by \operatorname{span}_{\mathbb R}(P-P)\cap N, we may assume that N\cong \mathbb Z^d and that P is full-dimensional. There are exactly 2^d congruence classes in N/2N\cong(\mathbb Z/2\mathbb Z)^d. If P had more than 2^d vertices, then two distinct vertices v,w would be congruent modulo 2N. Hence \frac{v+w}{2}\in N. Since this point lies in the relative interior of the segment [v,w]\subset P, it is a non-vertex lattice point of P, which contradicts the assumption that P is empty.

By Lemma 4.1, P has either three or four vertices. Since P is empty, every edge is primitive and P has no interior lattice points. Thus Pick’s formula [BR15, Theorem 2.8] gives \operatorname{Area}(P)=\frac{r}{2}-1, where r is the number of vertices of P.

If r=3, then \operatorname{Area}(P)=\frac12. Hence the normalized area of P is one, so P is a unimodular triangle and therefore lattice equivalent to \Delta_2.

Suppose that r=4, and write the vertices as v_0,v_1,v_2,v_3 in cyclic order. Then \operatorname{Area}(P)=1. Each diagonal divides P into two lattice triangles. Since every nondegenerate lattice triangle has area at least \frac12, all four triangles arising from the two diagonals have area \frac12 and are therefore unimodular.

After a lattice equivalence, we may assume that v_0=0, v_1=e_1, and v_2=e_2. Writing v_3=(a,b), the unimodularity and cyclic ordering of \operatorname{conv}\{v_0,v_2,v_3\} and \operatorname{conv}\{v_0,v_1,v_3\} give -a=1 and b=1. Thus v_3=e_2-e_1, and P is lattice equivalent to the unit square.

If \dim X\leq2, then X\cong \mathbb A^1 or X\cong \mathbb A^2. Hence X is smooth and T_X^2=0. Suppose now that \dim X=3. Then the associated Gorenstein polytope P is two-dimensional and, by terminality, empty.

By Definition 2.5 and Lemma 4.2, P is lattice equivalent either to the standard triangle \Delta_2

or to the unit square \square. If P is lattice equivalent to \Delta_2, then the primitive generators of the cone over P form a lattice basis. Hence the cone is unimodular and X\cong\mathbb A^3; thus T_X^2=0. If P is lattice equivalent to the unit square, then the Hilbert basis of \sigma^\vee\cap\widetilde M consists of four elements u_1,u_2,u_3,u_4, with the unique primitive relation u_1+u_4=u_2+u_3. Consequently,

which is the three-dimensional ordinary double point. In particular, X is a hypersurface and T_X^2=0. So X is unobstructed.

If there exists a vertex v of P such that \varphi_r(v)>1, then the assertion follows from Theorem 2.6(1). Suppose therefore that \varphi_r\leq1 on P. Since r\neq R^*, the section F_r=P\cap(\varphi_r=1) is a proper face of P, and hence \dim F_r\leq2. The assertion now follows from Theorem 2.6(3).

The assertion for \dim X_P\leq3 follows from Lemma 4.3. We therefore assume that \dim X_P=4, so that P is an empty three-dimensional lattice polytope.

By Proposition 4.5, for each square facet F\in\mathcal S(P), the corresponding nonzero tangent components occur in degrees -(R^*-ns_F), with n\geq 1. We call the component corresponding to n=1 the primitive tangent component associated with F.

We first restrict to the finite-dimensional primitive tangent subspace

Thus V_{\mathrm{prim}} consists of the Gorenstein-degree component together with the primitive tangent component associated with each square facet of P.

For every F\in\mathcal S(P), put m_F:=R^*-s_F. By (Proof),

Consequently, m_F\in\sigma^\vee\setminus\{0\}.

By Lemma 4.2, the facet F is lattice equivalent to the unit square. Hence there exist a vertex p\in F and primitive lattice vectors e_1,e_2\in N such that

Since e_1 and e_2 are parallel to F, they are annihilated by \pi_M(s_F), and therefore also by \pi_M(m_F)=-\pi_M(s_F). Thus Q_F:=[0,e_1] is a primitive lattice segment contained in \ker\pi_M(m_F), parallel to an edge of F, and a Minkowski summand of F.

Then (m_F,Q_F) is a deformation pair. Let t_F denote the parameter of the corresponding one-parameter deformation and set

By restricting (?) to the t_F-axis, its first-order part is

Hence \xi_F is the Minkowski-summand class determined by Q_F. Under the identification of Theorem 2.3, this class is nonzero because Q_F is a nontrivial Minkowski summand of the unit square F. Since T^1_{X_P}(-m_F) is one-dimensional, \xi_F is a generator.

Let d:=\dim_{\ensuremath{\mathbb{C}}}T^1_{X_P}(-R^*) and choose a basis \zeta_1,\ldots,\zeta_d of T^1_{X_P}(-R^*). Introduce variables z_1,\ldots,z_d of degree R^*, and put

where

Let

and, for k\geq1, set

Since T^1_{X_P} may be infinite-dimensional, no complete local noetherian base can carry independent parameters for all its tangent directions. We therefore begin with the finite-dimensional subspace V_{\mathrm{prim}} and construct a formal deformation in all orders.

More precisely, for each k\geq1, we construct inductively a graded k-th order deformation

of X_P. These deformations are compatible in the sense that the pullback of \mathcal X_{\mathrm{prim}}^{(k+1)} to \operatorname{Spec}R_{\mathrm{prim}}^{(k)} is isomorphic to \mathcal X_{\mathrm{prim}}^{(k)}. Their Kodaira–Spencer maps identify the coordinate tangent vectors with

The total degree-zero element

determines a graded first-order deformation \mathcal X_{\mathrm{prim}}^{(1)} \longrightarrow \operatorname{Spec}R_{\mathrm{prim}}^{(1)}.

Suppose inductively that \mathcal X_{\mathrm{prim}}^{(k)} has been constructed. The obstruction to lifting it to R_{\mathrm{prim}}^{(k+1)} is a sum of terms

where

and

By Proposition 4.7, this coefficient can be nonzero only if

Suppose first that |\boldsymbol\alpha|\geq1. Equation (?) is equivalent to

All the terms in this sum belong to the strongly convex cone \sigma^\vee. Hence every term is zero. Since R^*\neq0 and m_F\neq0, we obtain |\boldsymbol\alpha|=1 and

a_F=0 for every F. This gives a monomial of ordinary order one, whereas an obstruction to lifting R_{\mathrm{prim}}^{(k)} to R_{\mathrm{prim}}^{(k+1)} has ordinary order k+1\geq2. Consequently, no obstruction term can involve one of the variables z_i.

It remains to consider monomials of the form

Their degree is R^* precisely when

By Lemma 5.1, this can occur only when there are two parallel square facets F_1,F_2 such that

and s_{F_1}+s_{F_2}=R^*. Thus the only possible obstruction monomial is t_{F_1}t_{F_2}. In particular, such an obstruction can occur only in the passage from first to second order.

Let \omega_{F_1,F_2} \in T^2_{X_P}(-R^*) denote the coefficient of this monomial in the second-order obstruction. Put

Consider the graded homomorphism R_{\mathrm{prim}}^{(1)} \longrightarrow B^{(1)} given by

and by sending all the remaining variables to zero.

By Corollary 3.8, there is a formal two-parameter deformation over B whose Kodaira–Spencer classes in the u_1- and u_2-directions are \xi_{F_1} and \xi_{F_2}, respectively. Its first-order truncation therefore represents the pullback of \mathcal X_{\mathrm{prim}}^{(1)}, while its second-order truncation provides a lifting of that first-order deformation to B^{(2)}.

By functoriality of obstruction classes, the pullback of the second-order obstruction is consequently zero. On the other hand, this pullback is

Since

it follows that \omega_{F_1,F_2}=0.

Equivalently, after choosing compatible parameter coordinates, the comparison homomorphism has the form

Therefore

so a higher-order change of parameter cannot alter the coefficient of the mixed quadratic term.

We have proved that every second-order obstruction vanishes. For orders at least three, the degree calculation above shows that there is no monomial whose coefficient could lie in a nonzero homogeneous component of T^2_{X_P}. We may therefore construct inductively compatible graded liftings

for all k\geq1. For a complete local ring (R,\mathfrak m), we write \operatorname{Spf}R for its formal spectrum, namely the affine formal scheme determined by the infinitesimal thickenings \operatorname{Spec}(R/\mathfrak m^{k+1}), k\geq0. Passing to the inverse limit gives a formal deformation \mathcal X_{\mathrm{prim}} \longrightarrow \operatorname{Spf}R_{\mathrm{prim}} whose Kodaira–Spencer map is an isomorphism onto

We next incorporate the remaining homogeneous tangent directions. For every square facet F, let x_F\in\ensuremath{\mathbb{C}}[S_P] denote the character \chi^{s_F}. For n\geq1, put r_{F,n}:=R^*-ns_F. By (Proof),

It follows directly from the definition of the sets K_\tau^r that K_\tau^{r_{F,n}} = K_\tau^{r_{F,1}} for every face \tau\preceq\sigma. Indeed, at a vertex of F the upper bound is one for every n, while at a vertex outside F it is nonpositive, so the corresponding K-set is empty.

By [AS98, Theorem 3.3], multiplication by x_F^{n-1} is induced on the complexes computing T^1_{X_P} by the inclusions K_\tau^{r_{F,n}} \subseteq K_\tau^{r_{F,1}}. Since these inclusions are equalities, multiplication induces an isomorphism

Consequently,

We shall use the following flatness lemma for the coordinate-dependent substitution below.

We now perform the substitution at the level of equations and relations. Let

and put

After lifting the ambient coordinate functions to the primitive family, write

The ring \mathcal P_{\mathrm{prim}} is Noetherian, so we may choose finite homogeneous generators

of I_{\mathrm{prim}}. Since \mathcal A_{\mathrm{prim}} is flat over R_{\mathrm{prim}}, their reductions g_1,\ldots,g_e generate I_P. Choose finitely many homogeneous syzygies among the G_i as in Lemma ?, and let v_{\mathrm{prim}} be the matrix whose columns are these syzygies. Let u_{\mathrm{prim}} be the row matrix with components G_i, and let u_0,v_0 denote their reductions modulo \mathfrak m_{\mathrm{prim}}. We then have a homogeneous complex

with u_{\mathrm{prim}}\circ v_{\mathrm{prim}}=0, whose reduction is the finite homogeneous presentation

In particular, the columns of v_0 generate all relations among the central equations g_i.

For every F\in\mathcal S(P), choose a homogeneous polynomial, still denoted by x_F, in \mathcal P_0 whose image in A_0 is \chi^{s_F}. For N\geq1, introduce parameters

of degree \deg(t_{F,n})=R^*-ns_F, and put

Set

Every summand is homogeneous of degree

Let \mathcal P_N := R_N\widehat\otimes_{\ensuremath{\mathbb{C}}}\mathcal P_0. There is a continuous homogeneous \ensuremath{\mathbb{C}}-algebra homomorphism

which is the identity on \mathcal P_0 and on the variables z_i, and satisfies \Phi_N(t_F)=\widehat t_F^{(N)}. Notice that \Phi_N is not asserted to be a base change homomorphism: the coefficients x_F^{n-1} belong to the ambient coordinate ring.

Apply \Phi_N to both matrices in (?), and set

Then

Moreover, modulo the maximal ideal of R_N, the resulting complex reduces to (?). Thus

is a complex which is exact except possibly at \mathcal P_N^{\oplus e}, and whose reduction over the closed point is exact.

The homomorphism \Phi_N preserves the parameter ideals and induces the identity on the closed fibre. The equality u_N\circ v_N=0 shows that all substituted syzygies still vanish, and their reductions generate all syzygies among the central equations. Lemma ? therefore implies that \mathcal A_N is flat over R_N. Hence it defines a formal deformation \mathcal X_N\longrightarrow\operatorname{Spf}R_N of X_P. Writing G_i^{(N)}:=\Phi_N(G_i), we have

By the \ensuremath{\mathbb{C}}[S_P]-module structure on T^1_{X_P}, this gives

By (?), this is a generator of T^1_{X_P}\bigl(-(R^*-ns_F)\bigr). The z_i-directions map to the chosen basis \zeta_1,\ldots,\zeta_d. Consequently, the Kodaira–Spencer map of \mathcal X_N is an isomorphism onto

The families \mathcal X_N are compatible as N varies: \mathcal X_{N+1} restricts to \mathcal X_N after setting t_{F,N+1}=0 for every F. Furthermore,

We claim that every deformation of X_P over a local Artinian \ensuremath{\mathbb{C}}-algebra B is induced from \mathcal X_N for some N.

We prove this by induction on the length of B. The assertion is clear for B=\ensuremath{\mathbb{C}}. Choose a small extension

and let \mathcal Y be a deformation of X_P over B. By induction, its restriction \overline{\mathcal Y} := \mathcal Y\otimes_B\overline B is induced from \mathcal X_N by a homomorphism \varphi_N\colon R_N\longrightarrow\overline B for some N.

Since R_N is a formal power-series ring in finitely many variables, \varphi_N lifts to a homomorphism

Let \mathcal Y_0 be the corresponding pullback of \mathcal X_N. Then \mathcal Y and \mathcal Y_0 are two liftings of \overline{\mathcal Y} across the same small extension. Their difference is represented by an element \delta\in T^1_{X_P}\otimes_{\ensuremath{\mathbb{C}}}J.

Since T^1_{X_P} is the algebraic direct sum of its homogeneous components, \delta has finite homogeneous support. Hence there exists N'\geq N such that \delta\in V_{N'}\otimes_{\ensuremath{\mathbb{C}}}J. Regard \widetilde\varphi_N as a homomorphism R_{N'}\longrightarrow B by sending the additional parameters to zero. Since the Kodaira–Spencer map of \mathcal X_{N'} is an isomorphism onto V_{N'}, there exists

whose image under the Kodaira–Spencer map is \delta.

Because the extension is small, one has \mathfrak m_BJ=0. We may therefore modify the parameter homomorphism by setting

for each parameter q, where \overline q denotes its class in \mathfrak m_{R_{N'}}/\mathfrak m_{R_{N'}}^2. This again defines a local homomorphism

lifting the original homomorphism to \overline B. The corresponding change of the pullback deformation is exactly \delta. Hence the pullback defined by \widetilde\varphi_{N'}' is isomorphic to \mathcal Y, proving the claim.

Finally, let B'\longrightarrow B be a small extension of local Artinian \ensuremath{\mathbb{C}}-algebras and let \mathcal Y be a deformation of X_P over B. By the claim, \mathcal Y is induced from \mathcal X_N, for some N, by a homomorphism

Since R_N is a formal power-series ring in finitely many variables, this homomorphism lifts to a homomorphism R_N\longrightarrow B'. Pulling back \mathcal X_N gives a lifting of \mathcal Y to B'. Therefore

is surjective for every small extension B'\to B. Thus \operatorname{Def}_{X_P} is formally smooth, and X_P is unobstructed.