BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS

arXiv:2405.18516v1 [math.AG] 28 May 2024

JOAQUÍN MORAGA

Abstract. We study the birational complexity of log Calabi–Yau 3-folds. For such a pair pX, Bq of index
one and coregularity zero, we show that cbir pX, Bq P t0, 2, 3u. Further, we prove that pX, Bq has a log
Calabi–Yau crepant birational model that admits a crepant contraction to pP3´c , H0 ` ¨ ¨ ¨ ` H3´c q, where
c “ cbir pX, Bq. To prove this, we give a geometric characterization of standard P1 -links.

Contents
1. Introduction 1
2. Preliminaries 3
3. Geometry of standard P1 -links 8
4. Birational complexity of log Calabi–Yau 3-folds 13
5. Crepant fibrations to projective spaces 15
6. Examples and questions 16
References 17

1. Introduction
For a general type variety X its canonical model is the best possible birational model from almost any per-
spective. On the contrast, log Calabi–Yau pairs pX, Bq may have countably many birational non-isomorphic
models that satisfy different geometric properties. Thus, it is natural to introduce and study invariants that
allows us to detect better birational models of a given log Calabi–Yau pair. Toric log Calabi–Yau pairs, due
to their combinatorial nature, are the log Calabi–Yau pairs that we can understand the most. The complex-
ity cpX, Bq, of a log Calabi–Yau pair pX, Bq, is defined to be the value dim X ` dimQ ClQ pXq ´ |B|, where
|B| is the sum of the coefficients of B. This invariant is non-negative and pX, tBuq is toric whenever it is zero
(see [4]). In [20], Mauri and the author introduced the birational complexity cbir pX, Bq of a log Calabi–Yau
pair and proved several fundamental properties of this invariant. The birational complexity is the minimum
among the complexities of birational models of a given log Calabi–Yau pair. This invariant detects whether
a log Calabi–Yau pair has a toric model i.e., it is birationally equivalent to a toric log Calabi–Yau pair. Thus,
the birational complexity gives a natural invariant to find optimal models for log Calabi–Yau pairs.
In this article, we study the possible values of the birational complexity of log Calabi–Yau 3-folds pX, Bq
of index one and coregularity zero. 1 For each value of the birational complexity, we find a birational model
that computes the birational complexity and enjoys some extra geometric features.

2020 Mathematics Subject Classification. Primary 14E05, 14J32; Secondary 14E30.

1Log Calabi–Yau pairs of coregularity zero are called pairs of maximal intersection in the literature.

1
2 J. MORAGA

Throughout the article, we denote by Σn the reduced sum of the coordinate hyperplanes of Pn . Gross,
Hacking, and Keel proved that every log Calabi–Yau surface of index one and coregularity zero admits a
toric model, or equivalently, has birational complexity zero (see, e.g., [12, Proposition 1.3]).
Theorem 1.1. Let pX, Bq be a log Calabi–Yau surface of index one and coregularity zero. Then, we have
cbir pX, Bq “ 0. In particular, there is a crepant birational isomorphism pX, Bq »cbir pP2 , Σ2 q.
Our first theorem is a classification of the possible values of the birational complexity of log Calabi–Yau
3-folds of index one and coregularity zero. More precisely, we prove the following.
Theorem 1.2. Let pX, Bq be a log Calabi–Yau 3-fold of index one and coregularity zero. Then, we have
cbir pX, Bq P t0, 2, 3u.
In Example 6.1, Example 6.2, and Example 6.3, we show that all the values in Theorem 1.2 indeed
occur. Further, all such values already appear for pairs pX, Bq where X is a Fano hypersurface in P4 and
B is a suitable anti-canonical divisor. The next theorem states that a log Calabi–Yau 3-fold of index one,
coregularity zero, and birational complexity c has a birational model that admits a contraction to P3´c .
Theorem 1.3. Let pX, Bq be a log Calabi–Yau 3-fold of index one and coregularity zero. Let c :“ cbir pX, Bq
be the birational complexity of the pair. Then, there is a log Calabi–Yau crepant birational model pX 1 , B 1 q of
pX, Bq that admits a crepant contraction to pP3´c , Σ3´c q.
We refer the reader to Definition 2.4 for the concept of crepant birational model and to Definition 2.7 for
the concept of crepant contraction. The previous theorem can be regarded as the fact that the toric part
of such a log Calabi–Yau 3-fold can be factored in the base of the fibration from an appropiate birational
model. To prove the two previous theorems, we will need to study the birational geometry of certain special
fibrations; the so-called standard P1 -links (see Definition 2.19). These fibrations have appeared in the study
of dual complexes and log canonical centers [15, 9]. Our main theorem in this direction states that a standard
P1 -link is simply the projectivization of a split Q-vector bundle of rank 2.
Theorem 1.4. Let φ : pX, Bq Ñ Z be a standard P1 -link. Then, the variety X is the projectivization over Z
of a rank 2 split Q-vector bundle. Moreover, the divisor tBu consists of the reduced sum of the two sections
corresponding to the projections to the summands of the split Q-vector bundle.
Using the previous characterization of standard P1 -links, we will prove the following theorem about
standard P1 -links of dimension at most three.
Theorem 1.5. Let X be a normal variety of dimension at most 3. Let φ : pX, Bq Ñ Z be a standard P1 -
link. Let pZ, BZ q be the pair induced on Z by the canonical bundle formula. Then, we have a commutative
diagram:
ψ
pX, Bq o❴ ❴ ❴ pZ 1 , BZ 1 q ˆ pP1 , Σ1 q
φ π1
 
pZ, BZ q o π
pZ 1 , BZ 1 q
where π is a projective crepant birational morphism with BZ 1 effective, ψ is a crepant birational map, and
π1 is the projection to the first component.
Projectivizations of split Q-vector bundles over the same base are always birational equivalent to each
other. Thus, we emphasize that the interesting part of Theorem 1.5 is that we can realize such birational
 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 3

equivalence by keeping the pair log Calabi–Yau, i.e., not introducing boundary divisor with negative coeffi-
cients.

Acknowledgements. The author would like to thank Joshua Enwright, Fernando Figueroa, Konstantin
Loginov, and Stefano Filipazzi for very useful comments.

2. Preliminaries
In this section, we recall some preliminary results and prove some lemmas. For the concepts of singularities
of the MMP, we refer the reader to [14]. For the concepts related to the canonical bundle formula, we refer
the reader to [3, 10, 7]. We work over an algebraically closed field K of characteristic zero. Given a pair
pX, Bq and a closed irreducible subvariety Z, we write pX, B; Zq for the germ of pX, Bq at the generic point
of Z.

2.1. Crepant maps. In this subsection, we recall preliminaries regarding log Calabi–Yau pairs, crepant
birational maps, and crepant birational models.
Definition 2.1. A contraction f : X Ñ Y is a morphism for which f˚ OX “ OY . In particular, if X is
normal, then Y is normal. A fibration is a contraction with positive dimensional general fiber.
Definition 2.2. Let f : X Ñ Y be a finite morphism. Let pX, Bq and pY, BY q be two sub-pairs. We say
that f : pX, Bq Ñ pY, BY q is crepant if f ˚ pKY ` BY q “ KX ` B. Let us emphasize that we ask the equality
to hold and not just a Q-linear equivalence.
Definition 2.3. Let f : X Ñ Z be a projective contraction. Let pX, Bq be a log pair on X. We say that
pX, Bq is log Calabi–Yau over Z if pX, Bq is log canonical and KX ` B „Q,Z 0. The index of pX, Bq over Z is
the smallest positive integer i for which ipKX ` Bq „Z 0. We drop Z from the notation when Z “ SpecpKq.
Definition 2.4. Let X Ñ Z be a projective contraction and φ : X 99K X 1 be a birational map over Z.
Let pX, Bq and pX 1 , B 1 q be two sub-pairs. We say that φ : pX, Bq 99K pX 1 , B 1 q is crepant if the following
conditions are satisfied:
(1) there exists a common resolution Y with projective morphisms p : Y Ñ X and q : Y Ñ X 1 , and
(2) we have p˚ pKX ` Bq “ q ˚ pKX 1 ` B 1 q.
In the previous case, we say that φ is a crepant birational map for the sub-pairs pX, Bq and pX 1 , B 1 q.
Furthermore, we say that pX, Bq and pX 1 , B 1 q are crepant birational equivalent and write
pX, Bq »cbir pX 1 , B 1 q.
We say that pX 1 , B 1 q is a crepant birational model of pX, Bq over Z.
Given a birational map X 99K X 1 over Z and a sub-pair pX 1 , B 1 q, there is a unique sub-pair pX, Bq for
which pX, Bq »cbir pX 1 , B 1 q. This sub-pair pX 1 , B 1 q is called the log pull-back of pX, Bq to X 1 .
Remark 2.5. If pX, Bq is log Calabi–Yau over Z and pX 1 , B 1 q is crepant birational equivalent to pX, Bq over
Z, then pX 1 , B 1 q is log Calabi–Yau over Z provided that B 1 is effective. Furthermore, the index of pX, Bq
and pX 1 , B 1 q over Z agree (see, e.g., [8, Lemma 3.2]).
Definition 2.6. Let pX, Bq be a log Calabi–Yau pair over Z. A crepant birational model pX 1 , B 1 q over Z is
said to be a log Calabi–Yau crepant birational model if B 1 ě 0, or analogously, if pX 1 , B 1 q is log Calabi–Yau
over Z.
4 J. MORAGA

Definition 2.7. Let φ : X Ñ Z be a projective contraction. Let pX, Bq be a log Calabi–Yau pair over Z. Let
pZ, BZ q be the log canonical pair induced by the canonical bundle formula. We say that φ : pX, Bq Ñ pZ, BZ q
is a crepant contraction. Analogously, we say that φ is a crepant contraction for the log pairs pX, Bq and
pZ, BZ q.
Definition 2.8. Let pX, Bq be a log pair. The coregularity of pX, Bq, denoted by coregpX, Bq, is the
dimension of a minimal 2 log canonical center on a dlt modification of pX, Bq.
2.2. Complexity. In this subsection, we recall the concepts of complexity and birational complexity.
Definition 2.9. Let pX, Bq be a log Calabi–Yau pair. The complexity of pX, Bq is
cpX, Bq :“ dim X ` dimQ ClQ pXq ´ |B|,
where |B| is the sum of the coefficients of B.
Definition 2.10. Let pX, Bq be a log Calabi–Yau pair. The birational complexity of pX, Bq is
cbir pX, Bq :“ mintcpX 1 , B 1 q | pX 1 , B 1 q »cbir pX, Bq and B 1 ě 0u.
In other words, the birational complexity of pX, Bq is the smallest complexity among log Calabi–Yau crepant
birational models of pX, Bq.
By [4, Theorem 1.2], we know that cbir pX, Bq ě 0 for every log Calabi–Yau pair.
2.3. Degenerate divisors. We recall the definition of degenerate divisors for proper surjective morphisms
and prove a statement about contraction of degenerate divisors.
Definition 2.11. Let f : X Ñ Y be a proper surjective morphism and D be an effective Q-divisor on X.
(i) We say that D is exceptional over Y if the codimension of f pDq is at least two in Y .
(ii) We say that D is of insufficient fiber type over a prime divisor P of Y if there is a prime divisor on
X that dominates P and is not contained in the support of D.
(iii) We say that D is degenerate over Y if D is of insufficient fiber type over every prime divisor of Y .
Lemma 2.12. Let pX, Bq be a Q-factorial log canonical pair. Let X Ñ Z be a contraction of relative
dimension at most three. Assume that KX ` B „Q,Z E, where E is an effective divisor that is degenerate
over Z. Then, the pKX ` Bq-MMP over Z terminates after contracting all the components of E.
Proof. By [19, Lemma 2.9], we know that supppEq Ď Bs´ pKX ` B{Zq. Then, we can run a pKX ` Bq-MMP
over Z that contracts all the components of E. Since the relative dimension of X Ñ Z is at most three,
we know that this MMP terminates with a good minimal model over Z. Let X 99K X 1 be a birational
contraction induced by the MMP on which all the components of E are contracted. If B 1 is the push-forward
of B to X 1 , then pX 1 , B 1 q is log Calabi–Yau over Z. Thus, the MMP terminates on this model. 
Lemma 2.13. Let φ : X Ñ Z be a Mori fiber space of relative dimension one. Let pX, Bq be a log Calabi–
Yau pair over Z. Assume there is a normal prime component S Ď tBu for which φ|S : S Ñ Z is a finite
morphism of degree two. Let pZ, BZ q be the log Calabi–Yau pair induced by the canonical bundle formula.
Let Z 1 Ñ Z be a birational morphism that only extracts divisors with log discrepancy at most 21 with respect
to pZ, BZ q. Let pZ 1 , BZ 1 q be log pull-back of pZ, BZ q to Z 1 . Then, there exists a crepant birational model
pX 1 , B 1 q of pX, Bq, with B 1 effective, satisfying the following. The variety X 1 admits a Mori fiber space of
relative dimension one to Z 2 , where Z 2 Ñ Z 1 is a dlt modification of pZ 1 , BZ 1 q.
2minimal with respect to the inclusion.
 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 5

Proof. Let Y be the normalization of the fiber product S ˆZ X. Let pY, BY q be the log pull-back of pX, Bq to
pY, BY q. Then, the boundary divisor tBY u contains a prime component that dominates Y with a projective
birational morphism. Let S 1 be the normalization of the fiber product Z 1 ˆZ S and pS 1 , BS 1 q be the log
pull-back of pS, BS q to S 1 . Note that pS 1 , BS 1 q Ñ pZ 1 , BZ 1 q is a crepant finite surjective morphism of degree
2. Thus, by Riemann-Hurwitz we have that BS 1 ě 0. In particular, the morphism S 1 Ñ S is Z2 -equivariant
and only extract log canonical places and canonical places of pS, BS q. By adjunction, each such place
induces either a log canonical place or a canonical place of pY, BY q. Henceforth, we can find a Z2 -equivariant
projective birational morphism Y 1 Ñ Y that extracts all such places. Thus, we get a Z2 -equivariant fibration
Y 1 Ñ S 1 . Let pY 1 , BY 1 q be the log pull-back of pY, BY q to Y 1 . Without loss of generality, we may assume that
pY 1 , BY 1 q is log smooth. The divisor BY 1 may have negative coefficients, however, over each prime divisor of
S 1 there is a prime divisor of BY 1 with non-negative coefficient. Let pX0 , B0 q be the quotient of pY 1 , BY 1 q by
Z2 . The variety X0 admits a fibration to Z 1 . Some coefficients of B0 may be negative, however, over each
prime divisor of Z 1 there is a prime divisor of B0 with non-negative coefficient.
We construct a divisor B01 on X0 as follows. We increase all the negative coefficients of B0 to zero and the
coefficients of the prime degenerate divisors of B0 over Z 1 to one. By further blowing up, we may assume
that pX0 , B01 q is log smooth. We run a pKX0 ` B01 q-MMP over Z 1 . This MMP terminates after contracting
all the components of B0 with negative coefficients and all the exceptional divisors of B0 over Z 1 that are
not log canonical places. The previous statement follows by Lemma 2.12. Let X1 Ñ Z 1 be the model where
this MMP terminates and B1 be the push-forward of B0 to X1 . Then, the pair pX1 , B1 q is log Calabi–Yau
over Z 1 and is crepant birational equivalent to pX, Bq. We run a KX1 -MMP over Z 1 that terminates with a
Mori fiber space of relative dimension one X 1 Ñ Z0 over Z 1 . Any divisor extracted by Z0 Ñ Z 1 is the image
of a log canonical center of pX1 , B1 q. Thus, we conclude that Z0 Ñ Z 1 only extract log canonical places of
pZ 1 , BZ 1 q. Let pZ0 , BZ0 q be the log pull-back of pZ 1 , BZ 1 q to Z0 . Then, the statement follows by applying [20,
Lemma 2.10] to a dlt modification Z 2 of pZ0 , BZ0 q. 
2.4. The two-ray game. In this subsection, we prove two lemmas that are special versions of the two-ray
game.
Lemma 2.14. Let X be a normal variety of dimension at most three. Let φ : X Ñ Z be a Mori fiber space
to a Q-factorial variety. Let pX, Bq be a log Calabi–Yau pair over Z. Let pZ, BZ q be the pair induced on Z
by the canonical bundle formula. Assume that the support of φ˚ tBZ u is contained in the support of tBu. Let
Z Ñ Z 1 be a divisorial contraction that is defined by an extremal ray R that intersects some prime component
of tBZ u positively. Then, there exists a crepant birational model pX 1 , B 1 q of pX, Bq, with B 1 effective, that
admits a Mori fiber space to Z 1 .
Proof. Note that ρpX{Z 1 q “ 2. Let E be the divisor being contracted by Z Ñ Z 1 . Let S be a prime
component of tBZ u that intersects R positively. Then, for ǫ ą 0, we have that pX, B ´ ǫφ˚ Sq is a log
canonical pair. Furthermore, the divisor KX ` B ´ ǫφ˚ S contains φ˚ E on its relative diminished base locus
over Z 1 . Indeed, we have KX ` B ´ ǫφ˚ S „Q φ˚ pKZ ` BZ ´ δSq for a suitable δ ą 0 and φ˚ E is covered by
curves whose images on Z intersect the divisor KZ ` BZ ´ δS negatively and map to points on Z 1 . Then, we
may run a pKX ` B ´ ǫφ˚ Sq-MMP over Z 1 that will terminate with a good minimal model X 1 Ñ Z 1 after
contracting the divisor φ˚ E. This good minimal model has relative Picard rank over Z 1 and hence it is a
Mori fiber space for KX 1 . 
Lemma 2.15. Let X be a normal variety of dimension at most three. Let φ : X Ñ Z be a Mori fiber space
to a Q-factorial variety. Let pX, Bq be a log Calabi–Yau pair over Z. Let pZ, BZ q be the pair induced on Z
by the canonical bundle formula. Let Z Ñ Z 1 be a divisorial contraction that contracts the curve C. Assume
6 J. MORAGA

that C contains no log canonical centers of pZ, BZ q. Then, there is a crepant birational model pX 1 , B 1 q of
pX, Bq, with B 1 effective, that admits a Mori fiber space to Z 1 .
Proof. Note that ρpX{Z 1 q “ 2. Observe that pZ, BZ `ǫCq is log canonical for ǫ ą 0 small enough. Hence, the
pair pX, B`δφ˚ Cq is also log canonical for δ ą 0 small enough. The diminished base locus of KX `B`ǫδφ˚ C
over Z 1 contains φ˚ C. Indeed, we have KX ` B ` ǫδφ˚ C „Q δφ˚ C and the latter is covered by curves that
intersect φ˚ C negatively and map to points on Z 1 . Hence, we may run a pKX ` B ` δφ˚ Cq-MMP over Z 1
that terminates with a good minimal model X 1 Ñ Z 1 after contracting φ˚ C. As ρpX 1 {Z 1 q “ 1, this good
minimal model is a Mori fiber space for KX 1 . 
2.5. Torus actions. In this subsection, we prove lemmas regarding torus actions and birational contractions.
Definition 2.16. Let pX, Bq be a log pair. We write AutpX, Bq for the group of automorphisms of pX, Bq,
i.e., the group that consists of automorphisms φ : X Ñ X for which φ˚ B “ B. Let X Ñ Z be a contraction.
We write AutZ pX, Bq for the group of automorphisms of pX, Bq over Z.
Lemma 2.17. Let φ : X Ñ Z be a projective morphism. Let T be an algebraic torus. Let pX, Bq be a log
canonical pair for which T 6 AutZ pX, Bq. Let φ : X 99K X 1 be a birational contraction and let B 1 :“ φ˚ B.
Assume that pX 1 , B 1 q is plt. Then, we have T 6 AutZ pX 1 , B 1 q.
Proof. Let E be the reduced exceptional locus of φ. Note that κpEq “ 0 (see, e.g., [22, Lemma 2.28]).
We argue that E is T-invariant. First, note that t ¨ E „ E for every t P T as T is a connected group.
As κpEq “ 0, we get that supppt ¨ Eq “ supppEq for every t P T. Thus, t ¨ E “ E for every t P T. Let
Y Ñ X be a T-equivariant log resolution of pX, B ` supppEqq. Let ψ : Y Ñ X 1 be the associated birational
contraction. Note that every exceptional divisor of ψ and the strict transform of B on Y are T-invariant. Let
EY be the reduced exceptional divisor of ψ and BY be the strict transform of B on Y . Then, we may run a
T-equivariant MMP for KY ` EY ` BY over Z. By the plt condition of pX 1 , B 1 q this MMP terminates with
a model Y 1 Ñ Z after contracting all the exceptional divisors of ψ. The previous statement follows from
Lemma 2.12. Let BY 1 be the push-forward of BY to Y 1 . Then, we get T 6 AutZ pY 1 , BY 1 q and Y 1 is small
birational to X 1 over Z. As pX 1 , B 1 q is plt, the map Y 1 99K X 1 is a composition of pKY 1 ` BY 1 q-flops over
Z. Each such a flop is defined by a Cartier semiample divisor which is linearly equivalent to a T-invariant
Cartier divisor. Thus, each such a flop is itself T-equivariant. We conclude that T 6 AutZ pX 1 , B 1 q. 
Lemma 2.18. Let X Ñ Z be a fibration and pX, Bq be a plt pair. Assume that Gm 6 AutZ pX, Bq,
tBu “ S0 ` S8 , and Gm acts as the identity on both divisors S0 and S8 . Then, every exceptional canonical
place of pX, Bq is Gm -invariant.
Proof. First, we may replace pX, Bq with a Gm -equivariant log resolution. By doing so, we replace pX, Bq
with a sub-plt pair. Note that pX, Bq is sub-klt in the complement of S0 ` S8 . Thus, the pair pX, Bq has
only finitely many canonical valuations on the complement of S0 ` S8 (see, e.g., [16, Proposition 2.36]).
Such canonical valuations can be extracted by recursively blowing-up the intersections of the irreducible
components of intersections of divisors in B ą0 . Such irreducible components are Gm -invariant. Thus, after
finitely many Gm -equivariant blow-ups, we obtain a Gm -equivariant model pY, BY q that is terminal in the
complement of SY,0 ` SY,8 , the strict transform of S0 ` S8 . Note that pY, BY q is log smooth and the only
exceptional canonical valuations over Y are obtained by blowing-up centers of codimension 2 contained in
either SY,0 or SY,8 (see [16, Corollary 2.31]). Since Gm acts as the identity on both SY,0 and SY,8 , we
conclude that such centers are Gm -invariant. Therefore, every exceptional canonical place of pY, BY q is
Gm -invariant. So, every exceptional canonical place of pX, Bq is Gm -invariant. 
 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 7

2.6. Standard P1 -links. In this subsection, we recall the concepts of P1 -links and strict conic fibrations,
and prove some lemmata regarding these objects.

Definition 2.19. Let φ : pX, Bq Ñ Z be a log Calabi–Yau fibration. We say that φ is a standard P1 -link if
the following conditions are satisfied:
(1) the divisor tBu consists of two prime divisors S0 and S8 ,
(2) the restrictions φ|Si : Si Ñ Z are isomorphisms for i P t0, 8u,
(3) the pair pX, Bq is plt, and
(4) each reduced fiber of φ is isomorphic to P1 .

Lemma 2.20. Let φ : pX, Bq Ñ Z be a standard P1 -link. Assume that X is klt. Then φ is a Mori fiber
space.

Proof. Since X is klt, we may run a KX -MMP over Z. Note that every reduced fiber of φ is isomorphic to
P1 and intersects both S0 and S8 . This follows by conditions (2) and (4) of Definition 2.19. If X Ñ X 1
is either a divisorial contraction or flipping contraction, then the images of S0 and S8 would intersect on
X 1 , leading to a contradiction. We conclude that the KX -MMP over Z terminates with a Mori fiber space
X Ñ Z. Thus φ is itself a Mori fiber space. 

Lemma 2.21. Let φ : X Ñ Z be a Mori fiber space. Let pX, Bq be a log Calabi–Yau pair over Z. Assume
that pX, Bq is plt and tBu has two disjoint components S0 and S8 that are horizontal over Z. Furthermore,
assume that S0 and S8 are Q-Cartier. Then, the morphism φ : pX, Bq Ñ Z is a standard P1 -link.

Proof. Note that S8 is ample over Z. Hence, all the fibers of φ are one-dimensional. Otherwise, S0 would
intersect S8 . In particular, we conclude that tBu consists precisely of S0 and S8 . Indeed, any other
component of tBu would be vertical over Z and thus intersect S8 , violating the plt condition. This implies
condition (1) of Definition 2.19. Since pX, Bq is plt, both divisors S0 and S8 are normal. In particular,
pX, B ´ S0 q is plt Fano over Z, so the morphism φ|S0 : S0 Ñ Z has connected fibers. If φ|S0 contracts a
curve, then S8 would intersect such curve, contradicting the fact that S0 and S8 are disjoint. Thus, we get
that φ|S0 : S0 Ñ S8 is a bijection between normal varieties, and so it is an isomorphism. Anlogously, we
conclude that φ|S8 : S8 Ñ Z is an isomorphism. This proves condition (2) of Definition 2.19.
Finally, it suffices to show that condition (4) of Definition 2.19 is satisfied. For this regard, note that
pX, B ´ ǫpS0 ` S8 qq is klt and Fano over Z. Thus, by the relative version of Kawamata-Viehweg vanishing,
we get R1 φ˚ OX “ 0. In particular, for every fiber Xz , we have H 1 pXz , OXz q “ 0. Thus, each reduced
fiber is isomorphic to a tree of smooth rational curves. As both S0 and S8 are ample over Z, disjoint, and
intersect each fiber at a single point, we conclude that each reduced fiber is isomorphic to P1 . 

Definition 2.22. Let φ : X Ñ Z be a fibration. We say that φ : pX, Bq Ñ Z is a strict conic fibration if the
following conditions are satisfied:
(1) the morphism φ is a conic fibration with ρpX{Zq “ 1,
(2) the pair pX, Bq is log Calabi–Yau over Z,
(3) the divisor tBu has two prime components S0 and S8 that are horizontal over Z, and
(4) the prime divisors S0 and S8 are disjoint.

Lemma 2.23. Let φ : pX, Bq Ñ Z be a strict conic fibration. Let pZ, BZ q be the log pair induced by the
canonical bundle formula. Assume that pZ, BZ q is Q-factorial and dlt. Then, there exists 0 ď Γ ď B such
that φ : pX, Γq Ñ Z is a standard P1 -link.
8 J. MORAGA

Proof. Note that for ǫ ą 0 small enough, the pair pX, B ´ ǫφ˚ tBZ uq is plt. Set Γ :“ B ´ ǫφ˚ tBZ u. Note that
tΓu has two disjoint components that are horizontal over Z. Then, we can apply Lemma 2.21 to conclude
that φ : pX, Γq Ñ Z is a standard P1 -link. 

3. Geometry of standard P1 -links

In this section, we will prove Theorem 1.4 that gives a geometric characterization of standard P1 -links.
First, we prove two lemmata. The following lemma is proved by Kollár in the last paragraph of the proof
of [14, Proposition 4.37]. It states that a Q-factorial standard P1 -link is locally the finite quotients of a
product with pP1 , Σ1 q.

Lemma 3.1. Let X be a Q-factorial variety. Let φ : pX, Bq Ñ Z be a standard P1 -link. Let pZ, BZ q be the
log pair induced by the canonical bundle formula. For any closed point z P Z, there is an affine neighborhood
U of z in Z satisfying the following. Let pU, BU q be the restriction of pZ, BZ q to U and pXU , BXU q be the
restriction of pX, Bq to the preimage XU of U via φ. Then, there is a crepant finite surjective morphism
f : pV, BV q Ñ pU, BU q making the following diagram commutative:

fX
pXU , BU q o pV, BV q ˆ pP1 , Σ1 q
φU π1
 
pU, BU q o pV, BV q
f

where fX is a crepant finite surjective morphism, φU is the restriction of φ to XU , and π1 is the projection
to the first component.

Conversely, we show that a log Calabi–Yau fibration which is a finite quotient of a product with pP1 , Σ1 q
is indeed a standard P1 -link. In the converse statement, we do not require X to be Q-factorial.

Lemma 3.2. Let φ : pX, Bq Ñ pZ, BZ q be a crepant log Calabi–Yau fibration with pZ, BZ q klt. Assume there
exists a commutative diagram of log pairs:
fX
pX, Bq o pV, BV q ˆ pP1 , Σ1 q
φ π1
 
pZ, BZ q o pV, BV q
f

where the horizontal arrows are crepant finite surjective morphisms and π1 is the projection to the first
component. Then φ is a standard P1 -link.

Proof. Since pZ, BZ q is klt and f is crepant, we conclude that pV, BV q is klt. In particular, the pair pV, BV qˆ
pP1 , Σ1 q is plt. Since fX is crepant and finite, we have that pX, Bq is plt. This gives condition (3) of
Definition 2.19. Furthermore, by Riemann-Hurwitz we conclude that tBu contains precisely two prime
divisors S0 and S8 , namely the images of V ˆ t0u and V ˆ t8u, respectively. This proves condition (1) of
Definition 2.19. By [16, Proposition 5.51], we know that S0 and S8 are normal varieties that are disjoint.
We argue that S0 and S8 are Q-Cartier divisors. Let X̄ Ñ X be the Galois closure of fX , so we have a
 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 9

commutative diagram:

pX, Bq
fX ♦ ♦♦
♦♦♦ h
♦♦♦♦
w ♦♦ 
pX, Bq o pV, BV q ˆ pP1 , Σ1 q
fX

of crepant finite surjective morphisms. In the previous diagram, the morphisms h and fX are crepant Galois
morphisms and pX, Bq is a plt sub-pair. Let G be the Galois group associated with the finite Galois morphism
fX . Since V ˆ t0u is a Cartier divisors, we have that h˚ pV ˆ t0uq is a Cartier divisor on X. In particular,
h˚ pV ˆ t0uq is a G-invariant Cartier divisor. Hence, the divisor S0 is Q-Cartier. Analogously, the divisor
S8 is Q-Cartier.
Both S0 and S8 are ample over Z. Indeed, this follows from the projection formula and the fact that
both V ˆ t0u and V ˆ t8u are ample over V . Therefore, the pair pX, B ´ S0 q is plt and Fano over Z.
In particular, the morphism φ|S8 : S8 Ñ Z has connected fibers. If φ|S8 : S8 Ñ Z contracts a curve,
then S0 must intersect such a curve, contradicting the fact that S0 and S8 are disjoint. We conclude that
φ|S8 : S8 Ñ Z is a bijection between normal varieties and so it is an isomorphism. Analogously, we conclude
that φ|S0 : S0 Ñ Z is also an isomorphism. This gives condition (2) of Definition 2.19.
Note that all fibers of φ are curves, otherwise, S0 and S8 would intersect as they would induce ample
divisors on a positive-dimensional variety. Applying the relative version of Kawamata-Viehweg vanishing
theorem to the klt pair pX, B ´ ǫ1 S1 ´ ǫ2 S2 q, which is klt Fano over Z, we get R1 φ˚ OX “ 0. Thus, for
every closed point z P Z, the fiber Xz satisfies that H 1 pXz , OXz q “ 0. Thus, every fiber Xz is a tree of
smooth rational curves. Since both S0 and S8 can only intersect each fiber Xz at a single point, and they
are both ample over Z, we conclude that each reduced fiber is a rational curve. This gives condition (4) of
Definition 2.19. 

Now, we turn to give a proof of the geometric characterization of P1 -links.

Proof of Theorem 1.4. We set n :“ dim X. We will prove the statement in six steps. In the first three steps,
we perform birational modifications to produce a rank 2 split vector bundle over a suitable resolution of
Z. In the fourth step, we use this vector bundle to endow X with a torus action. In the fifth step, we use
the torus action on X to prove the statement in the Q-factorial case. In the last step, we deal with the
non-Q-factorial case.

Step 1: In this step, we produce a P1 -link on a resolution of singularities of Z.

Let S0 and S8 be the two components of tBu that dominate Z. We denote by pZ, BZ q be the klt pair
induced by the canonical bundle formula on Z. Let Z 1 Ñ Z be a log resolution of pZ, BZ q. Let X 1 be the
normalization of the fiber product Z 1 ˆZ X. Let pX 1 , B 1 q be the log pull-back of pX, Bq to X 1 . We denote
by S01 and S81
the pull-backs of S0 and S8 to X 1 , respectively. Note that B 1 may not be an effective divisor,
however, all its components with negative coefficients are vertical over Z 1 . Let U be an open affine in Z. Let
pU, BU q be the restriction of pZ, BZ q to U . Let U 1 be the preimage of U in Z 1 and let pU 1 , BU 1 q be the log
pull-back of pU, BU q to U 1 . Let pXU , BXU q be the restriction of pX, Bq to the preimage of U in X. We denote
by φU : XU Ñ U the restriction of φ to XU . By Lemma 3.1, up to shrinking U , we have a commutative
10 J. MORAGA

diagram:
fX
pXU , BU q o pV, BV q ˆ pP1 , Σ1 q
φU π1
 
pU, BU q o pV, BV q
f

where f and fX are crepant finite surjective morphism and π1 is the projection to the first component. Let
XU1 be the normalization of XU ˆU U 1 and let pXU1 , BXU1 q be the log pull-back of pXU , BU q to XU1 . Let V 1
be the normalization of V ˆU U 1 and let pV 1 , BV 1 q be the log pull-back of pU, BU q to V 1 . By the universal
property of fiber products, we obtain a commutative diagram of crepant morphisms:
g
(3.1) pXU1 , BXU1 q o pV 1 , BV 1 q ˆ pP1 , Σ1 q

φU 1 π1
 
pU 1 , BU 1 q o pV 1 , BV 1 q
h

where π1 is the projection to the first component, φU 1 is a fibration, and h is a crepant finite surjective
morphism.
We argue that g is a finite morphism. Let SXU1 ,0 and SXU1 ,8 be the restrictions of S01 and S8 1
to XU1 ,
respectively. Note that both SXU1 ,0 and SXU1 ,8 are Q-Cartier. Furthermore, the divisors SXU1 ,0 and SXU1 ,8
pull-back to positive multiples of V 1 ˆ t0u and V 1 ˆ t8u, respectively. Assume that g contracts a curve C.
If C is vertical over V 1 , then C is a fiber of the projection π1 and
0 “ SXU1 ,0 ¨ gpCq “ g ˚ pSXU1 ,0 q ¨ C “ m0 pV 1 ˆ t0uq ¨ C “ m0 ą 0,
leading to a contradiction. If C is horizontal over V 1 , then by the commutativity of the diagram (3.1), we
get that h contracts a curve. This leads to a contradiction as well. We conclude that g is finite.
Note that we have all the conditions to apply Lemma 3.2 except possibly one. Indeed, the couples in the
commutative diagram (3.1) may be sub-pairs. Hence, we need to add a suitable effective divisor to make
them log pairs. Note that pU 1 , BU 1 q is a log smooth sub-pair and the coefficients of BU 1 are strictly less than
one. Furthermore, h is a finite morphism. We pick Γ ě 0 on U 1 , a simple normal crossing divisor, such
that 1 ´ r1P ď coeff P pBU 1 ` Γq ă 1 for every prime divisor P of U 1 , where rP is the ramification index of
h at P . Then, we have that both pU 1 , BU1 ` Γq and pV 1 , BV 1 ` h˚ Γq are a klt pairs. Furthermore, the pair
pV 1 , BV 1 ` h˚ Γq ˆ pP1 , Σ1 q is a plt pair, so pXU 1 , BXU 1 ` φ˚U 1 Γq is a plt pair. Hence, we get a commutative
diagram:
g
pX 1 , BX 1 ` φ˚ 1 Γq o
U U U pV 1 , BV 1 ` h˚ Γq ˆ pP1 , Σ1 q

φU 1 π1
 
pU 1 , BU 1 ` Γq o pV 1 , BV 1 ` h˚ Γq
h
where the horizontal arrows are crepant finite surjective morphisms and π1 is the projection to the first
component. Note that φU 1 is a crepant contraction for the log Calabi–Yau pair pXU1 , BXU1 ` φ˚U 1 Γq and
pU 1 , BU 1 ` Γq is klt. By Lemma 3.2, we conclude that pXU1 , BXU1 ` φ˚U 1 Γq Ñ U 1 is a standard P1 -link. As U 1
is smooth, all the vertical divisors of XU 1 Ñ U 1 are Q-Cartier. This implies that pXU1 , SXU1 ,0 ` SXU1 ,8 q Ñ U 1
is a standard P1 -link as well. Thus, we conclude that pX 1 , S01 ` S8
1
q Ñ Z 1 is a standard P1 -link. This follows,
as being a standard P -link is a local property on the base of the fibration. Note tat both S01 and S8
1 1
are
 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 11

Q-Cartier divisors.

Step 2: In this step, we modify the standard P1 -link over Z 1 such that it has no multiple fibers over
codimension one points.

Assume that pX 1 , S01 ` S8 1

q Ñ Z 1 has a fiber of multiplicity m over the prime divisor P Ă Z 1 . Let
1˚
mEP :“ φ P . Then, the variety X 1 has a toric singularity of orbifold index m at the generic point of
˚
Q :“ S01 X φ1 P (see [23, Theorem 1]). In particular, there is a blow-up Y Ñ X 1 that extracts a Q-Cartier
divisor EQ with center Q and log discrepancy m 1
with respect to the pair pX 1 , S01 `S8
1
q (see e.g., [21, Theorem
1 1 1 1
1]). Let pY, SY,0 ` SY,8 ` p1 ´ m qEQ q be the log pull-back of pX , S0 ` S8 q to Y . Note that both SY,0 and
SY,8 are Q-Cartier divisors. By the canonical bundle formula, the divisor EQ appears with multiplicity 1 in
˚
φ1 P . Note that the strict transform of EP in Y is a degenerate divisor over Z 1 . Hence, we may contract
the strict transform of EP over Z 1 with a birational contraction Y 99K Y 1 . This birational contraction can
1 1
be obtained by running a pKY 1 ` S8 ` p1 ´ m qEQ q-MMP over Z 1 . Let EQ,Y 1 be the strict transform of EQ
in Y . Let SY 1 ,i be the strict transform of Si in Y 1 . Then, we have a commutative diagram
1 1

π ` ` 1
˘ ˘
pX 1 , S01 ` S81
q o❴ ❴ ❴ ❴ ❴ ❴ ❴ Y 1 , SY 1 ,0 ` SY 1 ,8 ` 1 ´ m EQ,Y 1
❑❑❑ ❥
❑❑❑
❥❥ ❥❥❥❥❥
❑❑❑❑ ❥❥❥
φ1 ❑% ❥❥❥❥ φY 1
1 u❥❥
Z

where π is a crepant birational map. Note that φ1 is a Mori fiber space by Lemma 2.20. Indeed, we know
that S01 and S81
are Q-Cartier. Henceforth, the morphism φY 1 is a Mori fiber space. Note that SY 1 ,0 and
SY 1 ,8 are Q-Cartier. By Lemma 2.21, we conclude that φY 1 is a standard P1 -link whose fiber over P is not
multiple. Since φ1 has only finitely many multiple fibers over codimension one points, we can repeat the
above procedure to obtain a standard P1 -link pX 2 , S02 ` S8
2
q Ñ Z 1 with no multiple fibers over codimension
one points. The pair pX , S0 ` S8 q is crepant equivalent to a log Calabi–Yau pair pX 2 , S02 ` S8
1 1 1 2
` Sq over
1 1
Z , where S is an effective divisor with standard coefficients that is vertical over Z .

Step 3: In this step, we show that X 2 is the projectivization of a rank 2 split vector bundle over Z 1 .

First, we argue that X 2 is smooth. Let z P Z 1 be a closed point. Let H1 , . . . , Hn´1 be Cartier divisors
containing z such that pZ 1 , H1 ` ¨ ¨ ¨ ` Hn1 ; zq is simple normal crossing. Let HX,i be the pull-back of Hi to
řn´1
X. Then, the pair pX 2 , S02 ` S8 2
` i“1 HX,i q is log canonical and log Calabi–Yau over Z 1 . By [23, Theorem
1], we conclude that φ2 : X 2 Ñ Z 1 is formally toric over a neighborhood of z P Z 1 . Further, all the reduced
fibers of the toric morphisms are isomorphic to P1 . Thus, the morphism X 2 Ñ Z 1 is formally isomorphic
to XpΣq Ñ Zpσq with σ “ xe1 , . . . , en´1 y. The fan Σ has maximal cones xpe1 , p1 q, . . . , pen´1 , pn q, p0, 1qy and
xpe1 , p1 q, . . . , pen´1 , pn´1 q, p0, ´1qy where the pi are integers. The fact that each pi is an integer follows from
the fact that φ2 has no multiple fibers over codimension one points. We conclude that XpΣq is smooth, so
X 2 is smooth on a neighborhood of the fiber over z P Z 1 . Since z P Z 2 is any closed point, we conclude that
X 2 is smooth. Therefore, S02 is a Cartier divisor that restricts to a single point on general fibers. By [11,
Lemma 2.12], we conclude that X 2 is a projectivized rank 2 vector bundle over Z 1 . Since φ2 admits a section
2
S8 , which is disjoint from S02 , we conclude that such vector bundle splits.
12 J. MORAGA

Step 4: In this step, we show that the pair pX, S0 ` S8 q admits a Gm -action.

Since X 2 Ñ Z 1 is the projectivization of a rank 2 split vector bundle, we have Gm 6 AutZ 1 pX 2 q.

Since S02 and S8 2
are sections of the vector bundle, we get Gm 6 AutZ 1 pX 2 , S02 ` S8 2
q. In particular,
we have Gm 6 AutZ pX , S0 ` S8 q. Let pX , B q be the log pull-back of pX, Bq to X 2 . Then, we get
2 2 2 2 2

Gm 6 AutZ pX 2 , B 2 q. Indeed, B 2 consists of S02 ` S8

2
and divisors that are vertical over Z 1 . The birational
map X 99K X only extract divisors with log discrepancy at most one with respect to pX 2 , B 2 q whose centers
2

are contained in either S02 or S8

2
. Note that both S02 and S82
are contained in the fixed point locus of Gm .
Thus, we can extract the exceptional divisors of X 99K X torus equivariantly from X 2 (see Lemma 2.18).
2

More precisely, there is a Gm -equivariant projective birational crepant morphism pY, BY q Ñ pX 2 , B 2 q such
that Y 99K X is a contraction over Z. By Lemma 2.17, we conclude that Gm 6 AutZ pX, Bq and so
Gm 6 AutZ pX, S0 ` S8 q.

Step 5: In this step, we finish the proof of the Q-factorial case by using the torus action on pX, S0 ` S8 q.

Note that all the reduced fibers of φ : pX, S0 ` S8 q Ñ Z are isomorphic to pP1 , t0u ` t8uq. Further-
more, we have Gm 6 AutZ pX, S0 ` S8 q. We conclude that the Gm -orbits are contained in fibers of φ.
In particular, one-dimensional orbit closures of the Gm -action on X are disjoint. Let Z0 be the normal-
ized Chow quotient of X by the Gm -action. Let S be a divisorial fan on Z0 such that there is a Gm -
equivariant isomorphism XpSq » X (see [2, Theorem 5.6]). Since orbit closures of X are disjoint, every
polyhedral divisor D P S has affine support. Let Z be the weighted monoid of the Gm -action on X. Let
D be a p-divisor on S with weighted cone Qě0 . Then, the variety XpDq r » XpDq is equivariantly iso-
À
morphic to SpecU0 p mě0 OX pmD0 qq for some Q-Cartier Q-divisor D0 on U0 an affine open of Z0 (see,
e.g., [24, Theorem 1]). By [1, Corollary 8.12], these Gm -invariant open affine varieties equivariantly glue
À
to SpecZ0 p mě0 OX pmDqq for some Q-Cartier Q-divisor D on Z0 . Analogously, the p-divisors of S with
À
weighted cone Qď0 equivariantly glue to SpecZ0 p mď0 OX pmDqq. We conclude that X is isomorphic to the
À À À
equivariant gluing of SpecZ0 p mě0 OX pmDqq and SpecZ0 p mď0 OX pmDqq along SpecZ0 p mPZ OX pmDqq
Thus, X is isomorphic to PpOZ0 ‘OZ0 pDqq. Note that the fibers of X Ñ Z0 are precisely the one-dimensional
orbit closures for the torus action. This implies that Z0 » Z. This finishes the proof in the Q-factorial case.

Step 6: In this step, we finish the proof in the general case.

Let pX, Bq Ñ Z be a standard P1 -link. Since pX, Bq is plt, it admits a small Q-factorialization Y Ñ X.
Let pY, BY q be the log pull-back of pX, Bq to Y . We run a KY -MMP over Z. Let Y 99K X 1 be the steps
of this MMP and X 1 Ñ Z 1 be the induced Mori fiber space. Note that every prime divisor on Y that is
vertical over Z is the closure of the pull-back of a prime divisor on Z. Then, the relative MMP Y 99K X 1
over Z contracts no divisors. Let B 1 be the push-forward of BY to X 1 so pX 1 , B 1 q is a plt pair and tB 1 u
has two disjoint sections over Z 1 . By Lemma 2.21, we conclude that pX 1 , B 1 q Ñ Z 1 is a standard P1 -link
with X 1 being Q-factorial. By the previous step, we have Gm 6 AutZ 1 pX 1 , B 1 q. In particular, we have
Gm 6 AutZ pX 1 , B 1 q. By Lemma 2.17, we get Gm 6 AutZ pX, Bq. Then, the same argument as in Step 5
finishes the proof. 

Now, we turn to prove a structural theorem for standard P1 -links for low-dimensional varieties.
 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 13

Proof of Theorem 1.5. Let X be a variety of dimension at most three. Let pX, Bq Ñ Z be a standard P1 -link.
Let pZ, BZ q be the log pair induced on Z by the canonical bundle formula. Note that Z is either a surface or
a curve. By Theorem 1.4, we know that X is the projectivization of a split Q-vector bundle of rank 2 on Z.
Thus X is isomorphic to PpOZ ‘ OZ pDqq for a Q-Cartier Q-divisor D on Z. Let π : Z 1 Ñ Z be a minimal
resolution of Z. Let pZ 1 , BZ 1 q be the log pull-back of pZ, BZ q to Z 1 . Let D1 be the pull-back of D to Z 1 and
˚
X 1 » PpOZ 1 ‘ OZ 1 pD1 qq. Let φ1 : X 1 Ñ Z 1 be the induced fibration. Let B 1 :“ φ1 BZ ` S01 ` S8 1
, where S01
1 1
and S8 are the strict transforms of S0 and S8 in X , respectively. Then, we have a commutative diagram:

π1
pX, Bq o pX 1 , B 1 q
φ φ1
 
pZ, BZ q o π
pZ 1 , BZ 1 q

where the vertical arrows are crepant contractions and the horizontal arrows are crepant birational mor-
phisms. Proceeding as in Step 2 of the proof of Theorem 1.4, we get a commutative diagram:
π1 π2
pX, Bq o pX 1 , B 1 q o pX 2 , B 2 q
r
φ φ1 rrrrr
r 2
  r x rr φ
1
pZ, BZ q o π
pZ , BZ 1 q

where and X 2 Ñ Z 1 is the projectivization of a rank 2 split vector bundle. In particular, we have that
tB 2 u “ S02 ` S82
are two disjoint smooth sections of X 2 Ñ Z 1 . Let k :“ pS02 q2 . If k “ 0, then X 2 is a
trivial projective bundle and we are done. Assume that k ą 0. We blow-up X 2 at k smooth points along
S02 and blow down the strict transform of the corresponding fiber over Z 1 . By doing so, we obtain a crepant
birational map π 3 : pZ 1 , BZ 1 q ˆ pP1 , Σ1 q 99K pX 2 , B 2 q over Z 1 . The product pZ 1 , BZ 1 q ˆ pP1 , Σ1 q carries the
structures of a trivial P1 -bundle over Z 1 and so we obtain a commutative diagram:
π1 π2 π 3
pX, Bq o pX 1 , B 1 q o pX 2 , B 2 q o pZ 1 , BZ 1 q ˆ pP1 , Σ1 q
2 qq ❣❣❣❣
φ qq ❣ ❣ ❣❣❣
1
φ φ q ❣❣
qq ❣❣❣❣❣π1
  xqqqs❣❣❣❣❣❣
pZ, BZ q o π
pZ 1 , BZ 1 q.

Note that the composition ψ ´1 :“ π 3 ˝ π 2 ˝ π 1 : pZ 1 , BZ 1 q ˆ pP1 , Σ1 q 99K pX, Bq is a crepant birational map.
This finishes the proof. 

4. Birational complexity of log Calabi–Yau 3-folds

In this section, we prove that a log Calabi–Yau 3-fold of index one and coregularity zero has birational
complexity different than 1.

Proof of Theorem 1.2. Let pX, Bq be a log Calabi–Yau 3-fold of index one and coregularity zero. Since
pX, Bq has index one its birational complexity is an integer. By [20, Theorem 1.4], we know that cbir pX, Bq P
t0, 1, 2, 3u. Thus, it suffices to show that the birational complexity of pX, Bq is not one. Assume, for the sake
of contradiction, that cbir pX, Bq “ 1. By [22, Theorem 1.1], there exists a crepant birational model pX 1 , B 1 q
14 J. MORAGA

of pX, Bq and a sequence of Mori fiber spaces

π1 π2 π3
X1 / Z1 / Z2 / SpecpKq.

where at least two of the three fibrations π1 , π2 and π3 are strict conic fibrations. Let pZ1 , BZ1 q and
pZ2 , BZ2 q be the log Calabi–Yau pairs induced by the canonical bundle formula. We may assume that
cbir pX, Bq “ cpX 1 , B 1 q.
First, assume that π1 is a strict conic fibration. By [22, Lemma 2.36], we may assume that pZ1 , BZ1 q
is Q-factorial and dlt. Hence, by Lemma 2.23, we conclude that pX 1 , Γ1 q Ñ Z1 is a standard P1 -link for a
suitable effective divisor 0 ď Γ1 ď B 1 . By Theorem 1.4, we conclude that X 1 Ñ Z1 is a projectivized split
Q-vector bundle. In particular, the log Calabi–Yau pair pZ1 , BZ1 q is a log Calabi–Yau pair of coregularity
zero and index one. Hence, we have pZ1 , BZ1 q »cbir pP2 , Σ2 q. Note that the birational map φ : P2 99K Z1 only
extracts log canonical places and canonical places. Each such canonical place (resp. log canonical place)
induce a canonical place (resp. log canonical place) on pX 1 , B 1 q by adjunction. Thus, there is a crepant
birational map φX : pX 2 , B 2 q 99K pX 1 , B 1 q, with B 2 ě 0, making the following diagram commutative:
φX
pX 1 , B 1 q o❴ ❴ ❴ pX 2 , B 2 q
π1 π11
 
pZ1 , BZ1 q o❴ ❴ ❴ pP2 , Σ2 q
φ

where π11 : pX 2 , B 2 q Ñ P2 is a strict conic fibration. In particular, we get ρpX 2 q “ 2 and |B 2 | “ 5. We

conclude that pX 2 , B 2 q is a toric pair and so cbir pX, Bq “ 0, leading to a contradiction.
From the previous paragraph, we know that π1 is not a strict conic fibration. Hence, B 1 contains a unique
prime divisor that dominates Z1 . In this case, we have cpX 1 , B 1 q “ 2 and ρpX 1 q “ ρpZ1 q ` 1. We conclude
that |B 1 | “ ρpZ1 q ` 3. Thus |B 1 | has at least ρpZ1 q ` 2 prime components that are vertical over Z1 . Since
X 1 Ñ Z1 is a Mori fiber space, the divisor tBZ1 u has at least ρpZ1 q ` 2 prime components. We conclude
that tBZ1 u has at least ρpZ1 q ` 2 prime components. Therefore, pZ1 , BZ1 q is a toric log Calabi–Yau pair. In
particular, there is a crepant birational map pP2 , Σ2 q 99K pZ1 , BZ1 q that only extract log canonical places.
By [20, Lemma 2.1], there is a crepant birational model pX 2 , B 2 q of pX 1 , B 1 q that admits a crepant fibration
˚
to pP2 , Σ2 q. Let π11 : pX 2 , B 2 q Ñ pP2 , Σ2 q be the corresponding fibration. We have that π11 pΣ2 q is contained
in the support of B 2 . In particular, B 2 has four prime components; three prime components that are vertical
over P2 and dominate the prime components of Σ2 , and a horizontal prime component that maps 2-to-1 to
P2 . Let S be the horizontal component of B 2 and S0 , S1 , and S2 be the vertical components. Note that
the pair pX 2 , Sq is plt. Indeed, it has no log canonical centers, different from S, that are horizontal over P2
and it has not log canonical centers that are vertical over P2 . In particular, the divisor S is normal (see [17,
Proposition 5.51]). Let pS, BS q be the log Calabi–Yau pair obtained by adjunction of pX 2 , B 2 q to S. Then
the morphism pS, BS q Ñ pP2 , Σ2 q is a finite crepant morphism. We conclude that pS, BS q is a toric log
Calabi–Yau pair.
Since pS, BS q is a toric log Calabi–Yau surface, each intersection S X Si is irreducible for i P t0, 1, 2u. We
argue that S X Si Ñ Hi is a 2-to-1 cover for each i P t0, 1, 2u. Here, Hi is the i-th coordinate hyperplane
of P2 . It suffices to show the statement for i “ 0. Let pS0 , BS0 q be the slc Calabi–Yau pair induced by
adjunction of pX 2 , B 2 q to S0 (see [10, Example 2.6]). Consider the fibration f0 : S0 Ñ H0 . Since S0 is prime,
a general fiber F of f0 is an irreducible slc curve. Let pF, BF q be the restriction of pS0 , BS0 q to F . Note that
pF, BF q is an slc log Calabi–Yau pair. Since S X S0 intersects F , we conclude that F is smooth. Otherwise,
 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 15

its normalization is not a curve of log Calabi–Yau type. There are two possibilities; either S X S0 Ñ H0 is
2-to-1 or there is a 1-dimensional log canonical center Q of pX 2 , B 2 q, that dominates H0 , contained in S0
but not contained in S, S1 , or S2 . In the latter case, the pair pX 2 , S0 ; Qq is not plt. By the classification of
2-dimensional log canonical singularities, we conclude that 2pKX 2 ` B 2 q „ 0 and KX 2 ` B 2 is not linearly
trivial around Q (see [18, §3, Figure 3]). This contradicts the fact that pX 2 , B 2 q has index one. We conclude
that S X Si Ñ H0 is 2-to-1 for each i.
Let S0_ Ñ S0 be its normalization and pS0_ , B0_ q be the log pull-back of pS0 , BS0 q to S0_ . Note that
pS0 , B0_ q is a log Calabi–Yau surface of index one and coregularity zero. Let S0_ Ñ P1 be the contraction
_

associated to the Stein factorization of S0_ Ñ H0 . Then, the log Calabi–Yau pair induced on P1 by the
canonical bundle formula is pP1 , Σ1 q. Up to a crepant birational transformation over P1 , we may assume
that S0_ is Q-factorial and B0_ has three prime components; two vertical components P1 and P2 that map
to t0u and t8u respectively, and one horizontal component P that maps 2-to-1 to P1 . In particular, P and
P1 intersect at a single point. This implies that P1 contains a closed point q that is a log canonical center
of pS0_ , B0_ q which is not contained in either P2 or P . Hence, the pair pS0_ , B0_ q has a D-type singularity
at q (see [18, §3, Figure 3]). This implies that KS_0 ` B0_ does not have index one at q. Then, we get a
contradiction. We conclude that the birational complexity of pX, Bq is not one. 

5. Crepant fibrations to projective spaces

In this section, we construct crepant fibration to projective spaces for log Calabi–Yau 3-folds of index one
and coregularity zero.

Proof of Theorem 1.3. Let pX, Bq be a log Calabi–Yau 3-fold of index one and coregularity zero. By Theo-
rem 1.2, we know that cbir pX, Bq P t0, 2, 3u. If cbir pX, Bq “ 3, then the statement is vacuous as we can simply
consider the structure morphism X Ñ SpecpKq. If cbir pX, Bq “ 0, then we know that pX, Bq »cbir pP3 , Σ3 q
by [20, Theorem 1.6]. Thus, it suffices to consider the case cbir pX, Bq “ 2 and show that pX, Bq has a crepant
birational model that admits a crepant contraction to pP1 , Σ1 q.
Let pX, Bq be a log Calabi–Yau 3-fold of index one and coregularity zero. Assume that cbir pX, Bq “ 2.
By [22, Theorem 1.1], there exists a crepant birational model pX 1 , B 1 q of pX, Bq and a sequence of Mori fiber
spaces
π1
X1 / Z1 π2 / . . . πk / SpecpKq,
where 2 ď k ď 3 and at least one of the Mori fiber spaces is a strict conic fibration. If πk is a strict conic
fibration, then the statement follows. If π1 is a strict conic fibration, then proceeding as in the proof of
Theorem 1.2, we get cbir pX, Bq “ 0, leading to a contradiction. We conclude that k “ 3 and π2 is a strict
conic fibration.
From now on, we assume that k “ 3 and π2 is a strict conic fibration. Let pZ1 , BZ1 q be the log Calabi–Yau
pair induced by the canonical bundle formula. Let π2 : pZ1 , BZ1 q Ñ P1 be the strict conic fibration. By [8,
Corollary 3], we know that pZ1 , BZ1 q has index two and coregularity zero. Furthermore, we may assume that
B 1 has a unique prime component S that dominates Z1 and π1˚ tBZ1 u is contained in the support of B 1 . Note
that S is a normal divisor as pX 1 , Sq is plt. Let pS, BS q be the log Calabi–Yau pair obtained by performing
adjunction of pX 1 , B 1 q to S. Then, we have a crepant finite surjective morphism pS, BS q Ñ pZ1 , BZ1 q of
degree 2. Let Z 1 Ñ Z1 be a minimal resolution. Since pZ1 , BZ1 q Ñ P1 is a strict conic fibration, all the
singularities of Z1 are contained in tBZ1 u. In particular Z 1 Ñ Z1 only extracts divisors with log discrepancy
ď 21 with respect to pZ1 , BZ1 q. Let pZ 1 , BZ 1 q be the log pull-back of pZ1 , BZ1 q to Z 1 . By Lemma 2.13, we
have a crepant birational model pX 2 , B 2 q of pX 1 , B 1 q, with B 2 effective, that admits a crepant fibration to
16 J. MORAGA

pZ2 , BZ2 q, where pZ2 , BZ2 q Ñ pZ1 , BZ1 q is a dlt modification. Further, the morphism X 2 Ñ Z2 is a Mori
fiber space. Note that Z2 is a smooth surface. Every fiber of Z2 Ñ P1 that contains a log canonical center
is contained in tBZ2 u. This follows from the connectedness of log canonical centers (see [9, Theorem 1.1]).
Thus, when we run a KZ2 -MMP over P1 , every curve contracted by the MMP is either disjoint from the log
canonical centers or intersects positively a curve with coefficient one in tBZ2 u. Thus, by Lemma 2.14 and
Lemma 2.15, we can find a crepant birational model pX p3q , B p3q q of pX 2 , B 2 q, with B p3q effective, that admits
a crepant contraction to pZ3 , BZ3 q. The morphism X p3q Ñ Z3 is a Mori fiber space and Z3 is a Hirzebruch
surface.
First, we assume that every vertical component of BZ3 has coefficient 21 . In this case, the pair pZ3 , BZ3 q
is plt. We argue that there exists a crepant birational map pP1 ˆ P1 , Γq 99K pZ3 , BZ3 q that only extract
divisors with log discrepancy ď 12 with respect to pZ3 , BZ3 q and tΓu contains two disjoint sections for the
projection to the first component. Indeed, the log Calabi–Yau pair pZ3 , BZ3 q has a prime component P that
is vertical over P1 and has coefficient 12 in BZ3 . The crepant birational map pP1 ˆ P1 , Γq 99K pZ2 , BZ2 q can
be obtained by inductively blowing up the intersection of P and one of the sections of the Hizebruch surface,
and then blowing down the strict transform of the corresponding fiber. By Lemma 2.13 and Lemma 2.15,
there exists a crepant model pX p4q , B p4q q of pX p3q , B p3q q, with B p4q effective, that admits a crepant fibration
to pP1 ˆ P1 , Γq. Note that pP1 ˆ P1 , Γq admits a crepant fibration to pP1 , Σ1 q by projecting to the second
component. Thus, the log Calabi–Yau pair pX p4q , B p4q q, which is crepant birational equivalent to pX, Bq,
admits a crepant fibration to pP1 , Σ1 q.
Finally, assume that there is a vertical component of BZ3 with coefficient 1. In this case, there exists a
crepant birational map pP1 ˆ P1 , Γq 99K pZ3 , BZ3 q that only extract log canonical places and tΓu contains
two disjoint sections for the projection to the first component. Then, by [20, Lemma 2.10] and Lemma 2.14,
there exists a crepant model pX p4q , B p4q q of pX p3q , B p3q q, with B p4q effective, that admits a crepant fibration
to pP1 ˆ P1 , Γq. The pair pP1 ˆ P1 , Γq admits a crepant fibration to pP1 , Σ1 q by projecting to the second
component. The log Calabi–Yau pair pX p4q , B p4q q is crepant birational equivalent to pX, Bq and admits a
crepant fibration to pP1 , Σ1 q. 

6. Examples and questions

In this section, we collect some examples and propose some questions. First, we give examples of log
Calabi–Yau pairs pX, Bq of dimension 3, index 1, and coregularity 0 with different birational complexities.

Example 6.1. Let pT, BT q be any toric log Calabi–Yau pair, i.e., BT is the reduced complement of the
algebraic torus on T . Then, we have cpT, BT q “ dim T ` ρpT q´ |BT | “ 0. Hence, pT, BT q is a log Calabi–Yau
pair of index 1, coregularity 0, and birational complexity 0.

Example 6.2. In this example, we construct a log Calabi–Yau pair pX, Bq of dimension 3, index 1, coreg-
ularity 0, and birational complexity 2.
Let X be a smooth cubic hypersurface in P4 . Then, X is not a rational variety [5, Theorem 0.12]. Let
H be a general hyperplane in P3 . The pair pP4 , X ` Hq is a dlt Fano pair of coregularity 2. Let Z be a
minimal log canonical center of pP4 , X ` Hq. Then, Z is a cubic surface in P3 . In particular, there exists
a 1-complement of coregularity zero pZ, Cq. By Kawamata-Viehweg vanishing theorem, the 1-complement
C lifts to a 1-complement CX of the log Fano pair pX, H|X q. Thus, we have that pX, H|X ` CX q is a log
Calabi–Yau 3-fold of index one and coregularity zero. Note that

cpX, H|C ` CX q “ dim X ` ρpXq ´ |H|C ` CX | ď 2.

 BIRATIONAL COMPLEXITY OF LOG CALABI–YAU 3-FOLDS 17

We argue that cbir pX, H|C `CX q “ 2. Assume otherwise that pX, H|C `CX q admits a log Calabi–Yau crepant
birational model pY, BY q for which cpY, BY q ă 2. Then, by Theorem 1.2, we would have cpY, BY q “ 0.
Indeed, the complexity of this log Calabi–Yau pair must be an integer and cannot be equal to one (see
Theorem 1.2). Thus, the log Calabi–Yau pair pX, H|C ` CX q would have a toric model, and hence X would
be rational. Leading to a contradiction. We conclude that cbir pX, H|C ` CX q “ 2.
Example 6.3. In this example, we construct a log Calabi–Yau pair pX, Bq of dimension 3, index 1, coreg-
ularity 0, and birational complexity 3.
Let X be the hypersurface defined by
trx0 : x1 : x2 : x3 : x4 s | x3 px30 ` x31 q ` x42 ` x0 x1 x2 x3 ` x4 px33 ` x34 q “ 0u Ă P4x0 ,...,x4 .
By [13, Example 3.2], we know that X is a terminal birationally superrigid Fano 3-fold. Let B :“ X X
tx4 “ 0u. Then, the log Calabi–Yau pair pX, Bq has index one and coregularity zero. As X is birationally
superrigid, [22, Corollary 1.2] implies that cbir pX, Bq “ 3.
We conclude by posing two natural questions about the values of the birational complexity of higher-
dimensional log Calabi–Yau pairs.
Question 6.4. Let pX, Bq be a log Calabi–Yau pair of dimension n ě 4, index one, and coregularity zero.
What are the possible values for the birational complexity cbir pX, Bq of pX, Bq? Are there such pairs with
birational complexity equal to one?
We know that every Fano variety X of coregularity zero admits either a 1-complement or a 2-complement
of coregularity zero (see, e.g., [6, Theorem 4]). Thus, we can extend the previous question to this setting.
Question 6.5. Let pX, Bq be a log Calabi–Yau pair of dimension n ě 2, index two, and coregularity zero.
What are the possible values for the birational complexity cbir pX, Bq of pX, Bq?
In the previous question, it is unclear which fractional numbers occur and how this reflects on the geometry
of the pair. For instance, in dimension one, the only possible birational complexity of pairs of coregularity
zero is zero.

References
[1] K. Altmann and J. Hausen. Polyhedral divisors and algebraic torus actions. Math. Ann., 334(3):557–607, 2006.
doi:10.1007/s00208-005-0705-8.
[2] K. Altmann, J. Hausen, and H. Süss. Gluing affine torus actions via divisorial fans. Transform. Groups, 13(2):215–242,
2008. doi:10.1007/s00031-008-9011-3.
[3] F. Ambro. Shokurov’s boundary property. J. Differential Geom., 67(2):229–255, 2004. URL
http://projecteuclid.org/euclid.jdg/1102536201.
[4] M. V. Brown, J. McKernan, R. Svaldi, and H. R. Zong. A geometric characterization of toric varieties. Duke Math. J.,
167(5):923–968, 2018. doi:10.1215/00127094-2017-0047.
[5] C. H. Clemens and P. A. Griffiths. The intermediate Jacobian of the cubic threefold. Ann. of Math. (2), 95:281–356, 1972.
doi:10.2307/1970801.
[6] F. Figueroa, S. Filipazzi, J. Moraga, and J. Peng. Complements and coregularity of Fano varieties, 2022, arXiv:2211.09187.
[7] S. Filipazzi. On a generalized canonical bundle formula and generalized adjunction. Ann. Sc. Norm. Super. Pisa Cl. Sci.,
21(5):1187–1221, 2020.
[8] S. Filipazzi, M. Mauri, and J. Moraga. The index of coregularity zero Calabi-Yau pairs, 2022, arXiv:2209.02925. To appear
in Algebra & Number Theory.
[9] S. Filipazzi and R. Svaldi. On the connectedness principle and dual complexes for generalized pairs. Forum Math. Sigma,
11:Paper No. e33, 2023. doi:10.1017/fms.2023.25.
18 J. MORAGA

[10] O. Fujino and Y. Gongyo. On canonical bundle formulas and subadjunctions. Michigan Math. J., 61(2):255–264, 2012.
doi:10.1307/mmj/1339011526.
[11] T. Fujita. On polarized manifolds whose adjoint bundles are not semipositive. In Algebraic geometry, Sendai, 1985, vol-
ume 10 of Adv. Stud. Pure Math., pages 167–178. North-Holland, Amsterdam, 1987. doi:10.2969/aspm/01010167.
[12] M. Gross, P. Hacking, and S. Keel. Mirror symmetry for log Calabi-Yau surfaces I. Publ. Math. Inst. Hautes Études Sci.,
122:65–168, 2015. doi:10.1007/s10240-015-0073-1.
[13] A.-S. Kaloghiros. Some examples of Calabi-Yau pairs with maximal intersection and no toric model. In Birational
geometry and moduli spaces, volume 39 of Springer INdAM Ser., pages 57–75. Springer, Cham, [2020] ©2020.
doi:10.1007/978-3-030-37114-2 5.
[14] J. Kollár. Singularities of the minimal model program, volume 200 of Cambridge Tracts in Mathematics. Cambridge
University Press, Cambridge, 2013. doi:10.1017/CBO9781139547895. With a collaboration of Sándor Kovács.
[15] J. Kollár and S. J. Kovács. Log canonical singularities are Du Bois. J. Amer. Math. Soc., 23(3):791–813, 2010.
doi:10.1090/S0894-0347-10-00663-6.
[16] J. Kollár and S. Mori. Classification of three-dimensional flips. J. Amer. Math. Soc., 5(3):533–703, 1992.
doi:10.2307/2152704.
[17] J. Kollár and S. Mori. Birational geometry of algebraic varieties, volume 134 of Cambridge Tracts in Mathematics. Cam-
bridge University Press, Cambridge, 1998. doi:10.1017/CBO9780511662560. With the collaboration of C. H. Clemens and
A. Corti, Translated from the 1998 Japanese original.
[18] J. e. a. Kollár. Flips and abundance for algebraic threefolds. Société Mathématique de France, Paris, 1992. Papers from
the Second Summer Seminar on Algebraic Geometry held at the University of Utah, Salt Lake City, Utah, August 1991,
Astérisque No. 211 (1992) (1992).
[19] C.-J. Lai. Varieties fibered by good minimal models. Math. Ann., 350(3):533–547, 2011. doi:10.1007/s00208-010-0574-7.
[20] M. Mauri and J. Moraga. Birational complexity and dual complexes, 2024, arXiv:2402.10136.
[21] J. Moraga. Extracting non-canonical places. Adv. Math., 375:107415, 12, 2020. doi:10.1016/j.aim.2020.107415.
[22] J. Moraga. Birational complexity and conic fibrations, 2024, arXiv:2403.17251.
[23] J. Moraga and R. Svaldi. A geometric characterization of toric singularities, 2021, arXiv:2108.01717.
[24] K. Watanabe. Some remarks concerning Demazure’s construction of normal graded rings. Nagoya Math. J., 83:203–211,
1981. doi:10.1017/S0027763000019498.

UCLA Mathematics Department, Box 951555, Los Angeles, CA 90095-1555, USA

Email address: jmoraga@math.ucla.edu