2011 09628

SMOOTH PROJECTIVE CALABI-YAU COMPLETE INTERSECTIONS
AND ALGORITHMS FOR THEIR FROBENIUS MANIFOLDS AND

HIGHER RESIDUE PAIRINGS
YOUNGGI LEE, JEEHOON PARK, AND JAEHYUN YIM

arXiv:2011.09628v1 [math.AG] 19 Nov 2020
Abstract. The goal of this article is to provide an explicit algorithmic construc-

tion of formal F -manifold structures, formal Frobenius manifold structures, and higher
residue pairings on the primitive middle-dimensional cohomology H of a smooth pro-
jective Calabi-Yau complete intersection variety X defined by homogeneous polyno-
mials G1 (x), . . . , Gk (x). Our main method is to analyze a certain dGBV (differential
Gerstenhaber-Batalin-Vilkovisky) algebra A obtained from the twisted de Rham com-
plex which computes H. More explicitly, we introduce a notion of a weak primitive
form associated to a solution of the Maurer-Cartan equation of A and the Gauss-Manin
connection, which is a weakened version of Saito’s primitive form ([19]). In addition,
we provide an explicit algorithm for a weak primitive form based on the Gröbner ba-
sis in order to achieve our goal. Our approach through the weak primitive form can
be viewed as a unifying link (based on Witten’s gauged linear sigma model, [24]) be-
tween the Barannikov-Kontsevich’s approach to Frobenius manifolds via dGBV algebras
(non-linear topological sigma model, [1]) and Saito’s approach to Frobenius manifolds
via primitive forms and higher residue pairings (Landau-Ginzburg model, [22]).
Contents
1. Introduction 2
1.1. Acknowledgement 6
2. The dGBV Algebra and Quantization 6
2.1. Cochain complex A 6
2.2. Gradings on A 7
2.3. The de Rham complex and gradings 8
2.4. Main theorem on cohomologies 8
2.5. The dGBV algebra associated to X 9
2.6. The 0-th cohomology of (A, KS ) and the Jacobian ideal of S 10
2.7. A quantization dGBV algebra 12
2.8. Deformation of dGBV algebra and the Gauss-Manin connection 13
3. An algorithm for formal F -manifolds 15
3.1. Definition of formal F -manifolds 15
3.2. A connection matrix for the Gauss-Manin connection 16
3.3. An explicit algorithm 18
3.4. A proof why the algorithm works 20
4. Weak primitive forms 22
4.1. Weak primitive forms and formal F -manifolds 22
4.2. An explicit algorithm 23
4.3. A proof why the algorithm works 25
5. An algorithm for formal Frobenius manifolds and higher residue pairings 27
5.1. Definition of formal Frobenius manifolds 27
2020 Mathematics Subject Classification. 14J32 (primary), 81T70, 53D45, 14J33, 14J81, 14F25
(secondary).
Key words and phrases. dGBV algebras, F -manifolds, Frobenius manifolds, weak primitive forms, higher
residue pairings, projective smooth complete intersections.
1
ALGORITHMS FOR FROBENIUS MANIFOLDS AND HIGHER RESIDUE PAIRINGS 2
5.2. Formal Frobenius manifold structures on H 28

5.3. Modified higher residue pairings 29
6. Appendix 33
6.1. Comparisons 33
6.2. The Riemann-Hilbert-Birkhoff problem 34
6.3. Comments on the algorithm 36
References 38
1. Introduction
Let M be a complex manifold. A Frobenius manifold structure on M is a commutative
algebra structure on the holomorphic tangent bundle TM with a metric satisfying certain
compatibility conditions: see Definition 5.1. Such a structure was first axiomatized by
Dubrovin in [5] and its first example was given by K. Saito (the universal unfolding of
an isolated hypersurface singularity) in [19] and [21]. It also plays an important role in
formulating the mirror symmetry conjecture: for example, see [1] and [3]. We refer to [15],
[14], and [8] for various constructions of Frobenius manifolds. There is a slight weakening:
an F -manifold, which appears in [9] and [13], is essentially a Frobenius manifold structure
without a metric. We refer to Definition 3.1 and Proposition 3.2 for its precise definition.
One can also think of formal Frobenius manifold structures and formal F -manifold struc-
tures. In this paper, we will provide a new construction of formal Frobenius manifolds and
formal F -manifolds on primitive middle-dimensional cohomology space H of a smooth pro-
jective Calabi-Yau complete intersection. Our construction is explicit and algorithmic so
that it can be implemented in a computer program.
On the B-side of the mirror symmetry conjecture, there are two well-known constructions
of Frobenius manifolds: the work [1] of Barannikov-Kontsevich and the work [19], [21] of K.
Saito and M. Saito. Barannikov-Kontsevich used the algebra P of holomorphic polyvector
fields on a compact Calabi-Yau manifold X with valued in anti-holomorphic forms in or-
der to construct Frobenius manifold structures on the cohomology group H • (X, C). They
equipped P with structure of the dGBV (differential Gerstenhaber-Batalin-Vilkovisky) alge-
bra, whose Maurer-Cartan functor governs generalized deformations of complex structures
on X. Then they proved the formality of the Maurer-Cartan functor attached to P, which
enables them to construct Frobenius manifold structures on H • (X, C) with the help of
¯
∂ ∂-lemma, and certain period integrals on X.
Saito’s theory deals with the universal unfolding of an isolated singularity of a holomor-
phic function; see [22] for a nice survey. When a holomorphic function f , defined on a
neighborhood of the origin 0 in Cn+1 with f (0) = 0, has an isolated critical point at 0,
then, under certain assumptions, Saito and his collaborators constructed Frobenius man-
ifold structures on a small open subset of OCn+1 ,0 /(∂f /∂x0 , . . . , ∂f /∂xn ) ≃ Cµ where µ
is an integer and OCn+1 ,0 is the germ of holomorphic functions on Cn+1 around 0. Their
main tools are higher residue pairings K (Theorem 5.1, [22]) and primitive forms associated
to K (Definition 6.1, [22]), which are realized on a certain quantization twisted de Rham
complex (section 4.1, [22]) involving d + df .
Our key idea to construct Frobenius manifolds and F -manifolds lies in the unification
based on Witten’s gauged linear sigma model (see section 5.4, [24] for detailed physical
explanations) of the above two different constructions. If the projective smooth Calabi-Yau
complete intersection variety XG ⊆ Pn is defined by homogeneous polynomial equations
G1 (x), . . . , Gk (x), where x = (x0 , . . . , xn ) is the homogeneous coordinate of the projective
Pk
n-space Pn , we will use a function S(y, x) := i=1 yi Gi (x) with a new set of variables
y1 , . . . , yk . This function S(y, x) will play a similar role as the defining equation of a
hypersurface of Saito’s theory. But S(y, x) has a non-isolated singularity locus differently
from Saito. In fact, the weight zero part of the singularity locus of S(y, x) will be exactly the
smooth projective complete intersection XG . As far as the authors know, the superpotential
S(y, x) was first introduced by B. Dwork in his proof of rationality of zeta functions of
algebraic varieties over a finite field; the auxiliary variables y1 , . . . , yk were introduced to
count the number of rational points of algebraic varieties over finite fields by using the
exponential sum (see (16), [6]).1 But later E. Witten explained its physical meaning by
looking at phase transitions which occur as parameters are varied in supersymmetric gauge
theories, section 5.4, [24].2
n−k
The two facts [(1) the primitive middle-dimensional cohomology H = Hpirm (X, C) is the
cohomology of the twisted de Rham complex involving d+dS (see Theorem 2.10 and Remark
2.11) and (2) we can modify the multiplication structure of the twisted de Rham complex
to obtain the dGBV structure (see Proposition 2.5)] link our study of formal Frobenius
manifolds and higher residue pairings on H to those in [1], [22]. In order to deal with
non-isolated singularities of S(y, x), we introduce the charge grading ch (see subsection 2.2)
on the dGBV algebra; for more detailed comparisons with [1], [22], we refer to subsection
6.1. Our key tools are dGBV algebras similar to [1] and weak primitive forms (Definition
1.1)3 which are weakened version of primitive forms in [22]. More explicitly, our procedure
can be summarized as follows:
(1) Construct a dGBV algebra which computes the primitive middle-dimensional coho-
mology H of XG . (This construction is motivated by the (0 + 0)-dimensional QFT
(Quantum Field Theory) viewpoint for the cohomology of XG , which was explained
in the introduction of [17] and Witten’s work [24].)
(2) Consider the Maurer-Cartan deformation problem associated to the shifted DGLA
(differential graded Lie algebra) coming from the dGBV algebra. In particular, find
a solution Γ of the corresponding Maurer-Cartan equation.
(3) Consider the Gauss-Manin connection associated to the solution Γ and provide an
algorithm to compute the connection matrix with respect to some particular basis
(the concept of weak primitive forms).
(4) If we choose Γ and a particular basis properly, then the connection matrix will give
rise to an associative and commutative structure on the space TH of the formal
tangent vector fields (a formal F -manifold structure on H) and a metric (a formal
Frobenius manifold structure on H).
The theory of primitive forms and higher residue pairings, which was developed by K.
Saito, provides a Frobenius manifold structure (an F-manifold structure and the metric),
by taking flat coordinates with respect to a metric and takes ~ → 0 (so called, a classical
limit). One may think of a concept which provides an F -manifold in a similar manner;
we introduce a notion of “weak primitive forms” (see Definition 1.1), which provides an
F -manifold structure when one takes flat coordinates, and ~ → 0. Note that the higher
residue pairing with ~ → 0 essentially corresponds to the metric. We summarize this in
Table 1.
For any dGBV algebra and its Maurer-Cartan solution, we now define the notion of
weak primitive forms associated to them. The adjective “weak” means that we drop some
conditions from the definition of primitive form; we take (S2) of Proposition 7.3, [22], which
is implied by (P1), (P2), and (P4) of Definition 6.1 (the original definition of primitive form),
[22], as the main ingredient of our definition. Let (A, ·, ∆, ℓ∆2 (·, ·), Q) be a dGBV algebra
(Definition 2.4), and {t = tα } be a coordinate of the tangent space for the Maurer-Cartan
deformation functor associated to the shifted DGLA (A, ∆ + Q, ℓ∆+Q 2 (·, ·)). Let Γ ∈ A[[t]]
1If one is familiar with the Griffiths-Dwork method (for calculating the Picard-Fuchs ideal), section 5.3,
[3], our use of the function S can also be viewed as a generalization of the G-D method to the algorithmic
construction of Frobenius manifolds.
2
The introduction of S(y, x) is also known as “the Cayley trick” to mathematicians; see subsection 2.4.
3In the table 2, we summarize how this simple notion of weak primitive forms unifies the approaches of
Saito-Takahashi [22], Barannikov-Kontsevich [1], and ours. There is also a categorical unification of both
approaches, as explained in the appendix of [22] but our unification is more explicit and down-to-earth.
Table 1. Weak primitive forms
Classical Complex (~ → 0) Quantization complex (~-perturbation)
Primitive forms and

Frobenius manifolds Take
higher residue pairings
flat coordinates
←−−−−−−−−−−−
F -manifolds ~→0 Weak primitive forms
be a Maurer-Cartan solution of (A, ∆ + Q, ℓ∆+Q 2 (·, ·)):

1
(∆ + Q)(Γ) + ℓ∆+Q (Γ, Γ) = 0,
2 2
where ∆ + Q and ℓ∆+Q 2 (·, ·) are assumed to be t-linear. Then quintuple (A[[t]], ·, ∆ + Q +
ℓ∆
2 (Γ, ·), ℓ ∆
2 (·, ·), Q + ℓ ∆
2 (Γ, ·)) is again a dGBV algebra.
One can consider the following quantization dGBV algebra:

A[[t]]((~)), ·, ~∆ + Q + ℓ∆ 1 ~∆ ∆
2 (Γ, ·), ~ ℓ2 (·, ·), Q + ℓ2 (Γ, ·)
~∆+Q+ℓ∆ (Γ,·)
with a formal parameter ~. Note that ~1 ℓ~∆ ∆ 1
2 (·, ·) = ℓ2 (·, ·) = ~ ℓ2
2
(·, ·). We define
the Gauss-Manin connection as follows (see Definition 2.20 for more details):
Γ Γ ∂ Γ
~ α~ := ~·e− ~ α e ~ : H r A[[t]]((~)), ~∆+Q+ℓ∆ r ∆
2 (Γ, ·) → H A[[t]]((~)), ~∆+Q+ℓ2 (Γ, ·) .
∂t
Definition 1.1 (weak primitive forms). An element ζ ∈ A0 [[t]][[~]] is called a weak primitive
form if
Γ
(1) the set {~ α~ ζ} is a C[[t]][[~]]-basis of H 0 A[[t]][[~]], ~∆ + Q + ℓ∆2 (Γ, ·) . In particular,
there are unique Aαβ ρ ∈ C[[t]][[~]] such that
Γ Γ X Γ
(1.1) ~ β~ ~ α~ ζ = Aαβ ρ · ~ ρ~ ζ + ~∆ + Q + ℓ∆ 2 (Γ, ·) (Λαβ ),
ρ
(2) the coefficient Aαβ is of the form Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~.

ρ
From the definition, we have

(Commutativity) Aαβ ρ = Aβα ρ .
Moreover, if the coefficient 1 Aαβ ρ = 0, then the following conditions hold:
X X
0 0
(Associativity) Aαβ ρ0 Aργ δ = Aβγ ρ0 Aρα δ ,
ρ ρ
∂ 0 ∂
(Potential) Aγβ ρ = γ 0 Aαβ ρ .
∂tα ∂t
The reason we put the restrictive condition (2) Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~ in the above
definition is its relevance to formal F -manifolds; see subsection 4.1 how weak primitive
forms provides formal F -manifolds. Moreover, if one compares (2) with equality (S2) in
Proposition 7.3, [22], then one can check that a weak primitive form equipped with a
compatible higher residue pairing becomes a primitive form in the sense of Saito-Takahashi
[22].
There are two main results and two auxiliary results in the current article:
(1) An explicit algorithmic construction of formal F -manifolds on H; see section 3.
(2) An explicit algorithmic construction of weak primitive forms related to H; see sec-
tion 4.
(3) An explicit algorithmic construction of Frobenius manifolds on H; see subsection
5.2.
(4) An explicit algorithmic construction of the modified higher residue pairings 4 related
to H; see subsection 5.3.
The reason why we call (3) and (4) auxiliary results is that their constructions are more or
less technically trivial. All these algorithms will be deduced by solving the equation (1.1)
for weak primitive forms. For different choices of Γ, ζ, Aαβ ρ , we will solve the differential
equation (1.1) algorithmically and connect them to the concepts of F-manifolds, Frobenius
manifolds, and modified higher residue pairings. We record such different choices in the
table 2. The equation (1.1) is subtle to solve; see Remarks 3.7 and Appendix 6.3. In fact,
the key technical obstacle for the current project was to solve algorithmically its special
case (3.3), which is also a main differential equation of [1] appeared in the proof of the
lemma 7.1, [1].
We need to mention the work [12] of Li-Li-Saito, the book of Hertling [8], and Sabbah’s
work [18]. There is a similarity between ours and [12] in that both use dGBV algebras (based
on the polyvector fields) to deal with higher residue pairings and to unify the methods
of [1] and [22]. But there is a notable difference: the main set-up of [12] consists of a
Stein domain M ⊂ Cm and a holomorphic function f : M → C with finite critical points
(section 1.2, [12]), which corresponds to the affine space Cn+k+1 itself and the polynomial
function S : Cn+k+1 → C, which does not have finite critical points, in our set-up.5 Thus
we can not apply [12] to study higher residue pairings and primitive forms relevant to
n−k
H = Hprim (X, C). This phenomenon regarding non-isolated singularities limits us to obtain
results only on “weak” primitive forms in our set-up. Our method, a systematic algorithmic
approach to the differential equation (1.1), which naturally appears in the study of primitive
forms, is not refined enough to obtain results on primitive forms. In [12], a good opposite
filtration plays a crucial role to produce analytic primitive forms; the Hodge filtration on
H seems to be related to good opposite filtrations (see subsection 6.2 for some evidence)
but we are not sure how those are related precisely. The book [8] on Frobenius manifolds
concentrates on the case of isolated singularities and [18] deals with tame functions with
isolated singularities, while we work with the case of non-isolated singularities.
Now we explain the contents of the paper briefly. Section 2 is devoted to explain how to
construct the dGBV algebra which computes the primitive middle-dimensional cohomology
H of XG and to study its properties. The materials from subsection 2.1 to subsection 2.7
are not limited to the Calabi-Yau case. But from subsection 2.8 throughout the article,
we will assume that XG is Calabi-Yau. In subsection 2.1, the underlying cochain complex
(A, KS = QS + ∆) of the dGBV algebra is constructed (see (2.2)). In subsection 2.2, the
grading structure of (A, KS ) is explained. In subsection 2.3, the grading structure of the
relevant twisted de Rham complex is explained. Then, in subsection 2.4, we explain how
(A, KS ) compares with the twisted de Rham complex and consequently construct an explicit
isomorphism between its 0-th cohomology and the primitive middle-dimensional cohomol-
ogy (Theorem 2.3) via the Cayley trick. In subsection 2.5, we make the dGBV algebra
(A, ·, KS , ℓKS , QS ) by introducing the ℓ2 -descendant ℓK
2
S
(Definition 2.4 and Proposition
2.5). In subsection 2.6, we compare the cohomologies of (A, KS ) and (A, QS ) and derive the
weak QS ∆-lemma (Lemma 2.8) which plays a crucial role in our algorithm. In subsection
2.7, a formal parameter ~ is introduced to construct the quantization dGBV algebra. In
subsection 2.8, we explain the deformation problem attached to (A, ·, KS , ℓKS , QS ) and the
corresponding Gauss-Manin connection on the deformation space.
Section 3 is devoted to constructing the algorithm for formal F -manifolds on H. In
subsection 3.1, we briefly recall the notion of formal F -manifolds. In subsection 3.2, we
4The modified higher residue pairing is essentially the same as the higher residue pairing but there is a
slight technical difference; see Definition 5.6 and the last paragraph of subsection 6.1 for details.
5The physical background of our theory is [24], which is different from the physical background of [12],
the BCOV (Bershadsky, Cecotti, Ooguri, and Vafa) theory.
explain how the connection matrix for the Gauss-Manin connection leads to the formal
F -manifolds. We provide an algorithm in subsection 3.3 and its proof in subsection 3.4.
Section 4 is devoted
P to constructing the algorithm for weak primitive forms when Γ is
of the form L = α tα uα (linear form). In subsection 4.1, we briefly explain how the weak
primitive form is related to Saito’s primitive form and gives rise to formal F -manifolds. We
provide an algorithm in subsection 4.2 and its proof in subsection 4.3.
Section 5 is about the algorithms for the formal Frobenius manifold and the modified
higher residue pairing on H. In subsection 5.1, we briefly recall the notion of formal Frobe-
nius manifolds. We give an algorithm for Frobenius manifolds in subsection 5.2. In sub-
section 5.3, we explain how to produce an algorithm of the modified higher residue pairing
which is essentially the same as the original higher residue pairing of Theorem 5.1, [22]; we
point out a subtle difference between them and explain what is missing for constructing an
algorithm of the original higher residue pairing.
Finally, in the appendix, we provide comparisons of our work with [1] and [22] (subsection
6.1). Then we give some remark on the Riemann-Hilbert-Birkhoff problem (subsection
6.2) and how to solve the equation 1.1 algorithmically (subsection 6.3), regardless of its
connection to F -manifolds and Frobenius manifolds.
1.1. Acknowledgement. The problem of constructing an algorithm for formal F -manifolds

arose from the co-work ([17]) and conversation of the second named author with Jae-Suk
Park, to whom we would like to thank deeply for sharing his ideas and insights in [16]. Jee-
hoon Park was partially supported by Samsung Science & Technology Foundation (SSTF-
BA1502) and BRL (Basic Research Lab) through the NRF (National Research Foundation)
of South Korea (NRF-2018R1A4A1023590). We appreciate Cheol-hyun Cho for providing
valuable comments on the first draft. Jeehoon Park would also like to thank KIAS (Korea
Institute for Advance Study) for its support and hospitality during his stay on sabbatical
leave, where the part of the paper was written.
2. The dGBV Algebra and Quantization

2.1. Cochain complex A. Let Pn be a n-dimensional projective space over C for n ≥ 1.
Denote by C[x] the usual homogeneous coordinate ring of Pn with x = (x0 , x1 , . . . , xn ).
For n ≥ k ≥ 1, let G1 (x), . . . , Gk (x) be homogeneous polynomials of degree d1 , . . . , dk
respectively. We consider a smooth projective variety XG embedded in Pn defined by
G1 (x), . . . , Gk (x), which satisfies the complete intersection property, i.e. dim XG = n − k.
Let X = XG (C) be the complex analytic manifold associated to XG and consider the
Dwork potential
k
X
(2.1) S(y, x) := yℓ · Gℓ (x),
ℓ=1
where we introduce the formal variables y1 , . . . , yk corresponding to G1 , . . . , Gk . Let N =

n + k + 1, q1 = y1 , . . . , qk = yk , qk+1 = x0 , . . . , qN = xn and define the Z-graded super-
commutative algebra A and differentials ∆, QS and KS as follows:
A := C[q][η] = C[q1 , q2 , . . . , qN ][η1 , η2 , . . . , ηN ],
XN
∂ ∂
∆ := : A → A,
i=1
∂qi ∂ηi
(2.2)
XN
∂S(q) ∂
QS := : A → A,
i=1
∂qi ∂ηi
KS := QS + ∆.
where ηi ’s are other variables (of cohomological degree −1) corresponding to qi ’s. The
additive cohomological Z-grading of A is given by the rules (|f | = m means f ∈ Am )
|qi | = 0, |ηi | = −1, i = 1, . . . , N.
6
Then we have the following cochain complex
K K K
0 → A−N −→
S
A−N +1 −→
S S
· · · −→ A0 → 0.
Let A := A0 = C[q1 , . . . , qN ]. We define the ℓ2 -descendant of KS with respect to the
product · as follows:
|a|
ℓK
2 (a, b) := KS (ab) − KS (a)b − (−1) aKS (b),
S
a, b ∈ A.
2.2. Gradings on A. We are going to introduce another two Z-gradings on A defined in

(2.2): charge and weight. For each qµ , assign a non-zero integer ch(qµ ) called charge of qµ
as follows:
ch(qi ) = ch(yi ) = −di , for i = 1, . . . , k, ch(qi ) = ch(xi−k−1 ) = 1, for i = k + 1, . . . , N.
Also assign ch(ηµ ) := −ch(qµ ). Define the background charge cX by
X k
X
cX := − ch(qµ ) = di − (n + 1).
µ i=1
Associated with ch(qµ )µ=1,...,N , we define the Euler vector

N
X X N
∂ ∂
Ech = ch(qµ )qµ + ch(ηµ )ηµ .
µ=1
∂q µ µ=1
∂η µ
For any monomial M ∈ A, we have Ech (M ) = ch(M )M . Hence we have the charge
decomposition of A:
M
A= Aλ ,
λ∈Z
where Aλ is the C-subspace of A generated by the monomials of charge λ. We also introduce
the notion of weight such that wt(yℓ ) = 1 for all ℓ = 1, . . . , k and wt(xi ) = 0 for i =
0, 1, . . . , n. Define wt(ηµ ) = 1 − wt(qµ ) for all µ = 1, . . . , N . Associated with the weight, we
define the Euler vector
N
X N
X k
X N
X
∂ ∂ ∂ ∂
Ewt = wt(qµ )qµ + wt(ηµ )ηµ = yℓ + ηℓ .
µ=1
∂qµ µ=1 ∂ηµ ∂yℓ ∂ηℓ
ℓ=1 ℓ=k+1
For any monomial M ∈ A, we have Ewt (M ) = wt(M )M . Hence we have the weight
decomposition of A:
M
A= A(w) ,
w≥0
where A(w) is the C-subspace of A generated by the monomials of weight w. Using the
•
weight, we consider the increasing filtration Fwt A defined by
j
M
(2.3) Fwt A= A(w) .
0≤w≤j
•
Proposition 2.1. The weight filtration induces a filtered cochain complex (Fwt A, KS ).
Proof. Note that KS = QS + ∆. Since QS preserves wt and ∆ decreases wt by 1, the result

follows.
6Note that the super-commutativity means that a · b = (−1)|a||b| b · a for homogeneous elements a, b.
Hence we see that ηi · ηj = −ηj · ηi , which implies that ηi2 = 0 and A−N−1 = A−N−2 = · · · = 0.
Note that A = A[η], where ηµ ην = −ην ηµ , and

|qµ | + |ηµ | = −1, ch(qµ ) + ch(ηµ ) = 0, wt(qµ ) + wt(ηµ ) = 1.
We then have decomposition
M MM
A= Ajλ,(w) ,
−N ≤j≤0 λ∈Z w≥0
where u ∈ Ajλ,(w)
if and only if Ech (u) = λ · u, Ewt (u) = w · u, and |u| = j. Note that
0

S(q) ∈ A0,(1) the Dwork potential S(q) was defined in (2.1) .
2.3. The de Rham complex and gradings. Now we consider an algebraic twisted de
Rham complex. For any commutative C-algebra C, the de Rham complex (Ω•C , d) is defined
and for any element S ∈ C, the twisted de Rham complex (Ω•C , d + dS∧) is defined. In
particular, we will consider the following de Rham complexes:
(Ω•A , dS ∧ +d), (Ω•A[S −1 ] , d), (Ω•A[S −1 ]0,(0) , d).
We also define the charge and the weight on the de Rham complex by
(
ch(qi ) = ch(dqi ), i = 1, 2, . . . , N
wt(qi ) = wt(dqi ). i = 1, 2, . . . , N
We have the weight and the charge decomposition of Ω such that
M MM j
Ω= Ωλ,(w) .
−N ≤j≤0 λ∈Z w≥0
2.4. Main theorem on cohomologies. We will find an explicit relationship between the
n−k
cochain complex (A, KS ) and the primitive middle dimensional cohomology Hprim (XG , C).
If we are interested in the cohomology of the smooth projective complete intersection va-
n−k
riety X of dimension n − k, then the primitive middle dimensional cohomology Hprim (X, C)
is the most interesting piece because the other degree cohomologies and non-primitive pieces
can be easily described in terms of the cohomology of the projective space Pn due to the
n−k
weak Lefschetz theorem and Poincare duality. For the computation of Hprim (X, C), the
Gysin sequence and “the Cayley trick” play important roles. There is a long exact se-
quence, called the Gysin sequence:
ResX
· · · → H n+k−1 (Pn , C) → H n+k−1 (Pn \ X, C) H n−k (X, C) → H n+k (Pn , C) → · · · ,
where ResX is the residue map (see p.96 of [4]). This sequence gives rise to an isomorphism
n−k ∼
ResX : H n+k−1 (Pn \ X, C) → Hprim (X, C).
The Cayley trick is about translating a computation of the cohomology of the complement
of a complete intersection into a computation of the cohomology of the complement of a
hypersurface in a bigger space. Let E := OPn (d1 ) ⊕ · · · ⊕ OPn (dk ) be the locally free sheaf of
OPn -modules with rank k. Let P be the projective bundle associated to E with fiber Pk−1
over Pn . Then P(E) is the smooth projective toric variety whose homogeneous coordinate
ring is given by A. Then S defines a hypersurface XS in P(E). The natural projection map
P(E) → Pn induces a morphism pr1 : P(E) \ XS → Pn \ X which can be checked to be a
homotopy equivalence. One can check that
e1 ek

s(x) = x, G1 (x)e1 −1 G1 (x) , . . . , Gk (x)ek −1 Gk (x) , ei = lcm(d1d,...,d
i
k)
is a well-defined section of pr1 . Hence there exists an isomorphism

∼
s∗ : H n+k−1 (P(E) \ XS , C) → H n+k−1 (Pn \ X, C).
Definition 2.2. We define a sequence of maps as follows:


(i) Let µ : (A•cX , QS + ∆) → (Ω•A )0 [−N ], dS + d be the cochain map defined by
d
µ(q u ηi1 · · · ηiℓ ) = (−1)i1 +···+iℓ +ℓ+1 q u dq1 · · · dq d
i1 · · · dqiℓ · · · dqN
where q u = q1u1 · · · qN
uN
.

•
(ii) Let ρ : (ΩA )0 , dS + d → Ω•A[S −1 ] , d be the cochain map7 defined by
0,(0)
xu y v dy β
ρ(xu yv dxα ∧ dy β ) = (−1)|v|+|β|−1 (|v| + |β| − 1)! dx α ∧ ,
S |v| S |β|
where xu = xu0 0 · · · xunn , yv = y1v1 · · · ykvk , dxα = dxα1 ∧ · · · ∧ dxαi , and dy β =
dyβ1 ∧ · · · ∧ dyβj with |v| = v1 + · · · + vk and |β| = j.
PN
(iii) Let θch be the contraction operator with the vector field i=1 ch(qi )qi ∂q∂ i , and θwt
PN
be the contraction operator with the vector field i=1 wt(qi )qi ∂q∂ i .
Therefore, we have the following maps:

quotient
µ ρ θwt ◦θch −2 −2 −3
A0cX ΩNA 0 Ω N
A[S −1 ] ΩN
B ΩN
B /d(ΩN
B )
0,(0)
and
−2 −3 N −2 ∼ N −2 ∼
ΩN
B /d(ΩN
B ) = HdR (P(E) \ XS ) HdR (Pn \ XG ) n−k
Hprim (XG , C)
s∗ ResG
where B := C[q, S −1 ]0,(0) and P(E) \ XS is a smooth affine variety whose coordinate ring is
n−k
given by B. Let ϕ : A0cX → Hprim (XG , C) be the composition of the above C-linear maps.
Theorem 2.3. The map ϕ is surjective and kerϕ ≃ (QS + ∆)(A−1 cX ). Thus we have an
induced isomorphism ϕ (using the same notation) of finite dimensional C-vector spaces
∼ n−k
A0cX /(QS + ∆)(A−1
cX ) ϕ Hprim (XG , C).
Moreover, ϕ sends the increasing filtration in (2.3) to the increasing Hodge filtration of
n−k
Hprim (X, C). The above explicit description of the map ϕ in terms of series of cochain
maps is our explicit reformulation8 of the results known to experts (see Theorem 1 in [4]
and see [7], [11], and [23]). This reformulation provides us an algebraic BV formalism for
cohomologies and period integrals of smooth projective complete intersection varieties. This
BV formalism will enable us to construct explicit algorithms for formal F -manifolds, weak
primitive forms, formal Frobenius manifolds, and the modified higher residue pairings.
2.5. The dGBV algebra associated to X. We explain that the quintuple (A, ·, KS , ℓK
2 , QS )
S
is a dGBV algebra. We briefly review the definitions of G-algebras, BV-algebras, GBV-

algebras, and dGBV-algebras.
Definition 2.4. Let k be a field. Let (C, ·) be a unital Z-graded super-commutative and
associative k-algebra. Let [·, ·] : C ⊗ C → C be a bilinear map of degree 1.
(a) (C, ·, [·, ·]) is called a G-algebra (Gerstenhaber algebra) over k if
[a, b] = (−1)|a||b| [b, a],
[a, [b, c]] = (−1)|a|+1 [[a, b], c] + (−1)(|a|+1)(|b|+1) [b, [a, c]],
[a, b · c] = [a, b] · c + (−1)(|a|+1)·|b| b · [a, c],
7We found this formula by trial and error. A simple computation confirms that it is indeed a cochain
map.
8We could not find any reference which contains explicit descriptions of cochain maps realizing the
n−k
isomorphism between Hprim (XG , C) and A0cX /(QS + ∆)(A−1
cX ).
for any homogeneous elements a, b, c ∈ C.

9
(b) (C, ·, K, ℓK
2 ) is called a GBV(Gerstenhaber-Batalin-Vilkovisky)-algebra over k where
|a|
(2.4) ℓK
2 (a, b) := K(a · b) − K(a) · b − (−1) a · K(b), a, b ∈ C,
if (C, K, ℓK K
2 ) is a shifted DGLA(differential graded Lie algebra) and (C, ·, ℓ2 ) is a
G-algebra,
(c) (C, ·, K, ℓK
2 , Q), where Q : C → C is a linear map of degree 1, is called a dGBV(differential
Gerstenhaber-Batalin-Vilkovisky) algebra if (C, ·, K, ℓK 2 (·, ·)) is a GBV algebra and
(C, ·, Q) is a cdga(commutative differential graded algebra), i.e.
Q2 = 0, Q(a · b) = Q(a) · b + (−1)|a| a · Q(b), a, b ∈ C.
We also introduce the C-linear map

XN
∂ ∂
∆ := K − Q = : A → A.
i=1
∂qi ∂ηi
∆ is also a differential of degree 1 (Q and K also have degree 1), i.e. ∆2 = 0. Therefore we
have
∆Q + Q∆ = 0.
Then a straightforward (lengthy) computation shows the following proposition:
Proposition 2.5. The quintuple (A, ·, KS , ℓK

2 , QS ) defined in (2.2) is a dGBV algebra
S
∆
over C and so is (A, ·, ∆, ℓ2 , QS ).
2.6. The 0-th cohomology of (A, KS ) and the Jacobian ideal of S. Associated with
Ech , we define
N
X
R := ch(qi )qi ηi ∈ A−1
0 .
i=1
Note that, for any f ∈ A,
N
X
∂ ∂
δR f := ℓK
2 (R, f ) =
S
ch(qi ) qi − ηi f.
i=1
∂qi ∂ηi
Hence for any f ∈ Aλ , we have δR f = λf . Note also that QS R = 0, while
N
X k
X
KS R = −cX , cX := −ch(qi ) = di − n − 1 = the background charge.
i=1 i=1
Note that cX = 0 means that X is a Calabi-Yau variety. Since KS preserves the charge
grading, we have the following proposition.
Proposition 2.6. For each charge λ ∈ Z, we have a cochain subcomplex (Aλ , KS ).
It follows that the cohomology H = H(A, KS ) of the cochain complex (A, KS ) has a
charge decomposition
M
H= Hλ , Hλ := H(Aλ , KS ).
λ∈Z
Proposition 2.7. The cohomology H = H(A, KS ) is concentrated in charge cX , i.e., H =

HcX . In particular, we have an isomorphism
A0cX /(QS + ∆)(A−1 0
cX ) ≃ A /(QS + ∆)(A
−1
).
9(C, ·, K) is called a BV algebra, if (C, ·, K, ℓK ) is a GBV algebra

2
Proof. From
KS (f · R) = ℓK
2 (R, f ) + KS R · f + KS f · R
S
= δR f − cX f + K S f · R
= (λ − cX )f + KS f · R (if f ∈ Aλ ),
we see that any f ∈ Aλ ∩ Ker KS belongs to Im KS unless λ = cX , since (λ − cX )f =

KS (f · R).
Lemma 2.8 (Weak QS ∆-lemma). For f ∈ A−1

cX , assume that QS (f ) = 0. Then ∆(f ) =
−2
−QS (∆(v)) for some v ∈ AcX .
Proof. Let {uα : α ∈ I} be a C-basis of A0cX /QS (A−1 cX ). Because there is an isomorphism
P
H 0 (AcX , QS ) ≃ H −1 (AcX , QS ) given by u 7→ Ru where R = i ch(qi )qi ηi (see [17], Propo-
sition 4.14), {Ruα : α ∈ I} is a C-basis of H −1 (AcX , QS ). Since QS (f ) = 0, f can be
written as
X
f= aρ · Ruρ + QS (v)
ρ∈I
for some v ∈ A−2

cX . A direct computation shows that
∆(Ruρ ) = 0.
Thus, we have ∆(f ) = ∆(QS (v)) = −QS (∆(v)).
Lemma 2.9. We have an isomorphism

∼
A0cX /(QS + ∆)(A−1
cX ) A0cX /QS (A−1
cX )
as C-vector spaces.
Proof. Let dim A0cX /QS (A−1 0

cX ) = µ. We can choose e1 , . . . , eµ ∈ AcX such that {eα +
0 −1
Im(QS ) : α ∈ I} is C-basis of AcX /QS (AcX ).
Suppose that
Xµ
aα · eα ≡ 0 mod QS + ∆.
α=1
We want to show aα = 0 for all α, i.e., {eα + Im(QS + ∆) : α ∈ I} is linearly independent

∆)(A−1
in A0cX /(QS +P cX ).
Let R(i) = wt(eα )=i aα · eα . Therefore
µ
X
aα · eα = R(0) + R(1) + · · · + R(n−k) = (QS + ∆)(η(0) + η(1) + · · · + η(n−k+1) ),
α=1
where wt(η(i) ) = i. Note that wt(∆) = −1 and QS (ηn−k+1 ) = 0.

Assuming that R(i) 6= 0 for some i, we can choose the largest r such that R(r) 6= 0. Then
R(r) = QS (η(r) ) + ∆(η(r+1) ).
But QS (ηr+1 ) = R(r+1) = 0. The weak QS ∆-lemma 2.8 says that ∆(η(r+1) ) ∈ Im(QS ).
Therefore,
R(r) ∈ Im(QS ), i.e., R(r) ≡ 0 mod QS .
Since {eα + Im(QS ) : α ∈ I} is C-basis of A0cX /QS (A−1
cX ), aα = 0 for all α with wt(eα ) = r.
It means that R(r) = 0, which contradicts to the assumption that R(i) 6= 0 for some i. Thus
R(i) = 0 for all i. Therefore aα = 0 for all α ∈ I.
Now we verify that {eα + Im(QS + ∆)} spans A0cX /(QS + ∆)(A−1 cX ). For that we write
x as
X
x= nα,0 · eα + QS (u0 ), nα,0 ∈ C, u0 ∈ Ac−1
X
,
α∈I
X
= nα,0 · eα − ∆(u0 ) + KS (u0 ).
α∈I
Then we can write −∆(u0 ) as

X
−∆(u0 ) = nα,1 · eα + QS (u1 ), nα,1 ∈ C, u1 ∈ A−1
cX ,
α∈I
X
= nα,1 · eα − ∆(u1 ) + KS (u1 ).
α∈I
We continue this process:
X
−∆(uk ) = nα,k+1 · eα + QS (uk+1 ).
α∈I
Since the differential QS increases the polynomial degree at least by min{di : i = 1, . . . , k},
but ∆ decreases the polynomial degree by 2, this process stops after finite steps, i.e.,
∆(uk ) = 0 for some k. This implies that
X
x= nα · eα + KS (u)
α∈I
for some nα ∈ C and u ∈ A−1

cX .
∼
Denote the isomorphism in Lemma 2.9 by q : A0cX /QS (A−1 0 −1
cX ) → AcX /(QS + ∆)(AcX ).
Theorem 2.10. There is an isomorphism

∼
Φ := ϕ ◦ q : H 0 (AcX , QS ) → Hprim
n−k
(X, C).
Note that ϕ is defined in the level of cochains but q is defined in the level of cohomologies.
Remark 2.11. It is well-known that there is an isomorphism between H 0 (AcX , QS ) and

n−k
Hprim (X, C) (for example, see Theorem 1 and relevant references in [4]). But we could not
find a suitable literature which realizes that isomorphism explicitly; we provide them here,
i.e., the maps ϕ and q. In fact, the middle bridge A0cX /(QS +∆)(A−1cX ), the 0-th cohomology
of the dGBV algebra, will play a crucial role in our theory.
2.7. A quantization dGBV algebra. Proposition 2.7 and Lemma 2.9 suggest the fol-
lowing quantization complex10 A•cX ((~)), QS + ~∆ with the formal parameter ~ satisfying
ch(~) = 1, wt(~) = 1, and |~| = 0: If ~ = 0, then we get the commutative differential graded
algebra (A•cX , QS ), and if ~ = 1, then we get the quantum complex (A•cX , QS + ∆) whose
N −2 n−k
0-th cohomology is canonically isomorphic to HdR (Pn \ XG ) ≃ Hprim (XG , C). Note that
wt(QS ) = 1 and wt(∆) = 1 imply that wt(QS + ~∆) = 0, and ch(QS ) = ch(∆) = 0 also
implies that ch(QS + ~∆) = 0.
The complex (AcX ((~)), QS + ~∆) is Z-filtered:

A•cX ((~))(m) , QS + ~∆ := A•cX [[~]]~−m , QS + ~∆
for m ∈ Z. We define a cochain map r(0)

r (0)
A•cX [[~]], QS + ~∆ (A•cX , QS )

10The quantization complex A•cX ((~)), QS + ~∆ can be viewed as a perturbative expansion of
(AcX , QS )
by taking the quotient11 by A•cX ((~))(−1) :

r (0)
0 A•cX ((~))(−1) A•cX ((~))(0) A•cX 0.
(m)
Definition 2.12. Let us define HS and HS as follows:
A0cX ((~))
HS := ,
(QS + ~∆)(A−1
cX ((~)))
(m) A0cX ((~))(m) A0cX [[~]]~−m

HS := =
(QS + ~∆)(A−1
cX ((~))
(m) ) (QS + ~∆)(A−1
cX [[~]]~
−m )
for m ∈ Z.
We also have an exact sequence:

(−1) (0) r (0)
0 HS HS A0cX /QS (A−1
cX ) 0.

Proposition 2.13. The triple A•cX ((~)), QS + ~∆, ℓ∆
2 ( , ) is a shifted differential graded
1 QS +~∆
Lie algebra. Note that ℓ∆
2 = ~ ℓ2 .
Proposition 2.14. If the background

charge cX is 0, i.e., XG is Calabi-Yau, the quintuple
A•0 ((~)), ·, QS + ~∆, ℓ∆
2 , Q S is a dGBV algebra.
n−k
Let H := Hprim (XG , C). From the previous discussion, we have C-linear isomorphisms:
A00
(0) ∼ ∼
r(0) HS ϕ H.
q
(QS + ~∆)(A−1
0 )
2.8. Deformation of dGBV algebra and the Gauss-Manin connection. From now
on throughout the paper,
we will assume that XG is Calabi-Yau,
i.e., the background charge cX = 0.
Definition 2.15. Let µ := dimC H = |I|. Then,

def
TH := The space of formal tangent vector fields on H.
Let us consider the deformation of dGBV algebra. If we let {tα : α ∈ I} to be the

coordinate of the affine manifold H, then TH ≃ H[[t]] in which we identify {∂/∂tα : α ∈ I}
as the basis of H, thus we have

TH ≃ H[[t]] ∼
Φ
A 0
0 /Q S (A −1
0 ) [[t]]
where Φ in Theorem 2.10 is assumed to be C[[t]]-linear. Note that QS (A−1 0 ) is the Jacobian
ideal of the commutative ring A00 = C[q]0 generated by ∂q ∂S
1
, . . . ∂S
∂qN . It suggests us to
consider the following complex:

A•0 [[t]]((~)), QS + ~∆ .
where QS and ∆ are assumed to be C[[t]]-linear. Clearly, we have the following proposition:

Proposition 2.16. The quintuple A•0 [[t]]((~)), ·, QS + ~∆, ℓ∆
2 , QS is a dGBV algebra.
11Sending ~ = 0 is the classical limit of the quantization complex

Now we deform differentials QS +~∆ and QS in order to make a deformed dGBV algebra.
Choose an arbitrary Γ ∈ A00 ⊗ C[[t]], then Γ is a solution to the Maurer-Cartan equation for
the DGLA A•0 [[t]]((~)), QS + ~∆, ℓ∆
2 ( , ) , i.e.,
1 S +~∆
(QS + ~∆)(eΓ − 1) := (QS + ~∆)(Γ) + ℓQ (Γ, Γ) = 0.
2 2
Define
XN
∂(S + Γ) ∂
QS+Γ := ℓ∆
2 (S + Γ, · ) = .
i=1
∂qi ∂ηi
Then, the operator QS+Γ + ~∆ is a C[[t]]((~))-linear map on A• [[t]]((~)) of degree 1, charge
0, weight 0, and satisfies (QS+Γ + ~∆)2 = 0 (see Lemma 3.3, [17]). In fact, a direct
computation (the deformation formalism) implies the following proposition:

Proposition 2.17. The quintuple A•0 [[t]]((~)), ·, QS+Γ + ~∆, ℓ∆
2 , QS+Γ is a dGBV algebra.
(m)
Definition 2.18. We define the 0-th cohomology HS+Γ and a filtered piece HS+Γ as follows:
A00 [[t]]((~)) (m) A00 [[t]][[~]]~−m
HS+Γ := , HS+Γ :=
(QS+Γ + ~∆)(A−1 0 [[t]]((~))) (QS+Γ + ~∆)(A−1 0 [[t]][[~]]~
−m )
for m ∈ Z. We call this as the ~-filtration.
Lemma 2.19. Let {uα : α ∈ I} be a C-basis of A00 /QS (A−1 0 ). Suppose that the set of
polynomials {Uα : Uα ∈ C[[t]][[~]], α ∈ I} satisfies the condtions Uα |t=0,~=0 = uα for all
(0)
α ∈ I, then {Uα : Uα ∈ C[[t]][[~]], α ∈ I} is a C[[t]][[~]]-basis of HS+Γ .
Proof. First, we prove that {uα : α ∈ I} is a C[[~]]-basis of A00 [[~]]/(QS + ~∆)(A−1

0 [[~]]). For
P r
all ψ = r≥0 ψr ~r! ∈ A00 [[~]], consider the following equation:
X ~r XX∞
~r X∞
~r X∞
~r
ψr = mψr ρ · uρ + QS λψr + ~∆ λψr .
r! ρ r=0
r! r=0
r! r=0
r!
r≥0
By comparing the ~-power terms, we get

X
ψ0 = mψ0 ρ uρ + QS (λψ0 ),
ρ
X
ψ1 − ∆(λψ0 ) = mψ1 ρ uρ + QS (λψ1 ),
ρ
ψ2 X1 λ
ψ2
− ∆(λψ1 ) = mψ2 ρ u ρ + Q S ,
2 ρ
2 2
..
.
ψr λ X 1 λ
ψr−1 ψr
−∆ = mψr ρ u ρ + Q S .
r! (r − 1)! ρ
r! r!
The coefficient mψr ρ is independent of the choice of λψr−1 by Lemma 2.8. Therefore, this
1
procedure uniquely determines r! mψr r .
(0)
For a proof that {Uα : Uα ∈ C[[t]][[~]], α ∈ I} is a C[[t]][[~]]-basis of HS+Γ , we use the same
procedure; it requires to compare not only ~-power terms, but also t-power terms, which
we leave to the reader.
Definition 2.20. We define the Gauss-Manin connection on HS+Γ . We use the notation
[·] to denote the cohomology class.
S+Γ
(i) For each α ∈ I, we define a connection α ~ : HS+Γ → HS+Γ by
−(S+Γ)
S+Γ
~ ([w]) := e
∂ S+Γ
α ~ e ~ w
∂tα

∂ 1 ∂(S + Γ)
= w+ w
∂tα ~ ∂tα
for w ∈ A00 [[t]]((~)).
S+Γ
(ii) We also define a connection ~
~−1 : HS+Γ → HS+Γ by
−(S+Γ)
S+Γ ∂ S+Γ
~
~
−1 ([w]) := e ~ e ~ w
∂~−1

∂
= w + (S + Γ)w
∂~−1
for w ∈ A00 [[t]]((~)).
S+Γ
These are well-defined on cohomology classes, since QS+Γ + ~∆ commutes with α
~
S+Γ
and ~
~−1 . Note that
S+Γ
S+Γ
(m) (m+1) (m) (m)
HS+Γ ⊆ HS+Γ and HS+Γ ⊆ HS+Γ , m ∈ Z.
~ ~
α ~−1
S+Γ
Thus, ~ α
~ preserves the ~-filtration.
3. An algorithm for formal F -manifolds

3.1. Definition of formal F -manifolds. In this subsection, we briefly review the defini-
tion of the formal F -manifold structure in [9]. We will only consider a pure even manifold
for simplicity, so the signs disappear.
Definition 3.1. Let M be a complex manifold of finite dimension, whose holomorphic

structure sheaf and holomorphic tangent sheaf are denoted by OM and TM respectively.
Let ◦ be a OM -bilinear operation ◦ : TM × TM → TM . Then (M, ◦) is called an F -manifold
if the following conditions are satisfied for all X, Y, Z, W ∈ TM :
(F1) (Associativity) The multiplication ◦ is associative:
(X ◦ Y ) ◦ Z = X ◦ (Y ◦ Z),
(F2) (Commutativity) The multiplication ◦ is commutative:
X ◦ Y = Y ◦ X,
(F3) (Potential)12 The following equality holds:
(3.1) [X ◦ Y, Z ◦ W ] − [X ◦ Y, Z] ◦ W − Z ◦ [X ◦ Y, W ] − X ◦ [Y, Z ◦ W ] + X ◦ [Y, Z] ◦ W

+ X ◦ Z ◦ [Y, W ] − Y ◦ [X, Z ◦ W ] + Y ◦ [X, Z] ◦ W + Y ◦ Z ◦ [X, W ] = 0.
One can similarly define its formal version: we consider the formal structure sheaf and
the formal tangent bundle instead of the holomorphic structure sheaf and the holomorphic
tangent bundle.
Let tM := {tα } be the formal coordinates on the open subset Ui . Choose sufficiently small
open subset U of Ui , then we could assume that O(U ) = C[[tM ]]. On the local coordinates
{tM , U }, the product ∂α ◦ ∂β is written as a linear combination of {∂α := ∂/∂tα } with
12We only consider a pure even manifold, so the equality (3.1) is revised accordingly. See 1.definition
in [9] for the equality with sign on a supermanifold.
coefficients in C[[tM ]]. Let Aαβ γ ∈ C[[tM ]] be a formal power series representing the 3-tensor
field such that X
∂α ◦ ∂β := Aαβ ρ · ∂ρ .
ρ
Then, consider the following conditions:
(C1) (Associativity) X X
Aαβ ρ Aργ δ = Aβγ ρ Aρα δ ,
ρ ρ
(C2) (Commutativity)
Aβα γ = Aαβ γ ,
(C3) (Potential)
∂α Aβγ δ = ∂β Aαγ δ .
Recall that TM is the space of formal tangent vector fields on a complex manifold M .
Proposition 3.2. Let ◦ : TM × TM → TM be a OM -bilinear operator. Assume that (M, ◦)

satisfies the conditions (C1), (C2), and (C3) in local coordinates. Then (M, ◦) is a formal
F -manifold.
Proof. It is sufficient to check for commuting vector fields (X, Y, Z, W ) = (∂α , ∂β , ∂γ , ∂δ ).

Then the left hand side of (F3) becomes
hX X i hX i hX i
Aαβ φ ∂φ , Aγδ ρ ∂ρ − Aαβ φ ∂φ , ∂γ ◦ ∂δ − ∂γ ◦ Aαβ φ ∂φ , ∂δ
φ ρ φ
X X
− ∂α ◦ [∂β , Aγδ ρ ∂ρ ] − ∂β ◦ [∂α , Aγδ ρ ∂ρ ]
X X X
= (Aαβ φ (∂φ Aγδ ρ )∂ρ − Aγδ ρ (∂ρ Aαβ φ )∂φ ) + (∂γ Aαβ φ )Aφδ ρ ∂ρ + (∂δ Aαβ φ )Aγφ ρ ∂ρ
φ,ρ φ,ρ φ,ρ
X X
ρ φ ρ φ
− (∂β Aγδ )Aαρ ∂φ − (∂α Aγδ )Aβρ ∂φ
φ,ρ φ,ρ
X
= (∂γ (Aαβ φ Aφδ ρ ) − ∂γ Aβδ φ Aαφ ρ − ∂α (Aγδ φ Aβφ ρ ) + ∂α Aδβ φ Aγφ ρ )∂ρ
φ,ρ
X
= (Aβδ φ ∂γ Aαφ ρ − Aδβ φ ∂α Aγφ ρ )∂ρ
φ,ρ
=0,
where the second equality holds by potential (C3), the third equality holds by associativity
(C1), and the last equality holds by commutativity (C2) and potential (C3). Therefore the
three conditions (C1), (C2), and (C3) induce (F3).
3.2. A connection matrix for the Gauss-Manin connection. In order to construct
n−k
a formal F -manifold structure on H = Hprim (X, C), we will calculate a connection matrix
for the Gauss-Manin connection. Choose a solution Γ ∈ A00 [[t]] of the Maurer-Cartan equa-
tion, and suppose that ∂α Γ|t=0 = uα for all α ∈ I, where {uα : α ∈ I} is a C-basis of
(m)
A00 /QS (A−1
0 ). Recall that Γ gives us the 0-th cohomology HS+Γ and a filtered piece HS+Γ
such that
A00 [[t]]((~)) (m) A00 [[t]][[~]]~−m
HS+Γ := −1 , HS+Γ :=
(QS+Γ + ~∆)(A0 [[t]]((~))) (QS+Γ + ~∆)(A−1 0 [[t]][[~]]~
−m )
S+Γ
(0)
for m ∈ Z. Since ∂α Γ = ~ α
~ 1 is a C[[t]][[~]]-basis of HS+Γ by Lemma 2.19, there is a
S+Γ
connection matrix Aαβ ρ ∈ C[[t]][[~]] with respect to the basis {~ ρ ~ 1 : ρ ∈ I} such that
S+Γ S+Γ X S+Γ
~ β ~ (~ α ~ 1) = Aαβ ρ · (~ ρ ~ 1) + (QS+Γ + ~∆)(Λαβ )
ρ∈I
for some Λαβ ∈ A−1 ρ

0 [[t]][[~]] and all α, β ∈ I. Suppose that there is Γ which makes Aαβ and
ρ
Λαβ have no ~-power terms, i.e., there is a connection matrix Aαβ ∈ C[[t]] such that
S+Γ S+Γ X S+Γ
(3.2) ~ β
~
(~ α
~ 1) = Aαβ ρ · (~ ρ
~ 1) + (QS+Γ + ~∆)(Λαβ ),
ρ∈I
for some Λαβ ∈ A−10 [[t]] and all α, β ∈ I. Then, by comparing ~-power terms, the equation
(3.2) reduces further to the following:
X
∂α Γ · ∂β Γ = Aαβ ρ · ∂ρ Γ + QS+Γ (Λαβ ),
(3.3) ρ∈I
∂β ∂α Γ = ∆(Λαβ ),
for all α, β ∈ I. We claim that the coefficients Aαβ ρ give a formal F -manifold structure on
H. For simplicity, we use the notation Γα1 α2 ···αℓ := ∂αℓ · · · ∂α2 ∂α1 Γ.
Remark 3.3. (1) The equation (3.3) is a special case of the equation (1.1) for weak
primitive forms as summarized in Table 2.
(2) The equation which appeared in the proof of Lemma 7.1 in [1] is the equation (3.3)
with α = β.
Proposition 3.4. Assume that Γ, Aαβ ρ , and Λαβ satisfy the equation (3.2). Then Aαβ ρ
satisfies (associativity), (commutativity), and (potential). In particular, Aαβ ρ equip H =
n−k
Hprim (X, C) with a formal F -manifold structure by Proposition 3.2.
S+Γ
Proof. By applying ~ γ
~ to the equation (3.2), we get
S+Γ S+Γ S+Γ S+Γ X S+Γ
~ γ
~ (~ β
~
(~ α
~ 1)) = ~ γ
~ Aαβ ρ Γρ + (QS+Γ + ~∆)(~ γ
~ Λαβ )
ρ∈I
Xh S+Γ
i S+Γ
(3.4) = ~∂γ Aαβ ρ Γρ + Aαβ ρ ~ γ

~ Γρ + (QS+Γ + ~∆)(~ γ
~ Λαβ )
ρ∈I
X X S+Γ X
δ ρ δ
= ~∂γ Aαβ + Aαβ Aγρ Γδ + (QS+Γ + ~∆) ~ γ
~ Λαβ + Aαβ ρ Λγρ .
δ∈I ρ∈I ρ∈I
Note that
S+Γ S+Γ S+Γ
~ γ
~ (~ β
~
(~ α
~ 1)) = ~2 Γαβγ + ~(Γα Γβγ + Γβ Γαγ + Γγ Γαβ ) + Γα Γβ Γγ .
Since the left hand side of equation (3.4) is invariant under the permutation of α, β, γ at
each ~-power term and the coefficient of Γδ in the right hand side is unique by Lemma 2.19,
X
Aαβ ρ Aγρ δ and ∂α Aβγ ρ
ρ
are also invariant under the permutation of α, β, γ. Therefore, the associativity and the
potential are satisfied. It is important that Aαβ ρ has no ~-power terms.
The commutativity follows from the equation (3.2) directly by the same argument.
n−k
Therefore, the key question to construct a formal F -manifold structure on H = Hprim (X, C)
is the following:
Can we find a Maurer-Cartan solution Γ whose connection matrix Aαβ ρ
(and Λαβ ) has no ~-power terms?
3.3. An explicit algorithm. We will provide an explicit algorithm which calculates Γ, Aαβ ρ
satisfying the equation (3.2). We use the following notations:
X X X 1
Γ= u α · tα + uα tα ∈ A00 [[t]],
m
m!
α∈I m≥2 α∈I
X X 1
(3.5) Aαβ ρ = aαβ ρ + aαβα ρ tα ∈ C[[t]],
m
m!
m≥1 α∈I
X X 1
Λαβ = λαβ + λαβα tα ∈ A−1
0 [[t]],
m
m!
m≥1 α∈I
where uα ∈ A00 , aαβ ρ , aαβα ρ ∈ C, λαβ , λαβα ∈ A−1 0 , and t

α1 ···αm
:= tα1 · · · tαm . We are
ρ
assuming that uα , aαβα , λαβα are invariant under the permutation of indices of α under
our notation. This assumption is important for our algorithm (see Remark 3.7). In order
(0)
to make {Γρ : ρ ∈ I} be a C[[t]][[~]]-basis of HS+Γ , we define the tα -term of Γ as uα
where {uα : α ∈ I} is a C-basis of A00 /QS (A−1 0 ); see Lemma 2.19. We have to determine
uα , aαβ ρ , aαβα ρ which satisfy the equation (3.2).
For the algorithm, Definition 3.5 and Lemma 2.8 (the weak QS ∆-lemma) are required.
(i)
Definition 3.5. Let |α| = m which means that α ∈ I m . We define the notation uα as
follows:
X 1
(i)
uα = uα · · · uαm−i , (0 ≤ i ≤ m − 1),
α ⊔···⊔α =α
(m − i)! 1
1 m−i
αj 6=∅
where the notation α1 ⊔ · · · ⊔ αm−i = α means α1 ∪ · · · ∪ αm−i = α and αk ∩ αℓ = ∅ for

ℓ 6= k.
(i)
Note that uα is invariant under the permutation of indices of α. For example,
(0)
(0)
uαβγδ = uα uβ uγ uδ ,
uαβγ = uα uβ uγ , (1)
(1)
uαβγδ = uα uβγδ + uβ uαγδ + uγ uαβδ + uδ uαβγ ,
uαβγ = uα uβγ + uβ uαγ + uγ uαβ and (2)
(2)
uαβγδ = uαβ uγδ + uαγ uβδ + uαδ uβγ ,
uαβγ = uαβγ , (3)
uαβγδ = uαβγδ .
Now we give the algorithm for Γ and Aαβ ρ as follows:

Step-0: Choose a C-basis {uα : α ∈ I} of JS := A00 /QS (A−1 0 ). Note that JS is a finite
dimensional C-vector space, and there is a partition of {uα : α ∈ I} in terms
Sn−k
of weights, i.e., we decompose {uα : α ∈ I} = r=0 {uα : α ∈ Ir } where I =
I0 ⊔ · · · ⊔ In−k . We use the notations umin = uα1 and umax = uα2 for α1 ∈ I0 and
α2 ∈ In−k where |I0 | = |In−k | = 1 (since XG is a Calabi-Yau manifold).
(0) (0)
Step-1: Determine aαβ and λαβ using a basis {uα : α ∈ I} of JS as follows:
X (0) (0)
uα uβ = aαβ ρ uρ + QS (λαβ ).
ρ∈I
(0) (0) (0)

Note that aαβ ρ ∈ C is unique and λαβ is unique up to ker QS . Therefore, ∆(λαβ )
(0)
is unique up to Im(QS ) by the weak QS ∆-lemma 2.8. Then, define uαβ := ∆(λαβ )
(0)
and aαβ ρ := aαβ ρ .
(i) ρ (i)
Step-ℓ (ℓ ≥ 2): Suppose that |α| = ℓ + 1. Determine aα and λα (0 ≤ i ≤ ℓ − 1) in
sequence as follows:
X
u(0)
α = a(0) ρ (0)
α uρ + QS (λα ),
ρ∈I
X
u(1)
α − ∆(λ(0)
α ) = a(1) ρ (1)
α uρ + QS (λα ),
ρ∈I
(3.6)
..
.
X
u(ℓ−1)
α − ∆(λ(ℓ−2)
α )= a(ℓ−1)
α
ρ
uρ + QS (λ(ℓ−1)
α ).
ρ∈I
(i) (i)
Since λα is unique up to ker QS for 0 ≤ i ≤ ℓ − 2, ∆(λα ) is unique up to Im(QS )
(ℓ−1) ρ
by the weak QS ∆-lemma 2.8. Therefore, aα is independent of the choices of
(i) ρ (ℓ−1) ρ (ℓ−1)
λα . Then, define aα := aα and uα := ∆(λα ).
By this inductive algorithm, we can completely determine Γ, Aαβ ρ , and Λαβ which turn out
to satisfy (see subsection 3.4) the equations (3.3).
Corollary 3.6. Suppose that Γ, Γ′ ∈ A00 [[t]] satisfy the equation (3.2) with the connection
matrices Aαβ ρ and A′αβ ρ respectively. If Γα |t=0 = Γ′α |t=0 = uα for some C-basis {uα :
α ∈ I} of A00 /QS (A−1 ρ ′ ρ ρ
0 ), then Aαβ = Aαβ . In other words, the result Aαβ of the above
0 −1
algorithm depends only on the choice of C-basis {uα : α ∈ I} of A0 /QS (A0 ).
Remark 3.7 (Non-triviality of solving the differential equations (3.3)). The symmetries
among the indices of α in uα , aαβα ρ , λαβα play an important role. If one tries to solve the
equations in a naive way by comparing the coefficients of the t-powers, then one would get
into trouble that the symmetries among the indices is not guaranteed. But our algorithm
guarantees that the relevant quantities such as uα , aαβα ρ , λαβα are invariant under the
permutation of α as the equations (3.6) indicate. Suppose that there is no condition such
as uαβ = uβα , i.e., we do not put symmetry restrictions on the index notation. Then, the
equation
∂β ∂α Γ = ∆(Λαβ )
in the equations (3.3) gives the following equation
uαβ + uβα = 2∆(λαβ ),
but then it cannot determine the values of uαβ , uβα individually from the data 2∆(λαβ )
of the previous step. This implies that the naive way of solving (3.3) inductively with
non-symmetric notations does not work well, either.
Remark 3.8 (Difficulty of constructing an algorithm for a flat metric). For a formal Frobe-
nius manifold structure (see Definition 5.1), we need to define a flat metric g on TH (i.e.,
∂γ g(∂α , ∂β ) = 0 for all α, β, γ ∈ I), which is compatible with the formal F -manifold struc-
ture. A natural candidate of a metric would be g(∂α , ∂β ) = Aαβ max using the weight
filtration (the Hodge filtration), since such a choice corresponds to a cup product on H and
satisfies the compatibility condition (Invariance) in Definition 5.1. To make g be flat, we
have to find Aαβ ρ satisfying the equation (3.2) with aα max = 0 for all |α| ≥ 3. However,
our algorithm for solving the equation (3.2) is not refined enough to satisfy the conditions
aα max = 0 for all |α| ≥ 3. For example, if we run the algorithm, we get non-zero uαβ and
the step 3 in our algorithm gives the following equations:

X (0) (0)
uα uβ uγ = aαβγ ρ uρ + QS (λαβγ ),
ρ∈I
(0)
X (1)
uα uβγ + uβ uαγ + uγ uαβ − ∆(λαβγ ) = aαβγ ρ uρ + QS (λαβγ ).
ρ∈I
It is hard to choose C-basis {uα : α ∈ I} satisfying aαβγ max = 0. Note that Corollary
3.6 implies that aαβγ max only depends on the choice of {uα :Pα ∈ I}. On the other
hand, in section 5, we will use a Maurer-Cartan solution L = α tα uα , i.e., Γ with the
condition uα =0 for |α| ≥ 2, and solve a differential equation (5.2) instead of (3.3) in order
to construct a formal Frobenius manifold. Note that both (5.2) and (3.3) are special cases
of the differential equation (1.1) for weak primitive forms.
3.4. A proof why the algorithm works. In this subsection, we give a proof of the
algorithm in subsection 3.3.
Definition 3.9. We define quantities U, B, C, which are our key players in the proof of
the algorithm, as follows:
S+Γ S+Γ
Uα1 ···αm := ~ ~
αm ···~ ~
α1 1,


 Bα1 α2 ρ :=Aα1 α2 ρ ,

 X

 Bα1 α2 α3 ρ := Bα1 α2 δ Bδα3 ρ + ~ · ∂α3 (Bα1 α2 ρ ),



 δ∈I
 .
..



 X



 B α ···α
ρ
:= Bα1 ···αm−1 δ Bδαm ρ + ~ · ∂αm (Bα1 ···αm−1 ρ ),
 1 m
δ∈I

 Cα1 α2 :=Λα1 α2 ,

 X

 δ S+Γ

 C α α α := B α α C δα + ~ ~ Cα1 α2 ,


1 2 3 1 2 3 α3
 δ∈I
 ..

 .

 X

 S+Γ
 Bα1 ···αm−1 δ Cδαm + ~ αm
Cα1 ···αm :=
 Cα1 ···αm−1 .
~
δ∈I
Theorem 3.10. The equations (3.2) hold for all α, β ∈ I if and only if the following
inductive equations hold for all m ≥ 2 and all α ∈ I m :
X
(3.7) Uα = Bα ρ Γρ + (QS+Γ + ~∆)(Cα ).
ρ∈I
Proof. It is trivial that the above inductive equations imply the equation (3.2) because the
m = 2 case of (3.7) is same as (3.2).
Suppose that the equation (3.2) holds for all α, β ∈ I. Therefore, the inductive equations
hold for all α ∈ I 2 . Assume that the inductive equations hold for all α ∈ I ℓ , i.e.,
X
(3.8) Uα1 ···αℓ = Bα1 ···αℓ ρ Γρ + (QS+Γ + ~∆)(Cα1 ···αℓ ),
ρ∈I
S+Γ
for all (α1 , . . . , αℓ ) ∈ I ℓ . Take ~ αℓ+1~ on the equation (3.8). Then we have the following
equations:
S+Γ X S+Γ S+Γ
~ αℓ+1~ Uα1 ···αℓ = ~ αℓ+1
~ (Bα1 ···αℓ ρ Γρ ) + (QS+Γ + ~∆)(~ αℓ+1
~ Cα1 ···αℓ ),
ρ∈I
Xh S+Γ S+Γ
i
Uα1 ···αℓ+1 = ~∂αℓ+1 Bα1 ···αℓ ρ Γρ + Bα1 ···αℓ ρ ~ ~
αℓ+1 ~ ~
αρ 1
ρ∈I
S+Γ
+ (QS+Γ + ~∆)(~ αℓ+1 ~ Cα1 ···αℓ )
X hX i
ρ ρ δ
= ~∂αℓ+1 Bα1 ···αℓ Γρ + Bα1 ···αℓ Aαℓ+1 ρ Γδ + (QS+Γ + ~∆)(Λαℓ+1 ρ )
ρ∈I δ∈I
S+Γ
+ (QS+Γ + ~∆)(~ αℓ+1~ Cα1 ···αℓ )
X X
ρ δ ρ
= ~∂αℓ+1 Bα1 ···αℓ + Bα1 ···αℓ Aαℓ+1 δ Γρ
ρ∈I δ∈I
X S+Γ
+ (QS+Γ + ~∆)( Bα1 ···αℓ δ Λαℓ+1 δ + ~ ~
αℓ+1 Cα1 ···αℓ ).
δ∈I
S+Γ
Recall that ~ α
~ and QS+Γ + ~∆ commute. The last equation implies
X
Uα1 ···αℓ+1 = Bα1 ···αℓ+1 ρ Γρ + (QS+Γ + ~∆)(Cα1 ···αℓ+1 )
ρ∈I
by the definitions of U, B, and C. By the mathematical induction, the result follows.
By a direct computation, we get the following lemma:
Lemma 3.11. For α ∈ I m , we have the following formula:

m−1
X X
i 1
Uα = U α ~i , where i
Uα = Γα · · · Γαm−i ,
i=0 α1 ⊔···⊔αm−i =α
(m − i)! 1
αj 6=∅
where α1 ⊔ · · · ⊔ αm−i = α means α1 ∪ · · · ∪ αm−i = α, and αk ∩ αℓ = ∅ for ℓ 6= k.
For example,
Uαβγ =Γα Γβ Γγ + ~(Γα Γβγ + Γβ Γαγ + Γγ Γαβ ) + ~2 Γαβγ ,

Uαβγδ =Γα Γβ Γγ Γδ + ~(Γα Γβγδ + Γβ Γαγδ + Γγ Γαβδ + Γδ Γαβγ )
+ ~2 (Γαβ Γγδ + Γαγ Γβδ + Γαδ Γβγ ) + ~3 Γαβγδ .
Therefore, we get the following result:

X 1
(3.9) i
Uα |t=0 = u(i)
α = uα · · · uαm−i ,
α1 ⊔···⊔αm−i =α
(m − i)! 1
αj 6=∅
(i)
where uα appeared in Definition 3.5.
By the definitions of B and C, the ~-degree of Bα ρ and Cα is m − 2 when α ∈ I m , i.e.,
we can write
m−2
X m−2
X
ρ i ρ i i
Bα = Bα ~ , C α = Cα ~i ,
i=0 i=0
for some i Bα ρ ∈ C[[t]], and i Cα ∈ A−1 0 [[t]]. Therefore, by comparing the ~-power terms of
(3.7), we get the following sequence of equations:
X
0 0
Uα = Bα ρ Γρ + QS+Γ (0 Cα ),
ρ∈I
X
1 1
Uα = Bα ρ Γρ + QS+Γ (1 Cα ) + ∆(0 Cα ),
ρ∈I
(3.10) ..
.
X
m−2 m−2
Uα = Bα ρ Γρ + QS+Γ (m−2 Cα ) + ∆(m−3 Cα ),
ρ∈I
m−1
Uα = ∆(m−2 Cα ).
Lemma 3.12. The constant term of tα -expansion of m−2 Bα ρ is aα ρ which appeared in

(3.5), i.e.,
m−2
Bα ρ |t=0 = aα ρ .
Proof. It is enough to show that

m−2
(3.11) Bα ρ = ∂αm ∂αm−1 · · · ∂α3 Aα1 α2 ρ ,
where α = α1 α2 · · · αm ∈ I m for m ≥ 2. Use the mathematical induction. When m = 2,
Bα1 α2 ρ = Aα1 α2 ρ by Definition 3.9, which implies that 0 Bα1 α2 ρ = Aα1 α2 ρ .
Suppose that equation (3.11) is true for |α| = ℓ. By Definition 3.9,
X
Bα1 ···αℓ+1 ρ = Bα1 ···αℓ δ Bδαℓ+1 ρ + ~ · ∂αℓ+1 (Bα1 ···αℓ ρ ).
δ∈I
ℓ−1
Therefore, the ~ -term of Bα1 ···αℓ+1 ρ is ∂αℓ+1 (ℓ−2 Bα1 ···αℓ ρ ) = ∂αℓ+1 · · · ∂α3 Aα1 α2 ρ .
According to Lemma 3.12 and the equation (3.9), we get the equations (3.6) by evaluating
t = 0 on the equations (3.10). This implies that the algorithm in subsection 3.3 works to
give a solution to (3.2).
4. Weak primitive forms

4.1. Weak primitive forms and formal F -manifolds. Here we explain how weak primi-
tive forms give arise to formal F -manifolds.
P For this, we consider a Maurer-Cartan solution
of a particular shape, Γ = L := ρ∈I u ρ t ρ
with wt(Γ) = 1. Let ζ := 0 ζ + 1 ζ~ where
S+L
0
ζ, 1 ζ ∈ A00 [[t]] with 0 ζ|t=0 = 1 and wt(ζ) = 0. Note that ~ ρ
~ ζ|t=0,~=0 = uρ for all
S+L
(0)
ρ ∈ I. Therefore, {~ ρ ζ : ρ ∈ I} is a C[[t]][[~]]-basis of
~ (Definition 2.18) by
HS+L
Lemma 2.19. It means that there exists a unique Aαβ ρ ∈ C[[t]][[~]] for all α, β, ρ ∈ I such
that
S+L S+L X S+L
(4.1) ~ β ~ (~ α ~ ζ) = Aαβ ρ · ρ ~ ζ + (QS+L + ~∆)(Λαβ ),
ρ∈I
for some Λαβ ∈ A−1

0 [[t]][[~]]. Now we assume that ζ is a weak primitive form (Definition
1.1), i.e., Aαβ ρ has a particular shape Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~ and Λαβ = 0 Λαβ + 1 Λαβ ~
where 0 Aαβ ρ , 1 Aαβ ρ ∈ A00 [[t]] and 0 Λαβ , 1 Λαβ ∈ A−10 [[t]] satisfy the equation (4.1). Then, we
can rewrite the equation (4.1) as:
S+L S+L X S+L
(4.2) ~ β ~ (~ α ~ ζ) = (0 Aαβ ρ + 1 Aαβ ρ ~)·~ ρ ~ ζ + (QS+L + ~∆)(0 Λαβ + 1 Λαβ ~).
ρ∈I
Note that the equation (4.2) and the equation (S2) (Proposition 7.3 in [22]) are the same
equations; this is the motivation for Definition 1.1 of weak primitive forms. In order to
explain its connection to formal F -manifolds, consider the following equation (with a sim-
S+L
plified notation α = α ~ ):
hX i
~ γ ~ β ~ α ζ =~ γ (Aαβ ρ ) ~ ρ ζ + (QS+L + ~∆)(~ γ Λαβ )
ρ∈I
Xh i
= ~∂γ (Aαβ ρ )~ ρ ζ + (Aαβ ρ )~ γ ~ ρ ζ + (QS+L + ~∆)(~ γ Λαβ )
ρ∈I
X X

= ~Aαβ,γ δ + Aαβ ρ Aγρ δ ~ δ ζ
δ∈I ρ∈I
X
+ (QS+L + ~∆) ~ γ Λ+ Aαβ ρ Λγρ .
ρ∈I
Therefore, the following quantities are invariant under the permutation of α, β, γ for all δ:
X
~Aαβ,γ δ + Aαβ ρ Aγρ δ
ρ∈I
X h X i
0
(4.3) = Aαβ ρ · 0 Aγρ δ + ~ 0 Aαβ,γ δ + 0
Aαβ ρ · 1 Aγρ δ + 1 Aαβ ρ · 0 Aγρ δ
ρ∈I ρ∈I
h X i
2 1 δ 1 ρ 1 δ
+~ Aαβ,γ + Aαβ · Aγρ .
ρ∈I
Suppose that 1 Aαβ ρ = 0 for all α, β, ρ ∈ I. Then, the equation (4.3) implies that the
following quantities are invariant under the permutation of α, β, γ for all δ:
X
0
Aαβ ρ · 0 Aγρ δ and 0 Aαβ,γ δ .
ρ∈I
Note that 0 Aαβ ρ = 0 Aβα ρ from (4.2). Therefore, 0 Aαβ ρ induces the structure of the formal
F -manifold structure when 1 Aαβ ρ = 0.
Remark 4.1. Suppose that we have the Levi-Civita connection represented by

X
1
∇∂α ∂β := Aαβ ρ ∂ρ ,
ρ∈ I
which is compatible with the metric h , i defined by

h∂α , ∂β i := 0 Aαβ max
where 0 Aαβ ρ and 1 Aαβ ρ appeared in the equation (4.2), i.e., 1 Aαβ ρ is the Christoffel symbol
of the Levi-Civita connection ∇. Then, the above argument implies that taking the flat
coordinates (1 Aαβ ρ = 0) for weak primitive forms (and ~ → 0) induces a formal F -manifold
structure, as we indicated in Table 1.
4.2. An explicit algorithm. We will give an explicit algorithm which calculates ζ =

0
ζ + 1 ζ~, Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~ satisfying the equation (4.2). For notational convenience,
let 12 · · · m denote α1 α2 · · · αm . We use the following notations:
X∞ ∞
i 1 X i X 1 X i
A12 ρ = i a12 ρ + a12α ρ · tα , i
ζ = i ζ̄ + ζ̄α · tα . (i = 0, 1)
m=1
m! m m=1
m! m
α∈I α∈I
(0)
Let 0 ζ̄ = 1, so that ~ ρ ζ is a basis of HS+L , and let 1 ζ̄ = 0, so that the weight condition
wt(ζ) = 0 holds. We have to determine i ζ̄α and i aα ρ (i = 0, 1) which satisfy the equation
(4.2). For the algorithm, we need some definition.
Definition 4.2. Let α = α1 · · · αm = 1 · · · m. We define the notations vα (ℓ, j) and bα ρ (ℓ, j)

as follows:
X
vα (ℓ, j) = u(α2 ) · j ζ̄α1 where u(α) := u1 u2 · · · um−1 um ,
α1 ⊔α2 =α
|α1 |=ℓ
and
X 1 X X X
bα ρ (ℓ, j) = ··· r1
a12α1 δ1 · · · rs+1 aδs αs+1 ρ .
α1 ⊔···⊔αs+1 =3···m
s!
r1 +···rs+1 =ℓ−j δs δ1
αi 6=∅ when i6=1 ri =0 or 1
where s = m − j − 2. We define bα ρ (ℓ, j) = 0 when s = m − j − 2 < 0.
Note that vα (ℓ, j) is invariant under the permutation of the indices of α. For example,
X
0
b123 ρ (0, 0) = a12 δ · 0 aδ3 ρ ,
0
v123 (0, 0) = u123 ζ̄ = u123 , δ
X
1
v123 (1, 0) = u12 0 ζ̄3 + u13 0 ζ̄2 + u23 0 ζ̄1 ,
ρ
b123 (1, 0) = a12 δ · 0 aδ3 ρ + 0 a12 δ · 1 aδ3 ρ ,
δ
v123 (1, 1) = u12 1 ζ̄3 + u13 1 ζ̄2 + u23 1 ζ̄1 , and
b123 ρ (1, 1) = 0 a123 ρ ,
v123 (2, 0) = u1 0 ζ̄23 + u2 0 ζ̄13 + u3 0 ζ̄12 , X
1
b123 ρ (2, 0) = a12 δ · 1 aδ3 ρ ,
v123 (2, 1) = u1 1 ζ̄23 + u2 1 ζ̄13 + u3 1 ζ̄12 , δ
b123 ρ (2, 1) = 1 a123 ρ .
Let us introduce the algorithm for ζ and Aαβ ρ as follows:
Step-0: Choose a C-basis {uα : α ∈ I} of JS := A00 /QS (A−1 0 1

0 ), and ζ̄ρ , ζ̄ρ . We can choose
0 1
arbitrary ζ̄ρ and ζ̄ρ with weight conditions.
(0) (0) (1) (1)
Step-1: Determine b12 , c12 , b12 , and c12 using a basis {uα : α ∈ I} of JS as follows:
X (0) (0)
v12 (0, 0) = b12 ρ · uρ + QS (c12 ),
ρ∈I
(4.4) X (0) (0)
X (1) (1)
0
v12 (1, 0) + v12 (0, 1) − b12 ρ · ζ̄ρ − ∆(c12 ) = b12 ρ · uρ + QS (c12 ).
ρ∈I ρ∈I
(i) (i)
Note that b12 ρ ∈ C is unique and c1 is unique up to ker QS for i = 0, 1. Therefore,
(i)
∆(c12 ) is unique up to Im(QS ) by the weak QS ∆-lemma 2.8. Then define
0 (0) 1 (1)
a12 ρ := b12 ρ , a12 ρ := b12 ρ ,
and
X (1) (0) (1)

X (1)
0
ζ̄12 := b12 ρ · 0 ζ̄ρ + b12 ρ · 1 ζ̄ρ + ∆(c12 ) − v12 (1, 1), 1
ζ̄12 := b12 ρ · 1 ζ̄ρ .
ρ∈I ρ∈I
(i) ρ (i)
Step-ℓ (ℓ ≥ 2): Suppose that |α| = ℓ + 1. Determine bα and cα (0 ≤ i ≤ ℓ) in sequence
as follows:
X
vα (0, 0) = b(0) ρ (0)
α uρ + QS (cα ),
ρ∈I
X X
vα (1, 0) + vα (0, 1) − b(0)
α
ρ 0
· ζ̄ρ − ∆(c(0)
α ) = b(1) ρ (1)
α uρ + QS (cα ),
ρ∈I ρ∈I
 
vα (2,h0) + vα (1, 1) i X
X (1)
 − bα ρ · 0 ζ̄ρ + bα ρ · 1 ζ̄ρ − ∆(cα )  =
(0) (1) b(2) ρ (2)
α uρ + QS (cα ),
(4.5)
ρ∈I ρ∈I
..
.
 
vα (ℓ, h0) + vα (ℓ − 1, 1) i X
X (ℓ−1)
 − bα ρ 0
· ζ̄ρ + bα
(ℓ−2) ρ 1 (ℓ−1)  =
· ζ̄ρ − ∆(cα ) b(ℓ) ρ (ℓ)
α uρ + QS (cα ),
ρ∈I ρ∈I
This inductive procedure works well, because the left hand sides of the equation
(ℓ−1) ρ
(4.5) are already determined by the previous steps for i ≤ ℓ − 1. Moreover, bα
(ℓ) ρ (i) (i)
and bα are independent of the choices of cα , since cα is unique up to ker QS for
(i)
0 ≤ i ≤ ℓ − 1, thus ∆(cα ) is unique up to Im(QS ) by Lemma 2.8. Then, define
ℓ−2
X ℓ−2
X
0
(4.6) ρ
aα := b(ℓ−1)
α
ρ
− ρ
bα (ℓ − 1, j), 1 ρ
aα := b(ℓ)
α
ρ
− n−1
bα ρ (ℓ, j),
j=0 j=0
and
X
0
ζ̄α := b(ℓ)
α
ρ 0
· ζ̄ρ + b(ℓ−1)
α
ρ 1
· ζ̄ρ + ∆(c(ℓ)
α ) − vα (ℓ, 1),
ρ∈I
(4.7) X
1
ζ̄α := b(ℓ)
α
ρ 1
· ζ̄ρ .
ρ∈I
By this algorithm, we can determine ζ and Aαβ ρ which turn out to satisfy (see subsection
4.3) the equation (4.2).
4.3. A proof why the algorithm works. In this subsection, we give a proof of the
algorithm in subsection 4.2.
Definition 4.3. We define quantities V, D, E, which are our key players in the proof of
the algorithm, as follows:
S+L S+L
Vα1 ···αm := ~ αm
~ ···~ α1
~ ζ,


 Dα1 α2 ρ :=Aα1 α2 ρ ,

 X

 D ρ
:= Dα1 α2 δ Dδα3 ρ + ~ · ∂α3 (Dα1 α2 ρ ),

 α 1 α2 α3

 δ∈I
 .
..



 X



Dα1 ···αm :=
ρ
Dα1 ···αm−1 δ Dδαm ρ + ~ · ∂αm (Dα1 ···αm−1 ρ ),

δ∈I

 Eα1 α2 :=Λα1 α2 ,

 X

 S+L

 Eα1 α2 α3 :=
 Dα1 α2 δ Eδα3 + ~ α3~ Eα1 α2 ,

 δ∈I
 .
..



 X

 S+L
Eα1 ···αm :=
 Dα1 ···αm−1 δ Eδαm + ~ αm~ Eα1 ···αm−1 .

δ∈I
Theorem 4.4. The equations (4.2) hold for all α, β ∈ I if and only if the following inductive
equations hold for all m ≥ 2 and all α ∈ I m :
X S+L
(4.8) Vα = Dα ρ · ~ ρ ~ ζ + (QS+L + ~∆)(Eα ).
ρ∈I
Proof. Use the same argument as in the proof of Theorem 3.10.
By the mathematical induction, we get the following two lemmas:
Lemma 4.5. Let α = α1 · · · αm , then

m
X
Vα = Vα (0, 0) + (Vα (ℓ, 0) + Vα (ℓ − 1, 1))~ℓ + Vα (m, 1)~m+1
ℓ=1
X
where Vα (ℓ, j) = u(α2 ) · j ζα1 .
α1 ⊔α2 =α
|α1 |=ℓ
Lemma 4.6. Let α = α1 · · · αm , then

m−1
X ℓ
X
Dα ρ = ℓ
D α ρ ~ℓ where ℓ Dα ρ = Bα ρ (ℓ, j),
ℓ=0 j=0
X 1 X X X
and Bα ρ (ℓ, j) = ··· r1
A12α1 δ1 · · · rs+1 Aδs αs+1 ρ
α1 ⊔···⊔αs+1 =3···m
s!
r1 +···rs+1 =ℓ−j δs δ1
αi 6=∅ when i6=1 ri =0 or 1
with s = m − j − 2.
Therefore, we obtain
(4.9) Vα (ℓ, j)|t=0 = vα (ℓ, j) and Bα ρ (ℓ, j)|t=0 = bα ρ (ℓ, j),
where vα (ℓ, j) and bα ρ (ℓ, j) appeared in Definition 4.2.
Lemma 4.7. We get the following formulas:

m−3
X m−3
X
m−1
Dα ρ |t=0 = 0 aα ρ + bα ρ (m − 2, j) and m−2
Dα ρ |t=0 = 1 aα ρ + bα ρ (m − 1, j).
j=0 j=0
Sketch of proof. Use Lemma 4.5, Lemma 4.6, and the equation (4.9).
We can write
m−1
X
ℓ
Eα = Eα ~ℓ
ℓ=0
for ℓ Eα ∈ A−1
0 [[t]]. Therefore, by comparing the ~-power terms of (3.7), we get the following
sequence of equations:
(4.10)
Xh i
0
Vα (0, 0) = Dα ρ uρ + QS+L (0 Eα ),
ρ∈I
Xh i
0
Vα (1, 0) + Vα (0, 1) = Dα ρ0 ζρ + 1 Dα ρ uρ + QS+L (1 Eα ) + ∆(0 Eα ),
ρ∈I
Xh i
0
Vα (2, 0) + Vα (1, 1) = Dα ρ1 ζρ + 1 Dα ρ0 ζρ + 2 Dα ρ uρ + QS+L (2 Eα ) + ∆(1 Eα ),
ρ∈I
..
.
Xh i
m−3
Vα (m − 1, 0) + Vα (m − 2, 1) = Dα ρ1 ζρ + m−2 Dα ρ0 ζρ + m−1 Dα ρ uρ
ρ∈I
+ QS+L (m−1 Eα ) + ∆(m−1 Eα ),

Xh i
m−2
Vα (m, 0) + Vα (m − 1, 1) = Dα ρ1 ζρ + m−2 Dα ρ1 ζρ + ∆(m−1 Eα ),
ρ∈I
Xh i
m−1
Vα (m, 1) = Dα ρ1 ζρ .
ρ∈I
By evaluating t = 0 on the equations (4.10), we get the equations (4.5) and (4.7). We
also get the equations (4.6) by Lemma 4.7. This says that the algorithm in subsection 4.2
provides us a solution to (4.2)
5. An algorithm for formal Frobenius manifolds and higher residue pairings

5.1. Definition of formal Frobenius manifolds. In this subsection we briefly review
the definition of Frobenius manifolds in [22], [1] and formal Frobenius manifolds. We will
only consider a pure even manifold same as an F -manifold case.
Definition 5.1. Let M be a complex manifold of finite dimension, and tM := {tα } be the
formal coordinates on the open subset Ui ⊂ M . Choose sufficiently small open subset U
of Ui , then we could assume that O(U ) = C[[tM ]]. On the local coordinates {tM , U }, the
product δ ◦ δ ′ is written as the linear combinations of {∂α := ∂/∂tα } with coefficients in
C[[tM ]]. Let Aαβ γ ∈ C[[tM ]] be a formal power series representing the 3-tensor field such
that X
∂α ◦ ∂β := Aαβ γ ∂γ .
γ
Let g be a non-degenerate symmetric pairing on C-vector space spanned by {∂α }. Then

one can extend g to the symmetric C[[t]]-pairing. Let gαβ := g(∂α , ∂β ). Then (M, ◦, g) is
called a formal Frobenius manifold if the following conditions are satisfied:
(D1) (Associativity)
X X
Aαβ ρ Aργ δ = Aβγ ρ Aρα δ .
ρ ρ
(D2) (Commutativity)
Aβα γ = Aαβ γ .
P
(D3) (Invariance) Put Aαβγ = ρ Aαβ ρ gργ ,
Aαβγ = Aβγα .
(D4) (Flat identity) The distinguished one ∂0 is the identity with respect to ◦:
A0α β = δ βα (where δ βα is the Kronecker delta).
(D5) (Potential)
∂α Aβγ δ = ∂β Aαγ δ .
Suppose that both the commutativity condition (D2) and the invariance condition (D3)
hold. Then the potential condition (D5) implies the existence of a single power series
function Φ ∈ C[[tM ]] satisfying
g(∂α ◦ ∂β , ∂γ ) = ∂α ∂β ∂γ Φ.
5.2. Formal Frobenius manifold structures on H. We now construct a formal manifold
structure on the complex manifold H ≃ Cµ by using Theorem 2.10 which says that H ≃
A00 /QS (A−1 0 −1
0 ). Let {uα : α ∈ I} be a C-basis of JS = A0 /QS (A0 ). Then, {uα : α ∈ I} is
t
also a C[[t]]((~))-basis of JS ((~)) := A00 [[t]]((~))/QS (A−10 [[t]]((~))) ≃ JS ⊗C C[[t]]((~)). Consider
the following equation:
X
(5.1) uα · uβ = aαβ ρ · uρ + QS (λαβ ),
ρ∈I
ρ
where aαβ ∈ C and λαβ ∈ A−1
Since {uρ } is a C-basis of JS , aαβ ρ is uniquely determined.
0 .
ρ t
We use the Aαβ to define a structure of C[[t]]-algebra on JS ((~)) as follows:
X
(5.2) uα · uβ = Aαβ ρ · uρ + QS (Λαβ ),
ρ∈I
where Aραβ∈ C[[t]]((~)) and Λαβ ∈ A−1 Since Aαβ ρ is also uniquely determined due
0 [[t]]((~)).
t
to the facts that {uρ : ρ ∈ I} is a C[[t]]((~))-basis of JS ((~)), Aαβ ρ = aαβ ρ ∈ C by (5.1) and
Λαβ is λαβ up to the kernel of QS .
t
We use gαβ in order to define the symmetric C[[t]]((~))-pairing on JS ((~)) as follows:
gαβ := Aαβ max .
Proposition 5.2. The 3-tensor field Aαβ ρ and gαβ satisfy the axioms of the formal Frobe-
nius manifold.
Proof. (i) (Commutativity/Associativity) Since uα · uβ = uβ · uα , Aαβ γ = Aβα γ for all

α, β, γ.
For associativity, consider the following computations:
 
X X
(uα · uβ ) · uγ =  Aαβ ρ · uρ + QS (Λαβ ) · uγ = Aαβ ρ · uρ · uγ + QS (Λαβ uγ )
ρ∈I ρ∈I
!
X X
= Aαβ ρ Aργ δ · uδ + QS (Λργ ) + QS (Λαβ uγ )
ρ∈I δ∈I
   
X X X
=  Aαβ ρ Aργ δ  uδ + QS  Λργ + Λαβ uγ  ,
and
 
X X
uα · (uβ · uγ ) = uα  Aβγ ρ · uρ + QS (Λβγ ) = Aβγ ρ · uρ · uα + QS (uα Λβγ )
ρ∈I ρ∈I
!
X X
ρ δ
= Aβγ Aρα · uδ + QS (Λρα ) + QS (uα Λβγ )
ρ∈I δ∈I
   
X X X
=  Aβγ ρ Aρα δ  uδ + QS  Λρα + uα Λβγ  .
t
Since (uα uβ )uγ = uα (uβ uγ ) and {uρ } is a C[[t]]-basis of JS , we get
X X
Aαβ ρ Aργ δ = Aβγ ρ Aρα δ
ρ∈I ρ∈I
for all α, β, γ, δ.
(ii) (Invariance) By associativity
X X
Aαβγ = Aαβ ρ Aργ max = Aβγ ρ Aρα max = Aβγα ,
ρ∈I ρ∈I
for all α, β, γ.
(iii) (Identity) Since we choose umin = u0 = 1, A0α β = δ βα for all α, β.
(iv) (Potential) Note that Aαβ γ = aαβ γ ∈ C for all α, β, γ. Hence, ∂δ Aαβ γ = 0 for all
α, β, γ, δ.
Remark 5.3. We can extend the solutions aαβ ρ and λαβ of (5.2) to the solutions Aαβ ρ
and Λαβ of the differential equation (1.1)
L L X
~ β~ ~ α~ 1 = uα uβ = Aαβ ρ · uρ + QS+L (Λαβ ) + ~∆(Λαβ ),
ρ
by setting ζ = 1, Γ = L. In other words, in terms of the language of the differential

equation for weak primitive forms, the above construction corresponds to ζ = 1, Γ = L,
and Aαβ ρ = aαβ ρ + 1 Aαβ ρ ~ + 2 Aαβ ρ ~2 + · · · , as we indicated in Table 2. But we have to
sacrifice the condition (2) of Definition 1.1. We could have relaxed the definition of weak
primitive forms by allowing higher ~-power terms in Aαβ ρ .
5.3. Modified higher residue pairings. The higher residue pairings are important tools
for constructing Frobenius manifolds in [22]. We introduce its slight variant, the modified
higher residue pairing for the formal Frobenius structure on H. Recall that TH is the space
of formal tangent vector fields on H given in Definition 2.15.
t
Definition 5.4. Let [u] := u + Im(QS ) in JS := A00 [[t]]/QS (A−1
0 [[t]]). Define the map η as
η : TH × TH −→ C[[t]],
([w1 ], [w2 ]) 7−→ Au1 u2 max , ui ∈ A00 [[t]] and [wi ] = [Φ(ui )]
where Au1 u2 max is uniquely determined by
X
u1 u2 = Au1 u2 ρ uρ + QS (Λu1 u2 )
ρ∈I
where Au1 u2 ρ ∈ C[[t]], Λu1 u2 ∈ A−1

0 [[t]].
The following proposition is due to Konno, Theorem 6.1, [11]; the hypersurface case was
proved in [2] and the case of the smooth projective complete intersection with equal degrees
d1 = · · · = dk was proved in [23].
Proposition 5.5. The following diagram commutes up to a numerical factor:

η|t=0
H×H C
cup product
∼
H 2(n−k) (XG , C) C.
The modified higher residue pairing is an extension of the cup product of H to the
quantization cohomology TH ((~)) so that it is compatible with the Gauss-Manin connection.
S+L
t
Note that Gauss-Manin connection ~ is also well defined on JS ((~)), since QS commutes
S+L S+L
with α
~ and ~
~−1 . For a Laurent series Φ(~) in ~, we set Φ∗ (~) = Φ(−~).
Definition 5.6 (modified higher residue pairing). A modified higher residue pairing for
TH ((~)) is a C[[t]]-bilinear symmetric pairing
K : TH ((~)) × TH ((~)) −→ C[[t]]((~))
X
(w1 , w2 ) 7−→ K(w1 , w2 ) = K (r) (w1 , w2 )~r , K (r) (w1 , w2 ) ∈ C[[t]]
r∈Z
such that
(H1) K(w1 , w2 ) = K(w2 , w1 )∗
(H2) g(~)K(w1 , w2 ) = K(g(~)w1 , w2 ) = K(w1 , g ∗ (~)w2 ), g(~) ∈ C((~))
S+L S+L
(H3) ∂α (K(w1 , w2 )) = K( α ~ w1 , w2 ) + K(w1 , α ~ w2 )
S+L S+L
(H4) ∂~−1 (K(w1 , w2 )) = K( ~−1
~
w1 , w2 ) − K(w1 , ~−1
~
w2 )
(H5) the following diagram commutes:
K
TH [[~]] × TH [[~]] C[[t]][[~]]
~=0 ~=0
η
TH × TH C[[t]],
where η is a C[[t]]-bilinear symmetric pairing coming from the cup-product pairing
n−k
of H = Hprim (XG , C) and the vertical maps substitute ~ with 0.
Theorem 5.7. If we define

K ([w1 ], [w2 ]) := Au1 u∗2 max ,
for [wi ] = [Φ(ui )], i = 1, 2, and
X
u1 u∗2 = Au1 u∗2 ρ uρ + QS (Λu1 u∗2 ),
ρ∈I
then K : TH ((~)) × TH ((~)) → C[[t]]((~)) is the modified higher residue pairing.
Proof. The conditions (H1) and (H2) are obvious by definition.

For (H3), consider the following equations:
S+L S+L
uα ∗ uα ∗
( α ~ u1 )u∗2 + u1 ( α ~ u2 )∗ = ∂α u1 + u1 u2 + u1 ∂α u2 + u2
~ ~
∗ ∗
= (∂α u1 )u2 + u1 ∂α (u2 )
= ∂α (u1 u∗2 ).
We know that ∂α (Au1 u∗2 ρ ) = A∂α (u1 u∗2 ) ρ , since
 
X X
∂α (u1 u∗2 ) = ∂α  Au1 u∗2 ρ uρ + QS (Λu1 u∗2 ) = (∂α Au1 u∗2 ρ )uρ + QS (∂α Λu1 u∗2 ).
ρ∈I ρ∈I
Therefore, condition (H3) holds.

For (H4), consider the following equations:
S+L S+L
( ~
~−1 u1 )u∗2 − u1 ( ~
~−1 u2 )∗ = (∂~−1 u1 + (S + L)u1 )u∗2 − u1 (∂~−1 u2 + (S + L)u2 )∗
= (∂~−1 u1 )u∗2 − u1 (−∂~−1 u∗2 )
= ∂~−1 (u1 u∗2 ).
By the same argument as before, condition (H4) also holds.
The condition (H5) follows from Proposition 5.5.
The reason why we call Definition 5.6 the modified higher residue pairing is its slight
difference from the original higher residue pairings in Theorem 5.1, [22]; the original one is
defined on the quantization cohomology HS+L but our pairing is defined on TH ((~)). Note
that TH ((~)) and HS+L are isomorphic as C[[t]]((~))-modules but it seems very hard to find an
explicit isomorphism between them as C[[t]]((~))-modules with the Gauss-Manin connections
(i.e., as differential modules). This is the essential reason why we can not find an algorithm
of the original higher residue pairing on HS+L , which we explain more details below.
(0) (0)
Definition 5.8. Define the map KS+L : HS+L × HS+L → C[[t]][[~]] as follows:
S+L S+L
KS+L (~ α
~ ζ, ~ β
~
ζ) := 0 Aαβ max ,
where ζ is a weak primitive form satisfying
S+L S+L X S+L
(5.3) ~ β ~ (~ α ~ ζ) = (0 Aαβ ρ + 1 Aαβ ρ ~) · ρ
~ ζ + (QS+L + ~∆)(0 Λαβ + 1 Λαβ ).
ρ∈I
S+L
(0)
Then, extend KS+L to HS+L using the fact that {~ ρ
~ ζ : ρ ∈ I} is a C[[t]][[~]]-basis of
(0)
HS+L as follows:
X
KS+L (Φ, Ψ) := AΦ ρ · (AΨ φ )∗ · 0 Aρφ max ,
ρ,φ∈I
where
X S+L
Φ= AΦ ρ ~ ρ
~ ζ + (QS+L + ~∆)(ΛΦ ),
ρ∈I
X S+L
Ψ= AΨ ρ ~ ρ
~ ζ + (QS+L + ~∆)(ΛΨ ).
ρ∈I
Definition 5.9. Define the map ηe as

A00 [[t]] A00 [[t]]
ηe : × −→ C[[t]],
QS+L (A−10 [[t]]) QS+L (A−10 [[t]])
(w1 , w2 ) 7−→ Aw1 w2 max , wi ∈ A00 [[t]],
where Aw1 w2 max is uniquely determined by
X
w1 w2 = Aw1 w2 ρ uρ · 0 ζ + QS+L (Λw1 w2 )
ρ∈I
and Aw1 w2 ρ ∈ C[[t]], Λw1 w2 ∈ A−1

0 [[t]].
Then, Proposition 5.10 follows from Definition 5.8 immediately:

(0)
Proposition 5.10. The map KS+L satisfies the following conditions for all w1 , w2 ∈ HS+L
and g(~) ∈ C[[~]]:
(h1) KS+L (w1 , w2 ) = KS+L (w2 , w1 )∗ .
(h2) g(~)KS+L (w1 , w2 ) = KS+L (g(~)w1 , w2 ) = K(w1 , g ∗ (~)w2 ).
(h5) The following diagram commutes:
(0) (0) KS+L
HS+L × HS+L C[[t]][[~]]
~=0 ~=0
A00 [[t]] A00 [[t]] e

η
QS+L (A−1
× QS+L (A−1
C[[t]],
0 [[t]]) 0 [[t]])
where the vertical maps substitute ~ with 0.

However, the map KS+L does not necessarily satisfy the following two conditions for all
(0)
w1 , w2 ∈ HS+L and g(~) ∈ C[[~]] in general:
S+L S+L
(h3) ∂α (KS+L (w1 , w2 )) = KS+L ( α
~ w1 , w2 ) + KS+L (w1 , α
~ w2 ).
S+L S+L
(h4) ∂~−1 (KS+L (w1 , w2 )) = KS+L ( ~
~−1 w1 , w2 ) − KS+L (w1 , ~
~−1 w2 ).
S+L S+L
From now on we use the simplified notations α = α and ~−1 = ~ ~
~−1 . We give some
criterions when the weak primitive form ζ guarantees (h3) and (h4):
Proposition 5.11. The map KS+L satisfies the conditions (h3) if the following equation
holds:
X
(5.4) ∂γ 0 Aαβ max = 1
Aγα ρ · 0 Aρβ max + 1 Aγβ ρ · 0 Aαρ max .
ρ∈I
where 0 Aαβ ρ and 1 Aαβ ρ appeared in the equation (5.3).

S+L
Proof. Since the map KS+L is extended by the C[[t]][[~]]-basis {~ ρ
~ ζ : ρ ∈ I}, the
condition (h3) is equivalent to
~∂γ KS+L (~ α ζ, ~ β ζ) = KS+L (~ γ ~ α ζ, ~ β ζ) − KS+L (~ α ζ, ~ γ ~ β ζ).
Suppose that the condition (5.4) holds. Then, consider the following equations:
KS+L (~ γ ~ α ζ, ~ β ζ) − KS+L (~ α ζ, ~ γ ~ β ζ)
X X
= Aγα ρ · 0 Aρβ max − (Aγβ ρ )∗ · 0 Aαρ max
ρ∈I ρ∈I
hX i hX i
0
= Aγα ρ · 0 Aρβ max − 0 Aγβ ρ ·0 Aαρ max +~ 1
Aγα ρ · 0 Aρβ max + 1 Aγβ ρ · 0 Aαρ max
ρ∈I ρ∈I

= 0 + ~ ∂γ 0 Aαβ max
= ~∂γ KS+L (~ α ζ, ~ β ζ).
Therefore, the condition (h3) holds.
Remark 5.12. Suppose that 1 Aαβ ρ is the Christoffel symbol of the Levi-Civita connection
∇ as in Remark 4.1. Then, the compatibility with the metric
∂γ h∂α , ∂β i = h∇∂γ ∂α , ∂β i + h∂α , ∇∂γ ∂β i
is equivalent to the condition (5.4).
For the condition (h4), we define the coefficient Bα ρ ∈ C[[t]][[~]] by

X
(5.5) ~ ~−1 α ζ = Bα ρ ~ ρ ζ + (QS+L + ~∆)(Bα ).
ρ∈I
Note that ~−1 ◦ α = α ◦ ~−1 , and

∂~−1 KS+L (~ α ζ, ~ β ζ) = ∂~−1 0 Aαβ max = 0.
We know that
KS+L ( ~−1 ~ α ζ, ~ β ζ) + KS+L (~ α ζ, ~−1 ~ β ζ)
= KS+L (~ ~−1 α ζ, ~ β ζ) + KS+L (~ α ζ, ~ ~−1 β ζ),
because
KS+L ( ~−1 ~ α ζ, ~ β ζ) = KS+L (−~2 α ζ+~ ~−1 α ζ, ~ β ζ)
= −~KS+L (~ α ζ, ~ β ζ) + KS+L (~ ~−1 α ζ, ~ β ζ)
and
KS+L (~ α ζ, ~−1 ~ β ζ) = KS+L (~ α ζ, −~2 β ζ+~ ~−1 β ζ)
= ~KS+L (~ α ζ, ~ β ζ) + KS+L (~ α ζ, ~ ~−1 β ζ).
By (5.5),
KS+L (~ ~−1 α ζ, ~ β ζ)+KS+L (~ α ζ, ~ ~−1 β ζ)
Xh i
(5.6) = Bα ρ · 0 Aρβ max + (Bβ ρ )∗ · 0 Aαρ max .
ρ∈I
Consider the following equation:

X
~ β ~ ~−1 α ζ =~ β Bα ρ ~ ρ ζ + (QS+L + ~∆)(~ β Bα )
ρ∈I
Xh i
= ~∂β Bα ρ ~ ρ ζ + Bα ρ ~ β ~ ρ ζ + (QS+L + ~∆)(~ β Bα )
ρ∈I
Xh X i
= ~∂β Bα δ + Bα ρ · Aβρ δ ~ δ ζ
δ∈I ρ∈I
X
+ (QS+L + ~∆) ~ β Bα + Bα ρ Λβρ .
ρ∈I
Therefore, the following quantity is invariant under the permutation of α, β for all δ:
X
(5.7) ~∂β Bα δ + Bα ρ · 0 Aβρ δ + ~ · 1 Aβρ δ .
ρ∈I
Proposition 5.13. If Bα ρ = ~Bα ρ with Bα ρ ∈ C[[t]] and for all α, ρ ∈ I, then the condition
(h4) holds.
Proof. Suppose that Bα ρ = ~Bα ρ with Bα ρ ∈ C[[t]] for all α, ρ ∈ I. Then (5.7) implies that
X
Bα ρ · 0 Aβρ δ
ρ∈I
has the same value up to the permutation of α, β for all δ. Therefore the RHS of the
equation (5.6) becomes zero, since
X X
Bα ρ · 0 Aρβ max + (Bβ ρ )∗ · 0 Aαρ max = ~Bα ρ · 0 Aρβ max − ~Bβ ρ · 0 Aαρ max = 0.
ρ∈I ρ∈I
It implies that (h4) holds.
Remark 5.14. If one can find the weak primitive form ζ satisfying the sufficient conditions
in Propositions 5.11 and 5.13, then the map KS+L in Definition 5.8 and ζ would lead to
the construction of the original higher residue pairing and the primitive form in the sense
of K. Saito. However, the algorithm in subsection 4.2 cannot provide the weak primitive
form satisfying those sufficient conditions.
6. Appendix
6.1. Comparisons. We compare our construction with the Barannikov-Kontsevich’s dGBV
algebra ([1]) for a compact Calabi-Yau manifold X = XG (C). In [1], the holomorphic
polyvector field valued in anti-holomorphic differential forms is twisted with a nowhere
vanishing holomorphic volume form Ω, so that the twisted algebra structure gives rise to
the dGBV algebra; ∂ becomes the BV operator ∆, and ∂¯ is the differential. The exterior
derivative d becomes ∆+ ∂. ¯ We don’t use polyvector fields on the compact Calabi-Yau man-
ifold (to construct dGBV algebras), which is a key tool in [1]; we rather use the polyvector
fields on the affine space An+k+1 (by looking at the complement of XG in Pn and using the
Table 2. Comparisons
[1] [22] Section 3 Section 4 Section 5
Qf
Q ∂ f has an isolated QS(q) QS(q) QS(q)
singularity
the twist of ∂
P ∂ ∂
P ∂ ∂
P ∂ ∂
P ∂ ∂
∆ using the ∂xi ∂ηi ∂qi ∂ηi ∂qi ∂ηi ∂qi ∂ηi
volume form Ω
Γ= X
X 1 X L= tα u α
Γ tα u α Γ L L
n! α α
n≥1 (linear)
(general)
0
0 0 aαβ ρ
ρ 0 ρ Aαβ ρ 0 ρ Aαβ ρ
Aαβ Aαβ 1 Aαβ 1 +1 Aαβ ρ ~
+ Aαβ ρ ~ + Aαβ ρ ~
+2 Aαβ ρ ~2 +···
ζ=0 ζ+1 ζ~+2 ζ~2 +··· 0

ζ 1 1 ζ + 1 ζ~ 1
(general)
Cayley trick) instead of the compact manifold X. The difference between our dGBV alge-
bra and Barannikov-Kontsevich’s dGBV algebra lies in the different decomposition of the
total differential d = ∆ + ∂¯ and K = ∆ + QS , and consequently, we can provide a relevant
n−k
theory on the primitive middle dimensional cohomology H = Hprim (X, C) of X instead of
•
the whole cohomology group H (X, C). A key difference coming from the different decom-
position of d results in the failure of the QS ∆-lemma (but there is its weak version; see
Lemma 2.8) in our theory, which prohibits us from applying Barannikov-Kontsevich’s the-
ory (where the ∂∂-lemma plays a key role) directly to our example TH . In the hypersurface
case, an explicit cochain map between the two constructions was given in [10].
Now we do a comparison with Saito-Takahashi [22]. The fact that the differential QS is
a derivation with respect to the multiplication of A[[t]] (so that H 0 A[[t]]((~)), QS is again
an algebra with the induced multiplication from A[[t]]((~))), enables us to develop an explicit
algorithm based on the Gröbner basis of the Jacobian ideal of the polynomial S. This feature
is not available in [1]: the differential ∂ (the Dolbeault differential) can not be studied by
the Jacobian ideal of some polynomial. The polynomial f in [22] has an isolated singularity
and our algorithm of weak primitive forms can be applied to [22] with the differential Qf .
On the other hand, our polynomial S has a non-isolated singularity which needs a new idea
to apply the Gröbner basis algorithm; assuming that XG is Calabi-Yau, we will restrict A to
its charge zero part A0 so that the cohomology algebra H 0 A0 , QS is finite dimensional13
over C and the relevant algorithm works. Another difference is that we have a filtration
(which we call the weight filtration) on the dGBV algebra which is compatible with the
Hodge filtration on H; this enables us to define the modified higher residue pairing rather
easily (see Theorem 5.7). Note that defining the higher residue pairing in the same way
(using the residue symbol) as [20] is difficult due to the fact that S has a non-isolated
singularity.
6.2. The Riemann-Hilbert-Birkhoff problem. The Riemann-Hilbert-Birkhoff problem

is important to prove the existence of primitive forms associated to the higher residue pairing
in [22]. In this appendix, we provide Lemma 6.1 which is a reinterpretation of the RHB
13Note that H 0 A, Q is infinite dimensional over C.

S
problem in the case t = 0 (Lemma 6.3 in [22]).14 Lemma 6.1 will be a simple consequence
of the weight (the Hodge filtration) homogeneity of the solution. Note that we can choose a
C-basis {uα : α ∈ I} = {u0 , . . . , uµ−1 } of A00 /QS (A−1
0 ) such that wt(uα ) corresponds to the
increasing Hodge filtration under the isomorphism A00 /QS (A−1 ∼ n−k
0 ) = Hprim (XG ) with u0 = 1.
Recall the following C[[~]]-module:
(0) A00 [[~]]
HS := .
(QS + ~∆)(A−1 0 [[~]])
Define the set {v0 , . . . , vµ−1 } as follows:

∞
X
ℓ
vi = ui + vi · ~ℓ , i = 0, . . . , µ − 1,
ℓ=1
where ℓ vi is an arbitrary polynomial in A00 satisfying the weight condition wt(ℓ vi ) + ℓ =

(0)
wt(ui ) for all ℓ ≥ 1. Note that {v0 , . . . , vµ−1 } is a C[[~]]-basis of HS by the same argument
in the proof of Lemma 2.19. Then, we get the following Lemma:
(0)
Lemma 6.1. The above C[[~]]-basis {v0 , . . . , vµ−1 } of HS satisfies the following conditions:
(i) v0 |~=0 = 1.
S
(0) t~
(ii) ~
~−1
~v ≡ CS ~v − ~N0 ~v in HS where v = (v0 , . . . , vµ−1 ), CS = 0 ∈ M (µ, C), and
 
k + wt(u0 ) ··· 0
 .. 
 . k + wt(u1 ) 
N0 =   ∈ M (µ, Q).
.. .. 
 . . 
0 ··· k + wt(uµ−1 )
Proof. The condition (i) is trivial by the definition of v0 .

For the condition (ii), define Λi ∈ A−10 [[~]] as
∞
X k
X
ℓ
Λi := Λ i · ~ℓ where ℓ Λi = (yj · ℓ vi )ηj .
ℓ=0 j=1
Then, we get the following consequences:

k
X k
X

QS (ℓ Λi ) = QS (yj · ℓ vi )ηj = yj ∂yj S)ℓ vi = S · ℓ vi ,
j=1 j=1
k
X k
X

∆(ℓ Λi ) = ∆ (yj · ℓ vi )ηj = (∂yj yj · ℓ vi + yj · ∂yj ℓ vi ) = (k + wt(ℓ vi ))ℓ vi
j=1 j=1
ℓ
= (k − ℓ + wt(ui )) vi .
This implies the following system of equations:
S · 0 vi = QS (0 Λi ),
S · 1 vi − 1 vi = −(k + wt(ui ))0 vi + QS (1 Λi ) + ∆(0 Λi ),
..
.
14The general RHB problem with non-zero t-power coefficients (Lemma 6.6 in [22]) will not be considered
here, since the main interest of this article is weak primitive forms rather than primitive forms.
The above equations are equivalent to
−~(k + wt(ui ))vi + (QS + ~∆)(Λi ) = S · vi − ~(1 vi ~ + 2 · 2 vi ~2 + 3 · 3 vi ~3 + · · · )

S
= Svi + ∂~−1 (vi ) = ~
~−1 vi .
Therefore,
S
(0)
~
~−1 vi ≡ −~(k + wt(ui ))vi in HS for all i = 0, . . . , µ − 1,
S
(0)
i.e., ~
~−1
~v ≡ CS ~v − ~N0 ~v in HS .
6.3. Comments on the algorithm. In the body of the paper, we concentrated on solving
X
(6.1) ~ β~ αζ = Aαβ ρ ~ ρ ζ + (QS+L + ~∆)(Λαβ ),
ρ
S+Γ
where α := α ~ , in certain cases (see Definition 1.1) which is related to some special
structures such as formal F -manifolds, Frobenius manifolds and higher residue pairings. In
this appendix, we provide some hints how to deal with P (6.1) in the other cases regardless
of those special structures, assuming that Γ = L := α tα uα .
We vary conditions on the vanishing of ~ℓ -coefficients of Aαβ ρ and ζ, to check whether we
can construct relevant algorithms. In the following, “unsolvable” means that our algorithm
does not work. Consider the following cases:
(i) (ζ = 0 ζ, Aαβ ρ = 0 Aαβ ρ : unsolvable) The equation (6.1) becomes
X
~ β ~ α 0ζ = 0
Aαβ ρ ~ ρ 0 ζ + (QS+L + ~∆)(Λαβ ),
ρ
X
0 0 0 2 0 0
uα uβ ζ + ~(uβ ζα + uα ζβ ) + ~ ( ζαβ ) = Aαβ ρ uρ 0 ζ + ~(0 Aαβ ρ · 0 ζρ )
ρ
+ (QS+L + ~∆)(Λαβ ).
Therefore, by comparing the ~-power terms, we have to solve the following equa-
tions:
X
uα uβ 0 ζ = 0
Aαβ ρ uρ 0 ζ + QS+L (0 Λαβ ),
ρ
X
0 0 0
uβ ζα + uα ζβ = Aαβ ρ · 0 ζρ + ∆(0 Λαβ ) + QS+L (1 Λαβ ),
ρ
0
ζαβ = ∆(1 Λαβ ).
However, it is difficult to solve the equation

X
(6.2) uβ 0 ζα + uα 0 ζβ = 0
Aαβ ρ · 0 ζρ + ∆(0 Λαβ ) + QS+L (1 Λαβ )
ρ
by the similar algorithm in P subsection 4.2, since it is hard to choose 0 ζα for all α ∈ I
such that uβ ζα + uα ζβ − ρ 0 Aαβ ρ · 0 ζρ − ∆(0 Λαβ ) ∈ Im(QS+L ). But if we allow
0 0
the ~-coefficient 1 Aαβ ρ to be non-zero, we can find the algorithm.

(ii) (ζ = 0 ζ, Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~ : solvable) The right hand side of the equation
(6.1) becomes
X
0
Aαβ ρ uρ 0 ζ + ~(0 Aαβ ρ · 0 ζρ + 1 Aαβ ρ uρ ) + ~2 (1 Aαβ ρ · 0 ζρ ) + (QS+L + ~∆)(Λαβ ).
ρ
Therefore, we have to solve the following equations:

X
uα uβ 0 ζ = 0
ρ
X
uβ 0 ζα + uα 0 ζβ = 0
Aαβ ρ · 0 ζρ + 1 Aαβ ρ uρ + ∆(0 Λαβ ) + QS+L (1 Λαβ ),
ρ
X
0 1
ζαβ = Aαβ ρ · 0 ζρ + ∆(1 Λαβ ).
ρ
We can solve the above equations because of the existence of the terms 1 Aαβ ρ and
0
ζαβ . For example, unlike the equation (6.2), we can choose 0 ζα for all α ∈ I such
that
X X
uβ 0 ζα + uα 0 ζβ − 0
Aαβ ρ · 0 ζρ − ∆(0 Λαβ ) = 1
Aαβ ρ uρ + QS+L (1 Λαβ ),
ρ ρ
which uniquely determines 1 Aαβ ρ . In addition, we can put 0 ζαβ as

X
0 1
ζαβ = Aαβ ρ · 0 ζρ + ∆(1 Λαβ ).
ρ
(iii) (ζ = 0 ζ + 1 ζ~, Aαβ ρ = 0 Aαβ ρ : unsolvable) The left hand side of the equation (6.1)
becomes
0
~ β~ α( ζ + 1 ζ~)
= uα uβ 0 ζ + ~(uα uβ 1 ζ + uβ 0 ζα + uα 0 ζβ ) + ~2 (uβ 1 ζα + uα 1 ζβ + 0 ζαβ ) + ~3 (1 ζαβ ),
and the right hand side of the equation (6.1) becomes

X
0
Aαβ ρ uρ 0 ζ + ~(0 Aαβ ρ (uρ 1 ζ + 0 ζρ )) + ~2 (0 Aαβ ρ · 1 ζρ ) + (QS+L + ~∆)(Λαβ ).
ρ
Then, we get the equation

X
uα uβ 1 ζ + uβ 0 ζα + uα 0 ζβ = 0
Aαβ ρ (uρ 1 ζ + 0 ζρ ) + QS+L (1 Λαβ ) + ∆(0 Λαβ ),
ρ
which is difficult to solve by the similar algorithm in subsection 4.2, since it is hard
to choose 0 ζα for all α ∈ I such that
X
uα uβ 1 ζ + uβ 0 ζα + uα 0 ζβ − 0
Aαβ ρ (uρ 1 ζ + 0 ζρ ) − ∆(0 Λαβ ) ∈ Im(QS+L ).
ρ
But again if we allow the ~-coefficient 1 Aαβ ρ to be non-zero, we can find the algo-
rithm.
(iv) (ζ = 0 ζ + 1 ζ~, Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~ : solvable) This case is the main algorithm
done in subsection 4.2.
(v) (ζ = 0 ζ + 1 ζ~ + 2 ζ~2 + · · · , Aαβ ρ = 0 Aαβ ρ : unsolvable) By the same argument in
(i) and (iii), the following equation from the equation 6.1
X
uα uβ 1 ζ + uβ 0 ζα + uα 0 ζβ = 0
Aαβ ρ (uρ 1 ζ + 0 ζρ ) + QS+L (1 Λαβ ) + ∆(0 Λαβ ),
ρ
is unsolvable, since it is difficult to choose 0 ζα for all α ∈ I satisfying the above

equation. Once again if we allow the ~-coefficient 1 Aαβ ρ to be non-zero, we can
find the algorithm.
(vi) (ζ = 0 ζ + 1 ζ~ + 2 ζ~2 + · · · + n ζ~n , Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~ : solvable for any
n ≥ 0) The case for n = 0, 1 is done in (ii) and (iv). For n = 2 case, by the same
argument in (ii) and (iv), we have the following equations:

X
uα uβ 0 ζ = 0
ρ
X
1 0 0 0
uα uβ ζ + uβ ζα + uα ζβ = Aαβ ρ (uρ 1 ζ + 0 ζρ ) + 1 Aαβ ρ uρ 0 ζ
ρ
+ ∆(0 Λαβ ) + QS+L (1 Λαβ ),

X
uα uβ 2 ζ + uβ 1 ζα + uα 1 ζβ + 0 ζαβ = 0
Aαβ ρ (uρ 2 ζ + 1 ζρ ) + 1 Aαβ ρ (uρ 1 ζ + 0 ζρ )
ρ
+ ∆(1 Λαβ ),
X
uβ 2 ζα + uα 2 ζβ + 1 ζαβ = 0
Aαβ ρ · 2 ζρ + 1 Aαβ ρ (uρ 2 ζ + 1 ζρ ),
ρ
X
2 1
ζαβ = Aαβ ρ · 2 ζρ .
ρ
They are solvable because of the existence of the terms 1 Aαβ ρ , 0 ζαβ , 1 ζαβ and 2 ζαβ .
The general case n ≥ 3 can be checked similarly.
These examples provide a general shape of how to solve (6.1) and we leave to the reader
to think about the case dealing with the higher coefficients ℓ Aαβ ρ of ~ℓ (ℓ ≥ 2) involved.
References
[1] Barannikov, S.; Kontsevich, M.: Frobenius manifolds and formality of Lie algebras of polyvector fields,
Internat. Math. Res. Notices 4 (1998) 201–215.
[2] Carlson J.; Griffiths P.: Infinitesimal variations of Hodge structure and the global Torelli problem, in
Journees de geometrie algebrique, Angers, juillet 1979, Sijthoff and Noordhoff, Alphen aan den Rijn,
1980, 51–76.
[3] Cox, David A.; Katz, Sheldon:Mirror symmetry and algebraic geometry. Mathematical Surveys and
Monographs, 68. American Mathematical Society, Providence, RI, 1999. xxii+469 pp.
[4] Dimca, A.: Residues and cohomology of complete intersections, Duke Math. J., 78 (1995) no. 1, 89–100.
[5] Dubrovin, B. A.: Geometry of 2D topological field theories, In Integrable systems and quantum groups
(Montecatini Terme, 1993), Lecture Notes in Math. vol. 1620, 120–348. Springer-Verlag, Berlin (1996).
[6] Dwork, B.: On the rationality of the zeta function of an algebraic variety, Amer. J. Math., vol. 82
(1960), pp. 631–648.
[7] Griffiths, Phillip A.: On the periods of certain rational integrals. I, II, Ann. of Math. (2) 90 (1969),
460–495; ibid. (2) 90 (1969), 496–541.
[8] Hertling, Claus: Frobenius manifolds and moduli spaces for singularities. Cambridge Tracts in Math-
ematics, 151. Cambridge University Press, Cambridge, 2002. x+270 pp.
[9] Hertling, C.; Manin, Y.: Weak Frobenius Manifolds, Internat. Math. Res. Notices 6 (1999) 277–286.
[10] Kim, Dokyoung; Kim, Yesule; Park, Jeehoon: Differential Gerstenhaber-Batalin-Vilkovisky algebras
for Calabi-Yau hypersurface complements. Mathematika 64 (2018), no. 3, 637–651.
[11] K. Konno: On the variational Torelli problem for complete intersections, Comp. Math., 78 (1991),
271–296.
[12] Li, Changzheng; Li, Si; Saito, Kyoji: Primitive forms via polyvector fields, available at
https://arxiv.org/abs/1311.1659
[13] Manin, Y.: F -manifolds with flat structure and Dubrovin’s duality. Adv. Math. 198 (2005), no. 1,
5–26.
[14] Manin, Y.: Frobenius manifolds, quantum cohomology, and Moduli spaces, American Mathematical
Society Colloquium Publications, 47. American Mathematical Society, Providence, RI, 1999. xiv+303
pp. ISBN: 0-8218-1917-8
[15] Manin, Y.: Three constructions of Frobenius manifolds: a comparative study Surv. Differ. Geom., 7,
Int. Press, 497–554.
[16] Park, Jae-Suk: Algebraic Principles of Quantum Field Theories. I, II, preprints arXiv:1101.4414,
arXiv:1102.1533.
[17] Park, Jae-Suk; Park, Jeehoon: Enhanced homotopy theory for period Integrals of smooth projective
hypersurfaces, Communications in Number theory and Physics, Volume 10 (2016), Number 2, Pages
235–337.
[18] Sabbah, C.: Déformations isomonodromiques et variétés de Frobenius, une introduction. Centre de
Mathematiques, Ecole Polytechnique, U.M.R. 7640 du C.N.R.S., no. 2000-05, 251 pages.
[19] Saito, K.: Primitive forms for an universal unfolding of a functions with isolated critical point, Journ.
Fac. Sci. Univ. Tokyo, Sect. IA Math. 28 no.3 (1981) 777–792.
(k)
[20] Saito, K.: The higher residue pairings KF for a family of hypersurface singular points, Proceedings
of Symposia in Pure Mathematics Vol. 40 (1983), part 2, 441–463.
[21] Saito, M.: On the structure of Brieskorn lattice, Ann. Inst. Fourier (Grenoble) 39 (1989), no.1, 27–72.
[22] Saito, K.; Takahashi, A.: From primitive forms to Frobenius manifolds, From Hodge theory to integra-
bility and TQFT tt*-geometry, 31–48, Proc. Sympos. Pure Math., 78, Amer. Math. Soc., Providence,
RI, 2008.
[23] Terasoma, T.: Infinitesimal variation of Hodge structures and the weak global Torelli theorem for
complete intersections, Ann. of Math., 132 (1990), 213–235.
[24] Witten, E.: Phases of N = 2 theories in two dimensions, Nuclear Physics B, Volume 403, Issues 1-2,
16 August 1993, Pages 159–222
Email address: yglee.math@gmail.com
Department of Mathematics, POSTECH (Pohang University of Science and Technology), San

31, Hyoja-Dong, Nam-Gu, Pohang, Gyeongbuk, 790-784, South Korea.
Email address: jeehoonpark@postech.ac.kr

Email address: yimjaehyun@postech.ac.kr


2011 09628

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2011 09628

Uploaded by

Copyright:

Available Formats

SMOOTH PROJECTIVE CALABI-YAU COMPLETE INTERSECTIONS

AND ALGORITHMS FOR THEIR FROBENIUS MANIFOLDS AND

YOUNGGI LEE, JEEHOON PARK, AND JAEHYUN YIM

Abstract. The goal of this article is to provide an explicit algorithmic construc-

5.2. Formal Frobenius manifold structures on H 28

Table 1. Weak primitive forms

Classical Complex (~ → 0) Quantization complex (~-perturbation)

Primitive forms and

be a Maurer-Cartan solution of (A, ∆ + Q, ℓ∆+Q 2 (·, ·)):

(2) the coefficient Aαβ is of the form Aαβ ρ = 0 Aαβ ρ + 1 Aαβ ρ ~.

From the definition, we have

1.1. Acknowledgement. The problem of constructing an algorithm for formal F -manifolds

2. The dGBV Algebra and Quantization

where we introduce the formal variables y1 , . . . , yk corresponding to G1 , . . . , Gk . Let N =

2.2. Gradings on A. We are going to introduce another two Z-gradings on A defined in

Associated with ch(qµ )µ=1,...,N , we define the Euler vector

Proof. Note that KS = QS + ∆. Since QS preserves wt and ∆ decreases wt by 1, the result

Note that A = A[η], where ηµ ην = −ην ηµ , and

is a well-defined section of pr1 . Hence there exists an isomorphism

Definition 2.2. We define a sequence of maps as follows:

Therefore, we have the following maps:

is a dGBV algebra. We briefly review the definitions of G-algebras, BV-algebras, GBV-

for any homogeneous elements a, b, c ∈ C.

We also introduce the C-linear map

Proposition 2.5. The quintuple (A, ·, KS , ℓK

Proposition 2.6. For each charge λ ∈ Z, we have a cochain subcomplex (Aλ , KS ).

Proposition 2.7. The cohomology H = H(A, KS ) is concentrated in charge cX , i.e., H =

9(C, ·, K) is called a BV algebra, if (C, ·, K, ℓK ) is a GBV algebra

we see that any f ∈ Aλ ∩ Ker KS belongs to Im KS unless λ = cX , since (λ − cX )f =

Lemma 2.8 (Weak QS ∆-lemma). For f ∈ A−1

for some v ∈ A−2

Lemma 2.9. We have an isomorphism

Proof. Let dim A0cX /QS (A−1 0

We want to show aα = 0 for all α, i.e., {eα + Im(QS + ∆) : α ∈ I} is linearly independent

where wt(η(i) ) = i. Note that wt(∆) = −1 and QS (ηn−k+1 ) = 0.

Then we can write −∆(u0 ) as

for some nα ∈ C and u ∈ A−1

Theorem 2.10. There is an isomorphism

Remark 2.11. It is well-known that there is an isomorphism between H 0 (AcX , QS ) and

for m ∈ Z. We define a cochain map r(0)

by taking the quotient11 by A•cX ((~))(−1) :

(m) A0cX ((~))(m) A0cX [[~]]~−m

We also have an exact sequence:

Proposition 2.14. If the background

Definition 2.15. Let µ := dimC H = |I|. Then,

Let us consider the deformation of dGBV algebra. If we let {tα : α ∈ I} to be the

11Sending ~ = 0 is the classical limit of the quantization complex

for m ∈ Z. We call this as the ~-filtration.

Proof. First, we prove that {uα : α ∈ I} is a C[[~]]-basis of A00 [[~]]/(QS + ~∆)(A−1

By comparing the ~-power terms, we get

3. An algorithm for formal F -manifolds

Definition 3.1. Let M be a complex manifold of finite dimension, whose holomorphic

(3.1) [X ◦ Y, Z ◦ W ] − [X ◦ Y, Z] ◦ W − Z ◦ [X ◦ Y, W ] − X ◦ [Y, Z ◦ W ] + X ◦ [Y, Z] ◦ W

Proposition 3.2. Let ◦ : TM × TM → TM be a OM -bilinear operator. Assume that (M, ◦)

Proof. It is sufficient to check for commuting vector fields (X, Y, Z, W ) = (∂α , ∂β , ∂γ , ∂δ ).

for some Λαβ ∈ A−1 ρ

(3.4) = ~∂γ Aαβ ρ Γρ + Aαβ ρ ~ γ

where uα ∈ A00 , aαβ ρ , aαβα ρ ∈ C, λαβ , λαβα ∈ A−1 0 , and t

where the notation α1 ⊔ · · · ⊔ αm−i = α means α1 ∪ · · · ∪ αm−i = α and αk ∩ αℓ = ∅ for

Now we give the algorithm for Γ and Aαβ ρ as follows:

(0) (0) (0)

in the equations (3.3) gives the following equation

uαβ + uβα = 2∆(λαβ ),