Dolgachev Surfaces: Fibrations & Categories

• Dolgachev surfaces are minimal elliptic surfaces over P¹ defined by p₉ = q = 0, κ = 1, and exactly two multiple fibers with coprime multiplicities.
• They are constructed via classical logarithmic transforms on rational elliptic surfaces or through Q-Gorenstein smoothings of singular models.
• Their derived categories exhibit exceptional collections, quiver embeddings, and phantom subcategories that reveal subtle categorical phenomena.

Dolgachev surfaces are minimal elliptic surfaces over $\PP^1$ with vanishing irregularity and geometric genus, equipped with exactly two multiple fibers of relatively prime multiplicities. In the formulations used in the recent literature, a Dolgachev surface of type $(p,q)$ or $D_{p,q}$ is a smooth projective surface $S$ or $Y$ with $p_g=q=0$, Kodaira dimension $\kappa=1$, and an elliptic fibration whose only multiple fibers have multiplicities $p$ and $q$; when $\gcd(p,q)=1$, the surface is simply connected (Cho et al., 2015, Canedo et al., 7 May 2026, Lee et al., 9 Sep 2025). The subject now spans several tightly linked themes: classical logarithmic transformations of rational elliptic surfaces, $(p,q)$0-Gorenstein smoothings of singular rational surfaces, classifications of rational homology disk degenerations, compactified moduli of elliptic surfaces with multiple fibers, and derived-categorical phenomena such as exceptional collections, quiver embeddings, and phantom subcategories (Cho et al., 2015, Tevelev et al., 2022, Canedo et al., 7 May 2026, Lee et al., 9 Sep 2025, Karzhemanov et al., 2023).

1. Definition and core invariants

A Dolgachev surface of type $(p,q)$1 is described as a simply-connected, minimal, nonsingular projective surface with

$(p,q)$2

Kodaira dimension $(p,q)$3, and an elliptic fibration $(p,q)$4 whose only multiple fibers have multiplicities $(p,q)$5 and $(p,q)$6; the multiplicities are assumed coprime, with $(p,q)$7 in one formulation and $(p,q)$8 in another (Cho et al., 2015, Tevelev et al., 2022). A parallel notation writes the two multiple fibers as

$(p,q)$9

for simple smooth curves $D_{p,q}$0 and $D_{p,q}$1 (Tevelev et al., 2022).

Several numerical formulas recur across the literature. For a general fiber $D_{p,q}$2 of the elliptic fibration, the canonical bundle formula is written

$D_{p,q}$3

in the sense of $D_{p,q}$4-divisors, equivalently

$D_{p,q}$5

(Cho et al., 2015). In a different notation, the canonical divisor is recorded as

$D_{p,q}$6

(Tevelev et al., 2022), and also as

$D_{p,q}$7

(Lee et al., 9 Sep 2025). The same sources also record

$D_{p,q}$8

and, by the Noether formula,

$D_{p,q}$9

so that $S$0 when $S$1 (Canedo et al., 7 May 2026, Lee et al., 9 Sep 2025).

The topological behavior is controlled by the multiplicities. One source states that $S$2 if $S$3, and more generally that $S$4 is cyclic of order $S$5 (Canedo et al., 7 May 2026). The same paper notes that $S$6 if $S$7, while the case $S$8 yields Enriques surfaces (Canedo et al., 7 May 2026). A section exists exactly when one multiplicity is $S$9, in which case the surface is a Halphen surface of index $Y$0 (Canedo et al., 7 May 2026). For the generic Dolgachev surface of type $Y$1, the Picard number is recorded as $Y$2 (Lee et al., 9 Sep 2025).

2. Constructions from rational elliptic surfaces and singular models

The classical construction starts from a rational elliptic surface obtained by blowing up the nine base points of a pencil of plane cubics. If

$Y$3

is such a pencil, then the blow-up $Y$4 at its nine base points carries an elliptic fibration whose general fiber $Y$5 is a smooth genus-one curve. Performing logarithmic transforms of orders $Y$6 and $Y$7 at two distinct smooth fibers produces a simply-connected minimal elliptic surface with two multiple fibers of multiplicities $Y$8 and $Y$9, recovering the canonical bundle formula and the invariants $p_g=q=0$0 and $p_g=q=0$1 (Cho et al., 2015). A related formulation chooses two points $p_g=q=0$2 and torsion line bundles

$p_g=q=0$3

of exact orders $p_g=q=0$4 and $p_g=q=0$5; the resulting elliptic surface $p_g=q=0$6 is isomorphic to the original rational Jacobian surface away from $p_g=q=0$7 and $p_g=q=0$8, but acquires multiple fibers of orders $p_g=q=0$9 and $\kappa=1$0 there (Lee et al., 9 Sep 2025).

A second construction replaces logarithmic transforms by $\kappa=1$1-Gorenstein smoothing of a singular rational surface. Starting again from a rational elliptic surface $\kappa=1$2, one contracts two chains of curves to obtain a normal projective surface

$\kappa=1$3

with two Wahl $\kappa=1$4 singularities, of types

$\kappa=1$5

with $\kappa=1$6 (Cho et al., 2015). In the Picard–Noether group one has

$\kappa=1$7

so $\kappa=1$8 is nef, $\kappa=1$9, and $p$0 for $p$1 (Cho et al., 2015). Wahl’s theory and Manetti’s vanishing $p$2 then yield a one-parameter $p$3-Gorenstein smoothing

$p$4

whose general fiber is a minimal simply-connected elliptic surface with $p$5, $p$6, and two multiple fibers of orders $p$7 and $p$8 (Cho et al., 2015).

The deformation-theoretic analysis also gives a detailed model of the central fiber after base change and weighted blow-up. After a base change of order $p$9, one resolves the family near the singular points and obtains a central fiber

$q$0

where

$q$1

and the components meet $q$2 along smooth rational curves $q$3 (Cho et al., 2015). This reducible model is the starting point for the later construction of exceptional bundles by gluing.

Construction	Starting data	Output
Logarithmic transforms	Rational elliptic surface from the blow-up of nine base points of a cubic pencil	Minimal elliptic surface with multiple fibers of orders $q$4
$q$5-Gorenstein smoothing	Singular rational surface with two Wahl singularities	General fiber $q$6 of Dolgachev type
Weighted blow-up central fiber	Base change and local resolution of the smoothing	$q$7 with $q$8 and $q$9

Taken together, these constructions show that Dolgachev surfaces can be accessed either through explicit elliptic-surface surgery or through the deformation theory of singular surfaces. The latter viewpoint is the one used most heavily in the derived-categorical papers (Cho et al., 2015, Tevelev et al., 2022).

3. Singular fibers, rational homology disk degenerations, and compactified moduli

Beyond the two multiple fibers, the remaining fibers are constrained by the elliptic fibration. One source states that all non-multiple fibers are irreducible of Kodaira type $\gcd(p,q)=1$0 $\gcd(p,q)=1$1, $\gcd(p,q)=1$2, $\gcd(p,q)=1$3, or $\gcd(p,q)=1$4 (Canedo et al., 7 May 2026). In the compactification framework for elliptic surfaces with a multiple fiber, the remaining singular fibers may be any reduced Kodaira fibers of types

$\gcd(p,q)=1$5

subject to the Euler–Kimura constraint that the total contribution to $\gcd(p,q)=1$6 is $\gcd(p,q)=1$7 (Lee et al., 9 Sep 2025).

A recent classification concerns $\gcd(p,q)=1$8HD degenerations of Dolgachev surfaces. For a $\gcd(p,q)=1$9 smoothing whose general fiber belongs to $(p,q)$00 with $(p,q)$01, the central fiber is a normal surface with one non–Du Val weighted-homogeneous $(p,q)$02 singularity and one Wahl singularity of type

$(p,q)$03

(Canedo et al., 7 May 2026). More precisely, the central fiber carries a minimal elliptic fibration with two multiple fibers of multiplicities $(p,q)$04 and $(p,q)$05, one supporting the non-log-canonical $(p,q)$06 singularity and the other the Wahl singularity. Up to “sliding,” the non-log-canonical singularities arise from $(p,q)$07-modifications of an elliptic fiber of type $(p,q)$08, $(p,q)$09, or $(p,q)$10 (Canedo et al., 7 May 2026).

The weighted-homogeneous singularity is encoded by a star-shaped resolution graph with central curve $(p,q)$11 of self-intersection $(p,q)$12 and Hirzebruch–Jung chains on the legs. Its Seifert invariants are written

$(p,q)$13

with $(p,q)$14, and the discrepancy of the central curve is

$(p,q)$15

The condition $(p,q)$16 forces one non-log-canonical leg of length at least $(p,q)$17 (Canedo et al., 7 May 2026). Every one-point $(p,q)$18HD degeneration of $(p,q)$19 appears in this classification (Canedo et al., 7 May 2026).

The same paper constructs minimal semi-log-canonical birational models of these degenerations. The procedure consists of a Seifert partial resolution extracting only the central curve

$(p,q)$20

followed by a semistable flip in a log-terminal threefold. The result is a central fiber

$(p,q)$21

glued along $(p,q)$22, with one new Wahl singularity on each component and with $(p,q)$23 nef (Canedo et al., 7 May 2026). Unobstructedness is proved by vanishing of the local-global obstructions, ultimately using $(p,q)$24 and vanishing statements on the components of the slc limit (Canedo et al., 7 May 2026).

From the moduli viewpoint, Dolgachev surfaces of fixed type $(p,q)$25 with $(p,q)$26 fit into an irreducible $(p,q)$27-dimensional parameter space $(p,q)$28 of quintuples

$(p,q)$29

where $(p,q)$30 is a rational elliptic surface of index $(p,q)$31, $(p,q)$32 are points over which the fibers are smooth or of type $(p,q)$33, and $(p,q)$34 are torsion line bundles of orders $(p,q)$35 and $(p,q)$36 (Lee et al., 9 Sep 2025). A stable-pair compactification is obtained by choosing a marking divisor $(p,q)$37 and the boundary

$(p,q)$38

and then invoking Birkar’s theory to construct a projective coarse moduli space

$(p,q)$39

(Lee et al., 9 Sep 2025). In this framework, when a multiple fiber degenerates to an additive Kodaira fiber, boundary components are described by semi-log-canonical limits.

Kawamata’s classification of moderate degenerations imposes further restrictions on permissible normal central fibers with $(p,q)$40. The only fiber types on such a central fiber are

$(p,q)$41

and no

$(p,q)$42

occur on the central fiber of a permissible normal degeneration (Lee et al., 9 Sep 2025). The invariants remain constant under degeneration: $(p,q)$43 and the log-canonical thresholds governing wall crossing are computed for the additive types (Lee et al., 9 Sep 2025).

4. Picard groups and explicit models of type $(p,q)$44

The case of the $(p,q)$45-Dolgachev surface is the setting in which the Picard group and derived category have been worked out most explicitly. For the $(p,q)$46-Gorenstein smoothing model with singularities of types

$(p,q)$47

certain divisors on the resolution $(p,q)$48 satisfying parity and divisibility conditions can be lifted to Cartier divisors on the smooth fiber $(p,q)$49 by Hacking’s gluing method (Cho et al., 2015). This produces ten divisors

$(p,q)$50

defined from lifts of $(p,q)$51 and of classes $(p,q)$52. Under the explicit definitions, the intersection matrix

$(p,q)$53

is

$(p,q)$54

and hence

$(p,q)$55

with basis $(p,q)$56 (Cho et al., 2015).

A complementary explicit model of a special $(p,q)$57-surface begins with $(p,q)$58 with coordinates $(p,q)$59, two lines $(p,q)$60 and $(p,q)$61, and a singular quartic chosen so that after blowing up nine suitably infinitely near points one obtains a Halphen pencil of index $(p,q)$62 (Karzhemanov et al., 2023). The resulting elliptic surface $(p,q)$63 has one fiber of type $(p,q)$64, three further nodal $(p,q)$65 fibers, and one multiple fiber $(p,q)$66 (Karzhemanov et al., 2023). On $(p,q)$67 one finds a divisor

$(p,q)$68

that is divisible by $(p,q)$69 in $(p,q)$70; taking the $(p,q)$71 cover branched along $(p,q)$72, then contracting and resolving, gives a smooth surface $(p,q)$73 with elliptic fibration $(p,q)$74 whose singular fibers are exactly four fibers of type $(p,q)$75, one multiple fiber $(p,q)$76, and one multiple fiber $(p,q)$77 (Karzhemanov et al., 2023).

For this explicit surface one has

$(p,q)$78

(Karzhemanov et al., 2023). The same paper exhibits ten divisor classes, denoted $(p,q)$79 together with $(p,q)$80, computes their full intersection matrix, and obtains a unimodular lattice of signature $(p,q)$81 (Karzhemanov et al., 2023). This explicit lattice-theoretic control is what makes a twelve-term line-bundle exceptional collection possible in that example.

5. Exceptional bundles, exceptional collections, and quiver realizations

The smoothing picture provides a natural source of exceptional bundles. On the weighted-projective component

$(p,q)$82

one constructs an exceptional rank-$(p,q)$83 bundle $(p,q)$84, and on

$(p,q)$85

one takes the rank-$(p,q)$86 tangent twist

$(p,q)$87

with restrictions

$(p,q)$88

(Cho et al., 2015). Gluing these to copies of $(p,q)$89 yields exceptional bundles on the reducible central fiber, and these deform uniquely to exceptional bundles on the smooth Dolgachev surface (Cho et al., 2015).

In the $(p,q)$90 case, setting $(p,q)$91 and $(p,q)$92, one obtains an exceptional sequence

$(p,q)$93

of length $(p,q)$94 in $(p,q)$95 (Cho et al., 2015). Semiorthogonality and exceptionality are proved by explicit cohomology vanishings, using upper semicontinuity, Riemann–Roch on $(p,q)$96, and degeneration to $(p,q)$97 and the components $(p,q)$98 (Cho et al., 2015). In the special $(p,q)$99 example constructed via the triple cover, the paper (Karzhemanov et al., 2023) gives an explicit line-bundle realization

$D_{p,q}$00

where the $D_{p,q}$01 are written as explicit $D_{p,q}$02-linear combinations of the lattice generators; Theorem 2.6 there states that these twelve line bundles form an exceptional collection of maximal possible length $D_{p,q}$03.

A different but related construction applies to all Dolgachev surfaces $D_{p,q}$04. The paper (Tevelev et al., 2022) proves the existence of a strong exceptional collection of length $D_{p,q}$05,

$D_{p,q}$06

where each $D_{p,q}$07 is a Hacking–Pak bundle obtained by extending the Kawamata bundle on the $D_{p,q}$08-resolution $D_{p,q}$09 to the smoothing. The Chern characters are computed as

$D_{p,q}$10

with

$D_{p,q}$11

and

$D_{p,q}$12

(Tevelev et al., 2022). The same paper establishes

$D_{p,q}$13

and for the unique nonzero $D_{p,q}$14-block,

$D_{p,q}$15

with all other $D_{p,q}$16 vanishing (Tevelev et al., 2022).

The endomorphism algebra of $D_{p,q}$17 is the path algebra of a star-shaped quiver $D_{p,q}$18 with ten vertices and

$D_{p,q}$19

arrows in the distinguished block (Tevelev et al., 2022). This yields an admissible embedding

$D_{p,q}$20

sending the simple modules to the exceptional bundles $D_{p,q}$21 (Tevelev et al., 2022). The same work shows that mutations of exceptional collections correspond exactly to antiflips in the passage from $D_{p,q}$22-resolutions to $D_{p,q}$23-resolutions.

6. Non-fullness, phantom categories, and categorical limits

The most striking categorical feature is that maximal or numerically maximal exceptional collections on Dolgachev surfaces need not be full. For the twelve-term line-bundle collection on the $D_{p,q}$24 surface arising from $D_{p,q}$25-Gorenstein smoothing, if

$D_{p,q}$26

then

$D_{p,q}$27

and the pseudoheight of the collection is at least $D_{p,q}$28; by Kuznetsov’s height formalism, $D_{p,q}$29, so $D_{p,q}$30 is a nontrivial phantom category (Cho et al., 2015).

The explicit construction in (Karzhemanov et al., 2023) gives a different proof of non-fullness for a special $D_{p,q}$31-surface. Let $D_{p,q}$32 be the Jacobian elliptic surface; the paper identifies $D_{p,q}$33 with the classical Hesse pencil rational elliptic surface, with Mordell–Weil group $D_{p,q}$34 and four $D_{p,q}$35 fibers (Karzhemanov et al., 2023). The Galois group $D_{p,q}$36 of the $D_{p,q}$37 cover acts on $D_{p,q}$38 by translation by a $D_{p,q}$39-torsion section $D_{p,q}$40, and restriction of the exceptional objects to the generic fiber $D_{p,q}$41 shows that they generate $D_{p,q}$42 (Karzhemanov et al., 2023). The $D_{p,q}$43-action exchanges a putative semiorthogonal decomposition

$D_{p,q}$44

contradicting the theorem of Kawatani–Okawa that no nontrivial semiorthogonal decomposition exists for the derived category of a smooth elliptic curve. The conclusion is again that the right orthogonal is a nonzero phantom with trivial $D_{p,q}$45 and Hochschild homology (Karzhemanov et al., 2023).

This phenomenon extends beyond a single special surface. The same paper states that any other $D_{p,q}$46-Dolgachev surface obtained by analytic logarithmic transforms from the same Jacobian surface inherits an exceptional collection of the same length, and the same $D_{p,q}$47 argument produces a phantom in its derived category (Karzhemanov et al., 2023). In parallel, (Tevelev et al., 2022) identifies for general $D_{p,q}$48 a rank-$D_{p,q}$49 Mukai lattice as the orthogonal complement to the length-$D_{p,q}$50 strong exceptional collection and states that a full exceptional collection on $D_{p,q}$51 is impossible unless $D_{p,q}$52, equivalently $D_{p,q}$53.

Within the literature summarized here, Dolgachev surfaces therefore serve as a meeting point of elliptic surface theory, singularity smoothing, and categorical birational geometry. The explicit exceptional collections on $D_{p,q}$54, the strong length-$D_{p,q}$55 collections on all $D_{p,q}$56, the quiver realizations, and the appearance of phantoms together show that the derived categories of Kodaira-dimension-$D_{p,q}$57 surfaces with multiple fibers are substantially subtler than their numerical invariants alone would suggest (Cho et al., 2015, Tevelev et al., 2022, Karzhemanov et al., 2023).