Two results regarding the variation of K-moduli

The proof is a combination of Propositions 3.5 and 3.7.

By Corollary 4.4, we directly conclude the first statement in Theorem 1.3. We divide the rest proof into two steps. First, we construct the connecting map \mathcal{M}^K_{n, \vec{d},\vec{a}}\to \mathcal{M}^K_{n,\vec{d}}. Then we show that it descends to a surjective map between moduli spaces.

Step 1. By Proposition 4.8, given (X, \sum_{j=1}^ka_jB_j)\in \mathcal{M}^K_{n, \vec{d}, \vec{a}}, we have that \cap_{j=1}^k B_j is also K-semistable of dimension n-k. Thus we can define the morphism

Actually, the above morphism is only defined on \mathbb{C}-points, however, we could extend it to be a morphism between stacks as follows. For convenience, we only define the morphism for \mathcal{M}^K_{n, \vec{d}, \vec{a}}(W)\to \mathcal{M}^K_{n, \vec{d}}(W), where W is a normal scheme(For arbitrary base, we take the concept of K-flatness introduced by [Kollar19families]. Refer to [XZ20b, Section 2.6] for the definition of the family.). Take [(\mathcal{X}, \sum_{j=1}^ka_j\mathcal{B}_j)\to W]\in \mathcal{M}^K_{n, \vec{d}, \vec{a}}(W), i.e.

1. \mathcal{X}\to W is a flat family such that -K_{\mathcal{X}/W} is \mathbb{Q}-Cartier and ample over W;

2. \mathcal{B}_j is a prime divisor on \mathcal{X} with \mathcal{B}_j\sim_{\mathbb{Q}, W} -r_jK_{\mathcal{X}/W} and r_j=\frac{d_j}{n+1};

3. (\mathcal{X}_t, \sum_{j=1}^ka_j\mathcal{B}_{j,t})\in \mathcal{M}^K_{n, \vec{d}, \vec{a}}(\mathbb{C}) for every closed point t\in W.

By Proposition 4.8 and Remark 4.9, we see that \mathcal{B}_{i_1,t}\cap \mathcal{B}_{i_2,t}\cap...\cap \mathcal{B}_{i_p,t} is a Fano variety for any closed point t\in W and non-repeating subset \{i_1,...,i_p\}\subset\{1,2,...,k\}, thus \mathcal{B}_{i_1}\cap \mathcal{B}_{i_2}\cap...\cap \mathcal{B}_{i_p}\to W is flat by [Kollar23, Lemma 10.58]. By Proposition 4.8 (2), we see that the family \mathcal{B}_1\cap \mathcal{B}_2\cap...\cap \mathcal{B}_k\to W could be obtained by taking k times adjunction, and every time we take the adjunction, there is no difference part. By Proposition 4.8 (3), we see that \mathcal{B}_1\cap \mathcal{B}_2\cap...\cap \mathcal{B}_k\to W is a \mathbb{Q}-Gorenstein family of K-semistable Fano varieties. Thus the connecting morphism could be defined by sending [(\mathcal{X}, \sum_{j=1}^ka_j\mathcal{B}_j)\to W] to [\mathcal{B}_1\cap \mathcal{B}_2\cap...\cap \mathcal{B}_k\to W].

Step 2. The morphism between stacks naturally descends to a morphism between K-moduli spaces by the universal property of good moduli spaces (e.g. [Alper13, Theorem 4.16 (vi)]):

We aim to show that \phi is surjective. Suppose [Y]\in M^K_{n, \vec{d}} is of the form S_1\cap...\cap S_k deduced by a log smooth pair (\mathbb{P}^n, \sum_{j=1}^k S_j), by Corollary 4.4, [(\mathbb{P}^n, \sum_{j=1}^ka_jS_j)] is a preimage. Suppose [Y] is not of the form S_1\cap...\cap S_k deduced by a log smooth pair (\mathbb{P}^n, \sum_{j=1}^k S_j). Then there is a flat family of K-semistable Fano varieties \mathcal{Y}\to C over a smooth pointed curve 0\in C such that -K_{\mathcal{Y}/C} is ample over C, \mathcal{Y}_0\cong Y, and \mathcal{Y}_t is of the form S_1\cap...\cap S_k deduced by a log smooth pair (\mathbb{P}^n, \sum_{j=1}^k S_j). Let \mathcal{Y}^*\to C^*:=C\setminus \{0\} be the family obtained via base change under the inclusion C^*\to C. We claim that there exist an étale morphism C'^*\to C^* and a family

such that for any t\in C'^*, the fiber (\mathcal{X}_t, \sum_{j=1}^k\mathcal{S}_{j, t}) is isomorphic to some log smooth pair (\mathbb{P}^n, \sum_{j=1}^k S_j) and \cap_{j=1}^k \mathcal{S}_{j, t}\cong \mathcal{Y}'^*_t, where \mathcal{Y}'^*\to C'^* is the pulling back of \mathcal{Y}^*\to C^* along C'^*\to C^*. To see this, we denote by \mathbb{P}_j:=|\mathcal{O}_{\mathbb{P}^n}(d_j)| and \mathcal{E}_j\subset \mathbb{P}^n\times \mathbb{P}_j the universal divisor. Putting \mathcal{D}_j:= \mathbb{P}_1\times...\mathbb{P}_{j-1}\times\mathcal{E}_j\times \mathbb{P}_{j+1}\times ...\times \mathbb{P}_k and T:=\mathbb{P}_1\times \mathbb{P}_2\times...\times \mathbb{P}_k, we consider the following universal family

Let Z\subset T be the subset of T parametrizing those log smooth fibers (\mathbb{P}^n, \sum_{j=1}^{k}\mathcal{D}_{j,t}) where \mathcal{D}_{j,t}, j=1,...,k, are mutually different. We denote the family by

Let \mathcal{U}\to U be the universal family of complete intersections of k different hypersurfaces in \mathbb{P}^n of degrees d_1, d_2,...,d_k respectively, then we naturally have the following commutative diagram

Note that there is a morphism C^*\to U such that \mathcal{Y}^*\to C^* is the pulling back of \mathcal{U}\to U via the base change C^*\to U. Taking base change again, we obtain a family of log smooth pairs

Up to shrinking C around 0\in C, one could find a section C'^*\to Z\times_U C^* such that C'^*\to C^* is an étale morphism. Pulling back the family \pi via this section, we get the required family (\mathcal{X}, \sum_{j=1}^k\mathcal{S}_j)\to C'^*.

Note that (\mathcal{X}, \sum_{j=1}^ka_j\mathcal{S}_j)\to C'^* is a family of K-semistable log Fano pairs, by the properness of K-moduli space, we could find a complete family

up to a finite base change and the central fiber (\tilde{\mathcal{X}}_{0'}, \sum_{j=1}^ka_j\tilde{\mathcal{S}}_{j,{0'}}) is also K-semistable. By Proposition 4.2, \cap_{j=1}^k \tilde{\mathcal{S}}_{j,{0'}} is K-semistable and [Y]=[\cap_{j=1}^k \tilde{\mathcal{S}}_{j,{0'}}] by the separatedness of K-moduli space (e.g. [BX19]). Thus [(\tilde{\mathcal{X}}_{0'}, \sum_{j=1}^ka_j\tilde{\mathcal{S}}_{j,{0'}})] is a preimage of [Y] and \phi is surjective. The proof is complete.

First, note that \sum_{j=1}^k c_j\mathcal{B}_j\sim_{\mathbb{Q}, T} -(\sum_{j=1}^k c_jl_j)K_{X\times T/T}. By Lemma 3.3, we have

It suffices to compute {\pi_{\vec{c}}}_*((-K_{X\times T/T})^n\mathcal{B}_j). Recall that

and {\pi_j}_*((-K_{X\times \mathbb{P}_j/\mathbb{P}_j})^n\mathcal{D}_j)=\mathcal{O}_{\mathbb{P}_j}(v) for \pi_{j}: X\times \mathbb{P}_j\to \mathbb{P}_j. Thus, we have

where 0_i means that the i-th position is 0. The proof is finished.

The idea of the proof is essentially the same as that of [Zhou21a, Theorem 1.1].

Suppose (X, \epsilon\sum_{j=1}^kc_jD_j) is K-semistable, to show that (D_1,...,D_k) is GIT-semistable under the {\rm {Aut}}(X)-action with respect to the linearization \mathcal{O}(c_1,...,c_k), by Remark 3.2, it suffices to show that \lambda_{{\rm {CM}}, \epsilon}(\vec{c}) is proportional to \mathcal{O}(c_1, ..., c_k). This is exactly Lemma 3.4. Now we suppose (X, \epsilon\sum_{j=1}^kc_jD_j) is K-polystable, then (D_1,...,D_k) is GIT-semistable as we have shown. Let \rho: \mathbb{C}^*\to {\rm {Aut}}(X) be any one parameter subgroup such that the limit point \lim_{t\to 0}\rho(t)\cdot (D_1,...,D_k) is GIT-polystable, then the GIT-weight corresponding to the \mathbb{C}^*-action is zero, which implies that the Futaki invariant of the test configuration induced by the \mathbb{C}^*-action vanishes. By [LWX21], we see that

Thus (D_1,...,D_k) is GIT-polystable.

Conversely, suppose (D_1,...,D_k) is GIT-semistable under {\rm {Aut}}(X)-action with respect to the linearization \mathcal{O}(c_1,...,c_k), we aim to show that (X, \epsilon\sum_{j=1}^kc_jD_j) is K-semistable. First, observe that there is a family (\mathcal{X}, \sum_{j=1}^k \mathcal{G}_j)\to C over a smooth pointed curve 0\in C such that

1. there is a morphism C\setminus{0}\to \mathbb{P}_j for each j;

2. (\mathcal{X}\setminus \mathcal{X}_0,\mathcal{G}_j\setminus \mathcal{G}_{j,0}) is obtained via pulling back (X\times \mathbb{P}_j,\mathcal{D}_j) under the morphism C\setminus \{0\}\to \mathbb{P}_j;

3. (\mathcal{X}_0, \sum_{j=1}^k\mathcal{G}_{j,0})\cong (X, \sum_{j=1}^kD_j);

4. (\mathcal{X}_t, \epsilon\sum_{j=1}^kc_j\mathcal{G}_{j,t}) is K-semistable for any t\in C\setminus \{0\}.

By the properness of K-moduli, up to a finite base change, one could replace the central fiber (\mathcal{X}_0, \epsilon\sum_{j=1}^kc_j\mathcal{G}_{j,0})\cong (X, \epsilon\sum_{j=1}^kc_jD_j) with a K-semistable log Fano pair (X', \epsilon\sum_{j=1}^kc_j D_j') and we denote the new family by (\mathcal{X}', \epsilon\sum_{j=1}^kc_j \mathcal{G}_j')\to C for convenience. We claim that it suffices to show that \mathcal{X}'_0\cong X. Suppose \mathcal{X}_0'\cong X, then (D_1',...,D_k')\in \mathbb{P}_1\times ...\times \mathbb{P}_k and it is GIT-semistable. By the separatedness of GIT moduli space, we know that (D_1,...,D_k) and (D_1',...,D_k') lie on the same S-equivalence class under {\rm {Aut}}(X)-action. Thus, (X, \epsilon\sum_{j=1}^kc_jD_j) is K-semistable. Next, we turn to prove that \mathcal{X}_0'\cong X. Since \epsilon is sufficiently small, by Theorem [Zhou21a, Theorem 3.3], we know that \mathcal{X}_0' is K-semistable. Note that X is K-polystable, thus \mathcal{X}_0'\cong X by [Zhou21a, Theorem 3.4]. By now, we have finished the proof from GIT-semistability to K-semistability. Finally, we suppose (D_1,...,D_k) is GIT-polystable, then (X, \epsilon\sum_{j=1}^kc_jD_j) is K-semistable as we have shown. To show that it is K-polystable, let (X_0, \epsilon\sum_{j=1}^kc_jD_{j,0}) be its K-polystable degeneration. By the same approach as above, we know that the ambient space of the K-polystable degeneration is preserved. Thus

is GIT-polystable by the previous paragraph. By the separatedness of GIT-moduli space, we see (D_1,...,D_k)\cong (D_{1,0},...,D_{k,0}). Therefore, (X, \epsilon\sum_{j=1}^kc_jD_j) is itself K-polystable.

The proof is a combination of Proposition 3.5 and [Zhou23b, Theorem 1.4 and Corollary 1.5].

By the choice of a_j, j=1,...,k, we know

for any 1\leq i\leq k. We show that S_1 is irreducible. Suppose not, there is an irreducible component of S_1, denoted by D, satisfying

The inequality holds since

Thus, we get a contradiction to the K-semistability. Similarly, all other S_j’s are irreducible. Next we show that S_j’s are mutually different. Otherwise we may suppose S_1=S_2 up to reordering, then

which is again a contradiction to the K-semistability. The proof is finished.

We divide the proof into several steps.

Step 1. Since \beta_{\mathbb{P}^n, \sum_{j=1}^ka_jS_j}(S_1)=0, by Remark 2.4, one could produce a special test configuration (\mathcal{X}, \sum_{j=1}^k a_j\mathcal{G}_j)\to \mathbb{A}^1 of (\mathbb{P}^n, \sum_{j=1}^{k}a_j S_j) such that the central fiber is the projective cone over (S_1, \sum_{j=2}^ka_j{S_j}|_{S_1}) with the polarization \mathcal{O}_{S_1}(d_1), and

Thus, the central fiber (\mathcal{X}_0, \sum_{j=1}^k a_j\mathcal{G}_{j,0}) is also K-semistable. Note that we have

for

By Lemma 2.2, we see that (S_1, \sum_{j=2}^ka_j{S_j}|_{S_1}) is K-semistable.

Step 2. Applying induction, we assume

is K-semistable. We aim to show that

is also K-semistable. First, observe that (a_{l+1}, a_{l+2}, ...,a_k) is the unique solution of the following linear equations:

By a similar proof of Proposition 4.1, one concludes that E_j^{(l)}, j=l+1,...,k, are different from each other and every E_j^{(l)} is irreducible. Note that

thus similar to Step 1, there exists a special test configuration (\mathcal{X}', \sum_{j=l+1}^k a_j\mathcal{G}'_j)\to \mathbb{A}^1 of (X_l, \sum_{j=l+1}^ka_jE_j^{(l)}) such that the central fiber is the projective cone over

with polarization \mathcal{O}_{X_{l+1}}(d_{l+1}), and

Hence, the central fiber (\mathcal{X}'_0, \sum_{j=l+1}^k a_j\mathcal{G}'_{j,0}) is also K-semistable. Note that we have

for

By Lemma 2.2, we see that the following log pair is K-semistable:

Step 3. By Step 2, we see that the following log pair

is K-semistable and S_{d_k}\cap S_{k-1}\cap...\cap S_1 is irreducible. Since

there exists a special test configuration (\mathcal{X}'', a_k\mathcal{G}_k'')\to \mathbb{A}^1 of

such that the central fiber is K-semistable and is the projective cone over S_1\cap ...\cap S_k with polarization \mathcal{O}_{S_1\cap ...\cap S_k}(d_k). Moreover, we have

for

Hence S_1\cap ...\cap S_k is K-semistable by Lemma 2.2. The proof is finished.

In the log smooth setting, the proof of Proposition 4.2 can be reversed by Lemma 2.2. To see this, first note that

can be degenerated to a projective cone over S_k\cap S_{k-1}\cap...\cap S_1, denoted by (\mathcal{X}''_0, a_k\mathcal{G}''_{k,0}) (see the notation in Step 3 of the proof of Prop 4.2). Since (\mathcal{X}''_0, a_k\mathcal{G}''_{k,0}) is K-semistable, we see

is also a K-semistable log Fano pair. Continuing to reverse the proof of Proposition 4.2, we see

is a K-semistable log Fano pair for any 1\leq l\leq k-1. Finally, reversing Step 1 of the proof of Prop 4.2, we see that (\mathbb{P}^n, \sum_{j=1}^ka_jS_j) can be degenerated to a projective cone over (S_1, \sum_{j=1}^ka_jS_j\cap S_1), denoted by (\mathcal{X}_0, \sum_{j=1}^k a_j\mathcal{G}_{j,0}). Since (\mathcal{X}_0, \sum_{j=1}^k a_j\mathcal{G}_{j,0}) is a K-semistable log Fano pair, it is deduced that (\mathbb{P}^n, \sum_{j=1}^ka_jS_j) is itself a K-semistable log Fano pair.

We can clearly see these equivalences from the proofs of Proposition 4.2 and Proposition 4.3.

Note that the family (\mathcal{X}, \mathcal{B})\to C is log smooth over C\setminus 0, we easily see that \mathcal{X}\to C is a \mathbb{Q}-Gorenstein family of Fano varieties by conditions (a) and (b). Since \mathcal{X} is smooth in codim 2, we have the following adjunction formulas

By condition (c), we see that -K_\mathcal{B} is ample over C. Combining condition (d), we see that \mathcal{B}\to C is also a \mathbb{Q}-Gorenstein family of Fano varieties if \mathcal{B}_0 has klt type singularities, which concludes (2). By adjunction, we have

We aim to show that {\rm {Diff}}_{\mathcal{B}_0}(0)=0 which concludes (3). To see this, first note that {\rm {Diff}}_{\mathcal{B}_0}(0)\geq 0 (e.g. [Kollar13, Proposition 4.5]). Observe the following:

which implies the following:

On the other hand, by formulas (\star) and (\star\star), we have

Thus, we derive -(K_{\mathcal{B}_0}+{\rm {Diff}}_{\mathcal{B}_0}(0))\sim_\mathbb{Q} -K_{\mathcal{B}_0} and it is clear that {\rm {Diff}}_{\mathcal{B}_0}(0)=0.

Let (\mathcal{X}, \sum_{j=1}^ka_j\mathcal{B}_j)\to C be the degeneration family over a pointed curve 0\in C as in Definition 4.6, for which we have (\mathcal{X}_0, \sum_{j=1}^ka_j\mathcal{B}_{j,0})\cong (X, \sum_{j=1}^ka_jB_j). Note that the degenerating family is log smooth over C\setminus 0, and each \mathcal{B}_j is a prime divisor on \mathcal{X} satisfying \mathcal{B}_j\sim_{\mathbb{Q}, C}-r_jK_{\mathcal{X}/C} for r_j=\frac{d_j}{n+1}<1.

Step 1. Since (X, \sum_{j=1}^k a_j B_j) is K-semistable, then the Proposition 4.1 stated for (\mathbb{P}^n, \sum_{j=1}^ka_jS_j) also applies to (X, \sum_{j=1}^ka_jB_j). Indeed, we observe that (a_1,...,a_k) is still the unique solution of the following linear equations

since we still have the following for a general fiber (\mathbb{P}^n, \sum_{j=1}^ka_jS_j):

Thus it is similar for (X, \sum_{j=1}^kB_j) to derive that B_j’s are irreducible and mutually different. Since \beta_{X, \sum_{j=1}^ka_jB_j}(B_1)=0, by [BLZ22], we know that B_1 induces a special test configuration of (X, \sum_{j=1}^ka_jB_j), denoted by

where the central fiber is also a K-semistable log Fano pair and (\mathcal{X}'_0, \sum_{j=1}^ka_j\mathcal{B}'_{j,0}) is the projective orbifold cone over B_1. Hence B_1 is a klt type divisor on X (e.g. [Kollar13, Section 3.1]). Applying Lemma 4.7 to (\mathcal{X}, \mathcal{B}_1)\to C, we see that B_1\cong \mathcal{B}_{1,0} is a Fano variety, and (K_X+B_1)|_{B_1}=K_{B_1}. Since there is no different part, we see that X is smooth along the codimension two points of B_1, which implies that B_j|_{B_1}=B_j\cap B_1 are all prime divisors on B_1 for 2\leq j\leq k. By the orbifold cone construction (e.g. [Kollar04seifert, Section 4] or [Zhuang24, section 3.1]), we see that (\mathcal{X}'_0, \sum_{j=1}^ka_j\mathcal{B}'_{j,0}) is the projective cone over (B_1, \sum_{j=2}^k a_j B_j\cap B_1). By Lemma 2.2, we see that (B_1, \sum_{j=2}^ka_j B_j\cap B_1) is a K-semistable log Fano pair.

Step 2. Now, we consider the degenerating family (\mathcal{B}_1, \sum_{j=2}^k a_j \mathcal{B}_j\cap \mathcal{B}_1)\to C, which is also a family of K-semistable log Fano pairs. Similarly as in Step 1 (see Step 2 of the proof of Proposition 4.2), we see that \mathcal{B}_{j,0}\cap \mathcal{B}_{1,0}=B_j\cap B_1 are mutually different prime divisors on B_1 for 2\leq j\leq k, and B_2\cap B_1 induces a special test configuration of (B_1, \sum_{j=2}^ka_j B_j\cap B_1) with vanishing generalized Futaki invariant. Using the same argument as in Step 1, we conclude that B_2\cap B_1 is a Fano variety and K_{B_1}+B_2\cap B_1=K_{B_2\cap B_1}. Continuing the process, we clearly complete the proofs of (1) and (2). Applying Lemma 2.2 in the final step, we conclude (3).

Consider the log pair (\mathbb{P}^4, c\cdot \frac{5}{6}S), where S is a cubic threefold and c\in (0,1). We show that the following statements are equivalent:

1. (\mathbb{P}^4, c\cdot \frac{5}{6}S) is K-semistable for 0<c\ll 1;

2. S is {\rm {GIT}}-semistable;

3. S is K-semistable;

4. (\mathbb{P}^4, \frac{5}{6}S) is K-semistable;

5. (\mathbb{P}^4, c\cdot\frac{5}{6}S) is K-semistable for any c\in (0,1).

Note that (1) implies (2) by [ADL19, Theorem 1.4]; (2) implies (3) by [LX19]. To see how (3) implies (4), let (X, \frac{5}{6}B) denote the projective cone over S with respect to the polarization S|_S, where B is the infinite section of the projective cone X. By Lemma 2.2, we see that (X, \frac{5}{6}B) is K-semistable. Note that (X, \frac{5}{6}B) is a degeneration of (\mathbb{P}^4, \frac{5}{6}S) by a special test configuration, so we get (\mathbb{P}^4, \frac{5}{6}S) is K-semistable by [BLX22]. (4) implies (5) by the interpolation property of K-stability and (5) implies (1) clearly.

These equivalences already imply that there is no wall for c\in (0,1). Indeed, suppose c_1\in (0,1) is the first wall, then there exists a K-semistable log Fano pair (\mathbb{P}^4, c_1\cdot \frac{5}{6}S) such that (\mathbb{P}^4, c\cdot \frac{5}{6}S) is not K-semistable for c>c_1, which is a contradiction.