Minimal Model Program on the generic fiber of log Calabi–Yau type fibration Donghyeon Kim, Dae-Won Lee Abstract . We study the minimal model program on the geometric generic fiber of a fibration f : X → S f:X\to S such that for a Zariski dense subset S ′ ⊆ S S^{\prime}\subseteq S , X s X_{s} is an ε \varepsilon -lc log Calabi–Yau type for every s ∈ S ′ s\in S^{\prime} . We prove that for a fibration f : X → S f:X\to S of varieties, if the fibers are of ε \varepsilon -lc log Calabi–Yau type, then the geometric generic fiber X η ¯ X_{\overline{\eta}} is pklt. In particular, for any big divisor D D on X η ¯ X_{\overline{\eta}} , we can run the anticanonical MMP and D D -MMP with scaling of an ample divisor on X η ¯ X_{\overline{\eta}} . 2010 Mathematics Subject Classification 14E05, 14E30 1. Introduction The minimal model program (MMP) provides a framework for studying the birational geometry of higher dimensional varieties. Controlling the singularities of fibers in a family is an important problem in birational geometry, so that one can run the MMP uniformly and understand the variation of Fano and Calabi–Yau type varieties. In this paper, we focus on fibrations whose closed fibers are of (log) Calabi–Yau type over a Zariski dense subset of the base, and we study the extent to which the geometric generic fiber inherits strong MMP properties from the closed fibers. Let us briefly recall the definition of log Calabi–Yau type. Let (X,\Delta) be a projective pair and \varepsilon>0 . We say that (X,\Delta) is of klt (resp. \varepsilon -lc) log Calabi–Yau type if there exists an effective \mathbb{Q} -Weil divisor \Delta^{\prime}\geq 0 on X such that (X,\Delta+\Delta^{\prime})\text{ is klt }(\text{resp. }\varepsilon\text{-lc) and }K_{X}+\Delta+\Delta^{\prime}\sim_{\mathbb{Q}}0. The Calabi–Yau type pairs naturally arise in birational geometry and behave nicely from the viewpoint of the MMP. Indeed, if (X,\Delta) is a klt log Calabi–Yau type pair, then by definition, there exists an effective \mathbb{Q} -divisor \Delta^{\prime} on X such that K_{X}+\Delta+\Delta^{\prime}\sim_{\mathbb{Q}}0 . For any effective divisor D on X , the pair (X,\Delta+\Delta^{\prime}+\varepsilon D) is also klt for sufficiently small \varepsilon>0 . Thus, one can run a D -MMP with scaling of an ample divisor. Moreover, if D is big, then this MMP terminates. Motivated by this observation, we formulate the following question regarding the behavior of log Calabi–Yau type singularities in families. Question 1.1 . Let \varepsilon>0 , f\colon X\to S be a fibration of varieties, and (X,\Delta) a pair. Suppose that there exists a Zariski dense subset S^{\prime}\subseteq S such that for each s\in S^{\prime} , the fiber (X_{s},\Delta_{s}) is of \varepsilon -lc log Calabi–Yau type. Is the geometric generic fiber (X_{\overline{\eta}},\Delta_{\overline{\eta}}) of klt log Calabi–Yau type? A complete answer to Question 1.1 seems to be subtle. Our first main result gives a weaker conclusion: we show that the geometric generic fiber is potentially klt (pklt for short). For the definition of pklt pairs, see Definition 2.2 . We note that if S^{\prime}\subseteq S is a Zariski dense open subset and the base field k is algebraically closed and of infinite transcendence degree over the prime field of any characteristic, then by [ 14 , Lemma 2.1] , Question 1.1 has an affirmative answer. However, our question concerns a more general setting. Theorem 1.2 . Let \varepsilon>0 , f:X\to S be a fibration of normal varieties, (X,\Delta) a pair, and let S^{\prime}\subseteq S be a Zariski dense subset such that (X_{s},\Delta_{s}) is of \varepsilon -lc log Calabi–Yau type for each s\in S^{\prime} . Then the geometric generic fiber (X_{\overline{\eta}},\Delta_{\overline{\eta}}) is pklt. The pklt condition is strong enough to run the anticanonical MMP by [ 2 ] and, more generally, a D -MMP with scaling of an ample divisor on the geometric generic fiber. Since every klt log Calabi–Yau type pair is pklt, the conclusion of our theorem still suffices to highlight many features of the MMP on the geometric generic fiber. As a consequence of Theorem 1.2 and Theorem 2.4 , we obtain the following corollary. Corollary 1.3 . Under the assumptions of Theorem 1.2 , the following holds: (a) there exists a -(K_{X_{\overline{\eta}}}+\Delta_{\overline{\eta}}) -MMP with scaling of an ample divisor, and (b) for any big \mathbb{Q} -Cartier divisor D on X_{\overline{\eta}} , there exists a D -MMP with scaling of an ample divisor that terminates. We note that determining whether a pair (X,\Delta) is pklt involves computing asymptotic invariants for every prime divisor E over X , which is often intractable in practice. Theorem 1.2 provides an effective way to bypass this difficulty. It demonstrates that the log Calabi–Yau structure of fibers over a Zariski dense subset is a sufficient condition for the geometric generic fiber to be pklt. This allows us to deduce the birational properties of the generic fiber, including the existence of the MMP, from the geometry of the fibration. We recall a conjecture due to Jiao, which predicts that for a klt Calabi–Yau type variety X , there exists a positive integer m such that m(K_{X}+\Delta)\sim 0 for some effective divisor \Delta on X under a boundedness assumption. Conjecture 1.4 (cf. [ 9 , Section 3] ) . There exists a positive integer m with the following property. Let \varepsilon,d,v>0 , X be a d -dimensional \varepsilon -lc Calabi–Yau type variety, and A an ample divisor on X such that A^{d}\leq v . Then there exists an effective \mathbb{Q} -Weil divisor \Delta on X such that (X,\Delta) is klt and m(K_{X}+\Delta)\sim 0 . Assuming Conjecture 1.4 , we can show that Question 1.1 has an affirmative answer for \Delta=0 case. Theorem 1.5 . If we assume Conjecture 1.4 , then Question 1.1 is true for \Delta=0 . Theorem 1.2 provides evidence for a positive answer to Question 1.1 in full generality. The rest of this paper is organized as follows. In Section 2 , we recall basic definitions and properties on Nakayama’s asymptotic order and potentially klt pairs, which will be used in the proof of our main theorems. In Section 3 , we prove Theorem 1.2 , deduce Corollary 1.3 , and give a proof of Theorem 1.5 . 2. Preliminaries Throughout this paper, we work over an algebraically closed field k of characteristic 0 , unless otherwise stated. By a variety , we mean an integral, separated scheme of finite type over k . All varieties are assumed to be quasi-projective. We now fix some notation and conventions that will be used throughout the paper. (1) A couple (X,\Delta) consists of a normal variety X and an effective \mathbb{Q} -Weil divisor \Delta on X . If, in addition, K_{X}+\Delta is \mathbb{Q} -Cartier, we call (X,\Delta) a pair . (2) For a scheme S and a point s\in S , we denote k(s) by the residue field of \mathcal{O}_{S,s} and regard s as the morphism s:\operatorname{Spec}k(s)\to S . The associated geometric point is \bar{s}:\operatorname{Spec}\overline{k(s)}\to S . For an integral scheme X , denote by k(X) its function field. (3) If \eta is the generic point of an integral scheme S , we write \bar{\eta}:\operatorname{Spec}\overline{k(\eta)}\to S and call X_{\bar{\eta}}\coloneqq X\times_{S}\bar{\eta} the geometric generic fiber of X\to S . (4) For a morphism f:X\to S of schemes and a (geometric) point s\to S , set X_{s}\coloneqq X\times_{S}s. If \Delta is a divisor on X , we write \Delta_{s} for its pullback to X_{s} whenever it is defined. (5) For a morphism f:X\to S between varieties, we say that f is a fibration if it is proper, surjective, and f_{*}\mathcal{O}_{X}=\mathcal{O}_{S} . (6) Let (X,\Delta) be a pair, f:X^{\prime}\to X a proper birational morphism between normal varieties, and E a prime divisor on X^{\prime} . The log discrepancy of E with respect to (X,\Delta) is A_{X,\Delta}(E)\coloneqq\mathrm{mult}_{E}\bigl(K_{X^{\prime}}-f^{*}(K_{X}+\Delta)\bigr)+1. This definition does not depend on the choice of f . We say that (X,\Delta) is kawamata log terminal (klt) if A_{X,\Delta}(E)>0 for every prime divisor E over X . (7) Let X be a normal variety and D a Cartier divisor on X . The base ideal of the complete linear system |D| is the image of the natural map H^{0}(X,\mathcal{O}_{X}(D))\otimes\mathcal{O}_{X}(-D)\longrightarrow\mathcal{O}_{X}, and is denoted by \mathfrak{b}(|D|) . (8) For a variety S , by shrinking S , we mean replacing S by a variety S^{\prime} equipped with a quasi-finite morphism S^{\prime}\to S (e.g., an open immersion or a generically finite cover), and replacing all data by their pullbacks to S^{\prime} while keeping the same notation. 2.1. Asymptotic order In this subsection, we recall the notion of asymptotic order for divisors. Let X be a normal projective variety and D a big Cartier divisor. For each integer m\geq 1 , denote by \mathfrak{b}_{m}\coloneqq\mathfrak{b}(|mD|) the base ideal of |mD| . Then the collection \mathfrak{b}_{\bullet}\coloneqq\left\{\mathfrak{b}_{m}\right\}_{m\in\mathbb{Z}_{\geq 0}} forms a graded sequence of ideals in \mathcal{O}_{X} , where we set \mathfrak{b}_{0}\coloneqq\mathcal{O}_{X} . For a prime divisor E over X , we define the asymptotic order of D with respect to E by \mathrm{ord}_{E}(\|D\|)\coloneqq\inf_{\begin{subarray}{c}m\in\mathbb{Z}_{\geq 1}\\ \mathfrak{b}_{m}\neq(0)\end{subarray}}\frac{\mathrm{ord}_{E}(\mathfrak{b}_{m})}{m}. If D is a big \mathbb{Q} -divisor on X , then we define \ord_{E}(\|D\|)\coloneqq\frac{1}{m}\ord_{E}(\|mD\|), where n is a positive integer such that mD is Cartier. More generally, if D is a pseudoeffective \mathbb{Q} -divisor on X , then for an ample Cartier divisor A on X , we define \sigma_{E}(D)\coloneqq\lim_{\begin{subarray}{c}\varepsilon\to 0^{+}\\ \varepsilon\text{ is rational }\end{subarray}}\mathrm{ord}_{E}(\|D+\varepsilon A\|). It is known that \sigma_{E}(D)<\infty and the definition does not depend on the choice of A (cf. [ 13 , Chapter 3.1] ). Moreover, if D is big, then \sigma_{E}(D)=\ord_{E}(\|D\|) . We also note that for any pseudoeffective \mathbb{Q} -divisors D and D^{\prime} on X , and any prime divisor E over X , we have the following inequality \sigma_{E}(D+D^{\prime})\leq\sigma_{E}(D)+\sigma_{E}(D^{\prime}). Let (X,\Delta) be a projective klt pair, and D a pseudoeffective \mathbb{Q} -divisor on X . We define the log canonical threshold as \mathrm{lct}_{\sigma}(X,\Delta,D)\coloneqq\inf_{E}\frac{A_{X,\Delta}(E)}{\sigma_{E}(D)}, where the \inf is taken over all prime divisors E over X . Let us recall the following lemma, which is essentially [ 7 , Proposition 2.1] and appears in the following form in [ 11 , Lemma 2.11] . Lemma 2.1 (cf. [ 7 , Proposition 2.1] and [ 11 , Lemma 2.11] ) . Let X be a normal projective variety, D a big \mathbb{Q} -Cartier \mathbb{Q} -divisor, and E a prime divisor on X . Suppose that a is a positive rational number. Then \sigma_{E}(D)0 such that A_{X,\Delta}(E)-\sigma_{E}(-(K_{X}+\Delta))>\varepsilon for every prime divisor E over X . The following is a valuative characterization of pklt pairs, which was recently proved in [ 2 ] . Theorem 2.3 (cf. [ 2 , Proposition 4.2] ) . Let (X,\Delta) be a pklt pair. Then \mathrm{lct}_{\sigma}(X,\Delta,-(K_{X}+\Delta))>1. The pklt condition is strong enough to guarantee the existence of MMPs. Theorem 2.4 (cf. [ 2 , Corollary 1.3] ) . Let (X,\Delta) be a pklt pair. Then (a) there exists a -(K_{X}+\Delta) -MMP with scaling of an ample divisor, and (b) for every big \mathbb{Q} -Cartier divisor D on X , there exists a D -MMP with scaling of an ample divisor that terminates. Proof. Part (a) is precisely [ 2 , Corollary 1.3] . For (b), we claim that there is a positive number \varepsilon>0 such that \mathrm{lct}_{\sigma}(X,\Delta,-(K_{X}+\Delta)+\varepsilon D)>1. Indeed, we have \displaystyle\frac{1}{\mathrm{lct}_{\sigma}(X,\Delta,-(K_{X}+\Delta)+\varepsilon D)} \displaystyle=\sup_{E}\frac{\sigma_{E}(-(K_{X}+\Delta)+\varepsilon D)}{A_{X,\Delta}(E)} \displaystyle\leq\sup_{E}\frac{\sigma_{E}(-(K_{X}+\Delta))+\varepsilon\sigma_{E}(D)}{A_{X,\Delta}(E)} \displaystyle\leq\sup_{E}\frac{\sigma_{E}(-(K_{X}+\Delta))}{A_{X,\Delta}(E)}+\varepsilon\sup_{E}\frac{\sigma_{E}(D)}{A_{X.\Delta}(E)} \displaystyle=\frac{1}{\mathrm{lct}_{\sigma}(X,\Delta,-(K_{X}+\Delta))}+\frac{\varepsilon}{\mathrm{lct}_{\sigma}(X,\Delta,D)}. By Theorem 2.3 , we have \mathrm{lct}_{\sigma}(X,\Delta,-(K_{X}+\Delta))>1 . By choosing \varepsilon>0 sufficiently small, we obtain the claim. Hence, by [ 15 , Lemma 1.60] , there exists an effective \mathbb{Q} -Weil divisor \Delta^{\prime}\sim_{\mathbb{Q}}-(K_{X}+\Delta)+\varepsilon D on X such that (X,\Delta+\Delta^{\prime}) is klt. Consequently, K_{X}+\Delta+\Delta^{\prime}\sim_{\mathbb{Q}}\varepsilon D and by [ 1 , Corollary 1.4.2] , we may run a (K_{X}+\Delta+\Delta^{\prime}) -MMP, i.e., a D -MMP with scaling that terminates. ∎ We also recall the following result, which shows that pklt pairs with a big anticanonical class are of Fano type. Theorem 2.5 (cf. [ 3 , Corollary 3.10] ) . Let (X,\Delta) be a pklt pair with -(K_{X}+\Delta) big. Then X is a variety of Fano type. Remark 2.6 . We expect that for a projective klt pair (X,\Delta) , the condition that (X,\Delta) is pklt is equivalent to the existence of the anticanonical minimal model of (X,\Delta) . The termination of an anticanonical MMP for pklt pairs would imply this equivalence. However, little is known about the termination of such MMPs. Theorem 2.5 was first stated in [ 5 , Theorem 5.1] and its proof relied on [ 5 , Proposition 4.4] . However, the proofs in both the arXiv and published versions of [ 5 , Proposition 4.4] contain a gap. Only recently, a complete proof has been provided in [ 3 , Corollary 3.10] by using valuation theory. For more details on this issue, we refer the reader to [ 3 , Remark 3.11] . 2.3. Models In this subsection, we introduce the notion of a model over a base, in a form analogous to the notion of a model in reduction mod p . Definition 2.7 . Let S be a variety and \eta\in S the generic point. Let X be a variety over \overline{k(\eta)} . (a) If X_{S} is a variety flat over S such that X_{S}\times_{S}\overline{k(\eta)}=X , then we call X_{S} a model of X over S . (b) Let X_{S} be a model of X over S and D\subseteq X a Weil divisor. If D_{S}\subseteq X_{S} is an effective Weil divisor on X_{S} flat over S with (D_{S}\subseteq X_{S})\times_{S}\overline{k(\eta)}=D\subseteq X , then we call D_{S} a model of D over S . (c) Let f:X^{\prime}\to X be a morphism between two varieties over \overline{k(\eta)} . If there exists an S -morphism f_{S}:X^{\prime}_{S}\to X_{S} such that f_{S}\times_{S}\overline{k(\eta)}=f , X^{\prime}_{S} , and X_{S} are flat over S , then we say that f_{S} is a model of f over S . (d) Let X_{S} be a model of X over S and D=\sum q_{i}D_{i} a \mathbb{Q} -Weil divisor on X with q_{i}\in\mathbb{Q} and prime D_{i} . If D_{iS} is a model of D_{i} over S for each i , then we say that D_{S}\coloneqq\sum q_{i}D_{iS} is a model of D over S . Note that if D is \mathbb{Q} -Cartier, then we may choose D_{S} as a \mathbb{Q} -Cartier divisor on X_{S} . (e) Let (X,\Delta) be a pair, X_{S} a model of X over S , and \Delta_{S} is so. Then we call (X_{S},\Delta_{S}) a model of (X,\Delta) over S . Remark 2.8 . Let S be a variety, and \eta\in S the generic point. Then, for every quasi-projective variety X over \overline{k(\eta)} , after shrinking S , there is a model X_{S} of X over S . Indeed, the variety X is determined by finitely many equations in \overline{k(\eta)}[x_{1},\cdots,x_{d+1}] for some d . Hence, there exists a finite extension k^{\prime}/k(\eta) and a scheme X_{k^{\prime}} over k^{\prime} such that X_{k^{\prime}}\times_{k^{\prime}}\overline{k(\eta)}=X . Let S^{\prime}\to S be a finite morphism, \eta_{S^{\prime}} the generic point of S^{\prime} such that k(\eta_{S^{\prime}})=k^{\prime} , and X_{S^{\prime}} be the closure of X_{k^{\prime}} in \mathbb{P}^{d}_{S^{\prime}} . By generic flatness (cf. [ 6 , Proposition 052A] ), after shrinking S^{\prime} , we may assume that X_{S^{\prime}} is a flat S^{\prime} -scheme. Replacing S by S^{\prime} gives a model X_{S} of S . A similar argument applies to constructing models of morphisms over \overline{k(\eta)} and a \mathbb{Q} -Weil divisor on X . Notation 2.9 . Fix a variety S with a (geometric) point s . Let X_{S} and X^{\prime}_{S} be flat S -schemes and f_{S}:X^{\prime}_{S}\to X_{S} be an S -morphism. We denote by X_{s}\coloneqq X_{S}|_{s},\penalty\ X^{\prime}_{s}\coloneqq X^{\prime}_{R}|_{s}\text{ and }f_{s}\coloneqq f_{S}|_{s}. If D=\sum r_{i}D_{i} is a Weil \mathbb{R} -divisor on X_{S} such that each D_{i} is a prime divisor and flat over S , then we denote D_{s}\coloneqq\sum r_{i}D_{i}|_{s} as a Weil \mathbb{R} -divisor on X_{s} . Note that if D is \mathbb{Q} -Cartier, then D_{s} is also \mathbb{Q} -Cartier. We next recall a property of log discrepancy which seems well known to experts. Proposition 2.10 (cf. [ 11 , Proposition 2.5] ) . Let (X,\Delta) be a pair, f:X\to S a fibration between normal varieties, X^{\prime}\to X a proper birational morphism, and E a prime divisor on X^{\prime} which is flat over S . Suppose \eta\in S is the generic point. Then after shrinking S , we obtain A_{X_{\overline{\eta}},\Delta_{\overline{\eta}}}(E_{\overline{\eta}})=A_{X_{s},\Delta_{s}}(E_{s}) for a closed point s\in S . 2.4. Volume In this subsection, we recall the definition of volume of a divisor and state results on the behavior of volumes in families and on Calabi–Yau type varieties due to Jiao [ 10 , 9 ] . Definition 2.11 . Let X be a normal projective variety and D a \mathbb{Q} -Cartier divisor on X . The volume of D is defined as \mathrm{vol}(D)\coloneqq\limsup_{m\to\infty}\frac{h^{0}(X,\mathcal{O}_{X}(\left\lfloor mD\right\rfloor))}{\frac{m^{\dim X}}{(\dim X)!}}. It is well known that the volume of a divisor has the upper-semicontinuous property. Theorem 2.12 (cf. [ 10 , Theorem 1.1] ) . Let X\to S be a fibration of normal varieties such that X_{\overline{\eta}} is a normal projective variety over \overline{k(\eta)} , where \eta\in S is the generic point, and let D be a Cartier divisor on X . Then for every Zariski dense subset S^{\prime}\subseteq S , we have \mathrm{vol}(D_{\overline{\eta}})=\inf_{s\in S^{\prime}}\mathrm{vol}(D_{s}). Moreover, Jiao proved a discreteness property of volumes on \varepsilon -lc log Calabi–Yau varieties. Theorem 2.13 (cf. [ 9 , Theorem 1.1] ) . Let \varepsilon>0 be a positive number and d a positive integer. Then there exists a discrete set \mathcal{C}\subseteq\mathbb{R}_{\geq 0} , depending only on \varepsilon and d with the following property: If (X,\Delta) is a d -dimensional \varepsilon -lc log Calabi–Yau pair, then for every Cartier divisor D on X , we have \mathrm{vol}(D)\in\mathcal{C} . 3. Proofs of main results Proof of Theorem 1.2 . Let f:X^{\prime}\to X_{\overline{\eta}} be a proper birational morphism with E a prime divisor on X^{\prime} . Then after shrinking S , there exist models f_{S}:X^{\prime}_{S}\to X of f and E_{S} of E over S . Let \varepsilon^{\prime}>0 be a positive number, and let A be an ample /S Cartier divisor on X . Choose an ample /S Cartier divisor A^{\prime} on X such that -(K_{X}+\Delta)+\varepsilon^{\prime}A\leq A^{\prime} . Then for every s\in S , we have \mathrm{vol}(-(K_{X_{s}}+\Delta_{s})+\varepsilon^{\prime}A_{s})\leq\mathrm{vol}(A^{\prime}_{s})=\mathrm{vol}(A^{\prime}_{\overline{\eta}}). Therefore, the set \left\{\mathrm{vol}(-(K_{X_{s}}+\Delta_{s})+\varepsilon^{\prime}A_{s})\right\}_{s\in S}\subseteq\mathbb{R}_{\geq 0} is bounded. Hence, by Theorem 2.13 , there exists S^{\prime}_{\varepsilon^{\prime}}\subseteq S^{\prime} that is Zariski dense in S such that \mathrm{vol}(-(K_{X_{s}}+\Delta_{s})+\varepsilon^{\prime}A_{s}) is constant for every s\in S^{\prime}_{\varepsilon^{\prime}} . Moreover, by Theorem 2.12 , we have (3.1) \mathrm{vol}(-(K_{X_{s}}+\Delta_{s})+\varepsilon^{\prime}A_{s})=\mathrm{vol}(-(K_{X_{\overline{\eta}}}+\Delta_{\overline{\eta}})+\varepsilon^{\prime}A)\text{ for every }s\in S^{\prime}_{\varepsilon^{\prime}}. Let D\coloneqq-(K_{X}+\Delta)+\varepsilon^{\prime}A_{S} for simplicity. Let a>\sigma_{E_{s}}(D_{s}) be a rational number. By Lemma 2.1 , we have \mathrm{vol}(f^{*}_{s}D_{s}-aE_{s})<\mathrm{vol}(f^{*}_{s}D_{s}). For any s\in S^{\prime}_{\varepsilon} , we then have \displaystyle\mathrm{vol}(f^{*}D_{\overline{\eta}}-aE) \displaystyle\leq\mathrm{vol}(f^{*}_{s}D_{s}-aE_{s}) \displaystyle(1) \displaystyle<\mathrm{vol}(f^{*}_{s}D_{s}) \displaystyle=\mathrm{vol}(D_{s}) \displaystyle=\mathrm{vol}(D_{\overline{\eta}}) \displaystyle(2) \displaystyle=\mathrm{vol}(f^{*}D_{\overline{\eta}}), where we used the upper semicontinuity of cohomology (cf. [ 8 , Theorem 12.8] ) in (1), and ( 3.1 ) in (2). Therefore, by Lemma 2.1 again, we obtain a>\sigma_{E}(D) . Thus, (3.2) \sigma_{E_{s}}(D_{s})\geq\sigma_{E}(D_{\overline{\eta}})\text{ for every }s\in S^{\prime}_{\varepsilon}. By Proposition 2.10 , after shrinking S , we may assume that (3.3) A_{X_{\overline{\eta}},\Delta_{\overline{\eta}}}(E)=A_{X_{s},\Delta_{s}}(E_{s}) for every s\in S . By the assumption that the fiber (X_{s},\Delta_{s}) is of \varepsilon -lc log Calabi–Yau type, there exists an effective \mathbb{Q} -Weil divisor \Delta^{\prime} on X_{s} such that (X_{s},\Delta_{s}+\Delta^{\prime}) is \varepsilon -lc and K_{X_{s}}+\Delta_{s}+\Delta^{\prime}\sim_{\mathbb{Q}}0 . \displaystyle A_{X_{\overline{\eta}},\Delta_{\overline{\eta}}}(E)-\sigma_{E}(D_{\overline{\eta}}) \displaystyle\geq A_{X_{s},\Delta_{s}}(E_{s})-\sigma_{E_{s}}(D_{s}) \displaystyle(1) \displaystyle=A_{X_{s},\Delta_{s}}(E_{s})-\mathrm{ord}_{E_{s}}(\|-(K_{X_{s}}+\Delta_{s})\|) \displaystyle\geq A_{X_{s},\Delta_{s}}(E_{s})-\mathrm{ord}_{E_{s}}\Delta^{\prime} \displaystyle=A_{X_{s},\Delta_{s}+\Delta^{\prime}}(E_{s})>\varepsilon for every s\in S^{\prime}_{\varepsilon} , where (1) is due to ( 3.2 ) and ( 3.3 ). Hence, A_{X_{\overline{\eta}},\Delta_{\overline{\eta}}}(E)-\sigma_{E}(-(K_{X_{\overline{\eta}}}+\Delta_{\overline{\eta}}))\geq\varepsilon. Hence, (X_{\overline{\eta}},\Delta_{\overline{\eta}}) is pklt, and we complete the proof. ∎ Remark 3.1 . We emphasize that the \varepsilon -lc condition on the fibers is crucial. If we replace this condition with klt or lc, then we do not know whether the volumes of Cartier divisors on a log Calabi–Yau type variety are contained in a discrete set or not. This discreteness is a key ingredient in our proof to ensure that the volume is constant over a Zariski dense subset. Proof of Corollary 1.3 . This follows directly from Theorems 1.2 and 2.4 . ∎ Proof of Theorem 1.5 . The argument follows the strategy of [ 4 , Step 4 in the proof of Theorem 1.2] . By Conjecture 1.4 , there exists a positive integer m that is independent of s\in S^{\prime} such that m(K_{X}+\Delta)\sim 0 . By [ 8 , Chapter III, Theorem 12.8 and Corollary 12.9] , after replacing S by a finite base change, we obtain a surjection f_{*}\mathcal{O}_{X}(-mK_{X})\to H^{0}(X_{s},\mathcal{O}_{X_{s}}(-mK_{X_{s}})). Thus, for every \Delta\in|-mK_{X_{s}}| , there exists a divisor \Delta^{\prime}\in|-mK_{X}| such that K_{X}+\frac{1}{m}\Delta^{\prime}\sim_{\mathbb{Q},S}0. Moreover, the generic fiber \left(X_{\overline{\eta}},\frac{1}{m}\Delta^{\prime}_{\overline{\eta}}\right) is klt by [ 12 , Theorem 9.5.19] . Since X_{s} is klt log Calabi–Yau type, we conclude the assertion. ∎